Type a command or search the notes. Tab cycles completions. Up and down recall history when blank. Escape dismisses, then leaves the prompt.
MATH
[1251]

Power series

Determine where an infinite polynomial converges and use calculus inside its radius.

A power series starts with coefficients and asks where their infinite polynomial converges. This reverses the Taylor-series viewpoint: a Taylor series starts with a function, derives coefficients and asks where the series equals that function.

From coefficients to a function

A power series defines a function only where it converges, so find that set before studying its algebra or calculus. Convergence has its partial-sum meaning and the convergence tests find where it holds.

[DEFINITION]

Infinite series

The series ∑n=1∞an\sum_{n=1}^{\infty}a_n converges to LL precisely when its partial sums satisfy

lim⁡k→∞sk=L,sk=∑n=1kan.\lim_{k\to\infty}s_k=L, \qquad s_k=\sum_{n=1}^{k}a_n.

In that case,

∑n=1∞an=L.\sum_{n=1}^{\infty}a_n=L.
[DEFINITION]

Power series

A power series centred at aa is a series of the form

∑k=0∞ck(x−a)k.\sum_{k=0}^{\infty}c_k(x-a)^k.

The series itself may define a new function at every value of xx where it converges.

[DEFINITION]

Radius and interval of convergence

The radius of convergence RR of

∑k=0∞ck(x−a)k\sum_{k=0}^{\infty}c_k(x-a)^k

is the number such that the series converges absolutely when ∣x−a∣<R|x-a|<R and diverges when ∣x−a∣>R|x-a|>R.

The interval of convergence is the open interval (a−R,a+R)(a-R,a+R) together with whichever endpoints converge.

With R=0R=0 the series converges only at its centre, while R=∞R=\infty gives convergence for every real xx. Here RR denotes a radius; the Taylor remainder is written separately as Rn+1(x)R_{n+1}(x).

Find the radius and test the endpoints

The radius gives the open interval where the power series converges, but it does not settle the boundary. At each endpoint the power series becomes an ordinary series, to be tested with the convergence tests.

[THEOREM]

Convergence has a central interval

Every real power series has a radius R∈[0,∞]R\in[0,\infty] with the following behaviour:

  • absolute convergence for ∣x−a∣<R|x-a|<R;
  • divergence for ∣x−a∣>R|x-a|>R;
  • separate endpoint behaviour when 0<R<∞0<R<\infty.

To see why the region has this shape, suppose the series converges at some x0≠ax_0\ne a. Then its terms satisfy ∣ck(x0−a)k∣≤M|c_k(x_0-a)^k|\le M for some M>0M>0. At any point with ∣x−a∣<∣x0−a∣|x-a|<|x_0-a|,

∣ck(x−a)k∣≤M∣x−ax0−a∣k.|c_k(x-a)^k| \le M\left|\frac{x-a}{x_0-a}\right|^k.

The right-hand side is geometric with ratio below 11, so convergence at one point forces absolute convergence at every point closer to the centre.

[METHOD]

Find the interval of convergence

For a power series centred at aa:

  1. apply the ratio or root test when its limit exists and is decisive; otherwise estimate the coefficients by another convergence test;
  2. rewrite that condition as ∣x−a∣<R|x-a|<R;
  3. record the open interval (a−R,a+R)(a-R,a+R);
  4. if 0<R<∞0<R<\infty, substitute x=a−Rx=a-R and x=a+Rx=a+R into the original series;
  5. apply the appropriate ordinary series test at each finite endpoint.

The endpoint series may be geometric, harmonic, alternating or something else and the two endpoints need not behave alike. If R=0R=0 there are no endpoints to test; if R=∞R=\infty every real xx is already inside the radius.

[EXAMPLE]

A power series with two divergent endpoints

Consider

∑k=0∞(x−3)k3k+2.\sum_{k=0}^{\infty}\frac{(x-3)^k}{3^k+2}.

