Characteristic surfaces and hyperbolicity

For the constant-speed wave equation $u_{tt}-c^2\Delta u=0$, with c positive, choose principal symbol $p(\tau,\xi)=\tau^2-c^2|\xi|^2$. A smooth hypersurface $\phi(t,x)=0$ is characteristic where its nonzero normal covector satisfies $p(\phi_t,\nabla\phi)=0$. The normal covector contains derivatives of the surface-defining function; it is not a point on the surface or a velocity vector. Multiplying the PDE or the defining function by a nonzero factor does not change this zero test.

A constant-time initial surface is noncharacteristic for this equation, so displacement and velocity can be prescribed there under the usual compatibility and regularity assumptions. In one spatial dimension the characteristic lines have slopes determined by the two traveling directions. A characteristic surface is instead adapted to transporting information along a front, and arbitrary independent Cauchy data cannot generally be prescribed on it in the same way. The noncharacteristic test is structural; global existence still requires a separate analysis of the equation and data.

Replacing the same spatial coefficient in every direction by different positive coefficients changes the propagation cone. A plane front is tested by inserting its spatial normal and temporal frequency into the symbol. Dividing the temporal frequency by the Euclidean length of the spatial normal gives the front's normal phase speed. A non-unit spatial normal must therefore be normalized before its coefficient is described as a physical speed; otherwise rescaling the surface equation would appear to change propagation.

In an anisotropic medium, the direction of a plane front's normal need not coincide with the direction in which a localized wave packet's group velocity points. Both arise from the same dispersion relation, but they are different derivatives or ratios of it. The examples here calculate normal front speed only. This precision matters when converting a symbol calculation into a domain-of-dependence shape or comparing it with an experiment that tracks packets rather than an extended plane crest.

For a first-order constant-coefficient system with time derivative isolated, spatial Fourier transformation gives an ordinary differential equation involving a matrix symbol for each spatial direction. Real eigenvalues describe real characteristic speeds, but a defective matrix can still create frequency-dependent growth. A Jordan block produces a factor proportional to frequency times time in its exponential. High-frequency perturbations then lose the uniform estimate expected of a well-posed evolution in a fixed Sobolev norm.

Strong hyperbolicity supplies a uniformly controlled diagonalization or an equivalent suitable symmetrizer, rather than real eigenvalues alone. Symmetric hyperbolic systems offer a concrete sufficient structure: a uniformly positive time matrix and symmetric spatial matrices yield energy estimates. These distinctions are especially important for repeated characteristic speeds. Strictly separated real speeds in a smooth finite-dimensional constant-coefficient setting are a convenient case, but a scalar classification slogan should not be substituted for a complete system estimate.