Pointwise and uniform convergence
A sequence of continuous functions can converge at every point to a function that is not continuous. What stronger kind of convergence keeps continuity, and when may a limit pass through an integral or a derivative?
Where you are 6 in Integration and sequences of functions
Why this exists
A power series is the limit of its partial sums (Taylor and Maclaurin series), and Representing functions as power series integrated one term by term, deferring the proof because it “needs uniform convergence”. For a limit f of functions f_n: is f continuous when every f_n is? Is \int_a^b f the limit of \int_a^b f_n? Is f' the limit of f_n'? Convergence at each point separately answers no to all three. Uniform convergence answers yes to the first two, and to the third once the derivatives converge uniformly. It asks of a sequence of functions the question Uniform continuity asked of one function: can one choice serve every x at once?
Definition
Throughout, S \subseteq \mathbb{R}, f_n : S \to \mathbb{R} for every n \ge 1, and f : S \to \mathbb{R}.
Pointwise convergence. f_n \to f pointwise on S if f_n(x) \to f(x) for every x \in S: for every x \in S and every \varepsilon > 0 there is N such that \lvert f_n(x) - f(x) \rvert < \varepsilon for every n > N. Here N may depend on x as well as on \varepsilon.
Uniform convergence. f_n \to f uniformly on S if for every \varepsilon > 0 there is N such that \lvert f_n(x) - f(x) \rvert < \varepsilon for every n > N and every x \in S. Now N is chosen before x is named (Nested quantifiers). Negated symbol by symbol, f_n \not\to f uniformly on S when
Uniform norm. For a bounded g : S \to \mathbb{R}, \lVert g \rVert_{\infty} = \sup_{x \in S} \lvert g(x) \rvert, also called the sup norm. Lebl writes \lVert g \rVert_u. It depends on S, which the text always names.
Series of functions. For g_k : S \to \mathbb{R}, the series \sum_{k=1}^{\infty} g_k converges pointwise or uniformly on S when its partial sums s_n = \sum_{k=1}^{n} g_k do, and its sum is the limit of the s_n.
Intuition
Around the graph of f draw the tube of points within \varepsilon of it, \{(x, y) : x \in S, \ \lvert y - f(x) \rvert < \varepsilon\}. Uniform convergence says that for each \varepsilon, every graph from some index on lies entirely inside the tube. Pointwise convergence says only that above each single x the values eventually enter and stay; the index can grow without bound as x moves. In Figure 1 the limit is 0 and every bump has height \displaystyle \frac{1}{2}: each point is eventually passed by, but the bump never fits in a thin band.
Each question above swaps two limits: continuity of f at c asks whether \lim_{x \to c} \lim_{n \to \infty} f_n(x) = \lim_{n \to \infty} \lim_{x \to c} f_n(x), and an integral is itself a limit of sums.
Results
Theorem 1 (uniform implies pointwise). If f_n \to f uniformly on S, then f_n \to f pointwise on S, and f_n \to f uniformly on every A \subseteq S. So a uniform limit, when it exists, is the pointwise limit.
Theorem 2 (uniform-norm test). (a) f_n \to f uniformly on S if and only if there is N_0 such that f_n - f is bounded on S for every n > N_0, and \lVert f_n - f \rVert_{\infty} \to 0. (b) If there are points x_n \in S such that f_n(x_n) - f(x_n) does not tend to 0, then f_n \not\to f uniformly on S.
Theorem 3 (Weierstrass M-test). Let g_k : S \to \mathbb{R} and numbers M_k satisfy \lvert g_k(x) \rvert \le M_k for every k \ge 1 and x \in S. If \sum_{k=1}^{\infty} M_k converges, then \sum_{k=1}^{\infty} g_k(x) converges absolutely for every x \in S, and the series converges uniformly on S.
Theorem 4 (continuity of the limit). If f_n \to f uniformly on S and every f_n is continuous at c \in S, then f is continuous at c. So a uniform limit of continuous functions is continuous.
