Summary
This lesson decides which functions satisfy the integrability criterion of the preceding lesson. It proves that every monotonic function on a closed and bounded interval is integrable, because the difference between its upper and lower sums with equal subintervals is a column of strips of width and height , even if the function has infinitely many jumps; and that every continuous function on a closed and bounded interval is integrable, by means of the Heine-Cantor theorem from the unit on sequences. It then proves that finitely many discontinuities do not prevent a bounded function from being integrable, and that changing the value of an integrable function at finitely many points does not change its integral. Using the density of the rationals and of the irrationals, it proves that the Dirichlet function, bounded and discontinuous at every point, is not integrable: its lower integral and its upper integral are different. Finally, it proves that the composition of an integrable function with a continuous function is integrable, and shows with Thomae’s function that the continuity of the outer function cannot be replaced by integrability.
Learning objectives
- Prove that monotonic functions and continuous functions on a closed and bounded interval are integrable, and bound in advance the difference between their sums.
- Prove that finitely many discontinuities do not prevent a bounded function from being integrable, and that finitely many values do not change the integral.
- Prove, using the density of the rationals and of the irrationals, that the Dirichlet function is not integrable, and compute lower and upper integrals of functions that are not integrable.
- Prove that composition with a continuous function preserves integrability, and recognize through examples which hypotheses cannot be omitted.
Monotonic functions and continuous functions
The lesson on the Riemann integral defined the integral of a function bounded on as the number common to its lower integral and its upper integral, and proved Riemann’s integrability criterion: is integrable if and only if, for every , some partition satisfies . It remains to find out which functions satisfy it. The lesson on the area problem observed that, for an increasing function, the difference between the sums with equal subintervals is a column of strips; that observation yields the first family of integrable functions. We say that is increasing on if whenever , decreasing if under the same conditions, and monotonic if it is increasing or decreasing; in what follows, is the uniform partition with subintervals of , with .
Let be a monotonic function on , with . Then is bounded and integrable on , and for every natural number :
On each subinterval, the extreme values of an increasing function are its values at the endpoints of the subinterval; when the two sums are subtracted, each strip begins at the height where the preceding one ends, and all of them together form a column. The figure shows the sums of an increasing function with a jump, the column formed by its strips and the narrower column of a partition with twice as many subintervals.
Proof
Let be increasing. For every in , , and is bounded. On subinterval of , the value is a lower bound of the values of , and does not exceed their infimum, which exists by the completeness axiom, from the lesson on number systems, and is the greatest one; and it is one of those values, and is not less than the infimum. It is, therefore, the infimum. Likewise, is the supremum. All the subintervals have length .
By the telescoping sum theorem, from the lesson on summations, progressions and induction, the sum equals , which is not negative.
If is decreasing, , the infimum on subinterval is and the supremum , and the telescoping sum gives , which is the same expression. In both cases, the difference is a constant times , which tends to by the proposition on the powers of the reciprocal, from the lesson on sequences and their limit, and by the algebra of limits. The corollary of the criterion with sequences of partitions, from the lesson on the Riemann integral, completes the proof.
The theorem does not require continuity: a monotonic function may have jumps, even infinitely many, as the solved problem “An increasing function with infinitely many jumps” shows. The equality also gives an error bound before any computation: since the integral lies between and , by the proposition on the order of the lower and upper integrals, each of the two sums is at a distance of at most from the integral. For functions that rise and fall, the column breaks, and another tool is needed: measuring how much the function varies on each subinterval.
Let be a function bounded on an interval , with , and let and be the infimum and the supremum of its values there, which exist by the completeness axiom. If and for all and in , then . Conversely, for all and in . If any two values are at most apart, so are the supremum and the infimum, which are limits of values.
Proof
Let be a fixed point of the interval. For every , : the number is an upper bound of the values, and is not less than the supremum, which is the least one. Since is arbitrary, is a lower bound of the values, and does not exceed the infimum, which is the greatest one.
Conversely, and ; subtracting, each of the two differences is bounded by .
The lesson on sequences, functions and uniform continuity announced that the Heine-Cantor theorem would be the basis of the following result: on a closed and bounded interval, a single response controls the variation of the function on all the subintervals at once.
Let be a function continuous on , with . Then is integrable on , and . If the subintervals are shorter than the response of uniform continuity, the function varies little on each of them, and the strips are low over the whole interval. The figure shows the strips of a function that rises and falls, for partitions with more and more subintervals.
Proof
By the boundedness theorem, from the lesson on the Weierstrass theorem, is bounded on . Let . By the Heine-Cantor theorem, from the lesson on sequences, functions and uniform continuity, is uniformly continuous on ; with the challenge , the definition of a uniformly continuous function, from the same lesson, gives a that depends only on .
By the lemma on the Archimedean property, from the lesson on the decimal expansion of rational numbers, there is a natural number . If , each subinterval of has length , and two points of the same subinterval are less than apart. The lemma on the oscillation on a subinterval, with and , bounds the difference between the supremum and the infimum on each subinterval.
Multiplying by and adding, with the theorem on the properties of summation, from the lesson on summations, progressions and induction, and with the sum of the lengths, , the difference between the sums is less than from on. Since it is not negative, it tends to , and the corollary of the criterion with sequences of partitions completes the proof.
The proof does not need to attain its maximum and its minimum on each subinterval, as the Weierstrass theorem would guarantee; the lemma on the oscillation suffices. No hypothesis is superfluous in the following sense: the integral is only defined for functions bounded on a closed and bounded interval, and a function continuous on an interval that is not closed may be unbounded, like on , or be bounded without extending continuously to the endpoint, like on , which has no limit at . For the latter, neither of the two theorems applies, and the following section settles it.
Loading the content…
Completed
Sign in to save your progress