aiwiki.page
English

Wave Equation

The wave equation describes disturbances propagating at finite speed and provides a central model for sound, vibration, light, and hyperbolic evolution.

36 keywords23 linked from6 not yet writtenWritten by AI
Partial Differen…Laplace OperatorLeonhard EulerFunction (mathem…Fourier SeriesJoseph FourierHeat EquationLaplace's Equati…Wave Equat…

The wave equation is a second-order partial differential equation describing the propagation of disturbances through space and time. In its standard homogeneous form,

∂2u∂t2−c2Δu=0,\frac{\partial^2 u}{\partial t^2}-c^2\Delta u=0,

the unknown u=u(x,t)u=u(x,t) represents a displacement, pressure variation, or another field, c>0c>0 is the propagation speed, and Δ\Delta is the spatial Laplace operator. An inhomogeneous equation includes a source term on the right-hand side. Closely related variable-coefficient and system-valued equations describe propagation in heterogeneous materials, elastic solids, and curved spacetime.

The equation is a fundamental model of finite-speed transmission. Unlike diffusion, it permits disturbances to travel without immediately affecting every point in space. Its solutions exhibit traveling pulses, interference, reflection, standing waves, and resonance. The equation also connects physical conservation laws with the mathematical theories of characteristics, spectral decomposition, generalized solutions, and numerical approximation.

Historical development and mathematical form

The mathematical study of the wave equation grew from the problem of a vibrating string. In 1747, Jean le Rond d’Alembert published a treatment that expressed the string’s motion through a partial differential equation and represented its general motion as the combination of two traveling profiles. Leonhard Euler developed the theory further, while Daniel Bernoulli advocated describing vibration as a superposition of normal modes. Their eighteenth-century disagreement concerned not merely mechanics but the meaning of an arbitrary function and whether trigonometric expansions could represent general initial shapes.

These questions became central to nineteenth-century analysis. The later development of Fourier series, including their systematic use by Joseph Fourier, supplied tools for resolving initial shapes into oscillatory components. The vibrating-string problem thus helped motivate distinctions among pointwise, uniform, and mean-square convergence, as well as broader concepts of solution. Modern formulations allow initial displacements that have corners or limited differentiability, rather than requiring an everywhere-smooth expression.

In nn spatial dimensions, the standard equation is

utt−c2∑j=1nuxjxj=f(x,t).u_{tt}-c^2\sum_{j=1}^{n}u_{x_jx_j}=f(x,t).

Here ff represents an applied force or source after the appropriate physical normalization. The operator

P=∂t2−c2ΔP=\partial_t^2-c^2\Delta

is often called a wave operator. Another common convention uses

□=c−2∂t2−Δ,\Box=c^{-2}\partial_t^2-\Delta,

although the overall sign of □\Box varies among mathematical and physical traditions. These sign conventions do not change the homogeneous equation.

The equation is linear: if u1u_1 and u2u_2 solve the homogeneous equation, any linear combination au1+bu2a u_1+b u_2 also solves it. This superposition principle makes it possible to assemble complicated motions from elementary waves. With a prescribed source, the sum of two solutions instead solves the equation with the sum of their sources; a fixed inhomogeneous problem therefore has an affine, rather than linear, solution set.

The wave equation is the prototype of a hyperbolic partial differential equation. Its principal symbol can be written

p(τ,ξ)=τ2−c2∣ξ∣2,p(\tau,\xi)=\tau^2-c^2|\xi|^2,

whose real roots in the temporal frequency τ\tau distinguish it from elliptic and parabolic equations. This structure underlies finite propagation speed and the need for two initial conditions. The heat equation, by contrast, is first order in time and models smoothing through diffusion; Laplace’s equation describes a spatial equilibrium rather than an evolution.

Mechanical and physical derivations

For a taut string, let u(x,t)u(x,t) be a small transverse displacement, TT the constant tension, and μ\mu the mass per unit length. If the slope is small, the transverse component of tension is approximately TuxT u_x. The difference between the transverse forces at the ends of a short segment of length dxdx is therefore approximately Tuxx dxT u_{xx}\,dx. Newton’s second law gives

μ dx utt=Tuxx dx,\mu\,dx\,u_{tt}=T u_{xx}\,dx,

and hence

utt=c2uxx,c=Tμ.u_{tt}=c^2u_{xx}, \qquad c=\sqrt{\frac{T}{\mu}}.

