The Riemann integral, defined
A Riemann sum depends on where each rectangle samples the function, and the integral was defined as a limit over all partitions and all samples at once. Can the integral be defined without sample points, from suprema and infima alone?
Where you are 6 in Integration and sequences of functions
Why this exists
The definite integral as a limit of Riemann sums defined \int_a^b f(x)\,dx as the number that every Riemann sum \sum f(x_i^*)\,\Delta x_i approaches as the mesh goes to 0, for every choice of sample points x_i^*. That definition is hard to use in proofs: the limit runs over all partitions and all sample points at once, and to show that it exists one must already know the value.
The sums named for Darboux remove both troubles. On each piece, use the infimum and the supremum of f in place of a sampled value. The least-upper-bound property of Least upper bounds and completeness turns the two families of sums into two numbers, and when they agree their common value is the integral, the same one as before (Result 6).
Definition
Let a < b. A partition of [a, b] is a finite set P = \{x_0, x_1, \dots, x_n\} with a = x_0 < x_1 < \cdots < x_n = b. Its pieces are the intervals [x_{i-1}, x_i], of widths \Delta x_i = x_i - x_{i-1}, and its mesh is \lVert P \rVert = \max_i \Delta x_i. A partition \tilde{P} is a refinement of P if P \subseteq \tilde{P}. Any two partitions P_1, P_2 have the common refinement P_1 \cup P_2.
Let f : [a, b] \to \mathbb{R} be bounded. On each piece set
which exist by the least-upper-bound property, because \{f(x) : x_{i-1} \le x \le x_i\} is nonempty and bounded. The lower and upper Darboux sums of f for P are
The lower integral and the upper integral of f are
Result 1 shows that both sets are bounded, so both numbers exist for every bounded f.
A bounded function f is Riemann integrable on [a, b] if \underline{\int_a^b} f = \overline{\int_a^b} f. The common value is the Riemann integral \int_a^b f(x)\,dx, also written \int_a^b f, and the set of such functions is \mathcal{R}[a, b]. An integrable function is bounded by definition. For f \in \mathcal{R}[a, b] we set \int_b^a f = -\int_a^b f, and \int_c^c f = 0 for every c.
Intuition
The figure shows the partition of Example 1. Each shaded rectangle reaches up to the lowest point of the graph over its piece, each outlined rectangle to the highest.
The difference U(P, f) - L(P, f) = \sum (M_i - m_i)\,\Delta x_i is the total area of the unshaded strips. Adding the point 2 splits the middle piece, the new shaded rectangle on \displaystyle [2, \frac{5}{2}] is taller, and the strips shrink.
A function is integrable exactly when the strips can be made as thin in total as we like (Result 4). Sample points are gone: L(P, f) and U(P, f) are fixed by P alone, so one sequence of partitions can prove integrability (Result 5), while one sequence of Riemann sums cannot.
Results
Throughout, f : [a, b] \to \mathbb{R} is bounded and P, P_1, P_2 are partitions of [a, b].
1. Bounds. If m \le f(x) \le M for every x \in [a, b], then m(b - a) \le L(P, f) \le U(P, f) \le M(b - a).
2. Refinement. If \tilde{P} is a refinement of P, then L(P, f) \le L(\tilde{P}, f) and U(\tilde{P}, f) \le U(P, f).
3. Lower below upper. L(P_1, f) \le U(P_2, f) for any two partitions. Hence, with m and M as in Result 1,
4. Riemann’s criterion. f \in \mathcal{R}[a, b] if and only if for every \varepsilon > 0 there is a partition P with U(P, f) - L(P, f) < \varepsilon. When f \in \mathcal{R}[a, b], its integral is the only real number I with L(P, f) \le I \le U(P, f) for every partition P.
