Riemann, Improper and Lebesgue Integration
CSIR NET real analysis: Riemann criterion, Lebesgue's criterion, FTC, Riemann-sum limits, improper integral tests, Lebesgue measure, MCT, Fatou and DCT.
- 20 min
On this page · 14 sections
Prerequisites
- Continuity, Differentiability and Bounded Variation: continuous, monotone and BV functions are the integrable classes, and the MVT proves the fundamental theorem.
- Real Numbers, Sets and Countability: countable sets and the Cantor set reappear as sets of measure zero.
- Sequences and Series of Functions: uniform convergence and integrals, which the Lebesgue convergence theorems generalise.
The Hook: A Function the Riemann Integral Cannot See
The Riemann Integral
Let \( f : [a, b] \to \mathbb{R} \) be bounded. For a partition \( P : a = x_0 < \dots < x_n = b \), with \( M_i = \sup_{[x_{i-1}, x_i]} f \), \( m_i = \inf_{[x_{i-1}, x_i]} f \) and \( \Delta x_i = x_i - x_{i-1} \),
$$U(f, P) = \sum M_i \Delta x_i, \qquad L(f, P) = \sum m_i \Delta x_i.$$\( f \) is Riemann integrable if \( \inf_P U(f, P) = \sup_P L(f, P) \); the common value is \( \int_a^b f \). Unbounded functions are never Riemann integrable on \( [a, b] \) (they are handled as improper integrals).
Riemann’s criterion. \( f \) is integrable iff for every \( \varepsilon > 0 \) there is a partition with \( U(f, P) - L(f, P) < \varepsilon \). Equivalently, the Riemann sums \( \sum f(t_i)\Delta x_i \) converge as the mesh \( \max \Delta x_i \to 0 \), for any choice of tags \( t_i \).
Which functions are integrable
Immediate consequences: continuous, monotone and BV functions are integrable; so is any bounded function with countably many discontinuities. Thomae’s function is integrable (discontinuous exactly on \( \mathbb{Q} \)), with integral 0. The indicator of the Cantor set is integrable (discontinuous exactly on the Cantor set, which has measure zero), with integral 0. \( 1_{\mathbb{Q}} \) is not (discontinuous everywhere).
Algebra of integrable functions
| Operation | Integrable? |
|---|---|
| \( f + g \), \( cf \), \( fg \), \( \max(f, g) \) | yes |
| \( \lvert f \rvert \) | yes, and \( \lvert \int f \rvert \le \int \lvert f \rvert \) |
| \( \frac{1}{f} \) | yes if \( \lvert f \rvert \ge c > 0 \) |
| \( \varphi \circ f \), \( \varphi \) continuous | yes |
| \( f \circ g \), both integrable | not always (below) |
| \( f \) with \( \lvert f \rvert \) integrable | not always: \( f = 1 \) on \( \mathbb{Q} \), \( -1 \) off it |
| \( f \) changed at finitely many points | yes, same integral |
| \( f \) changed at countably many points | not always: \( 0 \to 1_{\mathbb{Q}} \) |
Composition fails. Let \( g \) be Thomae’s function and \( \varphi(y) = 1 \) for \( y \ne 0 \), \( \varphi(0) = 0 \). Both are integrable on \( [0, 1] \), but \( \varphi \circ g = 1_{\mathbb{Q} \cap [0,1]} \) (taking \( g(0) = 1 \)), which is not.
The fundamental theorem of calculus
- If \( f \) is integrable on \( [a, b] \), then \( F(x) = \int_a^x f \) is Lipschitz (so continuous), and \( F'(c) = f(c) \) at every point \( c \) where \( f \) is continuous.
- If \( F \) is differentiable on \( [a, b] \) and \( F' \) is Riemann integrable, then \( \int_a^b F' = F(b) - F(a) \).
Part 2 needs \( F' \) integrable: \( F(x) = x^2\sin\frac{1}{x^2} \) (\( F(0) = 0 \)) is differentiable on \( [0, 1] \) with an unbounded derivative, so \( F' \) is not Riemann integrable.