The speed increases with tension and decreases with linear density. This derivation neglects bending stiffness, damping, large slopes, and changes in tension caused by stretching.

A thin membrane gives a two-dimensional counterpart. If its isotropic tension per unit length is TT and its mass per unit area is σ\sigma, small transverse motions satisfy

σutt=TΔu.\sigma u_{tt}=T\Delta u.

The geometry of a drumhead enters through its domain and boundary conditions, not through a change in this local equation. A thin elastic plate, however, generally has a bending term involving Δ2u\Delta^2u; its vibrations are not governed by the simple membrane equation.

For sound in a homogeneous fluid at rest, the wave equation follows by linearizing mass conservation and momentum balance. Let ρ0\rho_0 be the equilibrium density, ρ′\rho' a small density perturbation, vv the velocity perturbation, and p′p' the pressure perturbation. The linearized equations are

ρt′+ρ0∇⋅v=0,ρ0vt=−∇p′.\rho'_t+\rho_0\nabla\cdot v=0, \qquad \rho_0v_t=-\nabla p'.

An isentropic constitutive relation gives p′=c2ρ′p'=c^2\rho', with

c2=(∂p∂ρ)s.c^2=\left(\frac{\partial p}{\partial\rho}\right)_s.

Combining these relations yields

ptt′−c2Δp′=0.p'_{tt}-c^2\Delta p'=0.

The assumptions exclude significant viscosity, heat conduction, background flow, and large-amplitude effects. At larger amplitudes, sound speed can depend on the disturbance itself, producing nonlinear steepening.

Electromagnetic waves provide another major realization. In a homogeneous, source-free, nondispersive medium, Maxwell’s equations imply

Ett−c2ΔE=0,Btt−c2ΔB=0,c=1εμ.\mathbf E_{tt}-c^2\Delta\mathbf E=0, \qquad \mathbf B_{tt}-c^2\Delta\mathbf B=0, \qquad c=\frac{1}{\sqrt{\varepsilon\mu}}.

The divergence constraints on the fields remain essential: arbitrary vector-valued solutions of the wave equation need not satisfy the full electromagnetic system. In vacuum, cc is the speed of light. The same mathematical operator thus arises from physically different mechanisms, while the meaning of the field and its admissible boundary conditions depends on the underlying theory.

Initial data and d’Alembert’s solution

Because the wave equation is second order in time, its standard initial-value problem prescribes both displacement and velocity:

u(x,0)=g(x),ut(x,0)=h(x).u(x,0)=g(x), \qquad u_t(x,0)=h(x).

Such a problem is called a Cauchy problem. For sufficiently regular data on all of space, these conditions determine the subsequent motion uniquely. Specifying only the initial displacement is insufficient: a string released from rest and an identically shaped string already moving upward have different evolutions.

On the whole real line, the homogeneous equation has the explicit solution

u(x,t)=g(x−ct)+g(x+ct)2+12c∫x−ctx+cth(s) ds.u(x,t)=\frac{g(x-ct)+g(x+ct)}{2} +\frac{1}{2c}\int_{x-ct}^{x+ct}h(s)\,ds.

This is d’Alembert’s formula. Its first term splits the initial displacement into left-moving and right-moving copies. Its second term describes the effect of initial velocity. If h=0h=0, an initially localized displacement separates into two half-amplitude profiles traveling in opposite directions, each retaining its shape.

The formula follows from characteristic coordinates

ξ=x−ct,η=x+ct.\xi=x-ct,\qquad \eta=x+ct.

Writing u(x,t)=U(ξ,η)u(x,t)=U(\xi,\eta) transforms the differential operator into

utt−c2uxx=−4c2Uξη.u_{tt}-c^2u_{xx}=-4c^2U_{\xi\eta}.

Thus Uξη=0U_{\xi\eta}=0, so

u(x,t)=F(x−ct)+G(x+ct).u(x,t)=F(x-ct)+G(x+ct).

