Why we want a system in the first place
Up to now a problem meant one unknown function and one equation — a single y(x) obeying something like a y'' + b y' + c y = 0. But the real world rarely hands you one lonely variable. Two carts joined by springs have two positions to track; a predator and its prey have two populations rising and falling together; an electrical network has several currents that all push on each other. The honest model is several unknowns at once, each one's rate of change depending on the others. That is a system, and this rung is about solving the linear ones.
The defining feature is that the unknowns are tangled together — they are coupled unknowns. You cannot solve for the first one, finish it off, and only then turn to the second, because the second is sitting inside the first one's equation, and vice versa. Picture two people on a seesaw: neither one's height makes sense without the other's. To untangle them we will eventually borrow eigenvalues and eigenvectors from linear algebra — but before any of that machinery can run, we need every problem written in one standard shape. Producing that shape is the entire job of this first guide.
The standard shape: first order only
Here is the surprising promise we will cash in for the rest of the rung: no matter how high the order, every linear ODE can be rewritten using only first derivatives. A messy third-order equation, a coupled pair of second-order ones — all of it collapses into a flat list where the left side of each equation is just a single prime. We call that list a first-order system, and the price of admission to all the elegant theory ahead is getting your problem into exactly this form.
Why insist on first order? Because once everything is first-order, you can stack all the unknowns into a single vector x and all the right-hand sides into a single matrix A, and the entire system becomes one breathtakingly compact statement: x' = A x. That vector-matrix form looks exactly like the scalar growth law y' = k y you met long ago — and that resemblance is not a coincidence, it is the whole strategy. Everything you knew about a single exponential is about to be promoted, one dimension at a time, into the language of matrices.
The trick: name the derivatives
The whole conversion rests on a single, almost embarrassingly simple idea: give each derivative its own name. Suppose you have a second-order equation, say y'' + 3 y' + 2 y = 0. The reason it is second order is that y'' appears. So introduce a brand-new unknown for the velocity: let x1 = y and let x2 = y'. You have not changed the problem one bit — you have only handed the slope a name of its own. Now watch the two first-order equations fall out almost for free.
The first equation is true by sheer definition: since x2 was defined to be y', and x1 was defined to be y, we have x1' = y' = x2. The second equation is where the original ODE earns its keep — we solve it for the highest derivative, y'' = -3 y' - 2 y, and translate every symbol into the new names: x2' = y'' = -3 x2 - 2 x1. Two first-order equations, exactly as promised, and together they say precisely what the single second-order equation said. This bookkeeping move is the conversion to a first-order system, and it never fails for a linear equation.
Start (one 2nd-order eq): y'' + 3 y' + 2 y = 0
Name the derivatives: x1 = y , x2 = y'
The system (two 1st-order): x1' = x2
x2' = -2 x1 - 3 x2
In vector-matrix form: x' = A x , with
| x1 | | 0 1 |
x = | x2 | , A = | -2 -3 |Climbing to order n
Nothing about the trick was special to order two — it scales straight up, and that is exactly the connection between a scalar nth-order equation and a system. For an order-n equation you name the whole staircase of derivatives: x1 = y, x2 = y', x3 = y'', and so on up to xn = y^(n-1). The first n-1 equations are then pure definitions, each one declaring that the next variable is the derivative of the one before — x1' = x2, x2' = x3, all the way up. They simply pass the derivative down the line like a bucket brigade.
Only the very last equation, xn', carries any real content: it is the original ODE itself, solved for y^(n) and rewritten in the new names. So an order-n scalar equation becomes an n-by-n linear system of ODEs, x' = A x, where A has a stack of 1's just above its diagonal (the definitions) and a single bottom row built from the original coefficients (the physics). This is more than a convenience — it is why we are entitled to study every linear ODE through systems. The two pictures are genuinely the same object, and any truth proved for x' = A x is automatically a truth about the scalar equation it came from.
A new picture: motion in the phase plane
Recasting the problem as x' = A x does more than tidy the algebra — it changes what a solution *looks like*. A scalar solution y(x) was a curve you graphed against x. But a vector solution is a point that *moves*: at each instant, x is a single point in a plane (for a 2-by-2 system) whose coordinates are y and y'. As time runs, that point traces a path. The arena of all such points is the phase plane, and the path a solution carves through it is its trajectory.
Read x' = A x once more as a sentence about the plane: at the point x, the matrix A tells you the velocity arrow x'. So A turns every location into a little arrow, and a solution is simply a point flowing along, always obeying the arrow under its feet. A spring's oscillation, dull as a wiggling graph in y, becomes a clean loop spiralling around the origin in the phase plane. This geometric view is the soul of the whole subject of systems, and you will lean on it constantly once eigenvectors arrive to mark out the special straight-line directions of the flow.
What this unlocks next
With every problem now wearing the single uniform x' = A x, the rest of the rung is one focused campaign: solve that equation. The strategy is the analogy we already flagged. The scalar law y' = k y has solution y = c * e^(kt) — a constant times an exponential. So we *guess* that x' = A x has solutions of the same flavour, x = v * e^(lambda t), where v is now a constant vector pointing in a fixed direction and lambda a number setting the growth or decay rate. Substituting that guess is what summons the eigenvalue method, and it is the engine of the next guide.
Plugging x = v * e^(lambda t) into x' = A x will boil the entire differential equation down to the single algebraic demand A v = lambda v — the eigenvalue equation. From there the rung splits into the three cases that organize everything after: real distinct eigenvalues that give honest straight-line solutions, complex eigenvalues that bend the motion into rotation, and repeated eigenvalues that force us to invent generalized eigenvectors. You have now built the one bridge they all stand on. Carry one sentence forward: a higher-order linear equation and a first-order system are the same creature, and the system form is the one that lets linear algebra finish the job.