Boundaries, Non-Global Hyperbolicity, and Well-Posedness

Global hyperbolicity makes a strong promise: compactly supported Cauchy data determine one global solution, with no signal arriving from an unprescribed edge. Once that promise fails, “add a boundary condition” is sometimes the correct repair, but not a universal one. As of the evidence cutoff 2026-08-19, the defensible conclusion is conditional: timelike-boundary problems can be well posed after an admissible boundary dynamics is chosen, whereas chronology horizons exhibit obstructions that cannot be reduced to a missing Robin parameter.

For the local PDE bridge, a Green operator inverts a differential operator on compactly supported test sections and restricts support to the causal future or past of the source. Global hyperbolicity supplies the causal compactness used to prove uniqueness. With a timelike boundary, a causal curve can reach the boundary and return, so initial data alone do not determine the returning signal.

Required background. Partial Differential Equations and Green Functions provides weak solutions, energy estimates, and Green-operator support; Classical Field Theory supplies conserved energy and boundary flux. Helpful background. Flat-Spacetime Quantum Field Theory provides mode evolution and self-adjoint generators; Causality and Global Structure establishes domains of dependence, Cauchy horizons, and global hyperbolicity.

What the globally hyperbolic theorem actually buys

For a normally hyperbolic operator $P$ on a globally hyperbolic spacetime without boundary, smooth compact Cauchy data supported in $K$ give a unique smooth solution supported in the causal hull $J(K)=J^+(K)\cup J^-(K)$. The causal hull is different from the domain of dependence $D(K)$: a wave may spread beyond $D(K)$ while remaining inside $J(K)$. The same hypotheses give unique advanced and retarded Green operators. The existence and support statements are proved in Bär, Ginoux, and Pfäffle 2007, Theorems 3.2.11–3.2.12, pp. 85–87, and Corollary 3.4.3, p. 89.

Three ingredients are bundled together here: local hyperbolicity of the differential equation, a global causal geometry, and absence of an unprescribed timelike edge. Keeping only the first does not keep the theorem. Anti-de Sitter spacetime, for example, is smooth and the Klein–Gordon equation remains locally hyperbolic, but causal influence can reach conformal infinity and return in finite time. One must say what happens there.

This also clarifies a common overstatement. Failure of global hyperbolicity does not itself prove that no dynamics exists. It proves that the standard Cauchy theorem no longer selects one without further structure. The additional structure may be a boundary condition, an operator domain, a flux law, or—in causally pathological cases—something for which no acceptable choice is known.

Static timelike boundaries as extension problems

On a static spacetime, write a scalar equation schematically as

$$\partial_t^2\phi=-A\phi,$$

where $A$ is a positive symmetric spatial operator initially defined on compactly supported functions away from the boundary. If $A$ is essentially self-adjoint, its closure fixes the evolution. If it has several positive self-adjoint extensions $A_E$, each domain supplies a different admissible reflection law and

$$\phi(t)=\cos(tA_E^{1/2})\phi_0 +A_E^{-1/2}\sin(tA_E^{1/2})\dot\phi_0$$

defines the corresponding energy-conserving evolution on its spectral domain. The combination $A_E^{-1/2}\sin(tA_E^{1/2})$ means the spectral multiplier $\sin(t\sqrt a)/\sqrt a$, with value $t$ at $a=0$; it does not require a bounded inverse of $A_E$. Under explicit locality, time-translation, and positive-energy assumptions, Ishibashi and Wald show that satisfactory static dynamics arise this way; the result is a classification under hypotheses, not a theorem about arbitrary non-global hyperbolicity Ishibashi and Wald 2003, §§2–3, especially Theorem 3.1.

The half-line makes the nonuniqueness concrete. For $A=-d^2/dx^2$ on $x>0$, domains satisfying

$$u'(0)=\lambda u(0)$$

give self-adjoint Robin extensions on $L^2(0,\infty;dx)$ for every real $\lambda$, with Neumann at $\lambda=0$ and Dirichlet as a limiting condition. Here $u'(0)$ means the derivative toward increasing $x$, not the outward derivative $-u'(0)$. Integration by parts fixes the positivity condition explicitly:

$$\langle u,Au\rangle =\int_0^\infty\lvert u'\rvert^2dx+\lambda\lvert u(0)\rvert^2.$$

Thus $A$ is nonnegative for $\lambda\geq0$. If $\lambda\lt0$, the square-integrable mode $u(x)=e^{\lambda x}$ obeys the boundary condition and has eigenvalue $-\lambda^2$. Its wave amplitude grows or decays exponentially. Self-adjointness and perfectly reflecting real-frequency waves therefore do not guarantee stability.

