Series Solutions

Sometimes the tools we have developed fail us. Separation of variables requires a special structure. The characteristic equation works only for constant coefficients. Integrating factors demand linearity in a particular form. What happens when we encounter an equation like $y'' - xy = 0$? The variable coefficient $x$ defeats our usual approaches.

Power series provide a way forward. The idea is simple: assume the solution can be written as an infinite polynomial, substitute into the equation, and determine the coefficients. This method works for a remarkably broad class of differential equations, including many that resist every other technique.

The Core Idea

A power series centered at $x = 0$ has the form:

The coefficients $a_0, a_1, a_2, \ldots$ are constants to be determined. If we can find values for all these coefficients, we have our solution.

Why might this work? Many familiar functions have power series representations. The exponential, sine, and cosine all arise this way:

These series converge for all values of $x$. Other functions have series that converge only in some interval. The power series approach lets us discover solutions we might never have guessed, expressed as infinite series that we can evaluate to any desired precision.

A Motivating Example

Consider the harmonic oscillator equation:

We know the solutions are $\sin x$ and $\cos x$. But suppose we did not know this. Let us rediscover these solutions using power series.

Assume $y = \sum_{n=0}^{\infty} a_n x^n$. Then:

• $y' = \sum_{n=1}^{\infty} n a_n x^{n-1}$
• $y'' = \sum_{n=2}^{\infty} n(n-1) a_n x^{n-2}$

Substituting into $y'' + y = 0$:

To combine these sums, we need the powers of $x$ to match. In the first sum, let $m = n - 2$, so $n = m + 2$. When $n = 2$, $m = 0$:

Now both sums run over the same powers of $x$. Combining them:

For this equation to hold for all $x$, each coefficient must vanish:

This gives us the recurrence relation:

From this single formula, every coefficient can be computed once we know $a_0$ and $a_1$.

Adjust the number of terms to see how the partial sum converges to the true solution. Near $x = 0$, even a few terms capture the behavior well. Further from the origin, more terms are needed for accuracy.

The Recurrence Relation

The recurrence relation $a_{n+2} = -a_n / [(n+1)(n+2)]$ connects each coefficient to one two steps earlier. This creates two independent chains:

Even coefficients (from $a_0$):

Odd coefficients (from $a_1$):

These are exactly the coefficients of the Taylor series for $\cos x$ and $\sin x$. Setting $a_0 = 1, a_1 = 0$ gives $y = \cos x$. Setting $a_0 = 0, a_1 = 1$ gives $y = \sin x$. The general solution is:

Watch how each coefficient determines one two steps ahead. The even and odd subsequences evolve independently, generating the two linearly independent solutions that a second-order equation requires.

The Method Step by Step

The power series method follows a systematic procedure. For a differential equation with polynomial coefficients, the steps are:

The key insight is that substituting a power series into a differential equation transforms the problem from solving a differential equation to solving an algebraic recurrence. The recurrence relation tells us how to compute each coefficient from earlier ones.

Radius of Convergence

A power series does not necessarily converge everywhere. The radius of convergence $R$ determines the interval $|x| < R$ where the series converges. Outside this interval, the series diverges and does not represent the solution.

For the harmonic oscillator $y'' + y = 0$, we can use the ratio test:

Since this limit is $0$ for any $x$, the series converges for all $x$. The radius of convergence is infinite.

In general, for an equation $P(x)y'' + Q(x)y' + R(x)y = 0$ with polynomial coefficients, the radius of convergence extends at least to the nearest singular point where $P(x) = 0$. Near such points, the equation changes character, and ordinary power series may fail.

Even when a series converges everywhere, practical accuracy depends on how many terms we compute. The visualization shows how the region of good approximation expands as we include more terms.

When Series Methods Are Necessary

Some differential equations have no solutions in terms of elementary functions. Consider Bessel's equation:

This equation arises constantly in physics and engineering: heat conduction in cylinders, vibrations of circular membranes, electromagnetic wave propagation. For integer values of $\nu$, no combination of polynomials, exponentials, trigonometric functions, or logarithms can solve it.

Power series come to the rescue. The Bessel function $J_\nu(x)$ is defined by its series expansion:

For $\nu = 0$, this simplifies to:

The Bessel function oscillates like a sine wave but with decreasing amplitude. It cannot be expressed in closed form, yet we can compute its values to arbitrary precision using the series. Tables of Bessel functions filled entire books before computers made evaluation trivial.

