Reflecting, Transparent, and Mixed AdS Conditions

Anti-de Sitter infinity is timelike, so an exterior initial slice is not a Cauchy surface until a boundary domain is chosen. “Reflecting” is not one universal condition: the admissible falloffs, conserved energy, and positive self-adjoint extensions depend on the field, its mass, the spacetime dimension, and the symplectic form. Transparent conditions define a different, open system because flux is allowed to leave the bulk.

Helpful background. Black-Hole Geometry and Mechanics supplies AdS black-hole horizons; Partial Differential Equations and Green Functions supplies operator domains and self-adjoint extensions; Formulations of General Relativity supplies constrained field variables; Gravitational Actions and Boundary Terms supplies variational boundary terms; Causality and Global Structure explains why timelike infinity requires data; Black-Hole Perturbation Theory supplies the retarded and spectral framework.

Two asymptotic scalar branches

Use the inverse areal-radius defining coordinate $z=L/r$, not an exact Fefferman–Graham coordinate. The conformal boundary is $z=0$. For a scalar on AdS$_{d+1}$, $d\geq2$, let $m_{\rm eff}^2=m_0^2+\xi\mathcal R$, with $\mathcal R=-d(d+1)/L^2$. The real effective mass squared may be negative. Then

$$(\Box-m_{\rm eff}^2)\phi=0, \qquad \phi\sim \alpha(x)z^{\Delta_-}+\beta(x)z^{\Delta_+},$$

where

$$\Delta_\pm=\frac d2\pm\nu, \qquad \nu=\sqrt{\frac{d^2}{4}+m_{\rm eff}^2L^2}.$$

Reality of $\nu$ gives the Breitenlohner–Freedman bound $m_{\rm eff}^2L^2\geq-d^2/4$. At $\nu=0$ the two powers coalesce and the independent solution contains a logarithm. We first take $\nu\gt0$. Which falloffs are admissible is a question about a specified Hilbert space and energy, not just these exponents.

The exact global-AdS operator

For $f=1+r^2/L^2$, separate $\phi=R(t,r)Y_\ell$ with unit-norm spherical harmonic and eigenvalue $\ell(\ell+d-2)$. Set

$$\rho=\arctan(r/L),\quad \tau=t/L,\quad u=r^{(d-1)/2}R,\quad \sigma=\ell+\frac{d-2}{2}.$$

The radial wave equation becomes $\partial_\tau^2u=-Au$, where

$$A=-\frac{d^2}{d\rho^2} +\frac{\nu^2-\tfrac14}{\cos^2\rho} +\frac{\sigma^2-\tfrac14}{\sin^2\rho}, \qquad 0\lt\rho\lt\frac{\pi}{2}.$$

This follows by substituting $R=r^{-(d-1)/2}u$ in the scalar radial operator; the substitution removes its first derivative. Its static Hilbert measure is

$$\int_0^\infty |R|^2\frac{r^{d-1}}{f}\,dr =L\int_0^{\pi/2}|u|^2d\rho .$$

Fix smooth-center regularity first; arbitrary self-adjoint conditions at the origin are not extra physical choices. Near infinity, $u\sim(\cos\rho)^{1/2\pm\nu}$. The slow branch is square integrable precisely when $\nu\lt1$, whereas the fast branch is integrable for all $\nu\geq0$. At $\nu=0$ both the power and power-times-logarithm are integrable. Thus the boundary freedom in this Hilbert space occurs for $0\leq\nu\lt1$. It still remains to select a self-adjoint domain and check its positivity.

Ishibashi and Wald derive this reduction and classify the nonnegative extensions on exact AdS Ishibashi and Wald 2004, pp. 13–14, 25, 29–31, Theorem 3.2 and proof pp. 35–38, Open PDF. The general classification tests the center-regular negative-eigenvalue solution against the boundary condition; monotonicity of its Gamma-function coefficient ratio (or the logarithmic-limit digamma expression) identifies the parameters admitting a negative eigenvalue. We give a complete elementary version below. Maxwell and gravitational sectors have their own constraints and reductions, not an automatic identification with this scalar.

