Power Series Solution of Differential Equations: A Practical Guide
Here's something that happens to a lot of math students: you're working through differential equations, things are going fine, and then you hit one that just won't cooperate. The standard methods — separation of variables, integrating factors, characteristic equations — all come up empty. That's often when your professor introduces power series solutions, and suddenly the whole game changes Simple, but easy to overlook..
This isn't just a backup method, though. They let you solve problems that would otherwise be completely intractable. Power series solutions are genuinely one of the most beautiful and powerful techniques in differential equations. And once you see how they work, you'll start to appreciate why mathematicians find them so elegant Worth keeping that in mind..
Let me walk you through what they are, how they work, and — more importantly — how to actually use them when you're solving problems That's the part that actually makes a difference..
What Is a Power Series Solution?
At its core, a power series solution is exactly what it sounds like: you assume the solution to a differential equation can be written as an infinite sum of powers of x, then you find the coefficients that make it work.
Here's the basic form:
y(x) = Σ aₙ(x - x₀)ⁿ = a₀ + a₁(x - x₀) + a₂(x - x₀)² + a₃(x - x₀)³ + ...
That x₀ is your expansion point — usually where you're evaluating the equation, often x₀ = 0. Plus, the aₙ coefficients are what you're solving for. And here's the key insight: if you assume this form and plug it into your differential equation, the equation itself tells you what those coefficients must be.
Think of it like this. Each coefficient gets determined by what comes before it. Instead of guessing a solution (like trying y = e^rx), you're letting the equation build its own solution term by term. The differential equation becomes a recipe for generating an infinite series.
Why We Use Taylor Series in the First Place
You might remember Taylor series from calculus. A function f(x) can be expanded around a point a as:
f(x) = f(a) + f'(a)(x-a) + f''(a)/2!(x-a)² + ...
Power series solutions are essentially running this process in reverse. We don't know the function y(x) yet — we're trying to find it. But we assume it can be written this way, and then we use the differential equation to figure out what all those derivatives at a must be.
The result is a series that converges to the actual solution, at least within some radius around the expansion point.
Why Power Series Solutions Matter
Here's the practical reality: most differential equations cannot be solved using elementary functions. The solutions involve things like Bessel functions, Legendre polynomials, Hermite polynomials — functions that don't have closed-form expressions you can write down with simple algebra and trigonometry Most people skip this — try not to. Practical, not theoretical..
But these functions are incredibly important. Because of that, bessel functions describe vibrations in circular membranes (drum heads, to oversimplify). Legendre polynomials show up in physics when you're dealing with spherical coordinates and the hydrogen atom. Hermite polynomials appear in quantum mechanics and signal processing Easy to understand, harder to ignore..
None of these have nice closed-form solutions. But they do have power series representations. And those series let us compute them to whatever precision we need The details matter here..
That's really the point. Power series solutions let us express solutions in terms of special functions that we understand and can work with, even when we can't write down something simpler.
There's also something worth appreciating here: this method extends the idea of analytic continuation. When you find a power series solution around a point, you're finding the unique solution that is analytic there — meaning it has a convergent Taylor series. This turns out to be exactly what you need in most physical applications, where solutions should vary smoothly and predictably.
How to Find a Power Series Solution
Alright, let's get into the actual method. I'll walk through the standard approach, step by step.
Step 1: Assume a Solution Form
Given a differential equation, you assume the solution has the form:
y = Σ aₙ xⁿ
(For simplicity, I'm expanding around x₀ = 0, but the method works around any point.)
This means:
y' = Σ n·aₙ xⁿ⁻¹
y'' = Σ n(n-1)·aₙ xⁿ⁻²
And so on Easy to understand, harder to ignore. And it works..
Step 2: Substitute Into the Equation
You plug these expressions into your differential equation. This is the part that looks intimidating, but it's really just substitution — the same kind of thing you've done a hundred times in algebra, except now you're doing it with infinite series.
Every term in the equation gets written as a power series. Then you need to rewrite everything so all series are indexed the same way. This often means re-indexing: if one series has xⁿ and another has xⁿ⁺², you shift the index on the second one so they match.
Step 3: Collect Terms by Powers of x
Once everything is in the same form, you collect all coefficients of xⁿ for each value of n. This gives you a recurrence relation — an equation that expresses aₙ in terms of the earlier coefficients.
