The same guess, one degree higher each time
On the second-order rung the magic move was to guess y = e^(rx) for a constant-coefficient equation and watch the calculus collapse into algebra: a y'' + b y' + c y = 0 became a r^2 + b r + c = 0. Nothing about that trick cared that the order was two. Take the general nth-order linear homogeneous equation with constant coefficients, a_n y^(n) + ... + a_1 y' + a_0 y = 0, feed in the same guess, and the same collapse happens — just one degree higher.
Why it works is worth pausing on, because the reason is the whole point. Each derivative of e^(rx) just multiplies it by another r: y' = r e^(rx), y'' = r^2 e^(rx), and in general y^(k) = r^k e^(rx). So differentiating k times is the same as multiplying by r^k. Substitute, and every term carries a shared factor of e^(rx) — which is never zero — so you can divide it out and what remains is a pure polynomial equation in r. The exponential is the one function that survives differentiation unchanged in shape, and that is exactly why it is the right thing to guess.
ODE: a_n y^(n) + ... + a_1 y' + a_0 y = 0 Guess: y = e^(rx), so y^(k) = r^k e^(rx) Substitute & divide e^(rx): Characteristic poly: p(r) = a_n r^n + ... + a_1 r + a_0 = 0 Degree of p: exactly n -> n roots (counted with multiplicity)
n roots, n solutions: dimension accounting
The result of that substitution is the characteristic polynomial p(r), and its degree is exactly n — the order of the equation. Here is where a beautiful piece of bookkeeping lines up. From the previous guide you know the solution space of an nth-order linear homogeneous equation has dimension n, so you need n independent solutions. And the Fundamental Theorem of Algebra promises a degree-n polynomial has exactly n roots over the complex numbers, counted with multiplicity. The two n's are the same n, and that is no coincidence — it is the engine that makes the whole method close perfectly.
So the entire problem reduces to one sentence: find the n roots of p(r), and turn each root into a solution. If all n roots are real and distinct — call them r_1, r_2, ..., r_n — then the n functions e^(r_1 x), e^(r_2 x), ..., e^(r_n x) are independent, they form a fundamental set, and the general solution is just their weighted sum, y = c_1 e^(r_1 x) + ... + c_n e^(r_n x). This is the clean case, the direct generalization of two distinct real roots giving two exponentials at second order. The new richness — and the new care needed — appears only when roots repeat or go complex.
When a root repeats: multiplying by powers of x
Suppose a root r appears with multiplicity m — meaning the factor (r - root) shows up m times in p(r). A single root gave you e^(rx); but if you naively wrote e^(rx) down m times you would have one solution masquerading as m, and your fundamental set would be short by m - 1. At second order you met this exact snag with a repeated root and patched it by pairing e^(rx) with x e^(rx). The higher-order rule is simply that idea, extended as far as the multiplicity demands.
The rule for a root of multiplicity m is to take e^(rx) and multiply it by successive powers of x: e^(rx), x e^(rx), x^2 e^(rx), all the way up to x^(m-1) e^(rx). That is exactly m functions, they are independent, and together they cover the m-dimensional slice of the solution space that this one root is responsible for. A triple root r = 2, for instance, contributes the three solutions e^(2x), x e^(2x), x^2 e^(2x). Do this for every distinct root, real or complex, gather all the pieces, and you always land on precisely n functions — the multiplicities are guaranteed to sum to n.
Complex roots, paired and possibly repeated
When the coefficients a_0, ..., a_n are all real, any complex root r = alpha + i*beta cannot show up alone: its conjugate alpha - i*beta is forced to be a root too, with the same multiplicity. They travel in pairs. Just as at second order, you do not have to live with complex exponentials — Euler's formula e^(i*beta*x) = cos(beta*x) + i*sin(beta*x) lets you trade the conjugate pair e^((alpha+i*beta)x) and e^((alpha-i*beta)x) for the two real solutions e^(alpha*x) cos(beta*x) and e^(alpha*x) sin(beta*x). Same two-dimensional span, but written in honest real functions you can plot.
The genuinely new wrinkle at higher order is that a complex pair can itself repeat. The treatment for repeated complex roots simply stacks the two ideas you already have: if alpha plus or minus i*beta is a root of multiplicity m, you take the real pair e^(alpha*x) cos(beta*x) and e^(alpha*x) sin(beta*x) and multiply each by 1, x, x^2, ..., x^(m-1). That yields 2m real solutions — the right number, since the conjugate pair accounts for 2m roots. A fourth-order equation whose characteristic polynomial is (r^2 + 1)^2, for example, has the double pair r = plus or minus i, giving cos(x), sin(x), x cos(x), x sin(x).
The honest catch: finding the roots
Notice what this method does and does not promise. It promises that once you have the roots of p(r), assembling the general solution is mechanical and complete — the characteristic equation approach turns the differential equation entirely into the algebra of root-finding. What it does not promise is that the root-finding is easy. At second order the quadratic formula always delivers; at order three and four there are messy but real formulas; but from degree five onward, the Abel-Ruffini theorem says there is no general formula in radicals at all. So the bottleneck moves from calculus to algebra.
In practice this is far less grim than it sounds. Textbook problems are rigged so the polynomial factors nicely — you spot a rational root by the rational-root test, divide it out, and recurse, often peeling p(r) down to quadratics you can finish by hand. And when a polynomial genuinely will not factor, that is precisely what numerical root-finders are for: they hand you accurate decimal roots, which feed straight into the same solution forms. The theory stays exact; only the last step of locating the roots may go numerical, and that is a perfectly respectable place to stop.
Hold on to the through-line, because the next guides build straight on it. The polynomial p(r) is not just a root-finding gadget — it can be read as a polynomial in the differentiation operator D, and that reframing powers the operator algebra and the annihilator method coming next. The same factored form (D - r) acting on functions is what lets you handle forcing terms and, eventually, the variable-coefficient Cauchy-Euler equation. One polynomial, many doors.