Which functions are integrable
Riemann’s criterion decides integrability one function at a time, by building partitions by hand. Which functions pass it automatically, and how badly can a bounded function behave and still be integrable?
Where you are 6 in Integration and sequences of functions
Why this exists
The Riemann integral, defined defined the class \mathcal{R}[a, b] and proved Riemann’s criterion (Result 4 there): a bounded f is integrable if and only if for every \varepsilon > 0 some partition has U(P, f) - L(P, f) < \varepsilon. There each function was tested by computing its sums. A theorem about the integrals of all continuous functions needs to know, once and for all, that they are integrable.
This lesson proves that whole classes pass the test: continuous functions, monotone functions, and bounded functions with finitely many discontinuities of any kind. It also proves the rules that build integrable functions from others. Uniform continuity is what makes continuous functions work.
Definition
Throughout, a < b, and partitions, m_i, M_i, L(P, f), U(P, f) and the mesh \lVert P \rVert are as in The Riemann integral, defined.
A function f : [a, b] \to \mathbb{R} is increasing if x < y implies f(x) \le f(y), decreasing if x < y implies f(x) \ge f(y), and monotone if it is one or the other.
f has finitely many discontinuities if there is a finite set S \subseteq [a, b] such that f is continuous, in the sense of Continuity, defined and proved, at every point of [a, b] \setminus S. The kind of discontinuity does not matter: a jump, a removable one, or an oscillation.
For f bounded on an interval J, the oscillation of f on J is \operatorname{osc}_J f = \sup_J f - \inf_J f. With J_i = [x_{i-1}, x_i] the pieces of P, write
Intuition
The gap is a sum of strips, each an oscillation times a width. A strip is thin in one of two ways. Its oscillation can be small: on a narrow piece a continuous function cannot move much, and uniform continuity makes one width serve every piece at once (Result 1). Or its width can be small: wrap a bad point in a narrow piece and let f do what it likes there, since with \lvert f \rvert \le B that strip has area at most 2B times its width (Figure 1).
Monotone functions win a third way: their oscillations on the pieces add up to \lvert f(b) - f(a) \rvert, so the gap is at most that number times the mesh, however many jumps f has (Result 2, Figure 2).
Results
Throughout, f and g map [a, b] to \mathbb{R}.
1. Continuous functions. If f is continuous on [a, b], then f \in \mathcal{R}[a, b]. More precisely, if \delta > 0 and \eta > 0 are such that \lvert f(x) - f(y) \rvert < \eta whenever x, y \in [a, b] and \lvert x - y \rvert < \delta, then every partition with \lVert P \rVert < \delta has \operatorname{gap}(P, f) < \eta(b - a).
2. Monotone functions. If f is monotone on [a, b], then f \in \mathcal{R}[a, b], and every partition has \operatorname{gap}(P, f) \le \lVert P \rVert \, \lvert f(b) - f(a) \rvert, with equality when all pieces have the same width.
3. Additivity. Let f be bounded and a < s < b. Then f \in \mathcal{R}[a, b] if and only if f \in \mathcal{R}[a, s] and f \in \mathcal{R}[s, b], and in that case
By induction the same holds for any points a = s_0 < s_1 < \cdots < s_k = b.
4. Sums. If f, g \in \mathcal{R}[a, b], then f + g \in \mathcal{R}[a, b] and \int_a^b (f + g) = \int_a^b f + \int_a^b g. With constant multiples (The Riemann integral, defined, Exercise 12), the integral is linear.
5. Integrable away from an endpoint. Let \lvert f(x) \rvert \le B on [a, b], and let f \in \mathcal{R}[c, b] for every c \in (a, b). Then f \in \mathcal{R}[a, b], and for every c \in (a, b)
so \int_a^b f = \lim_{c \to a^+} \int_c^b f. The same holds at the right end, with f \in \mathcal{R}[a, d] for every d \in (a, b) and error at most B(b - d).
6. Finitely many discontinuities. If f is bounded on [a, b] and has finitely many discontinuities, then f \in \mathcal{R}[a, b].
7. Changing finitely many values. If f \in \mathcal{R}[a, b] and g(x) = f(x) for every x outside a finite set, then g \in \mathcal{R}[a, b] and \int_a^b g = \int_a^b f.
