Power series solutions

Exponentials solve every linear equation with constant coefficients. When a coefficient changes with x, as in y’’ + (1 + x)y = 0, what does a solution look like, and on how large an interval can we trust it?

Where you are 6 in Systems and transforms

Why this exists

In a cryogenics lab, Layla and Malik cool a small steel spring inside a cryostat while it vibrates. Cold steel is stiffer, and the stiffness per unit mass climbs steadily, so the displacement y (in centimeters, a function of the time t in seconds) obeys

y'' + (1 + t)\,y = 0.

With constant stiffness, Constant-coefficient equations would finish the job with exponentials. Here e^{rt} turns the left side into (r^2 + 1 + t)\,e^{rt}, and no constant r makes that zero for every t. The solutions are not elementary functions at all.

What survives is the idea of The common Maclaurin series: store a function as its power series coefficients. Substituting y = \sum a_k t^k turns the equation into a recurrence for the a_k (as in Solving and checking a two-term recurrence, but with coefficients depending on k). This lesson proves that the series converges, and on how large an interval.

Definition

Throughout, y is a function of x (of t for the spring). A power series about x_0 is \sum_{k=0}^{\infty} a_k (x - x_0)^k (Taylor and Maclaurin series). Its radius of convergence is the \rho \in [0, \infty] such that the series converges absolutely for |x - x_0| < \rho and diverges for |x - x_0| > \rho; the The ratio and root tests often finds it. A function f is analytic at x_0 if there is r > 0 with f(x) = \sum_{k=0}^{\infty} c_k (x - x_0)^k for all |x - x_0| < r. Polynomials, e^x, \sin x and \cos x are analytic at every point, with series converging on all of \mathbb{R}.

For the homogeneous equation in standard form

y'' + p(x)\,y' + q(x)\,y = 0,

a point x_0 is an ordinary point if p and q are both analytic at x_0, and a singular point otherwise. The equation P(x)\,y'' + Q(x)\,y' + R(x)\,y = 0 with polynomial coefficients has p = Q/P and q = R/P. If P, Q, R have no common factor, x_0 is ordinary exactly when P(x_0) \ne 0: Theorem 3 gives one direction, and if P(x_0) = 0, then Q/P or R/P is unbounded near x_0, which an analytic function cannot be. So, once common factors are cancelled, the singular points are the zeros of the leading coefficient, as in Constant-coefficient equations.

A series solution about x_0 is a solution y = \sum a_k (x - x_0)^k with \rho > 0. The rule giving a_{k+2} from earlier coefficients is the recurrence. The basic solutions at an ordinary point x_0 are y_1 with y_1(x_0) = 1, y_1'(x_0) = 0 and y_2 with y_2(x_0) = 0, y_2'(x_0) = 1.

Intuition

Take x_0 = 0. If y = \sum a_k x^k, then y', y'', p\,y' and q\,y are power series, so y'' + p\,y' + q\,y is one power series, zero on an interval only if every coefficient is zero. The coefficient of x^k contains a_{k+2} with the factor (k+2)(k+1) \ne 0, and otherwise only a_0, \dots, a_{k+1}. So a_0 and a_1, the value and slope at 0, are free, and everything after them is forced.

The series of \displaystyle \frac{1}{1 + x^2} stops converging at |x| = 1 although the function is smooth on \mathbb{R}: the denominator vanishes at \pm i, at distance 1 from 0. Theorem 3 makes this the rule for solutions. In the figure, both partial sums are excellent near t = 0 and each eventually leaves the solution.

0 1 2 3 4 5 6 −1 0 1 t y S₆ S₁₆ solution
Figure 1The cooling spring of Example 2: the solution (ink), the partial sum S_6 that the example computes (dashed) and S_{16} (gray). The series converges for every t, but each partial sum is accurate only near t = 0.

Results

Theorem 1 (operations on power series, recalled). Let \sum a_k (x - x_0)^k and \sum b_k (x - x_0)^k converge to f and g for |x - x_0| < \rho, with \rho > 0.