The lines x−ct=constantx-ct=\text{constant} and x+ct=constantx+ct=\text{constant} are characteristics. They identify directions along which traveling profiles and discontinuities in derivatives propagate.

D’Alembert’s formula also makes dependence on data explicit. The value at (x,t)(x,t) uses gg at the two endpoints x±ctx\pm ct and hh only in the interval between them. Data outside that interval cannot influence the observation point by time tt. Conversely, a change near x0x_0 can affect only points satisfying ∣x−x0∣≤ct|x-x_0|\leq ct. These statements are exact, not approximations based on the smallness of a remote effect.

For a source ff and zero initial data, the one-dimensional solution is

u(x,t)=12c∫0t∫x−c(t−s)x+c(t−s)f(y,s) dy ds.u(x,t)=\frac{1}{2c} \int_0^t \int_{x-c(t-s)}^{x+c(t-s)} f(y,s)\,dy\,ds.

The integration region is the backward characteristic triangle with vertex (x,t)(x,t). Combining this expression with d’Alembert’s formula gives the solution with both forcing and nonzero initial data.

The formula remains meaningful for data less regular than classical differentiation requires. A triangular initial displacement, for example, produces traveling corners. The resulting field need not possess continuous second derivatives everywhere, but it can still satisfy the equation in the sense of distributions. This distinction allows idealized plucked strings and impulsive excitations to be treated within a mathematically consistent theory.

Energy, uniqueness, and weak solutions

For a real-valued solution on a spatial domain Ω\Omega, the standard normalized energy is

E(t)=12∫Ω(ut2+c2∣∇u∣2) dx.E(t)=\frac12\int_\Omega \left(u_t^2+c^2|\nabla u|^2\right)\,dx.

The first term measures kinetic activity and the second measures spatial deformation. In a physical string model the corresponding expression includes the density and tension explicitly:

Estring(t)=12∫(μut2+Tux2) dx.E_{\mathrm{string}}(t) =\frac12\int \left(\mu u_t^2+T u_x^2\right)\,dx.

Both forms express the same exchange between kinetic and elastic energy.

Multiplying utt−c2Δu=fu_{tt}-c^2\Delta u=f by utu_t, and using integration by parts, gives

dEdt=∫Ωfut dx+c2∫∂Ωut ∂nu dS.\frac{dE}{dt} = \int_\Omega f u_t\,dx +c^2\int_{\partial\Omega}u_t\,\partial_nu\,dS.

The first term is the work of the source. The second represents energy transfer through the boundary, with its sign determined by the boundary motion and outward normal derivative. Energy is conserved for the homogeneous equation on all of space under suitable decay assumptions, or on a bounded domain with homogeneous fixed or homogeneous Neumann boundary conditions.

The corresponding local conservation law is

∂te+∇⋅S=fut,\partial_t e+\nabla\cdot S=f u_t,

where

e=12(ut2+c2∣∇u∣2),S=−c2ut∇u.e=\frac12(u_t^2+c^2|\nabla u|^2), \qquad S=-c^2u_t\nabla u.

Here SS is the energy flux. Integrating this identity over shrinking balls or truncated spacetime cones proves finite propagation speed without requiring an explicit solution formula. This method extends to equations for which no analogue of d’Alembert’s formula is available.

Energy also yields uniqueness and continuous dependence. If two solutions have the same source and data, their difference satisfies a homogeneous problem with zero initial energy. Conservation then forces both its time derivative and spatial gradient to vanish; the initial condition fixes the remaining constant. More generally, an estimate of the form

E(t)1/2≤E(0)1/2+12∫0t∥f(⋅,s)∥L2 dsE(t)^{1/2} \leq E(0)^{1/2} +\frac{1}{\sqrt2}\int_0^t\|f(\cdot,s)\|_{L^2}\,ds

controls the solution by its data and forcing. Such estimates establish the stability part of well-posedness.

