The Wave Equation

The wave equation, $u_{tt}=c^2\nabla^2u$ , is the one partial differential equation that every branch of physics eventually writes down. A plucked string, a drumhead, the pressure in an organ pipe, the fields of light in vacuum, the P- and S-waves of an earthquake, and the ripples of spacetime a gravitational-wave detector listens for are all governed by the same three symbols. It is the archetype of the hyperbolic class — the third member, alongside Laplace's elliptic equation and the heat equation's parabolic diffusion, of the trio of model second-order linear PDEs around which the whole subject is organized. What makes it hyperbolic is the single fact from which everything else follows: it has real characteristics, curves in spacetime along which signals travel at the finite speed $c$ . Factor the one-dimensional operator into $(\partial_t-c\partial_x)(\partial_t+c\partial_x)$ and the solution splits at once into a left-mover and a right-mover, rigid profiles that slide without changing shape. This is d'Alembert's 1747 discovery, and it hands us the deepest qualitative property of the equation: a finite domain of dependence, a strict speed limit on information, and — unlike the smoothing heat equation — faithful transport of every kink and corner in the initial data. Two complementary pictures then organize the whole theory. The travelling-wave picture, built on characteristics, describes propagation on unbounded domains; the standing-wave picture, built on separation of variables and Sturm–Liouville theory, describes the discrete normal modes of bounded bodies. Fourier analysis is the dictionary between them, and Green's functions make the duality — and its startling dependence on spatial dimension — explicit.

Key takeaways

The wave equation $u_{tt}=c^2\nabla^2u$ is the prototypical hyperbolic PDE: disturbances travel at a finite speed $c$ along real characteristics, with no smoothing and exact transport of the initial shape.
In one dimension d'Alembert's formula $u=f(x-ct)+g(x+ct)$ gives the general solution as a superposition of rigid left- and right-moving profiles.
The free-field energy $E=\tfrac12\int(u_t^2+c^2|\nabla u|^2)\,\mathrm{d}^n x$ is conserved, which forces uniqueness and stability of the Cauchy problem and marks the absence of dissipation.
On bounded domains, separation of variables produces a discrete spectrum of standing modes; a fixed–fixed string of length $L$ rings at $\omega_n=n\pi c/L$ , an exactly harmonic ladder.
Solution structure depends sharply on dimension: sharp, tail-free wavefronts (strong Huygens' principle) in odd $n\ge3$ , but a persistent trailing wake in even dimensions and in $n=1$ ; the time-harmonic ansatz $u=v\,e^{-i\omega t}$ collapses the equation to the Helmholtz equation $\nabla^2v+k^2v=0$ .
Key idea: factor the d'Alembertian into characteristic derivatives — solutions are rigid profiles transported at speed $c$ , so initial data propagate without distortion along the lines $x\mp ct=\text{const}$ , with a finite domain of dependence and conserved energy.

Where this lives today

The wave equation is not a historical relic; it is the workhorse of live science and engineering. In September 2015 the LIGO detectors recorded gravitational waves from a binary black-hole merger — ripples in spacetime that, in the weak-field limit, satisfy $\Box\,h_{\mu\nu}=0$ propagating at exactly $c$ , confirming a 1916 Einstein prediction and opening a new observational astronomy. Seismic imaging solves the elastic wave equation over and over in "full-waveform inversion," reconstructing the Earth's interior from surface arrival times through the domain-of-dependence structure of characteristics. Computational electromagnetics runs FDTD solvers whose stability is governed by the CFL condition of worked example 3. Medical ultrasound inverts the equation to image tissue, acousticians solve it to design concert halls, and the Helmholtz reduction underlies every laser cavity and waveguide built today. The three symbols d'Alembert wrote in 1747 now run on GPUs at petascale.

