Differential Equations › Section F · Series solutions › Topic F.2
Series solutions near ordinary points
When the coefficients of an equation vary with x, you can still build the solution one coefficient at a time as a power series. Near an ordinary point this always works, and the equation itself tells you how far the series can be trusted.
Solving one coefficient at a time
Take a second order linear equation whose coefficients are polynomials, such as \((1 + x^2)y'' + 2xy' - 2y = 0\). No trial exponential works, because the coefficients change with \(x\). Instead, assume the solution is a power series \(y = \sum c_n (x - a)^n\), substitute it, and use the fact from the topic Review: power series and Taylor series that a series equal to zero has every coefficient zero. That gives a formula linking the coefficients, and the solution emerges term by term. First, though, you need to know where the method is guaranteed to work.
Ordinary and singular points
Write the equation as \(p(x)y'' + q(x)y' + r(x)y = 0\) and divide by \(p\) to get \(y'' + P(x)y' + Q(x)y = 0\), with \(P = \frac{q}{p}\) and \(Q = \frac{r}{p}\). A point \(x = a\) is an ordinary point if \(P\) and \(Q\) are both analytic at \(a\); otherwise it is a singular point. When \(p\), \(q\) and \(r\) are polynomials with no common factor, the test is quick: \(a\) is ordinary exactly when \(p(a) \ne 0\). So for \((x^2 - 4)y'' + xy' + y = 0\) the singular points are \(\pm 2\), and every other point, real or complex, is ordinary.
What is guaranteed
If \(a\) is an ordinary point, the general solution can be written
where \(y_1\) and \(y_2\) are two independent power series solutions. The constants \(c_0 = y(a)\) and \(c_1 = y'(a)\) are free, which matches the two constants every second order equation needs; all later coefficients are forced. Both series converge at least for \(|x - a| < \rho\), where \(\rho\) is the distance from \(a\) to the nearest singular point, measured in the complex plane. The true radius can be larger, never smaller.
The method in five steps
1. Check that \(a\) is an ordinary point. 2. Write \(y\), \(y'\) and \(y''\) as series, substitute, and move every coefficient inside its sum. 3. Shift indices so every sum carries the power \((x - a)^n\), then strip out early terms until every sum starts at the same \(n\). 4. Set each coefficient to zero. The general one is the recurrence relation, which gives a later coefficient from earlier ones; record the values of \(n\) for which it holds. 5. Starting from \(c_0\) and \(c_1\), generate coefficients and collect the terms containing \(c_0\) into \(y_1\) and those containing \(c_1\) into \(y_2\).
Worked example: a recurrence two steps apart
Solve \(y'' + xy' + y = 0\) about \(a = 0\). The leading coefficient is 1, so every point is ordinary. 1. Substitute: \(xy' = \sum n c_n x^n\), and \(y''\) shifts to \(\sum (n+2)(n+1)c_{n+2}x^n\), all from \(n = 0\). 2. The coefficient of \(x^n\) is \((n+2)(n+1)c_{n+2} + (n+1)c_n\). 3. Setting it to zero and cancelling \(n + 1\):
4. Even terms: \(c_2 = -\frac{c_0}{2}\), \(c_4 = \frac{c_0}{8}\), \(c_6 = -\frac{c_0}{48}\). Odd terms: \(c_3 = -\frac{c_1}{3}\), \(c_5 = \frac{c_1}{15}\), \(c_7 = -\frac{c_1}{105}\). So
Recognising the result
Because the recurrence jumps by two, the even and odd coefficients never mix, which is why the solution splits neatly. Here the even pattern has a closed form,
which is the series for \(e^u\) with \(u = -\frac{x^2}{2}\) substituted, so \(y_1 = e^{-x^2/2}\). Check it directly: if \(y = e^{-x^2/2}\), then \(y' = -xy\) and \(y'' = (x^2 - 1)y\), so the left side is \((x^2 - 1 - x^2 + 1)y = 0\). With \(y(0) = 1\) and \(y'(0) = 0\), the solution is exactly \(e^{-x^2/2}\). The same method applied to \(y'' + y = 0\) gives \(c_{n+2} = -\frac{c_n}{(n+2)(n+1)}\), which rebuilds the series for \(\cos x\) and \(\sin x\). Usually no closed form appears, and the series itself is the answer.
