Normally Hyperbolic Operators

A normally hyperbolic operator is a wave operator whose highest-derivative part is fixed by the Lorentzian metric, even when the field has several components and lower-order terms mix them. Smooth coefficients and a smooth spacelike Cauchy hypersurface on a globally hyperbolic spacetime are enough for existence, uniqueness, continuous dependence, and finite propagation of compactly supported data. Formal self-adjointness is not required for this Cauchy theorem.

For the local PDE bridge, a differential operator acts distributionally by duality against compactly supported smooth test sections. Its principal symbol retains only the highest derivatives, and its support theorem says that a solution can be nonzero only in the causal set generated by the supports of its source and initial data. These notions are all that is assumed before the links below.

Required background. Partial Differential Equations and Green Functions provides principal symbols, distributional operators, Green functions, and support estimates; Classical Field Theory supplies linearized field equations, Cauchy data, and conserved-current reasoning. Helpful background. Flat-Spacetime Quantum Field Theory provides familiar scalar and spinor wave systems; Causality and Global Structure establishes Cauchy hypersurfaces, causal diamonds, and global hyperbolicity.

Metric principal part on a vector bundle

Let $E\to M$ be a finite-rank real or complex vector bundle. In the symbol convention used on this page, a second-order operator $P:C^\infty(M,E)\to C^\infty(M,E)$ is normally hyperbolic when

$$\sigma_2(P)(x,k) =g^{\mu\nu}(x)k_\mu k_\nu\,\operatorname{id}_{E_x}.$$

Equivalently, in a local frame its highest-derivative term is $g^{\mu\nu}\partial_\mu\partial_\nu$ times the identity; a symbol convention that replaces each derivative by $ik_\mu$ inserts an overall minus sign. The invariant content is the same: the characteristic covectors satisfy $g^{\mu\nu}k_\mu k_\nu=0$. The metric null cone, not a matrix eigenvalue of a lower-order potential, determines finite propagation.

Every such operator can be written uniquely as

$$P=\Box_\nabla+B,$$

where $\nabla$ is a connection on $E$ selected by $P$, $\Box_\nabla$ is its connection wave operator, and $B$ is a smooth endomorphism. This decomposition and the scalar, form, and squared-Dirac examples are given in Bär, Ginoux, and Pfäffle 2007, §1.5, pp. 33–35. First-order couplings can be absorbed into this connection; zero-order mass, curvature, or matrix potentials enter $B$. The connection determined by $P$ need not preserve a separately chosen fiber metric. Normal hyperbolicity alone makes no metric-compatibility or self-adjointness claim.

The global Cauchy theorem

Assume all of the following:

• $(M,g)$ is a smooth, time-oriented, globally hyperbolic Lorentzian manifold without timelike boundary;
• $\Sigma\subset M$ is a smooth spacelike Cauchy hypersurface with future unit normal $n$;
• $E\to M$ is a smooth finite-rank vector bundle and $P$ is normally hyperbolic with smooth coefficients; and
• $f\in C_c^\infty(M,E)$ and $u_0,u_1\in C_c^\infty(\Sigma,E)$.

Then there is a unique $u\in C^\infty(M,E)$ satisfying

$$Pu=f, \qquad u\rvert_\Sigma=u_0, \qquad \nabla_nu\rvert_\Sigma=u_1.$$

If $K=\operatorname{supp}f\cup\operatorname{supp}u_0\cup\operatorname{supp}u_1$, then

$$\operatorname{supp}u\subset J(K).$$

The map $(f,u_0,u_1)\mapsto u$ is linear and continuous in the standard compact-support and smooth topologies. This is the combined content of Bär, Ginoux, and Pfäffle 2007, Theorems 3.2.11–3.2.12, pp. 85–87. The compact-support assumption is the theorem’s clean test-section form; broader spacelike-compact or Sobolev data require the corresponding function-space statement rather than an unannounced extension.

Why the proof works

The proof has four distinct steps. First, a Green identity compares $P$ with its formal adjoint and converts the difference into a divergence. Integrated over a causal slab, this gives an energy-style uniqueness argument: zero source and zero Cauchy data force the solution to vanish in the relevant domain of dependence. The same argument yields the support inclusion.

