Skip to content
Topos Uranos

Summary

A power series is a series of functions whose terms are powers of x−ax - a multiplied by coefficients, and whose partial sums are polynomials. For each xx it is a numerical series, and the tests of the unit on numerical series decide its convergence; but the set where it converges always has the same shape. Abel’s lemma shows that, if the series converges at a point, it converges absolutely at every point closer to the center; from this, with the completeness axiom, comes the radius of convergence, which the Cauchy-Hadamard theorem computes with the limit superior of the roots of the coefficients, and the ratio test, when their ratios have a limit. Inside the interval of convergence, the Weierstrass M-test gives uniform convergence on each closed and bounded interior interval, and with it the sum is continuous and is differentiated and integrated term by term, with the same radius; it follows that the coefficients are determined by the derivatives of the sum at the center. At the endpoints of the interval anything can happen, and Abel’s theorem, proved with summation by parts, guarantees that if the series converges at an endpoint, its sum is continuous up to it.

Learning objectives

  1. Prove Abel’s lemma and the existence of the radius of convergence, and compute the radius with the Cauchy-Hadamard theorem or with the ratio of the coefficients.
  2. Prove that a power series converges uniformly on each closed and bounded interval interior to its interval of convergence, and that its sum is continuous and is differentiated and integrated term by term with the same radius.
  3. Decide convergence at the endpoints of the interval, and prove Abel’s theorem on the continuity of the sum up to an endpoint where the series converges.
  4. Sum numerical series by differentiating or integrating known power series and applying Abel’s theorem.

Power series and radius of convergence

The lesson on sequences and series of functions announced that this one would apply its M-test and its corollary on term-by-term continuity, integration and differentiation to the series whose terms are powers. They are the simplest series of functions, because their partial sums are polynomials, and the most useful, because the lesson on the Taylor series of the elementary functions will show that the elementary functions are sums of such series.

DefinitionPower series

Let aa be a real number and (cn)n≥0(c_n)_{n \geq 0} a sequence of real numbers. The series of functions ∑n=0∞cn(x−a)n\sum_{n=0}^{\infty} c_n (x - a)^n is called the power series with center aa and coefficients cnc_n, with the convention (x−a)0=1(x - a)^0 = 1, also at x=ax = a. Its partial sums are the polynomials Sn(x)=∑k=0nck(x−a)kS_n(x) = \sum_{k=0}^{n} c_k (x - a)^k; at each point where the series converges, its sum is S(x)=∑n=0∞cn(x−a)nS(x) = \sum_{n=0}^{\infty} c_n (x - a)^n, and its remainder is Rn(x)=S(x)−Sn(x)R_n(x) = S(x) - S_n(x).

At x=ax = a all the terms from n=1n = 1 on are zero, and the series converges, with sum c0c_0. A series that starts at n=1n = 1, such as ∑n=1∞xnn\sum_{n=1}^{\infty} \frac{x^n}{n}, is a power series with c0=0c_0 = 0. The remainder RnR_n is that of series, as in the lesson on sequences and series of functions; it is not the Taylor remainder Rn,aR_{n,a} of the lesson on the Taylor polynomial, which this lesson does not use. The theorems of the lessons on numerical series and on series of functions were stated for series from the index 11; the series from the index 00 is that of the terms cn−1(x−a)n−1c_{n - 1} (x - a)^{n - 1}, with n≥1n \geq 1, which has the same partial sums shifted by one index, and so they apply to it without further notice.

LemmaAbel’s lemma

Let x0≠ax_0 \neq a be a number such that the sequence of the terms cn(x0−a)nc_n (x_0 - a)^n is bounded, which happens, in particular, if the series converges at x0x_0. Then the series converges absolutely at every xx with ∣x−a∣<∣x0−a∣\abs{x - a} < \abs{x_0 - a}; more precisely, if K>0K > 0 satisfies ∣cn(x0−a)n∣≤K\abs{c_n (x_0 - a)^n} \leq K for every n≥0n \geq 0, and q=∣x−a∣∣x0−a∣q = \frac{\abs{x - a}}{\abs{x_0 - a}}, then:

∣cn(x−a)n∣≤Kqn,0≤q<1\abs{c_n (x - a)^n} \leq K q^n, \qquad 0 \leq q < 1

If the terms are bounded at a point, at the points closer to the center they do not exceed, in absolute value, those of a convergent geometric series.