Theorem 5 (integral of the limit). If f_n \in \mathcal{R}[a, b] for every n and f_n \to f uniformly on [a, b], then f \in \mathcal{R}[a, b] and
In particular, if \sum g_k converges uniformly on [a, b] and each g_k \in \mathcal{R}[a, b], then \int_a^b \sum_{k=1}^{\infty} g_k = \sum_{k=1}^{\infty} \int_a^b g_k.
Theorem 6 (derivative of the limit). Let each f_n : [a, b] \to \mathbb{R} be differentiable on [a, b] (one-sided at a and b) with f_n' continuous. Suppose f_n' \to g uniformly on [a, b] and \{f_n(c)\} converges for some c \in [a, b]. Then f_n converges uniformly on [a, b] to a function f that is differentiable on [a, b] with f' = g.
Theorem 6 stays true without assuming the f_n' continuous; that version, proved through the mean value theorem, is stated here without proof.
Why it is true
Theorem 1. Proof. Given x \in S and \varepsilon > 0, the N that serves \varepsilon for all of S serves it at x, and for all of A. A sequence of numbers has at most one limit (Theorem 2 of Limits of sequences, epsilon and N), so the pointwise limit is unique. \square
Theorem 2. Proof. (a) Suppose f_n \to f uniformly. With \varepsilon = 1 there is N_0 with \lvert f_n - f \rvert < 1 on S for n > N_0, so those f_n - f are bounded. Given \varepsilon > 0, take N \ge N_0 with \displaystyle \lvert f_n(x) - f(x) \rvert < \frac{\varepsilon}{2} for n > N and all x; the supremum gives \displaystyle \lVert f_n - f \rVert_{\infty} \le \frac{\varepsilon}{2} < \varepsilon. Conversely, given \varepsilon > 0, take N \ge N_0 with \lVert f_n - f \rVert_{\infty} < \varepsilon for n > N; then \lvert f_n(x) - f(x) \rvert \le \lVert f_n - f \rVert_{\infty} < \varepsilon for every x.
(b) If the convergence were uniform, then for n > N_0, 0 \le \lvert f_n(x_n) - f(x_n) \rvert \le \lVert f_n - f \rVert_{\infty} \to 0, and the squeeze theorem would give f_n(x_n) - f(x_n) \to 0. \square
Theorem 3. Proof. For each x, \sum \lvert g_k(x) \rvert converges by the comparison test (Series convergence, proved), so \sum g_k(x) converges (Absolute and conditional convergence); call its sum s(x). Let T_n = \sum_{k=n+1}^{\infty} M_k, which tends to 0 as the tail of a convergent series. For m > n, by the Triangle inequality,
Letting m \to \infty, the order limit theorem (Theorem 8 of Limits of sequences, epsilon and N) gives \lvert s(x) - s_n(x) \rvert \le T_n. The bound does not depend on x, so \lVert s - s_n \rVert_{\infty} \le T_n \to 0, and Theorem 2 applies. \square
Theorem 4. Proof. Let \varepsilon > 0. Choose one n with \displaystyle \lvert f_n(y) - f(y) \rvert < \frac{\varepsilon}{3} for every y \in S, then \delta > 0 with \displaystyle \lvert f_n(x) - f_n(c) \rvert < \frac{\varepsilon}{3} for x \in S, \lvert x - c \rvert < \delta. For such x,
\square
Theorem 5. Proof. By Theorem 2 there is N_0 such that \eta_n = \lVert f_n - f \rVert_{\infty} is finite for n > N_0, and \eta_n \to 0. Fix such an n. Each f_n is bounded, and \lvert f \rvert \le \lvert f_n \rvert + \eta_n, so f is bounded. On each piece of a partition P, f_n(x) - \eta_n \le f(x) \le f_n(x) + \eta_n, so m_i(f) \ge m_i(f_n) - \eta_n and M_i(f) \le M_i(f_n) + \eta_n. Multiplying by \Delta x_i and adding,
