Power series, proved
A power series is a limit of polynomials. Where does it converge, and why may it be differentiated and integrated one term at a time, as if it were a polynomial?
Where you are 6 in Integration and sequences of functions
Why this exists
Representing functions as power series differentiated and integrated power series one term at a time and deferred the proof, because it “needs uniform convergence”; Power series and the interval of convergence gave the radius of convergence by a proof sketch. The root test with the limit superior (Series convergence, proved), the M-test and the integral of a uniform limit (Pointwise and uniform convergence), and the fundamental theorem (The fundamental theorem of calculus, proved) now pay both debts: on its open interval of convergence a power series behaves like a polynomial.
Definition
A power series centered at a \in \mathbb{R} is a series of functions
with real coefficients c_n and the convention 0^0 = 1, so the series equals c_0 at x = a. Its partial sums s_N(x) = \sum_{n=0}^{N} c_n (x - a)^n are polynomials.
Its radius of convergence is the number R \in [0, \infty] of Theorem 1: the series converges absolutely when \lvert x - a \rvert < R and diverges when \lvert x - a \rvert > R. Only one R has both properties: a point between two candidates would both converge and diverge. On (a - R, a + R) the series defines its sum f(x) = \sum_{n=0}^{\infty} c_n (x - a)^n.
The derived series and the integrated series are
Intuition
Fix r < R. On [a - r, a + r] every term satisfies \lvert c_n (x - a)^n \rvert \le \lvert c_n \rvert r^n, and these bounds form a convergent series that does not depend on x. That is exactly the M-test, so the partial sums converge uniformly there, and uniform limits keep continuity and pass through integrals. As r grows toward R the bound weakens, and on the whole open interval it can fail: in Figure 1 the partial sums hug \displaystyle \frac{4}{4 - x} on [-3, 3] but fall away near 4.
Differentiation is reached through integration: integrating the derived series term by term returns the original series up to c_0, and the fundamental theorem does the rest.
Results
Theorem 1 (radius of convergence). Let \sum_{n=0}^{\infty} c_n (x - a)^n be a power series.
(a) If \{\lvert c_n \rvert^{1/n} : n \ge 1\} is unbounded, the series converges only at x = a, and R = 0.
(b) Otherwise let L = \limsup_{n \to \infty} \lvert c_n \rvert^{1/n}. Then \displaystyle R = \frac{1}{L}, with R = \infty when L = 0 (the Cauchy-Hadamard formula).
(c) If c_n \ne 0 for all large n and \displaystyle \left\lvert \frac{c_{n+1}}{c_n} \right\rvert \to L with 0 \le L < \infty, then \displaystyle R = \frac{1}{L} (R = \infty when L = 0); if \displaystyle \left\lvert \frac{c_{n+1}}{c_n} \right\rvert \to \infty, then R = 0.
Theorem 2 (uniform convergence and continuity). Let R > 0 and let f be the sum.
(a) For every r with 0 < r < R, the series converges absolutely and uniformly on [a - r, a + r].
(b) f is continuous on (a - R, a + R).
(c) If R < \infty and \sum_{n=0}^{\infty} \lvert c_n \rvert R^n converges, the series converges uniformly on [a - R, a + R], and its sum is continuous on that closed interval.
Theorem 3 (same radius). The derived series and the integrated series of a power series have the same radius of convergence R as the series itself.
Theorem 4 (term-by-term integration). If R > 0, then for every x \in (a - R, a + R),
Theorem 5 (term-by-term differentiation). If R > 0, then f is differentiable on (a - R, a + R) and
So f has derivatives of every order there, each the series differentiated term by term, and \displaystyle c_k = \frac{f^{(k)}(a)}{k!} for every k \ge 0: a power series is the Taylor series of its sum (Taylor polynomials). In particular (uniqueness), if \sum c_n (x - a)^n and \sum d_n (x - a)^n both converge with the same sum on some (a - \delta, a + \delta), \delta > 0, then c_n = d_n for every n.