  • (a) f is differentiable there and f'(x) = \sum_{k=1}^{\infty} k a_k (x - x_0)^{k-1}, a series with the same radius. Hence \displaystyle a_k = \frac{f^{(k)}(x_0)}{k!}.
  • (b) f g = \sum c_k (x - x_0)^k there, with c_k = \sum_{j=0}^{k} a_j b_{k-j}.
  • (c) If f = 0 on some interval |x - x_0| < \delta, then every a_k = 0.

These are stated without proof: (a) is the term by term differentiation of Representing functions as power series, proved in Real Analysis (Power series, proved), and (b) and (c) are the product and uniqueness results of The common Maclaurin series. All three hold, with the same proofs, for complex coefficients and complex x.

Theorem 2 (series solutions at an ordinary point). Let p(x) = \sum p_k (x - x_0)^k and q(x) = \sum q_k (x - x_0)^k converge for |x - x_0| < r, where 0 < r \le \infty. Let a_0, a_1 be any real numbers, and for k \ge 0 define

\begin{aligned} &(k+2)(k+1)\,a_{k+2} \\ &\quad = -\sum_{j=0}^{k} \big( (j+1)\,a_{j+1}\,p_{k-j} + a_j\,q_{k-j} \big). \end{aligned}
  • (a) The series y = \sum a_k (x - x_0)^k converges for |x - x_0| < r and solves y'' + p\,y' + q\,y = 0 there, with y(x_0) = a_0 and y'(x_0) = a_1.
  • (b) Every solution on an open interval K \ni x_0 inside (x_0 - r, x_0 + r) is the restriction of this series with a_0 = y(x_0), a_1 = y'(x_0).
  • (c) The basic solutions y_1, y_2 form a fundamental set on (x_0 - r, x_0 + r), and every solution there is a_0 y_1 + a_1 y_2.

For a forced equation y'' + p\,y' + q\,y = f with f = \sum f_k (x - x_0)^k on the same interval, add f_k to the right side of the recurrence; the proof below still works, with M also bounding |f_k|\,s^k and T_k + 1 in place of T_k.

Theorem 3 (radius from the zeros of P). Let P, Q, R be polynomials with P(x_0) \ne 0, and let d be the distance in the complex plane from x_0 to the nearest complex zero of P (d = \infty if P is constant). Then Q/P and R/P are analytic at x_0 with series converging for |x - x_0| < d. Hence every solution of P\,y'' + Q\,y' + R\,y = 0 near x_0 is a power series about x_0 converging at least for |x - x_0| < d.

The number d is a floor, not the exact radius: Example 3 beats it and Example 4 attains it.

Why it is true

Theorem 2. The shift u = x - x_0 changes no derivative, so take x_0 = 0.

Where the recurrence comes from. If y = \sum a_k x^k converges for |x| < \rho', then for |x| < \min(\rho', r) Theorem 1 gives

y'' + p\,y' + q\,y = \sum_{k=0}^{\infty} B_k x^k,

where B_k is (k+2)(k+1)a_{k+2}, from the shifted series of y'', plus the sum in the recurrence, from the products of y' = \sum (j+1)a_{j+1}x^j with p and of y with q. By Theorem 1(c) the left side is 0 exactly when every B_k = 0, which is the recurrence. So (a) comes down to convergence: once \sum a_k x^k has radius at least r, every B_k is 0 by construction.

Convergence. Fix s with 0 < s < r. The terms of the convergent series \sum p_k s^k and \sum q_k s^k tend to 0, so there is M > 0 with |p_k|\,s^k \le M and |q_k|\,s^k \le M for all k. Define A_0 = |a_0|, A_1 = |a_1| + 1 and, for k \ge 0,

\begin{aligned} (k+2)(k+1)\,A_{k+2} &= M s^{-k}\,T_k + M s\,A_{k+1}, \\ T_k &= \sum_{j=0}^{k} \big( (j+1)A_{j+1} + A_j \big) s^j. \end{aligned}

By induction |a_k| \le A_k: it holds for k = 0, 1, and if it holds up to k + 1, the recurrence, the triangle inequality and |p_{k-j}|, |q_{k-j}| \le M s^{j-k} give

\begin{aligned} (k+2)(k+1)\,|a_{k+2}| &\le M s^{-k}\,T_k \\ &\le (k+2)(k+1)\,A_{k+2}. \end{aligned}

Now T_k = T_{k-1} + \big((k+1)A_{k+1} + A_k\big)s^k, and for k \ge 1 the definition at k - 1 reads

M s^{-(k-1)}T_{k-1} = (k+1)k\,A_{k+1} - M s\,A_k.