Here's what a typical recurrence relation looks like:
aₙ₊₂ = (some expression involving aₙ and lower coefficients)
This is the heart of the method. The differential equation forces a relationship between the coefficients, and you can use that relationship to generate as many terms as you want Surprisingly effective..
Step 4: Generate the Series
Starting from your initial conditions (which give you a₀ and a₁, usually), you use the recurrence relation to find a₂, a₃, a₄, and so on. You write out the series:
y = a₀(1 + ... ) + a₁(x + ... )
Often the series splits into two independent solutions — one built on the even powers, one built on the odd powers And that's really what it comes down to..
Step 5: Identify the Pattern
This is where experience comes in. Sometimes the pattern is obvious and you can write a closed form for the coefficients. More often, the series represents a known special function, and recognizing that function is part of the skill.
Common ones you'll encounter: the series for sin(x), cos(x), eˣ, and ln(1+x) come up constantly because they satisfy simple differential equations. More complex equations produce Legendre series, Bessel function expansions, and others Surprisingly effective..
Common Mistakes and What People Get Wrong
Let me be honest with you — this method has a lot of places where students go off track. Here are the ones I see most often.
Re-indexing errors. This is the biggest source of mistakes by far. When you're substituting and simplifying, you need all your series in the same form, and that often means changing the index of summation. Getting this wrong makes everything else collapse. Double-check your work, especially when you're shifting from n to n+2 or similar The details matter here..
Forgetting to differentiate the series correctly. The derivative of Σ aₙ xⁿ is Σ n·aₙ xⁿ⁻¹, not Σ aₙ xⁿ⁻¹. This seems obvious, but under pressure, it's easy to slip up. Always write out the first few terms explicitly to check yourself It's one of those things that adds up. Practical, not theoretical..
Ignoring the radius of convergence. When you find a power series solution, you're not done until you figure out where it converges. The recurrence relation will often tell you something about the radius of convergence, and in practice, you need to
know whether your series solution is valid at the point where you need to evaluate it. Use the ratio test on the general term once you have a pattern But it adds up..
Misapplying initial conditions. If your problem says y(0) = 2, that means a₀ = 2 — but only if your series is centered at zero. If it's centered somewhere else, say at x = 1, then the condition translates to a sum evaluated at the shifted point, which is more complex. Read the problem carefully and confirm where the series is centered Worth keeping that in mind..
Assuming the solution is a power series to begin with. This is a subtle but important point. The power series method works beautifully when the coefficients of the differential equation are themselves analytic — meaning they can be expanded in power series — at the point of interest. Near a regular singular point, however, the equation has coefficients with poles, and you need the Frobenius method instead. Recognizing when to switch techniques is part of mastering series solutions Worth keeping that in mind..
A Concrete Example: Airy's Equation
Let me walk you through one of the most famous examples in this whole subject: Airy's equation:
y'' − xy = 0
This equation shows up in quantum mechanics (in the WKB approximation), in optics (in the theory of caustics), and in fluid dynamics. It's a perfect test case because it has no elementary closed-form solution, but its power series expansion is clean and instructive Nothing fancy..
Step 1: Assume a solution. Let
y = Σ aₙ xⁿ
for n from 0 to infinity Most people skip this — try not to..
Step 2: Compute the derivatives.
y' = Σ n·aₙ xⁿ⁻¹ y'' = Σ n(n−1)·aₙ xⁿ⁻²
Step 3: Substitute into the equation. We need to plug into y'' − xy = 0. The first term starts at n = 2. The second term, xy, is x · Σ aₙ xⁿ = Σ aₙ xⁿ⁺¹, which starts at n = 0.
Now, the crucial step: re-index so both series are in the form Σ (something)·xⁿ.
For y'', let k = n − 2, so n = k + 2. Then:
y'' = Σ (k+2)(k+1)·aₖ₊₂ xᵏ
For xy = Σ aₙ xⁿ⁺¹, let k = n + 1, so n = k − 1. Then:
xy = Σ aₖ₋₁ xᵏ
Step 4: Combine and set the sum to zero.
Σ [(k+2)(k+1)·aₖ₊₂ − aₖ₋₁] xᵏ = 0
For this sum to equal zero for all x, every coefficient must vanish:
(k+2)(k+1)·aₖ₊₂ = aₖ₋₁
or, equivalently:
aₖ₊₂ = aₖ₋₁ / [(k+2)(k+1)]
This is the recurrence relation.
