Reduction of Order: A Second Solution from a First

The gap that needs filling

By now the headline of this rung is second nature: the solutions of a y'' + b y' + c y = 0 fill a two-dimensional space, so you need two independent solutions to write the general one. For constant coefficients the characteristic equation hands you both at once, neatly sorted into three cases. But that gift is tied to constant coefficients. The moment a coefficient varies with x — an equation like x^2 y'' - 3x y' + 4y = 0, or the famous Bessel and Legendre equations of physics — guessing y = e^(rx) collapses, and you are left staring at a problem the algebra trick cannot touch.

Yet very often you can get *one* solution by other means — a lucky guess, a power-series computation, a known special function, or, in the constant-coefficient world, the single exponential that a repeated root coughs up. The trouble is that one solution only spans a line through the origin, not the whole plane: half the answer. Reduction of order is the method that takes that one solution, call it y1, and manufactures a second one y2 that is guaranteed to be independent of it. From a single seed it grows the missing dimension.

The idea: let the constant become a function

Here is the whole trick in one sentence. You already know that c * y1 is a solution for any constant c — but a constant multiple is *not* independent of y1, so it is useless for building a basis. So loosen the constant: instead of c, allow a *function* v(x), and look for a second solution of the form y2 = v(x) * y1(x). The hope is that the freedom in v will let you satisfy the equation while making y2 genuinely different from y1. This single move — promoting a constant to an unknown function — is the heart of the method, and it is the same instinct behind variation of parameters you will meet on the nonhomogeneous rung.

Why should so blunt a guess work? Because plugging y2 = v * y1 into the equation produces something magical: a new differential equation for v in which the plain v-term (the one with no derivative of v) cancels completely. It must cancel — its coefficient is exactly the original equation evaluated at y1, and that is zero precisely because y1 is a solution. What remains involves only v'' and v', never v itself. So if you rename w = v', the second-order equation in v becomes a *first-order* equation in w. That is where the name comes from: the order has been reduced, from two down to one.

The recipe, step by step

Put the equation in standard form first — divide through so the y'' coefficient is 1: y'' + p(x) y' + q(x) y = 0. That p(x) is the piece the method cares about. Then carry out the substitution y2 = v y1 and watch the order drop. Here is the sequence as a clean checklist.

1. Write the equation in standard form y'' + p(x) y' + q(x) y = 0, and have your known solution y1 in hand.
2. Substitute y2 = v y1. Compute y2' = v' y1 + v y1' and y2'' = v'' y1 + 2 v' y1' + v y1', then plug into the equation.
3. Collect terms. The coefficient of v is y1'' + p y1' + q y1 = 0 (because y1 solves the equation), so the bare v drops out, leaving only v'' and v'.
4. Set w = v'. The leftover equation is first-order linear in w — separable, in fact — so solve it: w = (1/y1^2) * exp(-integral of p dx).
5. Integrate once more to recover v from w = v', then form y2 = v y1. Drop any additive constant (it only rebuilds a multiple of y1) and any leading constant factor.

Steps four and five compress into a single formula worth knowing, the reduction-of-order formula: y2 = y1 * integral of [ exp(-integral of p dx) / y1^2 ] dx. You can memorize it, but it is healthier to re-derive it by the substitution a few times until the cancellation feels inevitable rather than magical. Notice the y1^2 in the denominator — that is why the method needs y1 to be nonzero on the interval you work in, a caveat worth keeping in your pocket.

A worked sketch and a hidden friend

Take a variable-coefficient case where the e^(rx) trick is useless: x^2 y'' - 3x y' + 4y = 0 for x > 0. Suppose someone tells you y1 = x^2 works — you can verify it in your head: y1' = 2x, y1'' = 2, and x^2*(2) - 3x*(2x) + 4*(x^2) = 2x^2 - 6x^2 + 4x^2 = 0. Good. Now standard form means dividing by x^2: y'' - (3/x) y' + (4/x^2) y = 0, so p(x) = -3/x. The integral of p is -3 ln x, so exp(-integral of p dx) = exp(3 ln x) = x^3. Then w = x^3 / (x^2)^2 = x^3 / x^4 = 1/x, and v = integral of 1/x dx = ln x. Hence y2 = v y1 = x^2 ln x.

Equation:        x^2 y'' - 3x y' + 4y = 0   (x > 0)
Known:           y1 = x^2
Standard form:   y'' - (3/x) y' + (4/x^2) y = 0   ->  p(x) = -3/x
Factor:          exp(-INT p dx) = exp(3 ln x) = x^3
Reduce:          w = v' = x^3 / (x^2)^2 = 1/x
Integrate:       v = INT (1/x) dx = ln x
Second solution: y2 = v*y1 = x^2 ln x
General:         y = c1*x^2 + c2*x^2 ln x

The general solution is therefore y = c1 * x^2 + c2 * x^2 ln x. That logarithm could never have come from any exponential guess; reduction of order found it because it asked the equation a smarter question. And the ln x factor instantly settles independence: x^2 ln x is not a constant multiple of x^2, so the Wronskian of the pair is nonzero and they form a genuine fundamental set. In fact this guarantee is built into the method, which leads to the friend hiding in the formula.

Look again at the term exp(-integral of p dx). You saw exactly that expression in the previous guides as Abel's identity for the Wronskian: W(x) = W(x0) * exp(-integral of p dx). That is no coincidence. Reduction of order works because v' = W / y1^2, where W is the Wronskian of y1 and the y2 you are building. Since Abel's identity says W is never zero (when it starts nonzero), v' is never zero, so v is genuinely non-constant — which is *precisely* the statement that y2 is independent of y1. The method cannot accidentally hand you a multiple of y1; independence is structurally guaranteed.

When it bites, and where it leads

Be clear-eyed about the limits. First, you must already possess y1 — reduction of order is a *second*-solution machine, not a first-solution oracle, and for genuinely hard variable-coefficient equations even finding y1 may demand power series or special-function theory from later rungs. Second, the formula divides by y1^2, so it stumbles at points where y1 = 0; you typically work on an interval where y1 keeps a constant sign, then patch across zeros separately if needed. Third, the surviving integral integral of [exp(-integral of p dx)/y1^2] dx may have no elementary closed form — and that is fine. A solution defined by an honest integral is still a solution.

It is also worth saying what reduction of order is *not*. It is not a way to solve nonlinear equations — the cancellation that drives it relies entirely on the linear, homogeneous structure, where 'y1 is a solution' makes a whole block of terms vanish. Throw in a (y')^2 and the trick disintegrates, just as superposition did back in the first guide. The technique lives and dies inside linearity.

With this guide the homogeneous rung is complete: you can read off the dimension of the solution space, test independence with the Wronskian and Abel's identity, solve every constant-coefficient case through the characteristic equation's three flavours, and now — even with variable coefficients — build a full basis from a single seed. The very same 'let the constant vary' idea is the doorway to the next rung, where a forcing term sits on the right-hand side and variation of parameters turns this trick into a general method for the nonhomogeneous equation. You are not closing a chapter so much as sharpening the tool you will carry into the next one.