A Riemann-integrable function need not be continuous. Consequently, applying a continuous operation to its values does not automatically produce a continuous function. The useful conclusion is instead that a continuous outer function preserves Riemann integrability. A proof through interval oscillations shows why continuity of the outer function matters, and why mere integrability of that outer function is insufficient.
The Darboux supplement proves stability under absolute value and products. The Riemann integration unit supplies the Darboux criterion, and the uniform convergence unit proves integrability of uniform limits. This chapter develops the first two foundations into general continuous composition, clipping and a square-root estimate. The finite-exception method used for discontinuous integrands reappears in the worked questions on composition and its limits, with a different purpose: testing composition with a discontinuous outer function. The worked questions on square roots and clipping give quantitative versions of the estimates below.
Reading route
First express a Darboux gap as a weighted sum of oscillations. Then prove the Lipschitz composition case, and only afterwards replace its uniform slope bound with uniform continuity. Prerequisites are suprema and infima, uniform continuity on a compact interval, common refinements, and elementary inequalities. Everything here concerns proper Riemann integrals on finite closed intervals.
Oscillation is the quantity to control
Definition 18.1.1
(Oscillation and the Darboux gap)
Let , let be bounded on , and let have cells of lengths . The oscillation of on and the Darboux gap of are
The Darboux criterion says that is integrable exactly when these gaps can be made arbitrarily small. Individual cell oscillations need not all be small. A jump can leave a large oscillation in a cell of very small length, with a small contribution to the total.
Concept lensGeometric
The gap as the area between two step functions
On each cell let equal and equal . These are step functions with , and . So is integrable exactly when some pair of step functions trapping encloses arbitrarily small area. In this picture the bad cells of the proof below are the thin tall columns, and their small total width is what keeps the area small.
Common mistake
(A small gap does not mean small oscillation on every cell)
It is tempting to read integrability as "on a fine enough partition every cell has small oscillation". That is false: for the indicator on , every partition has a cell containing on which , however fine the partition is. What becomes small is the weighted sum: with equal cells, at most two of them meet , so . This is why the proof below must separate good cells from bad cells instead of bounding every oscillation.
For bounded real functions, linear combinations are integrable when their summands are: cell oscillations satisfy the corresponding triangle bound, and a common refinement controls all the needed gaps. The Darboux supplement also proves the absolute-value estimate
This is the simplest example of composition stability. It does not prove a converse: an operation can erase the sign information responsible for oscillation.
First prove the case with a Lipschitz constant
Proposition 18.1.2
(Composition with a Lipschitz function)
Let be Riemann integrable on with values in an interval , and let satisfy for all . Then on every cell, for every partition, and is Riemann integrable.
Proof
(Differences of composite values)
For any two points in a partition cell, the difference of the composite values is at most times the difference of the inputs, hence at most . Taking suprema of such differences gives , and multiplying by and summing gives . The composite is bounded, for example by comparing every value with its value at one fixed point of . Thus the Darboux criterion proves its integrability whenever is integrable. No nearest extrema inside the cells need to exist; differences can approximate the supremum and infimum.
A clipping function that restricts real values to a fixed closed interval is 1-Lipschitz. It therefore preserves integrability, even when the original integrand has jumps. The worked questions on clipping prove this and quantify how clipping affects an error.
Worked example 18.1.3
(Clipping a step function)
Let on and on , and let , which is 1-Lipschitz. Then on and on . For a partition whose cell contains in its interior, with length , and on that cell, and both oscillations vanish elsewhere. So and . Refining to make small makes both gaps small, and . The inequality is not an equality: clipping loses part of the jump.
Continuous outer functions need no bounded derivative
Theorem 18.1.4
(Composition with a continuous function)
Let be Riemann integrable on with range in a compact interval , and let be continuous on . Then is Riemann integrable on .
Proof
(Good cells and bad cells)
If , all integrals vanish. Otherwise uniform continuity and boundedness of are available. Let , choosing , and put .
To make the composite gap less than a prescribed , first choose so that
Then choose a partition for with
Call a cell good when , and bad otherwise. On a good cell, all pairs of values differ by less than , so its composite oscillation is at most . The total good-cell contribution is at most .
For bad cells, , so their total length satisfies
Their composite oscillations are at most . Consequently,
Boundedness of the composite follows from . The Darboux criterion completes the proof. The strict inequality comes from the partition choice, even if a good-cell supremum attains the non-strict upper estimate.
