math.wiki
Analysis › Continuity › Uniform continuity
Sign in

Uniform continuity

Continuity lets the tolerance \delta depend on the point. When can one \delta serve every point of the domain at once, and why does a closed bounded interval always allow it?

Where you are 5 in Continuity

Why this exists

In Continuity, defined and proved the \delta that serves \varepsilon at c is chosen after c: for 3x^2 - 5 it shrank as \lvert c \rvert grew. Many proofs need one \delta for the whole domain: that continuous functions on [a, b] are integrable (Which functions are integrable), that arc length is a limit (Arc length and curvature). This lesson names that property, gives tests for and against it, and proves that it comes free on compact sets, by the subsequence argument of Extreme and intermediate value theorems, proved.

Definition

Throughout, S \subseteq \mathbb{R} and f : S \to \mathbb{R}.

Uniformly continuous. f is uniformly continuous on S if for every \varepsilon > 0 there is \delta > 0 such that \lvert f(x) - f(y) \rvert < \varepsilon for all x, y \in S with \lvert x - y \rvert < \delta.

Continuity at every point of S says: for every c \in S and every \varepsilon > 0 there is \delta > 0 with \lvert f(x) - f(c) \rvert < \varepsilon for all x \in S with \lvert x - c \rvert < \delta. There “for every c” comes first, so \delta may depend on c (Nested quantifiers); here \delta is fixed before either point is named. Negated symbol by symbol (Negating a quantified statement), f is not uniformly continuous on S when

\begin{aligned} &\exists \varepsilon > 0 \ \forall \delta > 0 \ \exists x, y \in S : \\ &\lvert x - y \rvert < \delta \text{ and } \lvert f(x) - f(y) \rvert \ge \varepsilon \end{aligned}

Lipschitz. f is Lipschitz on S with constant K if \lvert f(x) - f(y) \rvert \le K \lvert x - y \rvert for all x, y \in S: every chord of the graph has slope between -K and K.

Both properties depend on S as well as on the formula for f.

Intuition

Fix \varepsilon and ask, at each point c, how wide a window around c keeps f within \varepsilon of f(c). Where the graph is steep the window is narrow. Uniform continuity asks that the windows not shrink to width 0. In Figure 1 they narrow toward h = 0 but stop at a positive width; in Figure 2 they shrink to nothing.

h r 0.25 1 2.25 3.5 2 4 6 8
Figure 1The bowl profile \displaystyle r(h) = 4\sqrt{h} of Example 2 with \varepsilon = 2. At c = 2.25 every h within 1.25 keeps r(h) in the band (4, 8); at c = 0 only h < 0.25 keeps r(h) < 2. A single \delta for every point must be the narrowest window, 0.25.

On a closed bounded interval they cannot: where they shrink, points would pile up, by Bolzano-Weierstrass, at a point of the interval, and f would be discontinuous there.

Results

Theorem 1 (basic facts). (a) If f is uniformly continuous on S, then f is continuous at every point of S. (b) If f is uniformly continuous on S and A \subseteq S, then f|_A is uniformly continuous on A.

Theorem 2 (sequential test). f is not uniformly continuous on S if and only if there are \varepsilon_0 > 0 and sequences \{x_n\}, \{y_n\} in S with x_n - y_n \to 0 and \lvert f(x_n) - f(y_n) \rvert \ge \varepsilon_0 for every n.

Theorem 3 (Lipschitz functions). (a) If f is Lipschitz on S, then f is uniformly continuous on S. (b) Let I be an interval and f continuous on I and differentiable at every interior point of I, with \lvert f'(x) \rvert \le K there. Then f is Lipschitz on I with constant K. (c) If f is Lipschitz on S with constant K and differentiable at a limit point c \in S, then \lvert f'(c) \rvert \le K. So for f differentiable on an interval, the smallest Lipschitz constant is \sup \lvert f' \rvert.

Theorem 4 (compact domains). If K \subseteq \mathbb{R} is compact and f : K \to \mathbb{R} is continuous, then f is uniformly continuous on K. In particular every continuous function on [a, b] is uniformly continuous.

