When the Quadratic Refuses to Factor
In the previous guide you met the characteristic equation a r^2 + b r + c = 0, the quadratic you read off a constant-coefficient equation a y'' + b y' + c y = 0 by guessing y = e^(rt). Two distinct real roots gave you two clean exponentials; a repeated root gave you e^(rt) and t e^(rt). But a quadratic does not always have real roots. When the discriminant b^2 - 4ac is *negative*, the formula r = (-b ± sqrt(b^2 - 4ac)) / (2a) asks you to take the square root of a negative number — and the real line has no answer.
Take the cleanest example: y'' + y = 0, the equation that says "my second derivative is the negative of myself". Its characteristic equation is r^2 + 1 = 0, so r^2 = -1. No real number squares to -1. Yet you already *know* this equation has solutions — sin(t) and cos(t) both have the property that differentiating twice flips the sign. So real-world wiggles are sitting right there, even though the algebra has gone imaginary. The two stories must be reconciled.
The fix is not to give up on e^(rt) but to allow r to be a complex number. The same substitution still works; we just let the roots live in the complex plane. When b^2 - 4ac < 0 the two roots come as a complex-conjugate pair r = alpha ± i beta, where alpha = -b / (2a) is real and beta = sqrt(4ac - b^2) / (2a) is the (real, positive) leftover. The two roots are mirror images across the real axis. The whole trick of this guide is to translate those two complex exponentials back into the language of real oscillations.
Euler's Formula: The Bridge Back to the Real World
The single fact that makes everything work is Euler's formula: e^(i theta) = cos(theta) + i sin(theta). It says that raising e to an imaginary power is not exotic at all — it is exactly a cosine plus an imaginary sine. You can almost feel why it must be true: differentiating cos(theta) + i sin(theta) with respect to theta gives -sin(theta) + i cos(theta), which equals i times the original. The only function whose derivative is i times itself is e^(i theta). So the exponential and the trig pair are the *same object* wearing two costumes.
Now feed a complex root into e^(rt). With r = alpha + i beta we get e^((alpha + i beta) t) = e^(alpha t) · e^(i beta t) = e^(alpha t) (cos(beta t) + i sin(beta t)). The real exponential e^(alpha t) controls overall growth or decay; the e^(i beta t) factor, unpacked by Euler, is the pure oscillation at angular frequency beta. The conjugate root r = alpha - i beta gives the same thing with the sine flipped: e^(alpha t) (cos(beta t) - i sin(beta t)).
We want *real* solutions, not complex ones. The superposition principle rescues us: any linear combination of solutions is again a solution, and we are free to combine the two complex exponentials so the imaginary parts cancel. Adding the conjugate pair and dividing by 2 leaves e^(alpha t) cos(beta t); subtracting and dividing by 2i leaves e^(alpha t) sin(beta t). Both are genuinely real, and both solve the original equation. The complex numbers were just scaffolding — once the building stands, we take them away.
Reading the General Solution
Putting it together, the complex-roots case has a single clean recipe. When the characteristic equation gives r = alpha ± i beta, the general solution is y = e^(alpha t) (C1 cos(beta t) + C2 sin(beta t)). The two pieces e^(alpha t) cos(beta t) and e^(alpha t) sin(beta t) are linearly independent — their Wronskian never vanishes — so they form a valid pair to span the two-dimensional solution space, exactly as guides 1 and 2 demanded. The constants C1 and C2 are the two free parameters you fix with initial conditions y(0) and y'(0).
discriminant b^2 - 4ac roots general solution ---------------------------------------------------------------------- > 0 (positive) r1, r2 real, distinct C1 e^(r1 t) + C2 e^(r2 t) = 0 (zero) r real, repeated (C1 + C2 t) e^(r t) < 0 (negative) alpha +/- i beta e^(alpha t)(C1 cos(beta t) + C2 sin(beta t))
Look at what alpha and beta each control. The sign of alpha decides the envelope: alpha < 0 means the oscillation decays inside a shrinking exponential cushion (this is damped vibration), alpha > 0 means it grows without bound, and alpha = 0 means it neither grows nor decays — a steady, undying wiggle. The value of beta sets how fast it oscillates. So a single complex root carries two independent meanings at once: its real part is a growth rate, its imaginary part is a frequency.
Amplitude and Phase: One Wave, Not Two
The form C1 cos(beta t) + C2 sin(beta t) is correct but looks like two separate waves. It is really *one* shifted cosine. There is a tidy identity: C1 cos(beta t) + C2 sin(beta t) = R cos(beta t - phi), where R = sqrt(C1^2 + C2^2) and the angle phi satisfies tan(phi) = C2 / C1. The number R is the amplitude — the peak height of the swing — and phi is the phase, a horizontal shift that says where in its cycle the wave starts. This amplitude-phase form is how engineers and physicists actually read the answer.
The cleanest instance is alpha = 0, where the exponential envelope is just 1 and the solution is a pure R cos(beta t - phi) that swings forever at the same height. This is [[simple-harmonic-motion|simple harmonic motion]] — a mass on a frictionless spring, a pendulum at small angles, an ideal LC circuit. Here beta is the system's natural frequency: the rate at which it *wants* to oscillate when left alone, set entirely by the equation's coefficients, not by how hard you started it. For m y'' + k y = 0 it works out to beta = sqrt(k / m) — stiffer spring, faster swing; heavier mass, slower swing.
Add a little friction and alpha becomes slightly negative: the cosine now lives under a shrinking roof e^(alpha t), so each swing is a touch smaller than the last. That is underdamped motion — a plucked guitar string fading, a child's swing left to coast — the most musical of the three cases. Crucially, this happens whenever damping is *light* (b^2 < 4ac); you do not need zero friction to oscillate, only little enough that the system overshoots its rest point before settling. Heavier damping eventually kills the wiggle entirely and pushes you into the real-root cases of the previous guide.
A Worked Example and the Honest Caveats
Let us run the whole machine on y'' + 2 y' + 5 y = 0 with y(0) = 0 and y'(0) = 4. Watch the three numbers alpha, beta, and the constants fall out in order, and notice how every step echoes the structure from the earlier guides.
- Write the characteristic equation: r^2 + 2 r + 5 = 0. Its discriminant is 4 - 20 = -16 < 0, so we are in the complex-roots case.
- Find the roots: r = (-2 ± sqrt(-16)) / 2 = -1 ± 2 i. So alpha = -1 (decay) and beta = 2 (frequency).
- Write the general solution: y = e^(-t) (C1 cos(2 t) + C2 sin(2 t)). A decaying oscillation — underdamped.
- Apply y(0) = 0: at t = 0 the exponential is 1, cos is 1, sin is 0, so y(0) = C1 = 0.
- Differentiate and apply y'(0) = 4. With C1 = 0 this forces 2 C2 = 4, so C2 = 2. The answer is y = 2 e^(-t) sin(2 t).