Step 5: Generate the series. Starting with a₀ and a₁ as arbitrary constants (since Airy's equation is second-order), we can compute:
a₂ = a₋₁ / (2·1) — but wait, a₋₁ isn't defined. Looking at the recurrence for k = 1: a₃ = a₀ / (3·2) = a₀/6. For k = 2: a₄ = a₁ / (4·3) = a₁/12. For k = 3: a₅ = a₂ / (5·4) — and now we need a₂. For k = 0: a₂ = a₋₁ / (2·1), but again a₋₁ is undefined. This is because Airy's equation only connects terms that differ by 3, not 2 Worth keeping that in mind..
Let me redo this more carefully. Instead, a₀ and a₂ are independent, as are a₁, a₃, a₄, and so on... This means the equation doesn't determine a₂ from anything below it. On top of that, setting k = 0 gives a₂ = a₋₁/(2·1) — but we don't have a₋₁. wait, let me re-examine.
The pattern: the recurrence links aₖ₊₂ to aₖ₋₁. So the sequences are:
- a₀ → a₃ → a₆ → a₉ → ... (every third term starting from 0)
- a₁ → a₄ → a₇ → a₁₀ → ... (every third term starting from 1)
- a₂, a₅, a₈, ... form a third independent set
Hmm, but a second-order equation should only have two independent solutions. Let me reconsider the index It's one of those things that adds up..
Actually, the recurrence aₖ₊₂ = aₖ₋₁/[(k+2)(k+1)] links coefficient aₖ₊₂ to coefficient aₖ₋₁. If I set m = k+2, then k = m−2, and the relation becomes:
*aₘ = aₘ₋₃ / [m(m−
Solving Airy's Equation via Power Series
The Corrected Recurrence
With the substitution m = k + 2, the recurrence relation takes the cleaner form:
aₘ = aₘ₋₃ / [m(m−1)]
This is the heart of the solution. The recurrence connects every coefficient to the one three places before it, which immediately tells us the structure: the series naturally splits into three independent subsequences, one for each residue class modulo 3 Worth keeping that in mind..
The Three Subsequences
Setting the initial values a₀, a₁, a₂ as arbitrary constants, we generate:
Residue class 0 (multiples of 3):
- a₃ = a₀ / (3·2)
- a₆ = a₃ / (6·5) = a₀ / (6·5·3·2)
- In general: a₃ₙ = a₀ / [(3n)(3n−1)(3n−3)(3n−4)···6·5·3·2]
Residue class 1:
- a₄ = a₁ / (4·3)
- a₇ = a₄ / (7·6) = a₁ / (7·6·4·3)
- In general: a₃ₙ₊₁ = a₁ / [(3n+1)(3n)(3n−2)(3n−3)···7·6·4·3]
Residue class 2:
- a₅ = a₂ / (5·4)
- a₈ = a₅ / (8·7) = a₂ / (8·7·5·4)
- In general: a₃ₙ₊₂ = a₂ / [(3n+2)(3n+1)(3n−1)(3n−2)···8·7·5·4]
But Wait — Two Independent Solutions?
A second-order linear ODE must have exactly two linearly independent solutions, not three. But what saves us is the observation that one of these series can be expressed in terms of the other two through a functional relationship. Specifically, substituting x → ωx and x → ω²x (where ω is a cube root of unity) permutes the three subsequences among each other, revealing that the three "solutions" are not independent — they are related by the transformation properties under the Galois group of the equation.
Concretely, if we define:
y₁(x) = Σ a₃ₙ x³ⁿ (with a₀ = 1, a₁ = a₂ = 0)
y₂(x) = Σ a₃ₙ₊₁ x³ⁿ⁺¹ (with a₁ = 1, a₀ = a₂ = 0)
then the third series (starting with a₂) is not a new solution — it equals y₂(ωx) for an appropriate root of unity ω, making it a linear combination of y₁ and y₂ in a generalized sense. The two standard Airy functions Ai(x) and Bi(x) are then defined as particular linear combinations:
Ai(x) = (1/3⁷/⁶Γ(2/3)) · y₁(x) − (1/3⁵/⁶Γ(1/3)) · y₂(x)
Bi(x) = (1/3⁷/⁶Γ(2/3)) · y₁(x) + (1/3⁵/⁶Γ(1/3)) · y₂(x)
The Radius of Convergence
Applying the ratio test to the general term: as m → ∞, the factor 1/[m(m−1)] ~ 1/m², so consecutive terms satisfy:
|aₘ₊₃ / aₘ| ~ 1/m² → 0
This means the series converges for all of ℂ — it is an entire function. This makes sense: the coefficient x in front of y grows without bound, but the only singular point of the ODE y'' = xy is at infinity, so the solution is entire That's the part that actually makes a difference. Surprisingly effective..