5. A sequence of partitions. If partitions P_1, P_2, \dots satisfy U(P_n, f) - L(P_n, f) \to 0, then f \in \mathcal{R}[a, b] and
6. Agreement with Riemann sums. If f \in \mathcal{R}[a, b] with \int_a^b f = I, then for every \varepsilon > 0 there is \delta > 0 such that every partition P with \lVert P \rVert < \delta and every choice of x_i^* \in [x_{i-1}, x_i] give
So a function integrable here is integrable in the sense of The definite integral as a limit of Riemann sums, with the same integral, and the rule of computing with right endpoints and equal widths holds for it.
Why it is true
Result 1. Proof. On a piece, m is a lower bound of the values, so m \le m_i; each value f(x) there satisfies m_i \le f(x) \le M_i; and M_i \le M likewise. Multiply m \le m_i \le M_i \le M by \Delta x_i > 0 and add over i; the widths add to b - a. \square
Result 2. Proof. Suppose first that \tilde{P} = P \cup \{y\} with x_{j-1} < y < x_j. Only the jth term changes. Let m' and m'' be the infima of f on [x_{j-1}, y] and [y, x_j]. Each is the infimum of a subset of the values on [x_{j-1}, x_j], so m' \ge m_j and m'' \ge m_j, and
So L(\tilde{P}, f) \ge L(P, f). A refinement with k new points is reached by adding them one at a time, and induction on k gives the claim. The upper sums are Exercise 11. \square
Result 3. Proof. P_1 \cup P_2 refines both partitions, so by Result 2 and m_i \le M_i,
Every lower sum is at most every upper sum, so Result 4 of Least upper bounds and completeness gives \sup_P L(P, f) \le \inf_P U(P, f). The outer inequalities follow from Result 1. \square
Result 4. Proof. Write \underline{I} and \overline{I} for the lower and upper integrals. By definition L(P, f) \le \underline{I} and \overline{I} \le U(P, f), and Result 3 gives \underline{I} \le \overline{I}, so
If partitions with U - L < \varepsilon exist for every \varepsilon > 0, then 0 \le \overline{I} - \underline{I} < \varepsilon for every \varepsilon > 0, so \overline{I} = \underline{I}.
Conversely, let f \in \mathcal{R}[a, b] with integral I, and let \varepsilon > 0. The approximation property (Least upper bounds and completeness, Result 2) gives partitions P_1 with \displaystyle L(P_1, f) > I - \frac{\varepsilon}{2} and P_2 with \displaystyle U(P_2, f) < I + \frac{\varepsilon}{2}. For P = P_1 \cup P_2, Result 2 gives U(P, f) - L(P, f) \le U(P_2, f) - L(P_1, f) < \varepsilon.
Finally, I lies between every L(P, f) and U(P, f). If J does too, then \lvert I - J \rvert \le U(P, f) - L(P, f) for every P, which is below every \varepsilon > 0; so J = I. \square
Result 5. Proof. Given \varepsilon > 0, some P_n has U(P_n, f) - L(P_n, f) < \varepsilon, so f \in \mathcal{R}[a, b] by Result 4. With I = \int_a^b f, Result 4 also gives 0 \le I - L(P_n, f) \le U(P_n, f) - L(P_n, f), and the right side tends to 0, so L(P_n, f) \to I by the squeeze theorem of Convergence of sequences. In the same way 0 \le U(P_n, f) - I \le U(P_n, f) - L(P_n, f) gives U(P_n, f) \to I. \square
Result 6. Fix B > 0 with \lvert f(x) \rvert \le B on [a, b].
Lemma. If P' is P with k points added, then U(P, f) - U(P', f) \le 2Bk\lVert P \rVert and L(P', f) - L(P, f) \le 2Bk\lVert P \rVert.
Proof. Add one point y inside piece j. As in the proof of Result 2, the term M_j\,\Delta x_j is replaced by M'(y - x_{j-1}) + M''(x_j - y), where M', M'' \ge -B. So U drops by at most M_j\,\Delta x_j + B\,\Delta x_j \le 2B\,\Delta x_j \le 2B\lVert P \rVert. Adding a point never increases the mesh, so after k points the drop is at most 2Bk\lVert P \rVert. Lower sums are the same. \square
Proof of Result 6. Every Riemann sum S = \sum f(x_i^*)\,\Delta x_i for P satisfies L(P, f) \le S \le U(P, f), because m_i \le f(x_i^*) \le M_i.