8. Controlled functions. For f bounded on an interval J,
Hence if f \in \mathcal{R}[a, b], K \ge 0, and \lvert g(x) - g(y) \rvert \le K\lvert f(x) - f(y) \rvert for all x, y \in [a, b], then \operatorname{gap}(P, g) \le K\operatorname{gap}(P, f) for every partition P, and g \in \mathcal{R}[a, b].
Why it is true
Riemann’s criterion, refinement (Result 2 of The Riemann integral, defined) and the bounds m(b - a) \le \int_a^b f \le M(b - a) (Result 3 there) are used freely, as is the fact that the integral is the only number between every lower and every upper sum.
Result 1. Proof. Let \lVert P \rVert < \delta. On each piece J_i, f is continuous on a closed bounded interval, so by the extreme value theorem (Extreme and intermediate value theorems, proved, Theorem 2) there are s, t \in J_i with M_i = f(s) and m_i = f(t). As \lvert s - t \rvert \le \Delta x_i < \delta, M_i - m_i < \eta. Multiply by \Delta x_i and add: \operatorname{gap}(P, f) < \eta(b - a).
Now let \varepsilon > 0. By Theorem 4 of Uniform continuity, f is uniformly continuous on [a, b], so some \delta > 0 serves \displaystyle \eta = \frac{\varepsilon}{b - a}. For an integer \displaystyle n > \frac{b - a}{\delta} (The Archimedean property and density of the rationals), the partition into n equal pieces has mesh below \delta and gap below \varepsilon. By Riemann’s criterion, f \in \mathcal{R}[a, b]. \square
Result 2. Proof. Let f be increasing. Then f(a) \le f(x) \le f(b), and on each piece m_i = f(x_{i-1}) and M_i = f(x_i). With d_i = f(x_i) - f(x_{i-1}) \ge 0,
the sum telescoping, with equality when every \Delta x_i = \lVert P \rVert. Equal pieces with n large make the gap below any \varepsilon > 0. If f is decreasing, -f is increasing, and \sup(-f) = -\inf f, \inf(-f) = -\sup f on each piece give \operatorname{gap}(P, -f) = \operatorname{gap}(P, f). \square
Result 3. Proof. If P_1 partitions [a, s] and P_2 partitions [s, b], then P = P_1 \cup P_2 partitions [a, b] with L(P, f) = L(P_1, f) + L(P_2, f), and likewise for U and the gap.
Suppose f \in \mathcal{R}[a, s] and f \in \mathcal{R}[s, b], with integrals I_1 and I_2. Given \varepsilon > 0, pick P_1 and P_2 with gaps below \displaystyle \frac{\varepsilon}{2}; then \operatorname{gap}(P, f) < \varepsilon, so f \in \mathcal{R}[a, b]. For any partition Q of [a, b], split Q \cup \{s\} at s into P_1 and P_2. By refinement,
so \int_a^b f = I_1 + I_2.
Conversely, let f \in \mathcal{R}[a, b], \varepsilon > 0, and \operatorname{gap}(Q, f) < \varepsilon. Splitting Q \cup \{s\} as before, \operatorname{gap}(P_1, f) + \operatorname{gap}(P_2, f) \le \operatorname{gap}(Q, f) < \varepsilon, and both gaps are \ge 0, so each is below \varepsilon. For k points, use induction on k. \square
Result 4. Proof. On a piece, f(x) + g(x) \le M_i(f) + M_i(g) for every x, so M_i(f + g) \le M_i(f) + M_i(g); likewise m_i(f + g) \ge m_i(f) + m_i(g). Hence for every P
Given \varepsilon > 0, take P_1 and P_2 with \operatorname{gap}(P_1, f) and \operatorname{gap}(P_2, g) below \displaystyle \frac{\varepsilon}{2}, and P = P_1 \cup P_2. Refining shrinks gaps, so \operatorname{gap}(P, f + g) < \varepsilon. And \int f + \int g lies between L(P, f + g) and U(P, f + g) for every P, so it equals \int (f + g). \square
Result 5. Proof. If B = 0, then f = 0. Let B > 0 and \varepsilon > 0. Choose c \in (a, b) with \displaystyle c - a < \frac{\varepsilon}{4B}, a partition Q of [c, b] with \displaystyle \operatorname{gap}(Q, f) < \frac{\varepsilon}{2}, and P = \{a\} \cup Q. On the first piece [a, c], M_1 - m_1 \le 2B, so
