Unit 7, lesson 37

Power-Series Solutions at Ordinary Points

How to solve a linear equation near an ordinary point with a power series, shift indices so the powers match, derive a coefficient recurrence and let the two initial coefficients fix the solution.

  • Video 37
  • Power-Series and Frobenius Solutions

Take the equation y′′+2xy′=0y'' + 2xy' = 0, where xx is the independent variable and y′=dy/dxy' = dy/dx, with the data y(0)=0y(0) = 0 and y′(0)=1y'(0) = 1. The coefficient 2x2x changes with xx, so the exponential trial erxe^{rx} that works for constant coefficients is of no use here. Which function solves this problem? Writing yy as a power series about 00 turns the equation into a rule that produces the coefficients one at a time. This lesson explains when that approach is justified, how to line up the powers, and how the two starting values pick out one solution, which here turns out to be ∫0xe−u2 du\int_0^x e^{-u^2}\,du.

Key terms

Ordinary point

Definition Write a second-order linear homogeneous equation in the independent variable xx in normalized form,

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

A function is analytic at x0x_0 if it equals a power series in x−x0x - x_0 that converges on some open interval around x0x_0. The point x0x_0 is an ordinary point of the equation if pp and qq are both analytic at x0x_0, and a singular point otherwise. For an equation P(x) y′′+Q(x) y′+R(x) y=0P(x)\,y'' + Q(x)\,y' + R(x)\,y = 0 whose coefficients are polynomials with no common factor, the normalized coefficients are p=Q/Pp = Q/P and q=R/Pq = R/P, and x0x_0 is an ordinary point exactly when P(x0)≠0P(x_0) \ne 0.

In plain words Near an ordinary point, the coefficients of the normalized equation are themselves given by power series. Nothing in the equation breaks down there, and its solutions can be built as power series too.

Example In y′′+2xy′=0y'' + 2xy' = 0 the coefficients p(x)=2xp(x) = 2x and q(x)=0q(x) = 0 are polynomials, so every real number is an ordinary point. For (1−x2) y′′−2x y′+6y=0(1 - x^2)\,y'' - 2x\,y' + 6y = 0 the leading coefficient vanishes only at x=1x = 1 and x=−1x = -1. So x0=0x_0 = 0 is an ordinary point, while 11 and −1-1 are singular points, where the normalized coefficients

p(x)=−2x1−x2,q(x)=61−x2\begin{gathered} p(x) = \frac{-2x}{1 - x^2}, \\[0.4em] q(x) = \frac{6}{1 - x^2} \end{gathered}

are not even defined.

Power-series solution

Definition A power-series solution about x0x_0 is a solution of the form

y=∑n=0∞an(x−x0)ny = \sum_{n=0}^{\infty} a_n (x - x_0)^n

whose series converges on an open interval ∣x−x0∣<ρ\lvert x - x_0 \rvert < \rho with ρ>0\rho > 0 and satisfies the equation there. Inside that interval the series can be differentiated term by term as often as needed, and the differentiated series converge on the same interval. The first derivative is

y′=∑n=1∞n an(x−x0)n−1,y' = \sum_{n=1}^{\infty} n\,a_n (x - x_0)^{n-1},

and differentiating once more gives a series for y′′y'' that starts at n=2n = 2. It follows that the coefficients are the Taylor coefficients an=y(n)(x0)/n!a_n = y^{(n)}(x_0)/n!. Substituting the trial series into the equation and matching powers is a formal computation, and its result is a solution once the series is known to converge. At an ordinary point this is guaranteed. For every choice of a0a_0 and a1a_1 there is exactly one power-series solution about x0x_0 with these first two coefficients, and its series converges on some interval around x0x_0. The next lesson shows how large that interval is guaranteed to be.

In plain words Instead of guessing a formula for yy, treat its Taylor coefficients as the unknowns and let the equation decide them.