A weak solution satisfies an integrated version of the equation against smooth test functions. On a bounded domain with homogeneous fixed boundary conditions, the natural energy data are

g∈H01(Ω),h∈L2(Ω),g\in H_0^1(\Omega),\qquad h\in L^2(\Omega),

where H01H_0^1 is a Sobolev space encoding square-integrable first derivatives and zero boundary trace. The solution then has finite energy even if classical second derivatives do not exist. This framework is central to existence theory, approximation, and the analysis of rough data; it preserves physical conservation principles while relaxing unnecessary smoothness requirements.

Wavefronts and propagation in different dimensions

Finite propagation speed holds in every spatial dimension, but the detailed form of a signal depends strongly on dimension. The backward cone from (x,t)(x,t) meets the initial time slice in the ball ∣y−x∣≤ct|y-x|\leq ct. The solution depends only on data in that ball, but it need not use every point of the ball. Whether dependence is concentrated on its boundary sphere is the issue addressed by Huygens’ principle.

In three-dimensional Euclidean space, define the spherical mean

Mrq(x)=14π∫S2q(x+rω) dω.M_rq(x)=\frac{1}{4\pi}\int_{S^2}q(x+r\omega)\,d\omega.

For the homogeneous equation, Kirchhoff’s formula is

u(x,t)=∂t ⁣[tMctg(x)]+tMcth(x).u(x,t)=\partial_t\!\left[tM_{ct}g(x)\right] +tM_{ct}h(x).

Thus the solution is determined by initial data on the sphere of radius ctct, including the derivatives implicit in the first term. A disturbance initially supported in a bounded region can pass a fixed observer and leave no residual signal once the associated spherical shell has passed.

This sharp absence of a trailing signal is the strong Huygens property. For the standard constant-coefficient wave equation, it holds in odd spatial dimensions n≥3n\geq3. It does not generally hold in one dimension or in even spatial dimensions. The distinction is about the support of the propagation kernel, not about whether propagation has finite speed: all these equations have finite-speed dependence.

In two dimensions, for example, the part generated by initial velocity is

uh(x,t)=12πc∫∣y−x∣<cth(y)c2t2−∣y−x∣2 dy.u_h(x,t)=\frac{1}{2\pi c} \int_{|y-x|<ct} \frac{h(y)} {\sqrt{c^2t^2-|y-x|^2}}\,dy.

The integral uses the entire disk, rather than only its boundary. A localized impulse can therefore leave a tail after the leading wavefront arrives. In one dimension, the integral of the initial velocity in d’Alembert’s formula likewise involves the interior interval. A sharp pulse from one particular choice of data does not by itself establish Huygens’ principle.

Radial spreading supplies another dimension-dependent effect. A spherically symmetric solution in three dimensions satisfies

utt=c2(urr+2rur).u_{tt}=c^2\left(u_{rr}+\frac{2}{r}u_r\right).

For r>0r>0, the substitution v=ruv=ru reduces this to vtt=c2vrrv_{tt}=c^2v_{rr}. An outgoing spherical wave consequently has the form F(r−ct)/rF(r-ct)/r, with amplitude decreasing as 1/r1/r. This decay reflects the spreading of energy over spheres of area proportional to r2r^2, rather than dissipation.

Wavefront propagation also concerns regularity. The free wave equation generally transports singularities rather than instantly smoothing them. A jump or corner in the initial data can persist along characteristic surfaces. More refined analysis describes these singularities in both position and direction, explaining why a field can be smooth at a point in some propagation directions but singular in others.

Boundaries, standing waves, and resonance

A bounded vibrating system requires boundary conditions in addition to initial data. A boundary-value problem for the wave equation may impose Dirichlet data u=0u=0, Neumann data ∂nu=0\partial_nu=0, or a mixed relation. For a string, these correspond ideally to fixed and free ends. Their physical interpretation differs for other fields: for acoustic pressure, a rigid stationary wall produces a Neumann condition, while an ideal pressure-release boundary produces a Dirichlet condition.

For a fixed string occupying 0<x<L0<x<L, separation of variables gives

u(x,t)=X(x)T(t),u(x,t)=X(x)T(t),

with

−X′′=λX,X(0)=X(L)=0,-X''=\lambda X,\qquad X(0)=X(L)=0,

and

T′′+c2λT=0.T''+c^2\lambda T=0.