Symplectic flux defines the domain

For two linearized scalar solutions, the Klein–Gordon symplectic current is

$$j^a(\phi_1,\phi_2) =i\left(\phi_1^*\nabla^a\phi_2-\phi_2\nabla^a\phi_1^*\right).$$

For $0\lt\nu\lt1$, the outward radial flux density per unit time and unit harmonic pairing has the finite raw limit

$$\begin{aligned} \mathcal J_{12}(r) &=i r^{d-1}f \left(R_1^*\partial_rR_2-R_2\partial_rR_1^*\right),\\ \lim_{r\to\infty}\mathcal J_{12} &=-2i\nu L^{d-2} \left(\alpha_1^*\beta_2-\beta_1^*\alpha_2\right). \end{aligned}$$

The minus sign follows from outward increasing-$r$ orientation and $z=L/r$. The cross powers give $\Delta_--\Delta_+=-2\nu$; equal-power terms cancel. For arbitrary two-branch solutions, subleading slow-branch terms vanish in this limit only in the stated window. Integrating the density over time and angles gives the flux through a boundary segment. Renormalized prescriptions outside this window require their own boundary terms; the displayed expression is not a general counterterm theorem.

A closed, conservative bulk evolution requires this antisymmetric form to vanish for every pair of fields in the domain. The common linear choices are

$$\alpha=0\quad\text{(fast-falloff or Dirichlet type)},$$ $$\beta=0\quad\text{(slow-falloff or Neumann type, when admissible)},$$

and

$$\beta=\kappa\alpha \quad\text{(Robin or mixed)},$$

with real, fixed $\kappa$ for a time-translation-invariant local condition. More generally, $\kappa$ may be a self-adjoint boundary operator. Vanishing symplectic flux is necessary for a symmetric realization; positivity of the associated energy is an additional restriction. Some Robin parameters produce a negative-energy bound state even though the formal flux vanishes.

Two different positive domains

Take a conformally coupled scalar on AdS$_4$ with $m_0=0$, $\xi=1/6$, so $m_{\rm eff}^2L^2=-2$. In the $\ell=0$ sector, $\nu=\sigma=1/2$, and the entire potential vanishes:

$$A=-\partial_\rho^2,\qquad 0\lt\rho\lt b,\qquad b=\frac{\pi}{2},\qquad u(0)=0.$$

The origin condition excludes the physical field’s singular $1/r$ branch. Smooth compatible data give smooth-center solutions. Consider two domains in $L^2(0,b)$:

$$\begin{aligned} D_D&=\{u\in H^2(0,b):u(0)=u(b)=0\},\\ D_N&=\{u\in H^2(0,b):u(0)=u'(b)=0\}. \end{aligned}$$

Integration by parts leaves the boundary form $[\bar u v'-\bar u'v]_0^b$, which vanishes on either domain. These separated real endpoint conditions give self-adjoint interval operators. Their complete sine bases show positivity directly:

$$\begin{aligned} e_n^D&=\frac2{\sqrt\pi}\sin[(2n+2)\rho], &\lambda_n^D&=(2n+2)^2,\\ e_n^N&=\frac2{\sqrt\pi}\sin[(2n+1)\rho], &\lambda_n^N&=(2n+1)^2,\qquad n=0,1,\ldots . \end{aligned}$$

Since $u=rR=L\alpha+L\beta(b-\rho)+\cdots$ at infinity, $u(b)=L\alpha$ and $u'(b)=-L\beta$. Thus $D_D$ selects the fast branch and $D_N$ the slow branch. They are two distinct strictly positive domains for the same field and geometry. Up to constant amplitudes, their lowest physical radial profiles are

$$R_D=\frac1{1+r^2/L^2},\quad \omega_D=\frac2L; \qquad R_N=\frac1{\sqrt{1+r^2/L^2}},\quad \omega_N=\frac1L.$$

Both satisfy the complete scalar radial equation. The changed frequency is not an error: boundary data are part of the problem.