Given \varepsilon > 0, choose n with \displaystyle \eta_n(b - a) < \frac{\varepsilon}{3}, then by Riemann’s criterion (Result 4 of The Riemann integral, defined) a partition P with \displaystyle U(P, f_n) - L(P, f_n) < \frac{\varepsilon}{3}. Then U(P, f) - L(P, f) < \varepsilon, so f \in \mathcal{R}[a, b] by the same criterion. Now f - f_n is integrable (Result 4 of Which functions are integrable) with \lvert f - f_n \rvert \le \eta_n, so the bound in Result 1 of The fundamental theorem of calculus, proved gives
For a series, apply this to the partial sums, which are integrable by the sum rule, with \int_a^b s_n = \sum_{k=1}^{n} \int_a^b g_k. \square
Theorem 6. Proof. By Theorem 4, g is continuous, so g \in \mathcal{R}[a, b]. Let L = \lim f_n(c). For x \in [a, b], the evaluation theorem (Result 2 of The fundamental theorem of calculus, proved) on the interval between c and x, with the orientation of Result 1 there, gives f_n(x) = f_n(c) + \int_c^x f_n'. Define
Then f_n(x) - f(x) = \big(f_n(c) - L\big) + \int_c^x (f_n' - g), and Result 1 of The fundamental theorem of calculus, proved bounds the integral by \lVert f_n' - g \rVert_{\infty}(b - a):
The right side does not depend on x and tends to 0, so f_n \to f uniformly. As g is continuous, Result 4(b) of The fundamental theorem of calculus, proved (base point c) gives f' = g at every point of [a, b]. \square
Worked examples
A bump that slides into the axis
Let \displaystyle f_n(x) = \frac{nx}{1 + n^2x^2} on [0, \infty). Find the pointwise limit and \lVert f_n \rVert_{\infty}, and decide whether the convergence is uniform. Show it is uniform on [a, \infty) for every a > 0, and for \displaystyle a = \frac{1}{2} and \varepsilon = 0.1 find the smallest N that works.
Solutionf_n(0) = 0, and for x > 0, \displaystyle 0 < f_n(x) < \frac{nx}{n^2x^2} = \frac{1}{nx} \to 0. So f_n \to 0 pointwise. Next,
positive for \displaystyle x < \frac{1}{n} and negative after, so \displaystyle \lVert f_n \rVert_{\infty} = f_n(\frac{1}{n}) = \frac{1}{2} for every n. This does not tend to 0, so by Theorem 2 the convergence is not uniform on [0, \infty) (Figure 1).
On [a, \infty) with \displaystyle n \ge \frac{1}{a}, f_n is decreasing, so its supremum there is \displaystyle f_n(a) < \frac{1}{na} \to 0: uniform. For \displaystyle a = \frac{1}{2} and n \ge 2 the supremum is \displaystyle f_n(\frac{1}{2}) = \frac{2n}{n^2 + 4}, and \displaystyle \frac{2n}{n^2 + 4} < 0.1 exactly when n^2 - 20n + 4 > 0, that is \displaystyle n > 10 + \sqrt{96} \approx 19.80 (the other root is below 1). So N = 19 works: every n > 19 has \lvert f_n(x) \rvert < 0.1 on \displaystyle [\frac{1}{2}, \infty). No smaller N does, as \displaystyle \lVert f_{19} \rVert_{\infty} = \frac{38}{365} \approx 0.1041.
Continuity lost in the limit
Let \displaystyle f_n(x) = \frac{2}{\pi}\arctan(nx) on \mathbb{R}. Find the pointwise limit and show the convergence is not uniform on \mathbb{R}, nor on (0, 1], where the limit is continuous. Show it is uniform on [a, \infty) for a > 0, and for a = 0.1 and \varepsilon = 0.01 find the smallest N.
SolutionAs t \to \infty, \displaystyle \arctan t \to \frac{\pi}{2}. So f_n(x) \to 1 for x > 0, f_n(x) \to -1 for x < 0 (the arctangent is odd), and f_n(0) = 0: the limit is the sign function, discontinuous at 0 (Figure 2). Each f_n is continuous, so by Theorem 4 the convergence cannot be uniform on \mathbb{R}.
