Boundary Conditions, Stability, and Censorship

Boundary conditions are part of the dynamical law in asymptotically anti-de Sitter gravity. They decide which fluxes enter or leave, which charges are conserved, which modes are allowed, and even which formulation of weak cosmic censorship is being tested. De Sitter, anti-de Sitter, and asymptotically flat collapse therefore cannot be compared by changing only the sign of $\Lambda$ in a local field equation.

Helpful background. Black-Hole Geometry and Mechanics supplies horizon definitions; Black-Hole Perturbation Theory supplies boundary spectra; Formulations of General Relativity supplies initial-boundary problems and charges; Exact Solutions of Einstein’s Equations supplies de Sitter and AdS families; Causality and Global Structure supplies conformal infinity; and Computational Relativity supplies nonlinear validation.

Three causal boundaries, three evolution problems

For $\Lambda=0$, future null infinity $\mathscr I^+$ is null, and isolated outgoing radiation can carry Bondi energy away. For $\Lambda\gt 0$, future conformal infinity is spacelike; a static observer instead loses signals across a cosmological horizon. For $\Lambda\lt 0$, conformal infinity is timelike. Data on an initial spacelike slice do not determine the AdS future until boundary conditions at infinity have also been imposed.

A schematic balance for the energy $E_{\rm ext}$ stored outside any black-hole horizon is

$$E_{\rm ext}(t_2)-E_{\rm ext}(t_1) =-\int_{t_1}^{t_2}\!dt \int_{\mathscr I_t}\!\mathcal F_{\infty}\,dS -\Delta E_H,$$

where $\mathcal F_{\infty}$ is outward flux through conformal infinity and $\Delta E_H$ is the net energy crossing a future black-hole horizon during $[t_1,t_2]$. Thus $\Delta E_H$ is already integrated over horizon time and area; it vanishes when no horizon is present. Under stationary, source-free reflecting data, the total asymptotic charge—black-hole energy plus exterior energy—is conserved, while horizon absorption is an internal transfer. The labels “closed,” “dissipative,” and “driven” refer to this distinction:

• Reflecting data set the conformal-boundary flux to zero. The total asymptotic charge is conserved, but $E_{\rm ext}$ can decrease when energy enters a horizon.
• Absorbing or transparent data allow positive outward flux and can suppress repeated refocusing.
• Time-dependent sources can inject or extract energy, so the boundary charge obeys a sourced Ward identity rather than an isolated conservation law.
• Mixed or designer data relate the independent asymptotic coefficients of a field. Whether they define positive, stable evolution depends on mass, dimension, sector, and the chosen boundary functional.

At linear order, admissible conservative dynamics are associated with positive self-adjoint extensions of the reduced spatial operator. Ishibashi and Wald derive this statement and the AdS mode reduction in Ishibashi and Wald 2004, §1, pp. 2–5, and §§2.3–2.5, pp. 13–24, PDF. Their scalar, electromagnetic, and gravitational classifications show why “use Robin conditions” is incomplete without a field sector and parameter range Ishibashi and Wald 2004, §3.3, Theorem 3.2, pp. 32–38, PDF. Linear positivity is necessary evidence for stability, not a nonlinear-collapse theorem.

A scalar example of admissible boundary data

For $(\Box-m_{\rm eff}^2)\phi=0$ on AdS$_4$, where $m_{\rm eff}^2$ includes any curvature coupling, the asymptotic radial equation admits powers

$$\phi\sim\alpha(t,\Omega)\left(\frac Lr\right)^{\Delta_-} +\beta(t,\Omega)\left(\frac Lr\right)^{\Delta_+}, \qquad \Delta_\pm=\frac32\pm\nu, \quad \nu^2=\frac94+m_{\rm eff}^2L^2.$$

For $0\lt\nu\lt1$, both branches are normalizable in the Hilbert space used for static wave evolution, and a real relation between $\alpha$ and $\beta$ can define a self-adjoint extension. Only a restricted range gives a positive operator. At $\nu=0$ the second solution is logarithmic; at $\nu\ge1$ there is a unique extension in this class. For $\nu^2\lt0$, below the Breitenlohner–Freedman bound, the operator is unbounded below. These are linear statements with regularity imposed at the center Ishibashi and Wald 2004, §3.3, Theorem 3.2, pp. 32–38, PDF.

In particular, a minimally coupled massless scalar has $\nu=3/2$. It does not admit the same one-parameter Robin family as a scalar in $0\lt\nu\lt1$ under these hypotheses. This is why a boundary-condition label cannot be specified independently of the field and its function space.

Visibility must be defined relative to the boundary

Weak cosmic censorship is an asymptotic visibility statement. In asymptotically flat collapse one commonly asks whether singular behavior can communicate with $\mathscr I^+$. In AdS the relevant observers live at a timelike conformal boundary, and their access depends on the boundary metric, sources, and allowed falloffs. In de Sitter asymptotics, cosmological horizons restrict the observers who can receive a signal, while future conformal infinity is spacelike.

Consequently, an AdS censorship claim should declare:

1. the field equations and matter model;
2. global or Poincaré AdS and the conformal completion;
3. boundary conditions and whether the boundary is driven;
4. the initial-data topology and regularity;
5. the visible diagnostic—geodesic incompleteness, divergent curvature, or arbitrarily large curvature seen from the boundary;
6. whether the phenomenon occurs at finite boundary time or only asymptotically.

These distinctions are consequential. A boundary source that continually performs work is not an isolated collapse experiment. Conversely, excluding a source does not make all reflecting data stable; confinement can amplify small nonlinear interactions.

What the censorship experiments establish

Evidence cutoff: 12 August 2026. Several influential examples point in different directions because they pose different problems.