The physical picture

Picture a stretched string under tension $T$ with mass per length $\rho$ . Lift one point into a bump and it pulls on its neighbours; they accelerate, pull theirs, and the disturbance walks along the string. Newton's law for a small element is exactly $\rho\,u_{tt}=T\,u_{xx}$ — the wave equation with $c^2=T/\rho$ . Its message is that curvature drives acceleration: where the string is concave down it accelerates down, a self-correcting feedback that makes a disturbance travel rather than merely spread. Nothing dissipates; the bump keeps its shape at the fixed speed $c$ . The hallmark is finite propagation speed. The value of $u$ at $(x,t)$ can be influenced only by the initial data inside the backward cone $|x'-x|\le ct$ — the domain of dependence. This is the soul of hyperbolicity and the reason the wave equation respects causality, while the heat equation, whose signals travel infinitely fast, does not. The first figure shows the cleanest consequence of the characteristic structure: an initial pulse splitting into two rigid half-height copies that ride off in opposite directions.

Figure

A localized profile splits into two half-height copies that ride left and right at speed c: the d'Alembert rule u = ½·(φ(x−ct) + φ(x+ct)) transports rigid shapes without distortion, so a kink stays a kink forever

Two complementary pictures then describe the very same field. The travelling-wave picture says any signal on the line is a sum of rigid shapes sliding left and right. The standing-wave picture says any vibration of a bounded body is a superposition of normal modes, each oscillating in place at its own frequency while its shape stays fixed. For a string clamped at both ends the modes are pure sinusoids whose frequencies form an exactly harmonic ladder — the reason a bowed string sounds like one pitch with overtones rather than an inharmonic clang. The second figure shows that ladder with real numbers for a concert-tuned string.

Figure

The standing-wave modes of a fixed-fixed string of length L = 0.65 m: eigenfrequencies fall on the harmonic ladder fₙ = n·c/(2L) with c = 572 m/s, so f₁ = 440 Hz (concert A), f₂ = 880 Hz, f₃ = 1320 Hz — integer multiples, which is why a string sounds musical

Energy is the bookkeeper that closes the theory. Kinetic density $\tfrac12u_t^2$ and potential density $\tfrac12c^2|\nabla u|^2$ trade back and forth as the wave moves, but their spatial integral is exactly constant for the free equation. That one conservation law delivers uniqueness and stability of the Cauchy problem, and it is the sharpest way to see how a wave differs from a diffusion: nothing decays, nothing is forgotten.

The core equations

The wave equation itself is the statement that the second time derivative of the field equals a constant times its spatial curvature. Written in its canonical, coordinate-free form it is the compact object every physical instance specializes.

u_{tt}=c^2\nabla^2 u \qquad (1)

where: $u(\mathbf{x},t)$ — the wave field (displacement, pressure, potential, or a Cartesian field component); $c$ — propagation speed set by the medium, $c>0$ ; $\nabla^2$ — the spatial Laplacian; $u_{tt}=\partial^2u/\partial t^2$ . Reading it physically: acceleration is proportional to curvature, so a peak (negative curvature) is pushed back toward equilibrium. Collecting all the derivatives on one side defines the d'Alembertian, the relativistically natural operator whose vanishing is the wave equation. It is the Minkowski-signature cousin of the Laplacian.

\Box\,u \equiv \frac{1}{c^2}\partial_t^2 u-\nabla^2 u = 0 \qquad (2)

where: $\Box$ — the d'Alembertian (wave operator), Lorentz-invariant for the choice $c=$ speed of light. Reading it: replacing $\partial_t\to\tau$ and $\partial_{x_j}\to\xi_j$ gives the principal symbol $c^{-2}\tau^2-|\xi|^2$ , an indefinite (Lorentzian) quadratic form — the algebraic fingerprint of hyperbolicity. In one spatial dimension the operator factors over the reals into two commuting first-order transport operators, and this factorization is the engine of the entire travelling-wave picture.

\partial_t^2-c^2\partial_x^2=(\partial_t-c\partial_x)(\partial_t+c\partial_x) \qquad (3)

where: $\partial_t\mp c\partial_x$ — directional derivatives along the characteristics $x\pm ct=\text{const}$ . Reading it: in the characteristic coordinates $\xi=x-ct$ , $\eta=x+ct$ the equation collapses to the single mixed derivative $u_{\xi\eta}=0$ , whose general solution is a left-mover plus a right-mover, $u=F(x-ct)+G(x+ct)$ — nothing else can survive. Fitting the general solution to Cauchy data — initial displacement and initial velocity — produces d'Alembert's explicit formula, the exact solution of the pure initial-value problem on the line.

