Limits of functions, epsilon and delta
Calculus took limits only where a function is defined on both sides of the point. How do you take a limit at the end of a domain, at a point outside it, or on a domain full of gaps, and how do the theorems about sequences carry over?
Where you are 5 in Continuity
Why this exists
The epsilon delta definition of a limit took \lim_{x \to a} f(x) for a function defined on an open interval around a, except possibly at a. Analysis needs limits at the endpoint of [-2, 2], at \displaystyle \sqrt{2} for a function on \mathbb{Q}, at 0 for a function on the points \displaystyle \frac{1}{n}. The right setting is f : S \to \mathbb{R} on any S \subseteq \mathbb{R}, with the limit taken at a limit point of S (Open, closed and compact sets in R). This lesson gives that definition, proves the sequential criterion, which turns the theorems of Limits of sequences, epsilon and N into theorems about functions, and treats one-sided limits as limits of restrictions.
Definition
Throughout, S \subseteq \mathbb{R} and f : S \to \mathbb{R}.
Limit point. As in Open, closed and compact sets in R, c \in \mathbb{R} is a limit point of S if every neighborhood (c - \varepsilon, c + \varepsilon) contains a point of S other than c. It may or may not belong to S. Lebl’s Basic Analysis calls it a cluster point.
Limit of a function. Let c be a limit point of S and L \in \mathbb{R}. Then f(x) converges to L as x \to c, written \lim_{x \to c} f(x) = L or f(x) \to L as x \to c, if for every \varepsilon > 0 there is \delta > 0 such that \lvert f(x) - L \rvert < \varepsilon for every x \in S with 0 < \lvert x - c \rvert < \delta. If no such L exists, the limit does not exist. Only points of S are tested; whether c \in S, and the value f(c), play no part.
Restriction. For A \subseteq S, the restriction f|_A : A \to \mathbb{R} is f|_A(x) = f(x) for x \in A.
One-sided limits. If c is a limit point of S \cap (c, \infty), the right-hand limit \lim_{x \to c^+} f(x) is the limit of f|_{S \cap (c, \infty)} at c, when it exists. If c is a limit point of S \cap (-\infty, c), the left-hand limit \lim_{x \to c^-} f(x) is the limit of f|_{S \cap (-\infty, c)}.
If S contains (c - r, c) \cup (c, c + r), then with \delta \le r the definition says exactly what The epsilon delta definition of a limit says.
Intuition
The picture is the band and strip of Calculus 1, but the strip is searched only for points of S. At the end of an interval, half the strip is empty (Figure 1).
A sequence \{x_n\} in S \setminus \{c\} with x_n \to c is a route into c through the domain. The limit is L exactly when f(x_n) \to L along every route. One route proves nothing, but two routes with different limits prove there is no limit (Figure 2).
Results
Throughout, f, g, h : S \to \mathbb{R} and c is a limit point of S.
Theorem 1 (uniqueness). If f(x) \to L and f(x) \to L' as x \to c, then L = L'.
Theorem 2 (sequential criterion). \lim_{x \to c} f(x) = L if and only if f(x_n) \to L for every sequence \{x_n\} with x_n \in S \setminus \{c\} for every n and x_n \to c.
Theorem 3 (divergence criterion). The limit at c does not exist if (a) some sequence \{x_n\} in S \setminus \{c\} with x_n \to c has \{f(x_n)\} divergent, or (b) two such sequences have f(x_n) \to L_1 and f(y_n) \to L_2 with L_1 \ne L_2.
Theorem 4 (limit laws). If f(x) \to A and g(x) \to B as x \to c, then f + g \to A + B, f - g \to A - B, fg \to AB and kf \to kA for every constant k. If also B \ne 0 and g(x) \ne 0 for every x \in S \setminus \{c\}, then \displaystyle \frac{f}{g} \to \frac{A}{B}.
Theorem 5 (order and squeeze). (a) If both limits exist and f(x) \le g(x) for every x \in S \setminus \{c\}, then \lim_{x \to c} f(x) \le \lim_{x \to c} g(x). (b) If f(x) \le g(x) \le h(x) for every x \in S \setminus \{c\} and f(x) \to L, h(x) \to L, then g(x) \to L.