Why it is true
Theorem 1. Proof. At x = a every term after c_0 is 0. Fix x \ne a and let \rho_n = \lvert c_n (x - a)^n \rvert^{1/n} = \lvert c_n \rvert^{1/n} \lvert x - a \rvert for n \ge 1. (a) \{\rho_n\} is unbounded, so the series diverges by Theorem 5(b) of Series convergence, proved. (b) The suprema of the tails scale: \sup\{\rho_k : k \ge n\} = \lvert x - a \rvert \sup\{\lvert c_k \rvert^{1/k} : k \ge n\}, so by the product law \limsup \rho_n = L \lvert x - a \rvert. If L\lvert x - a \rvert < 1, Theorem 5(a) there gives absolute convergence; if L \lvert x - a \rvert > 1, Theorem 5(b) gives divergence. When L = 0 the first case holds for every x. (c) The ratio of consecutive terms is \displaystyle \left\lvert \frac{c_{n+1}}{c_n} \right\rvert \lvert x - a \rvert \to L \lvert x - a \rvert, and the The ratio and root tests gives absolute convergence below 1 and divergence above 1 (or when the ratio tends to \infty). The radius is unique, so (b) and (c) agree. \square
Theorem 2. Proof. (a) The point a + r lies inside, so \sum \lvert c_n \rvert r^n converges by Theorem 1. For \lvert x - a \rvert \le r, \lvert c_n (x - a)^n \rvert \le M_n = \lvert c_n \rvert r^n, and the Weierstrass M-test (Theorem 3 of Pointwise and uniform convergence) applies. (c) is the same argument with M_n = \lvert c_n \rvert R^n.
(b) Fix x_0 with \lvert x_0 - a \rvert < R and choose r with \lvert x_0 - a \rvert < r < R. The partial sums are polynomials, so continuous (Continuity, defined and proved), and by (a) and Theorem 4 of Pointwise and uniform convergence the restriction of f to [a - r, a + r] is continuous at x_0. Since x_0 lies inside (a - r, a + r), a \delta that serves this restriction, shrunk below r - \lvert x_0 - a \rvert, serves f itself. The same argument gives continuity on [a - R, a + R] in (c). \square
Theorem 3. Proof. Let R' be the radius of the derived series. Take x with \lvert x - a \rvert < R and choose r with \lvert x - a \rvert < r < R. The series converges at a + r, so c_n r^n \to 0 and \lvert c_n \rvert r^n \le B for some B and every n. With \displaystyle q = \frac{\lvert x - a \rvert}{r} < 1,
and \sum n q^{n-1} converges by the ratio test (the ratio \displaystyle \frac{n+1}{n} q \to q). By comparison the derived series converges at x; so R' \ge R. Conversely, if \lvert x - a \rvert < R', the derived series converges absolutely at x (Theorem 1), and \lvert c_n (x - a)^n \rvert \le \lvert x - a \rvert \, \lvert n c_n (x - a)^{n-1} \rvert for n \ge 1, so the series converges at x by comparison, and R \ge R'. The derived series of the integrated series is the original series, so the first part, applied to the integrated series, shows that the two have the same radius. \square
Theorem 4. Proof. By Theorem 2(b), f is continuous, so integrable on the closed interval I with ends a and x. Let r = \lvert x - a \rvert < R (if r = 0 both sides are 0). By Theorem 2(a) the partial sums converge to f uniformly on [a - r, a + r], so on I (Theorem 1 of Pointwise and uniform convergence). Each s_N is a polynomial, and the evaluation theorem (Result 2 of The fundamental theorem of calculus, proved) gives
Theorem 5 of Pointwise and uniform convergence on I gives \int_a^x s_N \to \int_a^x f; when x < a, both integrals are the negatives of integrals over [x, a] and the limit passes the sign. \square
Theorem 5. Proof. Write the derived series as \sum_{m=0}^{\infty} (m + 1)c_{m+1}(x - a)^m. By Theorem 3 it has radius R; let g be its sum, continuous on (a - R, a + R) by Theorem 2(b). Theorem 4, applied to this series, gives for \lvert x - a \rvert < R
Fix x_0 with \lvert x_0 - a \rvert < R and r with \lvert x_0 - a \rvert < r < R. On [a - r, a + r], f(x) = c_0 + \int_a^x g with g continuous, so Result 4(b) of The fundamental theorem of calculus, proved (base point a) gives f'(x_0) = g(x_0); as x_0 is inside the interval, this is the two-sided derivative of f.