Theorem 5 (Cauchy sequences). If f is uniformly continuous on S and \{x_n\} is a Cauchy sequence of points of S, then \{f(x_n)\} is a Cauchy sequence.

Theorem 6 (continuous extension). Let a < b be real and f : (a, b) \to \mathbb{R} continuous. Then f is uniformly continuous if and only if the limits L_a = \lim_{x \to a} f(x) and L_b = \lim_{x \to b} f(x) exist. In that case F, equal to f on (a, b) with F(a) = L_a and F(b) = L_b, is continuous on [a, b].

Why it is true

Theorem 1. Proof. (a) Given c \in S and \varepsilon > 0, the uniform \delta serves \varepsilon at c: take y = c. (b) The \delta that serves \varepsilon on S serves it on A, which tests fewer pairs. \square

Theorem 2. Proof. Suppose f is not uniformly continuous, and take \varepsilon_0 from the negation. For each n \ge 1, the negation with \displaystyle \delta = \frac{1}{n} gives x_n, y_n \in S with \displaystyle \lvert x_n - y_n \rvert < \frac{1}{n} and \lvert f(x_n) - f(y_n) \rvert \ge \varepsilon_0; by the squeeze theorem x_n - y_n \to 0. Conversely, given such sequences and any \delta > 0, some n has \lvert x_n - y_n \rvert < \delta, and that pair has \lvert f(x_n) - f(y_n) \rvert \ge \varepsilon_0. So no \delta serves \varepsilon_0. \square

Theorem 3. Proof. (a) If K = 0, f is constant. If K > 0, given \varepsilon > 0 let \displaystyle \delta = \frac{\varepsilon}{K}; then \lvert x - y \rvert < \delta gives \lvert f(x) - f(y) \rvert \le K \lvert x - y \rvert < \varepsilon.

(b) Let x < y in I. Then [x, y] \subseteq I, f is continuous on [x, y] and differentiable on (x, y), whose points are interior to I. The mean value theorem (The mean value theorem) gives \xi \in (x, y) with f(y) - f(x) = f'(\xi)(y - x), so \lvert f(y) - f(x) \rvert \le K \lvert y - x \rvert.

(c) Every difference quotient \displaystyle \frac{f(x) - f(c)}{x - c}, x \in S \setminus \{c\}, lies in [-K, K], and they tend to f'(c), so -K \le f'(c) \le K by the order property of limits (Theorem 5 of Limits of functions, epsilon and delta). With (b), on an interval every K \ge \sup \lvert f' \rvert works and no smaller one does. \square

Theorem 4. Proof. Suppose f is continuous on K but not uniformly continuous. Theorem 2 gives \varepsilon_0 > 0 and x_n, y_n \in K with x_n - y_n \to 0 and \lvert f(x_n) - f(y_n) \rvert \ge \varepsilon_0. As K is compact (Open, closed and compact sets in R), a subsequence x_{n_k} converges to some c \in K. Then

y_{n_k} = x_{n_k} - (x_{n_k} - y_{n_k}) \to c - 0 = c

since a subsequence of \{x_n - y_n\} still tends to 0. By the sequential criterion (Theorem 3 of Continuity, defined and proved), f(x_{n_k}) and f(y_{n_k}) both tend to f(c), so their difference tends to 0 and is below \varepsilon_0 for large k, a contradiction. A closed bounded interval is compact by the Heine-Borel theorem on \mathbb{R}. \square

Boundedness was used to extract the subsequence, closedness to keep its limit in the domain, where continuity applies.

Theorem 5. Proof. Let \varepsilon > 0 and take \delta serving \varepsilon. As \{x_n\} is Cauchy, there is N with \lvert x_m - x_n \rvert < \delta for all m, n > N; for those, \lvert f(x_m) - f(x_n) \rvert < \varepsilon. \square

Theorem 6. Proof. Suppose the limits exist. F agrees with f near each c \in (a, b), so is continuous there. Near a, the points of [a, b] other than a lie in (a, b), so \lim_{x \to a} F(x) = L_a = F(a), and F is continuous at a by the limit form (Theorem 2 of Continuity, defined and proved); likewise at b. By Theorem 4, F is uniformly continuous, and so is its restriction f, by Theorem 1(b).