Theorem 6 (restrictions). Let A \subseteq S. (a) If c is a limit point of A and f(x) \to L, then f|_A(x) \to L. (b) If for some \alpha > 0 the sets A and S have the same points in (c - \alpha, c) \cup (c, c + \alpha), then c is a limit point of A, and f(x) \to L if and only if f|_A(x) \to L.
Theorem 7 (one-sided limits). If c is a limit point of both S \cap (-\infty, c) and S \cap (c, \infty), then \lim_{x \to c} f(x) = L if and only if \lim_{x \to c^-} f(x) = \lim_{x \to c^+} f(x) = L.
By Theorem 6(b) the limit is local. And two functions equal on S \setminus \{c\} have the same limit at c, since c is never tested.
Why it is true
Theorem 1. Proof. Let \varepsilon > 0, choose \delta_1 and \delta_2 serving \displaystyle \frac{\varepsilon}{2} for L and for L', and let \delta = \min\{\delta_1, \delta_2\}. Since c is a limit point, some x \in S has 0 < \lvert x - c \rvert < \delta, and by the Triangle inequality
A nonnegative number below every positive number is 0, so L = L'. \square
The limit point is needed to find x. If c is not one, some (c - \delta_0, c + \delta_0) holds no point of S but possibly c; with \delta = \delta_0 nothing is tested, and every L would pass.
Theorem 2. Proof. Such sequences exist by Theorem 2 of Open, closed and compact sets in R. Suppose f(x) \to L, let \{x_n\} be such a sequence and \varepsilon > 0. Choose \delta serving \varepsilon, then N with \lvert x_n - c \rvert < \delta for n > N. For n > N, x_n \in S and 0 < \lvert x_n - c \rvert < \delta, so \lvert f(x_n) - L \rvert < \varepsilon.
For the converse, take the contrapositive. If f(x) \not\to L, negating the definition (Negating a quantified statement) gives \varepsilon > 0 such that for every \delta > 0 some x \in S has 0 < \lvert x - c \rvert < \delta and \lvert f(x) - L \rvert \ge \varepsilon. For each n \ge 1 use \displaystyle \delta = \frac{1}{n} and call the point x_n. Then x_n \in S \setminus \{c\} and \displaystyle \lvert x_n - c \rvert < \frac{1}{n}, so x_n \to c, but \lvert f(x_n) - L \rvert \ge \varepsilon for every n, so f(x_n) \not\to L (Theorem 1 of Limits of sequences, epsilon and N). \square
Theorem 3. Proof. If f(x) \to L, Theorem 2 makes f(x_n) \to L along every such sequence, contradicting divergence in (a) and giving L_1 = L = L_2 in (b). \square
Theorem 4. Proof. Let \{x_n\} lie in S \setminus \{c\} with x_n \to c. By Theorem 2, f(x_n) \to A and g(x_n) \to B. The sum law of Convergence of sequences and the product, difference and quotient laws (Theorems 4 and 6 of Limits of sequences, epsilon and N, with every g(x_n) \ne 0 for the quotient) give the limits along \{x_n\}, and Theorem 2 converts them back. \square
Theorem 5. Proof. (a) Along one sequence \{x_n\} as in Theorem 2, f(x_n) \le g(x_n) and the two sequences tend to the two limits; apply the order limit theorem (Theorem 8 of Limits of sequences, epsilon and N). (b) Along every such sequence the squeeze theorem of Convergence of sequences gives g(x_n) \to L; apply Theorem 2. \square
Theorem 6. Proof. (a) A \delta serving \varepsilon for f serves it for f|_A, which is tested at fewer points. (b) For \varepsilon \le \alpha the neighborhood (c - \varepsilon, c + \varepsilon) holds the same points of A as of S apart from c, so it meets A \setminus \{c\} because it meets S \setminus \{c\}; wider ones contain it, so c is a limit point of A. One direction is (a). For the other, let \delta' serve \varepsilon for f|_A and \delta = \min\{\delta', \alpha\}. A point x \in S with 0 < \lvert x - c \rvert < \delta lies in A, so \lvert f(x) - L \rvert < \varepsilon. \square
Theorem 7. Proof. Let S^- = S \cap (-\infty, c) and S^+ = S \cap (c, \infty). If f(x) \to L, apply Theorem 6(a) to A = S^- and A = S^+. Conversely, given \varepsilon > 0, let \delta^- and \delta^+ serve \varepsilon for the two restrictions and \delta = \min\{\delta^-, \delta^+\}. A point x \in S with 0 < \lvert x - c \rvert < \delta is not c, so it lies in S^- or S^+, and \lvert f(x) - L \rvert < \varepsilon. \square
Worked examples
A limit at the end of the domain
Let \displaystyle f(x) = \sqrt{4 - x^2} on S = [-2, 2]. Prove that \lim_{x \to 2} f(x) = 0. For \varepsilon = 0.8, compare the \delta the proof gives with the largest \delta that works.