Applying this to the derived series, and so on by induction, \displaystyle f^{(k)}(x) = \sum_{n \ge k} \frac{n!}{(n - k)!} c_n (x - a)^{n-k}, each series with radius R. At x = a only the term n = k survives: f^{(k)}(a) = k!\,c_k. For uniqueness, both radii are at least \delta by Theorem 1, and sums that agree near a have the same derivatives of every order at a. Hence c_k = d_k. \square
Worked examples
A radius the ratio test cannot find
Find the radius and the interval of convergence of \sum_{n=0}^{\infty} c_n x^n, where c_n = 2^n for even n and c_n = 3^n for odd n, its sum f(x), and \displaystyle f(\frac{1}{6}).
SolutionThe ratio \displaystyle \frac{c_{n+1}}{c_n} is \displaystyle 3\left(\frac{3}{2}\right)^n for even n and \displaystyle 2\left(\frac{2}{3}\right)^n for odd n, so it has no limit, finite or infinite, and Theorem 1(c) is silent. The roots \lvert c_n \rvert^{1/n} are 2 for even n \ge 2 and 3 for odd n: bounded, with every tail having supremum 3. So L = 3, and Theorem 1(b) gives \displaystyle R = \frac{1}{3}. At \displaystyle x = \pm\frac{1}{3} every odd-indexed term has absolute value 1, so the terms do not tend to 0 and both endpoints diverge (Theorem 2 of Series convergence, proved). The interval of convergence is \displaystyle \left(-\frac{1}{3}, \frac{1}{3}\right).
For \displaystyle \lvert x \rvert < \frac{1}{3}, split the series into a series carrying the even-indexed terms with zeros in the odd places, and one carrying the odd-indexed terms. Inserting zeros only repeats partial sums, so each is a convergent geometric series, and the sum law gives
The first fraction alone would allow \displaystyle \lvert x \rvert < \frac{1}{2}; the odd terms fix the radius. Finally \displaystyle f(\frac{1}{6}) = \frac{9}{8} + \frac{2}{3} = \frac{43}{24}.
Uniform on smaller intervals, not on the whole
Let \displaystyle f(x) = \sum_{n=0}^{\infty} \frac{x^n}{4^n} with partial sums s_n. Find the radius and the sum. Find the smallest n with \lvert f(x) - s_n(x) \rvert < 0.01 for every x \in [-3, 3], and show that the convergence is not uniform on (-4, 4).
SolutionThe series is geometric with ratio \displaystyle \frac{x}{4}, so R = 4 and \displaystyle f(x) = \frac{4}{4 - x}. The tail is again geometric:
For \lvert x \rvert = u \le 3, the factor \displaystyle \frac{4}{4 - x} is largest at x = u, and \displaystyle \frac{4}{4 - u}\left(\frac{u}{4}\right)^{n+1} increases with u. So on [-3, 3], \displaystyle \lVert f - s_n \rVert_{\infty} = 4\left(\frac{3}{4}\right)^{n+1}, exactly the M-test bound of Theorem 2(a) with r = 3, since every term is positive at x = 3. This is below 0.01 exactly when \displaystyle n + 1 > \frac{\ln 400}{\ln(4/3)} \approx 20.83. So n = 20, with norm \approx 0.00951, while n = 19 gives \approx 0.01268; every later n works too, since the norms decrease.
