When the elementary tricks run out
Every method you have trusted so far quietly assumed something special about the coefficients. For a constant-coefficient equation a y'' + b y' + c y = 0 you guessed y = e^(rx) and the calculus collapsed into a characteristic polynomial. For a Cauchy-Euler equation you guessed y = x^r and got lucky again. But these are gifts of *special* coefficients. The moment you face an honest variable-coefficient equation like y'' - 2x y' + 2 y = 0, where the coefficient 2x genuinely moves with x, every clever guess fails: no single exponential, no single power, no finite combination of elementary functions solves it.
This is not a rare predicament; it is the *normal* state of affairs. Be honest about it: most differential equations have no solution expressible in closed form, and variable coefficients are exactly where that wall appears. The way through is a change of ambition. Instead of demanding a finished formula, we will accept the solution as an infinite power series — a sum y = a0 + a1 (x - x0) + a2 (x - x0)^2 + ... — and our job becomes finding the coefficients a0, a1, a2, ... one after another. A power series is, after all, a polynomial that never stops; if we can pin down its every coefficient, we have pinned down the function.
What makes a point ordinary
The method does not work everywhere, so the first task is to find a safe place to expand around. Put the second-order linear equation into its standard form y'' + p(x) y' + q(x) y = 0, with the leading coefficient divided out to 1. A point x0 is an ordinary point when both p(x) and q(x) are *analytic* there — meaning each one has its own convergent power series around x0, with no blow-up, no division by zero, nothing torn. In practice, if p and q are polynomials, or ratios of polynomials whose denominators are nonzero at x0, then x0 is an ordinary point. Everywhere else — wherever a coefficient explodes — is a singular point, and the clean theory of this guide does not reach it.
The promise attached to an ordinary point is exact and reassuring. At an ordinary point a power-series solution always exists, and you can find two independent ones — enough for a full general solution of a second-order equation, by the superposition you already trust. There is no risk of the method silently failing or producing a divergent nonsense series; the ordinary-point condition is precisely the certificate that the series will behave. That guarantee is what makes this the right place to start; the harder case of expanding *at* a singular point is the method of Frobenius, the next major step in this rung.
The machine: substitute, shift, and match
Here is the engine of the whole power-series method. Write y = sum of a_n (x - x0)^n. Differentiating a power series is done term by term, so y' and y'' are themselves power series with shifted indices and extra factors out front. You then substitute all three series into the differential equation. The trick that turns this from a mess into a method is re-indexing: you shift the summation index in each series so that every term carries the same power of (x - x0). Once they are aligned, the entire equation becomes a single power series set equal to zero.
And now the decisive fact: a power series equals zero for all x in an interval only if every single coefficient equals zero. (This is the identity theorem for power series — the same uniqueness that lets you compare coefficients on two sides of an equation.) So setting the coefficient of each power (x - x0)^k to zero hands you an equation relating the a's. Collected across all k, these equations form a recurrence relation: a rule that computes each new coefficient from earlier ones. The differential equation has been transmuted into an algebra problem you can crank by hand, term after term.
Equation: y' = y (use it as a tiny demo; ordinary point x0 = 0) Try: y = a0 + a1 x + a2 x^2 + a3 x^3 + ... Then: y' = a1 + 2 a2 x + 3 a3 x^2 + ... Match powers of x (coefficient of x^k on each side must agree): x^0 : a1 = a0 x^1 : 2 a2 = a1 -> a2 = a1 / 2 = a0 / 2! x^2 : 3 a3 = a2 -> a3 = a2 / 3 = a0 / 3! general recurrence : (n+1) a_(n+1) = a_n So a_n = a0 / n! and y = a0 ( 1 + x + x^2/2! + ... ) = a0 e^x. The series rebuilt e^x from scratch -- no exponential was ever assumed.
Two solutions from two free constants
In that tiny demo only a0 was free, because y' = y is first order. For a genuine second-order equation the recurrence relation leaves two coefficients undetermined — almost always a0 and a1 — and every later coefficient is computed from them. That is no accident: a0 is just y(x0) and a1 is just y'(x0), the two pieces of initial data a second-order problem always needs. The two free constants are the power-series face of the two arbitrary constants in a general solution.
To extract the two independent building blocks, run the recurrence twice with clean choices. Set a0 = 1, a1 = 0 and grind out one series — call it y1; then set a0 = 0, a1 = 1 and grind out the other — y2. Because their starting data differ, y1 and y2 are automatically independent, so by superposition the general solution is y = a0 y1 + a1 y2 with a0, a1 now playing the role of the free constants. You have built a fundamental set of solutions without ever writing them in closed form.
Where the series is trustworthy
A series solution is only as good as the region where it converges, so the last honest question is: how far from x0 can you trust it? The answer is beautifully geometric. The series for each solution converges at least within a circle reaching out to the *nearest singular point* of the equation — and the radius of convergence is at least that distance. You do not even have to compute the series to know how far it will carry: just locate where the coefficients p(x) and q(x) misbehave, and the closest such trouble spot caps your guaranteed radius.
One subtlety makes this lovely rather than merely useful: the nearest singular point may be a *complex* number, off the real line entirely, and it still sets the radius. The classic case is y'' + y = 0, whose solutions are cos x and sin x. Their power series at 0 converge for every real x — infinite radius — and indeed p = 0, q = 1 have no singular points anywhere in the complex plane. By contrast, an equation whose only trouble sits at x = i and x = -i would have a real-centered series guaranteed only for |x| less than 1, because those imaginary points lie a distance 1 from the origin even though they never appear on the real axis you are working on.
Hold the whole arc of this guide in one line. At an ordinary point you trade the search for a formula for the patient growth of coefficients: substitute a power series, match powers to get a recurrence relation, run it from two free constants to get two independent solutions, and trust the result out to the nearest singularity. The next guide sharpens the second half of that craft — how to read a recurrence's pattern and pin the radius of convergence precisely — before we dare to expand *at* a singular point and meet the Bessel and special functions that the elementary methods could never reach.