Second-order linear ODEs — superposition principle, general theory: Visual walkthrough

We will use a few symbols. Before we do, here is the promise for each:

$y$ is a function — a machine that takes a number $x$ and returns a number $y (x)$ . Think of it as a curve drawn on paper.
$y^{'}$ means "the slope of that curve" (how steep it is) and $y^{''}$ means "how fast the slope itself is changing" (the bending). We earn these fully in Step 1.
$L$ is a machine that eats a function and returns a function. We build it in Step 2.

Let's go from zero.

Step 1 — What $y^{'}$ and $y^{''}$ actually look like

WHY we need them. An "ODE" is just a rule that ties a curve's height, slope, and bending together. To read such a rule you must first see those three quantities on one picture. Everything after this depends on recognising them.

PICTURE. In the figure, the blue curve is $y = cos x$ . At the marked point the short orange segment is the tangent — its tilt is $y^{'}$ . The green shading shows where the curve bends downward ( $y^{''} < 0$ ) versus upward ( $y^{''} > 0$ ).

Figure — Second-order linear ODEs — superposition principle, general theory

Step 2 — Packaging the ODE into one machine $L$

Each symbol, right where it sits:

$y^{''}$ — the bending of the curve (from Step 1).
$p (x)$ — a given function; a dial that weights how much the slope contributes.
$y^{'}$ — the slope.
$q (x)$ — a given function; a dial weighting the height.
$y$ — the height of the curve.

WHY box it. Because then "solving the ODE" becomes the single sentence: find every function $y$ that the machine sends to the target $g$ . And the machine has a magic property we can draw.

PICTURE. The figure shows $L$ as a box: a curve $y$ goes in the left, a new curve $L [y]$ comes out the right. Feeding in different curves gives different outputs.

Step 3 — The magic property: $L$ is linear

Term by term:

$c_{1}, c_{2}$ — plain numbers (scaling factors, e.g. "make this curve twice as tall").
$y_{1}, y_{2}$ — two input curves.
Left side: feed the combined curve into the machine.
Right side: feed each curve separately, then combine the outputs.
The equals sign is the whole claim: doing it either way lands on the same output.

WHY it's true. Because differentiation itself already obeys this: the slope of a sum is the sum of the slopes, and scaling a curve scales its slope. The parent note shows the three-line algebra; here we just see it.

PICTURE. Left panel: build the input by stretching $y_{1}$ by $c_{1}$ , stretching $y_{2}$ by $c_{2}$ , and stacking them — then run the machine. Right panel: run each through first, then stack. The two red output curves coincide exactly.

This one identity — see Linear algebra — vector spaces and bases for the general idea of a "linear map" — is the seed of everything below.

Step 4 — Superposition: solutions add up

The two under-braces are zero because $y_{1}, y_{2}$ are assumed to be solutions.
Therefore the whole thing is zero, so $c_{1} y_{1} + c_{2} y_{2}$ is also a solution, for any numbers $c_{1}, c_{2}$ .

WHY this is huge. It means solutions can be mixed in any proportion and you stay a solution. Geometrically: the solutions live on a flat sheet through the origin — a plane.

PICTURE. Each solution is drawn as an arrow from the origin in an abstract "solution space". $y_{1}$ points one way, $y_{2}$ another. Every dot $c_{1} y_{1} + c_{2} y_{2}$ on the shaded plane is again a valid solution. Nothing escapes the plane.

Step 5 — Why we need independence (the parallel-arrows trap)

$c_{1} + 3 c_{2}$ collapses two knobs into one effective knob.
So you'd be missing half the solutions.

WHY it matters. A second-order ODE needs two free constants (Step 8 explains why). Parallel solutions give only one. We call two non-parallel solutions linearly independent.

PICTURE. Left: two parallel arrows — all combinations stay on one line (bad). Right: two spread-apart arrows — combinations fill the plane (good).

Step 6 — The Wronskian: a machine that detects parallelness

$W$ is called the Wronskian.
$W (x_{0}) \neq = 0$ at even one point $\Rightarrow$ no hidden relation $\Rightarrow$ arrows are independent.
$W$ measures how much $y_{1}, y_{2}$ "spread apart" in the height-and-slope plane.