On (-4, 4), each s_n is bounded, by n + 1, while f(x) \to \infty as x \to 4^-. So f - s_n is unbounded for every n, and by Theorem 2 of Pointwise and uniform convergence the convergence is not uniform (Figure 1).
Maya’s cornhole tosses
At a cornhole league night, Maya’s bag drops through the hole with probability \displaystyle \frac{1}{5} on each toss, independently. Let X be the number of tosses up to and including her first bag through the hole, so \displaystyle P(X = n) = \frac{1}{5}\left(\frac{4}{5}\right)^{n-1} for n \ge 1. Diego, who keeps the league’s stats, sets G(x) = \sum_{n=1}^{\infty} P(X = n)\,x^n. Find G and its radius, then the mean \sum n P(X = n) and the variance \sum n^2 P(X = n) - \left(\sum n P(X = n)\right)^2.
SolutionFactoring out \displaystyle \frac{x}{5} leaves a geometric series with ratio \displaystyle \frac{4x}{5}, so the radius is \displaystyle \frac{5}{4} and
The point x = 1 lies inside \displaystyle \left(-\frac{5}{4}, \frac{5}{4}\right), so Theorem 5, used twice, gives G'(1) = \sum n P(X = n) and G''(1) = \sum n(n - 1)P(X = n). Differentiating the closed form, \displaystyle G'(x) = \frac{5}{(5 - 4x)^2} and \displaystyle G''(x) = \frac{40}{(5 - 4x)^3}. The mean is G'(1) = 5 tosses, \sum n^2 P(X = n) = 40 + 5 = 45, and the variance is 45 - 25 = 20. The exchange of limits behind G'(1) = \sum n P(X = n) is justified only because \displaystyle 1 < \frac{5}{4}.
Two integrations in a row
Prove that \displaystyle \arctan u = \sum_{n=0}^{\infty} \frac{(-1)^n u^{2n+1}}{2n + 1} for \lvert u \rvert < 1, then find \int_0^{1/2} \arctan(t^2)\,dt to six decimal places.
SolutionThe geometric series \sum_{n=0}^{\infty} (-1)^n s^{2n} has sum \displaystyle \frac{1}{1 + s^2} and radius 1. For \lvert u \rvert < 1, Theorem 4 with a = 0 integrates it from 0 to u, and the evaluation theorem (Result 2 of The fundamental theorem of calculus, proved), with \displaystyle (\arctan s)' = \frac{1}{1 + s^2}, identifies the integral as \arctan u.
For \lvert t \rvert < 1, u = t^2 is allowed, so \displaystyle \arctan(t^2) = \sum (-1)^n \frac{t^{4n+2}}{2n + 1}. This is a power series in t that converges for \lvert t \rvert < 1 and whose terms are unbounded for \lvert t \rvert > 1, so its radius is 1, and Theorem 4 applies again:
The terms b_n decrease to 0: \displaystyle b_0 = \frac{1}{24}, \displaystyle b_1 = \frac{1}{2688}, \displaystyle b_2 = \frac{1}{112640}, \displaystyle b_3 = \frac{1}{3440640}. By the alternating series test (Theorem 6 of Series convergence, proved) the sum differs from b_0 - b_1 + b_2 \approx 0.0413035 by at most b_3 < 3 \times 10^{-7}, so the integral lies in [0.0413032, 0.0413036] and equals 0.041303 to six decimal places.
Even sums have only even powers
Let f(x) = \sum_{n=0}^{\infty} c_n x^n for \lvert x \rvert < R, with R > 0. Prove that if f(-x) = f(x) for every x \in (-R, R), then c_n = 0 for every odd n.