Proof

  1. ∣cn(x0−a)n∣≤K\dato{cota}{\abs{c_n (x_0 - a)^n} \leq K}

    If the series converges at x0x_0, its terms tend to 00, by the proposition on the necessary condition for convergence, from the lesson on numerical series and their convergence, and they are bounded, by the theorem stating that every convergent sequence is bounded, from the lesson on sequences and their limit. In both cases, a bound can be enlarged, and there is a K>0K > 0 like the one in the statement.

  2. ∣cn(x−a)n∣=∣cn∣∣x0−a∣n(∣x−a∣∣x0−a∣)n=∣cn(x0−a)n∣ qn≤Kqn\abs{c_n (x - a)^n} = \abs{c_n} \abs{x_0 - a}^n \left(\frac{\abs{x - a}}{\abs{x_0 - a}}\right)^n = \abs{c_n (x_0 - a)^n} \, q^n \leq \resaltar{K q^n}

    Let xx with ∣x−a∣<∣x0−a∣\abs{x - a} < \abs{x_0 - a}. If x≠ax \neq a, since the absolute value of a product is the product of the absolute values, and by the theorem on the laws of rational exponents, from the lesson on powers with rational exponents and radicals, with the positive bases ∣x0−a∣\abs{x_0 - a} and qq, the term at xx is the term at x0x_0 multiplied by qnq^n, and 0<q<10 < q < 1. If x=ax = a, then q=0q = 0, the terms from n=1n = 1 on are zero and, with the convention q0=1q^0 = 1, the one with index 00 is ∣c0∣≤K\abs{c_0} \leq K: the bound holds as well.

  3. 0≤∣cn(x−a)n∣≤Kqn→∑n=1∞∣cn(x−a)n∣≤Kq1−q0 \leq \abs{c_n (x - a)^n} \leq K q^n \rightarrow \resaltar{\sum_{n=1}^{\infty} \abs{c_n (x - a)^n} \leq \frac{K q}{1 - q}}

    By the theorem on the geometric series, from the lesson on numerical series and their convergence, with first term K≠0K \neq 0 and ratio qq, the series of the Kqn−1K q^{n - 1} converges, and by the proposition on the linearity and monotonicity of series, from the same lesson, so does that of the KqnK q^n. The theorem on the comparison test for series, from the lesson on series with positive terms, gives the convergence of the series of absolute values, and with the term of index 00, absolute convergence at xx.

The lemma only requires the terms to be bounded at x0x_0, not the series to converge there; the theorem below and the solved problem “The Cauchy product and Abel’s theorem” use it with convergence. It follows from it that the set of points where the series converges has no gaps around the center.

TheoremExistence of the radius of convergence

For every power series ∑n=0∞cn(x−a)n\sum_{n=0}^{\infty} c_n (x - a)^n there is a unique ρ\rho, a nonnegative real number or ρ=∞\rho = \infty, such that the series converges absolutely at every xx with ∣x−a∣<ρ\abs{x - a} < \rho and diverges at every xx with ∣x−a∣>ρ\abs{x - a} > \rho. The number ρ\rho is the supremum of the distances to the center of the points where the series converges; Abel’s lemma gives absolute convergence at everything that lies closer to the center.

Proof

  1. 0∈E,ρ=sup⁡E or ρ=∞0 \in E, \qquad \resaltar{\rho = \sup E} \text{ or } \rho = \infty

    Let EE be the set of the numbers ∣x−a∣\abs{x - a}, with xx a point where the series converges. It contains 00, because the series converges at aa. If EE is not bounded above, let ρ=∞\rho = \infty; if it is, by the completeness axiom, from the lesson on number systems, it has a supremum, and let ρ\rho be that supremum.

