From a line to a room: the same equation, more dimensions
Every earlier guide in this rung lived on a line. We derived u_tt = c^2 u_xx for a vibrating string, split any start into left- and right-moving halves with d'Alembert's formula u(x,t) = F(x - c t) + G(x + c t), and watched signals travel at the single wave speed c with a clean domain of dependence. Real waves, though, spread across drumheads and pools and the open air. The honest model just swaps the single second derivative u_xx for the Laplacian: u_tt = c^2 (u_xx + u_yy) on a membrane, u_tt = c^2 (u_xx + u_yy + u_zz) in air.
1D string : u_tt = c^2 u_xx 2D drum : u_tt = c^2 (u_xx + u_yy) = c^2 (Laplacian u) 3D air : u_tt = c^2 (u_xx + u_yy + u_zz) = c^2 (Laplacian u) Cauchy data (same in every dimension): u(x, 0) = phi(x) initial displacement u_t(x, 0) = psi(x) initial velocity
You might hope the higher-dimensional story is just d'Alembert with extra letters. It is not. The same Cauchy data — an initial displacement phi and an initial velocity psi — still pins the solution, and the equation is still hyperbolic, so it still has a finite speed of propagation. But how a localized disturbance spreads, and crucially whether it leaves a trailing wake behind, depends sharply on whether you are in one, two, or three dimensions. That contrast is the whole point of this guide.
Spherical means: averaging a wave back onto a line
How do you even solve a 3D wave equation? The classic trick refuses to attack the full mess head-on. Instead, fix a point and look at the average of u over an expanding sphere of radius r around it. This is the method of spherical means: define M(r,t) to be the mean value of u(x,t) over the sphere of radius r centred at your chosen point. Averaging tames the Laplacian — by symmetry the angular wiggles cancel — and a short calculation shows that this average obeys a one-dimensional wave equation in the radial variable r.
That is the magic: a problem in three messy dimensions becomes a problem on a half-line, where we already own d'Alembert. In 3D the radial equation works out so cleanly that r·M(r,t) satisfies the plain string equation, and unwinding the average as r → 0 recovers u at the centre. The output is a closed formula for u in terms of integrals of phi and psi over a single sphere — the sphere of radius c·t. This is Kirchhoff's formula, the 3D analogue of d'Alembert.
Huygens' principle: why 3D signals stay sharp
Now the payoff. Suppose at t = 0 you make a sharp, localized disturbance — a tiny clap of sound concentrated near one point. In 3D, Kirchhoff says the signal reaching a far point P at time t draws only on the data on the sphere of radius c·t. The clap touches that sphere for one instant, when c·t equals the distance to P, and then never again. So a listener hears a crisp pop and then silence: the wave passes, leaves no wake, and the air falls quiet behind it. This clean cut is Huygens' principle (in its strong form), and it holds in 3D — and in every odd dimension three and up.
This is not a small technicality; it is why you exist as a hearing creature. Because 3D obeys Huygens' principle, a spoken word arrives as the same crisp word, syllable after syllable, with no smeared echo of the previous one piled on top. If air behaved like a 2D membrane, every sound would trail a fading hum and speech would dissolve into mush. The sharpness of the 3D characteristic cone — data living only on its surface, never its interior — is what keeps a signal a signal.
The 2D drumhead: descent, and a stubborn wake
Two dimensions get solved by a lovely cheat called the method of descent: a 2D problem is just a 3D problem whose data happens not to depend on the z-coordinate. So take the 3D Kirchhoff answer, feed it z-independent data, and the sphere-integral collapses to an integral over a flat disc — the whole solid disc of radius c·t, not merely its boundary circle. That gives Poisson's formula for the 2D wave equation, the membrane's version of d'Alembert.
And here is the catch that makes 2D special. Because the integral runs over the whole disc, not just its rim, the value at P keeps drawing on data from every point closer than c·t — including the original disturbance, long after the wavefront has swept past. So a stone dropped in still water does not produce one clean ring and silence: it makes an expanding ripple followed by a long, decaying train of trailing waves. The wake never fully shuts off. Huygens' principle fails in 2D — strong Huygens holds only in odd dimensions.