u(x,t)=\tfrac12\!\left[\phi(x-ct)+\phi(x+ct)\right]+\frac{1}{2c}\int_{x-ct}^{x+ct}\!\psi(s)\,\mathrm{d}s \qquad (4)

where: $\phi(x)=u(x,0)$ — initial displacement; $\psi(x)=u_t(x,0)$ — initial velocity; $[x-ct,\,x+ct]$ — the domain of dependence. Reading it: the value at $(x,t)$ is fixed entirely by data in a finite interval, the sharp statement of finite signal speed. The free field conserves an energy that is manifestly non-negative, and its constancy is what makes the Cauchy problem well posed.

E(t)=\frac12\int_{\mathbb{R}^n}\!\left(u_t^2+c^2|\nabla u|^2\right)\mathrm{d}^n x,\qquad \frac{\mathrm{d}E}{\mathrm{d}t}=0 \qquad (5)

where: $\tfrac12u_t^2$ — kinetic density; $\tfrac12c^2|\nabla u|^2$ — potential/field density. Reading it: two solutions with identical Cauchy data have zero-energy difference, hence are identical — uniqueness follows in one line. On a bounded interval the same equation yields a discrete spectrum; the fixed–fixed string is the canonical example, with exactly harmonic eigenfrequencies.

u(x,t)=\sum_{n=1}^{\infty}\sin\!\frac{n\pi x}{L}\Big(a_n\cos\omega_n t+b_n\sin\omega_n t\Big),\quad \omega_n=\frac{n\pi c}{L} \qquad (6)

where: $L$ — string length with $u(0,t)=u(L,t)=0$ ; $\omega_n=n\pi c/L$ — the $n$ -th eigenfrequency; $a_n,b_n$ — Fourier coefficients set by $\phi,\psi$ . Reading it: the fundamental is $f_1=c/2L$ and every overtone is an integer multiple of it. Separating out a single time frequency reduces the wave equation to the Helmholtz equation, the master equation of resonators, waveguides, and scattering.

u=v(\mathbf{x})e^{-i\omega t}\;\Longrightarrow\;\nabla^2 v+k^2 v=0,\qquad k=\frac{\omega}{c} \qquad (7)

where: $v(\mathbf{x})$ — the spatial amplitude; $k$ — wavenumber. Reading it: the eigenvalues of $-\nabla^2$ on the domain are the squared wavenumbers $k^2$ , so they hand you the resonant frequencies $\omega=ck$ directly.

Derivation Ad'Alembert's solution from operator factorization

The goal is to solve the one-dimensional Cauchy problem $u_{tt}-c^2u_{xx}=0$ on the whole line, with prescribed initial displacement $u(x,0)=\phi(x)$ and initial velocity $u_t(x,0)=\psi(x)$ , obtaining not merely a solution but the explicit formula that exhibits the domain of dependence. The strategy is to factor the second-order operator into two commuting first-order transport operators, integrate each along its characteristics to get the general solution as a left-mover plus a right-mover, and then pin down the two arbitrary profiles by matching the Cauchy data.

Step 1. Factor the operator as $(\partial_t-c\partial_x)(\partial_t+c\partial_x)u=0$ and set $w=(\partial_t+c\partial_x)u$ , so that $(\partial_t-c\partial_x)w=0$ . Why: the factors commute, so the single second-order equation becomes a pair of first-order transport equations that can each be solved by inspection.

Step 2. Integrate along characteristics. A transport equation $(\partial_t-c\partial_x)w=0$ is constant along $x+ct=\text{const}$ , and assembling both families gives the general solution $u=F(x-ct)+G(x+ct)$ with arbitrary $C^2$ profiles $F,G$ . Why: this is the whole travelling-wave picture — rigid shapes moving right ( $F$ ) and left ( $G$ ) with no change of form.

Step 3. Apply the initial displacement at $t=0$ : $F(x)+G(x)=\phi(x)$ . Why: at $t=0$ both profiles sit at argument $x$ , so their sum is the initial shape — one equation for the two unknowns.