SolutionOnly points of S are tested, so the right half of every strip around 2 is empty (Figure 1). For x \in [-2, 2], 2 - x \ge 0 and 2 + x \le 4, so
Let \varepsilon > 0 and \displaystyle \delta = \frac{\varepsilon^2}{4}. If x \in S and 0 < \lvert x - 2 \rvert < \delta, then 0 < 2 - x < \delta, and since the square root is increasing,
For \varepsilon = 0.8 the proof gives \delta = 0.16; at the far end, \displaystyle f(1.84) = \sqrt{0.6144} \approx 0.784. Exactly, f(x) < 0.8 when x^2 > 3.36, which near 2 means \displaystyle x > \sqrt{3.36} \approx 1.8330. So the largest \delta is \displaystyle 2 - \sqrt{3.36} \approx 0.1670; a wider strip tests \displaystyle x = \sqrt{3.36}, where f(x) = 0.8. That f(2) = 0 was never used.
A sequence is a limit of a function
Let \{a_n\} be a sequence, \displaystyle S = \left\{\frac{1}{n} : n \ge 1\right\}, and f : S \to \mathbb{R} with \displaystyle f\left(\frac{1}{n}\right) = a_n. Prove that \lim_{x \to 0} f(x) = L if and only if a_n \to L.
SolutionProof. The points \displaystyle \frac{1}{n} \ne 0 tend to 0, so 0 is a limit point of S, and the points of S in (0, \delta) are the \displaystyle \frac{1}{n} with \displaystyle n > \frac{1}{\delta}. Suppose a_n \to L and \varepsilon > 0; choose a positive integer N with \lvert a_n - L \rvert < \varepsilon for n > N, and let \displaystyle \delta = \frac{1}{N}. A tested x is \displaystyle \frac{1}{n} with n > N, so \lvert f(x) - L \rvert = \lvert a_n - L \rvert < \varepsilon. Conversely, if \delta serves \varepsilon for f, take an integer \displaystyle N \ge \frac{1}{\delta}: for n > N, \displaystyle 0 < \frac{1}{n} < \delta, so \lvert a_n - L \rvert < \varepsilon. \square
So the \varepsilon-N definition is the \varepsilon-\delta definition on one particular domain, with \displaystyle \delta = \frac{1}{N}.
Every value along some route
Let \displaystyle f(x) = \cos\frac{\pi}{x} on S = (0, \infty). Show that \lim_{x \to 0} f(x) does not exist. For each L \in [-1, 1] find a sequence \{z_n\} in S with z_n \to 0 and f(z_n) \to L, and show that no L outside [-1, 1] arises this way.
SolutionLet \displaystyle x_n = \frac{1}{2n} and \displaystyle y_n = \frac{1}{2n + 1}. Both lie in S and tend to 0, and
Two routes give 1 and -1 (Figure 2), so by Theorem 3(b) the limit does not exist.
For L \in [-1, 1] let \theta = \arccos L \in [0, \pi] and \displaystyle z_n = \frac{1}{2n + \theta/\pi}. Then \displaystyle 0 < z_n \le \frac{1}{2n}, so z_n \to 0, and f(z_n) = \cos(2n\pi + \theta) = L for every n. Conversely, -1 \le f(z_n) \le 1 along any route, so the order limit theorem gives -1 \le L \le 1.
On \displaystyle A = \left\{\frac{1}{2n} : n \ge 1\right\} the restriction f|_A is the constant 1, so it has limit 1 at 0; Theorem 6(b) cannot pass this to f, since every (0, \alpha) holds points of S outside A.