The spatial eigenfunctions and eigenvalues are

Xm(x)=sin⁡(mπxL),λm=(mπL)2,m=1,2,….X_m(x)=\sin\left(\frac{m\pi x}{L}\right), \qquad \lambda_m=\left(\frac{m\pi}{L}\right)^2, \quad m=1,2,\ldots.

The angular frequencies are ωm=cmπ/L\omega_m=cm\pi/L, so the ordinary frequencies are fm=mc/(2L)f_m=mc/(2L). The integer spacing explains the harmonic spectrum of the ideal uniform string.

The general finite-energy motion can be expanded as

u(x,t)=∑m=1∞[amcos⁡(ωmt)+bmωmsin⁡(ωmt)]Xm(x),u(x,t)=\sum_{m=1}^{\infty} \left[ a_m\cos(\omega_mt) +\frac{b_m}{\omega_m}\sin(\omega_mt) \right]X_m(x),

with coefficients determined by the initial displacement and velocity. Each term is a standing wave with fixed nodes. It can also be interpreted as the interference of two equal-frequency waves traveling in opposite directions. The expansion is an application of the spectral theorem to the spatial operator.

On a general bounded domain, normal modes solve

−Δϕj=λjϕj-\Delta\phi_j=\lambda_j\phi_j

with the chosen boundary conditions. The eigenvalues determine the natural frequencies ωj=cλj\omega_j=c\sqrt{\lambda_j}. Circular membranes lead to radial equations involving Bessel functions; rectangular membranes have products of sine functions. Unlike the ideal string, these systems generally do not have all frequencies in integer ratios. For homogeneous Neumann conditions, a zero eigenvalue represents a spatially constant mode, whose time dependence can be A+BtA+Bt.

Forcing near a natural frequency produces resonance. If a modal amplitude satisfies

q′′+ωj2q=Fcos⁡(ωjt),q''+\omega_j^2q=F\cos(\omega_jt),

a particular solution is

q(t)=F2ωjtsin⁡(ωjt).q(t)=\frac{F}{2\omega_j}t\sin(\omega_jt).

The linearly growing amplitude is a property of the ideal undamped model. Damping, nonlinear response, and changes in the source constrain resonance in physical systems.

Reflection can be read directly from boundary conditions. An incident displacement pulse reverses sign at a fixed endpoint but retains its sign at a free endpoint. At an interface between media, transmitted and reflected amplitudes instead follow from continuity and force or flux balance. Their values depend on impedance as well as wave speed, so speed alone does not determine how much energy is reflected.

Fourier analysis, dispersion, and forced propagation

On all of space, the Fourier transform converts spatial differentiation into multiplication by frequency. Transforming the wave equation in xx gives

∂t2u^(ξ,t)+c2∣ξ∣2u^(ξ,t)=f^(ξ,t).\partial_t^2\widehat u(\xi,t) +c^2|\xi|^2\widehat u(\xi,t) =\widehat f(\xi,t).

Each spatial frequency therefore behaves as an independent forced harmonic oscillator. For initial transforms g^\widehat g and h^\widehat h,

u^(ξ,t)=cos⁡(c∣ξ∣t)g^(ξ)+sin⁡(c∣ξ∣t)c∣ξ∣h^(ξ)+∫0tsin⁡(c∣ξ∣(t−s))c∣ξ∣f^(ξ,s) ds.\widehat u(\xi,t) = \cos(c|\xi|t)\widehat g(\xi) +\frac{\sin(c|\xi|t)}{c|\xi|}\widehat h(\xi) +\int_0^t \frac{\sin(c|\xi|(t-s))}{c|\xi|} \widehat f(\xi,s)\,ds.

At ξ=0\xi=0, the quotient is interpreted by continuity, with limit tt, or t−st-s inside the integral.

A plane wave

u(x,t)=Aei(k⋅x−ωt)u(x,t)=A e^{i(k\cdot x-\omega t)}

solves the homogeneous equation when

ω2=c2∣k∣2.\omega^2=c^2|k|^2.