Example For y′′+2xy′=0y'' + 2xy' = 0 with y(0)=0y(0) = 0 and y′(0)=1y'(0) = 1, the steps in the cards below give

y=∑k=0∞(−1)kx2k+1k! (2k+1)=x−x33+x510−⋯ .\begin{aligned} y &= \sum_{k=0}^{\infty} \frac{(-1)^k x^{2k+1}}{k!\,(2k+1)} \\[0.4em] &= x - \frac{x^3}{3} + \frac{x^5}{10} - \cdots. \end{aligned}

Differentiating term by term cancels each factor 2k+12k + 1 and leaves the exponential series with −x2-x^2 in place of the variable,

y′=∑k=0∞(−1)kx2kk!=e−x2,y' = \sum_{k=0}^{\infty} \frac{(-1)^k x^{2k}}{k!} = e^{-x^2},

which converges for every real xx. Together with y(0)=0y(0) = 0 this identifies the solution as

y(x)=∫0xe−u2 du,y(x) = \int_0^x e^{-u^2}\,du,

which also equals π2erf⁡(x)\tfrac{\sqrt{\pi}}{2}\operatorname{erf}(x) in terms of the error function. A direct check confirms it. The fundamental theorem of calculus gives y′=e−x2y' = e^{-x^2}, then y′′=−2xe−x2y'' = -2xe^{-x^2}, so y′′+2xy′=0y'' + 2xy' = 0 for every real xx, and the integral has value 00 and slope 11 at x=0x = 0.

Index shift

Definition An index shift relabels the terms of a sum so that it shows a chosen power. Substituting n=m+kn = m + k in a sum that starts at n=kn = k gives

∑n=k∞cn(x−x0)n−k=∑m=0∞cm+k(x−x0)m.\begin{aligned} &\sum_{n=k}^{\infty} c_n (x - x_0)^{n-k} \\[0.4em] &\qquad = \sum_{m=0}^{\infty} c_{m+k} (x - x_0)^{m}. \end{aligned}

Both sides list the same terms in the same order. The index of the coefficient and the lower limit change together with the exponent, and afterward the index may be called nn again. Terms that are zero, such as a term with a factor nn at n=0n = 0, may be added or dropped so that all sums start at the same index.

In plain words Shifting the index is bookkeeping. It changes how the terms are labeled, never which terms are there, and it lets every sum in the equation talk about the same power.

Example With y=∑anxny = \sum a_n x^n about 00, the second derivative is

y′′=∑n=2∞n(n−1) anxn−2.y'' = \sum_{n=2}^{\infty} n(n-1)\,a_n x^{n-2}.

Replacing nn by n+2n + 2 throughout writes the same sum as

∑n=0∞(n+2)(n+1) an+2 xn.\sum_{n=0}^{\infty} (n+2)(n+1)\,a_{n+2}\,x^{n}.

In the other term, the factor xx raises each power of y′y' by one, so no shift is needed,

2xy′=∑n=1∞2n anxn=∑n=0∞2n anxn,\begin{aligned} 2xy' &= \sum_{n=1}^{\infty} 2n\,a_n x^{n} \\[0.4em] &= \sum_{n=0}^{\infty} 2n\,a_n x^{n}, \end{aligned}

where the second line adds the term for n=0n = 0, which is zero. A common slip changes the exponent but not the coefficient, writing ∑n(n−1) anxn\sum n(n-1)\,a_n x^n for y′′y''. That sum has constant term 00, while the true constant term of y′′y'' is 2a22a_2.

Coefficient recurrence

Definition Once every sum in the equation is written with the power (x−x0)n(x - x_0)^n and starts at the same index, the equation reads ∑cn(x−x0)n=0\sum c_n (x - x_0)^n = 0 on an interval around x0x_0, where each cnc_n is a combination of the unknown coefficients. A power series that is zero at every point of an open interval around its center has all its coefficients zero, so cn=0c_n = 0 for every nn. Solving cn=0c_n = 0 for the coefficient with the highest index gives the coefficient recurrence, a formula for a later coefficient in terms of earlier ones. Low powers that do not fit the general pattern give separate equations, which must be solved one by one.

In plain words A power series is zero on an interval only if every one of its coefficients is zero. Setting each coefficient of the equation to zero gives a rule that builds the solution one coefficient at a time.

Example Adding the two shifted sums for y′′+2xy′=0y'' + 2xy' = 0 gives (n+2)(n+1) an+2+2n an=0(n+2)(n+1)\,a_{n+2} + 2n\,a_n = 0 for every n≥0n \ge 0, so

an+2=−2n an(n+2)(n+1).a_{n+2} = \frac{-2n\,a_n}{(n+2)(n+1)}.