Substituting, the A_k terms cancel:

\begin{aligned} &(k+2)(k+1)\,A_{k+2} \\ &\quad = \Big( \frac{k(k+1)}{s} + M(k+1) + M s \Big) A_{k+1}. \end{aligned}

Since A_1 > 0, every A_k with k \ge 1 is positive, and

\begin{aligned} \frac{A_{k+2}}{A_{k+1}} &= \frac{k}{(k+2)s} + \frac{M}{k+2} \\ &\quad + \frac{M s}{(k+2)(k+1)} \to \frac{1}{s}. \end{aligned}

By the ratio test \sum A_k x^k converges for |x| < s, and since |a_k x^k| \le A_k |x|^k, so does \sum a_k x^k. As s < r was arbitrary, the radius is at least r, proving (a).

Uniqueness. By Theorem 1(a), p and q are continuous on (-r, r). If y solves the equation on K, the series of (a) with a_0 = y(0), a_1 = y'(0) solves the same initial value problem, so the two agree on K by Result 2 of Constant-coefficient equations. This proves (b). For (c), W(y_1, y_2)(0) = 1, so Result 4 of that lesson makes y_1, y_2 a fundamental set, and a_0 y_1 + a_1 y_2 has value a_0 and slope a_1 at 0. \square

Theorem 3. Over the complex numbers P splits into linear factors, so partial fractions (as in Integration by partial fractions, with linear factors only) write Q/P as a polynomial plus finitely many terms \displaystyle \frac{c}{(x - z)^m}, each z a zero of P. With \displaystyle w = \frac{x - x_0}{z - x_0},

\frac{1}{(x - z)^m} = \frac{(-1)^m}{(z - x_0)^m}\cdot\frac{1}{(1 - w)^m},

and \displaystyle \frac{1}{(1 - w)^m} is the m-fold product of the geometric series \sum w^n, a power series converging for |w| < 1 by Theorem 1(b), that is, for |x - x_0| < |z - x_0|, which includes |x - x_0| < d. Adding finitely many such series gives a series for Q/P on |x - x_0| < d; its coefficients are real by Theorem 1(a), being derivatives of the real function Q/P divided by k!. The same holds for R/P, and Theorem 2 applies with r = d. \square

Worked examples

Example 1

Two series, one of them a bell curve

Find the general solution of y'' + x y' + y = 0 as power series about 0.

Solution

Here p = x and q = 1 are polynomials, so by Theorem 2 with r = \infty every solution is a power series converging on all of \mathbb{R}. Substitute y = \sum a_k x^k. Shifting the index by two, y'' has x^k coefficient (k+2)(k+1)a_{k+2}; and x y' = \sum_{k \ge 0} k a_k x^k, where the k = 0 term is zero. So the coefficient of x^k in the left side is (k+2)(k+1)a_{k+2} + (k+1)a_k. By Theorem 1(c) it vanishes; dividing by k + 1,

a_{k+2} = -\frac{a_k}{k+2}.

Even and odd coefficients do not mix. From a_0: \displaystyle a_2 = -\frac{a_0}{2}, \displaystyle a_4 = \frac{a_0}{2 \cdot 4}, and in general \displaystyle a_{2n} = \frac{(-1)^n a_0}{2^n n!}. From a_1: \displaystyle a_3 = -\frac{a_1}{3}, \displaystyle a_5 = \frac{a_1}{3 \cdot 5}, and \displaystyle a_{2n+1} = \frac{(-1)^n a_1}{1 \cdot 3 \cdots (2n+1)}. So

\begin{aligned} y &= a_0 \sum_{n=0}^{\infty} \frac{(-1)^n x^{2n}}{2^n n!} \\ &\quad + a_1 \sum_{n=0}^{\infty} \frac{(-1)^n x^{2n+1}}{1 \cdot 3 \cdots (2n+1)}. \end{aligned}