So f \in \mathcal{R}[a, b]. For any c \in (a, b), Result 3 gives \int_a^b f = \int_a^c f + \int_c^b f, and -B \le f \le B gives \left\lvert \int_a^c f \right\rvert \le B(c - a). The right end is the same with the narrow piece [d, b] last. \square
Result 6. Proof. Order the discontinuities together with a and b: a = s_0 < s_1 < \cdots < s_k = b. Fix j and let t be the midpoint of [s_{j-1}, s_j]. For every c \in (s_{j-1}, t), f is continuous at each point of [c, t], so f \in \mathcal{R}[c, t] by Result 1. Result 5 gives f \in \mathcal{R}[s_{j-1}, t], its right-end form gives f \in \mathcal{R}[t, s_j], and Result 3 joins all the pieces. \square
Result 7. Proof. Let h = g - f, which is 0 outside a finite set S, so bounded. Each x \notin S has a neighborhood free of S, on which h = 0, so h is continuous at x, and h \in \mathcal{R}[a, b] by Result 6. Every piece of every partition contains a point outside S, where h = 0, so L(P, h) \le 0 \le U(P, h) and \int_a^b h = 0. Then g = f + h and Result 4 finish. \square
Result 8. Proof. Let M = \sup_J f, m = \inf_J f and D the supremum on the right. For x, y \in J, \lvert f(x) - f(y) \rvert \le M - m, so D \le M - m. Given \eta > 0, the approximation property of Least upper bounds and completeness gives x with \displaystyle f(x) > M - \frac{\eta}{2} and y with \displaystyle f(y) < m + \frac{\eta}{2}, so D > M - m - \eta. Hence D = M - m.
If \lvert f \rvert \le B, then \lvert g(x) \rvert \le \lvert g(a) \rvert + K\lvert f(x) - f(a) \rvert \le \lvert g(a) \rvert + 2KB, so g is bounded. On a piece J_i, every \lvert g(x) - g(y) \rvert \le K\operatorname{osc}_{J_i} f, so by the identity \operatorname{osc}_{J_i} g \le K\operatorname{osc}_{J_i} f. Multiply by \Delta x_i, add, and apply Riemann’s criterion. \square
Worked examples
Example 1: a jump and a stray value, by the theorems
Let f(x) = x^2 for 0 \le x < 1, f(1) = 7, and f(x) = 4 - x for 1 < x \le 3. Show that f \in \mathcal{R}[0, 3] and find \int_0^3 f.
Solutionf is bounded, by 7, and continuous except at 1, where the left limit is 1 and the right limit 3. By Result 6, f \in \mathcal{R}[0, 3], and by Result 3 the integral splits at 1. On [0, 1], f agrees with x^2 except at 1; on [1, 3], with 4 - x except at 1. Result 7 and The second fundamental theorem of calculus give
The value f(1) = 7 plays no part.
Example 2: two bounds for the same function
Let \displaystyle f(x) = \sqrt{\lvert x \rvert} on [-1, 4], and P_n the partition into n equal pieces of width \displaystyle h = \frac{5}{n}. (a) Use the estimate of Result 1 to find n guaranteeing \displaystyle \operatorname{gap}(P_n, f) < \frac{1}{10}. (b) Do better with monotonicity on each side of 0, for n divisible by 5.
Solution(a) For 0 \le t \le s, \displaystyle (\sqrt{s} - \sqrt{t})^2 \displaystyle \le (\sqrt{s} - \sqrt{t})(\sqrt{s} + \sqrt{t}) = s - t. So \displaystyle \left\lvert \sqrt{\lvert x \rvert} - \sqrt{\lvert y \rvert} \right\rvert \displaystyle \le \sqrt{\lvert x - y \rvert} when x and y have the same sign; if x < 0 < y, the difference is at most \displaystyle \max(\sqrt{\lvert x \rvert}, \sqrt{y}) \displaystyle \le \sqrt{\lvert x - y \rvert}. So on each piece \displaystyle M_i - m_i \le \sqrt{h}, as in the proof of Result 1, and
This is below \displaystyle \frac{1}{10} exactly when \displaystyle \frac{5}{n} < \frac{1}{2500}, that is n > 12{,}500; so n = 12{,}501 is guaranteed to work.