At n=0n = 0 the recurrence gives a2=0a_2 = 0, whatever a0a_0 is. At n=1n = 1 it gives a3=−26a1=−13a1a_3 = -\tfrac{2}{6}a_1 = -\tfrac13 a_1, and at n=3n = 3 it gives a5=−620a3a_5 = -\tfrac{6}{20}a_3.

Initial coefficients

Definition At an ordinary point x0x_0 of a second-order equation, the recurrence leaves a0a_0 and a1a_1 free and determines every later coefficient from them. These two initial coefficients carry the initial data,

a0=y(x0),a1=y′(x0),a_0 = y(x_0), \quad a_1 = y'(x_0),

so the initial-value problem with y(x0)=k0y(x_0) = k_0 and y′(x0)=k1y'(x_0) = k_1 is solved by the series with a0=k0a_0 = k_0 and a1=k1a_1 = k_1. Let y1y_1 be the series solution with (a0,a1)=(1,0)(a_0, a_1) = (1, 0) and y2y_2 the one with (a0,a1)=(0,1)(a_0, a_1) = (0, 1). On an interval around x0x_0 where both series converge, every solution is

y=a0 y1+a1 y2,y = a_0\,y_1 + a_1\,y_2,

and the pair is a fundamental set, since its Wronskian at x0x_0 is y1(x0) y2′(x0)−y1′(x0) y2(x0)=1y_1(x_0)\,y_2'(x_0) - y_1'(x_0)\,y_2(x_0) = 1.

In plain words A second-order equation needs two numbers to fix a solution. In the series method those two numbers are simply the first two coefficients, the value and the slope at the center.

Example For y′′+2xy′=0y'' + 2xy' = 0, the data y(0)=0y(0) = 0 and y′(0)=1y'(0) = 1 give a0=0a_0 = 0 and a1=1a_1 = 1. The recurrence then produces a3=−13a_3 = -\tfrac13, a5=110a_5 = \tfrac{1}{10} and a7=−142a_7 = -\tfrac{1}{42}, while every even coefficient is zero. The two basic choices give y1=1y_1 = 1, because (a0,a1)=(1,0)(a_0, a_1) = (1, 0) makes every coefficient after a0a_0 zero, and y2=∫0xe−u2 duy_2 = \int_0^x e^{-u^2}\,du, the solution just found. Both series converge for every real xx, so the general solution on the whole real line is

y=C0+C1∫0xe−u2 du,y = C_0 + C_1\int_0^x e^{-u^2}\,du,

and the data above select C0=0C_0 = 0 and C1=1C_1 = 1.

Even and odd coefficient chains

Definition When the recurrence expresses an+2a_{n+2} through ana_n alone, the coefficients split into two chains that never interact. The even chain a0,a2,a4,…a_0, a_2, a_4, \dots consists of multiples of a0a_0, and the odd chain a1,a3,a5,…a_1, a_3, a_5, \dots consists of multiples of a1a_1. The solution then splits as y=a0 y1+a1 y2y = a_0\,y_1 + a_1\,y_2, where y1y_1 contains only even powers of x−x0x - x_0 and y2y_2 only odd powers. If the factor in front of ana_n vanishes at some index, every later coefficient of that chain is zero and the chain sums to a polynomial. A recurrence that links indices three apart, or one that involves more than two coefficients, splits differently or not at all, so the chains have to be read off the recurrence of the equation at hand.

In plain words A two-step recurrence hops over every other coefficient. One chain of hops starts at a0a_0 and the other at a1a_1, so the value and the slope at the center each control their own half of the series.

Example For y′′+2xy′=0y'' + 2xy' = 0 the even chain starts with a2=0⋅a0=0a_2 = 0 \cdot a_0 = 0, so a4a_4, a6a_6 and all later even coefficients vanish, and the even part of every solution is the constant a0a_0. The odd chain with a1=1a_1 = 1 follows the formula

a2k+1=(−1)kk! (2k+1)a_{2k+1} = \frac{(-1)^k}{k!\,(2k+1)}

for every k≥0k \ge 0. A direct check confirms it. With n=2k+1n = 2k + 1 the recurrence multiplies this coefficient by −2(2k+1)/((2k+3)(2k+2))-2(2k + 1)/((2k + 3)(2k + 2)), and the product simplifies to (−1)k+1/((k+1)! (2k+3))(-1)^{k+1}/((k+1)!\,(2k + 3)), which is the same formula with k+1k + 1 in place of kk.