Conversely, let f be uniformly continuous and \{x_n\} in (a, b) tend to a. Being convergent, \{x_n\} is Cauchy, so \{f(x_n)\} is Cauchy by Theorem 5 and converges (Theorem 4 of Cauchy sequences). For another such sequence \{y_n\}, the interleaved x_1, y_1, x_2, y_2, \dots also tends to a, so its values converge, and their subsequences \{f(x_n)\} and \{f(y_n)\} share the limit. So all such sequences give one limit L_a, which is \lim_{x \to a} f(x) by the sequential criterion (Theorem 2 of Limits of functions, epsilon and delta). Likewise at b. \square

Worked examples

Example 1

A uniform delta for a cube

Prove that f(x) = x^3 is uniformly continuous on [-1, 2]. For \varepsilon = 0.03, compare the \delta of the proof with the largest \delta that works on the whole interval.

Solution

Factor: \lvert x^3 - y^3 \rvert = \lvert x - y \rvert \, \lvert x^2 + xy + y^2 \rvert. On [-1, 2] each of x^2, \lvert xy \rvert, y^2 is at most 4, so \lvert x^3 - y^3 \rvert \le 12 \lvert x - y \rvert: f is Lipschitz with constant 12, and Theorem 3(a) gives \displaystyle \delta = \frac{\varepsilon}{12}. For \varepsilon = 0.03 that is \delta = 0.0025.

For the largest \delta, let y = x + h with 0 < h \le 1 and -1 \le x \le 2 - h. Then y^3 - x^3 = h(3x^2 + 3xh + h^2), and the factor in parentheses is a convex quadratic in x, so its maximum on [-1, 2 - h] is at an end: 3 - 3h + h^2 at x = -1 or 12 - 6h + h^2 at x = 2 - h, the larger. So the largest change over a gap of h is

h(12 - 6h + h^2) = 8 - (2 - h)^3

which increases with h. Pairs with \lvert x - y \rvert < \delta give changes up to, but not including, 8 - (2 - \delta)^3, so \delta works exactly when 8 - (2 - \delta)^3 \le 0.03. The largest is \displaystyle \delta = 2 - \sqrt[3]{7.97} \approx 0.0025031, barely above 0.0025: 12 is the slope at x = 2.

Example 2

Mason’s bowl

Mason turns wooden bowls in a woodturning studio. The outside of one bowl has radius \displaystyle r(h) = 4\sqrt{h} inches at height h inches above its lowest point, for 0 \le h \le 4. Lily programs the cutter, whose height is known only to within \delta, and wants the radius right to within \varepsilon = 0.05 inch at every height. (a) Prove \displaystyle \lvert \sqrt{x} - \sqrt{y} \rvert \le \sqrt{\lvert x - y \rvert} for x, y \ge 0. (b) Find the largest \delta that works on [0, 4]. (c) Show that r is not Lipschitz on [0, 4]. (d) Find a \delta that works on [1, 4].

Solution

(a) Proof. Let x \ge y \ge 0. Then \displaystyle \left(\sqrt{y} + \sqrt{x - y}\right)^2 = x + 2\sqrt{y(x - y)} \ge x, so \displaystyle \sqrt{x} \le \sqrt{y} + \sqrt{x - y}, which is the claim. \square

(b) By (a), \displaystyle \lvert r(x) - r(y) \rvert \le 4\sqrt{\lvert x - y \rvert}, which is less than \varepsilon when \displaystyle \lvert x - y \rvert < \left(\frac{\varepsilon}{4}\right)^2. For \varepsilon = 0.05 this is \delta = 0.0125^2 = 0.00015625. No larger \delta works: the pair 0 and h = 0.00015625 is closer than any larger \delta, and r(h) - r(0) = 0.05, not less than \varepsilon. Figure 1 shows the same narrowing with \varepsilon = 2, where \displaystyle \left(\frac{2}{4}\right)^2 = 0.25.