Problem	Result supported by the evidence	Essential boundary of the claim
Reflecting global AdS with spherical massless scalar data	Repeated returns produce delayed collapse for broad numerical families, while stable islands prevent a universal-collapse claim	Finite sampled families, finite runtime, spherical matter model
Spherical scalar domain-wall data with a negative potential	A proposed naked-singularity scenario instead formed a small black hole in the numerical evolution	One potential, symmetry class, and initial-data construction; not a general censorship theorem
Poincaré AdS$_4$ Einstein–Maxwell with a slowly increased boundary electric field	Supercritical driving produced curvature growing without bound where it remained visible to boundary observers	Axisymmetric, boundary-driven, no charged matter, numerical, and no finite-time singularity was demonstrated
The same electric setup with a sufficiently charged scalar	A linear scalar instability and regular static hairy branches supply a candidate route to bounded fields	No complete time-dependent settling evolution was performed; selection of a regular endpoint and other censorship mechanisms remain open
de Sitter collapse	Cosmological redshift and spacelike future infinity change decay and visibility; exact dust collapse can still form a black hole inside a cosmological horizon	Observer patch, matter model, and endpoint definition must be fixed

The domain-wall example is a useful negative control. Garfinkle evolved thick-wall data designed to challenge censorship and found a concentrated ingoing pulse forming a small black hole rather than a naked singularity Garfinkle 2004, §§II–IV, pp. 1–6. The conclusion and its failure of the assumed homogeneous approximation are given in Garfinkle 2004, §V, pp. 6–7. One failed counterexample increases confidence only within its declared family.

The boundary-driven Einstein–Maxwell experiment is different. Crisford and Santos begin in the Poincaré AdS vacuum, turn on a localized axisymmetric boundary electric field, and evolve beyond the largest amplitude admitting the relevant smooth stationary solution. They observe gauge-invariant electromagnetic and curvature quantities growing without bound along the horizon while remaining visible from the boundary Crisford and Santos 2017, Eqs. (1)–(9) and Figs. 1–3, pp. 1–4. The growth is inferred as boundary time tends late; the calculation does not exhibit a curvature singularity forming at a finite boundary time.

Adding matter changes the theory. For a charged scalar above a mass-dependent charge threshold, the dangerous electric background is linearly unstable to scalar condensation. Crisford, Horowitz, and Santos construct regular static hairy branches with bounded fields for the mechanism they test Crisford, Horowitz, and Santos 2018, §§III–V, pp. 4–20. They explicitly leave the complete time-dependent charged-scalar evolution undone Crisford, Horowitz, and Santos 2018, §V, p. 18. The result therefore supplies a candidate regular endpoint and shows that the particle spectrum can change one censorship experiment; it does not establish dynamical selection of that endpoint or prove that a weak-gravity condition repairs every possible AdS violation.

Stability is conditional, not a property of the metric alone

The same local AdS geometry can participate in inequivalent dynamical systems. Reflecting conditions produce a conservative cavity; dissipation competes with nonlinear focusing; a source can drive curvature growth; mixed data can add or remove stable branches. Stability must therefore be written as a statement of the form:

Solutions starting in a specified neighborhood of a background, in a specified norm, remain controlled or converge in a specified sense under a specified boundary evolution.

Changing any of these ingredients changes the proposition. A mode spectrum is an especially sensitive check: reflecting massless scalar data in global AdS have exact resonances, whereas admissible alternate boundary problems for other fields or mass ranges can shift frequencies or introduce unstable modes. Yet spectral nonresonance alone does not prove nonlinear stability, and resonance alone does not force every datum to collapse.

The major unresolved question is a generic nonlinear classification for vacuum and realistic matter across admissible AdS boundary conditions. A decisive result would state the data topology, boundary phase space, conserved charges, and visibility criterion, then prove or robustly compute the basin of censored and uncensored outcomes. An update should be triggered by such a theorem, by a validated counterexample stable under admissible perturbations and matter completion, or by a correction that changes the long-time curvature inference in an existing calculation.

Exercise: compare two apparent counterexamples

Experiment A evolves compact initial data with reflecting boundary conditions and no later source. Experiment B starts from vacuum but increases a boundary electric field forever toward a supercritical value. Both develop curvature visible from the boundary. Can they be treated as the same counterexample to one censorship conjecture?

Solution

No. Experiment A is an isolated initial-boundary-value problem: after the initial slice, its reflecting boundary carries no energy flux. Experiment B is driven; the boundary source performs work and changes the charge balance. Their admissible perturbations, conserved quantities, and genericity questions differ.

Each can test a precisely formulated visibility claim, but the formulation must include its boundary data. One should additionally report whether curvature diverges at finite or infinite boundary time, which invariant diverges, whether a horizon shields an interior singularity, and whether the result persists when matter allowed by the theory is included. Only after those choices match can the outcomes be compared as evidence about the same conjecture.

References

• Crisford, Toby, and Jorge E. Santos. “Violating Weak Cosmic Censorship in AdS$_4$.” Physical Review Letters 118 (2017): 181101. doi:10.1103/PhysRevLett.118.181101. Open PDF.
• Crisford, Toby, Gary T. Horowitz, and Jorge E. Santos. “Testing the Weak Gravity–Cosmic Censorship Connection.” Physical Review D 97 (2018): 066005. doi:10.1103/PhysRevD.97.066005. Open PDF.
• Garfinkle, David. “Gravitational Collapse in Anti de Sitter Space.” Physical Review D 70 (2004): 104015. doi:10.1103/PhysRevD.70.104015. Open PDF.
• 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. doi:10.1088/0264-9381/21/12/012. Open PDF.