Zero flux does not imply positive energy

Replace the outer condition by $u'(b)=h u(b)$ with real $h$, equivalently $\beta=-h\alpha$. The quadratic form and conserved real-field energy are

$$\begin{aligned} Q_h[u]&=\int_0^b|u'|^2d\rho-h|u(b)|^2,\\ E_h&=\frac12\int_0^b \left[(\partial_\tau u)^2+(\partial_\rho u)^2\right]d\rho -\frac h2u(b)^2 . \end{aligned}$$

The form domain has $u\in H^1(0,b)$ and $u(0)=0$. Cauchy–Schwarz gives $|u(b)|^2\leq b\int|u'|^2$, proving $Q_h\geq0$ for $0\leq h\leq1/b$; negative $h$ adds a nonnegative boundary term. Conversely the trial function $u=\rho$ gives $Q_h=b-hb^2\lt0$ when $h\gt1/b$. Hence

$$Q_h\geq0\ \text{for all form-domain data} \quad\Longleftrightarrow\quad h\leq\frac2\pi .$$

At equality $u=\rho$ is a zero mode and may evolve linearly in $\tau$. For the explicit choice $h=\coth b$, $u=\sinh\rho$ has eigenvalue $-1$. The solution $e^\tau\sinh\rho$ grows exponentially, despite zero boundary symplectic flux. Its conserved energy is zero, not negative. Initial data $(u,u_\tau)=(\sinh\rho,0)$ instead give negative energy and evolution $\cosh\tau\,\sinh\rho$. These examples demonstrate an indefinite energy and a negative eigenvalue without confusing them with the energy of each individual growing solution.

The energy above belongs to the reduced, boundary-completed scalar action (physical-time generator $E_h/L$), not the unmodified minimal-stress bulk integral. This matters for the slow branch. For example, writing $r_*=L\rho$ and $u_*=\partial_{r_*}u$, integration by parts uses

$$-\left(u_*-\frac{fu}{r}\right)^2+\frac{2f}{L^2}u^2 =-u_*^2+\partial_{r_*}\left(\frac{fu^2}{r}\right).$$

The endpoint term cannot be discarded when $u$ tends to a nonzero constant.

Evolution and the black-hole qualification

For a nonnegative self-adjoint $A_E$, finite-energy initial data $u_0\in D(A_E^{1/2})$, $u_1\in L^2$ evolve uniquely by spectral calculus:

$$u(\tau)=\cos(\tau\sqrt{A_E})u_0+f_\tau(A_E)u_1,\qquad f_\tau(\lambda)= \begin{cases} \sin(\tau\sqrt\lambda)/\sqrt\lambda,&\lambda\gt0,\\ \tau,&\lambda=0. \end{cases}$$

The zero-eigenvalue definition is important: this expression does not assume that $A_E^{-1/2}$ exists. Smooth compatible data remain smooth, and $\tfrac12(\|u_\tau\|^2+Q_E[u])$ is conserved. Uniqueness holds after the domain is fixed, not before it. This is the spectral evolution construction described by Ishibashi and Wald 2003, §1, p. 3, Eq. (4), Open PDF, with the zero-eigenvalue limit made explicit here.

For Schwarzschild–AdS or another static AdS black-hole exterior, the same local falloffs do not determine positivity of the full operator. The event horizon changes the potential and the retarded problem allows horizon loss. The exact-AdS extension theorem does not supply a black-hole stability classification Ishibashi and Wald 2004, concluding discussion, pp. 38–39, Open PDF.