(c) For h > 0, \displaystyle \frac{r(h) - r(0)}{h} = \frac{4}{\sqrt{h}}, which exceeds every K once \displaystyle h < \frac{16}{K^2}. So r is uniformly continuous on [0, 4] but not Lipschitz there.

(d) On [1, 4], \displaystyle r'(h) = \frac{2}{\sqrt{h}} \le 2, so Theorem 3(b) gives K = 2 and \displaystyle \delta = \frac{0.05}{2} = 0.025, which is 160 times the \delta of (b). The bottom decides the tolerance.

Example 3

Bounded, continuous, and still not uniform

Show that f(x) = \sin(x^2) is not uniformly continuous on \mathbb{R}, and for \displaystyle \varepsilon = \frac{1}{2} and \delta = 0.1 find a pair that defeats \delta. On which sets [-M, M] is f uniformly continuous?

Solution

Let \displaystyle x_n = \sqrt{n\pi} and \displaystyle y_n = \sqrt{n\pi + \pi/2} for n \ge 1 (Figure 2). Then \sin(x_n^2) = 0 and \sin(y_n^2) = \cos(n\pi) = \pm 1, so \lvert f(y_n) - f(x_n) \rvert = 1. Rationalizing,

y_n - x_n = \frac{\pi/2}{\sqrt{n\pi + \pi/2} + \sqrt{n\pi}} < \frac{\sqrt{\pi}}{4\sqrt{n}}

which tends to 0. Theorem 2 with \varepsilon_0 = 1 shows f is not uniformly continuous; no \delta serves \displaystyle \varepsilon = \frac{1}{2}. For \delta = 0.1 the bound is below 0.1 once n > 6.25\pi \approx 19.6. At n = 20, x_{20} \approx 7.9267 and y_{20} \approx 8.0251 are about 0.0985 apart, with values 0 and 1. (At n = 19 the gap is about 0.1010, too wide.)

x y 1 −1
Figure 2Example 3. The points \displaystyle x_n = \sqrt{n\pi} (on the axis) and \displaystyle y_n = \sqrt{n\pi + \pi/2} (at height \pm 1) draw together, yet their values stay 1 apart, so no single \delta serves \varepsilon = \tfrac{1}{2}.

On every [-M, M], f is uniformly continuous by Theorem 4.

Example 4

Two functions on an open interval

Decide whether each function is uniformly continuous on (0, 1): (a) \displaystyle g(x) = x\sin\frac{1}{x}; (b) \displaystyle k(x) = \cos\frac{\pi}{x}.

Solution

Both are continuous on (0, 1) by Theorems 4 to 6 of Continuity, defined and proved.

(a) Since \lvert g(x) \rvert \le x, the squeeze theorem gives \lim_{x \to 0} g(x) = 0, and as g is continuous on (0, \infty), \lim_{x \to 1} g(x) = \sin 1. Both limits exist, so g is uniformly continuous by Theorem 6.

(b) Let \displaystyle z_n = \frac{1}{n + 1}, a sequence in (0, 1) that converges, hence is Cauchy. But k(z_n) = \cos((n + 1)\pi) = (-1)^{n+1}, whose consecutive terms differ by 2, is not Cauchy. By Theorem 5, k is not uniformly continuous, although it is bounded.

Example 5

Uniformly continuous on a bounded set means bounded

Prove that if S is bounded and f is uniformly continuous on S, then f is bounded. Deduce that \displaystyle \frac{1}{x^2} is not uniformly continuous on (0, 1].

Solution

Proof. Let S \ne \varnothing lie in [a, b], and take \delta serving \varepsilon = 1. By the The Archimedean property and density of the rationals some integer m \ge 1 has \displaystyle \frac{b - a}{m} < \delta. Cut [a, b] into m closed pieces of length \displaystyle \frac{b - a}{m}, and in each piece that meets S choose a point s_j of S. Let B = 1 + \max_j \lvert f(s_j) \rvert. Each x \in S shares a piece with some s_j, so \lvert x - s_j \rvert < \delta and \lvert f(x) \rvert < \lvert f(s_j) \rvert + 1 \le B. \square

The function \displaystyle \frac{1}{x^2} is continuous and unbounded on the bounded set (0, 1], so it is not uniformly continuous there.