Step 4. Apply the initial velocity. Differentiating $u$ in $t$ and setting $t=0$ gives $-cF'(x)+cG'(x)=\psi(x)$ ; integrate to get $G(x)-F(x)=\tfrac1c\int_0^x\psi(s)\,\mathrm{d}s+\text{const}$ . Why: the time derivative distinguishes the two movers by sign, so the velocity data supply the second, independent relation.

Step 5. Solve the pair for $F$ and $G$ and substitute back into $u=F(x-ct)+G(x+ct)$ . The integration constant cancels, leaving $u(x,t)=\tfrac12[\phi(x-ct)+\phi(x+ct)]+\tfrac1{2c}\int_{x-ct}^{x+ct}\psi(s)\,\mathrm{d}s$ . Why: the shifted arguments $x\mp ct$ show the value at $(x,t)$ depends only on data in $[x-ct,x+ct]$ — the finite domain of dependence, made explicit. *Verified: substituting the boxed expression, both $u_{tt}$ and $c^2u_{xx}$ reduce to $\tfrac{c^2}{2}[\phi''(x-ct)+\phi''(x+ct)]+\tfrac{c}{2}[\psi'(x+ct)-\psi'(x-ct)]$ , and $u(x,0)=\phi$ , $u_t(x,0)=\psi$ hold identically. Matches Stone & Goldbart and Evans Ch. 2.*

Go deeper · Derivation B · 1 minSeparation of variables and the standing-wave spectrum

Derivation BSeparation of variables and the standing-wave spectrum

The goal is to construct the complete standing-wave solution of a string of length $L$ clamped at both ends, $u_{tt}=c^2u_{xx}$ on $0\le x\le L$ with $u(0,t)=u(L,t)=0$ , and to show that the boundary conditions quantize the allowed frequencies into the harmonic ladder $\omega_n=n\pi c/L$ . The method is to seek product solutions $X(x)T(t)$ , which turns the PDE into a Sturm–Liouville eigenvalue problem in space and a harmonic oscillator in time, then superpose the modes and fix the coefficients by orthogonality against the initial data.

Step 1. Substitute $u(x,t)=X(x)T(t)$ and divide by $XT$ to get $T''/(c^2T)=X''/X=-\lambda$ , a constant. Why: a function of $t$ alone can equal a function of $x$ alone only if both equal the same separation constant $-\lambda$ , splitting one PDE into two ODEs.

Step 2. Solve the spatial eigenvalue problem $X''+\lambda X=0$ with $X(0)=X(L)=0$ . Nontrivial solutions require $\lambda_n=(n\pi/L)^2$ with $X_n(x)=\sin(n\pi x/L)$ , $n=1,2,\dots$ . Why: this is a regular Sturm–Liouville problem, so its eigenvalues are real and positive and its eigenfunctions are mutually orthogonal on $[0,L]$ — the property that lets us expand any initial shape.

Step 3. Solve the temporal factor $T_n''+c^2\lambda_n T_n=0$ , giving $T_n=a_n\cos\omega_n t+b_n\sin\omega_n t$ with $\omega_n=c\sqrt{\lambda_n}=n\pi c/L$ . Why: each spatial mode oscillates in place at its own eigenfrequency, and because $\omega_n\propto n$ the overtones are exact integer multiples of the fundamental.

Step 4. Superpose and match data: $u=\sum_n X_nT_n$ , and orthogonality fixes $a_n=\tfrac2L\int_0^L\phi\sin\tfrac{n\pi x}{L}\,\mathrm{d}x$ and $b_n=\tfrac{2}{n\pi c}\int_0^L\psi\sin\tfrac{n\pi x}{L}\,\mathrm{d}x$ . Why: projecting the initial conditions onto each eigenfunction isolates its coefficient — Fourier analysis translating between standing modes and arbitrary data. *Verified: each term satisfies the PDE and both boundary conditions; the spectrum $\omega_n=n\pi c/L$ is the standard string overtone series with fundamental $f_1=c/2L$ . Consistent with Arfken & Weber and Riley–Hobson–Bence.*

Go deeper · Derivation C · 2 minSpherical means and Huygens' principle in three dimensions

Derivation CSpherical means and Huygens' principle in three dimensions

The goal is to explain why a compact flash of light or sound in three-dimensional space is registered at a distant point sharply and only once — the strong Huygens' principle — by solving $u_{tt}=c^2\nabla^2u$ in $\mathbb{R}^3$ with Cauchy data $\phi,\psi$ and reading off the geometry of influence. The route is to average the field over spheres about the observation point, show the average obeys a one-dimensional wave equation, solve it with d'Alembert's machinery, and take the radius to zero to recover the field as an integral over an expanding shell rather than a solid ball.