(b) If 5 divides n, then 0 is a partition point. f is decreasing on [-1, 0] and increasing on [0, 4], so on each side the gap telescopes as in Result 2:
This is below \displaystyle \frac{1}{10} exactly when n > 150, so n = 155 is the least multiple of 5 that works. Result 1’s estimate sees only how fast f can change, fastest at 0; the telescoping sees the total change.
Example 3: an oscillating discontinuity
Let \displaystyle f(x) = \sin\frac{1}{x} for 0 < x \le 1 and f(0) = 0. Prove that f \in \mathcal{R}[0, 1], and find c for which \int_c^1 f is guaranteed to be within \displaystyle \frac{1}{100} of \int_0^1 f.
SolutionProof. \lvert f \rvert \le 1, and for each c \in (0, 1), f is continuous on [c, 1], so f \in \mathcal{R}[c, 1] by Result 1. Result 5 with B = 1 gives f \in \mathcal{R}[0, 1] and \left\lvert \int_0^1 f - \int_c^1 f \right\rvert \le c. So \displaystyle c = \frac{1}{100} works. \square
The discontinuity at 0 is no jump: f is 1 at the points \displaystyle \frac{1}{2\pi k + \pi/2} and -1 at \displaystyle \frac{1}{2\pi k + 3\pi/2}, which tend to 0. The integral exists with no elementary antiderivative.
Example 4: infinitely many jumps
Let f(0) = 0 and \displaystyle f(x) = \frac{1}{k} for \displaystyle \frac{1}{k+1} < x \le \frac{1}{k}, k \ge 1 (Figure 2). Prove that f \in \mathcal{R}[0, 1], find the gap of the partition into n equal pieces, and show that \displaystyle \int_0^1 f = \frac{\pi^2}{6} - 1.
Solutionf is increasing: for 0 < x < y \le 1, with x on step k and y on step j, y > x forces j \le k, so f(y) \ge f(x); and f(0) = 0 is below every other value. f jumps at every \displaystyle \frac{1}{k} with k \ge 2, so Result 6 does not apply, but Result 2 does: f \in \mathcal{R}[0, 1], with equal-partition gap \displaystyle \frac{1}{n}\big(f(1) - f(0)\big) = \frac{1}{n}.
On \displaystyle \left[\frac{1}{k+1}, \frac{1}{k}\right], f equals \displaystyle \frac{1}{k} except at the left end, so by Result 7 its integral there is
By Result 3, summing over k = 1, \dots, N and telescoping the last two terms,
By Result 5 with B = 1, \int_0^1 f is within \displaystyle \frac{1}{N+1} of this. Let N \to \infty: \displaystyle \sum \frac{1}{k^2} converges (The integral test and p series) to Euler’s value \displaystyle \frac{\pi^2}{6}, derived in Fourier series, so \displaystyle \int_0^1 f = \frac{\pi^2}{6} - 1 \approx 0.645.
Example 5: reciprocals
Let f \in \mathcal{R}[a, b] with f(x) \ge m > 0 for every x. Prove that \displaystyle \frac{1}{f} \in \mathcal{R}[a, b].
SolutionProof. For x, y \in [a, b],
Result 8 with \displaystyle K = \frac{1}{m^2} gives \displaystyle \frac{1}{f} \in \mathcal{R}[a, b]. \square
Positivity alone is not enough: f(x) = x for 0 < x \le 1 with f(0) = 1 is positive and integrable, but \displaystyle \frac{1}{f} is unbounded.
Example 6: Thomae’s function
Let \displaystyle \tau(x) = \frac{1}{q} if \displaystyle x = \frac{p}{q} in lowest terms with q \ge 1, and \tau(x) = 0 for irrational x: the function of Example 6 of Continuity, defined and proved, there called P, discontinuous at every rational. Prove that \tau \in \mathcal{R}[0, 1] with \int_0^1 \tau = 0, and for \displaystyle \varepsilon = \frac{1}{2} find n for which the equal partition has U(P_n, \tau) < \varepsilon.
SolutionProof. Every piece of every partition contains an irrational, where \tau = 0, and \tau \ge 0; so every L(P, \tau) = 0.