The first series is e^u with \displaystyle u = -\frac{x^2}{2}, so y_1 = e^{-x^2/2}. Check: y_1' = -x e^{-x^2/2} and y_1'' = (x^2 - 1)e^{-x^2/2}, and (x^2 - 1) - x^2 + 1 = 0. The second series has no elementary closed form; Exercise 10 writes it with an integral.

Example 2

The cooling spring

Solve y'' + (1 + t)\,y = 0, y(0) = 1, y'(0) = 0 through the term in t^6, and estimate y(0.5).

Solution

The coefficients are polynomials, so the series converges for every t. With y = \sum a_k t^k, the product (1 + t)y is \sum_{k \ge 0} a_k t^k plus \sum_{k \ge 1} a_{k-1}t^k. The constant term gives 2a_2 + a_0 = 0, and the coefficient of t^k for k \ge 1 gives

(k+2)(k+1)\,a_{k+2} = -(a_k + a_{k-1}).

Each coefficient uses the two before it, so there is no parity split and no simple formula; we compute. With a_0 = 1, a_1 = 0: \displaystyle a_2 = -\frac12; 6a_3 = -(a_1 + a_0) gives \displaystyle a_3 = -\frac16; \displaystyle 12a_4 = -(a_2 + a_1) = \frac12 gives \displaystyle a_4 = \frac{1}{24}; \displaystyle 20a_5 = -(a_3 + a_2) = \frac23 gives \displaystyle a_5 = \frac{1}{30}; \displaystyle 30a_6 = -(a_4 + a_3) = \frac18 gives \displaystyle a_6 = \frac{1}{240}. So

\begin{aligned} y &\approx 1 - \frac{t^2}{2} - \frac{t^3}{6} \\ &\quad + \frac{t^4}{24} + \frac{t^5}{30} + \frac{t^6}{240}. \end{aligned}

At t = 0.5 the six terms sum to 0.857878. A numerical solution gives 0.857861; the next term, \displaystyle a_7 t^7 = -\frac{1}{560 \cdot 128} \approx -0.000014, accounts for most of the gap.

Example 3

A shifted center and a radius from complex zeros

For (x^2 - 2x + 5)\,y'' + x y' - y = 0 about x_0 = 1, give the radius Theorem 3 guarantees, find the recurrence, and find the general solution, with y_1 through (x - 1)^4.

Solution

P(x) = (x - 1)^2 + 4 has no real zero, but its complex zeros 1 \pm 2i lie at distance 2 from 1. So the series converge at least for |x - 1| < 2, that is, on (-1, 3).

Put u = x - 1; the equation reads (u^2 + 4)\,y'' + (u + 1)\,y' - y = 0, with primes in u. With y = \sum a_k u^k, the coefficient of u^k collects k(k-1)a_k from u^2 y'', 4(k+2)(k+1)a_{k+2} from 4y'', k a_k from u y', (k+1)a_{k+1} from y', and -a_k. Since k(k-1) + k - 1 = (k-1)(k+1), setting it to 0 and dividing by k + 1 gives

a_{k+2} = -\frac{a_{k+1} + (k-1)\,a_k}{4(k+2)}.

For y_1 (a_0 = 1, a_1 = 0): \displaystyle a_2 = \frac18, \displaystyle a_3 = -\frac{a_2}{12} = -\frac{1}{96}, \displaystyle a_4 = -\frac{a_3 + a_2}{16} = -\frac{11}{1536}. Now try a_0 = a_1 = 1: a_2 = 0 and a_3 = 0, and since each coefficient uses the two before it, all later ones vanish. So y = 1 + u = x is a solution, as substitution confirms: x \cdot 1 - x = 0. Its Wronskian with y_1 at 1 is -1 \ne 0, so the general solution is

\begin{aligned} y &= c_1 x + c_2\Big(1 + \frac{(x-1)^2}{8} - \frac{(x-1)^3}{96} \\ &\qquad - \frac{11(x-1)^4}{1536} + \cdots\Big). \end{aligned}

Theorem 3 promised radius at least 2; the solution x converges everywhere.