Other Special Functions

The power series method gives birth to a rich family of special functions, each defined as the solution to a particular differential equation:

Legendre polynomials $P_n(x)$: Solutions to $(1-x^2)y'' - 2xy' + n(n+1)y = 0$. These arise in problems with spherical symmetry, like the gravitational potential of a planet or the quantum mechanics of the hydrogen atom.

Hermite polynomials $H_n(x)$: Solutions to $y'' - 2xy' + 2ny = 0$. These describe the quantum harmonic oscillator, fundamental to quantum mechanics.

Laguerre polynomials $L_n(x)$: Solutions to $xy'' + (1-x)y' + ny = 0$. These appear in the radial part of the hydrogen atom wave functions.

Airy functions $\text{Ai}(x)$, $\text{Bi}(x)$: Solutions to $y'' - xy = 0$. These describe the behavior of quantum particles near turning points.

Each of these began as a power series solution to a differential equation that resisted elementary methods. Over time, mathematicians studied their properties so thoroughly that they became building blocks in their own right, referenced in tables and implemented in software.

Singular Points and the Frobenius Method

The basic power series method assumes the solution is analytic at $x = 0$: it has a convergent Taylor series. But some equations have singular points where this fails.

Consider:

At $x = 0$, the coefficient of $y''$ vanishes. This is a regular singular point. A plain power series will not work, but a modified approach does.

The Frobenius method assumes a solution of the form:

where $r$ is an unknown exponent to be determined. Substituting this into the equation leads to an indicial equation for $r$, plus a recurrence for the coefficients. The value(s) of $r$ from the indicial equation determine the leading behavior near the singular point.

Bessel's equation, for instance, has a regular singular point at $x = 0$. The Frobenius method reveals solutions proportional to $x^\nu$ and $x^{-\nu}$ (when $\nu$ is not an integer), or involving logarithms (when $\nu$ is an integer or half-integer). These subtleties make singular points a rich topic, but the core idea remains: expand in a series and match coefficients.

Computing with Power Series

In practice, power series solutions are evaluated by truncation. We compute $a_0, a_1, \ldots, a_N$ for some large $N$, then approximate:

The error depends on how fast the coefficients decay and how large $|x|$ is. Near the center of convergence, a few terms suffice. Far from the center, many terms may be needed.

Modern mathematical software (Mathematica, Maple, SciPy) can symbolically compute recurrence relations and numerically evaluate series to high precision. The method is not just theoretical; it is a practical computational tool.

Why This Matters

Power series solutions reveal something deep about differential equations. The solution is encoded in a sequence of numbers $a_0, a_1, a_2, \ldots$ satisfying algebraic relations. Solving the differential equation becomes a matter of algebraic bookkeeping.

This perspective extends to numerical methods. Finite difference schemes, spectral methods, and other computational approaches all share the spirit of converting continuous problems into discrete algebraic ones. Power series is the original version of this idea, and understanding it illuminates the entire field.

Moreover, the special functions that emerge from power series solutions appear everywhere in science. You cannot study quantum mechanics without Hermite and Laguerre polynomials. You cannot analyze wave propagation in cylindrical geometries without Bessel functions. These are not obscure mathematical curiosities; they are the language of physical phenomena.

Key Takeaways

• Assume a series solution $y = \sum a_n x^n$ when elementary methods fail, especially for equations with variable coefficients
• Substitute and match powers of x: Derivatives of power series are power series. Substitution into the DE yields a relation that must hold coefficient by coefficient
• Extract the recurrence relation: Setting coefficients of each power to zero gives algebraic equations relating the $a_n$
• Two initial coefficients determine everything: For second-order equations, $a_0$ and $a_1$ (often related to initial conditions $y(0)$ and $y'(0)$) determine all higher coefficients through the recurrence
• Check convergence: Use the ratio test or other methods to find the radius of convergence. The series represents the solution only where it converges
• Many important functions are defined this way: Bessel, Legendre, Hermite, and Airy functions all arise as power series solutions to equations that have no elementary closed form
• Singular points require care: At regular singular points, the Frobenius method (series with a non-integer leading power) extends the approach