Let f \in \mathcal{R}[a, b] and \varepsilon > 0. The proof of Result 4 gives a partition P_0 (the P_1 \cup P_2 there) with \displaystyle L(P_0, f) > I - \frac{\varepsilon}{2} and \displaystyle U(P_0, f) < I + \frac{\varepsilon}{2}; let k be its number of points, and \displaystyle \delta = \frac{\varepsilon}{4Bk}. If \lVert P \rVert < \delta, then P \cup P_0 is P with at most k points added, so the lemma and Result 2 give
and likewise L(P, f) > I - \varepsilon. Every Riemann sum for P lies between, so within \varepsilon of I. \square
Worked examples
Example 1: Darboux sums, then a refinement
Let f(x) = 4x - x^2 on [0, 3] and \displaystyle P = \{0, 1, \frac{5}{2}, 3\}. Compute L(P, f) and U(P, f), then both sums for P' = P \cup \{2\}.
SolutionCompleting the square, f(x) = 4 - (x - 2)^2, so f increases on (-\infty, 2] and decreases on [2, \infty). On a piece where f is monotone, the infimum and supremum are the end values; on a piece containing 2 the supremum is f(2) = 4 and the infimum is the smaller end value. With f(0) = 0, f(1) = 3, \displaystyle f(\frac{5}{2}) = \frac{15}{4} and f(3) = 3:
| piece | \Delta x_i | m_i | M_i |
|---|---|---|---|
| [0, 1] | 1 | 0 | 3 |
| \displaystyle [1, \frac{5}{2}] | \displaystyle \frac{3}{2} | 3 | 4 |
| \displaystyle [\frac{5}{2}, 3] | \displaystyle \frac{1}{2} | 3 | \displaystyle \frac{15}{4} |
So
In P' the middle piece splits into [1, 2], with m = 3 and M = 4, and \displaystyle [2, \frac{5}{2}], with \displaystyle m = \frac{15}{4} and M = 4:
The lower sum rose by \displaystyle \frac{3}{8}; the upper sum did not move, since the supremum 4 is attained at 2, in both new pieces. The integral, 9 by The second fundamental theorem of calculus, lies between \displaystyle \frac{51}{8} and \displaystyle \frac{87}{8}.
Example 2: integrability from equal partitions
Prove that f(x) = 9 - x^2 is integrable on [0, 3] and find \int_0^3 f. How large must n be for the partition into n equal pieces to have \displaystyle U - L < \frac{1}{10}?
SolutionLet P_n have \displaystyle x_i = \frac{3i}{n} and \displaystyle \Delta x_i = \frac{3}{n}. For 0 \le s < t, f(s) - f(t) = t^2 - s^2 > 0, so f is decreasing on [0, 3], and on each piece M_i = f(x_{i-1}) and m_i = f(x_i), both attained. Using \displaystyle \sum_{j=0}^{n-1} j^2 = \frac{(n-1)n(2n-1)}{6},
and in the same way, with \sum_{i=1}^{n} i^2,
For n = 3 these are 22 and 13. The difference telescopes:
which tends to 0. By Result 5, f \in \mathcal{R}[0, 3], and since \displaystyle \frac{(n \pm 1)(2n \pm 1)}{n^2} \to 2,
Finally \displaystyle \frac{27}{n} < \frac{1}{10} exactly when n > 270, so the least such n is 271.
Example 3: draining a tank at the winery
At a winery, Ben drains a fermentation tank through a valve, and the outflow rate r(t), in liters per minute, falls as the tank empties. Grace reads the meter at these times:
| t (minutes) | 0 | 2 | 5 | 9 | 15 |
|---|---|---|---|---|---|
| r(t) (liters per minute) | 12 | 9 | 6 | 4 | 2.5 |
The volume drained in the first 15 minutes is \int_0^{15} r(t)\,dt. Using only that r is decreasing, find guaranteed bounds for it. With equally spaced readings, how many pieces guarantee bounds less than 5 liters apart?