On (0, 1] the limit is the constant 1, and Theorem 4 says nothing. Take \displaystyle x_n = \frac{1}{n}: \displaystyle 1 - f_n(x_n) = 1 - \frac{2}{\pi} \cdot \frac{\pi}{4} = \frac{1}{2}, which does not tend to 0, so the convergence is not uniform there by Theorem 2(b).
For t > 0, \displaystyle \arctan t + \arctan\frac{1}{t} = \frac{\pi}{2}, so on [a, \infty)
with equality at x = a. This tends to 0, so the convergence is uniform. For a = 0.1, \displaystyle \frac{2}{\pi}\arctan\frac{10}{n} < 0.01 exactly when \displaystyle \frac{10}{n} < \tan\frac{\pi}{200}, that is \displaystyle n > \frac{10}{\tan(\pi/200)} \approx 636.6. So N = 636, and no smaller N works.
Area that does not follow the values
Let g_n(x) = n^2 x e^{-nx} on [0, 1]. Show g_n \to 0 pointwise, compute \int_0^1 g_n, and conclude that the convergence is not uniform. Find \lVert g_n \rVert_{\infty}.
Solutiong_n(0) = 0. For x > 0, \displaystyle e^{nx} \ge \frac{(nx)^3}{6} (one term of its series, all terms positive), so \displaystyle 0 < g_n(x) \le \frac{6}{n x^2} \to 0. With u = nx,
by parts. The limit function has integral 0 \ne 1, so by Theorem 5 the convergence is not uniform (Figure 3). Directly: g_n'(x) = n^2 e^{-nx}(1 - nx), so the maximum is \displaystyle g_n(\frac{1}{n}) = \frac{n}{e}, which grows without bound.
Kevin’s furnace
At a copper smelter, Kevin tests a new controller on an induction furnace over two-hour melts. On the nth trial the furnace draws \displaystyle P_n(t) = 3 + \frac{\sin(nt)}{\sqrt{n}} megawatts at time t hours, 0 \le t \le 2. Carlos, who plans the power budget, asks: (a) Does P_n converge uniformly, and past which N does every trial stay within 0.05 megawatt of the limit at every moment? (b) What do the energies E_n = \int_0^2 P_n tend to? (c) Do the rates of change P_n' converge?
Solution(a) \displaystyle \lvert P_n(t) - 3 \rvert = \frac{\lvert \sin(nt) \rvert}{\sqrt{n}} \le \frac{1}{\sqrt{n}}, with equality at \displaystyle t = \frac{\pi}{2n}, which lies in [0, 2]. So \displaystyle \lVert P_n - 3 \rVert_{\infty} = \frac{1}{\sqrt{n}} \to 0: P_n \to 3 uniformly. The tolerance holds exactly when \displaystyle \frac{1}{\sqrt{n}} < 0.05, that is n > 400, so N = 400 (trial 400 touches 0.05 exactly).
(b) By Theorem 5, E_n \to \int_0^2 3\,dt = 6 megawatt-hours, and its proof gives \displaystyle \lvert E_n - 6 \rvert \le \frac{2}{\sqrt{n}} without computing E_n. Exactly, \displaystyle E_n = 6 + \frac{1 - \cos 2n}{n^{3/2}}.
(c) No. \displaystyle P_n'(t) = \sqrt{n}\cos(nt), and \displaystyle P_n'(0) = \sqrt{n} \to \infty, while the limit 3 has derivative 0. Small wiggles can be steep, which is why Theorem 6 asks for uniform convergence of the derivatives.
A series integrated and differentiated term by term
Let \displaystyle G(x) = \sum_{k=1}^{\infty} \frac{\sin kx}{k^2} and \displaystyle H(x) = \sum_{k=1}^{\infty} \frac{\cos kx}{k^3}. Prove that G is continuous on \mathbb{R}, compute \int_0^{\pi} G, and prove that H is differentiable with H' = -G.
Solution(a) \displaystyle \left\lvert \frac{\sin kx}{k^2} \right\rvert \le \frac{1}{k^2}, and \displaystyle \sum \frac{1}{k^2} converges (The integral test and p series). By Theorem 3 the series converges uniformly on \mathbb{R}; its partial sums are continuous, so G is continuous by Theorem 4.