Example 6

Periodic functions

Let f : \mathbb{R} \to \mathbb{R} be continuous and periodic with period p > 0: f(x + p) = f(x) for every x. Prove that f is uniformly continuous on \mathbb{R}.

Solution

Proof. By Proof by induction, applied forward and backward, f(x + kp) = f(x) for every integer k. Let \varepsilon > 0. By Theorem 4, f is uniformly continuous on the compact set [0, 2p]; take \delta serving \varepsilon there, and replace it by \min\{\delta, p\}, which still serves.

Let x \le y with y - x < \delta. With \displaystyle k = \left\lfloor \frac{x}{p} \right\rfloor, put x' = x - kp \in [0, p) and y' = y - kp. Then x' \le y' < x' + \delta < 2p, so x', y' \in [0, 2p] with \lvert x' - y' \rvert = \lvert x - y \rvert < \delta. Hence

\lvert f(x) - f(y) \rvert = \lvert f(x') - f(y') \rvert < \varepsilon

\square

Boundaries

  • Theorem 4 needs both halves of compact. \sin(x^2) on the closed unbounded [0, \infty) and \displaystyle \cos\frac{\pi}{x} on the bounded open (0, 1) are continuous and not uniformly continuous. Some noncompact domains still work: Lipschitz functions, periodic continuous functions.
  • The implications go one way. Lipschitz implies uniformly continuous implies continuous. \displaystyle 4\sqrt{h} on [0, 4] is uniformly continuous and not Lipschitz; \sin(x^2) on \mathbb{R} is continuous and not uniformly continuous.
  • Products can fail. Sums and compositions keep uniform continuity; x \cdot x on \mathbb{R} does not (Exercise 9).
  • Theorem 6 is for bounded intervals. On (0, \infty), f(x) = x is uniformly continuous with no finite limit at \infty.
  • Beyond \mathbb{R}. The definition and Theorem 4 carry over to compact sets of a metric space, with \lvert x - y \rvert replaced by a distance (Metric spaces, a first look).

Where this goes next

Which functions are integrable uses Theorem 4 to make the upper and lower sums of a continuous function close on every fine partition. Pointwise and uniform convergence asks the same “one choice for every x” question of a sequence of functions. The mean value theorem and its consequences proves the mean value theorem behind Theorem 3(b) from the axioms.

Exercises

Uniform deltas and Lipschitz constants

  1. Prove that f(x) = x^2 - 3x is uniformly continuous on [0, 5] with \displaystyle \delta = \frac{\varepsilon}{7}. For \varepsilon = 0.07, find the largest \delta that works on [0, 5].
  2. Show that \sin 3x is Lipschitz on \mathbb{R} with constant 3. For \varepsilon = 0.3, find the largest \delta that works on \mathbb{R}.
  3. Mason’s smaller bowl has radius \displaystyle r(h) = 3\sqrt{h} for 0 \le h \le 4. For \varepsilon = 0.03, find the largest \delta that works on [0, 4]. On [1, 4], find the \delta that Theorem 3 gives and the largest \delta that works.
  4. Find the smallest Lipschitz constant K of \displaystyle \frac{1}{x^2} on [2, \infty), the \displaystyle \delta = \frac{\varepsilon}{K} for \varepsilon = 0.01, and the largest \delta that works for that \varepsilon.
  5. Find the smallest Lipschitz constant on the set given: (a) \arctan 2x on \mathbb{R}; (b) \displaystyle \frac{1}{1 + x^2} on \mathbb{R}; (c) x\lvert x \rvert on [-3, 3].

Deciding uniform continuity

  1. Use Theorem 2, giving the sequences and \varepsilon_0, to show that each function is not uniformly continuous: (a) e^x on \mathbb{R}; (b) x\sin x on [0, \infty); (c) \ln x on (0, 1).
  2. Decide which are uniformly continuous on (0, 1): (a) \displaystyle \frac{\sin x}{x}; (b) \displaystyle \sqrt{x}\cos\frac{1}{x}; (c) \displaystyle \sin\frac{1}{1 - x}.
  3. Find a Cauchy sequence \{z_n\} in \displaystyle \left(-\frac{\pi}{2}, \frac{\pi}{2}\right) for which \{\tan z_n\} is not Cauchy, and conclude. Then find the smallest Lipschitz constant of \tan on [-1.5, 1.5].