Example 4

Legendre’s equation with a polynomial solution

For (1 - x^2)\,y'' - 2x y' + 6y = 0 about 0, find the basic solutions. Show that one is a polynomial and that the other has radius exactly 1.

Solution

P = 1 - x^2 vanishes at \pm 1, so the guaranteed radius is 1. The term -x^2y'' contributes -k(k-1)a_k to the coefficient of x^k and -2xy' contributes -2k a_k. Since k(k-1) + 2k - 6 = (k - 2)(k + 3),

a_{k+2} = \frac{(k - 2)(k + 3)}{(k + 2)(k + 1)}\,a_k.

Even indices from a_0 = 1: a_2 = -3, and k = 2 makes the factor 0, so a_4 = a_6 = \dots = 0 and y_1 = 1 - 3x^2. Check: (1 - x^2)(-6) - 2x(-6x) + 6(1 - 3x^2) = 0. Odd indices from a_1 = 1: \displaystyle a_3 = -\frac{2}{3}, \displaystyle a_5 = \frac{6}{20}a_3 = -\frac15, \displaystyle a_7 = \frac{24}{42}a_5 = -\frac{4}{35}, so

y_2 = x - \frac{2}{3}x^3 - \frac{1}{5}x^5 - \frac{4}{35}x^7 - \cdots.

For odd k the factor (k - 2)(k + 3) is never 0, so no odd coefficient vanishes, and the ratio of consecutive terms \displaystyle \frac{a_{k+2}x^{k+2}}{a_k x^k} tends to x^2. The radius is exactly 1: the floor is attained. Indeed \displaystyle y_2 = \frac{3x}{4} + \frac{1 - 3x^2}{8}\ln\frac{1 + x}{1 - x}, unbounded as x \to \pm 1.

Example 5

A coefficient that is not a polynomial

Solve y'' + e^x y = 0, y(0) = 1, y'(0) = 1 through the term in x^5.

Solution

\displaystyle e^x = \sum \frac{x^k}{k!} converges everywhere, so by Theorem 2 with r = \infty the solution converges for every x. By Theorem 1(b) the x^k coefficient of e^x y is \displaystyle \sum_{j=0}^{k} \frac{a_j}{(k-j)!}, so

(k+2)(k+1)\,a_{k+2} = -\sum_{j=0}^{k} \frac{a_j}{(k-j)!}.

Every earlier coefficient feeds each new one. With a_0 = a_1 = 1:

  • k = 0: 2a_2 = -1, so \displaystyle a_2 = -\frac12;
  • k = 1: 6a_3 = -(a_0 + a_1) = -2, so \displaystyle a_3 = -\frac13;
  • k = 2: \displaystyle 12a_4 = -(\frac12 + 1 - \frac12) = -1, so \displaystyle a_4 = -\frac{1}{12};
  • k = 3: \displaystyle 20a_5 = -(\frac16 + \frac12 - \frac12 - \frac13) = \frac16, so \displaystyle a_5 = \frac{1}{120}.
\begin{aligned} y &= 1 + x - \frac{x^2}{2} - \frac{x^3}{3} \\ &\quad - \frac{x^4}{12} + \frac{x^5}{120} + \cdots \end{aligned}
Example 6

Proof: odd p and even q split the series by parity

Let p be odd and q even, with series converging for |x| < r. Prove that y_1 is even and y_2 is odd, and that the recurrence ties a_{k+2} only to coefficients whose index has the parity of k.

Solution

Let z(x) = y_1(-x) on (-r, r). By the chain rule z' = -y_1'(-x) and z'' = y_1''(-x), so

\begin{aligned} &z'' + p(x)z' + q(x)z \\ &\ = y_1''(-x) - p(x)y_1'(-x) + q(x)y_1(-x) \\ &\ = y_1''(-x) + p(-x)y_1'(-x) + q(-x)y_1(-x) = 0, \end{aligned}

the last because y_1 solves the equation at the point -x. Also z(0) = 1 and z'(0) = 0, so z = y_1 by Theorem 2(b): y_1 is even. Likewise w(x) = -y_2(-x) solves the equation with w(0) = 0 and w'(0) = 1, so w = y_2, which is odd.