Let \varepsilon > 0, and let F be the set of x \in [0, 1] with \displaystyle \tau(x) \ge \frac{\varepsilon}{2}: fractions \displaystyle \frac{p}{q} with \displaystyle q \le \frac{2}{\varepsilon} and 0 \le p \le q, finitely many, say K. Take \displaystyle n > \frac{4K}{\varepsilon}. A point lies in at most two pieces of P_n, so at most 2K pieces meet F; there M_i \le 1, and they contribute at most \displaystyle \frac{2K}{n} < \frac{\varepsilon}{2}. On every other piece \displaystyle M_i \le \frac{\varepsilon}{2}, and those pieces, of total width at most 1, contribute at most \displaystyle \frac{\varepsilon}{2}. So \operatorname{gap}(P_n, \tau) = U(P_n, \tau) < \varepsilon: \tau is integrable, and its integral lies in [0, \varepsilon) for every \varepsilon > 0, so it is 0. \square
For \displaystyle \varepsilon = \frac{1}{2}, F is the set of fractions with q \le 4: \displaystyle 0, 1, \frac{1}{2}, \frac{1}{3}, \frac{2}{3}, \frac{1}{4}, \frac{3}{4}, so K = 7, and the argument guarantees \displaystyle U(P_n, \tau) < \frac{1}{2} for every n > 56, for instance n = 57.
Boundaries
- Bounded is a hypothesis. Result 5 fails for unbounded f (Exercise 8); such limits are the improper integrals of Improper integrals with an infinite integrand.
- Finitely many is not the edge. Examples 4 and 6 have infinitely many discontinuities, while 1 on the rationals and 0 on the irrationals is integrable on no interval. Lebesgue’s criterion, proved in a later analysis course, settles it: a bounded f is integrable if and only if its discontinuities can be covered by countably many intervals of total length as small as we like.
- Changing values. Result 7 allows finitely many changes; changing 0 to 1 at every rational destroys integrability.
- Composition. If f \in \mathcal{R}[a, b] and \varphi is continuous on an interval containing f([a, b]), then \varphi \circ f is integrable, proved in a later analysis course. If \varphi is only integrable, \varphi \circ f can fail (Exercise 9).
Where this goes next
The fundamental theorem of calculus, proved uses Results 1, 3 and 5 to prove both forms of the fundamental theorem of calculus. Pointwise and uniform convergence adds one more class: a uniform limit of integrable functions is integrable.
Exercises
Bounding the gap
- For \displaystyle f(x) = \sqrt{x} on [0, 9], find \operatorname{gap}(P_n, f) for the partition into n equal pieces, and the least n making it less than \displaystyle \frac{1}{100}.
- For \displaystyle f(x) = \frac{1}{1 + x^2} on [-2, 3] and n divisible by 5, find \operatorname{gap}(P_n, f) for n equal pieces, and the least such n making it less than \displaystyle \frac{1}{100}.
Integrals of discontinuous functions
- In a darkroom, Sarah exposes a print under an enlarger whose lamp warms up: the light on the paper is I(t) = 20 + 10t lux for 0 \le t \le 2 seconds and 40 lux for 2 < t < 8. At t = 8 David slides a card over the paper, and I(t) = 15 lux for 8 \le t \le 12. Explain why the exposure \int_0^{12} I(t)\,dt exists and does not depend on the value at t = 8, and find it in lux-seconds.
- Prove that f(x) = \lfloor x^2 \rfloor is integrable on [0, 2] and find \int_0^2 f.
- Let f(0) = 0 and f(x) = 2^{-k} for 2^{-k-1} < x \le 2^{-k}, k \ge 0. Show f \in \mathcal{R}[0, 1], find \int_0^1 f, and find the least n for which the partition into n equal pieces has gap below \displaystyle \frac{1}{1000}.
- For Thomae’s function \tau of Example 6, compute U(P_4, \tau) and U(P_6, \tau) for the partitions of [0, 1] into 4 and 6 equal pieces.
- Let \displaystyle f(x) = \cos\frac{\pi}{x} for 0 < x \le 2 and f(0) = 5. Find c for which \int_c^2 f is guaranteed to be within \displaystyle \frac{1}{1000} of \int_0^2 f, as large as Results 5 and 7 allow.