For a real wave the conserved energy includes a boundary term,

$$\mathcal E =\frac12\int_0^\infty(\dot\phi^2+\phi_x^2)\,dx +\frac\lambda2\phi(t,0)^2.$$

The bulk energy alone changes at rate $-\lambda\phi(t,0)\dot\phi(t,0)$: a Robin wall can store and return energy. Its symplectic flux nevertheless vanishes for two solutions obeying the same Robin condition.

The credible competing positions are best understood as choices of system boundary rather than contradictory verdicts. The extension formulation is powerful for static, conservative systems. Energy estimates can also handle time-dependent geometries, inhomogeneous data, and boundary laws that exchange energy with an exterior. A dissipative boundary condition may yield well-posed evolution through a contraction semigroup, but it is not a positive self-adjoint extension and does not describe a closed unitary system.

Anti-de Sitter boundaries and regularity

Exact AdS separates into modes, and the radial operator can be essentially self-adjoint in one mass range but admit a family of positive extensions in another. In the latter range, ratios of the two allowed asymptotic falloffs encode generalized Dirichlet, Neumann, or Robin behavior. Ishibashi and Wald derive this dependence for scalar, electromagnetic, and gravitational perturbations, including the positivity restriction on the extension Ishibashi and Wald 2004, §3, especially §3.3 and Theorem 3.2.

Exact symmetry is not essential to every well-posedness result. For a class of asymptotically AdS spacetimes that need not be stationary or separable, Warnick formulates weighted weak problems using a renormalized energy. Dirichlet and Neumann existence and uniqueness are established at the $H^1$ level, with higher regularity under stronger compatibility assumptions; inhomogeneous and Robin data are then treated as boundary problems Warnick 2013, §§2–4, especially Theorems 4.1–4.2, §6.2.

Thus “well posed” must always name the function space, boundary regularity, allowed mass range, and energy estimate. Smooth mode solutions on exact AdS do not by themselves establish a general asymptotically AdS theorem. Conversely, an $H^1$ weak theorem does not automatically grant smoothness at the conformal boundary or a unitary quantum evolution.

Chronology horizons are a different adverse case

A compactly generated Cauchy horizon is not merely a reflecting wall. Null generators repeatedly accumulate in a compact region, and singularities propagate along the resulting almost-closed null geometry. Kay, Radzikowski, and Wald prove two sharply localized obstructions for a linear scalar field: the usual algebra on the initial globally hyperbolic region cannot be extended while retaining their local compatibility condition at base points, and an extended bisolution that was Hadamard initially fails the local Hadamard property there Kay, Radzikowski, and Wald 1997, §5, Theorems 1–2.

The scope matters. The theorem concerns base points of compactly generated Cauchy horizons and specified algebraic and state conditions. It does not say that every non-globally-hyperbolic spacetime is quantum mechanically impossible, nor that the stress tensor diverges at every point of every chronology horizon. It does show why a universal prescription “choose a self-adjoint extension at the horizon” is untenable: the obstruction is microlocal and causal, not simply missing boundary flux data.