(b) By Theorem 5, with \displaystyle \int_0^{\pi} \sin kx\,dx = \frac{1 - \cos k\pi}{k}, which is \displaystyle \frac{2}{k} for odd k and 0 for even k,
since the even terms \displaystyle \frac{1}{(2j)^3} sum to \displaystyle \frac{1}{8} of the whole.
(c) Fix M > 0 and let H_n be the nth partial sum of H on [-M, M]. Each H_n is differentiable with continuous derivative \displaystyle H_n' = -\sum_{k=1}^{n} \frac{\sin kx}{k^2}, which converges uniformly to -G by (a) and Theorem 1. Also \displaystyle H_n(0) = \sum_{k=1}^{n} \frac{1}{k^3} converges. Theorem 6 gives a uniform limit with derivative -G on [-M, M]; by Theorem 1 it is H, so H' = -G, and M was arbitrary. One step further fails: G’s derived series \displaystyle \sum \frac{\cos kx}{k} is the harmonic series at x = 0.
Dini’s theorem
Let f_n : [a, b] \to \mathbb{R} be continuous, with f_{n+1}(x) \le f_n(x) for every n and x, and f_n \to f pointwise with f continuous. Prove that f_n \to f uniformly.
SolutionProof. Let g_n = f_n - f, continuous on [a, b]. For each x, \{g_n(x)\} is nonincreasing with limit 0, so g_n(x) \ge 0 by the order limit theorem. By the extreme value theorem (Theorem 2 of Extreme and intermediate value theorems, proved) there is x_n with g_n(x_n) = \lVert g_n \rVert_{\infty}. Since g_{n+1} \le g_n, the norms are nonincreasing and bounded below by 0, so they converge to some \varepsilon \ge 0 (The monotone convergence theorem) with \lVert g_n \rVert_{\infty} \ge \varepsilon for all n.
Suppose \varepsilon > 0. By Bolzano-Weierstrass (Theorem 5 of Subsequences and Bolzano-Weierstrass) a subsequence x_{n_k} converges to some c, in [a, b] as the interval is closed. Fix m. For every k with n_k \ge m, g_m(x_{n_k}) \ge g_{n_k}(x_{n_k}) \ge \varepsilon. As g_m is continuous at c, letting k \to \infty gives g_m(c) \ge \varepsilon (Continuity, defined and proved, Theorem 3, and the order limit theorem). This holds for every m, contradicting g_m(c) \to 0. So \varepsilon = 0, and Theorem 2 gives uniform convergence. \square
Boundaries
- Pointwise limits keep little. Continuity is lost in Example 2, the integral in Example 3.
- Theorem 4 runs one way. A continuous limit does not make convergence uniform: Example 1, Example 2 on (0, 1], Exercise 18. Dini’s theorem adds monotonicity and a closed bounded interval.
- Uniform is enough for integrals, not necessary: Exercise 3. The Lebesgue integral, in a later analysis course, needs only a common bound on the \lvert f_n \rvert.
- Theorem 5 needs a bounded interval. The functions equal to \displaystyle \frac{1}{n} on [0, n] and 0 beyond converge uniformly to 0 on [0, \infty), yet each has improper integral 1 (Improper integrals).
- Derivatives need their own uniform convergence: Example 4, Exercise 11.
- The set matters. Example 1 converges uniformly on every [a, \infty) with a > 0 but not on (0, \infty); power series behave alike inside their interval of convergence.
Where this goes next
Power series, proved uses the M-test and Theorems 4 to 6 to show that a power series is continuous, and can be integrated and differentiated term by term, on its open interval of convergence. Metric spaces, a first look treats \lVert f - g \rVert_{\infty} as a distance between bounded functions, so that uniform convergence is convergence in that metric.
Exercises
Pointwise limits and uniform norms
- Let \displaystyle f_n(x) = \frac{x}{1 + nx} on [0, \infty). Find the pointwise limit and \lVert f_n \rVert_{\infty}, and decide whether the convergence is uniform.