SolutionSince r is decreasing, on each piece M_i is the reading at the left end and m_i the reading at the right end, whatever r does between readings. A decreasing function is integrable (Which functions are integrable), so Result 4 places the integral between L(P, r) and U(P, r) for P = \{0, 2, 5, 9, 15\}, with widths 2, 3, 4, 6:
So the tank lost between 67 and 99 liters. For n equal pieces of width \displaystyle \frac{15}{n} the gap telescopes as in Example 2:
which is less than 5 exactly when n > 28.5. So 29 equal pieces, that is 30 readings, suffice.
Example 4: a jump and a stray value
Let f(x) = 2 for 0 \le x < 1, f(1) = 5, and f(x) = -1 for 1 < x \le 3. Prove that f \in \mathcal{R}[0, 3] and \int_0^3 f = 0.
SolutionProof. For 0 < \varepsilon < 1 let P_\varepsilon = \{0, 1 - \varepsilon, 1 + \varepsilon, 3\}. On [0, 1 - \varepsilon], f = 2; on [1 + \varepsilon, 3], f = -1; the middle piece takes the values 2, 5 and -1, so there m_2 = -1 and M_2 = 5. Hence
Given \eta > 0, the choice \displaystyle \varepsilon = \min\{\frac{1}{2}, \frac{\eta}{24}\} gives U - L = 12\varepsilon < \eta, so f \in \mathcal{R}[0, 3] by Result 4. By Result 4 again, -3\varepsilon \le \int_0^3 f \le 9\varepsilon for every \varepsilon \in (0, 1), so \int_0^3 f = 0. \square
The value 5 enters only through a piece of width 2\varepsilon, which can be made as thin as we like.
Example 5: bounded but not integrable
Let g(x) = 9 - x^2 for rational x and g(x) = 0 for irrational x, on [0, 3]. Find both integrals of g and decide whether g \in \mathcal{R}[0, 3].
Solutiong is bounded, with 0 \le g \le 9. Every piece of a partition contains an irrational number (density of the irrationals, The Archimedean property and density of the rationals), where g = 0, and g \ge 0; so every m_i = 0, every L(P, g) = 0, and the lower integral is 0.
For the upper sums compare g with f(x) = 9 - x^2 of Example 2. On a piece [x_{i-1}, x_i], g \le f \le f(x_{i-1}), because f is decreasing. Given \varepsilon > 0, let \displaystyle \eta = \min\{\Delta x_i, \frac{\varepsilon}{6}\} and pick a rational q with x_{i-1} < q < x_{i-1} + \eta (density of \mathbb{Q}). Then q lies in the piece, q + x_{i-1} \le 6, and
By the approximation property M_i(g) = f(x_{i-1}) = M_i(f). So U(P, g) = U(P, f) for every P, and by Example 2
The lower and upper integrals are 0 and 18, so g \notin \mathcal{R}[0, 3].
Boundaries
- Boundedness is built in. An unbounded f is unbounded on some piece of every partition, where m_i or M_i does not exist. So \displaystyle f(x) = \frac{1}{\sqrt{x}}, with f(0) = 0, is not in \mathcal{R}[0, 1]; such integrals are limits of integrals, as in Improper integrals with an infinite integrand.
- Bounded is not enough. The functions of Example 5 and Exercise 9 are bounded, yet their lower integral is below their upper integral. The Lebesgue integral of a later analysis course integrates them.
- Refining need not help. Result 2 gives \le, not <: in Example 1 the upper sum stayed at \displaystyle \frac{87}{8}.
- Absolute values. \lvert f \rvert \in \mathcal{R}[a, b] does not give f \in \mathcal{R}[a, b] (Exercise 10).