  2. ∣x−a∣<∣x0−a∣→∣cn(x−a)n∣≤Kqn\abs{x - a} < \abs{x_0 - a} \rightarrow \resaltar{\abs{c_n (x - a)^n} \leq K q^n}

    Let xx with ∣x−a∣<ρ\abs{x - a} < \rho. The number ∣x−a∣\abs{x - a} is not an upper bound of EE: if ρ=∞\rho = \infty, because EE is not bounded, and otherwise, because it is less than the supremum. There is then a point x0x_0 where the series converges and which lies farther from the center; in particular, x0≠ax_0 \neq a, and Abel’s lemma gives absolute convergence at xx.

  3. ∣x−a∣>ρ=sup⁡E→∣x−a∣∉E\abs{x - a} > \rho = \sup E \rightarrow \resaltar{\abs{x - a} \notin E}

    Let xx with ∣x−a∣>ρ\abs{x - a} > \rho; then ρ\rho is a real number. If the series converged at xx, the number ∣x−a∣\abs{x - a} would belong to EE and would not exceed the supremum ρ\rho. Therefore, it diverges.

  4. ρ1<t<ρ2→ρ1=ρ2\rho_1 < t < \rho_2 \rightarrow \resaltar{\rho_1 = \rho_2}

    If two numbers ρ1<ρ2\rho_1 < \rho_2 satisfied what the statement says, let t=ρ1+ρ22t = \frac{\rho_1 + \rho_2}{2} if ρ2\rho_2 is a real number, and t=ρ1+1t = \rho_1 + 1 if ρ2=∞\rho_2 = \infty. At the point x=a+tx = a + t, the series would diverge, because t>ρ1t > \rho_1, and would converge, because t<ρ2t < \rho_2. Therefore, ρ\rho is unique.

DefinitionRadius and interval of convergence

The number ρ\rho of the preceding theorem is the radius of convergence of the series. If ρ>0\rho > 0, its interval of convergence is (a−ρ,a+ρ)\ioo{a - \rho}{a + \rho}, or the whole real line if ρ=∞\rho = \infty.

If ρ=0\rho = 0, the series converges only at aa. If ρ\rho is a positive real number, the set of points where the series converges is the interval of convergence plus none, one or both of the endpoints a−ρa - \rho and a+ρa + \rho, about which the theorem says nothing. The radius is computed with the root test applied to the absolute values of the terms. The lesson on ratio and root tests called ρ\rho the limit superior of the roots; here that letter denotes the radius, and the limit superior is called LL.

TheoremCauchy-Hadamard theorem

Let L=lim sup⁡n→∞∣cn∣nL = \limsup_{n \to \infty} \sqrt[n]{\abs{c_n}}, which is a nonnegative real or ∞\infty, because the roots are not negative. Then the radius of convergence of the series ∑n=0∞cn(x−a)n\sum_{n=0}^{\infty} c_n (x - a)^n is ρ=1L\rho = \frac{1}{L} if 0<L<∞0 < L < \infty; it is ρ=∞\rho = \infty if L=0L = 0, and ρ=0\rho = 0 if L=∞L = \infty. The nnth root of ∣cn(x−a)n∣\abs{c_n (x - a)^n} is ∣x−a∣∣cn∣n\abs{x - a} \sqrt[n]{\abs{c_n}}, and its limit superior lies below or above 11 according as ∣x−a∣\abs{x - a} lies below or above 1L\frac{1}{L}.