When the hypotheses fail
- Refute: “If f \in \mathcal{R}[c, 1] for every c \in (0, 1), then f \in \mathcal{R}[0, 1].” Use \displaystyle f(x) = \frac{1}{x} with f(0) = 0, and find \int_c^1 f.
- Refute: “If f, g \in \mathcal{R}[0, 1] and f([0, 1]) \subseteq [0, 1], then g \circ f \in \mathcal{R}[0, 1].”
Proofs
- Prove that if f \in \mathcal{R}[a, b] and a \le c < d \le b, then f \in \mathcal{R}[c, d].
- Prove that if f \in \mathcal{R}[a, b], then \lvert f \rvert \in \mathcal{R}[a, b] and \left\lvert \int_a^b f \right\rvert \le \int_a^b \lvert f \rvert.
- Prove that if f, g \in \mathcal{R}[a, b], then f^2 \in \mathcal{R}[a, b] and fg \in \mathcal{R}[a, b].
- Let f be continuous on [a, b] with f(x) \ge 0 for every x and \int_a^b f = 0. Prove that f(x) = 0 for every x \in [a, b].
- Suppose a = s_0 < s_1 < \cdots < s_k = b and f is monotone on each [s_{j-1}, s_j]. Prove that f \in \mathcal{R}[a, b].
- Let f be bounded on [0, 1] and continuous at every point except 0 and the points \displaystyle \frac{1}{k}, k \ge 1. Prove that f \in \mathcal{R}[0, 1].
Answers
- \displaystyle \frac{27}{n}, and n = 2701. \displaystyle \sqrt{x} is increasing, so by Result 2 the gap is \displaystyle \frac{9}{n}\big(f(9) - f(0)\big) = \frac{27}{n}, below \displaystyle \frac{1}{100} exactly when n > 2700.
- \displaystyle \frac{17}{2n}, and n = 855. With 5 \mid n, 0 is a partition point; f is increasing on [-2, 0] and decreasing on [0, 3], so the gap telescopes on each side: \displaystyle \frac{5}{n}\big((1 - \frac{1}{5}) + (1 - \frac{1}{10})\big) \displaystyle = \frac{5}{n} \cdot \frac{17}{10}. It is below \displaystyle \frac{1}{100} exactly when n > 850, and the next multiple of 5 is 855.
- 360 lux-seconds. I is bounded by 40; at t = 2 both formulas give 40, so the only discontinuity is at 8, and Result 6 gives existence. Split at 2 and 8 (Result 3): \int_0^2 (20 + 10t)\,dt = 60; on [2, 8], I = 40 except at t = 8, giving 240 by Result 7; on [8, 12], 15 \cdot 4 = 60. Another value at t = 8 changes I at one point only, so by Result 7 the exposure is unchanged.
- \displaystyle 5 - \sqrt{2} - \sqrt{3}. f is increasing, so integrable by Result 2. On [0, 1], \displaystyle [1, \sqrt{2}], \displaystyle [\sqrt{2}, \sqrt{3}], \displaystyle [\sqrt{3}, 2] it equals 0, 1, 2, 3 except at the right end, so by Results 3 and 7 the integral is \displaystyle (\sqrt{2} - 1) + 2(\sqrt{3} - \sqrt{2}) \displaystyle + 3(2 - \sqrt{3}).
- \displaystyle \int_0^1 f = \frac{2}{3} and n = 1001. As in Example 4, f is increasing with f(1) - f(0) = 1, so the equal gap is \displaystyle \frac{1}{n}, below \displaystyle \frac{1}{1000} exactly when n > 1000. On [2^{-k-1}, 2^{-k}] the integral is 2^{-k} \cdot 2^{-k-1} = 2^{-2k-1} (Result 7). Summing k = 0, \dots, N gives \int_{2^{-N-1}}^1 f, within 2^{-N-1} of \int_0^1 f (Result 5), and \displaystyle \sum_{k \ge 0} 2^{-2k-1} = \frac{1/2}{1 - 1/4}.