A function on the rationals
The equation x^2 = 2 has no rational solution (Irrational numbers), so g(x) = \dfrac{x^2 + 2}{x^2 - 2} is defined at every x \in \mathbb{Q}. Let g : \mathbb{Q} \to \mathbb{R} be this function. Find \displaystyle \lim_{x \to \sqrt{3}} g(x), and prove that \displaystyle \lim_{x \to \sqrt{2}} g(x) does not exist.
SolutionEvery real c is a limit point of \mathbb{Q}, since each \displaystyle \left(c, c + \frac{1}{n}\right) contains a rational (The Archimedean property and density of the rationals). So both limits are defined, though \displaystyle \sqrt{3} and \displaystyle \sqrt{2} are not in the domain.
On \mathbb{Q}, x \to c (take \delta = \varepsilon) and constants tend to themselves. At \displaystyle \sqrt{3}, Theorem 4 gives x^2 + 2 \to 5 and x^2 - 2 \to 1; the denominator never vanishes on \mathbb{Q}, so
At \displaystyle \sqrt{2}, suppose g(x) \to L. On \mathbb{Q}, x^2 + 2 = g(x)\,(x^2 - 2) and x^2 - 2 \to 0, so the product law gives x^2 + 2 \to L \cdot 0 = 0. But x^2 + 2 \to 4, and Theorem 1 would force 4 = 0. So the limit does not exist.
Olivia’s karting track
Olivia runs a karting track. A ride costs \$12 for the first 10 minutes or part of them and \$5 for each further 5 minutes or part, and no ride runs past 30 minutes. So the charge C(t) in dollars for t minutes is defined on S = (0, 30]: it is 12 on (0, 10], 17 on (10, 15], 22 on (15, 20], 27 on (20, 25] and 32 on (25, 30]. Ethan, who keeps the books, asks for (a) \lim_{t \to 0} C(t); (b) the one-sided limits at 15, and whether \lim_{t \to 15} C(t) exists; (c) \lim_{t \to 12} C(t) and the largest \delta that works for \varepsilon = 1; (d) \lim_{t \to 30} C(t).
Solution(a) 0 \notin S, but it is a limit point of S. With \delta = 10 the tested t lie in (0, 10), where C = 12, so the limit is 12.
(b) With \delta = 5, the tested points of S \cap (-\infty, 15) lie in (10, 15), where C = 17: the left-hand limit is 17. The tested points of S \cap (15, \infty) lie in (15, 20), where C = 22: the right-hand limit is 22. Since 15 is a limit point of both sides, Theorem 7 says \lim_{t \to 15} C(t) does not exist (Figure 3). That C(15) = 17 is irrelevant.
(c) The limit is 17. If \delta \le 2, the tested t lie in (10, 14), where C = 17. If \delta > 2, the strip contains t = 10, where \lvert C(10) - 17 \rvert = 5 \ge 1. So the largest \delta is 2; the jump at 15 alone would allow 3.
(d) 30 is a limit point of S \cap (-\infty, 30) but not of S \cap (30, \infty) = \varnothing, so there is no right-hand limit and Theorem 7 does not apply. The definition decides: with \delta = 5 the tested t lie in (25, 30), so \lim_{t \to 30} C(t) = 32.
Rational and irrational inputs
Let h(x) = x^2 for x \in \mathbb{Q} and h(x) = 2x - 1 for x \notin \mathbb{Q}. Prove that \lim_{x \to c} h(x) exists for c = 1, with value 1, and for no other c.
SolutionProof. Let c \ne 1. Each interval \displaystyle \left(c, c + \frac{1}{n}\right) contains a rational r_n and an irrational s_n (The Archimedean property and density of the rationals). Both sequences avoid c and tend to c, and by the sequence laws
Since c^2 - (2c - 1) = (c - 1)^2 > 0, Theorem 3(b) rules out a limit at c.