Reflecting, transparent, and dissipative problems

A self-adjoint zero-flux domain is reflecting in the energetic sense. A pulse can change phase or mix boundary components, but its net bulk symplectic flux through infinity vanishes. The word does not imply a pointwise hard wall or select $\alpha=0$ automatically.

A transparent boundary condition instead permits outward flux. The bulk alone is then an open system: its generator is not self-adjoint in the conserved bulk inner product. Well-posedness can still be recovered by coupling boundary degrees of freedom, adjoining another spacetime, or specifying a causal dissipative boundary response. The enlarged system may be conservative even though the bulk subsystem is not. A frequency-dependent condition $\beta(\omega)=K(\omega)\alpha(\omega)$ must obey causal analyticity and passivity conditions if it represents such a response.

Mixed boundary conditions can mean either conservative real Robin data or a dissipative response; the distinction must be stated. For quasinormal modes of an AdS black hole, future-horizon regularity is combined with precisely one chosen boundary domain. Changing that domain changes the poles. An “AdS QNM” quoted without field and falloff data is incomplete.

Scope, counterexample, and limiting cases

The explicit conformal example proves nonuniqueness of positive domains. Outside the two-normalizable-falloff window one must not extrapolate its raw flux formula. For a massless $\ell=0$ scalar on AdS$_4$, $\nu=3/2$. Two exact radial solutions are

$$R_1=1,\quad \omega_1=0;\qquad R_2=1+\frac{2L^2}{r^2+L^2},\quad \omega_2=\frac2L.$$

At $t=0$ their paired raw flux is $\mathcal J_{12}=-4ir^3/(r^2+L^2)$, which diverges as $-4ir$. Both have leading slow coefficient $\alpha=1$ and fast coefficient $\beta=0$; the divergent contribution comes from a slow-branch descendant. Neither belongs to the ordinary static Hilbert space. This is a counterexample to an unrestricted raw-flux formula, not to the theorem for admissible domains. Adding boundary counterterms would define a different prescription that must be analyzed separately.

The Poincaré patch, global AdS, and a black-hole exterior can share local falloffs while differing globally in spectrum and horizon conditions. Nonlinear boundary conditions may also exchange energy among bulk modes even when total flux vanishes. Those questions belong to a specified nonlinear theory; the present result concerns linear evolution in a declared domain.

Common pitfalls

Calling every AdS boundary reflecting. Timelike infinity supplies no condition by itself. State the two falloffs, the symplectic flux, and the selected domain.

Confusing symmetry with positivity. A real Robin parameter can kill the antisymmetric boundary form while still allowing a negative eigenvalue. Check the quadratic energy as well as the flux.

Exercises

Test a Robin domain

Let the boundary data of every field satisfy $\beta=\kappa\alpha$. Insert two such fields into the symplectic flux and determine the condition on $\kappa$ for the flux to vanish. Why does this calculation not prove stability?

Solution

Substitution gives

$$\alpha_1^*\beta_2-\beta_1^*\alpha_2 =\alpha_1^*\kappa\alpha_2-(\kappa\alpha_1)^*\alpha_2 =(\kappa-\kappa^*)\alpha_1^*\alpha_2.$$

It vanishes for arbitrary boundary data when $\kappa$ is real, or more generally when the boundary operator is self-adjoint in the boundary inner product. This establishes the boundary symmetry condition. It does not show that the spatial operator is positive: an allowed real Robin parameter may create a negative eigenvalue and an exponentially growing solution. Positivity or a lower energy bound must be checked independently.

References

• Ishibashi, A., and Wald, R. M. (2003). “Dynamics in non-globally-hyperbolic static spacetimes: II. General analysis of prescriptions for dynamics.” Classical and Quantum Gravity 20, 3815–3826. DOI. Open PDF.
• Ishibashi, A., and Wald, R. M. (2004). “Dynamics in non-globally-hyperbolic static spacetimes. III. Anti-de Sitter spacetime.” Classical and Quantum Gravity 21, 2981–3014. DOI. Open PDF.