SolutionProof. The series \sum (-1)^n c_n x^n has the same coefficients in absolute value, so the same radius by Theorem 1(b), and its sum at x is f(-x). By hypothesis this equals f(x) on (-R, R), so uniqueness (Theorem 5) gives (-1)^n c_n = c_n for every n. For odd n this says c_n = -c_n, so c_n = 0. \square
Continuous on the closed interval, differentiable only inside
Let \displaystyle f(x) = \sum_{n=1}^{\infty} \frac{x^n}{n(n + 1)}. Show that f is continuous on [-1, 1], find a closed form, evaluate \displaystyle \sum_{n=1}^{\infty} \frac{(-1)^n}{n(n + 1)}, and show that f is differentiable on (-1, 1) but has no left derivative at 1.
SolutionThe coefficient ratio is \displaystyle \frac{n}{n + 2} \to 1, so R = 1 by Theorem 1(c). Since \displaystyle \sum \frac{1}{n(n + 1)} converges (it telescopes to 1), Theorem 2(c) makes the convergence uniform on [-1, 1] and f continuous there.
For the closed form, Theorem 4 applied to \displaystyle \sum_{n=0}^{\infty} t^n = \frac{1}{1 - t} gives \displaystyle \sum_{n=0}^{\infty} \frac{x^{n+1}}{n + 1} = -\ln(1 - x) for \lvert x \rvert < 1. With \displaystyle \frac{1}{n(n + 1)} = \frac{1}{n} - \frac{1}{n + 1}, for 0 < \lvert x \rvert < 1,
Both sides are continuous at -1 from the right, so the formula also holds at x = -1:
At x = 1 the formula tends to 1, matching f(1) = 1.
By Theorem 5, \displaystyle f'(x) = \sum_{n=1}^{\infty} \frac{x^{n-1}}{n + 1} on (-1, 1), and this derived series diverges at x = 1, where it is a tail of the harmonic series. Each of its terms is nondecreasing on [0, 1), so f' is too. Given B > 0, choose K with \displaystyle \sum_{n=1}^{K} \frac{1}{n + 1} > B + 1 (the harmonic series diverges, The integral test and p series), then x_0 \in (0, 1) with \displaystyle \sum_{n=1}^{K} \frac{x_0^{n-1}}{n + 1} > B, by continuity of this polynomial at 1. For x_0 \le x < 1, the mean value theorem (The mean value theorem and its consequences) on [x, 1], where f is continuous, gives \xi \in (x, 1) with
So the left difference quotients at 1 tend to \infty, and f has no left derivative at 1, although the series converges there.
Boundaries
- Endpoints are outside every theorem but Theorem 2(c). Example 6 converges on [-1, 1], yet its derived series diverges at 1 and f has no derivative there. Abel’s theorem, used in Representing functions as power series, says that if \sum c_n R^n converges, then f(x) \to \sum c_n R^n as x \to (a + R)^-. Theorem 2(c) proves the case of absolute convergence; the general case, by summation by parts, is stated here without proof and is proved in a later analysis course.
- Uniform convergence on closed subintervals, not on the open interval. Example 2, Exercises 4 and 16.
- Derivatives of every order do not make a power series sum. The function equal to e^{-1/x^2} for x \ne 0 and 0 at 0 (Taylor's theorem with remainder) has every derivative 0 at 0, so by Theorem 5 a power series at 0 with that sum would be the zero series.
- The real line does not show the radius. \displaystyle \frac{1}{1 + x^2} has derivatives of every order on \mathbb{R}, yet its series at 0 has R = 1. The reason lives in the complex plane, at \pm i (Holomorphic functions are power series).
Where this goes next
Theorem 5 makes Power series solutions of differential equations legitimate. In Holomorphic functions are power series the same proofs run on disks \lvert z - a \rvert < R, and much more is true: every function differentiable on a disk is the sum of its Taylor series there.
Exercises
Radius of convergence
- Find the radius and the interval of convergence of \sum_{k=1}^{\infty} 2^k x^{k^2}.