Step 1. Define the spherical mean $M(r,t)$ of $u$ over a sphere of radius $r$ centred at the observation point $\mathbf{x}$ . Using $\nabla^2$ in spherical coordinates, the wave equation forces $rM$ to satisfy $(rM)_{tt}=c^2(rM)_{rr}$ . Why: averaging kills the angular derivatives, and the radial Laplacian's $\tfrac1r\partial_r^2(r\,\cdot)$ structure turns the average into an exact one-dimensional wave in the variable $r$ .

Step 2. Solve the reduced 1D problem. $rM$ propagates like a d'Alembert solution on the half-line $r\ge0$ , with the odd-extension boundary behaviour $rM\to0$ as $r\to0$ . Why: this reduces a genuinely three-dimensional problem to the one case we have already solved completely, at the cost of tracking the factor $r$ .

Step 3. Recover the field by letting $r\to0$ , which yields Kirchhoff's formula $u(\mathbf{x},t)=\partial_t(t\,\overline{\phi})+t\,\overline{\psi}$ , where $\overline{f}=\tfrac1{4\pi}\int_{|\mathbf{n}|=1}f(\mathbf{x}+ct\,\mathbf{n})\,\mathrm{d}\Omega$ is the average over the sphere of radius $ct$ . Why: the field at $(\mathbf{x},t)$ depends only on data on the single spherical shell $|\mathbf{y}-\mathbf{x}|=ct$ , not on the interior of the ball.

Step 4. Read off Huygens' principle. Because only the shell contributes, a source of compact support is felt at $\mathbf{x}$ during a finite window and then falls silent — no afterglow. In two dimensions the average is over a disk, so a ripple trails a decaying tail. Why: the sharp-versus-trailing distinction is purely whether the influence set is a shell (odd $n\ge3$ ) or a solid region (even $n$ and $n=1$ ). *Verified: Kirchhoff's formula solves the 3D Cauchy problem and is the textbook statement of the strong Huygens' principle (Courant & Hilbert, Vol. II; Evans Ch. 2.4).*

Limiting cases

Special and limiting regimes locate the wave equation among the model PDEs and expose how its behaviour depends on frequency and on spatial dimension.

Case	Result	Meaning
Static / zero-frequency	$u_{tt}\to0\;\Rightarrow\;\nabla^2u=0$	time-independent solutions are harmonic — the equation degenerates to Laplace's elliptic equation
Time-harmonic	$u=v\,e^{-i\omega t}\;\Rightarrow\;\nabla^2v+k^2v=0,\ k=\omega/c$	each frequency component obeys Helmholtz; superposing them rebuilds any solution
One dimension	$u=f(x-ct)+g(x+ct)$	exact general solution: rigid left- and right-movers, sharp features preserved forever
Odd $n\ge3$ (strong Huygens)	influence confined to the cone $\|\mathbf{x}-\mathbf{x}_0\|=ct$	a 3D flash is seen sharply once and is gone — clean wavefronts, no reverberation
Even $n$ and $n=1$ (tails)	influence fills the solid cone $\|\mathbf{x}-\mathbf{x}_0\|\le ct$	a 2D ripple leaves a decaying wake; the disturbance lingers behind the front
Damped variant	$u_{tt}+\gamma u_t=c^2\nabla^2u$	adding $\gamma u_t$ attenuates the wave and breaks energy conservation — a parabolic flavour creeps in

Common trapTreating the wave equation like the heat equation

The wave equation does not smooth its data — the single most common conceptual error. A kink in the initial profile propagates as a kink forever; d'Alembert's formula transports $\phi$ and $\psi$ rigidly and never blurs them, unlike the heat equation, which instantly smooths any corner into an analytic bump. Two further traps follow. First, the equation is second order in time, so a well-posed initial-value problem needs both $u(\mathbf{x},0)$ and $u_t(\mathbf{x},0)$ ; supplying only the displacement, as for diffusion, leaves the motion undetermined. Second, you may not pose data anywhere you like: prescribing $u$ on a characteristic surface, or fixing conditions at both ends of a time interval, generally makes the problem ill-posed, because characteristics carry the data and cannot both start and end where you demand. Hyperbolic means causal — respect the cone.