Proof X-ray
Order of choices and the two kinds of cell
The tolerances must be chosen in dependency order: first, then from the uniform continuity of , and only then the partition for , whose gap bound contains . Choosing the partition first would leave undetermined, so the argument would be circular.
Each hypothesis has a distinct job. Uniform continuity of , which follows from continuity on the compact , turns small input oscillation into small output oscillation on good cells, with one for all cells. Integrability of is used only through being small. Boundedness of gives the crude bound on bad cells. The non-obvious move is Chebyshev-style: a cell with oscillation at least costs at least of the small gap , so the bad cells cannot occupy much length. The proof closes because bad cells have small total length and bounded damage, while good cells have uniformly small damage.
If is only continuous on a non-compact interval, need not be choosable uniformly for all values, which is why is compact.
For a nonnegative integrable function, square root is a valid continuous outer function on a compact interval containing its range. Its derivative is unbounded near zero, so a bounded-derivative argument would not justify that application. Uniform continuity is the appropriate hypothesis. The worked questions on square roots give a direct quantitative bound for its Darboux gaps.
Worked example 18.1.5
(Cosine of the indicator of a point)
Let at and elsewhere on , and let , which is continuous on . Then except at , where it equals . For the uniform partition into cells, lies in at most two cells (it is a node when is even), so and . Both tend to , so is integrable with : a single point does not change an integral. This is the bad-cell estimate in miniature, since the bad cells are the one or two cells touching .
What this theorem does and does not reverse
If a continuous outer function is invertible on the relevant compact value interval, with continuous inverse on its image, composition with that inverse can recover integrability of the original function. For instance, cube and real cube root give an equivalence for bounded real : integrability of is equivalent to integrability of .
Squaring has no inverse on an interval containing both signs. It can remove the oscillation of a nonintegrable sign-changing function. The worked questions on composition test this with a nonconstant magnitude, so integrability of a transformed function cannot simply be read backward.
Quick check
(Let for rational and for irrational on . Is integrable? Is ?)
Compute the square, then the oscillation of on any cell.
Solution · Answer
is integrable, but is not: every cell contains a rational point (value ) and an irrational point (value ), so and for every partition. Squaring discards the sign that carries the oscillation, so integrability of with does not return to .
Composition stability also says nothing like
Worked example 18.1.6
(A nonlinear function does not commute with the integral)
Take and on . The left side is . The right side is . Since , the two sides differ. The theorem guarantees the existence of the left integral, not a nonlinear rule for its value.
Finally, continuity of the outer function is a substantive sufficient hypothesis. An outer function with a single jump can be integrable and still turn an integrable inner function into the rational indicator. The composition questions verify both original integrals and directly disprove integrability of the composite; they do not rely on a general discontinuity-set criterion.
Counterexample mode
An integrable outer function is not enough
Claim: if and are both Riemann integrable then is. Let list the rationals of without repetition, put and at irrationals, and let , for . The function is integrable with integral (proved in the questions on composition), and is integrable because it has a single jump, with integral . At a rational point , so ; at an irrational point and . Hence is the rational indicator, with supremum and infimum on every cell, so for every partition and it is not integrable. The hypothesis broken is continuity of at the value taken by .
Quick check
(Is the square root Lipschitz on ? Does the composition theorem still apply to for integrable with values in ?)
Look at the difference quotient between and .
Solution · Answer
The difference quotient is unbounded as , so is not Lipschitz on . The continuous-outer-function theorem still applies, because is continuous on the compact interval , so is integrable.
Quick check
(The floor function is discontinuous. Is integrable for on ?)
Compute on .
Solution · Answer
for and , so the composite is except at one point and is integrable with integral . Continuity of the outer function is sufficient, not necessary; here the discontinuities of are not values that takes on a set of positive length.
Checklist
- Specify boundedness, the finite closed integration interval and the interval containing the values supplied to the outer function.
- For a Lipschitz operation, bound every cell oscillation directly. For a merely continuous operation, use uniform continuity and control the total length of bad cells.
- Choose uniform-continuity tolerance before choosing the partition for the inner function. Keep their dependencies explicit.
- Integrability of the inner function does not mean continuity of that function.
- Both functions being Riemann integrable does not suffice for arbitrary composition. Check which function is outer.
- Do not infer integrability backward through an operation that discards information, or interchange a nonlinear operation with an integral.