Corner Compatibility and Anti-de Sitter Boundaries

Anti-de Sitter asymptotics turn conformal infinity into a timelike boundary. Initial data alone therefore do not determine the future: one must choose boundary data at infinity and make them compatible with the initial data where the two meet. Local existence, unique continuation, and dynamical instability answer different questions and cannot be combined into a single verdict about “the AdS boundary problem.”

This comparison includes theorem versions available through 2026-08-12. A later finite-regularity Einstein IBVP theorem, counterexample to a corner condition, or sharper unique-continuation threshold could change which data classes the results cover.

Required background. Analysis for Mathematical Relativity supplies boundary regularity and compatibility; The Einstein Equation supplies the conformal equations; Hypersurface Normals and Tangential Projectors supplies corner geometry.

Helpful background. Causality and Global Structure distinguishes null and timelike conformal infinity, while PDEs and Green Functions for Gravitation supplies boundary estimates.

Corner jets are part of the data

Let $\Sigma$ be an initial spacelike surface and $\mathcal I$ a timelike boundary, meeting at $S$. Write an evolution schematically as

$$\partial_tU=F(U,\partial_iU), \qquad B(U,\partial U)\big|_{\mathcal I}=q.$$

At $S$, the initial value $U_0$ must satisfy $B(U_0,\partial U_0)=q(0)$. Differentiate the boundary equation in time and use the evolution equation to eliminate $\partial_tU$; this gives the first compatibility condition. Repeating produces a hierarchy relating the normal and tangential jets of $U_0$ and $q$. A $C^k$ theorem needs the corresponding finite hierarchy, while a smooth theorem assumes compatibility to every order.

This is not a technical afterthought. If the corner conditions fail, a solution may lose differentiability at $S$ even though the initial and boundary data are separately smooth. In the conformally compactified Einstein equations, the constraints and gauge conditions also enter this hierarchy.

Why AdS requires a boundary choice

For a physical asymptotically AdS metric $g$, choose $\widetilde g=\Omega^2g$ with $\Omega=0$ and $d\Omega\ne0$ at $\mathcal I$. Unlike asymptotically flat null infinity, $\mathcal I$ is timelike. Null curves can reach it and return in finite conformal time, so the bulk is not globally hyperbolic without a boundary condition.

Pure AdS makes this causal distinction explicit. In four dimensions, let $\Lambda=-3/a^2$, use the universal cover with $\tau\in\mathbb R$, and put $r=a\tan\chi$. The metric is

$$g=\frac{a^2}{\cos^2\chi} \left(-d\tau^2+d\chi^2+\sin^2\chi\,d\omega^2\right), \qquad 0\leq\chi\lt\frac{\pi}{2}.$$

Multiplication by $\Omega^2=\cos^2\chi$ gives a regular metric up to $\chi=\pi/2$, where the induced metric is $a^2(-d\tau^2+d\omega^2)$, explicitly Lorentzian. A radial null ray has $d\tau=\pm d\chi$: it goes from the center to the boundary in $\Delta\tau=\pi/2$ and, under reflection, back in total time $\pi$. This is finite global time; it is not finite physical affine parameter to infinity. The latter diverges along a complete AdS null geodesic.

Prescribing the conformal class of the Lorentzian metric induced on $\mathcal I$ is one geometric choice. Other matter or perturbation problems admit reflecting, dissipative, or mixed conditions. These define genuinely different dynamics: reflecting conditions return flux to the bulk, while dissipative conditions permit flux through the boundary. A stability or instability result for one choice does not settle another.

Results compared on common axes

Result	Domain and data	Boundary/corner conditions	Conclusion and status
Friedrich–Nagy vacuum IBVP	Four-dimensional vacuum; compact spatial slice with a finite timelike boundary; smooth data	Gauge-source and curvature boundary data; smooth corner compatibility	Local smooth well-posedness in the specified frame reduction; peer-reviewed theorem, not an AdS-infinity theorem.
Enciso–Kamran asymptotically AdS construction	$(n+1)$-dimensional vacuum Lorentzian Einstein metrics; asymptotically hyperbolic constrained initial data with the weighted $C^{n-1}$ and polyhomogeneous regularity of Theorem 1.1	A $C^p$ conformal boundary metric compatible with the initial data through order $p$, with $p\ge 2q+r_n$	A finite-time solution, unique up to diffeomorphism, whose conformal compactification has the theorem’s $C^{n-1}$ and polyhomogeneous $C^q$ regularity; peer-reviewed theorem.
Holzegel–Shao bulk–boundary uniqueness	Vacuum Fefferman–Graham asymptotically AdS segments in dimension $n+1$ with $n>2$, regular to a sufficiently high order	Gauge-equivalent pairs of coefficients $(g^{(0)},g^{(n)})$ on an open boundary domain with compact closure satisfying the generalized null-convexity condition	The two metrics are isometric near that boundary domain; peer-reviewed unique-continuation theorem, not forward existence or stability.
Moschidis null-dust instability	$3+1$ spherical Einstein–radial-massless-Vlasov, equivalently ingoing and outgoing null dust, on an exterior domain	Reflecting conformal infinity and a timelike inner mirror of radius $r_{0\epsilon}>0$ that shrinks with the perturbation size	A family tending to AdS in the theorem’s norm whose developments contain trapped spheres and black-hole regions; peer-reviewed theorem for this reduced matter model, not vacuum gravity.
Einstein–scalar numerical mechanism	$3+1$ spherical Einstein–massless-scalar evolution	Reflecting AdS boundary conditions and selected smooth one-parameter data families	Numerical evidence for weakly turbulent transfer and black-hole formation, with stable islands and model dependence; not a theorem for unrestricted vacuum gravity.