Where this goes next
Which functions are integrable proves from Result 4 that continuous and monotone functions are integrable, covering Examples 2 and 3, and that finitely many jumps do no harm. The fundamental theorem of calculus, proved combines Darboux sums with the mean value theorem to prove \int_a^b F' = F(b) - F(a). Pointwise and uniform convergence asks when \int_a^b f_n \to \int_a^b f.
Exercises
Darboux sums for a given partition
- Compute L(P, f) and U(P, f) for f(x) = x^3 - 3x on [-2, 2] with P = \{-2, -1, 0, 2\}.
- Compute L(P, f) and U(P, f) for f(x) = \lvert x - 1 \rvert on [0, 3] with \displaystyle P = \{0, \frac{1}{2}, 2, 3\}. Then add the point 1 and compute both sums again.
- Compute L(P, f) and U(P, f) for f(x) = \sin x on [0, \pi] with \displaystyle P = \{0, \frac{\pi}{6}, \frac{\pi}{2}, \pi\}.
Integrability from equal partitions
- For f(x) = 2x + 3 on [1, 4] and the partition P_n into n equal pieces, find L(P_n, f) and U(P_n, f). Conclude that f \in \mathcal{R}[1, 4] and find the integral.
- Do the same for f(x) = e^x on [0, 1], using the geometric sum.
- For \displaystyle f(x) = \frac{1}{x} on [1, 4], show that \displaystyle U(P_n, f) - L(P_n, f) = \frac{9}{4n}, and find the least n with \displaystyle U(P_n, f) - L(P_n, f) < \frac{1}{1000}.
Jumps and readings
- Let f = 1 on [0, 1), f = 4 on [1, 3] and f = 2 on (3, 4]. For 0 < \varepsilon < 1 compute both sums for the partition \{0, 1 - \varepsilon, 1 + \varepsilon, 3 - \varepsilon, 3 + \varepsilon, 4\}, and conclude that \int_0^4 f = 11.
- At the winery, juice from the press fills a second tank at a rate that rises over time. Grace’s readings are 1, 2.5, 3 and 5 liters per minute at t = 0, 4, 6 and 10 minutes. Give guaranteed bounds for the volume added in the first 10 minutes, and the least number of equal pieces that guarantees bounds less than 1 liter apart.
Functions that are not integrable
- Let f(x) = x for rational x and f(x) = x + 1 for irrational x, on [0, 2]. Find both integrals of f.
- Refute with a counterexample: “If \lvert f \rvert \in \mathcal{R}[a, b], then f \in \mathcal{R}[a, b].”
Proofs
- Prove the upper-sum half of Result 2: if \tilde{P} refines P, then U(\tilde{P}, f) \le U(P, f).
- Prove that if f \in \mathcal{R}[a, b] and c is real, then cf \in \mathcal{R}[a, b] and \int_a^b cf = c\int_a^b f.
- Prove that if f, g \in \mathcal{R}[a, b] and f(x) \le g(x) for every x \in [a, b], then \int_a^b f \le \int_a^b g.
- Prove that if f \in \mathcal{R}[a, b], there are partitions P_1 \subseteq P_2 \subseteq \cdots with U(P_n, f) - L(P_n, f) \to 0.
- Suppose \lvert f(x) - f(y) \rvert \le K\lvert x - y \rvert for all x, y \in [a, b]. Prove that U(P, f) - L(P, f) \le K(b - a)\lVert P \rVert for every partition P, and conclude that f \in \mathcal{R}[a, b].
Answers
- L = -6 and U = 8. f'(x) = 3x^2 - 3 vanishes at \pm 1, and f(-2) = -2, f(-1) = 2, f(0) = 0, f(1) = -2, f(2) = 2. The pieces have (m_i, M_i) = (-2, 2), (0, 2), (-2, 2) and widths 1, 1, 2.