Worked examples

1 · A concert-tuned string (fully worked)

A steel string of length $L=0.65\ \mathrm{m}$ and linear mass density $\rho=4.0\times10^{-4}\ \mathrm{kg/m}$ is tensioned to $T=131\ \mathrm{N}$ . Find the wave speed, the fundamental pitch and its first overtones, and the energy stored when the fundamental vibrates with amplitude $A=1.0\ \mathrm{mm}$ .

Step 1. Wave speed from the medium. The transverse wave speed is $c=\sqrt{T/\rho}=\sqrt{131/(4.0\times10^{-4})}=\sqrt{3.275\times10^{5}}=\mathbf{572\ m/s}$ — the $c$ that appears in $u_{tt}=c^2u_{xx}$ for this string, set by tension, not the longitudinal sound speed of steel.

Step 2. Fundamental and overtones. With fixed ends the eigenfrequencies are $\omega_n=n\pi c/L$ , i.e. $f_n=nc/(2L)$ . The fundamental is $f_1=572/(2\times0.65)=572/1.30=\mathbf{440\ Hz}$ — concert-pitch A. The overtones are exact integer multiples: $f_2=\mathbf{880\ Hz}$ , $f_3=\mathbf{1320\ Hz}$ , $f_4=\mathbf{1760\ Hz}$ . This perfect harmonicity is why the note sounds like a single musical pitch.

Step 3. Energy in the fundamental mode. For a mode $u_1=A\sin(\pi x/L)\cos\omega_1 t$ the total (kinetic plus potential) energy is $E_1=\tfrac14\rho L\,\omega_1^2 A^2$ , with $\omega_1=2\pi f_1=2\pi(440)=2766\ \mathrm{rad/s}$ . Plugging in, $E_1=\tfrac14(4.0\times10^{-4})(0.65)(2766)^2(1.0\times10^{-3})^2=\mathbf{5.0\times10^{-4}\ J}\approx0.50\ \mathrm{mJ}$ . c = 572 m/s, f₁ = 440 Hz with harmonics 880 / 1320 / 1760 Hz, and E₁ ≈ 0.50 mJ.

2 · A pulse splitting on the line (d'Alembert with numbers)

The same string (now treated as effectively infinite) is given an initial displacement $\phi(x)$ shaped as a single bump of height $4.0\ \mathrm{cm}$ and zero initial velocity, $\psi=0$ . Describe the field at $t=0.50\ \mathrm{ms}$ .

Step 1. Apply d'Alembert with $\psi=0$ : $u(x,t)=\tfrac12[\phi(x-ct)+\phi(x+ct)]$ . The single bump splits into two copies, each of half the original height: $4.0/2=\mathbf{2.0\ cm}$ apiece. No integral term survives because the string starts at rest.

Step 2. Locate the copies. In $t=0.50\ \mathrm{ms}$ each travels $ct=572\times5.0\times10^{-4}=0.286\ \mathrm{m}=28.6\ \mathrm{cm}$ , one left and one right, so their peaks are $2\times28.6=\mathbf{57.2\ cm}$ apart. Each copy keeps the exact shape of the original bump — no spreading.

Step 3. Domain of dependence check. A point at distance $50\ \mathrm{cm}$ from the origin has not yet been reached (its backward interval $[x-ct,x+ct]$ does not yet include the bump), so it stays flat — a direct display of finite signal speed. The bump becomes two 2.0 cm half-copies moving apart at 572 m/s, 57.2 cm apart after 0.50 ms.