- Let f_n(x) = x^n(1 - x) on [0, 1]. Find \lVert f_n \rVert_{\infty}, show f_n \to 0 uniformly, and give \lVert f_3 \rVert_{\infty}.
- Let f_n(x) = nx^n(1 - x) on [0, 1]. Show f_n \to 0 pointwise but not uniformly, find \lim \lVert f_n \rVert_{\infty}, and compute \lim \int_0^1 f_n.
- Let \displaystyle f_n(x) = x + \frac{x^2}{n}. Find the smallest N with \lvert f_n(x) - x \rvert < 0.01 for all n > N and x \in [0, 3]. Show the convergence is not uniform on [0, \infty).
- Let f_n(x) = e^{-n\lvert x \rvert} on [-1, 1]. Find the pointwise limit and explain why the convergence is not uniform. On \displaystyle [\frac{1}{10}, 1], find the smallest N for \varepsilon = 0.01.
- On a second controller at the smelter, trial n draws \displaystyle Q_n(t) = 5 + \frac{t^2}{n + t} megawatts for 0 \le t \le 3 hours. Find the smallest N past which every trial stays within 0.05 megawatt of 5 throughout, and the limit of the energies \int_0^3 Q_n.
Interchanging limits
- Compute \displaystyle \lim_{n \to \infty} \int_0^{\pi} \frac{n\sin x}{n + x^2}\,dx, justifying the swap.
- Let h_n(x) = nxe^{-nx^2} on [0, 1]. Find the pointwise limit, \int_0^1 h_n and \lVert h_n \rVert_{\infty}, and decide whether the convergence is uniform.
- Show that \displaystyle \sum_{k=1}^{\infty} \frac{\cos kx}{2^k} converges uniformly on \mathbb{R}, and compute \int_0^{\pi/2} of its sum as a series. Identify the value using the Maclaurin series of \arctan.
- Let \displaystyle F(x) = \sum_{k=1}^{\infty} \frac{e^{-kx}}{k^2} on [1, 3]. Prove F'(x) = \ln(1 - e^{-x}) and give F'(1) to five decimal places.
- Let \displaystyle f_n(x) = \frac{x}{1 + nx^2} on \mathbb{R}. Show f_n \to 0 uniformly, find \lim f_n'(x) for each x, and compare with the derivative of the limit.
Proofs
- Prove that if f_n \to f and g_n \to g uniformly on S, then f_n + g_n \to f + g uniformly on S.
- Prove the Cauchy criterion: \{f_n\} converges uniformly on S to some function if and only if for every \varepsilon > 0 there is N with \lvert f_m(x) - f_n(x) \rvert < \varepsilon for all m, n > N and all x \in S.
- Show that \displaystyle f_n(x) = x + \frac{1}{n} converges uniformly on \mathbb{R} but f_n^2 does not converge uniformly to x^2.
- Prove that a uniform limit on S of uniformly continuous functions is uniformly continuous on S.
- Let f_n \to f uniformly on S with each f_n bounded. Prove that f is bounded and that one M has \lvert f_n(x) \rvert \le M for every n and every x \in S.
- Let f_n be continuous on S with f_n \to f uniformly, and x_n \to x with all x_n and x in S. Prove that f_n(x_n) \to f(x).
- Refute: “if continuous f_n converge pointwise on [0, 1] to a continuous function, the convergence is uniform.”
Answers
- Limit 0, \displaystyle \lVert f_n \rVert_{\infty} = \frac{1}{n}, uniform. For x \ge 0, \displaystyle \frac{x}{1 + nx} = \frac{1}{n} - \frac{1}{n(1 + nx)}, which lies in \displaystyle [0, \frac{1}{n}) and increases toward \displaystyle \frac{1}{n} as x \to \infty. So the supremum is \displaystyle \frac{1}{n} (not attained), and Theorem 2 applies.