Proofs

  1. Prove that if f and g are uniformly continuous on S, so is f + g. Show that f(x) = x on \mathbb{R} is uniformly continuous but f \cdot f is not.
  2. Prove that if f and g are bounded and uniformly continuous on S, then fg is uniformly continuous on S.
  3. Let f : S \to \mathbb{R} and g : T \to \mathbb{R} be uniformly continuous with f(S) \subseteq T. Prove that g \circ f is uniformly continuous on S.
  4. Let f be continuous on [0, \infty) with \lim_{x \to \infty} f(x) = L. Prove that f is uniformly continuous on [0, \infty).
  5. Let I and J be intervals with b the largest point of I and the smallest point of J, and let f be uniformly continuous on I and on J. Prove that f is uniformly continuous on I \cup J. Deduce that \displaystyle \sqrt[3]{x} is uniformly continuous on \mathbb{R}.
  6. Let a < b and let f be continuous, monotone and bounded on (a, b). Prove that f is uniformly continuous.
  7. Prove that if f is uniformly continuous on \mathbb{R}, there are constants A and B with \lvert f(x) \rvert \le A + B\lvert x \rvert for every x. Deduce that x^3 is not uniformly continuous on \mathbb{R}.
  8. Refute: “if f is uniformly continuous on (0, 1) and on (1, 2), then it is uniformly continuous on (0, 1) \cup (1, 2).”
Answers
  1. Largest \displaystyle \delta = \frac{7 - \sqrt{48.72}}{2} \approx 0.010014. \lvert f(x) - f(y) \rvert = \lvert x - y \rvert \, \lvert x + y - 3 \rvert and -3 \le x + y - 3 \le 7 on [0, 5], so K = 7 works. For a gap h \le 5, x + y ranges over [h, 10 - h], so the largest change is h(7 - h), increasing for h < 3.5. Solve \delta(7 - \delta) = 0.07.
  2. Largest \displaystyle \delta = \frac{2}{3}\arcsin 0.15 \approx 0.10038. \lvert (\sin 3x)' \rvert \le 3 and Theorem 3(b). Also \displaystyle \sin 3x - \sin 3y = 2\cos\frac{3(x + y)}{2}\sin\frac{3(x - y)}{2}; for a gap \displaystyle h \le \frac{\pi}{3} the largest change is \displaystyle 2\sin\frac{3h}{2}, reached when the cosine is \pm 1, and increasing in h. Solve \displaystyle 2\sin\frac{3\delta}{2} = 0.3.
  3. On [0, 4]: \delta = (0.03/3)^2 = 0.0001, as in Example 2. On [1, 4]: \displaystyle r' = \frac{3}{2\sqrt{h}} \le \frac{3}{2}, so \delta = 0.02; since \displaystyle \sqrt{x + h} - \sqrt{x} decreases in x, the largest change over a gap h is \displaystyle 3(\sqrt{1 + h} - 1), and setting it to 0.03 gives the largest, \delta = 0.0201.
  4. \displaystyle K = \frac{1}{4}, \delta = 0.04, largest \displaystyle \delta = \frac{5}{\sqrt{6}} - 2 \approx 0.041241. \displaystyle \left\lvert -\frac{2}{x^3} \right\rvert \le \frac{1}{4} with equality at 2; Theorem 3(b) and (c). For a gap h, \displaystyle \frac{1}{x^2} - \frac{1}{(x + h)^2} decreases in x, so the largest change is at x = 2, and \displaystyle \frac{1}{4} - \frac{1}{(2 + \delta)^2} = 0.01 gives \displaystyle (2 + \delta)^2 = \frac{25}{6}.
