Why Global Hyperbolicity Matters for Field Propagation

Global hyperbolicity is the standard geometric hypothesis that turns a local wave operator into a global propagation theory without extra boundary data. For a normally hyperbolic linear operator on a smooth boundaryless spacetime, it gives a well-posed Cauchy problem, finite propagation speed, and unique future- and past-supported Green operators. These conclusions belong to a specified operator class; they are not automatic statements about every gauge system, nonlinear equation, or spacetime with a timelike boundary.

Required background. Geodesics and Jacobi Fields supplies the causal geodesics that generate propagation boundaries. Helpful background. Lorentzian Metrics and Causal Cones fixes the principal cone; Causality and Global Structure supplies compact diamonds; Partial Differential Equations and Green Functions supplies distributions, test functions, and Green-function language.

Normally hyperbolic operators

Let $E\to M$ be a real or complex vector bundle over a smooth Lorentzian spacetime. A second-order linear differential operator

$$P:C^\infty(M,E)\longrightarrow C^\infty(M,E)$$

is normally hyperbolic when its principal symbol is the metric quadratic form times the identity on the fiber:

$$\sigma_P(x,\xi) =-g^{\mu\nu}(x)\xi_\mu\xi_\nu\,\operatorname{id}_{E_x}.$$

The overall sign is a convention; the invariant content is that the characteristic covectors are precisely null covectors and that every bundle component has the same metric principal cone. Lower-order terms may mix components without changing normal hyperbolicity.

Examples include the scalar Klein–Gordon operator with smooth mass and curvature coupling, connection wave operators on vector bundles, and the square of a Dirac-type operator. A first-order Dirac operator is not itself normally hyperbolic, although it has a related Green-hyperbolic theory. Maxwell’s equations and the linearized Einstein equation require gauge treatment and constraints before their reduced evolution operators take normally hyperbolic form. The theorem below should therefore not be quoted for an unreduced gauge system merely because waves are present.

The global Cauchy theorem

Assume that $M$ is globally hyperbolic and has no boundary, $S\subset M$ is a smooth spacelike Cauchy hypersurface, and $n$ is its future-directed unit normal. For compactly supported smooth sections $u_0,u_1$ on $S$ and a compactly supported smooth source $f$ on $M$, the problem

$$Pu=f, \qquad u\big|_S=u_0, \qquad \nabla_nu\big|_S=u_1$$

has a unique smooth solution $u$. It depends continuously on $(f,u_0,u_1)$ in the standard test-function and smooth-section topologies. If

$$K=\operatorname{supp}(f)\cup\operatorname{supp}(u_0) \cup\operatorname{supp}(u_1),$$

then finite propagation gives

$$\operatorname{supp}(u)\subset J(K)=J^+(K)\cup J^-(K).$$

The exact statement and proof are Bär, Ginoux, and Pfäffle 2007, Theorems 3.2.11–3.2.12, pp. 85–87. Local energy estimates give uniqueness and local existence. A Cauchy temporal splitting covers the spacetime by local solution regions, while compact causal influence allows those solutions to be patched and extended without an escaping sequence of support. Global hyperbolicity is doing global work; normal hyperbolicity alone is only a local principal-symbol condition.

The support inclusion means there is no propagation outside the metric causal cone. It does not assert Huygens’ principle. In curved spacetime, a solution can have a tail throughout the timelike interior of the cone even though its singularities propagate along null bicharacteristics. “Finite speed,” “support only on the null cone,” and “singular support on null directions” are three different claims.

Green operators and support orientation

Let $C_c^\infty(M,E)$ denote compactly supported smooth sections. To avoid a widespread naming ambiguity, define operators by their support rather than by a sign label. A future-supported Green operator $G_{\mathrm{fut}}$ satisfies, for every $f\in C_c^\infty(M,E)$,

$$PG_{\mathrm{fut}}f=f, \qquad G_{\mathrm{fut}}Pf=f, \qquad \operatorname{supp}(G_{\mathrm{fut}}f) \subset J^+(\operatorname{supp}f).$$

The past-supported operator $G_{\mathrm{past}}$ obeys the same inverse identities with support in $J^-(\operatorname{supp}f)$. Common physics terminology calls the future-supported operator retarded and the past-supported operator advanced, because a retarded response occurs after its source. Bär, Ginoux, and Pfäffle instead denote the future-supported operator $G_+$ and call it advanced, and denote the past-supported operator $G_-$ and call it retarded. Their Definition 3.4.1 and Corollary 3.4.3, pp. 88–89 are unambiguous because the support inclusions are explicit. Whenever labels differ, the $J^+$ or $J^-$ condition is the statement to preserve.