- Find the radius of convergence of \displaystyle \sum_{n=0}^{\infty} \frac{(3n)!}{(n!)^3} x^n.
- Find the radius of convergence of \displaystyle \sum_{n=1}^{\infty} \left(1 + \frac{1}{n}\right)^{n^2} x^n, and show that the series diverges at both endpoints.
Uniform convergence on closed intervals
- Let s_N(x) = \sum_{n=0}^{N} (-1)^n x^{2n}. Find the smallest N with \displaystyle \left\lvert \frac{1}{1 + x^2} - s_N(x) \right\rvert < 10^{-4} for every \displaystyle x \in [-\frac{1}{2}, \frac{1}{2}], and show that the convergence is not uniform on (-1, 1).
- Let \displaystyle s_N(x) = \sum_{n=0}^{N} \frac{x^n}{n!}. Find the smallest N with \lvert e^x - s_N(x) \rvert < 10^{-3} for every x \in [-2, 2], and show that the convergence is not uniform on \mathbb{R}.
Differentiating and integrating term by term
- Find \displaystyle \sum_{n=1}^{\infty} \frac{n^2}{4^n}.
- Find \displaystyle \sum_{n=0}^{\infty} \frac{(-1)^n}{(2n + 1)\,3^n}.
- Find \displaystyle \int_0^{1/2} \frac{1 - e^{-t}}{t}\,dt to four decimal places, the integrand taking the value 1 at t = 0.
- Let f(x) = x^2 e^{x^3}. Find f^{(8)}(0) and f^{(9)}(0).
- Let \displaystyle h(x) = \sum_{n=0}^{\infty} \frac{x^n}{(n!)^2}. Show that its radius is \infty and that x h''(x) + h'(x) = h(x) for every x.
- Diego’s own bag drops through the hole with probability \displaystyle \frac{1}{3} on each toss, independently. Let Y be the number of tosses up to his first, and H(x) = \sum_{n=1}^{\infty} P(Y = n)\,x^n. Find H and its radius, the mean and the variance of Y, and, using H(-1), the probability that Y is even.
Proofs
- Let \sum c_n x^n have radius R with 0 < R < \infty. Prove that \sum c_n x^{2n} has radius \displaystyle \sqrt{R}.
- Prove that if \{c_n\} is bounded, then \sum c_n x^n has radius at least 1, and that if moreover every c_n is an integer and infinitely many are nonzero, the radius is exactly 1.
- Let f be the sum of \sum c_n (x - a)^n, with R > 0, and suppose f(x) = 0 for every x \in (a, a + \delta), for some \delta > 0. Prove that c_n = 0 for every n.
- Let \displaystyle E(x) = \sum_{n=0}^{\infty} \frac{x^n}{n!}. Prove that E is defined on \mathbb{R}, that E' = E, and that E(x)E(-x) = 1 for every x.
- Let 0 < R < \infty. Prove that if \sum c_n (x - a)^n converges uniformly on (a - R, a + R), then it converges at a + R and at a - R. Deduce again that the convergence in Example 2 is not uniform on (-4, 4).
Answers
- R = 1, interval (-1, 1). The coefficient of x^m is 2^k when m = k^2 and 0 otherwise, so no ratio exists. The roots are 2^{1/k} at m = k^2 and 0 elsewhere, so the tail suprema tend to 1 and L = 1. At x = \pm 1 the terms have absolute value 2^k, which does not tend to 0.
- \displaystyle R = \frac{1}{27}. \displaystyle \frac{c_{n+1}}{c_n} = \frac{(3n + 3)(3n + 2)(3n + 1)}{(n + 1)^3} \to 27; Theorem 1(c).