Writing its term as aka_k, the ratio test gives

∣ak+1ak∣=∣x−3∣3k+23k+1+2⟶∣x−3∣3.\left|\frac{a_{k+1}}{a_k}\right| =|x-3|\frac{3^k+2}{3^{k+1}+2} \longrightarrow\frac{|x-3|}{3}.

Therefore the series converges when ∣x−3∣<3|x-3|<3, giving the possible interval (0,6)(0,6).

At x=0x=0,

ak=(−3)k3k+2,∣ak∣⟶1,a_k=\frac{(-3)^k}{3^k+2}, \qquad |a_k|\longrightarrow1,

so the terms do not tend to zero. At x=6x=6,

ak=3k3k+2⟶1,a_k=\frac{3^k}{3^k+2}\longrightarrow1,

so the terms again fail the nn-th term test. The interval of convergence is

(0,6).{(0,6)}.
[EXAMPLE]

A power series with one convergent endpoint

Consider

∑k=1∞4kk(x−2)k.\sum_{k=1}^{\infty}\frac{4^k}{k}(x-2)^k.

The ratio test gives

∣ak+1ak∣=4∣x−2∣kk+1⟶4∣x−2∣.\left|\frac{a_{k+1}}{a_k}\right| =4|x-2|\frac{k}{k+1} \longrightarrow4|x-2|.

Thus R=1/4R=1/4, with possible interval (7/4,9/4)(7/4,9/4).

At x=7/4x=7/4,

4kk(−14)k=(−1)kk,\frac{4^k}{k}\left(-\frac14\right)^k =\frac{(-1)^k}{k},

which converges by the alternating-series test. At x=9/4x=9/4,

4kk(14)k=1k,\frac{4^k}{k}\left(\frac14\right)^k =\frac1k,

which diverges. Hence

[74,94).{\left[\frac74,\frac94\right)}.

Algebra inside a common radius

Inside a common interval of absolute convergence, power series can be added and multiplied like polynomials.

[THEOREM]

Algebra of power series

Inside a common interval where the power series converge absolutely, if

f(x)=∑k=0∞ak(x−a)k,g(x)=∑k=0∞bk(x−a)k,f(x)=\sum_{k=0}^{\infty}a_k(x-a)^k, \qquad g(x)=\sum_{k=0}^{\infty}b_k(x-a)^k,

then

f(x)±g(x)=∑k=0∞(ak±bk)(x−a)k,f(x)\pm g(x) =\sum_{k=0}^{\infty}(a_k\pm b_k)(x-a)^k,

and

f(x)g(x)=∑k=0∞(∑j=0kajbk−j)(x−a)k.f(x)g(x) =\sum_{k=0}^{\infty} \left(\sum_{j=0}^{k}a_jb_{k-j}\right)(x-a)^k.

The second formula is the Cauchy product. The coefficient of (x−a)k(x-a)^k collects every pair of terms whose powers add to kk. Products at endpoints require separate justification when endpoint convergence is only conditional.

Calculus inside the radius

Inside the open interval of convergence, a power series can be differentiated and integrated term by term.

[THEOREM]

Differentiate and integrate term by term

Inside the radius of convergence,

ddx∑k=0∞ak(x−a)k=∑k=1∞kak(x−a)k−1,\frac{d}{dx}\sum_{k=0}^{\infty}a_k(x-a)^k =\sum_{k=1}^{\infty}ka_k(x-a)^{k-1},

and

∫∑k=0∞ak(x−a)k dx=C+∑k=0∞akk+1(x−a)k+1.\int\sum_{k=0}^{\infty}a_k(x-a)^k\,dx =C+\sum_{k=0}^{\infty}\frac{a_k}{k+1}(x-a)^{k+1}.

Both resulting power series have the same open radius of convergence as the original one.

On every closed interval [a−r,a+r][a-r,a+r] with 0<r<R0<r<R, the original series and its differentiated series converge uniformly. This local uniform control justifies the termwise operations; pointwise convergence alone would not.