- \displaystyle L = \frac{5}{4} and U = 4. On \displaystyle [0, \frac{1}{2}], \displaystyle m = \frac{1}{2} and M = 1; on \displaystyle [\frac{1}{2}, 2], m = f(1) = 0 and M = f(2) = 1; on [2, 3], m = 1 and M = 2. So \displaystyle L = \frac{1}{4} + 0 + 1 and \displaystyle U = \frac{1}{2} + \frac{3}{2} + 2. With 1 added, \displaystyle [\frac{1}{2}, 1] has m = 0, \displaystyle M = \frac{1}{2} and [1, 2] has m = 0, M = 1: \displaystyle L = \frac{5}{4} again and \displaystyle U = \frac{1}{2} + \frac{1}{4} + 1 + 2 = \frac{15}{4}.
- \displaystyle L = \frac{\pi}{6} and \displaystyle U = \frac{11\pi}{12}. \sin increases on \displaystyle [0, \frac{\pi}{2}] and decreases on \displaystyle [\frac{\pi}{2}, \pi], so the pieces, of widths \displaystyle \frac{\pi}{6}, \displaystyle \frac{\pi}{3}, \displaystyle \frac{\pi}{2}, have \displaystyle (m_i, M_i) = (0, \frac{1}{2}), \displaystyle (\frac{1}{2}, 1), (0, 1). So \displaystyle L = \frac{1}{2} \cdot \frac{\pi}{3} and \displaystyle U = \frac{\pi}{12} + \frac{\pi}{3} + \frac{\pi}{2}.
- f is increasing, \displaystyle \Delta x_i = \frac{3}{n} and \displaystyle f(x_i) = 5 + \frac{6i}{n}. So \displaystyle U(P_n, f) = \frac{3}{n}\sum_{i=1}^{n}\left(5 + \frac{6i}{n}\right) \displaystyle = 24 + \frac{9}{n} and \displaystyle L(P_n, f) = \frac{3}{n}\sum_{i=0}^{n-1}\left(5 + \frac{6i}{n}\right) \displaystyle = 24 - \frac{9}{n}. The gap \displaystyle \frac{18}{n} tends to 0, so by Result 5 the integral is 24.
- e^x is increasing. With \displaystyle h = \frac{1}{n}, the geometric sum gives L(P_n, f) = h\sum_{i=0}^{n-1} e^{ih} \displaystyle = \frac{h(e - 1)}{e^h - 1} and U(P_n, f) = e^h L(P_n, f), so \displaystyle U - L = \frac{e - 1}{n} \to 0. As h \to 0, \displaystyle \frac{e^h - 1}{h} \to 1 (the derivative of e^x at 0), so both sums tend to e - 1, the integral.
- f is decreasing, so the gap telescopes as in Example 2: \displaystyle U - L = \frac{3}{n}\big(f(1) - f(4)\big) = \frac{3}{n} \cdot \frac{3}{4} = \frac{9}{4n}. Then \displaystyle \frac{9}{4n} < \frac{1}{1000} exactly when n > 2250, so the least is n = 2251.
- L = 11 - 5\varepsilon and U = 11 + 5\varepsilon. The pieces have widths 1 - \varepsilon, 2\varepsilon, 2 - 2\varepsilon, 2\varepsilon, 1 - \varepsilon and (m_i, M_i) equal to (1, 1), (1, 4), (4, 4), (2, 4), (2, 2). Given \eta > 0, \displaystyle \varepsilon = \min\{\frac{1}{2}, \frac{\eta}{20}\} makes U - L = 10\varepsilon < \eta, so f is integrable, and 11 - 5\varepsilon \le \int_0^4 f \le 11 + 5\varepsilon for every \varepsilon \in (0, 1) forces \int_0^4 f = 11.
- Between 21 and 36 liters. The rate is increasing, so m_i is the left reading and M_i the right one: L = 1 \cdot 4 + 2.5 \cdot 2 + 3 \cdot 4 = 21 and U = 2.5 \cdot 4 + 3 \cdot 2 + 5 \cdot 4 = 36. With n equal pieces, \displaystyle U - L = \frac{10}{n}(5 - 1) = \frac{40}{n}, which is below 1 exactly when n > 40: 41 pieces.