For the recurrence: p_m = 0 for even m and q_m = 0 for odd m (Theorem 1(c) applied to p(x) + p(-x) and q(x) - q(-x)). So (j+1)a_{j+1}p_{k-j} can be nonzero only when k - j is odd, that is, when j + 1 has the parity of k; and a_j q_{k-j} only when j has the parity of k. \square

Boundaries

  • Singular points. There Theorem 2 says nothing; the recurrence may even force radius 0 (Exercise 14). The method of Frobenius, which tries (x - x_0)^m times a power series, handles the mildest singular points.
  • The interval is local. Exercise 15 has solutions defined on all of \mathbb{R} whose series about 0 have radius 1. To go farther, expand again about another ordinary point.
  • Partial sums are local too. Even with \rho = \infty, a fixed partial sum is accurate only near x_0 (the figure); trusting a value needs a bound on the tail, as in Exercise 10.
  • Analytic coefficients only. If p or q is merely smooth, solutions need not be analytic and the method does not apply.

Where this goes next

Legendre’s, Hermite’s, Bessel’s and Airy’s equations are solved by exactly these series (Bessel’s needs Frobenius’s extension at 0). Fourier series writes solutions as another infinite sum, suited to periodic forcing.

Exercises

Series solutions and recurrences

  1. Solve y'' + 2x y' + 4y = 0 about 0. Give the recurrence, both basic solutions, and a closed form for the odd one.
  2. Solve (1 + x^2)\,y'' + 4x y' + 2y = 0 about 0 and sum the series in closed form.
  3. For y'' + x y' - 2y = 0 about 0, find a polynomial solution, and the odd basic solution through x^7.
  4. Solve x y'' + y' - y = 0, y(1) = 1, y'(1) = 0 about x_0 = 1 through (x - 1)^4, and give an interval on which the series surely converges.
  5. Solve y'' + (\cos x)\,y = 0, y(0) = 1, y'(0) = 0 through the term in x^6.

Ordinary points and radii

  1. Without solving, give the radius Theorem 3 guarantees for series solutions of (x^2 + 9)\,y'' + x y' - 4y = 0 about x_0 = 0, x_0 = 2 and x_0 = -4.
  2. Show that 0 is an ordinary point of x y'' + x y' + (\sin x)\,y = 0 although the leading coefficient vanishes there. What radius does Theorem 2 give? Find a_2 and a_3 for y(0) = 1, y'(0) = 0.
  3. Find both basic solutions of (1 - x^2)\,y'' - 2x y' + 2y = 0 about 0, sum the even one in closed form, and give its radius.

Using the series

  1. Solve y'' + x y = 1 + x, y(0) = 0, y'(0) = 0 through the term in x^6.
  2. Let y_2 be the odd basic solution of Example 1. Show that y_2(x) = e^{-x^2/2}\int_0^x e^{s^2/2}\,ds, and compute y_2(1) to four decimal places from the series.
  3. For (1 - x^2)\,y'' - 3x y' + \lambda y = 0, find the recurrence about 0, every \lambda for which there is a polynomial solution, and that solution when \lambda = 8.

