From a wall of equations to one rule
In the previous guide you saw the whole ritual of the power-series method: write the unknown as y = a_0 + a_1 x + a_2 x^2 + ..., differentiate term by term, substitute into the equation, and collect by powers of x. What comes back is not a tidy answer but a wall of conditions — one equation for the coefficient of x^0, one for x^1, one for x^2, and so on, forever. Faced with infinitely many equations you might despair. The escape is that they are not independent: they are all the same rule, repeated. That single repeating rule is the recurrence relation.
A recurrence relation is a formula that expresses each new coefficient in terms of earlier ones — say a_(n+2) in terms of a_n. Think of it as a chain of dominoes: you tip over the first one or two by hand (those are your free starting values), and every later coefficient then falls automatically, each determined by the ones before it. The equation has compressed an infinite list of demands into one portable rule you can turn like a crank. Solving the differential equation, in this whole subject, really means finding and turning this crank.
Cranking the recurrence
Let us actually turn the crank on a clean example, y'' + y = 0 — an equation you already know answers to (sine and cosine), which makes it perfect for watching the machine work. Substituting the series and matching the coefficient of x^n to zero produces the recurrence a_(n+2) = -a_n / ((n + 2)(n + 1)), valid for every n >= 0. This is a two-term recurrence: it links a coefficient only to the one two steps behind it, never to its immediate neighbour. That gap of two is about to do something pretty.
- Pick the two free constants a_0 and a_1; everything else is forced from them.
- Feed even indices: a_2 = -a_0/(2·1) = -a_0/2!, then a_4 = -a_2/(4·3) = a_0/4!, then a_6 = -a_0/6!, ...
- Feed odd indices separately: a_3 = -a_1/3!, then a_5 = a_1/5!, then a_7 = -a_1/7!, ...
- Read the two patterns: the a_0-chain sums to cos(x), the a_1-chain sums to sin(x). The general solution is a_0 cos(x) + a_1 sin(x).
Notice the lovely structural fact a two-term recurrence gives you for free: because each step jumps by two, the even-indexed coefficients form one self-contained chain (built entirely from a_0) and the odd-indexed coefficients form a completely separate chain (built entirely from a_1). The series splits cleanly into an even-power solution and an odd-power solution — and those two pieces are precisely your two independent solutions. You did not have to hunt for the second solution; the recurrence handed it to you in the same breath as the first.
When the chain stops: polynomials appear
Here is where the recurrence earns its keep beyond rebuilding sine and cosine. In many physical equations the recurrence carries a parameter — a constant baked into the equation — and for special values of that parameter, one of the chains hits a coefficient that is forced to zero, after which everything downstream is zero too. The infinite series terminates into a finite polynomial. This is not a happy accident; it is the mechanism behind the most important special functions in physics.
Take the Legendre equation (1 - x^2) y'' - 2x y' + l(l + 1) y = 0, whose recurrence works out to a_(n+2) = a_n · (n(n+1) - l(l+1)) / ((n+2)(n+1)). Stare at the numerator: when the running index n equals the parameter l, the factor n(n+1) - l(l+1) becomes exactly zero, so a_(l+2) = 0, and the whole chain from there onward dies. If l is a non-negative integer, one of the two chains terminates and that solution is a polynomial of degree l — a Legendre polynomial. The other chain runs forever and, as it happens, blows up at x = +-1.
How far does the answer reach?
A series solution is only honest where the infinite sum actually converges to a finite value. Outside that region the terms grow without bound and the 'solution' is meaningless symbols. The reach of a series — the distance out from your expansion point within which it faithfully represents a genuine solution — is its radius of convergence. A series for an ODE is not a free pass to all of x; it is a local object, valid on a disc, and you need to know how big that disc is.
The beautiful part — the headline result of this rung — is that you can predict the radius before computing a single coefficient. The theorem says: for an equation expanded about an ordinary point x0, the radius of convergence is at least the distance from x0 to the nearest singular point of the equation. So the recipe is pure bookkeeping: divide into standard form, find where the coefficients blow up, measure from x0 to the closest such place, and that distance is your guaranteed minimum reach. The classification of singular points you learned earlier was not abstract taxonomy — it literally fences in your answer.
equation x0 nearest singular pt guaranteed R -------- -- ------------------- ------------ y'' + y = 0 0 none (entire) infinity (1 - x^2) y'' - 2x y' + 6y = 0 0 x = 1, x = -1 1 (1 + x^2) y'' + ... = 0 0 x = i, x = -i (!) 1 (x - 3) y'' + y = 0 0 x = 3 3
The singularity you cannot see
Look again at (1 + x^2) y'' + ... = 0 expanded about x0 = 0. Along the real axis the coefficient 1 + x^2 never vanishes — it is a smooth, friendly bump with no visible trouble anywhere. Yet the radius is only 1, not infinity. The reason is that 1 + x^2 = 0 at x = i and x = -i, two points sitting up in the complex plane, each a distance 1 from the origin. The theorem measures distance in the complex plane, so a singularity hiding off the real line still controls how far a perfectly real-looking series will reach.
This explains a phenomenon that genuinely puzzles newcomers. You can have a real function that looks utterly smooth on the whole real line — no spike, no corner, no infinity — yet its Taylor series suddenly stops converging at some finite radius for no reason you can see by looking at the real graph. The reason is always there; it is just hiding in the complex plane. It is an honest reminder that the real line is a thin slice of a richer geometry, and convergence answers to that geometry, not to what your eye can see along x.
Two honest caveats round this out. First, the theorem gives a guaranteed minimum, not the exact value — the true radius can be larger (the series may keep converging past the nearest singularity), but it is never smaller. Second, this whole clean story lives at an ordinary point; at a singular point the plain series can fail and you need the heavier machinery of the next guides. Within the interval of validity the series solution is a real, dependable answer — just remember it is a local one, with a reach you can compute in advance.