Proof

  1. n≥N→∣cn(x−a)n∣<qntn=(qt)n,0<qt<1n \geq N \rightarrow \dato{geo}{\abs{c_n (x - a)^n} < q^n t^n = (q t)^n}, \qquad 0 < q t < 1

    Let x≠ax \neq a and t=∣x−a∣>0t = \abs{x - a} > 0, and suppose that LL is a real with tL<1t L < 1. The number q=12(L+1t)q = \frac{1}{2} \left(L + \frac{1}{t}\right) satisfies L<q<1tL < q < \frac{1}{t}. By the lemma on bounds from some index on, from the lesson on ratio and root tests, there is a natural number NN with ∣cn∣n<q\sqrt[n]{\abs{c_n}} < q for every n≥Nn \geq N. Since (∣cn∣n)n=∣cn∣\left(\sqrt[n]{\abs{c_n}}\right)^n = \abs{c_n} and, by the lemma on the monotonicity of rational powers, from the lesson on powers with rational exponents and radicals, with exponent nn, the nnth power preserves strict order between nonnegative numbers (if the root is zero, so is cnc_n), the bound follows.

  2. ∑n=0∞∣cn(x−a)n∣≤∣c0∣+∑k=1N∣ck∣tk+(qt)N+11−qt\resaltar{\sum_{n=0}^{\infty} \abs{c_n (x - a)^n} \leq \abs{c_0} + \sum_{k=1}^{N} \abs{c_k} t^k + \frac{(q t)^{N + 1}}{1 - q t}}

    By the lemma on the geometric majorant, from the lesson on ratio and root tests, with C=1C = 1 and ratio qtq t, the series of the absolute values of the terms from n=1n = 1 on converges, and with the term of index 00, the series converges absolutely at xx.

  3. (∀N)(∃n≥N)(∣cn(x−a)n∣>(qt)n>1)(\forall N)(\exists n \geq N)\left(\abs{c_n (x - a)^n} > (q t)^n > \resaltar{1}\right)

    Suppose now that tL>1t L > 1, including the case L=∞L = \infty. Let q=12(1t+L)q = \frac{1}{2} \left(\frac{1}{t} + L\right) if LL is a real, and q=1t+1q = \frac{1}{t} + 1 if L=∞L = \infty; in both cases, 1t<q<L\frac{1}{t} < q < L. By the lemma on bounds from some index on, for every natural number NN there is an n≥Nn \geq N with ∣cn∣n>q\sqrt[n]{\abs{c_n}} > q, and as before, ∣cn∣>qn\abs{c_n} > q^n. Since qt>1q t > 1, also (qt)n>1(q t)^n > 1: the terms do not tend to 00, and by the proposition on the necessary condition for convergence, the series diverges.

  4. t<1L→tL<1,t>1L→tL>1→ρ=1Lt < \frac{1}{L} \rightarrow t L < 1, \qquad t > \frac{1}{L} \rightarrow t L > 1 \rightarrow \resaltar{\rho = \frac{1}{L}}

    If L=0L = 0, the first case holds for every x≠ax \neq a, and the series converges on the whole line: by the uniqueness of the radius, from the theorem on the existence of the radius of convergence, ρ=∞\rho = \infty. If L=∞L = \infty, the second case holds for every x≠ax \neq a, and ρ=0\rho = 0. If 0<L<∞0 < L < \infty, the series converges absolutely if ∣x−a∣<1L\abs{x - a} < \frac{1}{L} and diverges if ∣x−a∣>1L\abs{x - a} > \frac{1}{L}, and by the same uniqueness, ρ=1L\rho = \frac{1}{L}.

The theorem always gives the radius, even if the roots have no limit. When the ratios of the coefficients have a limit, the ratio test gives the same result with a simpler computation; it is the corollary on the tests for terms of arbitrary sign, from the lesson on ratio and root tests, applied to the terms cn(x−a)nc_n (x - a)^n for each xx, as that lesson announced.

CorollaryThe radius by the ratio

Suppose that there is an index n1n_1 such that cn≠0c_n \neq 0 for every n≥n1n \geq n_1, and that ∣cn+1∣∣cn∣→ℓ\frac{\abs{c_{n + 1}}}{\abs{c_n}} \to \ell, where ℓ\ell is a real or ℓ=∞\ell = \infty. Then the radius of convergence is ρ=1ℓ\rho = \frac{1}{\ell} if 0<ℓ<∞0 < \ell < \infty; it is ρ=∞\rho = \infty if ℓ=0\ell = 0, and ρ=0\rho = 0 if ℓ=∞\ell = \infty. The ratio of two consecutive terms at xx is ∣x−a∣\abs{x - a} multiplied by the ratio of the coefficients, and tends to ∣x−a∣ ℓ\abs{x - a} \, \ell.