At c = 1, let 0 < \lvert x - 1 \rvert < 1, so 0 < x < 2. For rational x, \lvert h(x) - 1 \rvert = \lvert x - 1 \rvert \, \lvert x + 1 \rvert < 3\lvert x - 1 \rvert; for irrational x, \lvert h(x) - 1 \rvert = 2\lvert x - 1 \rvert. Given \varepsilon > 0, \displaystyle \delta = \min\left\{1, \frac{\varepsilon}{3}\right\} gives \lvert h(x) - 1 \rvert < 3\delta \le \varepsilon. \square
Boundaries
- Isolated points have no limit. At 7 in (0, 5] \cup \{7\}, or at any point of \mathbb{Z}, no limit is defined; the definition would accept every L.
- Particular routes prove nothing. In Example 3 the route \displaystyle \frac{1}{2n} gives 1, yet there is no limit. Theorem 2 needs every route.
- Restrictions go one way. f|_A can have a limit when f has none (Examples 3 and 5). Theorem 6(b) needs A to agree with S near c, which a one-sided restriction never does.
- Composition needs a hypothesis. Let f(x) = 0 for every x, g(y) = 1 for y \ne 0 and g(0) = 0. Then f(x) \to 0 and g(y) \to 1 as y \to 0, but g(f(x)) = 0 for every x. Exercise 14 repairs this.
Where this goes next
Continuity, defined and proved defines continuity at c \in S by \varepsilon and \delta; at a limit point of S it means \lim_{x \to c} f(x) = f(c), and the sequential criterion becomes the main tool of Extreme and intermediate value theorems, proved and Uniform continuity. Differentiability and the derivative rules, proved takes limits of difference quotients, functions on S \setminus \{c\}. Limits and continuity in the complex plane reads the same definition in the plane, where routes come from every direction.
Exercises
Limit points and limits on a domain
- Decide whether c is a limit point of S: (a) S = (0, 5] \cup \{7\}, c = 0; (b) the same S, c = 7; (c) \displaystyle S = \left\{\frac{n}{n + 1} : n \ge 1\right\}, c = 1; (d) S = \mathbb{Z}, c = 2; (e) S = \mathbb{Q} \cap [0, 1], \displaystyle c = \frac{1}{\sqrt{2}}.
- Let \displaystyle f(x) = \sqrt{9 - x^2} on [-3, 3]. Prove that \lim_{x \to -3} f(x) = 0 with \displaystyle \delta = \frac{\varepsilon^2}{6}, and for \varepsilon = 0.6 compare this \delta with the largest \delta that works, to four decimal places.
- Let \displaystyle S = \left\{\frac{1}{n} : n \ge 1\right\} and \displaystyle f\left(\frac{1}{n}\right) = \dfrac{2n + 3}{n + 4}. Find \lim_{x \to 0} f(x) and the largest \delta that works for \varepsilon = 0.01.
Routes, one-sided limits and restrictions
- Let \displaystyle k(x) = \sin\frac{\pi}{x} on (0, 1]. Find sequences in (0, 1] tending to 0 along which k is constantly 1, constantly -1 and constantly 0. What do they show?
- Let \displaystyle m(x) = x \left\lfloor \frac{1}{x} \right\rfloor on (0, 1]. (a) Prove that \lim_{x \to 0} m(x) = 1. (b) Find the one-sided limits of m at \displaystyle \frac{1}{2}.
- Let p(x) = \dfrac{x^2 + 2x}{\lvert x \rvert} for x \ne 0. Find the one-sided limits of p at 0, and decide whether \lim_{x \to 0} p(x) exists.
- Let \displaystyle S = [-1, 0] \cup \left\{\frac{1}{n} : n \ge 1\right\}, with q(x) = x + 1 for x \in [-1, 0] and \displaystyle q\left(\frac{1}{n}\right) = \frac{(-1)^n}{n}. Find the one-sided limits of q at 0 and decide whether \lim_{x \to 0} q(x) exists. Is \lim_{x \to 1/3} q(x) defined?
- Ethan books track time at Olivia’s karting track for a team: \$40 for each 15 minutes or part of them, up to 60 minutes, so \displaystyle R(t) = 40\left\lceil \frac{t}{15} \right\rceil on (0, 60]. Find (a) \lim_{t \to 0} R(t); (b) the one-sided limits at 30; (c) \lim_{t \to 20} R(t) and the largest \delta that works for \varepsilon = 30; (d) \lim_{t \to 60} R(t).