WHY a determinant. The quantity $y_{1} y_{2}^{'} - y_{2} y_{1}^{'}$ is exactly the signed area of the parallelogram made by the two vectors $(y_{1}, y_{1}^{'})$ and $(y_{2}, y_{2}^{'})$ . Zero area = squashed flat = parallel = dependent. Non-zero area = genuine spread = independent. See Linear algebra — vector spaces and bases.

PICTURE. The two vectors $(y_{1}, y_{1}^{'})$ and $(y_{2}, y_{2}^{'})$ drawn tail-to-origin; the shaded parallelogram between them; its area is $∣ W ∣$ . When they line up the parallelogram collapses to a segment of area $0$ .

Step 7 — Edge case: all-or-nothing (Abel's theorem)

The exponential of any real number is strictly positive — it never touches zero.
So $W (x)$ is $W (x_{0})$ multiplied by a strictly positive factor: if $W (x_{0}) \neq = 0$ then $W$ is never zero; if $W (x_{0}) = 0$ then $W$ is always zero.

WHY we care. You only ever check the Wronskian at one convenient point. See Abel's theorem and First-order linear ODEs for where that little equation comes from.

PICTURE. The exponential envelope curve $exp (- \int p)$ stays strictly above the axis; $W (x)$ (blue) is that curve scaled by a constant, so it either lives entirely above or entirely below the axis — it cannot cross.

Step 8 — Why exactly two constants, and the forced case

$y - y_{p}$ is homogeneous $\Rightarrow$ equals some $c_{1} y_{1} + c_{2} y_{2}$ .
So every forced solution = (the whole homogeneous plane) shifted by the single vector $y_{p}$ .

WHY it fails to be a plane through the origin. The forced solutions form the same plane, lifted off the origin by $y_{p}$ — an affine plane. Two initial conditions $y (x_{0}) = a, y^{'} (x_{0}) = b$ then pin one exact point (guaranteed unique by Existence and uniqueness theorems for ODEs).

PICTURE. The blue homogeneous plane passes through the origin; the orange forced plane is a parallel copy shifted by the green vector $y_{p}$ . A red dot marks the unique solution selected by the initial conditions.

Where each piece comes from in practice: $y_{1}, y_{2}$ from the Characteristic equation — constant coefficient ODEs, and $y_{p}$ from Method of undetermined coefficients or Variation of parameters.

The one-picture summary

The full logical chain: linearity of $L$ $\Rightarrow$ solutions add (a plane) $\Rightarrow$ need two independent arrows $\Rightarrow$ Wronskian detects independence $\Rightarrow$ Abel makes it all-or-nothing $\Rightarrow$ forced case is the same plane shifted by $y_{p}$ .

Recall Feynman retelling — say it like a story

The ODE is a machine. Feed it a curve, it hands you back a curve. The machine is fair: stretch the input, the output stretches the same; add two inputs, the outputs add. Because it's fair, if two curves each come out as flat zero, then any mix of them also comes out zero — so all the "zero solutions" live on a flat sheet. To lay out that whole sheet you need two curves that don't point the same way; if one is just a scaled copy of the other you only get a line and you're missing half. To check they truly point different ways, compute the little area between the vectors $(y_{1}, y_{1}^{'})$ and $(y_{2}, y_{2}^{'})$ — that's the Wronskian. If that area is non-zero anywhere, Abel's rule says it's non-zero everywhere, so one check suffices. Finally, if the machine is asked to output something non-zero, find just one curve that works and slide the whole flat sheet over to sit on it; two initial conditions then poke a finger onto the exact point you want.

Recall Test yourself

Why can you scale-and-add homogeneous solutions but not forced ones? ::: For $L [y] = 0$ , $L [c_{1} y_{1} + c_{2} y_{2}] = c_{1} \cdot 0 + c_{2} \cdot 0 = 0$ . For $L [y] = g$ , you'd get $(c_{1} + c_{2}) g$ , which equals $g$ only if $c_{1} + c_{2} = 1$ — so free scaling breaks. What does the Wronskian geometrically measure? ::: The signed area of the parallelogram formed by the vectors $(y_{1}, y_{1}^{'})$ and $(y_{2}, y_{2}^{'})$ ; zero area = parallel = dependent. Why is checking $W$ at one point enough for ODE solutions? ::: Abel: $W (x) = W (x_{0}) exp (- \int p)$ , and the exponential is never zero, so $W$ is all-or-nothing. What shape do the forced solutions form? ::: The homogeneous plane through the origin, shifted (translated) by any one particular solution $y_{p}$ — an affine plane.