Leibniz rule. For continuous \( f \) and differentiable \( u, v \):
$$\frac{d}{dx}\int_{u(x)}^{v(x)} f(t)\,dt = f(v(x))\,v'(x) - f(u(x))\,u'(x).$$Mean value theorem for integrals. If \( f \) is continuous and \( g \ge 0 \) integrable on \( [a, b] \), then \( \int_a^b fg = f(c)\int_a^b g \) for some \( c \in [a, b] \).
Limits of sums as integrals
$$\lim_{n\to\infty} \frac{1}{n}\sum_{k=1}^{n} f\!\left(\frac{k}{n}\right) = \int_0^1 f(x)\,dx \quad (f \text{ integrable on } [0, 1]).$$| Limit | Integral | Value |
|---|---|---|
| \( \sum_{k=1}^n \frac{1}{n + k} \) | \( \int_0^1 \frac{dx}{1 + x} \) | \( \log 2 \) |
| \( \sum_{k=1}^n \frac{n}{n^2 + k^2} \) | \( \int_0^1 \frac{dx}{1 + x^2} \) | \( \frac{\pi}{4} \) |
| \( \frac{1}{n^{p+1}}\sum_{k=1}^n k^p \), \( p > -1 \) | \( \int_0^1 x^p\,dx \) | \( \frac{1}{p+1} \) |
| \( \left(\frac{n!}{n^n}\right)^{1/n} \) | \( \exp \int_0^1 \log x\,dx \) | \( \frac{1}{e} \) |
For \( -1 < p < 0 \) the integrand is unbounded at 0, so the last-but-one row is an improper integral; the limit still holds (compare the sum with the integral, using monotonicity).
Improper Integrals
\( \int_a^\infty f = \lim_{R\to\infty}\int_a^R f \) (type I) and, for \( f \) unbounded near \( a \), \( \int_a^b f = \lim_{\varepsilon \to 0^+}\int_{a+\varepsilon}^b f \) (type II). A two-sided integral \( \int_{-\infty}^{\infty} \) converges only if both halves converge separately; the symmetric limit \( \lim_R \int_{-R}^R \) is the Cauchy principal value, a weaker notion: \( \int_{-\infty}^\infty x\,dx \) diverges, with principal value 0.
Tests
The tests mirror those for series: comparison and limit comparison for non-negative integrands, absolute convergence implies convergence, and the Dirichlet test (if \( \int_a^R g \) is bounded in \( R \) and \( f \) decreases to 0, then \( \int_a^\infty fg \) converges).
| Integral | Converges iff |
|---|---|
| \( \int_1^\infty \frac{dx}{x^p} \) | \( p > 1 \) |
| \( \int_0^1 \frac{dx}{x^p} \) | \( p < 1 \) |
| \( \int_2^\infty \frac{dx}{x(\log x)^p} \) | \( p > 1 \) |
| \( \int_0^\infty x^{s-1}e^{-x}\,dx = \Gamma(s) \) | \( s > 0 \) |
| \( \int_0^1 x^{a-1}(1-x)^{b-1}\,dx = B(a, b) \) | \( a > 0 \) and \( b > 0 \) |
| \( \int_0^\infty \frac{x^p}{1 + x^q}\,dx \), \( q > 0 \) | \( -1 < p < q - 1 \) |
| \( \int_0^\infty \frac{\sin x}{x^p}\,dx \) | \( 0 < p < 2 \) (absolutely iff \( 1 < p < 2 \)) |
| \( \int_0^1 \log x\,dx \) | converges, value \( -1 \) |
Standard values: \( \int_0^\infty \frac{\sin x}{x}\,dx = \frac{\pi}{2} \), \( \int_0^\infty e^{-x^2}dx = \frac{\sqrt{\pi}}{2} \), \( \Gamma(\frac{1}{2}) = \sqrt{\pi} \), \( B(a, b) = \frac{\Gamma(a)\Gamma(b)}{\Gamma(a + b)} \).
