Differential Equations › Section F · Series solutions › Topic F.1

Review: power series and Taylor series

Equations with variable coefficients usually have no solution built from familiar functions, but they can often be solved one coefficient at a time as a power series. This review gathers the series skills that method depends on, with extra practice in the bookkeeping it needs.

Why series come back now

Every method so far relied on constant coefficients, where a trial exponential turns the equation into algebra. An equation such as \((1 + x^2)y'' + 2xy' - 2y = 0\) gives that trick nothing to grip. Instead, assume the solution is a power series and work out its coefficients. That needs four calculus skills: radius of convergence, term-by-term differentiation, lining up powers in sums, and Taylor series. The calculus topics Power series, Representing functions as power series, and Taylor and Maclaurin series taught them; here they are tuned for differential equations.

Power series and the radius

A power series with centre \(a\) is

\sum_{n=0}^{\infty} c_n (x - a)^n = c_0 + c_1(x - a) + c_2(x - a)^2 + \cdots

It converges absolutely for \(|x - a| < R\) and diverges for \(|x - a| > R\), where \(R\) is the radius of convergence, usually found with the ratio test. For \(\sum \frac{2^n (x-1)^n}{n+1}\) the ratio tends to \(2|x - 1|\), so \(R = \frac{1}{2}\); checking the endpoints gives the interval \([\frac{1}{2}, \frac{3}{2})\). This matters for differential equations because a series solution is only a solution where the series converges.

Adding series and moving factors inside

Two series can be merged into one sum, by simply adding coefficients, only when they start at the same index and carry the same power of \(x - a\). A factor in front of a sum can be moved inside by adding exponents,

x^2 \sum_{n=0}^{\infty} c_n x^n = \sum_{n=0}^{\infty} c_n x^{n+2}

which works directly when the factor is a constant or a power of \(x - a\). With centre 0 a factor \(x\) is fine. With centre \(a = 2\), first write \(x\) as \((x - 2) + 2\), which splits one sum into two.

Differentiating term by term

Inside its radius, a power series can be differentiated term by term, and the radius does not change:

y' = \sum_{n=1}^{\infty} n c_n (x-a)^{n-1}, \qquad y'' = \sum_{n=2}^{\infty} n(n-1) c_n (x-a)^{n-2}

The constant term disappears in \(y'\), so that sum starts at \(n = 1\), and \(y''\) loses two terms, so it starts at \(n = 2\). Because the missing terms are exactly zero, you may start these sums lower, at \(n = 0\), without changing their value. That freedom is often the quickest way to make several sums start at the same place.

Shifting the index

To combine sums you usually want every power written as \((x - a)^n\). An index shift does this without changing the series. The rule: if you lower the starting value by \(k\), raise every \(n\) inside the sum by \(k\), and the reverse. For example, lowering the start of \(y''\) by 2 gives

\sum_{n=2}^{\infty} n(n-1) c_n x^{n-2} = \sum_{n=0}^{\infty} (n+2)(n+1) c_{n+2} x^n

Write out the first terms of each side to check: both begin \(2c_2 + 6c_3 x + \cdots\). This one check catches nearly every shifting slip.

Worked example: one single series

Write

\sum_{n=2}^{\infty} n(n-1)c_n x^{n-2} + x^2 \sum_{n=0}^{\infty} c_n x^n

as one series in powers \(x^n\). 1. Move the \(x^2\) inside: the second sum becomes \(\sum c_n x^{n+2}\), from \(n = 0\). 2. Shift the first sum down by 2, as in the previous card, so it starts at \(n = 0\) with general term \((n+2)(n+1)c_{n+2}x^n\). 3. Shift the second sum up by 2: its general term becomes \(c_{n-2} x^n\), from \(n = 2\). 4. The first sum starts at 0 and the second at 2, so strip the \(n = 0\) and \(n = 1\) terms out of the first, \(2c_2\) and \(6c_3 x\). The result:

2c_2 + 6c_3 x + \sum_{n=2}^{\infty} \big[(n+2)(n+1)c_{n+2} + c_{n-2}\big] x^n

When a series is zero everywhere

If \(\sum c_n (x - a)^n = 0\) for every \(x\) in an interval around \(a\), then every coefficient is zero. Put \(x = a\) to get \(c_0 = 0\), differentiate and put \(x = a\) again to get \(c_1 = 0\), and so on. The series in the worked example is exactly \(y'' + x^2 y\) for \(y = \sum c_n x^n\). Setting it equal to zero forces \(c_2 = 0\), \(c_3 = 0\), and

(n+2)(n+1)c_{n+2} + c_{n-2} = 0 \quad (n \ge 2)

One equation has become infinitely many simple ones; solving them is the next topic.

