Promote d/dx to an object in its own right
Until now you have written y', y'', y''' as things that happen TO a function. The first move of this guide is a change of viewpoint: package the act of differentiating into a single symbol D, so that D y means y', D^2 y means y'', and in general D^k y means y^(k). The letter D no longer points at any particular function — it is an operator, a machine that eats one function and spits out another. This linear differential operator is linear in the honest sense: D(c1 f + c2 g) = c1 D f + c2 D g, so it distributes over sums and pulls constants straight out.
With D in hand, a whole constant-coefficient equation collapses into one tidy expression. The equation a y'' + b y' + c y becomes (a D^2 + b D + c) y, where the chunk in parentheses, L = a D^2 + b D + c, is itself an operator built by adding scaled copies of D. We have quietly turned a differential equation into the statement L y = (forcing) — the operator L applied to the unknown equals the right-hand side. The entire game from here is learning to manipulate L as if it were an algebraic creature.
The polynomial in D: factor it and the answer falls out
Here is the payoff for the bookkeeping. Because L = a D^2 + b D + c is a polynomial in D, it factors over its roots exactly like a^2 + b a + c factors over numbers. If the characteristic polynomial a r^2 + b r + c has roots r1 and r2, then L = a (D - r1)(D - r2) — and crucially the order of those factors does not matter, because constant-coefficient operators commute. The roots of the characteristic polynomial and the factors of the operator are the very same numbers wearing two costumes.
Each linear factor solves a tiny first-order problem you already know. The equation (D - r) y = 0 is just y' = r y, whose solution is y = e^(rt). So a factor (D - r) of L 'announces' the building block e^(rt) in the homogeneous solution. Watch a repeated factor: (D - r)^2 y = 0 means applying (D - r) twice gives zero, and that is exactly the situation whose extra solution is t e^(rt). This is where the mysterious t-factor of the repeated-root solution forms finally has a clean origin — a root repeated m times comes from a factor (D - r)^m, which contributes e^(rt), t e^(rt), ..., t^(m-1) e^(rt).
Notice we have re-derived the entire higher-order homogeneous theory of the previous two guides without a single new idea — just by factoring the operator and reading each factor as a first-order piece. That is the test of a good viewpoint: it makes old results look inevitable. But the real reason to set all this up is what it does for NONhomogeneous equations, and that is where the annihilator enters.
The annihilator: an operator that kills the forcing term
Suppose you face a nonhomogeneous equation L y = g(x), where g is the forcing term. The key observation is simple and a little playful: many of the functions g you meet in practice — polynomials, e^(ax), sin(bx), cos(bx), and their products — are themselves solutions of some constant-coefficient homogeneous equation. That means there is an operator A such that A g = 0. Such an A is an annihilator of g: apply it, and the forcing term vanishes into thin air.
The annihilators are a short, memorizable dictionary, and every entry is just 'which factor produces this function as a homogeneous solution'. A polynomial of degree m is killed by D^(m+1); the exponential e^(ax) is killed by (D - a); the pair sin(bx), cos(bx) is killed by (D^2 + b^2); and a product like x e^(ax) is killed by (D - a)^2. You do not memorize these as arbitrary facts — you read them straight off the root-to-factor correspondence you just built. The table below is the whole toolkit.
forcing term g(x) annihilator A (because g solves A y = 0) ----------------------------- -------------------- ------------------------- c (a constant) D root r = 0 x^m (polynomial deg m) D^(m+1) root r = 0, mult. m+1 e^(ax) D - a root r = a x^m e^(ax) (D - a)^(m+1) root r = a, mult. m+1 cos(bx), sin(bx) D^2 + b^2 roots r = +/- i*b e^(ax) cos(bx), e^(ax)sin(bx) (D - a)^2 + b^2 roots r = a +/- i*b
The method, step by step
Now combine the two ideas. To solve L y = g, apply the annihilator A of g to BOTH sides. The right side dies (A g = 0), and you are left with a single, larger, purely homogeneous equation A L y = 0. You can solve that by the characteristic-polynomial machinery you already own — and its general solution must contain every piece of the true answer to the original problem, both the complementary solution (from L's own roots) and the particular solution (from A's roots). The annihilator method is exactly this trick of trading one nonhomogeneous problem for one bigger homogeneous one.
- Solve the homogeneous equation L y = 0 first to get the complementary solution y_c — you need its roots to spot overlaps later.
- Pick the annihilator A of the forcing term g from the dictionary, so that A g = 0.
- Apply A to both sides to form A L y = 0, then list the full general solution of this larger homogeneous equation from its characteristic roots.
- Throw away the terms that already appear in y_c; the survivors are the correct form for the particular solution y_p, with undetermined coefficients.
- Substitute that y_p back into L y = g and match coefficients to pin the constants down; the full answer is y = y_c + y_p.
Make it concrete with y'' - y = e^(x), that is (D^2 - 1) y = e^(x). Here L = (D - 1)(D + 1), so y_c = C1 e^(x) + C2 e^(-x). The forcing e^(x) is annihilated by A = (D - 1), giving (D - 1)^2 (D + 1) y = 0. Its roots are r = 1 (now doubled!) and r = -1, so the enlarged solution is C1 e^(x) + C2 e^(-x) + C3 x e^(x). The first two terms are already y_c, so the only genuinely new survivor is the particular form y_p = A x e^(x) — and the extra x appeared automatically because e^(x) was already a homogeneous solution.
What it explains, and where it stops
The deepest gift of this method is that it demystifies the method of undetermined coefficients. Where that method told you to GUESS a trial form and then handed you a modification rule — 'multiply by x if your guess clashes with the homogeneous solution' — the annihilator method DERIVES that rule. The clash is just a repeated root in A L: when the forcing's root coincides with one of L's roots, the factor's power climbs, and a higher power of that factor is precisely what produces the extra x. The famous 'multiply by x' is not a patch; it is what the bigger characteristic polynomial says must happen.
Be honest about the wall this method runs into. The annihilator only exists when g is itself a solution of some constant-coefficient homogeneous equation — that is, a polynomial, an exponential, a sine or cosine, or a finite product and sum of these. The moment the forcing is something like tan(x), ln(x), or 1/x, NO finite annihilator exists, the whole approach is silent, and you must fall back on variation of parameters — the genuinely universal method that the final guide of this rung is devoted to. The annihilator method is sharp and elegant, but it is a specialist, not a generalist.