- \displaystyle R = \frac{1}{e}. \displaystyle \lvert c_n \rvert^{1/n} = \left(1 + \frac{1}{n}\right)^n \to e (The monotone convergence theorem); Theorem 1(b). At \displaystyle x = \pm\frac{1}{e} the terms have absolute value \displaystyle \exp\left(n^2 \ln\left(1 + \frac{1}{n}\right) - n\right). For u > 0, \displaystyle u - \frac{u^2}{2} < \ln(1 + u) < u - \frac{u^2}{2} + \frac{u^3}{3} (Taylor's theorem with remainder), so with \displaystyle u = \frac{1}{n} the exponent lies in \displaystyle \left(-\frac{1}{2}, -\frac{1}{2} + \frac{1}{3n}\right), and the terms tend to e^{-1/2} \ne 0.
- N = 6. The tail is \displaystyle \frac{(-1)^{N+1} x^{2N+2}}{1 + x^2}, whose absolute value increases on \displaystyle [0, \frac{1}{2}] (its derivative has the sign of (2N + 2)(1 + x^2) - 2x^2 > 0), so the norm is \displaystyle \frac{(1/4)^{N+1}}{5/4} = \frac{1}{5 \cdot 4^N}, below 10^{-4} exactly when 4^N > 2000, that is N \ge 6. On (-1, 1) the norm is at least \displaystyle \lim_{x \to 1} \frac{x^{2N+2}}{1 + x^2} = \frac{1}{2} for every N.
- N = 9. For 0 \le x \le 2, \displaystyle e^x - s_N(x) = \sum_{n > N} \frac{x^n}{n!} increases with x; for -2 \le x < 0, Theorem 4 of Series convergence, proved gives \lvert e^x - s_N(x) \rvert \le e^{\lvert x \rvert} - s_N(\lvert x \rvert). So the norm is e^2 - s_N(2), which is \approx 0.00175 for N = 8 and \approx 0.000344 for N = 9. On \mathbb{R}, \displaystyle e^x - s_N(x) \ge \frac{x^{N+1}}{(N + 1)!} for x > 0, unbounded. (The series is that of The common Maclaurin series.)
- \displaystyle \frac{20}{27}. Theorem 5 on \displaystyle \sum x^n = \frac{1}{1 - x}, times x, gives \displaystyle \sum n x^n = \frac{x}{(1 - x)^2}; once more, \displaystyle \sum n^2 x^n = x \cdot \frac{d}{dx}\frac{x}{(1 - x)^2} = \frac{x(1 + x)}{(1 - x)^3} for \lvert x \rvert < 1. At \displaystyle x = \frac{1}{4}: \displaystyle \frac{(1/4)(5/4)}{27/64}.
- \displaystyle \frac{\pi}{2\sqrt{3}} \approx 0.9069. By Example 4 at \displaystyle u = \frac{1}{\sqrt{3}} < 1: \displaystyle \frac{\pi}{6} = \arctan\frac{1}{\sqrt{3}} = \frac{1}{\sqrt{3}}\sum \frac{(-1)^n}{(2n + 1)3^n}.
- 0.4438. \displaystyle \frac{1 - e^{-t}}{t} = \sum_{n=1}^{\infty} \frac{(-1)^{n+1} t^{n-1}}{n!} with radius \infty, so Theorem 4 gives \displaystyle \sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n \cdot n!\,2^n}. The terms \displaystyle \frac{1}{2}, \frac{1}{16}, \frac{1}{144}, \frac{1}{1536}, \frac{1}{19200} give \approx 0.443845, with error at most \displaystyle b_6 = \frac{1}{276480} by the alternating series test.
- f^{(8)}(0) = 20160 and f^{(9)}(0) = 0. \displaystyle f(x) = \sum_{n=0}^{\infty} \frac{x^{3n+2}}{n!} on \mathbb{R}; the coefficient of x^8 is \displaystyle \frac{1}{2!} and there is no x^9 term, so by Theorem 5, \displaystyle f^{(8)}(0) = \frac{8!}{2}.
