Reduction of Order

Sometimes you already know one solution of a second-order equation — you spotted it, someone handed it to you, or a factored characteristic equation coughed one up — but a second-order equation needs two independent solutions to be fully solved. Reduction of order is the trick that squeezes the second solution out of the first. The idea is almost cheeky: take your known solution and let its coefficient vary. For the linear homogeneous equation

y'' + p(x)\,y' + q(x)\,y = 0,

suppose you already know one solution y_1(x). We look for a partner of the form

y_2 = v(x)\,y_1(x),

where v(x) is an unknown function to be found. If v were merely a constant we'd just get a multiple of y_1 — nothing new — so the whole game is to let v genuinely vary with x. Substitute this guess and something wonderful happens: the equation collapses from second order to first order. That is where the method gets its name. Unlike the guess-an-exponential method for constant-coefficient equations, reduction of order works even when p and q are honest functions of x.

Why the order drops: the substitution

Let's actually do it. With y_2 = v\,y_1, the product rule gives

y_2' = v'\,y_1 + v\,y_1', \qquad y_2'' = v''\,y_1 + 2v'\,y_1' + v\,y_1''.

Feed these into y_2'' + p\,y_2' + q\,y_2 = 0 and group the terms by which derivative of v they carry:

v''\,y_1 + v'\big(2y_1' + p\,y_1\big) + v\underbrace{\big(y_1'' + p\,y_1' + q\,y_1\big)}_{=\,0} = 0.

Look at the coefficient of v (no derivatives): it is exactly the original equation evaluated at y_1 — and that is zero, because y_1 is a solution! The plain-v term vanishes, and we are left with an equation involving only v'' and v':

v''\,y_1 + v'\big(2y_1' + p\,y_1\big) = 0.

There is no bare v left — only its derivatives. So if we set w = v', this is a first-order equation in w:

w'\,y_1 + w\big(2y_1' + p\,y_1\big) = 0,

which is separable (or first-order linear). Solve it for w = v', then integrate once more to recover v, and finally y_2 = v\,y_1. Two integrations, and the second solution is yours.

Worked example 1 — the repeated root, demystified

Take y'' - 2y' + y = 0. Its characteristic equation is r^2 - 2r + 1 = (r-1)^2 = 0, a repeated root r = 1, so we know one solution: y_1 = e^{x}. The exponential method stalls here — it only produced one solution. Reduction of order rescues us. Try

y_2 = v(x)\,e^{x}.

Here p(x) = -2 and y_1 = e^x, so y_1' = e^x and 2y_1' + p\,y_1 = 2e^x - 2e^x = 0. The reduced equation v''\,y_1 + v'(2y_1' + p\,y_1) = 0 becomes simply

v''\,e^{x} = 0 \;\Longrightarrow\; v'' = 0.

Integrate twice: v' = c_1, then v = c_1 x + c_2. Drop the trivial part c_2 (it just rebuilds a multiple of y_1) and take c_1 = 1, giving v = x. Therefore

y_2 = v\,y_1 = x\,e^{x}.

So the mysterious extra factor of x in the repeated-root formula y = (C_1 + C_2 x)e^{rx} is not pulled from a hat — it is exactly what reduction of order produces. This is where that x\,e^{rx} comes from, every single time.

Worked example 2 — variable coefficients

Now a case the exponential guess cannot touch. Consider the Euler–Cauchy equation

x^2 y'' - 3x\,y' + 4y = 0,

and suppose we already know one solution y_1 = x^2 (check: x^2(2) - 3x(2x) + 4x^2 = 2x^2 - 6x^2 + 4x^2 = 0 ✓). First put the equation in the standard form y'' + p\,y' + q\,y = 0 by dividing by x^2:

y'' - \frac{3}{x}\,y' + \frac{4}{x^2}\,y = 0, \qquad p(x) = -\frac{3}{x}.

Try y_2 = v\,x^2. With y_1 = x^2, y_1' = 2x, the reduced equation v''\,y_1 + v'(2y_1' + p\,y_1) = 0 is

v''\,x^2 + v'\!\left(4x - \frac{3}{x}\cdot x^2\right) = v''\,x^2 + v'\,(4x - 3x) = v''\,x^2 + v'\,x = 0.

Set w = v': then w' x^2 + w\,x = 0, i.e. \dfrac{w'}{w} = -\dfrac{1}{x}. Integrating gives \ln|w| = -\ln|x|, so

w = v' = \frac{1}{x} \;\Longrightarrow\; v = \int \frac{1}{x}\,dx = \ln x.

Hence the second solution is

y_2 = v\,y_1 = x^2 \ln x,

and the general solution is y = C_1 x^2 + C_2 x^2 \ln x. Notice the \ln x: reduction of order happily produces solutions of a completely different shape from y_1 — something no exponential-guess method would ever have found.

Given one known solution y_1(x) of y'' + p(x)\,y' + q(x)\,y = 0, a second independent solution is y_2 = v(x)\,y_1(x), where:

Substituting collapses the ODE to a first-order equation in w = v' (the bare-v term drops because y_1 solves the ODE): \;w'\,y_1 + w\,(2y_1' + p\,y_1) = 0.
Solving gives the compact closed form y_2 = y_1(x)\int \frac{e^{-\int p(x)\,dx}}{y_1(x)^2}\,dx.
The pair y_1, y_2 is linearly independent, so the general solution is y = C_1 y_1 + C_2 y_2.

It feels like getting something for nothing. The secret is that a second-order linear equation only has a two-dimensional space of solutions, and it is very structured: once you pin down one direction y_1 in that space, the equation itself constrains the other direction almost completely. Writing y_2 = v\,y_1 is like saying "the second solution can only differ from the first by a slowly changing scale factor v" — and the ODE, robbed of one whole degree of freedom by knowing y_1, has only enough left to determine a first-order equation for that factor. That is exactly why the order drops by one. The same reasoning powers variation of parameters, which lets all the constants vary to attack non-homogeneous equations.

You need a known solution to start. Reduction of order is a bootstrapping method, not a from-scratch solver — no y_1, no method. It converts "one solution" into "two", but it cannot conjure the first one out of nothing.
The second solution is v\cdot y_1, not v alone. A very common slip is to solve for v and stop. In example 1 that would leave you with x, which is not a solution of the ODE — the solution is x\,e^{x}. Always multiply back by y_1.
Drop the trivial constant. Integrating v' throws up a constant of integration; that piece rebuilds a plain multiple of y_1 and adds nothing new. Keep only the genuinely new, non-constant part of v.
Put the equation in standard form first. The reduction formula and the value of p(x) assume the leading coefficient of y'' is 1. In example 2 we had to divide by x^2 before reading off p = -3/x.

See the two solutions, and their family

Here are the two solutions of y'' - 2y' + y = 0 from example 1: the known y_1 = e^{x} and the one reduction of order handed us, y_2 = x\,e^{x} (faint). The bold curve is the general solution (C_1 + C_2 x)\,e^{x} — slide C_1 and C_2 to blend them. Notice that y_2 is genuinely a different shape from y_1 (it dips negative for x < 0): that difference is exactly the linear independence that makes it a legitimate second solution rather than a rescaled copy.