  5. (a) 2: the derivative \displaystyle \frac{2}{1 + 4x^2} has supremum 2 at 0. (b) \displaystyle \frac{3\sqrt{3}}{8} \approx 0.6495: \displaystyle \lvert f'(x) \rvert = \frac{2\lvert x \rvert}{(1 + x^2)^2} is largest where 1 - 3x^2 = 0. (c) 6: x\lvert x \rvert has derivative 2\lvert x \rvert (also at 0), largest at \pm 3. Each by Theorem 3(b) and (c).
  6. (a) x_n = n, \displaystyle y_n = n + \frac{1}{n}, \varepsilon_0 = 1: \displaystyle e^{y_n} - e^{x_n} = e^n\left(e^{1/n} - 1\right) \ge \frac{e^n}{n} \ge 1, using e^t \ge 1 + t twice. (b) x_n = 2\pi n, \displaystyle y_n = 2\pi n + \frac{1}{n}, \varepsilon_0 = 4: f(x_n) = 0 and \displaystyle f(y_n) = \left(2\pi n + \frac{1}{n}\right)\sin\frac{1}{n} \ge 2\pi n \cdot \frac{2}{\pi n} = 4, as \displaystyle \sin t \ge \frac{2t}{\pi} for \displaystyle 0 \le t \le \frac{\pi}{2} (the sine is concave there). (c) \displaystyle x_n = \frac{1}{n + 1}, \displaystyle y_n = \frac{1}{2n + 2}, \varepsilon_0 = \ln 2: the gap \displaystyle \frac{1}{2n + 2} \to 0 and \ln x_n - \ln y_n = \ln 2.
  7. (a) Yes: \displaystyle \frac{\sin x}{x} \to 1 as x \to 0 (The squeeze theorem and the limit of sine x over x) and \to \sin 1 as x \to 1; Theorem 6. (b) Yes: \displaystyle \left\lvert \sqrt{x}\cos\frac{1}{x} \right\rvert \le \sqrt{x} \to 0, and the limit at 1 is \cos 1; Theorem 6. (c) No: \displaystyle z_n = 1 - \frac{1}{n\pi + \pi/2} lies in (0, 1) and tends to 1, so is Cauchy, while \displaystyle \sin\frac{1}{1 - z_n} = (-1)^n is not; Theorem 5.
  8. \displaystyle z_n = \frac{\pi}{2} - \frac{1}{n} for n \ge 1 converges, so is Cauchy, and \displaystyle \tan z_n = \cot\frac{1}{n} is unbounded, so not Cauchy (Theorem 2 of Cauchy sequences). By Theorem 5, \tan is not uniformly continuous on \displaystyle \left(-\frac{\pi}{2}, \frac{\pi}{2}\right). On [-1.5, 1.5], \sec^2 x is largest at \pm 1.5, so K = \sec^2 1.5 \approx 199.85.
  9. Proof. Given \varepsilon > 0, take \delta_1 serving \displaystyle \frac{\varepsilon}{2} for f and \delta_2 for g, and \delta = \min\{\delta_1, \delta_2\}. If \lvert x - y \rvert < \delta, the Triangle inequality gives \lvert (f + g)(x) - (f + g)(y) \rvert < \varepsilon. The function x is Lipschitz with K = 1. For x^2, take x_n = n, \displaystyle y_n = n + \frac{1}{n}: \displaystyle y_n^2 - x_n^2 = 2 + \frac{1}{n^2} \ge 2, and Theorem 2 with \varepsilon_0 = 2. \square
  10. Proof. Let M > 0 bound \lvert f \rvert and \lvert g \rvert on S. Adding and subtracting f(x)g(y), the change \lvert f(x)g(x) - f(y)g(y) \rvert is at most \lvert f(x) \rvert \lvert g(x) - g(y) \rvert + \lvert g(y) \rvert \lvert f(x) - f(y) \rvert, so at most M \lvert g(x) - g(y) \rvert + M \lvert f(x) - f(y) \rvert. Given \varepsilon > 0, take \delta_1 serving \displaystyle \frac{\varepsilon}{2M} for f, \delta_2 for g, and \delta = \min\{\delta_1, \delta_2\}; then \lvert x - y \rvert < \delta makes the left side below \varepsilon. \square
  11. Proof. Given \varepsilon > 0, take \eta > 0 with \lvert g(u) - g(v) \rvert < \varepsilon for u, v \in T, \lvert u - v \rvert < \eta, then \delta > 0 with \lvert f(x) - f(y) \rvert < \eta for x, y \in S, \lvert x - y \rvert < \delta. For such x, y, u = f(x) and v = f(y) are within \eta, so \lvert g(f(x)) - g(f(y)) \rvert < \varepsilon. \square