Proofs

  1. Prove that (1 - x^2)\,y'' - 2x y' + \lambda y = 0 has a polynomial solution of degree exactly n (an integer n \ge 0) if and only if \lambda = n(n+1).
  2. Prove that if \lambda \ne n(n+1) for every integer n \ge 0, every nonzero series solution of the equation in Exercise 12 about 0 has radius exactly 1.
  3. Prove that 0 is a singular point of x^2 y'' + (3x - 1)\,y' + y = 0, and that a power series satisfying it formally must be a_0\sum_{k=0}^{\infty} k!\,x^k. Conclude that the only series solution with positive radius is y = 0.
  4. Prove that every solution of (1 + x^2)\,y'' + 4x y' + 2y = 0 is defined on all of \mathbb{R}, yet every nonzero solution’s series about 0 has radius exactly 1. So the radius in Theorem 3 cannot be enlarged in general.
Answers
  1. \displaystyle a_{k+2} = -\frac{2a_k}{k+1}, since the coefficient of x^k is (k+2)(k+1)a_{k+2} + 2(k+2)a_k. So \displaystyle y_1 = \sum_{n \ge 0} \frac{(-2)^n x^{2n}}{1 \cdot 3 \cdots (2n-1)} = 1 - 2x^2 + \frac43 x^4 - \cdots (empty product 1), and \displaystyle a_{2n+1} = \frac{(-1)^n}{n!} gives y_2 = x e^{-x^2}.
  2. \displaystyle y = \frac{a_0 + a_1 x}{1 + x^2}. The coefficient of x^k is (k+2)(k+1)(a_{k+2} + a_k), so a_{k+2} = -a_k and y = (a_0 + a_1x)(1 - x^2 + x^4 - \cdots), with radius 1 (Exercise 15).
  3. y = 1 + x^2 and \displaystyle y_2 = x + \frac{x^3}{6} - \frac{x^5}{120} + \frac{x^7}{1680} - \cdots. The recurrence is \displaystyle a_{k+2} = -\frac{(k - 2)a_k}{(k+2)(k+1)}; from a_0 = 1, a_2 = 1 and a_4 = 0.
  4. \displaystyle y = 1 + \frac{(x-1)^2}{2} - \frac{(x-1)^3}{3} + \frac{7(x-1)^4}{24} + \cdots, converging at least on (0, 2). With u = x - 1 the equation is (u + 1)y'' + y' - y = 0, so (k+2)(k+1)a_{k+2} equals a_k - (k+1)^2 a_{k+1}. The only singular point is 0, at distance 1.
  5. \displaystyle y = 1 - \frac{x^2}{2} + \frac{x^4}{12} - \frac{x^6}{80} + \cdots. With \displaystyle \cos x = 1 - \frac{x^2}{2} + \frac{x^4}{24} - \cdots: 2a_2 = -1, then \displaystyle 12a_4 = -(a_2 - \frac12) = 1, then \displaystyle 30a_6 = -(a_4 - \frac{a_2}{2} + \frac{1}{24}), which is \displaystyle -\frac38. The odd coefficients are 0, as Example 6 predicts.
  6. 3, \displaystyle \sqrt{13} and 5: the zeros of x^2 + 9 are \pm 3i, and \displaystyle |2 - 3i| = \sqrt{13}, |-4 - 3i| = 5.
  7. Radius \infty; \displaystyle a_2 = -\frac12, \displaystyle a_3 = \frac16. In standard form p = 1 and \displaystyle q = \frac{\sin x}{x} = 1 - \frac{x^2}{6} + \cdots (value 1 at 0), both analytic on \mathbb{R}. Then 2a_2 + a_1 + a_0 = 0 and 6a_3 + 2a_2 + a_1 = 0.
  8. y_2 = x, and \displaystyle y_1 = 1 - x^2 - \frac{x^4}{3} - \frac{x^6}{5} - \cdots = 1 - \frac{x}{2}\ln\frac{1 + x}{1 - x}, radius 1. The recurrence is \displaystyle a_{k+2} = \frac{(k - 1)a_k}{k + 1}, so a_3 = 0 and \displaystyle a_{2n} = -\frac{1}{2n - 1} for n \ge 1, which is -x times the series of \displaystyle \frac12\ln\frac{1+x}{1-x}.
  9. \displaystyle y = \frac{x^2}{2} + \frac{x^3}{6} - \frac{x^5}{40} - \frac{x^6}{180} + \cdots. Matching x^k: (k+2)(k+1)a_{k+2} + a_{k-1} = f_k with f_0 = f_1 = 1, so \displaystyle a_2 = \frac12, \displaystyle a_3 = \frac16, a_4 = 0, \displaystyle a_5 = -\frac{a_2}{20}, \displaystyle a_6 = -\frac{a_3}{30}.