Laws and two kinds of inputs
- Find \lim_{x \to 2} \dfrac{x^3 - 8}{x^2 - 4} on S = \mathbb{R} \setminus \{-2, 2\}, naming the theorems used.
- Let u(x) = x^3 for x \in \mathbb{Q} and u(x) = x for x \notin \mathbb{Q}. Find every c at which \lim_{x \to c} u(x) exists, and the limit there.
Proofs
- Prove: if \lim_{x \to c} f(x) = L > 0, there is \delta > 0 with \displaystyle f(x) > \frac{L}{2} for every x \in S with 0 < \lvert x - c \rvert < \delta.
- Prove that f(x) \to L implies \lvert f(x) \rvert \to \lvert L \rvert. Show by an example that the converse fails, and prove that it holds when L = 0.
- Suppose that for every sequence \{x_n\} in S \setminus \{c\} with x_n \to c, the sequence \{f(x_n)\} converges. Prove that \lim_{x \to c} f(x) exists.
- Let f : S \to \mathbb{R} and g : T \to \mathbb{R} with f(S) \subseteq T. Suppose f(x) \to b as x \to c, b is a limit point of T, g(y) \to M as y \to b, and f(x) \ne b for every x \in S \setminus \{c\}. Prove that g(f(x)) \to M as x \to c.
- Prove: if f(x) \to L and g(x) \to M as x \to c with L < M, there is \delta > 0 with f(x) < g(x) for every x \in S with 0 < \lvert x - c \rvert < \delta.
Answers
- (a) Yes: \displaystyle \frac{1}{2}\min\{\varepsilon, 5\} \in S is within \varepsilon of 0. (b) No: (6, 8) \cap S = \{7\}, so no limit is defined at 7. (c) Yes: \displaystyle \frac{n}{n + 1} = 1 - \frac{1}{n + 1} \to 1, never equal to 1. (d) No: (1.5, 2.5) holds no integer but 2. (e) Yes: \displaystyle \frac{1}{\sqrt{2}} \approx 0.707 is irrational, in (0, 1), and every neighborhood of it contains rationals of (0, 1).
- Largest \displaystyle \delta = 3 - \sqrt{8.64} \approx 0.0606, just above 0.06. For x \in [-3, 3], 3 - x \le 6, so 9 - x^2 = (3 + x)(3 - x) \le 6(3 + x). If 0 < 3 + x < \delta, then \displaystyle f(x) \le \sqrt{6(3 + x)} < \sqrt{6\delta} = \varepsilon. For \varepsilon = 0.6, f(x) < 0.6 exactly when x^2 > 8.64, which near -3 means \displaystyle x < -\sqrt{8.64} \approx -2.9394.
- Limit 2; largest \displaystyle \delta = \frac{1}{496}. By Example 2 the limit is that of \displaystyle \frac{2n + 3}{n + 4}. The distance to 2 is \displaystyle \frac{5}{n + 4}, below 0.01 exactly when n > 496. A \delta tests the \displaystyle \frac{1}{n} with \displaystyle n > \frac{1}{\delta}, so it works exactly when \displaystyle \frac{1}{\delta} \ge 496.
- \displaystyle x_n = \frac{2}{4n + 1} gives \displaystyle \frac{\pi}{x_n} = 2n\pi + \frac{\pi}{2} and k = 1; \displaystyle y_n = \frac{2}{4n + 3} gives \displaystyle 2n\pi + \frac{3\pi}{2} and k = -1; \displaystyle z_n = \frac{1}{n} gives \sin n\pi = 0. By Theorem 3(b) the limit at 0 does not exist.
- (a) \displaystyle \frac{1}{x} - 1 < \left\lfloor \frac{1}{x} \right\rfloor \le \frac{1}{x}; multiplying by x > 0 gives 1 - x < m(x) \le 1, and Theorem 5(b) gives 1. (b) Left-hand limit 1, right-hand limit \displaystyle \frac{1}{2}: on \displaystyle \left(\frac{1}{3}, \frac{1}{2}\right), m(x) = 2x, and on \displaystyle \left(\frac{1}{2}, 1\right], m(x) = x.