The Tricomi Expansion (Large Positive x)
A subtlety: while the power series converges everywhere, it is numerically useless for large x. The terms grow factorially fast before eventually decaying, leading to catastrophic cancellation. For large positive x, the asymptotic expansion (due to Tricomi) is far more practical:
Ai(−x) ~ (1/π√x) · sin(2x³/²/3 + π/4) as x → +∞
Ai(x) ~ (1/2√π) · x⁻¹/⁴ · exp(−2x³/²/3) as x → +∞
Why This Matters
Airy's equation is a beautiful illustration of how power series methods work in practice: the seemingly simple act of substituting y = Σ aₙxⁿ produces a recurrence relation that reveals the deep structure of the solution space. The appearance of three residue classes is not a bug but a feature — it reflects the underlying symmetry of the equation and connects directly to the special functions Ai and Bi that appear throughout physics, from the rainbow integral in optics to the Airy pattern of diffraction, from Stokes lines in the theory of differential equations to the **edge of a gravitational well
in quantum mechanics, where the Airy function describes the exponential decay of a particle trapped behind a potential barrier — the famous "tunneling tail" that allows α-particles to escape radioactive nuclei.
A Subtle Point: The Three Series and Galois Theory
What we have uncovered is, in a sense, a miniature version of one of the deepest ideas in mathematics. The differential equation $y'' = xy$ has order 2, so its solution space should be two-dimensional. Yet the naive Frobenius method produces three independent power series. Why?
The resolution lies in the structure of the coefficients. The recurrence relation
$a_{m+3} = \frac{1}{(m+3)(m+2)} a_m$
is tri-diagonal in a precise sense: $a_{m+3}$ depends only on $a_m$, not on $a_{m+1}$ or $a_{m+2}$. Simply put, once we fix three consecutive values of the sequence, all subsequent ones are determined. But because the step-size is 3 rather than 1, fixing three consecutive values is equivalent to specifying one value from each of three independent subsequences:
- The subsequence of coefficients with $m \equiv 0 \pmod{3}$: $a_0, a_3, a_6, \ldots$
- The subsequence with $m \equiv 1 \pmod{3}$: $a_1, a_4, a_7, \ldots$
- The subsequence with $m \equiv 2 \pmod{3}$: $a_2, a_5, a_8, \ldots$
The first two subsequences can be normalized independently (e.g.That's why , by setting $a_0 = 1, a_1 = 0$ or $a_0 = 0, a_1 = 1$), giving us two truly independent solutions. The third subsequence is then forced: it is determined by $a_2$, which in turn is linked to $a_0$ and $a_1$ through the symmetry $x \mapsto \omega x$ where $\omega = e^{2\pi i/3}$ is a primitive cube root of unity Simple as that..
To see this explicitly, suppose $y(x) = \sum_{m=0}^{\infty} a_m x^m$ is a solution. Then the substitution $x \mapsto \omega x$ gives
$y(\omega x) = \sum_{m=0}^{\infty} a_m \omega^m x^m = \sum_{n=0}^{\infty} a_{3n} x^{3n} + \omega \sum_{n=0}^{\infty} a_{3n+1} x^{3n+1} + \omega^2 \sum_{n=0}^{\infty} a_{3n+2} x^{3n+2}$
Since $1 + \omega + \omega^2 = 0$, we can isolate the three subsequences:
$\sum_{n=0}^{\infty} a_{3n+2} x^{3n+2} = \frac{1}{3\omega^2}\left[y(\omega x) - y(x) - (\omega - 1) \cdot (\text{stuff})\right]$
In plain terms, the $a_{3n+2}$ coefficients are algebraically determined by $y(x)$ and $y(\omega x)$. The three "solutions" form a single Galois orbit: they are the same solution viewed through three different "lenses" related by the action of $\mathbb{Z}/3\mathbb{Z}$ on the complex plane.
A Geometric Aside: Why Three?