A Green operator maps a source to a smooth solution. A fundamental solution at a point is instead a distributional kernel solving an adjoint delta-source equation. The two are related by integration and adjunction, but they are not the same object. Their difference,

$$G=G_{\mathrm{fut}}-G_{\mathrm{past}},$$

is the causal propagator: on compactly supported sections $PG=GP=0$, and its image consists of spacelike-compact homogeneous solutions under the standard exactness hypotheses. None of these constructions selects a quantum state; they provide the classical causal structure used to impose commutation relations.

Why boundaries change the theorem

Suppose instead that the spacetime has a timelike boundary. A causal curve can enter from that boundary without ever meeting the proposed initial surface. Zero initial data therefore do not imply a zero solution unless boundary data have also been fixed. Dirichlet, Neumann, Robin, dissipative, transparent, or mixed conditions define different problems; some are well posed, some are not, and their admissibility can depend on the field, mass, dimension, and energy form.

Anti-de Sitter spacetime is the canonical geometric example: its conformal infinity is timelike, so ordinary global hyperbolicity fails. For free fields, dynamics can be recovered in suitable regimes by choosing an admissible self-adjoint extension, but there is no universal instruction “use Dirichlet” independent of the operator. The field- and mass-dependent classification is developed in Ishibashi and Wald 2004, §2 and §§3.1–3.3, pp. 2985–3012.

A finite numerical domain has the same logical structure. A boundary condition may reflect, absorb, or inject radiation. The resulting domain of influence can include reflected characteristics, so the simple boundaryless inclusion in $J_M^\pm(\operatorname{supp}f)$ must be reformulated for that problem. Global hyperbolicity of the interior geometry does not choose the boundary operator.

Beyond the sufficient theorem

Global hyperbolicity is a powerful uniform sufficient condition for every normally hyperbolic operator of the stated type. It is not logically necessary for a particular operator to possess some Green operator. A causally compatible open subset of a globally hyperbolic spacetime can inherit Green operators, and special non-globally-hyperbolic geometries can admit them after extra choices. Existence may then come without uniqueness; examples and the precise extension construction appear in Bär, Ginoux, and Pfäffle 2007, §3.5, pp. 92–102.

Conversely, global hyperbolicity does not by itself prove well-posedness for the nonlinear Einstein equation. One must choose a hyperbolic reduction, solve the constraints, control gauge propagation, and show that the reduced solution satisfies the original equations. Those analytic tasks belong to Formulations of General Relativity. The role of this chapter is the causal foundation: Cauchy slices and compact diamonds make a global hyperbolic evolution theorem possible.

Exercise

On the Minkowski half-space $x>0$ in $1+1$ dimensions, consider the scalar wave equation $(-\partial_t^2+\partial_x^2)u=0$. Construct a nonzero smooth solution whose field and first time derivative both vanish on $t=0$, and explain why this does not contradict uniqueness on globally hyperbolic Minkowski spacetime.

Solution

Choose a nonzero smooth function $F$ supported in the positive half-line and set

$$u(t,x)=F(t-x).$$

This is a right-moving wave and satisfies the wave equation. At $t=0$ and $x>0$, the argument is $-x<0$, so both $F(-x)$ and $F'(-x)$ vanish. Thus $u$ and $\partial_tu$ have zero initial data. Nevertheless, $u$ becomes nonzero when $t-x$ enters the support of $F$: a signal has entered through the timelike boundary $x=0$.

The half-space without a boundary condition is not the boundaryless globally hyperbolic setting of the theorem, and $t=0$ is not a Cauchy surface for it. Imposing, for example, homogeneous Dirichlet data would exclude this particular solution unless $F(t)=0$ at the boundary for all $t$.

References

• Bär, Christian, Nicolas Ginoux, and Frank Pfäffle, Wave Equations on Lorentzian Manifolds and Quantization, European Mathematical Society, Chapter 3, especially §§3.2–3.5, pp. 73–102, 2007.
• Ishibashi, Akihiro, and Robert M. Wald, “Dynamics in Non-Globally-Hyperbolic Static Spacetimes III: Anti-de Sitter Spacetime”, Classical and Quantum Gravity 21, §2 and §§3.1–3.3, pp. 2981–3014, 2004.
• Minguzzi, Ettore, “Lorentzian Causality Theory”, Living Reviews in Relativity 22, article 3, §§3–4, 2019.