Results and limitations as of 2026-08-19

Regime and source	Assumptions	Established result	Limitation or unresolved discriminator
Globally hyperbolic baseline — Bär, Ginoux, and Pfäffle 2007, §3.2, Theorems 3.2.11–3.2.12	Normally hyperbolic operator; globally hyperbolic spacetime; no timelike boundary	Unique Cauchy evolution and causal Green operators	Does not cover timelike boundaries or chronology horizons
Static non-globally-hyperbolic dynamics — Ishibashi and Wald 2003, §§2–3, Theorem 3.1	Staticity, locality assumptions, time translations, positive conserved energy	Admissible prescriptions correspond to positive self-adjoint spatial extensions	Does not classify dissipative, time-dependent, or arbitrary causal pathologies
Asymptotically AdS initial-boundary problem — Warnick 2013, §§2–4 and §6	Specified mass window, weighted spaces, compatible Dirichlet, Neumann, or Robin data	Existence and uniqueness of weak solutions; higher regularity with stronger data	Quantum state selection and interacting backreaction are outside the theorem
Compactly generated Cauchy horizon — Kay, Radzikowski, and Wald 1997, §5, Theorems 1–2	Linear scalar field; initial globally hyperbolic region; F-locality or Hadamard bisolution; horizon base point	Algebraic extension and Hadamard continuation fail in the stated senses	Not a blanket result for every horizon point, field, algebra, or non-global geometry

The strongest common conclusion is therefore regime-dependent. Reflecting boundary choices are genuine physical inputs, not technical decorations. Dissipative choices require a different generator class. At chronology horizons, even locally conventional quantum behavior can be obstructed. Confidence is high for these delimited statements and low for any attempt to collapse them into one universal rule.

Diagnostic: two Robin dynamics from one equation

For a massless mode on $x>0$, take

$$\phi(t,x)=e^{-i\omega t} \left(e^{-ikx}+R(k)e^{ikx}\right), \qquad \omega=k>0,$$

and impose $\partial_x\phi(t,0)=\lambda\phi(t,0)$ with real $\lambda$. Find $R(k)$, verify energy-conserving reflection, and identify the Dirichlet and Neumann limits.

Solution

At $x=0$ the boundary condition gives $-ik+ikR=\lambda(1+R)$, hence

$$R(k)=\frac{ik+\lambda}{ik-\lambda}.$$

For real $k$ and $\lambda$, numerator and denominator have equal modulus, so $\lvert R\rvert=1$; incoming and reflected monochromatic waves have equal time-averaged energy flux. The instantaneous bulk flux need not vanish, because the Robin boundary term can store energy. Neumann data give $R=1$ at $\lambda=0$, while $\lambda\to+\infty$ gives $R=-1$, the Dirichlet phase. For $\lambda\lt0$, the pole at $k=-i\lambda$ corresponds to the unstable bound mode found above; real-frequency unit reflection does not exclude it. The same bulk equation therefore admits inequivalent dynamics until the boundary parameter is supplied. This diagnostic does not model a chronology horizon, because it presumes an ordinary timelike edge and a conservative local reflection law.

What could change these conclusions

The conclusions could change if a theorem enlarges or narrows the admissible spacetime, field, state, algebra, regularity, or energy class; if an independent calculation changes the Robin or AdS limiting behavior; if a counterexample invalidates the causal or microlocal obstruction; or if a new boundary or regulator analysis establishes well-posed evolution outside the regimes in the matrix. A claimed chronology-horizon state must be checked specifically against the hypotheses and base-point conclusion of Kay–Radzikowski–Wald, while a new AdS proposal must state its flux, function space, positivity, and compatibility conditions.

References

• Bär, Christian, Nicolas Ginoux, and Frank Pfäffle. Wave Equations on Lorentzian Manifolds and Quantization. European Mathematical Society, 2007. Open PDF, arXiv:0806.1036.
• Ishibashi, Akihiro, and Robert M. Wald. “Dynamics in Non-Globally-Hyperbolic Static Spacetimes II: General Analysis of Prescriptions for Dynamics.” Classical and Quantum Gravity 20 (2003): 3815–3826. arXiv:gr-qc/0305012.
• Ishibashi, Akihiro, and Robert M. Wald. “Dynamics in Non-Globally-Hyperbolic Static Spacetimes III: Anti-de Sitter Spacetime.” Classical and Quantum Gravity 21 (2004): 2981–3014. arXiv:hep-th/0402184.
• Kay, Bernard S., Marek J. Radzikowski, and Robert M. Wald. “Quantum Field Theory on Spacetimes with a Compactly Generated Cauchy Horizon.” Communications in Mathematical Physics 183 (1997): 533–556. arXiv:gr-qc/9603012.
• Warnick, Claude M. “The Massive Wave Equation in Asymptotically AdS Spacetimes.” Communications in Mathematical Physics 321 (2013): 85–111. arXiv:1202.3445.