4.6.14 · D2Ordinary Differential Equations

Visual walkthrough — Non-homogeneous — method of undetermined coefficients (annihilator method)

2,192 words10 min readBack to topic

Step 0 — The words we need before we start

Before a single symbol appears, let me name the three things we will draw over and over.


Step 1 — Picture the equation as "machine eats , spits out "

WHAT. We are solving with and .

WHY. Every method in this chapter is a way of inverting : we know the output , we want the input . Drawing as a box makes the goal concrete — find the arrow that goes backwards.

PICTURE. On the left, the input (unknown, a question mark). The box transforms it. On the right, the known output (the amber curve). Our job is the dashed backward arrow.

The general solution splits as (this is the Superposition principle for linear ODEs): is everything the machine crushes to zero, is one input that hits the target .


Step 2 — Find : the inputs the machine crushes to zero

WHAT. Solve , i.e. . Factor: , so the roots are and .

WHY. The homogeneous solutions are the machine's "silent inputs" — feed them in and nothing comes out. We must know them first, because later we will have to avoid copying them.

PICTURE. Two seeds on the root-number-line, at and . Each seed grows upward into its function: and .

The are free constants — knobs the initial conditions will later set. Right now they just mark which functions the machine annihilates.


Step 3 — Read backwards to build its annihilator

WHAT. We want the smallest machine with . Our is grown from the single root . Reading that backwards, the seed comes from the factor . So .

WHY. This is the heart of the method. If crushes to zero, then applying to both sides of our equation will make the right-hand side vanish — turning a hard non-homogeneous problem into an easy homogeneous one.

PICTURE. The output curve enters the small machine and comes out flat at zero. Let us verify the crush on the picture and in symbols:

The first term is ; the second is times . They cancel exactly — that cancellation is why is the annihilator.


Step 4 — Apply to the whole equation: both sides now homogeneous

WHAT. Hit every part of with :

WHY. The right side is now zero — because was designed to kill . We have paid a price: the equation is now higher order. But higher-order homogeneous with constant coefficients is something we can always solve by listing roots. We traded "hard, small" for "easy, big."

PICTURE. The equation collapses. absorbs on the right; on the left the operators stack up. Group the repeated factor:

Notice the amber alarm bell: the factor now appears twice. The root was already in (from ), and added another copy. That doubling is the whole story — hold that thought for Step 6.


Step 5 — List every root of the big equation

WHAT. The big operator has roots: (once), and (twice — a repeated root).

WHY. We need the general solution of the big homogeneous ODE, and its shape is dictated entirely by these roots. A simple root gives ; a repeated root of multiplicity gives and — this is the rule from Characteristic equation and repeated roots.

PICTURE. Three seeds on the number line: one at , and a stacked pair at . The stacked pair grows into two functions, (short) and (taller — the makes it climb).

The first two terms are exactly from Step 2. Only the last term, , is genuinely new.


Step 6 — Keep only the new term: that IS the shape of

WHAT. Discard and (they are ). The survivor is the trial form:

WHY — the deepest "why" on this page. The terms from the roots of satisfy ; feed them to the machine and you get zero, never . So they cannot produce the target. Only the roots that newly contributed can hit . Here 's contribution was the second copy of root , and a second copy always shows up as an extra . That is why overlap forces the factor — not by a memorized rule, but because the number line literally stacked two seeds at the same spot.

PICTURE. Two piles of terms. The grey pile () is labelled "already — throw away." The amber pile () is labelled "NEW — this is ."


Step 7 — Tune the dial inside the real machine

WHAT. Substitute into the original and solve for .

WHY. The annihilator gave us the shape but not the size. To fix the size we return to the true machine and demand the output equal exactly.

PICTURE. A dial marked ; we rotate it until the output curve (cyan) lands exactly on the target (amber). Compute the derivatives term by term:

Now assemble :

The -terms cancel to zero (they must — that piece was a silent input), leaving a clean constant:

So , and the full answer:


Step 8 — The degenerate & edge cases, on one number line

WHAT. What if there were no overlap? What if root were already double in ? The number-line picture answers both at a glance.

WHY. A method you only trust in one case is not a method. Every scenario is just "how many seeds already sit at the target root, and how many does add."

PICTURE. Three mini number-lines, side by side:

  • No overlap (e.g. , root not in ): plants a fresh seed at , multiplicity → trial , no .
  • Simple overlap (our example, root once in ): stack becomes → trial , one .
  • Double overlap (if had root twice already): makes it → trial , .

The one-picture summary

Everything above, compressed: the equation flows from hard non-homogeneous (top) through the annihilator into an easy homogeneous problem (middle), whose root-line is read off (bottom), and only the new seed survives as .

Recall Feynman retelling — explain the whole walkthrough to a friend

You have a machine that eats a function and spits out . You want the input. First, list the machine's silent inputs — the ones it turns into pure zero: here and . That's . Now look at the output and ask: "what tiny machine would grind down to nothing?" Since grows from the seed , that tiny machine is . Run the whole equation through . The right side becomes zero — the problem is now homogeneous, just longer. List its seeds: , and twice, because the seed was already there and dropped another on top. Two stacked seeds always sprout an extra , so the new plant is . The old plants are just — throw them out. The new plant is the shape of your answer. Feed into the real machine and turn the dial until the output hits exactly — it lands at . Full answer: . The extra was never a rule to memorize; it was two seeds standing in the same place.


Active recall

Recall Why does the right-hand side become zero after applying

? Because is built to annihilate : by construction, so .

Where does the factor come from geometrically?
The overlapping root gets a second stacked seed on the root-line; a doubled root of multiplicity 2 sprouts and , and the new term carries the .
Which terms of the big homogeneous solution do we discard?
The ones matching the roots of — they are , satisfy , and cannot produce .
If and is not a root of , what is the trial?
(no , because the seed at 3 is fresh, multiplicity 1).