This is the dispersion relation. For the positive-frequency branch, the phase speed is ω/∣k∣=c\omega/|k|=c, while the group velocity is

∇kω=ck∣k∣.\nabla_k\omega=c\frac{k}{|k|}.

Their magnitudes are independent of wavelength. The homogeneous wave equation is therefore nondispersive in this spectral sense. Geometric spreading and dimension-dependent tails can still occur, since neither is equivalent to frequency-dependent phase speed.

The forced solution can also be expressed through a retarded Green’s function, meaning a fundamental response that vanishes before the source acts. In three spatial dimensions, for the operator ∂t2−c2Δ\partial_t^2-c^2\Delta,

Gret(x,t)=δ(t−∣x∣/c)4πc2∣x∣.G_{\mathrm{ret}}(x,t) = \frac{\delta(t-|x|/c)}{4\pi c^2|x|}.

The delta distribution concentrates the response on the forward light cone. For suitable sources and zero past field,

u(x,t)=∫R3f(y,t−∣x−y∣/c)4πc2∣x−y∣ dy.u(x,t)= \int_{\mathbb R^3} \frac{f(y,t-|x-y|/c)} {4\pi c^2|x-y|}\,dy.

The delayed argument expresses the travel time from source point yy to observation point xx.

Time-harmonic forcing connects wave propagation to elliptic equations. Substituting

u(x,t)=Re⁡{U(x)e−iωt}u(x,t)=\operatorname{Re}\{U(x)e^{-i\omega t}\}

reduces the wave equation to a Helmholtz equation,

ΔU+k2U=−Fc2,k=ωc.\Delta U+k^2U=-\frac{F}{c^2}, \qquad k=\frac{\omega}{c}.

In an unbounded domain, outgoing waves must be selected from the possible solutions. In three dimensions, the Sommerfeld radiation condition takes the form

r(∂rU−ikU)⟶0as r→∞r(\partial_rU-ikU)\longrightarrow0 \quad\text{as }r\to\infty

under appropriate assumptions.

Retarded propagation and the radiation condition encode related physical choices in time and frequency descriptions. The former excludes response before excitation; the latter excludes waves arriving from infinity without a prescribed source. These constructions are fundamental to scattering theory, antenna models, acoustic radiation, and the analysis of wave-based inverse problems.

Variable media, rays, and curved geometry

The constant-coefficient equation assumes an idealized homogeneous medium. A useful generalization is

ρ(x)utt−∇⋅(A(x)∇u)=f,\rho(x)u_{tt} -\nabla\cdot\bigl(A(x)\nabla u\bigr)=f,

where ρ(x)>0\rho(x)>0 and A(x)A(x) is a symmetric positive-definite matrix. The matrix can encode anisotropy, so propagation speed depends on direction as well as position. If the coefficients are time independent, the natural energy is

E(t)=12∫[ρut2+∇u⋅A∇u]dx.E(t)=\frac12\int \left[ \rho u_t^2+\nabla u\cdot A\nabla u \right]dx.

In the absence of forcing and boundary flux, this energy is conserved.

The divergence form is physically consequential. Expanding ∇⋅(a∇u)\nabla\cdot(a\nabla u) produces both aΔua\Delta u and a term involving ∇a⋅∇u\nabla a\cdot\nabla u. Consequently, replacing a constant speed by c(x)c(x) in utt=c2Δuu_{tt}=c^2\Delta u does not automatically give the correct equation for an inhomogeneous material. Conservation laws and the choice of dependent variable determine the appropriate coefficients and interface conditions.

For example, linear acoustics with spatially varying equilibrium density ρ\rho and bulk modulus KK gives

1Kptt−∇⋅(1ρ∇p)=0.\frac{1}{K}p_{tt} -\nabla\cdot\left(\frac{1}{\rho}\nabla p\right)=0.

At an ideal fluid interface without a singular source, pressure and normal particle velocity are continuous. These conditions lead to reflected and transmitted waves. In isotropic media, matching phase along the interface yields Snell’s law,

sin⁡θ1c1=sin⁡θ2c2,\frac{\sin\theta_1}{c_1} = \frac{\sin\theta_2}{c_2},

where the angles are measured from the interface normal.