Lebesgue Measure
The outer measure of \( A \subseteq \mathbb{R} \) is
$$m^*(A) = \inf\left\{\sum_{n} \ell(I_n) : A \subseteq \bigcup_n I_n,\ I_n \text{ open intervals}\right\}.$$It is defined for every set, gives intervals their length, is translation invariant and countably subadditive. A set \( E \) is measurable if \( m^*(A) = m^*(A \cap E) + m^*(A \setminus E) \) for every \( A \) (Carathéodory). On measurable sets, \( m = m^* \) is countably additive.
| Fact | Detail |
|---|---|
| Measurable sets form a \( \sigma \)-algebra | contains all open, closed and Borel sets |
| Null sets | every set with \( m^* = 0 \) is measurable; every subset of a null set is null |
| Countable sets | measure 0 (cover \( q_n \) by an interval of length \( \frac{\varepsilon}{2^n} \)) |
| Cantor set | measure 0, uncountable, closed, nowhere dense |
| Fat Cantor (Smith–Volterra–Cantor) set | closed, nowhere dense, positive measure |
| Regularity | for measurable \( E \) and \( \varepsilon > 0 \): open \( G \supseteq E \) and closed \( F \subseteq E \) with \( m(G \setminus F) < \varepsilon \) |
| Continuity from below | \( E_n \uparrow E \Rightarrow m(E_n) \to m(E) \) |
| Continuity from above | \( E_n \downarrow E \) and \( m(E_1) < \infty \Rightarrow m(E_n) \to m(E) \) |
| Non-measurable sets | exist (Vitali’s construction, using the axiom of choice) |
Continuity from above needs the finiteness: \( E_n = [n, \infty) \) decreases to \( \emptyset \) while \( m(E_n) = \infty \).
An open dense set of small measure. Cover the \( n \)-th rational in \( [0, 1] \) by an open interval of length \( \frac{\varepsilon}{2^n} \). The union is open, dense, and has measure at most \( \varepsilon \). Its complement in \( [0, 1] \) is closed, nowhere dense and has measure at least \( 1 - \varepsilon \). “Dense” and “large” are unrelated.
The Lebesgue Integral
\( f : E \to \mathbb{R} \) is measurable if \( \{x : f(x) > c\} \) is measurable for every \( c \). Continuous functions, monotone functions, and pointwise limits, sups and infs of sequences of measurable functions are measurable.
For a non-negative simple function \( s = \sum c_i 1_{E_i} \), \( \int s = \sum c_i\,m(E_i) \). For measurable \( f \ge 0 \), \( \int f = \sup\{\int s : 0 \le s \le f,\ s \text{ simple}\} \). For general \( f \), \( \int f = \int f^+ - \int f^- \), and \( f \) is Lebesgue integrable when \( \int \lvert f \rvert < \infty \).
| Sequence on \( \mathbb{R} \) or \( [0, 1] \) | What happens |
|---|---|
| \( n \cdot 1_{(0, 1/n)} \) on \( [0, 1] \) | \( \to 0 \) everywhere, \( \int = 1 \) always; no integrable dominator; Fatou strict |
| \( 1_{[n, n+1]} \) on \( \mathbb{R} \) | \( \to 0 \) everywhere, \( \int = 1 \) always (mass escapes to infinity) |
| \( \frac{1}{n}1_{[0, n]} \) on \( \mathbb{R} \) | \( \to 0 \) uniformly, \( \int = 1 \) always (uniform convergence is not enough on infinite measure) |
| \( 1_{[n, \infty)} \) | decreasing to 0, \( \int = \infty \): MCT has no decreasing version without integrability |
Riemann versus Lebesgue
- Every Riemann integrable function on \( [a, b] \) is Lebesgue integrable, with the same integral.
- \( 1_{\mathbb{Q}} \) on \( [0, 1] \) is Lebesgue integrable (integral 0) but not Riemann integrable.