  10. y_2(1) \approx 0.7248. With g = e^{-x^2/2}\int_0^x e^{s^2/2}\,ds, the The fundamental theorem of calculus gives g' = 1 - x g and g'' = -g - x g', so g solves the equation with g(0) = 0, g'(0) = 1, and g = y_2 by Theorem 2(b). The series at 1 is \displaystyle 1 - \frac13 + \frac{1}{15} - \frac{1}{105} + \frac{1}{945} - \frac{1}{10{,}395} + \cdots, and these six terms sum to 0.724772. It alternates with decreasing terms, so y_2(1) lies between it and \displaystyle 0.724772 + \frac{1}{135{,}135} < 0.724780.
  11. \displaystyle a_{k+2} = \frac{k(k+2) - \lambda}{(k+2)(k+1)}\,a_k; polynomial solutions exactly for \lambda = n(n+2), n \ge 0 an integer; for \lambda = 8, y = 1 - 4x^2. If \lambda = n(n+2), start from a_0 or a_1 by the parity of n, and a_{n+2} = 0 ends the series. Conversely, for a polynomial of degree n the x^n coefficient of the left side is (\lambda - n(n+2))a_n.
  12. If y has degree n and leading coefficient c \ne 0, then, since 1 \cdot y'' has degree below n, the x^n coefficient of the left side is c\big(\lambda - n(n-1) - 2n\big), that is, c\big(\lambda - n(n+1)\big); so \lambda = n(n+1). Conversely, with \lambda = n(n+1) the recurrence is \displaystyle a_{k+2} = \frac{k(k+1) - n(n+1)}{(k+2)(k+1)}a_k. Start from a_0 = 1 or a_1 = 1 (the other 0) by the parity of n. The coefficients of the other parity are all 0; those of n’s parity are nonzero up to a_n, because k(k+1) \ne n(n+1) for 0 \le k < n, and a_{n+2} = 0 makes all later ones 0. So y is a polynomial of degree exactly n.
  13. Theorem 3 gives radius at least 1, since 1 - x^2 vanishes only at \pm 1. Let y = \sum a_k x^k be nonzero; say a_0 \ne 0 (else a_1 \ne 0, and the odd case is the same). Since k(k+1) \ne \lambda for every k, the recurrence makes every a_{2n} \ne 0, and the ratio of the terms a_{2n+2}x^{2n+2} and a_{2n}x^{2n} has absolute value \displaystyle \frac{|2n(2n+1) - \lambda|}{(2n+2)(2n+1)}\,x^2, which tends to x^2. For |x| > 1 these terms eventually grow in absolute value, so they do not tend to 0, and \sum a_k x^k diverges. The radius is exactly 1.
  14. P = x^2, Q = 3x - 1, R = 1 have no common factor and P(0) = 0, so 0 is singular. The coefficient of x^k in the left side is k(k-1)a_k + 3k a_k + a_k - (k+1)a_{k+1}, which equals (k+1)\big((k+1)a_k - a_{k+1}\big). So a_{k+1} = (k+1)a_k and a_k = k!\,a_0. If a_0 \ne 0, the ratio of consecutive terms has absolute value (k+1)|x| \to \infty for every x \ne 0, so the radius is 0. A series solution with positive radius must then have a_0 = 0, hence every a_k = 0.
  15. Expanding, \big((1 + x^2)y\big)'' equals (1 + x^2)y'' + 4xy' + 2y. So y is a solution on an interval exactly when (1 + x^2)y = c_0 + c_1x there, and every solution is the restriction of \displaystyle \frac{c_0 + c_1x}{1 + x^2}, defined on \mathbb{R}. By Exercise 2 its series about 0 has a_{2n} = (-1)^n c_0 and a_{2n+1} = (-1)^n c_1. If (c_0, c_1) \ne (0, 0), these coefficients do not tend to 0, so for |x| \ge 1 the terms a_k x^k do not tend to 0 and the series diverges; with Theorem 3 the radius is exactly 1.
math.wiki · CC-BY-SA-4.0