3 · The CFL stability limit of a solver

*A 1D finite-difference acoustic solver uses $c=343\ \mathrm{m/s}$ and grid spacing $\Delta x=1.0\ \mathrm{mm}$ . The Courant–Friedrichs–Lewy condition demands the numerical domain of dependence contain the physical one, $\nu\equiv c\,\Delta t/\Delta x\le1$ , so the largest stable step is $\Delta t\le\Delta x/c=1.0\times10^{-3}/343=\mathbf{2.92\ \mu s}$ . At $\Delta t=3.0\ \mu\mathrm{s}$ , $\nu=1.03>1$ and the scheme blows up; at $\Delta t=2.5\ \mu\mathrm{s}$ , $\nu=0.86$ and it is stable. Δt_max = 2.92 μs — the characteristic geometry directly sets the solver's timestep.

Physics of the portrait

The wave equation is inseparable from the young mathematician who first wrote it down and solved it, and from the century-long argument his solution touched off — a dispute that, in being resolved, gave physics both Fourier analysis and the modern notion of a function.

Origins · 1 minJean-Baptiste le Rond d'Alembert and how this was discovered

Jean-Baptiste le Rond d'Alembert

1717–1783 · French mathematician and Encyclopédiste who gave the first general solution of the wave equation

Abandoned as an infant on the steps of a Paris church and raised by a glazier's family, d'Alembert rose to co-edit, with Diderot, the great Encyclopédie. In 1747 he took up the vibrating string and did something new: instead of guessing modes, he derived the PDE $u_{tt}=c^2u_{xx}$ from the mechanics and then solved it in general, showing the motion is $u=f(x-ct)+g(x+ct)$ — an arbitrary right-mover plus an arbitrary left-mover. This was the birth of the method of characteristics and the first general solution of a PDE of physics. Euler extended the analysis but disagreed sharply about admissible initial shapes: d'Alembert insisted the profiles be smooth, Euler wanted to allow a plucked string's corner, and Daniel Bernoulli argued in 1753 that every vibration is a superposition of sinusoidal modes — the physical seed of Fourier series, which both men rejected as too restrictive. The three-way debate over what counts as an allowable function ran for decades; only Fourier's 1822 theory of heat and the rigorous definitions of function and convergence that followed settled it. d'Alembert's rigid travelling profiles and Bernoulli's standing modes were, in the end, two faces of the same solution space.

History

1747

— d'Alembert derives and gives the first general solution of the vibrating-string wave equation

1753

— Daniel Bernoulli proposes that any vibration is a superposition of sinusoidal modes

1822

— Fourier's theory of heat legitimizes trigonometric series, resolving the vibrating-string debate

1865

— Maxwell shows light is an electromagnetic wave obeying the same equation with c the speed of light

1882

— Kirchhoff casts Huygens' principle as an exact surface integral

Going further

Texts. Stone & Goldbart, Mathematics for Physics, chapters on PDEs and Green functions · Arfken & Weber, Mathematical Methods for Physicists, chapters on PDEs and the Helmholtz equation · Hassani, Mathematical Physics, PDE and Green's-function chapters · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering · Courant & Hilbert, Methods of Mathematical Physics, Vol. II · Evans, Partial Differential Equations (AMS GSM 19).

Provenance. Equations are standard textbook results, re-verified here. A verifier should confirm: (1) d'Alembert's formula solves the 1D Cauchy problem by direct substitution (Stone & Goldbart; Evans Ch. 2); (2) the string spectrum $\omega_n=n\pi c/L$ and fundamental $f_1=c/2L$ , and the worked value $c=\sqrt{T/\rho}=572\ \mathrm{m/s}\Rightarrow f_1=440\ \mathrm{Hz}$ (Arfken & Weber; Riley–Hobson–Bence); (3) the energy identity via multiply-by- $u_t$ and integration by parts; (4) Kirchhoff's formula and the strong Huygens' principle in 3D versus the 2D tail (Courant & Hilbert Vol. II; Evans); (5) the CFL number $c\,\Delta t/\Delta x\le1$ giving $\Delta t_{\max}=2.92\ \mu\mathrm{s}$ for $c=343\ \mathrm{m/s}$ , $\Delta x=1\ \mathrm{mm}$ .

Prerequisites. Classification of Second-Order Linear PDEs · Boundary Conditions and Well-Posedness · Separation of Variables in Cartesian Coordinates. Leads to. d'Alembert's Solution and Characteristics · The Helmholtz Equation · Green's Functions for PDEs · The Method of Characteristics.