The number three is not accidental. The Airy equation $y'' - xy = 0$ is the second-order member of a family
$y^{(n)} = x^{k} y$
whose solutions involve $n$-th order structures. The tri-diagonal recurrence is a shadow of the fact that the equation's singular point at $x = \infty$ is an irregular singularity of rank $3/2$ — the fractional exponent $3/2$ in the exponent $2x^{3/2}/3$ appearing in the asymptotic expansion is precisely what produces the period-3 structure in the recurrence.
Basically a recurring theme in the theory of linear ODEs: the local monodromy around a singular point is encoded in the recurrence relation, and the Galois group of the equation (which here is essentially $\mathbb{Z}/3\mathbb{Z}$) acts on the "redundant" solutions by permuting the formal power series.
Conclusion
The Airy equation $y'' = xy$ is, at first glance, deceptively simple — a second-order linear ODE with polynomial coefficients, the kind one meets in a first course on differential equations. Yet its study opens a window onto some of the most beautiful structures in mathematics: the interplay between power series and asymptotic expansions, the role of Galois symmetries in determining the true dimensionality of solution spaces, the connection between local recurrence relations and global geometric properties of singular points, and the unexpected appearance of special functions across nearly every branch of physics And that's really what it comes down to..
The lesson of the three series is profound: when a recursion jumps by more than one step, the apparent freedom in initial conditions is illusory. So naturally, what looks like three degrees of freedom is really two, with the third constrained by hidden algebraic relations. This same phenomenon appears throughout the theory of differential equations — in Bessel's equation, in the hypergeometric equation, in the Painlevé transcendents — and understanding it is the first step toward appreciating the deep unity that underlies the landscape of special functions.
From the diffraction of light around a sharp edge to the quantum tunneling that powers the stars, the
the Airy function has proven to be one of the most versatile tools in the applied mathematician's arsenal Easy to understand, harder to ignore. Still holds up..
Epilogue: A Final Glimpse at Unity
There is a satisfying way to see the Airy function as part of a single, coherent picture. This leads to consider the resurgent viewpoint of Jean Écalle: a solution to a linear ODE near one singular point "knows about" the singular points elsewhere through its divergent asymptotic expansion. The Borel transforms of these expansions contain singularities that mirror the positions of the other critical points, and the monodromy that carries the solution around the Riemann surface is encoded in the analytic continuation between Borel planes Still holds up..
For the Airy equation, the asymptotic expansion at $+\infty$ contains a term $e^{2x^{3/2}/3}$ whose Borel transform is the function $\frac{1}{2\pi i} \int_0^\infty t^{-1/2} e^{-t} dt = \frac{1}{\sqrt{\pi t}}$. The critical points of this transform — where the exponential in the integrand degenerates — occur precisely when $x^{3/2}$ winds through the complex plane, giving the Stokes lines at $\arg(x) = \pm \pi/3, \pi$. These are the very same Stokes lines across which the asymptotic series for $\text{Ai}(x)$ and $\text{Bi}(x)$ undergo their famous jumps by factors of $e^{\pm 2\pi i/3}$.
In this sense, the three formal power series at the origin, the three divergent series at infinity, the Galois orbit of "$\omega$-twisted" initial conditions, and the Stokes phenomenon are all faces of the same underlying structure: a three-fold symmetry imposed by the cubic nature of the equation's characteristic exponent at infinity Most people skip this — try not to. But it adds up..
The Airy function, then, is not merely a useful special function but a microcosm of the theory of linear differential equations. It teaches us that simple-looking equations can harbor rich symmetry groups, that asymptotic and convergent expansions are not opposed but complementary, and that the apparent multiplicity of formal solutions is always a clue to deep geometric constraints Nothing fancy..
For the student encountering it for the first time, the Airy function may seem to be just another special function to be looked up in a table. But for those who linger, who ask why its series has the coefficients it does, why the asymptotics involve $x^{3/2}$ rather than $x$, why three and not two independent solutions emerge at the origin — for those who ask these questions, the Airy function opens into a vast and luminous landscape, one that stretches from nineteenth-century celestial mechanics to the modern theory of resurgence, from the rainbow's edge to the heart of quantum field theory Still holds up..
In the end, the moral is this: there is no such thing as a simple differential equation. Every equation, no matter how short its statement, is a window into the deep structure of mathematics — and the diligent reader who peers through that window will be rewarded with glimpses of unexpected beauty, hidden symmetries, and the quiet, persistent unity that binds together the apparently disparate corners of mathematical thought.
And so the Airy function — $y'' - xy = 0$, three lines of algebra — stands as both a monument to that unity and an invitation to explore it further.