- A conditionally convergent improper Riemann integral is not a Lebesgue integral: \( \frac{\sin x}{x} \) is not Lebesgue integrable on \( (0, \infty) \), since \( \int \frac{\lvert \sin x \rvert}{x} = \infty \). Absolutely convergent improper integrals of functions Riemann integrable on each \( [a, R] \) are Lebesgue integrals.
- Functions equal almost everywhere have the same Lebesgue integral. \( \int \lvert f \rvert = 0 \) iff \( f = 0 \) a.e.
Differentiation and the Cantor function
- If \( f \) is Lebesgue integrable on \( [a, b] \) and \( F(x) = \int_a^x f \), then \( F' = f \) almost everywhere.
- \( F(x) - F(a) = \int_a^x F' \) for all \( x \in [a, b] \) holds exactly when \( F \) is absolutely continuous. Absolutely continuous functions are continuous and BV.
- The Cantor function is continuous, increasing (so BV), constant on each removed interval, so \( F' = 0 \) a.e., yet \( F(0) = 0 \), \( F(1) = 1 \). Here \( \int_0^1 F' = 0 \ne 1 \): it is not absolutely continuous.
Counter-examples Table
| False claim | Counter-example |
|---|---|
| Bounded \( \Rightarrow \) Riemann integrable | \( 1_{\mathbb{Q}} \) on \( [0, 1] \) |
| Riemann integrable \( \Rightarrow \) finitely many discontinuities | Thomae’s function; \( 1_{\text{Cantor set}} \) |
| \( \lvert f \rvert \) integrable \( \Rightarrow f \) integrable | \( 1 \) on \( \mathbb{Q} \), \( -1 \) off \( \mathbb{Q} \) |
| Composition of integrable functions is integrable | \( \varphi \circ \) Thomae with \( \varphi = 1_{\mathbb{R} \setminus \{0\}} \) |
| \( F \) differentiable \( \Rightarrow \int F' = F(b) - F(a) \) (Riemann) | \( x^2\sin\frac{1}{x^2} \): \( F' \) unbounded |
| Continuous, increasing \( \Rightarrow \int F' = F(b) - F(a) \) | Cantor function |
| \( \int_1^\infty f \) converges \( \Rightarrow f \to 0 \) | \( \sin(x^2) \); spikes |
| Improper Riemann integrable \( \Rightarrow \) Lebesgue integrable | \( \frac{\sin x}{x} \) on \( (0, \infty) \) |
| Measure zero \( \Rightarrow \) countable | Cantor set |
| Nowhere dense \( \Rightarrow \) measure zero | fat Cantor set |
| Dense open set has full measure | union of small intervals around the rationals |
| Pointwise convergence \( \Rightarrow \int f_n \to \int f \) | \( n \cdot 1_{(0, 1/n)} \) |
| Uniform convergence on \( \mathbb{R} \) \( \Rightarrow \int f_n \to \int f \) | \( \frac{1}{n}1_{[0, n]} \) |
| Every subset of \( \mathbb{R} \) is measurable | Vitali set |
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Riemann integrable (bounded, null discontinuity set) | Lebesgue integrable (measurable, \( \int \lvert f \rvert < \infty \)) | |
| improper Riemann integral, conditionally convergent | Lebesgue integral (needs absolute convergence) | |
| measure zero | countable; nowhere dense | |
| \( F(x) = \int_a^x f \) is continuous always | differentiable with \( F' = f \) guaranteed at continuity points of \( f \) (Riemann) and a.e. (Lebesgue) | |
| principal value of \( \int_{-\infty}^\infty \) | convergence of \( \int_{-\infty}^\infty \) (both halves separately) |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C, one or more options may be correct.
Level 1: Foundation
- Is Thomae’s function Riemann integrable on \( [0, 1] \)? If so, what is its integral?
- For which \( p \) does \( \int_0^1 x^{-p}\,dx \) converge?
- What is the Lebesgue measure of the Cantor set?