- The ratio \displaystyle \frac{(n!)^2}{((n + 1)!)^2} = \frac{1}{(n + 1)^2} \to 0, so R = \infty. By Theorem 5 twice, for n \ge 1 the coefficient of x^{n-1} in h' + x h'' is \displaystyle \frac{n + n(n - 1)}{(n!)^2} = \frac{1}{((n - 1)!)^2}, the coefficient of x^{n-1} in h.
- \displaystyle H(x) = \frac{x}{3 - 2x}, radius \displaystyle \frac{3}{2}; mean 3, variance 6, \displaystyle P(Y \text{ even}) = \frac{2}{5}. \displaystyle H'(x) = \frac{3}{(3 - 2x)^2} and \displaystyle H''(x) = \frac{12}{(3 - 2x)^3}, so H'(1) = 3, \sum n(n - 1)P(Y = n) = 12 and the variance is 12 + 3 - 9. Both \pm 1 lie inside the radius, and \displaystyle \sum_{n \text{ even}} P(Y = n) = \frac{H(1) + H(-1)}{2} = \frac{1 - 1/5}{2}.
- Proof. At a real x the series is \sum c_n u^n with u = x^2. By Theorem 1 it converges absolutely when x^2 < R, that is \displaystyle \lvert x \rvert < \sqrt{R}, and diverges when x^2 > R, that is \displaystyle \lvert x \rvert > \sqrt{R}. The radius is the only number with these two properties, so it is \displaystyle \sqrt{R}. \square
- Proof. Let \lvert c_n \rvert \le B. For \lvert x \rvert < 1, \lvert c_n x^n \rvert \le B\lvert x \rvert^n, and comparison with a geometric series gives convergence, so R \ge 1. If the c_n are integers and infinitely many are nonzero, then \lvert c_n \rvert \ge 1 for infinitely many n, so at x = 1 the terms do not tend to 0 and the series diverges. Were R > 1, it would converge at 1; so R = 1. \square
- Proof. Let \delta' = \min\{\delta, R\}. By Theorem 5, every f^{(k)} is differentiable on (a - R, a + R), so continuous (Differentiability and the derivative rules, proved). On (a, a + \delta'), f is constant, so by induction every f^{(k)} is 0 there. By continuity at a, f^{(k)}(a) = \lim_{x \to a^+} f^{(k)}(x) = 0, and Theorem 5 gives \displaystyle c_k = \frac{f^{(k)}(a)}{k!} = 0. \square
- Proof. The ratio \displaystyle \frac{1}{n + 1} \to 0, so R = \infty by Theorem 1(c). By Theorem 5, \displaystyle E'(x) = \sum_{n \ge 1} \frac{n x^{n-1}}{n!} = \sum_{m \ge 0} \frac{x^m}{m!} = E(x). Let p(x) = E(x)E(-x). By the product and chain rules, p'(x) = E(x)E(-x) - E(x)E(-x) = 0 on \mathbb{R}, so p is constant (The mean value theorem and its consequences), and p(0) = E(0)^2 = 1. \square
- Proof. Let f be the sum and \varepsilon > 0. Take N with \displaystyle \lvert f(x) - s_n(x) \rvert < \frac{\varepsilon}{3} for every n > N and every x in the open interval. For m > n > N, the triangle inequality gives \displaystyle \lvert s_m(x) - s_n(x) \rvert < \frac{2\varepsilon}{3} there. Now s_m - s_n is a polynomial, so continuous, and letting x \to (a + R)^- the order limit theorem (Limits of sequences, epsilon and N) gives \displaystyle \left\lvert \sum_{k=n+1}^{m} c_k R^k \right\rvert \le \frac{2\varepsilon}{3} < \varepsilon. As \varepsilon was arbitrary, the Cauchy criterion (Theorem 1 of Series convergence, proved) gives convergence at a + R; letting x \to (a - R)^+ gives it at a - R. In Example 2 the series diverges at x = 4, where every term is 1, so the convergence is not uniform on (-4, 4). \square