Check yourself

  1. Use a power series about 00 to solve y′′−y=0y'' - y = 0 with y(0)=1y(0) = 1 and y′(0)=1y'(0) = 1. Which function is it?
Show the answer

Write y=∑anxny = \sum a_n x^n. The shifted second derivative gives (n+2)(n+1) an+2−an=0(n + 2)(n + 1)\,a_{n+2} - a_n = 0 for every n≥0n \ge 0, so

an+2=an(n+2)(n+1).a_{n+2} = \frac{a_n}{(n+2)(n+1)}.

The data give a0=1a_0 = 1 and a1=1a_1 = 1. If an=1/n!a_n = 1/n!, the recurrence gives an+2=1/(n+2)!a_{n+2} = 1/(n+2)!, so by induction along both chains an=1/n!a_n = 1/n! for every nn and

y=∑n=0∞xnn!=ex.y = \sum_{n=0}^{\infty} \frac{x^n}{n!} = e^x.

The even chain alone sums to cosh⁡x\cosh x and the odd chain alone to sinh⁡x\sinh x, so the solution is cosh⁡x+sinh⁡x=ex\cosh x + \sinh x = e^x, and indeed (ex)′′−ex=0(e^x)'' - e^x = 0.

  1. For y′′+2xy′=0y'' + 2xy' = 0 with y(0)=0y(0) = 0 and y′(0)=1y'(0) = 1, a classmate keeps the first two terms of the series and claims that y=x−x3/3y = x - x^3/3 solves the problem. Is the claim right?
Show the answer

No. For this polynomial y′=1−x2y' = 1 - x^2 and y′′=−2xy'' = -2x, so y′′+2xy′=−2x+2x−2x3y'' + 2xy' = -2x + 2x - 2x^3. The terms in xx cancel and leave

y′′+2xy′=−2x3.y'' + 2xy' = -2x^3.

The polynomial has the right value and slope at 00, and the equation is satisfied through degree two, but the residual −2x3-2x^3 is not zero, so it is only an approximation near 00. Adding terms pushes the residual to higher powers. With x5/10x^5/10 added it becomes x5x^5, and with −x7/42-x^7/42 also added it becomes −x7/3-x^7/3. Only the full series, whose sum is ∫0xe−u2 du\int_0^x e^{-u^2}\,du, makes the residual vanish.

  1. Explain why x0=0x_0 = 0 is an ordinary point of Airy’s equation y′′−xy=0y'' - xy = 0. Find its coefficient recurrence and the solution with y(0)=1y(0) = 1 and y′(0)=0y'(0) = 0 through the term in x6x^6. Do the coefficients split into even and odd chains?
Show the answer

The normalized coefficients p(x)=0p(x) = 0 and q(x)=−xq(x) = -x are polynomials, so every point is ordinary. With y=∑anxny = \sum a_n x^n, a shift gives

xy=∑n=0∞anxn+1=∑n=1∞an−1xn.\begin{aligned} xy &= \sum_{n=0}^{\infty} a_n x^{n+1} \\[0.4em] &= \sum_{n=1}^{\infty} a_{n-1} x^{n}. \end{aligned}

The constant term of the equation is 2a2=02a_2 = 0, so a2=0a_2 = 0. For n≥1n \ge 1 the coefficient of xnx^n gives

an+2=an−1(n+2)(n+1).a_{n+2} = \frac{a_{n-1}}{(n+2)(n+1)}.

This recurrence links indices three apart, so the chains are a0,a3,a6,…a_0, a_3, a_6, \dots and a1,a4,a7,…a_1, a_4, a_7, \dots and a2,a5,a8,…a_2, a_5, a_8, \dots, and the last one is zero because a2=0a_2 = 0. With a0=1a_0 = 1 and a1=0a_1 = 0 the recurrence gives a3=16a_3 = \tfrac16, a6=16⋅30=1180a_6 = \tfrac{1}{6 \cdot 30} = \tfrac{1}{180}, and a4=a5=0a_4 = a_5 = 0, so

y=1+x36+x6180+⋯ .y = 1 + \frac{x^3}{6} + \frac{x^6}{180} + \cdots.

The chains are not even and odd, since this one solution already mixes the odd power x3x^3 with the even power x6x^6.

How this lesson connects

It also uses power series and term-by-term differentiation.