Second, one proves local existence on sufficiently small relatively compact causal neighborhoods. Third, a partition of unity decomposes compact data into pieces supported in such neighborhoods. Global hyperbolicity supplies a smooth splitting by Cauchy hypersurfaces, allowing the local solutions to be extended forward and backward. Compactness of a causal diamond intersected with a finite time slab prevents the support from escaping while the extension is continued. If a maximal extension stopped at finite time, one more uniform local step would extend it, a contradiction.

Finally, continuous dependence follows by viewing the Cauchy map as a continuous linear bijection between appropriate Fréchet spaces and applying the open mapping theorem on each fixed compact support. The concise scalar version of existence, causal dependence, and the Green-operator construction appears in Hollands and Wald 2015, §2.1.

Finite propagation is not stability

The support bound says where a disturbance may occur; it does not bound its amplitude. A smooth negative matrix potential in $B$ may produce exponentially growing modes while leaving the principal symbol and causal cone unchanged. Conversely, an energy estimate may control growth but cannot repair missing global data at a timelike boundary. Causality, existence, uniqueness, continuity, and stability are related but logically separate properties.

For the flat Klein–Gordon operator, spatial Fourier transformation gives independent equations

$$\ddot{\widehat u}(t,\mathbf k) +\left(\lvert\mathbf k\rvert^2+m^2\right)\widehat u(t,\mathbf k) =-\widehat f(t,\mathbf k),$$

whose two initial values determine each mode. This verifies uniqueness and the mass sign. Finite propagation is less obvious in Fourier space, but follows from the hyperbolic support theorem: the Fourier decomposition must not be mistaken for instantaneous spatial influence.

What fails when a hypothesis is dropped

On the half-space $x>0$ in two-dimensional Minkowski spacetime, take the wave equation with zero data at $t=0$ but no boundary condition at $x=0$. If $h(s)$ is smooth, vanishes for $s\leq0$, and is not identically zero, then

$$u(t,x)=h(t-x)$$

solves the wave equation, has zero initial data for every $x>0$, and later enters from the timelike boundary. Thus the same initial data admit both $u=0$ and a nonzero solution. The local operator remains normally hyperbolic; what failed is that the stated hypersurface is not Cauchy for a boundaryless globally hyperbolic problem. Boundary conditions are additional data, not a consequence of the principal symbol.

Check your understanding

Let $P=\Box_\nabla+A^\mu\nabla_\mu+B$ on a smooth vector bundle, where $A^\mu$ and $B$ are arbitrary smooth matrix-valued coefficients. Must $A^\mu$ and $B$ be Hermitian for the Cauchy theorem above to apply?

Solution

No. Hermiticity is relevant when constructing conserved inner products, adjoints, or a quantum star-algebra, but it is not one of the Cauchy theorem’s hypotheses. The matrices multiply at most one derivative, so they do not alter the metric principal symbol. Smoothness, normal hyperbolicity, a globally hyperbolic spacetime, a genuine Cauchy hypersurface, and the stated data class give existence and uniqueness. Non-Hermitian lower-order terms may cause growth or loss of a positive energy estimate, yet that is a stability issue rather than failure of local causal well-posedness. To diagnose the operator, first discard every term below second differential order and inspect the remaining quadratic form in the covector. Only after that causal check should one ask whether an inner product makes the full operator symmetric or whether its evolution is bounded. Therefore, conflating those questions would add an unnecessary hypothesis to the theorem.

References

• Bär, Christian, Nicolas Ginoux, and Frank Pfäffle. Wave Equations on Lorentzian Manifolds and Quantization. ESI Lectures in Mathematics and Physics. Zürich: European Mathematical Society, 2007. DOI 10.4171/037.
• Hollands, Stefan, and Robert M. Wald. “Quantum Fields in Curved Spacetime.” Physics Reports 574 (2015): 1–35. DOI 10.1016/j.physrep.2015.02.001.