One lucky guess that turns calculus into algebra
From the first two guides in this rung you already know the shape of the answer: a homogeneous second-order equation has a two-dimensional solution space, so if you can find two solutions that are linearly independent — confirmed by a nonzero Wronskian — their combinations give every solution. What you do not yet have is a way to actually FIND those two solutions. This guide supplies it for the most important special case, and it does so with one astonishingly cheap trick.
Restrict attention to a constant-coefficient equation a y'' + b y' + c y = 0, where a, b, c are plain numbers, not functions of x. The trick is to GUESS that the solution is an exponential, y = e^(rt), for some constant r yet to be found. Why this guess? Because the exponential is the one function whose derivative is just a copy of itself times a constant: differentiating e^(rt) never produces a new shape, only a new factor of r. So all three terms of the equation will share the same e^(rt), and it will factor out.
Watch it happen. With y = e^(rt) we get y' = r e^(rt) and y'' = r^2 e^(rt). Substitute: a r^2 e^(rt) + b r e^(rt) + c e^(rt) = 0. Every term carries the factor e^(rt), which is never zero, so we may divide it out and the calculus simply evaporates. What is left is a r^2 + b r + c = 0 — an ordinary quadratic in r. The differential equation has been traded for an algebra problem you have solved since school.
The characteristic equation and its discriminant
That quadratic a r^2 + b r + c = 0 is the characteristic equation of the differential equation (some books call it the auxiliary equation — same object). It carries all the information about the unforced behaviour of the system. There is a clean dictionary at work: the coefficients of the equation become the coefficients of the quadratic, term for term. You can write the characteristic equation down by inspection — replace y'' by r^2, y' by r, and y by 1 — without any substitution at all once you trust the pattern.
a y'' + b y' + c y = 0 -> a r^2 + b r + c = 0
roots: r = ( -b +/- sqrt(b^2 - 4ac) ) / (2a)
discriminant D = b^2 - 4ac decides everything:
D > 0 two distinct real roots r1, r2
D = 0 one repeated real root r
D < 0 complex conjugate pair alpha +/- i*betaThe quadratic formula gives r = (-b +/- sqrt(b^2 - 4ac)) / (2a), and the single quantity under the square root, the discriminant D = b^2 - 4ac, governs the whole story. It can be positive, zero, or negative, and these three possibilities are not minor variations — they produce genuinely different KINDS of motion. A positive discriminant means pure decay or growth; zero sits on a knife edge; negative brings oscillation. The next three sections take them one at a time.
Case 1 — two distinct real roots
When D > 0 the quadratic has two different real roots r1 and r2. Each one feeds the guess back: y1 = e^(r1 t) and y2 = e^(r2 t) are both solutions. Are they independent? Yes — two exponentials with different growth rates can never be constant multiples of each other, so this is the distinct real roots case and the pair {e^(r1 t), e^(r2 t)} forms a fundamental set of solutions. The general solution is then y = C1 e^(r1 t) + C2 e^(r2 t).
Make it concrete with y'' - 5 y' + 6 y = 0. Its characteristic equation is r^2 - 5 r + 6 = 0, which factors as (r - 2)(r - 3) = 0, so r1 = 2 and r2 = 3. The general solution is y = C1 e^(2t) + C2 e^(3t). Picture it physically: this is the heavily damped regime, where any disturbance simply slides back to rest without ever swinging past it — the overdamped door-closer that eases shut rather than slamming and bouncing.
Case 2 — a repeated root, and where the second solution hides
When D = 0 the two roots collide into a single repeated value r = -b/(2a). Now the exponential trick hands you only ONE solution, y1 = e^(rt). But the solution space is two-dimensional — you are still one independent solution short, and you cannot simply reuse e^(rt) with a different constant, because C1 e^(rt) + C2 e^(rt) is just (C1 + C2) e^(rt), a single one-parameter family pretending to be two. This is the genuinely tricky repeated root case.
The missing partner turns out to be y2 = t e^(rt) — the same exponential multiplied by t. You can verify by substitution that it solves the equation, but the deeper reason it appears is the method of the very next guide in this rung, reduction of order: feed in the one solution you have, and it manufactures the second by the systematic step y2 = v(t) y1, solving a simpler equation for v. For a repeated root that procedure delivers v = t every time, which is exactly why the t factor is not a magic trick but the predictable output of an honest method.
So when r is a repeated root, the fundamental set is {e^(rt), t e^(rt)} and the general solution is y = (C1 + C2 t) e^(rt). Concretely, y'' - 4 y' + 4 y = 0 gives r^2 - 4 r + 4 = (r - 2)^2 = 0, a repeated root r = 2, so y = (C1 + C2 t) e^(2t). Physically this is the critically damped case: the fastest possible return to rest without overshoot, the razor's edge between the sluggish overdamped slide and the oscillation that case 3 is about to unleash.
Case 3 — complex roots, previewed
When D < 0 the square root of a negative number forces the roots into a complex conjugate pair, r = alpha +/- i*beta, with no real solutions to the quadratic at all. Formally the exponential trick still works and produces e^((alpha + i*beta) t) — but that is a complex-valued function, and the original equation had real coefficients, so we expect real motion. This is the complex conjugate roots case, and it is exactly where oscillation comes from: physically, D < 0 is the underdamped regime, the bell that rings and the pendulum that swings.
The bridge from the complex exponential back to honest real solutions is Euler's formula, e^(i*beta*t) = cos(beta t) + i sin(beta t), which trades the imaginary exponential for sines and cosines. The upshot, developed fully in the next guide, is the real fundamental set {e^(alpha t) cos(beta t), e^(alpha t) sin(beta t)}: an oscillation at frequency beta inside an envelope e^(alpha t) that decays when alpha < 0. We stop at the doorway here on purpose — case 3 earns its own guide.
The recipe, the limits, and the honest fine print
- Write the characteristic equation by inspection: replace y'' with r^2, y' with r, y with 1 to get a r^2 + b r + c = 0.
- Compute the discriminant D = b^2 - 4ac and read off the case: D > 0 distinct real, D = 0 repeated, D < 0 complex.
- Build the fundamental set: {e^(r1 t), e^(r2 t)}, or {e^(rt), t e^(rt)}, or {e^(alpha t) cos(beta t), e^(alpha t) sin(beta t)}.
- Combine with two constants for the general solution; if initial conditions are given, solve for C1 and C2 to pin down the particular solution.
Be honest about the boundary of this method. The whole trick rests on one assumption: the coefficients a, b, c are CONSTANTS. The instant a coefficient depends on x — as in x^2 y'' + x y' + y = 0 — the substitution e^(rt) no longer factors cleanly, the characteristic equation evaporates, and you need other tools (a Cauchy-Euler substitution, or power series). The three-case picture is gorgeous precisely because it is special; it is not a universal solver for second-order equations.
Two finer points worth carrying forward. First, the same idea scales: an nth-order constant-coefficient equation has an nth-degree characteristic polynomial, and a root repeated m times contributes the family e^(rt), t e^(rt), ..., t^(m-1) e^(rt) — the t-factor pattern of case 2 generalizes. Second, do not confuse 'discriminant zero' with 'no damping': D = 0 is the critically damped knife edge, whereas genuine undamped oscillation (b = 0) lands squarely in case 3 with purely imaginary roots. The discriminant, not the damping coefficient alone, is what sorts the cases.