Worked example: a non-constant leading coefficient
Solve \((1 + x^2)y'' + 2xy' - 2y = 0\) about 0, an ordinary point since \(1 + 0^2 \ne 0\). 1. Split \((1 + x^2)y''\) into \(y'' + x^2 y''\); the second part is \(\sum n(n-1)c_n x^n\). 2. Shift \(y''\) as before; every sum can now start at \(n = 0\), because the added terms are zero. 3. Collecting \(x^n\): \(n(n-1) + 2n - 2 = (n+2)(n-1)\), so the coefficient is \((n+2)\) times \((n+1)c_{n+2} + (n-1)c_n\), giving
4. Odd terms: \(n = 1\) gives \(c_3 = 0\), so every later odd coefficient is 0 and \(y_2 = x\) exactly. 5. Even terms: \(c_2 = c_0\), \(c_4 = -\frac{c_0}{3}\), \(c_6 = \frac{c_0}{5}\), so
How far the series reaches
In the last example \(1 + x^2\) is never zero for real \(x\), yet it vanishes at \(\pm i\), one unit from 0. The theorem therefore promises convergence for \(|x| < 1\), and that is exactly the radius of the arctangent series. The function \(1 + x\arctan x\) exists for every \(x\), but this series stops representing it beyond 1, so a series answer should always state where it is valid. Legendre's equation,
is singular at \(\pm 1\), giving radius at least 1 about 0; when \(\ell\) is a whole number one series stops, and the topic Special functions in applications develops these Legendre polynomials.
Airy's equation
Some recurrences link coefficients three steps apart. For Airy's equation \(y'' - xy = 0\) about 0, the \(x^0\) coefficient gives \(2c_2 = 0\), and for \(n \ge 1\)
So \(c_3 = \frac{c_0}{6}\), \(c_4 = \frac{c_1}{12}\), \(c_5 = 0\), \(c_6 = \frac{c_0}{180}\) and \(c_7 = \frac{c_1}{504}\): every third coefficient vanishes. There are no singular points, so both series converge for all \(x\). Their sums, the Airy functions, have no formula in elementary functions, yet they describe light near the bright edge of a rainbow and a quantum particle in a uniform force field. The series is their definition.
Common mistakes to avoid
First, combining sums before their powers and starting indices match: shift first, then strip out terms. Second, forgetting the low-order equations that sit outside the general recurrence, such as \(2c_2 = 0\) for Airy's equation; dropping them gives wrong coefficients. Third, treating \(c_0\) and \(c_1\) as unknowns to be solved for: they are free, and initial conditions fix them as \(y(a)\) and \(y'(a)\). Fourth, judging the radius only by real singular points: \(1 + x^2\) has complex zeros that cap the radius at 1. Fifth, using this method at a singular point, where it can fail.
Key takeaways
For \(p y'' + q y' + r y = 0\) with polynomial coefficients and no common factor, \(a\) is an ordinary point when \(p(a) \ne 0\). There the general solution is a power series with \(c_0 = y(a)\) and \(c_1 = y'(a)\) free, splitting into two independent series. Substitute, shift indices, strip out early terms and set coefficients to zero to get a recurrence. The series converges at least out to the nearest singular point, complex ones included. Sometimes the series is a known function, like \(e^{-x^2/2}\) or \(1 + x\arctan x\); often, as for Airy's equation, the series is the function.
Flashcards
Flashcard
y″ + P(x)y′ + Q(x)y = 0 with P = q/p, Q = r/p
Test your understanding
Question 1 of 5
What are the singular points of (x² - 4)y″ + xy′ + y = 0?
Question 2 of 5
Substituting y = Σ cₙxⁿ into y″ + xy′ + y = 0 gives which recurrence?
Question 3 of 5
For y″ + xy′ + y = 0 with y(0) = 2 and y′(0) = 0, what is c₄, the coefficient of x⁴?
Question 4 of 5
Series solutions of (x² + 9)y″ + y = 0 about x = 0 are guaranteed to converge for |x| less than what?
Question 5 of 5
In a series solution about an ordinary point a, what are c₀ and c₁?