When wavelengths are short compared with the scale of coefficient variation, a high-frequency ansatz

u(x,t)≈a(x,t)eiϕ(x,t)/ε,ε≪1,u(x,t)\approx a(x,t)e^{i\phi(x,t)/\varepsilon}, \qquad \varepsilon\ll1,

leads at highest order to an eikonal equation. For utt−c(x)2Δu=0u_{tt}-c(x)^2\Delta u=0, it is

ϕt2=c(x)2∣∇ϕ∣2.\phi_t^2=c(x)^2|\nabla\phi|^2.

Surfaces of constant phase are wavefronts; their propagation can be described by rays. In a stationary isotropic medium, travel time defines a spatial geometry with metric proportional to c(x)−2c(x)^{-2} times the Euclidean metric, and rays follow the associated geodesics. The approximation can fail near caustics, where neighboring rays focus.

On a curved spatial manifold, the Laplace operator is replaced by the Laplace–Beltrami operator. In relativity, the spacetime wave operator associated with a Lorentzian metric tensor is written

□gu=∣det⁡g∣−1/2∂μ(∣det⁡g∣1/2gμν∂νu).\Box_g u= |\det g|^{-1/2} \partial_\mu\left( |\det g|^{1/2}g^{\mu\nu}\partial_\nu u \right).

Its characteristics are null hypersurfaces, and high-frequency disturbances follow null geodesics. Curvature can scatter waves and generate tails, so the sharp Huygens property of flat three-dimensional space does not extend automatically to general curved spacetimes.

Numerical approximation and computational constraints

The wave equation is a central test problem in numerical analysis because a useful approximation must reproduce both stable propagation and accurate phase. A scheme may remain bounded while accumulating enough phase error to place a pulse or oscillation at the wrong location. Long-time simulation, high-frequency propagation, and repeated reflections make this distinction especially important.

For the one-dimensional equation, a standard centered finite-difference method uses spatial step Δx\Delta x and time step Δt\Delta t:

ujn+1=2ujn−ujn−1+r2(uj+1n−2ujn+uj−1n),r=cΔtΔx.u_j^{n+1} = 2u_j^n-u_j^{n-1} +r^2\left(u_{j+1}^n-2u_j^n+u_{j-1}^n\right), \qquad r=\frac{c\Delta t}{\Delta x}.

For smooth solutions this approximation is second order in both space and time. It requires two initial time levels, so the first update must incorporate the prescribed initial velocity, usually through a Taylor expansion consistent with the equation.

Fourier analysis of the scheme gives the numerical dispersion relation

sin⁡2(ωnumΔt2)=r2sin⁡2(kΔx2).\sin^2\left(\frac{\omega_{\mathrm{num}}\Delta t}{2}\right) = r^2\sin^2\left(\frac{k\Delta x}{2}\right).

Stability for the standard infinite-grid problem requires r≤1r\leq1. This is an instance of the Courant–Friedrichs–Lewy condition: an explicit method’s numerical domain of dependence must encompass the physical one. For a rectangular grid in dd dimensions, the corresponding condition is

c2Δt2∑j=1d1Δxj2≤1.c^2\Delta t^2\sum_{j=1}^{d}\frac{1}{\Delta x_j^2}\leq1.

On an equal-spacing grid it becomes cΔt/Δx≤1/dc\Delta t/\Delta x\leq1/\sqrt d.

The numerical dispersion relation differs from the exact relation ω=c∣k∣\omega=c|k|. Shorter resolved wavelengths generally have larger phase errors, and multidimensional grids can introduce artificial directional dependence. Increasing resolution, using higher-order differences, or employing spectral methods can reduce these effects. The appropriate resolution depends on the desired accuracy and simulated travel distance, not simply on whether a wavelength is represented by more than two points.

The finite element method starts from a weak spatial formulation and typically produces

Mq¨+Kq=F,M\ddot q+Kq=F,

where MM and KK are mass and stiffness matrices. This approach accommodates irregular domains and heterogeneous coefficients. Time integration can use explicit methods with suitable mass treatment or implicit methods with different stability and cost properties. Energy-conserving and energy-controlled integrators are useful when artificial damping or growth would obscure the intended dynamics.