- \displaystyle \lVert f_n \rVert_{\infty} = \left(\frac{n}{n + 1}\right)^n \frac{1}{n + 1} \le \frac{1}{n + 1} \to 0, and \displaystyle \lVert f_3 \rVert_{\infty} = \frac{27}{256}. f_n'(x) = x^{n-1}\big(n - (n + 1)x\big) vanishes inside at \displaystyle \frac{n}{n + 1}, and f_n(0) = f_n(1) = 0.
- \displaystyle \lim \lVert f_n \rVert_{\infty} = \frac{1}{e} and \lim \int_0^1 f_n = 0. For x < 1, nx^n \to 0 (ratio \displaystyle \frac{n + 1}{n}x \to x < 1); f_n(1) = 0. The maximum, at \displaystyle \frac{n}{n + 1}, is \displaystyle \left(\frac{n}{n + 1}\right)^{n+1} = \left(1 + \frac{1}{n}\right)^{-(n+1)} \to e^{-1} (The monotone convergence theorem), so not uniform. \displaystyle \int_0^1 f_n = \frac{n}{(n + 1)(n + 2)} \to 0.
- N = 900. \displaystyle \lvert f_n(x) - x \rvert = \frac{x^2}{n} \le \frac{9}{n} on [0, 3], attained at 3, and \displaystyle \frac{9}{n} < 0.01 exactly when n > 900. On [0, \infty), x_n = n gives f_n(x_n) - x_n = n, which does not tend to 0; Theorem 2(b).
- The limit is 1 at 0 and 0 elsewhere, discontinuous, while each f_n is continuous: not uniform by Theorem 4. On \displaystyle [\frac{1}{10}, 1], \lVert f_n \rVert_{\infty} = e^{-n/10} < 0.01 exactly when n > 10\ln 100 \approx 46.05, so N = 46.
- N = 177; energies tend to 15 megawatt-hours. \displaystyle \frac{t^2}{n + t} has derivative \displaystyle \frac{t^2 + 2nt}{(n + t)^2} \ge 0, so its maximum on [0, 3] is \displaystyle \frac{9}{n + 3}, below 0.05 exactly when n > 177. Theorem 5 gives \int_0^3 Q_n \to \int_0^3 5\,dt, with error at most \displaystyle \frac{27}{n + 3}.
- 2. \displaystyle \left\lvert \frac{n\sin x}{n + x^2} - \sin x \right\rvert = \frac{x^2 \sin x}{n + x^2} \le \frac{\pi^2}{n} on [0, \pi], so the convergence to \sin x is uniform, and Theorem 5 gives \int_0^{\pi} \sin x\,dx.
- Limit 0; \displaystyle \int_0^1 h_n = \frac{1 - e^{-n}}{2} \to \frac{1}{2}; \displaystyle \lVert h_n \rVert_{\infty} = \sqrt{\frac{n}{2e}} at \displaystyle x = \frac{1}{\sqrt{2n}}. Not uniform, by Theorem 5. Pointwise: h_n(0) = 0, and for x > 0, \displaystyle e^{nx^2} \ge \frac{n^2x^4}{2} gives \displaystyle 0 < h_n(x) \le \frac{2}{nx^3} \to 0.
- \displaystyle \arctan\frac{1}{2} \approx 0.46365. M-test with M_k = 2^{-k}. By Theorem 5 the integral is \displaystyle \sum_{k \ge 1} \frac{\sin(k\pi/2)}{k 2^k}; only odd k = 2j + 1 survive, with sign (-1)^j, giving \displaystyle \sum_{j \ge 0} \frac{(-1)^j (1/2)^{2j+1}}{2j + 1}, the arctangent series at \displaystyle \frac{1}{2}.
- F'(1) = \ln(1 - e^{-1}) \approx -0.45868. The partial sums F_n have continuous derivatives \displaystyle -\sum_{k=1}^{n} \frac{e^{-kx}}{k}, bounded termwise on [1, 3] by \displaystyle \frac{e^{-k}}{k} \le e^{-k}, a convergent geometric series, so they converge uniformly (Theorem 3), as does F_n(1). Theorem 6 gives \displaystyle F' = -\sum \frac{e^{-kx}}{k} = \ln(1 - e^{-x}), from \displaystyle -\ln(1 - u) = \sum \frac{u^k}{k} for \lvert u \rvert < 1.