  12. Proof. Let \varepsilon > 0 and choose R \ge 0 with \displaystyle \lvert f(x) - L \rvert < \frac{\varepsilon}{2} for x > R. By Theorem 4, f is uniformly continuous on [0, R + 1]; take \delta_1 serving \varepsilon there and \delta = \min\{\delta_1, 1\}. Let \lvert x - y \rvert < \delta. If both lie in [0, R + 1], \lvert f(x) - f(y) \rvert < \varepsilon. Otherwise one exceeds R + 1 and the other is within 1 of it, so both exceed R and \lvert f(x) - f(y) \rvert \le \lvert f(x) - L \rvert + \lvert L - f(y) \rvert < \varepsilon. \square
  13. Proof. Given \varepsilon > 0, take \delta_1 and \delta_2 serving \displaystyle \frac{\varepsilon}{2} on I and on J, and \delta = \min\{\delta_1, \delta_2\}. Let x < y in I \cup J with y - x < \delta. If both lie in I or both in J, the change is below \displaystyle \frac{\varepsilon}{2}. Otherwise x \in I and y \in J, so x \le b \le y, and the triangle inequality through f(b) bounds the change by \displaystyle \frac{\varepsilon}{2} + \frac{\varepsilon}{2}. The cube root is continuous (Exercise 13 of Continuity, defined and proved), so uniformly continuous on [-1, 1] by Theorem 4, and Lipschitz with \displaystyle K = \frac{1}{3} on [1, \infty) and (-\infty, -1], where its derivative is \displaystyle \frac{1}{3}\lvert x \rvert^{-2/3} \le \frac{1}{3}. Glue twice. \square
  14. Proof. Replacing f by -f if needed, let f be nondecreasing, and let L = \inf f((a, b)), which exists as f is bounded. Given \varepsilon > 0, the approximation property (Least upper bounds and completeness) gives x_0 \in (a, b) with f(x_0) < L + \varepsilon; for a < x < x_0, L \le f(x) \le f(x_0) < L + \varepsilon. So \lim_{x \to a} f(x) = L. Likewise \lim_{x \to b} f(x) = \sup f((a, b)), and Theorem 6 applies. \square
  15. Proof. Take \delta serving \varepsilon = 1. For x \ne 0, choose an integer n with \displaystyle \frac{\lvert x \rvert}{\delta} < n \le \frac{\lvert x \rvert}{\delta} + 1, and let \displaystyle t_k = \frac{kx}{n} for 0 \le k \le n. Consecutive points are \displaystyle \frac{\lvert x \rvert}{n} < \delta apart, so by the triangle inequality \lvert f(x) - f(0) \rvert is less than the number of steps, \displaystyle n \le \frac{\lvert x \rvert}{\delta} + 1. So A = \lvert f(0) \rvert + 1 and \displaystyle B = \frac{1}{\delta} work. For x^3, once \lvert x \rvert > \max\{1, A + B\}, \lvert x \rvert^3 > (A + B)\lvert x \rvert \ge A + B\lvert x \rvert. \square
  16. Proof. Let f = 0 on (0, 1) and f = 1 on (1, 2). Each piece is constant, so uniformly continuous. On the union, \displaystyle x_n = 1 - \frac{1}{2n} and \displaystyle y_n = 1 + \frac{1}{2n} have \displaystyle x_n - y_n = -\frac{1}{n} \to 0 and \lvert f(x_n) - f(y_n) \rvert = 1, so Theorem 2 with \varepsilon_0 = 1 applies. In Exercise 13 the pieces share a point. \square
math.wiki · CC-BY-SA-4.0