Ask the tutor: Ask me to walk through d'Alembert's splitting for your own initial data, derive a cavity's normal modes via Helmholtz, or show why 3D wavefronts stay sharp while 2D ripples trail a tail.

Why does the wave equation need two initial conditions but the heat equation only one?

Because it is second order in time. Integrating twice in $t$ introduces two arbitrary functions of position — exactly the pair $u(\mathbf{x},0)=\phi$ and $u_t(\mathbf{x},0)=\psi$ that d'Alembert's formula consumes. Physically, to launch a wave you must say both where the medium starts and how fast it is moving: a plucked string (displacement, zero velocity) and a struck string (zero displacement, velocity) sound audibly different on the same instrument. The heat equation is only first order in time, so a single initial temperature field determines all later evolution.

What does it really mean that the wave equation is "hyperbolic"?

It is a statement about the equation's characteristics — the surfaces along which the highest derivatives can jump, and along which signals travel. Replacing derivatives by symbols, $\partial_t\to\tau$ , $\partial_x\to\xi$ , gives the principal form $\tau^2-c^2\xi^2=(\tau-c\xi)(\tau+c\xi)$ : two distinct real characteristic directions, which is the definition of hyperbolic. Elliptic equations (Laplace) have none and describe equilibrium; parabolic equations (heat) have one repeated family and describe diffusion. The real characteristics are precisely why waves propagate at finite speed with a domain of dependence, while the other two classes do not.

Why is energy conservation such a big deal for a PDE?

Because it hands you uniqueness and stability almost for free. Multiply the equation by $u_t$ and integrate over space to get $\mathrm{d}E/\mathrm{d}t=0$ with $E=\tfrac12\int(u_t^2+c^2|\nabla u|^2)$ , a manifestly non-negative quantity. Take two solutions with identical Cauchy data: their difference has zero initial energy, and since energy is conserved and non-negative it stays zero, forcing the difference to vanish — uniqueness. The same estimate shows small changes in the data produce small changes in the solution, the stability half of well-posedness. Conservation is not a bonus fact; it is the structural backbone of the theory.

Is Huygens' principle really different in 2D and 3D?

Yes, and strikingly so. In three dimensions the field at a point depends only on data on the shell at distance $ct$ , so a compact flash is registered sharply and then vanishes — you can hold a crisp conversation across a room because each syllable arrives and ends. In two dimensions (ripples on a pond) the influence comes from the whole disk out to radius $ct$ , so a disturbance leaves a decaying tail: the water keeps sloshing after the front has passed. This "strong Huygens' principle" holds only in odd dimensions $\ge3$ ; even dimensions and $n=1$ always carry tails — the cleanest example of one equation behaving differently with geometry.

How do the travelling-wave and standing-wave pictures fit together?

They are two coordinate systems on the same solution space. On an unbounded domain the natural description is d'Alembert's: rigid profiles sliding along characteristics. On a bounded domain the walls reflect those profiles back on themselves, and a right- and a left-mover of matching wavelength interfere into a shape that oscillates in place — a standing mode. Algebraically, $\sin(n\pi x/L)\cos(\omega_n t)=\tfrac12\sin(k_n(x-ct))+\tfrac12\sin(k_n(x+ct))$ , a standing wave written as two travelling ones. Fourier analysis is the dictionary: expanding data in the eigenmodes is the bounded-domain analogue of splitting it into movers.

What is the connection between the wave equation and the Helmholtz equation?

The Helmholtz equation is what the wave equation becomes at a single frequency. Substitute $u(\mathbf{x},t)=v(\mathbf{x})e^{-i\omega t}$ and the time derivatives pull down $-\omega^2$ , leaving $\nabla^2v+k^2v=0$ with $k=\omega/c$ . Because a general solution is a Fourier superposition over $\omega$ , solving Helmholtz at every frequency and summing reconstructs the full time-dependent field. This is a workhorse move: it turns scattering, waveguide, cavity, and antenna problems into boundary-value problems for one elliptic-looking equation whose eigenvalues are the resonant frequencies. Nearly every "modes of a cavity" calculation is a Helmholtz eigenproblem in disguise.