- Right-hand limit 2, left-hand limit -2, no limit: p(x) = x + 2 for x > 0 and -(x + 2) for x < 0, and Theorem 7 applies.
- Left-hand limit 1, right-hand limit 0, no limit. On [-1, 0), q(x) = x + 1 \to 1; on \displaystyle \left\{\frac{1}{n}\right\}, Example 2 gives the limit of \displaystyle \frac{(-1)^n}{n}, which is 0; Theorem 7 applies. At \displaystyle \frac{1}{3} no limit is defined: \displaystyle \left(\frac{1}{4}, \frac{5}{12}\right) holds no other point of S.
- (a) 40, with \delta = 15. (b) Left-hand limit 80 (R = 80 on (15, 30]), right-hand limit 120 (R = 120 on (30, 45]), so no limit at 30. (c) The limit is 80 and the largest \delta is 5: for \delta \le 5 the tested t lie in (15, 25), and for \delta > 5 the strip contains t = 15, where \lvert 40 - 80 \rvert = 40 \ge 30. (d) 160, since R = 160 on (45, 60].
- 3. On S, \displaystyle \frac{x^3 - 8}{x^2 - 4} = \frac{(x - 2)(x^2 + 2x + 4)}{(x - 2)(x + 2)} = \frac{x^2 + 2x + 4}{x + 2}, and equal functions on S \setminus \{2\} have equal limits. Theorem 4 gives x^2 + 2x + 4 \to 12 and x + 2 \to 4 \ne 0, with x + 2 \ne 0 on S; the quotient law gives \displaystyle \frac{12}{4}.
- Exactly at c = -1, 0, 1, with limit c. As in Example 6, rational routes give c^3 and irrational ones c, so a limit needs c^3 = c. At such c, \lvert x - c \rvert < 1 gives \lvert x \rvert < 2 and \lvert x^3 - c^3 \rvert = \lvert x - c \rvert \, \lvert x^2 + xc + c^2 \rvert < 7\lvert x - c \rvert, so \displaystyle \delta = \min\left\{1, \frac{\varepsilon}{7}\right\} works for both kinds of x.
- Proof. Take \displaystyle \varepsilon = \frac{L}{2} in the definition. Every tested x has \displaystyle f(x) > L - \frac{L}{2} = \frac{L}{2}. \square
- Proof. By the reverse triangle inequality, \big\lvert \lvert f(x) \rvert - \lvert L \rvert \big\rvert \le \lvert f(x) - L \rvert, so a \delta serving \varepsilon for f serves it for \lvert f \rvert. The converse fails for \displaystyle s(x) = \frac{x}{\lvert x \rvert} at 0: \lvert s \rvert = 1, but the one-sided limits of s are -1 and 1. If L = 0, \lvert f(x) - 0 \rvert = \big\lvert \lvert f(x) \rvert - 0 \big\rvert, so the statements coincide. \square
- Proof. Fix one such sequence \{x_n\} (Theorem 2 of Open, closed and compact sets in R) and let L = \lim f(x_n). For any other \{y_n\}, let z_{2n-1} = x_n and z_{2n} = y_n. If \lvert x_n - c \rvert < \varepsilon for n > N_1 and \lvert y_n - c \rvert < \varepsilon for n > N_2, then \lvert z_k - c \rvert < \varepsilon for k > 2\max\{N_1, N_2\}, so z_k \to c in S \setminus \{c\} and f(z_k) converges, say to L'. Its subsequences \{f(x_n)\} and \{f(y_n)\} also tend to L' (Theorem 2 of Subsequences and Bolzano-Weierstrass), so L' = L and f(y_n) \to L. Theorem 2 finishes. \square
- Proof. Let \{x_n\} lie in S \setminus \{c\} with x_n \to c. By Theorem 2, y_n = f(x_n) \to b, and y_n \in T \setminus \{b\} since f(x) \ne b. Theorem 2 for g gives g(y_n) \to M. The sequence was arbitrary, so Theorem 2 gives g(f(x)) \to M. \square
- Proof. By Theorem 4, g(x) - f(x) \to M - L > 0, and Exercise 11 gives \delta with \displaystyle g(x) - f(x) > \frac{M - L}{2} > 0 for every tested x. \square