Proof

  1. ∣cn+1(x−a)n+1∣∣cn(x−a)n∣=t ∣cn+1∣∣cn∣→tℓ\frac{\abs{c_{n + 1} (x - a)^{n + 1}}}{\abs{c_n (x - a)^n}} = t \, \frac{\abs{c_{n + 1}}}{\abs{c_n}} \to \resaltar{t \ell}

    Let x≠ax \neq a and t=∣x−a∣t = \abs{x - a}. For n≥n1n \geq n_1, the term cn(x−a)nc_n (x - a)^n does not vanish. If ℓ\ell is a real, the algebra of limits gives the limit of the ratios of the absolute values; if ℓ=∞\ell = \infty, given M>0M > 0, the ratios of the coefficients exceed Mt\frac{M}{t} from some index on, and those of the terms exceed MM.

  2. lim sup⁡n→∞∣cn+1(x−a)n+1∣∣cn(x−a)n∣=tℓ<1\limsup_{n \to \infty} \frac{\abs{c_{n + 1} (x - a)^{n + 1}}}{\abs{c_n (x - a)^n}} = t \ell < \resaltar{1}

    If tℓ<1t \ell < 1, the tail of the series from n1n_1, written from the index 11, has no zero terms, and by the lemma on shifting the indices and changing finitely many terms, from the lesson on sequences and their limit, its ratios tend to tℓt \ell. Since they converge, they are bounded, and by the theorem on convergence and the limits superior and inferior, from the lesson on subsequences and the Bolzano-Weierstrass theorem, their limit superior is tℓt \ell. The corollary on the tests for terms of arbitrary sign gives the absolute convergence of the tail, and the proposition on the tail of a series, from the lesson on numerical series and their convergence, that of the series.

  3. n≥N→∣cn+1(x−a)n+1∣≥∣cn(x−a)n∣n \geq N \rightarrow \abs{c_{n + 1} (x - a)^{n + 1}} \geq \resaltar{\abs{c_n (x - a)^n}}

    If tℓ>1t \ell > 1, including the case ℓ=∞\ell = \infty, the definition of the limit, with ε=tℓ−1\varepsilon = t \ell - 1, or that of the infinite limit, with M=1M = 1, gives an index from which on the ratios are greater than 11. The second item of the corollary on the tests for terms of arbitrary sign, applied to the tail, gives its divergence, and that of the series.

  4. t<1ℓ→tℓ<1,t>1ℓ→tℓ>1→ρ=1ℓt < \frac{1}{\ell} \rightarrow t \ell < 1, \qquad t > \frac{1}{\ell} \rightarrow t \ell > 1 \rightarrow \resaltar{\rho = \frac{1}{\ell}}

    As in the last move of the Cauchy-Hadamard theorem, the uniqueness of the radius gives its value in the three cases.

For example, the geometric series ∑n=0∞xn\sum_{n=0}^{\infty} x^n has ratios equal to 11 and radius 11; the series ∑n=0∞n! xn\sum_{n=0}^{\infty} n! \, x^n, ratios n+1n + 1, which tend to ∞\infty, and radius 00; and the series ∑n=0∞xnn!\sum_{n=0}^{\infty} \frac{x^n}{n!}, ratios 1n+1\frac{1}{n + 1}, which tend to 00, and radius ∞\infty: it converges on the whole line, and the lesson on the Taylor series of the elementary functions will prove that its sum is exe^x; nothing in this lesson depends on that. The corollary does not apply if infinitely many coefficients are zero, as in ∑n=0∞x2n4n\sum_{n=0}^{\infty} \frac{x^{2n}}{4^n}, nor if the ratios have no limit, as in the solved problem “A radius the ratio does not give”; the Cauchy-Hadamard theorem does.