Answers
- Yes; integral 0. Discontinuities are the rationals (measure zero); every lower sum is 0 because irrationals are dense.
- \( p < 1 \).
- 0. The removed intervals have total length \( \frac{1}{3} + \frac{2}{9} + \frac{4}{27} + \dots = 1 \).
Level 2: Part B Style
Practice Q1. \( \lim_{n\to\infty} \sum_{k=1}^{n} \frac{n}{n^2 + k^2} \) equals:
(a) \( \frac{\pi}{4} \) (b) \( \frac{\pi}{2} \) (c) \( \log 2 \) (d) 1
Answer: (a)
\( \frac{n}{n^2 + k^2} = \frac{1}{n}\cdot\frac{1}{1 + (k/n)^2} \), so the sum tends to \( \int_0^1 \frac{dx}{1 + x^2} = \frac{\pi}{4} \).
Practice Q2. \( \int_0^\infty \frac{x^p}{1 + x^3}\,dx \) converges if and only if:
(a) \( p > -1 \) (b) \( p < 2 \) (c) \( -1 < p < 2 \) (d) \( 0 < p < 3 \)
Answer: (c)
Near 0 the integrand is \( \sim x^p \): need \( p > -1 \). Near infinity it is \( \sim x^{p-3} \): need \( p - 3 < -1 \), that is \( p < 2 \).
Practice Q3. Let \( F(x) = \int_x^{x^2} e^{-t^2}\,dt \). Then \( F'(1) \) equals:
(a) 0 (b) \( e^{-1} \) (c) \( 2e^{-1} \) (d) \( 3e^{-1} \)
Answer: (b)
\( F'(x) = e^{-x^4}\cdot 2x - e^{-x^2}\cdot 1 \). At \( x = 1 \): \( 2e^{-1} - e^{-1} = e^{-1} \). Option (a) comes from noticing \( F(1) = 0 \) and wrongly concluding \( F'(1) = 0 \).
Practice Q4. Let \( f_n = n \cdot 1_{(0, 1/n)} \) on \( [0, 1] \). Which statement is correct?
(a) \( f_n \to 0 \) pointwise and \( \int_0^1 f_n \to 0 \) (b) \( f_n \to 0 \) pointwise, \( \int_0^1 f_n = 1 \) for all \( n \), and no integrable \( g \) satisfies \( f_n \le g \) for all \( n \) (c) \( (f_n) \) is increasing, so MCT applies (d) Fatou’s lemma fails for this sequence
Answer: (b)
At \( x = 0 \), \( f_n(0) = 0 \); for \( x > 0 \), \( f_n(x) = 0 \) once \( \frac{1}{n} \le x \). Each integral is \( n \cdot \frac{1}{n} = 1 \). A dominator would satisfy \( g(x) \ge \sup_n f_n(x) \ge \frac{1}{2x} \) for small \( x \) (take the largest \( n \) with \( n < \frac{1}{x} \), so \( n \ge \frac{1}{x} - 1 \ge \frac{1}{2x} \) for \( x \le \frac{1}{2} \)), which is not integrable; DCT would otherwise force \( \int f_n \to 0 \). (c) is false (not monotone). (d) is false: Fatou holds, with \( 0 \le 1 \).
Level 3: Part C Style
Practice Q5. Which of the following functions are Riemann integrable on \( [0, 1] \)?
(a) Thomae’s function (b) the indicator function of the Cantor set (c) the indicator function of \( \mathbb{Q} \cap [0, 1] \) (d) \( f(x) = \sin\frac{1}{x} \) for \( x \ne 0 \), \( f(0) = 0 \)
Answer: (a), (b) and (d)
- (a) Integrable. Bounded, discontinuous exactly on the rationals (measure zero).
- (b) Integrable. The Cantor set \( C \) is closed with empty interior, so \( 1_C \) is discontinuous exactly at points of \( C \), a null set. Integral 0.