It is worth being precise so you do not over-claim. Even in 2D the leading edge of the disturbance still travels at exactly speed c — there is no signal ahead of the wavefront, so finite propagation speed survives in every dimension. What fails in 2D is only the trailing edge: the back of the signal is smeared, not sharp. Finite speed is about the front; Huygens is about the back. Keep those two ideas apart and the dimension-by-dimension table stops feeling paradoxical.
Energy: the bookkeeping that survives every dimension
One thing the wave equation never gives up, in any dimension, is energy. Define the total energy as the integral over space of (1/2) u_t^2 plus (c^2/2) |grad u|^2 — a kinetic part from how fast the medium is moving, and a potential part from how stretched it is. Differentiate in time, use the equation, and the integrand collapses into a pure boundary flux. On the whole space, or with fixed or insulated edges, that flux is zero, so the energy is constant. This is conservation of wave energy.
Conservation is exactly what separates a wave from heat. Diffusion dissipates: energy bleeds away and sharp features smooth out. A wave does neither — it conserves energy and refuses to smooth, which is precisely why a pulse can travel undistorted and why running the wave equation backward in time is perfectly fine (just send t to -t; the equation does not notice). Compare that to the infinite speed and irreversibility of the heat equation, whose backward problem is wildly ill-posed. Same toolbox, opposite personalities.
- Write the energy. E(t) = integral of (1/2) u_t^2 + (c^2/2) |grad u|^2 over the region.
- Differentiate in time. dE/dt = integral of u_t u_tt + c^2 grad u · grad u_t.
- Use the equation and integrate by parts. Substitute u_tt = c^2 (Laplacian u); the interior terms cancel and only a boundary flux of u_t against the outward normal derivative is left.
- Kill the boundary. With fixed ends (u fixed, so u_t = 0) or insulated ends (normal derivative = 0), the flux vanishes, so dE/dt = 0 and energy is conserved.
Energy conservation also does quiet, heavy lifting in the theory: it forces uniqueness. If two solutions share the same Cauchy data, their difference starts with zero energy and conserves it, so it stays zero forever — the two solutions must agree. The same bookkeeping localizes to cones and re-proves finite speed, ties back to the linear structure you used in d'Alembert, and underwrites the standing-wave and normal-mode picture from the previous guide, where each mode carries a fixed, undecaying packet of energy. One identity, many jobs.
Honest caveats and where the road leads next
A few honest limits keep this from becoming folklore. First, strong Huygens (no wake, signal lives on the wavefront) is genuinely special: it holds for the plain wave equation in odd dimensions n ≥ 3, fails in all even dimensions, and even in 1D the d'Alembert formula keeps a non-decaying contribution from the initial velocity, so a pure velocity kick leaves a lasting plateau rather than a clean pulse. So '1D is sharp' deserves an asterisk: the displacement part is sharp, the velocity part can linger.
Second, Huygens is fragile. Add even a tiny lower-order term — a damping u_t, a mass term, a wave in a curved or layered medium where c varies — and the strong principle generally breaks, growing a small wake even in 3D. Real acoustics, seismic waves, and electromagnetism in matter all show such tails. And third, if the speed depends on frequency, distinct wavelengths travel at different phase and group velocities and a pulse spreads out — that is dispersion, a separate effect from the wake, and the wave equation u_tt = c^2 Laplacian u is the special non-dispersive case where it is absent.
Where does this rung hand you off? You have now seen the wave equation whole: derived from a string, solved in 1D by d'Alembert, bounded by a finite signal speed, built from standing waves, and lifted to a room by Kirchhoff and Poisson with Huygens' principle as the prize. The next steps generalize each thread: separation of variables on a circular drum brings Bessel functions and the dispersion relation; steady waves at a single frequency give the Helmholtz equation; and nonlinear waves can steepen into shocks. You leave this rung knowing not just the formulas, but why a clap is crisp and a ripple is not.