- Lower integral 2, upper integral 4. Every value on a piece is at least x_{i-1}, and rationals near x_{i-1} give values near it, so m_i = x_{i-1}; likewise irrationals near x_i give M_i = x_i + 1. With h(x) = x, L(P, f) = L(P, h) and U(P, f) = U(P, h) + 2. Equal partitions give \displaystyle L(P_n, h) = 2 - \frac{2}{n} and \displaystyle U(P_n, h) = 2 + \frac{2}{n}, so \int_0^2 h = 2 (Result 5), and the integrals of f are 2 and 4.
- On [0, 1] let f(x) = 1 for rational x and f(x) = -1 for irrational x. Then \lvert f \rvert = 1 is constant, with integral 1. But every piece contains numbers of both kinds, so m_i = -1 and M_i = 1, every L(P, f) = -1 and every U(P, f) = 1. The lower and upper integrals are -1 and 1, and f \notin \mathcal{R}[0, 1].
- Proof. Let \tilde{P} = P \cup \{y\} with x_{j-1} < y < x_j, and let M' and M'' be the suprema of f on [x_{j-1}, y] and [y, x_j]. Each is the supremum of a subset of the values on [x_{j-1}, x_j], so M', M'' \le M_j (Least upper bounds and completeness, Result 4), and M'(y - x_{j-1}) + M''(x_j - y) \le M_j\,\Delta x_j. The other terms do not change, so U(\tilde{P}, f) \le U(P, f). A general refinement adds finitely many points one at a time, and induction on their number finishes the proof. \square
- Proof. For c = 0 every sum is 0. For c < 0 and a nonempty bounded set E, \sup(cE) = c\inf E and \inf(cE) = c\sup E (Least upper bounds and completeness, Result 5). So M_i(cf) = c\,m_i(f) and m_i(cf) = c\,M_i(f) on each piece, hence U(P, cf) = c\,L(P, f) and L(P, cf) = c\,U(P, f). The same rule, applied to the sets of all lower and all upper sums of f, makes the upper integral of cf equal to c\sup_P L(P, f) = c\int_a^b f and its lower integral c\inf_P U(P, f) = c\int_a^b f. For c > 0 the same steps run with \sup(cE) = c\sup E, and nothing trades places. In each case the two integrals agree, so cf \in \mathcal{R}[a, b] with integral c\int_a^b f. \square
- Proof. Fix a partition P. On each piece m_i(f) \le f(x) \le g(x) for every x, so m_i(f) is a lower bound of the values of g there, and m_i(f) \le m_i(g). Hence L(P, f) \le L(P, g) \le \int_a^b g, the last step by Result 4. So \int_a^b g is an upper bound of all lower sums of f, and \int_a^b f = \underline{\int_a^b} f \le \int_a^b g. \square
- Proof. By Result 4, for each n \ge 1 there is a partition Q_n with \displaystyle U(Q_n, f) - L(Q_n, f) < \frac{1}{n}. Let P_n = Q_1 \cup \cdots \cup Q_n. Then P_n \subseteq P_{n+1} and P_n refines Q_n, so by Result 2 the gap U(P_n, f) - L(P_n, f) lies between 0 and U(Q_n, f) - L(Q_n, f), which is below \displaystyle \frac{1}{n}. By the squeeze theorem the gap tends to 0. \square
- Proof. f is bounded, by \lvert f(a) \rvert + K(b - a). On a piece, f(x) \le f(y) + K\,\Delta x_i for all x, y in it. Fixing y and taking the supremum over x gives M_i - K\,\Delta x_i \le f(y) for every y in the piece, so M_i - K\,\Delta x_i \le m_i. Then U(P, f) - L(P, f) = \sum (M_i - m_i)\,\Delta x_i, which is at most K\sum (\Delta x_i)^2 \le K\lVert P \rVert \sum \Delta x_i = K(b - a)\lVert P \rVert. For equal partitions this is \displaystyle \frac{K(b - a)^2}{n} \to 0, so f \in \mathcal{R}[a, b] by Result 5. \square