Taylor series, briefly

If \(f\) has derivatives of every order at \(a\), its Taylor series about \(a\) is

\sum_{n=0}^{\infty} \frac{f^{(n)}(a)}{n!}(x - a)^n

and with \(a = 0\) it is a Maclaurin series. The library from calculus still does most of the work: \(e^x\), \(\sin x\) and \(\cos x\) for every \(x\), and \(\frac{1}{1-x} = \sum x^n\) for \(|x| < 1\). A new centre needs only new derivative values. Every derivative of \(e^x\) at 3 equals \(e^3\), so

e^x = e^3 \sum_{n=0}^{\infty} \frac{(x - 3)^n}{n!}

valid for all \(x\).

Worked example: a polynomial about a new centre

Write \(p(x) = x^3 - 3x^2 + 4\) as a Taylor series about \(a = 2\). 1. \(p(2) = 8 - 12 + 4 = 0\). 2. \(p'(x) = 3x^2 - 6x\), so \(p'(2) = 0\). 3. \(p''(x) = 6x - 6\), so \(p''(2) = 6\), and \(p'''(x) = 6\); higher derivatives are 0. 4. Divide by factorials: \(\frac{6}{2!} = 3\) and \(\frac{6}{3!} = 1\), so \(p(x) = 3(x-2)^2 + (x-2)^3\). The series stops, as it must for a polynomial. Check: the cube expands to \(x^3 - 6x^2 + 12x - 8\), and adding \(3x^2 - 12x + 12\) returns \(p(x)\). You need this move when a coefficient such as \(x^2\) must be rewritten about a centre other than 0.

Analytic functions

A function is analytic at \(a\) if its Taylor series about \(a\) has a positive radius and equals the function near \(a\). Polynomials, \(e^x\), \(\sin x\) and \(\cos x\) are analytic everywhere. A rational function in lowest terms is analytic wherever its denominator is not zero, and its radius about \(a\) is the distance from \(a\) to the nearest zero of the denominator, complex zeros included. So \(\frac{1}{1+x^2}\), smooth on the whole real line, has radius 1 about 0, because its denominator vanishes at \(\pm i\). The next topic uses this to predict where series solutions converge.

Common mistakes to avoid

First, adding sums that start at different indices or carry different powers: line up both the power and the start before you combine. Second, shifting the start without shifting the inside, or shifting the inside the same way as the start; the check is to write out two terms of each version. Third, moving a factor \(x\) inside a series centred at \(a \ne 0\) without first writing it as \((x - a) + a\). Fourth, dropping the factorial in a Taylor coefficient: the coefficient of \((x - 2)^2\) above is 3, not 6. Fifth, assuming a function smooth on the real line has an infinite radius.

Key takeaways

A power series \(\sum c_n (x - a)^n\) converges inside its radius \(R\), and a series solution means something only there. Inside the radius you may differentiate term by term. To combine sums, move factors inside, shift indices (lower the start by \(k\), raise each \(n\) by \(k\)), and strip out early terms until every sum has the same start and power. A series that is zero on an interval has every coefficient zero, which turns an equation into a recurrence. Taylor series, analytic functions and complex zeros of denominators complete the toolkit.

Flashcards

Flashcard

Power series

Σ cₙ(x - a)ⁿ with centre a

Same notation as the calculus Power series topic.

Test your understanding

Question 1 of 5

Which series equals Σ n(n - 1)cₙxⁿ⁻² summed from n = 2?

Question 2 of 5

What is the radius of convergence of Σ 2ⁿ(x - 1)ⁿ/(n + 1)?

Question 3 of 5

What is the Taylor series of x³ - 3x² + 4 about a = 2?

Question 4 of 5

If Σ cₙxⁿ = 0 for every x in an interval around 0, what follows?

Question 5 of 5

Why is the radius of the Maclaurin series of 1/(1 + x²) equal to 1, even though the function is smooth for all real x?

← Previous topicNext topic →