- \displaystyle U(P_4, \tau) = \frac{3}{4} and \displaystyle U(P_6, \tau) = \frac{11}{18}. \displaystyle M_i = \frac{1}{q} for the least q with a fraction \displaystyle \frac{p}{q} in the piece. For P_4 the M_i are \displaystyle 1, \frac{1}{2}, \frac{1}{2}, 1 (from \displaystyle 0, \frac{1}{2}, \frac{1}{2}, 1), summing to 3. For P_6 they are \displaystyle 1, \frac{1}{3}, \frac{1}{2}, \frac{1}{2}, \frac{1}{3}, 1 (from \displaystyle 0, \frac{1}{3}, \frac{1}{2}, \frac{1}{2}, \frac{2}{3}, 1), summing to \displaystyle \frac{11}{3}. Multiply by the widths \displaystyle \frac{1}{4} and \displaystyle \frac{1}{6}.
- \displaystyle c = \frac{1}{1000}. f is bounded by 5 and continuous on (0, 2], so Result 5 applies, but B = 5 only gives \displaystyle c = \frac{1}{5000}. Let \tilde{f} equal f except \tilde{f}(0) = 0. By Result 7, \tilde{f} has the same integrals over [0, 2] and every [c, 2], and \lvert \tilde{f} \rvert \le 1, so Result 5 with B = 1 bounds the difference by c.
- On [c, 1], f is continuous, so f \in \mathcal{R}[c, 1], with \int_c^1 f = -\ln c (The second fundamental theorem of calculus). But f is unbounded on [0, 1], and integrable functions are bounded by definition, so f \notin \mathcal{R}[0, 1]. Result 5 needs one bound B on all of [a, b].
- Let f = \tau of Example 6, and g(0) = 0, g(y) = 1 for y \ne 0, integrable by Result 6. Then g(\tau(x)) is 1 for rational x and 0 for irrational x. Every piece contains both kinds, so every L = 0 and every U = 1, and g \circ \tau \notin \mathcal{R}[0, 1].
- Proof. If a < c, Result 3 at s = c gives f \in \mathcal{R}[c, b]. If d < b, Result 3 on [c, b] at s = d gives f \in \mathcal{R}[c, d]. When c = a or d = b, skip that step. \square
- Proof. By the reverse triangle inequality (Result 11 of The real numbers as an ordered field), \big\lvert \lvert f(x) \rvert - \lvert f(y) \rvert \big\rvert \le \lvert f(x) - f(y) \rvert, so Result 8 with K = 1 gives \lvert f \rvert \in \mathcal{R}[a, b]. As -\lvert f \rvert \le f \le \lvert f \rvert, monotonicity and constant multiples (The Riemann integral, defined, Exercises 13 and 12) give -\int_a^b \lvert f \rvert \le \int_a^b f \le \int_a^b \lvert f \rvert. \square
- Proof. Let \lvert f \rvert \le B. Then \lvert f(x)^2 - f(y)^2 \rvert = \lvert f(x) + f(y) \rvert \lvert f(x) - f(y) \rvert \le 2B\lvert f(x) - f(y) \rvert, and Result 8 with K = 2B gives f^2 \in \mathcal{R}[a, b]. By Result 4 and constant multiples, f + g and f - g are integrable, so are their squares, and \displaystyle fg = \frac{1}{4}\big((f + g)^2 - (f - g)^2\big) is integrable by Result 4. \square
- Proof. Suppose f(c) > 0. By continuity there is \delta > 0 with \displaystyle f(x) > \frac{f(c)}{2} for x \in [a, b], \lvert x - c \rvert < \delta. Then \displaystyle J = [c - \frac{\delta}{2}, c + \frac{\delta}{2}] \cap [a, b] is an interval of length \ell > 0. Let P consist of a, b and the ends of J. On J, \displaystyle m_i \ge \frac{f(c)}{2}; on the other pieces m_i \ge 0. So \int_a^b f \ge L(P, f) \displaystyle \ge \frac{f(c)}{2}\ell > 0, a contradiction. \square
- Proof. On each [s_{j-1}, s_j], f lies between its end values, so f is bounded, and f \in \mathcal{R}[s_{j-1}, s_j] by Result 2. Result 3 for s_0, \dots, s_k gives f \in \mathcal{R}[a, b]. \square
- Proof. Let 0 < c < 1. In [c, 1] the only possible discontinuities are the \displaystyle \frac{1}{k} with \displaystyle k \le \frac{1}{c}, finitely many, so by Result 6, f \in \mathcal{R}[c, 1]. As f is bounded on [0, 1], Result 5 gives f \in \mathcal{R}[0, 1]. \square