- (c) Not integrable. Discontinuous everywhere; \( U = 1 \), \( L = 0 \) for every partition.
- (d) Integrable. Bounded by 1 and discontinuous only at 0.
Practice Q6. Which of the following improper integrals converge?
(a) \( \int_0^\infty \frac{\sin x}{x}\,dx \) (b) \( \int_0^\infty \frac{\lvert \sin x \rvert}{x}\,dx \) (c) \( \int_0^1 \log x\,dx \) (d) \( \int_0^\infty \sin(x^2)\,dx \)
Answer: (a), (c) and (d)
- (a) Converges (to \( \frac{\pi}{2} \)): no problem at 0, Dirichlet at infinity.
- (b) Diverges. \( \int_{k\pi}^{(k+1)\pi}\frac{\lvert \sin x \rvert}{x}dx \ge \frac{2}{(k+1)\pi} \), and \( \sum \frac{1}{k+1} = \infty \).
- (c) Converges to \( [x\log x - x]_0^1 = -1 \), since \( x\log x \to 0 \).
- (d) Converges. With \( u = x^2 \), \( \int_1^\infty \sin(x^2)dx = \int_1^\infty \frac{\sin u}{2\sqrt{u}}du \), convergent by Dirichlet. The integrand does not tend to 0, which is the trap.
Practice Q7. Let \( m \) denote Lebesgue measure on \( \mathbb{R} \). Which statements are true?
(a) Every countable subset of \( \mathbb{R} \) has measure zero. (b) Every subset of \( \mathbb{R} \) of measure zero is countable. (c) There is an open dense subset of \( [0, 1] \) with measure less than \( \frac{1}{2} \). (d) Every Lebesgue integrable function on \( [0, 1] \) is Riemann integrable.
Answer: (a) and (c)
- (a) True. Cover the \( n \)-th point by an interval of length \( \frac{\varepsilon}{2^n} \).
- (b) False. The Cantor set is uncountable with measure zero.
- (c) True. Put an open interval of length \( \frac{1}{4 \cdot 2^n} \) around the \( n \)-th rational of \( [0, 1] \) and intersect the union with \( (0, 1) \): open, dense, measure at most \( \frac{1}{4} \).
- (d) False. \( 1_{\mathbb{Q}} \) is Lebesgue integrable but not Riemann integrable (also, unbounded Lebesgue integrable functions such as \( x^{-1/2} \) are never Riemann integrable).
Quick Revision Table
| Idea | Fact |
|---|---|
| Riemann criterion | \( U - L < \varepsilon \) for some partition |
| Lebesgue’s criterion | bounded + discontinuities of measure zero |
| FTC | \( F(x) = \int_a^x f \) Lipschitz; \( F' = f \) at continuity points |
| Leibniz | \( f(v)v' - f(u)u' \) |
| Riemann sum | \( \frac{1}{n}\sum f(\frac{k}{n}) \to \int_0^1 f \) |
| p-integrals | \( \int_1^\infty x^{-p} \): \( p > 1 \); \( \int_0^1 x^{-p} \): \( p < 1 \) |
| \( \int_0^\infty \frac{\sin x}{x^p} \) | conv \( 0 < p < 2 \); abs \( 1 < p < 2 \) |
| Null sets | countable sets, Cantor set |
| MCT / Fatou / DCT | \( 0 \le f_n \uparrow \); \( \int\liminf \le \liminf\int \); \( \lvert f_n \rvert \le g \in L^1 \) |
| Riemann \( \subset \) Lebesgue | on \( [a, b] \), same integral |
| AC | \( F(x) - F(a) = \int_a^x F' \) for all \( x \); Cantor function fails |
Key Takeaways
What’s Next
- Metric Spaces and Several Variables: \( L^1 \) and sup metrics on \( C[a, b] \), completeness and compactness.
- Sequences and Series of Functions for uniform convergence and integrals.
- Real Analysis: Quick Revision.
- Statistics candidates will meet these integrals again as densities and expectations in Mathematical Sciences Unit 4.