Uniform convergence on a set means that for every tolerance ε>0\varepsilon>0, there is one index NN such that ∣Sn(x)−f(x)∣<ε|S_n(x)-f(x)|<\varepsilon for every n>Nn>N and every input xx in the set. Pointwise convergence may require a different NN at each input.

The theorem concerns only the open interval. Uniform convergence at either endpoint is a separate question and endpoint behaviour may change after differentiating or integrating.

If

f(x)=∑k=0∞ak(x−a)k,f(x)=\sum_{k=0}^{\infty}a_k(x-a)^k,

then evaluating derivatives at the centre gives

f(j)(a)=j!aj.f^{(j)}(a)=j!a_j.

Consequently, the power series is the Taylor series of the function it defines.

[EXAMPLE]

Recovering cos⁡x\cos x from sin⁡x\sin x

Termwise differentiation of

sin⁡x=∑k=0∞(−1)kx2k+1(2k+1)!\sin x=\sum_{k=0}^{\infty}(-1)^k\frac{x^{2k+1}}{(2k+1)!}

gives

cos⁡x=∑k=0∞(−1)kx2k(2k)!.\cos x =\sum_{k=0}^{\infty}(-1)^k\frac{x^{2k}}{(2k)!}.
[EXAMPLE]

Find ∑k=0∞k2xk\sum_{k=0}^{\infty}k^2x^k

Start with the geometric series:

11−x=∑k=0∞xk,∣x∣<1.\frac1{1-x}=\sum_{k=0}^{\infty}x^k, \qquad |x|<1.

Differentiate and multiply by xx:

x(1−x)2=∑k=0∞kxk.\frac{x}{(1-x)^2}=\sum_{k=0}^{\infty}kx^k.

Differentiate once more and multiply by xx:

x(1+x)(1−x)3=∑k=0∞k2xk,∣x∣<1.\frac{x(1+x)}{(1-x)^3} =\sum_{k=0}^{\infty}k^2x^k, \qquad |x|<1.

Therefore

∑k=0∞k2xk=x(1+x)(1−x)3,∣x∣<1.{ \sum_{k=0}^{\infty}k^2x^k =\frac{x(1+x)}{(1-x)^3}, \qquad |x|<1. }

Why pointwise convergence is not enough

The termwise calculus theorem is special to power series. For a general sequence of functions, a limit at each fixed input does not control the derivatives.

[PROPOSITION]

Pointwise convergence is not enough

Convergence at each fixed xx (called pointwise convergence) does not, by itself, justify differentiating the limit term by term.

Consider

fn(x)=sin⁡(n2x)n.f_n(x)=\frac{\sin(n^2x)}n.

For every fixed xx,

∣fn(x)∣≤1n⟶0,|f_n(x)|\le\frac1n\longrightarrow0,

so fn(x)→f(x)=0f_n(x)\to f(x)=0. However,

fn′(x)=ncos⁡(n2x),fn′(0)=n,f_n'(x)=n\cos(n^2x), \qquad f_n'(0)=n,

which does not converge to f′(0)=0f'(0)=0. In general,

lim⁡n→∞lim⁡h→0fn(x+h)−fn(x)handlim⁡h→0lim⁡n→∞fn(x+h)−fn(x)h\lim_{n\to\infty}\lim_{h\to0} \frac{f_n(x+h)-f_n(x)}h \quad\text{and}\quad \lim_{h\to0}\lim_{n\to\infty} \frac{f_n(x+h)-f_n(x)}h

need not agree. Power-series theorems supply the extra control inside the radius of convergence.

Because a power series is the Taylor series of its sum, its partial sums are the Taylor polynomials of that sum. The remainder in Taylor's theorem controls their error and the Taylor series page decides when a Taylor series built from a given function equals it. The next chapter carries Taylor approximation into several variables.

On this page