The existence result and its finite compatibility hierarchy are given in Enciso–Kamran 2019, Theorem 1.1 and Appendix A. The distinct inverse problem is treated in Holzegel–Shao 2023, Theorem 6.7: both $g^{(0)}$ and the undetermined coefficient $g^{(n)}$, modulo their stated gauge relation, are required. The finite-boundary comparison uses Friedrich–Nagy 1999, §§3–7 and Theorem 8.1.

What the competing statements do not imply

Smooth local existence does not imply global stability. It guarantees a solution only for a time controlled by the norms in the theorem. Conversely, a collapse or turbulent-cascade example does not contradict local existence: the solution may exist smoothly for a substantial interval before concentrating.

Unique continuation asks whether sufficiently rich boundary information determines the nearby bulk. It does not say that arbitrary boundary information is admissible initial-boundary data, nor that reconstructing the bulk is stable under noise. Its geometric hypotheses can fail for short boundary time intervals or unfavorable null geometry.

Numerical evidence for the Einstein–massless-scalar system with reflecting AdS boundary conditions, initiated by Bizoń–Rostworowski 2011, pp. 1–4, identifies a weakly turbulent transfer to high frequencies and black-hole formation for families of small spherical data. Islands of stability and model dependence prevent the slogan “AdS is unstable” from serving as a general theorem. Moschidis 2020, Theorem 1 is rigorous but changes the equations to spherical radial massless Vlasov, equivalently null dust, and introduces a timelike inner mirror whose radius shrinks with the data. Neither qualification can be dropped when comparing it with the vacuum conjecture.

The unrestricted nonlinear vacuum stability problem remains open and depends on the boundary condition, data class, topology, and regularity. This unresolved status is compatible with every local theorem in the comparison table.

Solved diagnostic: a first corner mismatch

For $u_{tt}-u_{xx}=0$ on $x\geq0$, with initial data $(f,g)$ and Dirichlet boundary value $q(t)$, derive the compatibility conditions at $(t,x)=(0,0)$ for a $C^2$ solution.

Solution

For $u_{tt}-u_{xx}=0$ on $x\ge0$, prescribe $u(0,x)=f(x)$, $u_t(0,x)=g(x)$ and Dirichlet boundary data $u(t,0)=q(t)$. Continuity requires

$$f(0)=q(0).$$

Differentiating the boundary condition gives $u_t(t,0)=q'(t)$, so a $C^1$ solution requires $g(0)=q'(0)$. Differentiating twice and using the wave equation gives

$$f''(0)=u_{xx}(0,0)=u_{tt}(0,0)=q''(0).$$

Thus smoothness of $f$, $g$, and $q$ separately does not ensure even $C^2$ compatibility. Einstein’s corner hierarchy is more coupled because the constraints and gauge equations determine parts of these jets.

The chapter began with free data on intersecting null hypersurfaces and ends with the opposite lesson at a timelike conformal boundary: causality tells us exactly where additional data are needed, while the field equations determine which additional data are admissible.

References

• Bizoń, Piotr, and Andrzej Rostworowski. “On Weakly Turbulent Instability of Anti-de Sitter Space.” Physical Review Letters 107 (2011): 031102. DOI.
• Enciso, Alberto, and Niky Kamran. “Lorentzian Einstein Metrics with Prescribed Conformal Infinity.” Journal of Differential Geometry 112 (2019): 505–554. DOI. Open PDF.
• Friedrich, Helmut, and Gabriel Nagy. “The Initial Boundary Value Problem for Einstein’s Vacuum Field Equation.” Communications in Mathematical Physics 201 (1999): 619–655. Open PDF.
• Holzegel, Gustav, and Arick Shao. “The Bulk-Boundary Correspondence for the Einstein Equations in Asymptotically Anti-de Sitter Spacetimes.” Archive for Rational Mechanics and Analysis 247 (2023): article 56. DOI. Open PDF.
• Moschidis, Georgios. “A Proof of the Instability of AdS for the Einstein–Null Dust System with an Inner Mirror.” Analysis & PDE 13 (2020): 1671–1754. DOI. Open PDF.