- \displaystyle \lVert f_n \rVert_{\infty} = \frac{1}{2\sqrt{n}} at \displaystyle x = \pm\frac{1}{\sqrt{n}}, so uniform. \displaystyle f_n'(x) = \frac{1 - nx^2}{(1 + nx^2)^2} tends to 1 at x = 0 and to 0 elsewhere. The limit 0 has derivative 0 everywhere, so at 0 the limit of derivatives is not the derivative of the limit. The f_n' converge to a discontinuous function, so not uniformly near 0 (Theorem 4).
- Proof. Given \varepsilon > 0, take N_1 and N_2 serving \displaystyle \frac{\varepsilon}{2} for the two sequences and N = \max\{N_1, N_2\}. For n > N and x \in S, the triangle inequality gives \displaystyle \lvert (f_n + g_n)(x) - (f + g)(x) \rvert < \frac{\varepsilon}{2} + \frac{\varepsilon}{2}. \square
- Proof. If f_n \to f uniformly, take N serving \displaystyle \frac{\varepsilon}{2}; for m, n > N, the triangle inequality through f(x) gives \lvert f_m(x) - f_n(x) \rvert < \varepsilon. Conversely, for each x, \{f_n(x)\} is Cauchy, so converges (Theorem 4 of Cauchy sequences); call the limit f(x). Given \varepsilon > 0, take N with \displaystyle \lvert f_m(x) - f_n(x) \rvert < \frac{\varepsilon}{2} for m, n > N and all x. Fix n > N and x and let m \to \infty: the order limit theorem gives \displaystyle \lvert f(x) - f_n(x) \rvert \le \frac{\varepsilon}{2} < \varepsilon. \square
- Proof. \displaystyle \lVert f_n - x \rVert_{\infty} = \frac{1}{n} \to 0. But \displaystyle f_n(x)^2 - x^2 = \frac{2x}{n} + \frac{1}{n^2}, and at x_n = n this is \displaystyle 2 + \frac{1}{n^2}, which does not tend to 0; Theorem 2(b). \square
- Proof. Let \varepsilon > 0. Choose n with \displaystyle \lvert f_n - f \rvert < \frac{\varepsilon}{3} on S, then \delta with \displaystyle \lvert f_n(x) - f_n(y) \rvert < \frac{\varepsilon}{3} for x, y \in S, \lvert x - y \rvert < \delta. For such x, y, the triangle inequality through f_n(x) and f_n(y) gives \lvert f(x) - f(y) \rvert < \varepsilon. \square
- Proof. Let \lvert f_n \rvert \le B_n on S. Take N with \lvert f_n - f \rvert < 1 on S for n > N. Then \lvert f \rvert < B_{N+1} + 1 = B, so f is bounded, and for n > N, \lvert f_n \rvert < \lvert f \rvert + 1 < B + 1. So M = \max\{B_1, \dots, B_N, B + 1\} works. \square
- Proof. f is continuous by Theorem 4. For n beyond the N_0 of Theorem 2, \lvert f_n(x_n) - f(x) \rvert \le \lVert f_n - f \rVert_{\infty} + \lvert f(x_n) - f(x) \rvert. The first term tends to 0 by Theorem 2, the second by the sequential criterion (Theorem 3 of Continuity, defined and proved), so the squeeze theorem finishes. \square
- Proof. Let f_n(x) = n^2x^2e^{-nx}. Each is continuous; f_n(0) = 0, and for x > 0, \displaystyle e^{nx} \ge \frac{(nx)^3}{6} gives \displaystyle f_n(x) \le \frac{6}{nx} \to 0, so the limit is 0, continuous. But \displaystyle f_n(\frac{2}{n}) = 4e^{-2} \approx 0.54 for n \ge 2, so \lVert f_n \rVert_{\infty} does not tend to 0 and the convergence is not uniform (Theorem 2). The sequence is not monotone in n, so Dini’s theorem does not apply. \square