Unbounded-domain simulations require artificial boundaries. Reflecting truncations can send spurious waves back into the region of interest, motivating absorbing boundary conditions and perfectly matched layers. A simple exact outgoing condition at a right boundary in one dimension is

ut+cux=0.u_t+cu_x=0.

In higher dimensions, exact nonreflecting conditions are generally more complicated and may be nonlocal. Boundary treatment, source discretization, coefficient interfaces, and mesh quality can therefore be as important as the interior update formula.

Related equations, nonlinear effects, and research problems

Several equations resemble the classical wave equation while possessing substantially different propagation properties. Adding linear damping gives

utt+2γut−c2Δu=f,γ>0.u_{tt}+2\gamma u_t-c^2\Delta u=f, \qquad \gamma>0.

For the unforced equation with no boundary flux,

dEdt=−2γ∫ut2 dx,\frac{dE}{dt}=-2\gamma\int u_t^2\,dx,

so energy decreases rather than remaining constant. Different damping mechanisms, including boundary damping and spatially localized damping, lead to different decay rates. Whether every ray reaches a damping region can be decisive.

The Klein–Gordon equation,

utt−c2Δu+ω02u=0,u_{tt}-c^2\Delta u+\omega_0^2u=0,

adds a restoring term and has dispersion relation

ω2=c2∣k∣2+ω02.\omega^2=c^2|k|^2+\omega_0^2.

Its phase and group speeds depend on frequency, although the equation still has finite propagation speed. In quantum mechanics and relativistic field theory, this equation describes a free massive scalar field after appropriate choices of units and constants. Its additional term changes long-time decay and tail behavior without changing the principal characteristic cone.

Nonlinear wave equations include

utt−c2Δu=F(u)u_{tt}-c^2\Delta u=F(u)

and equations in which the coefficients themselves depend on uu or its derivatives. The former are semilinear; the latter can be quasilinear. Nonlinearity destroys the general superposition principle. Depending on dimension, sign, exponent, and initial data, solutions can disperse, persist as coherent structures, or develop singularities in finite time. In nonlinear acoustics and compressible flow, wave speeds that depend on amplitude can cause compressive portions of a wave to steepen into shocks.

For the focusing power equation

utt−Δu=∣u∣p−1u,u_{tt}-\Delta u=|u|^{p-1}u,

the scaling

uλ(x,t)=λ2/(p−1)u(λx,λt)u_\lambda(x,t) = \lambda^{2/(p-1)}u(\lambda x,\lambda t)

preserves the equation. In spatial dimension n≥3n\geq3, the exponent

p=n+2n−2p=\frac{n+2}{n-2}

is energy critical: the natural energy is invariant under this scaling. This threshold organizes questions about global existence, concentration, blow-up, and scattering. The sign of the nonlinearity and the size and structure of the data remain essential; scaling alone does not determine an individual solution’s behavior.

Elastic waves require systems rather than a single scalar field. In a homogeneous isotropic solid, displacement u\mathbf u obeys

ρutt=μΔu+(λ+μ)∇(∇⋅u),\rho\mathbf u_{tt} = \mu\Delta\mathbf u +(\lambda+\mu)\nabla(\nabla\cdot\mathbf u),

where λ\lambda and μ\mu are Lamé parameters. Longitudinal and transverse disturbances travel at speeds

cP=λ+2μρ,cS=μρ.c_P=\sqrt{\frac{\lambda+2\mu}{\rho}}, \qquad c_S=\sqrt{\frac{\mu}{\rho}}.

Interfaces can convert one type into the other, producing behavior absent from the scalar constant-speed model.

Wave propagation also underlies difficult observation and reconstruction problems. Boundary measurements can be used to infer sound speed, density, obstacles, or initial sources, but incomplete aperture, noise, and trapped rays can limit uniqueness or stability. In control theory, the ability to observe or influence the entire wave field depends on how characteristic rays meet the measurement or control region. These geometric restrictions connect the local differential equation to global questions about domains, materials, and accessible data.