Boundary Conditions and AdS Evolution

Boundary conditions are part of the definition of an AdS evolution problem. Because conformal infinity is timelike, a field can exchange data and energy with it in finite global time. The admissible conditions depend on the field, mass, dimension, symplectic form, energy norm, and desired boundary dynamics; “AdS uses reflecting Dirichlet conditions” describes one important phase space, not a universal law.

Helpful background. Lorentzian Metrics and Causal Cones supplies timelike boundaries; Geodesics and Jacobi Fields supplies propagation; Causality and Global Structure supplies predictability; Exact Solutions of Einstein’s Equations supplies AdS patches; Formulations of General Relativity supplies hyperbolic evolution; Gravitational Actions and Boundary Terms supplies a differentiable variational principle.

Two asymptotic scalar modes

Use Poincaré-like coordinates near the boundary,

$$d\widetilde s^2 =\frac{\ell^2}{z^2} \left(dz^2+g_{(0)ij}dx^i dx^j+\cdots\right), \qquad z\to0.$$

For a scalar obeying $(\widetilde\Box-m^2)\phi=0$ in asymptotically AdS$_{d+1}$, the leading radial indicial equation gives

$$\ell^2\widetilde\Box\phi =z^2\partial_z^2\phi-(d-1)z\partial_z\phi +z^2\Box_{(0)}\phi+\cdots.$$

Substituting $\phi=z^\Delta a(x)$ makes tangential derivatives subleading and yields $\Delta(\Delta-d)=m^2\ell^2$. The two roots give

$$\phi =z^{\Delta_-}\alpha(x)+z^{\Delta_+}\beta(x)+\cdots, \qquad \Delta_\pm=\frac d2\pm\nu, \qquad \nu=\sqrt{\frac{d^2}{4}+m^2\ell^2}.$$

Real exponents require the Breitenlohner–Freedman bound $m^2\ell^2\geq-d^2/4$. At special integer separations, and at the degenerate bound $\nu=0$, logarithms modify the displayed form. Backreaction and interactions also add nonlinear terms.

The two coefficients are not automatically both free. The finite symplectic norm, energy, and self-adjointness determine which mode belongs to the phase space. In the standard scalar problem, a range equivalent to $0\leq\nu\lt1$ permits more than one positive self-adjoint extension under suitable conditions; for larger $\nu$ the slower mode is normally excluded and the extension is essentially fixed. Details depend on the exact energy and field sector. Ishibashi and Wald classify scalar, electromagnetic, and gravitational cases in Ishibashi and Wald 2004, §2 and §§3.1–3.3, pp. 2985–3012.

For $\nu\gt 0$, the threshold can be checked from the static scalar inner product. Its near-boundary radial measure is proportional to $z^{1-d}dz$, so a mode $z^\Delta$ contributes $\int_0 z^{2\Delta+1-d}dz$. The slow mode is integrable precisely when $1-2\nu\gt-1$, or $\nu\lt1$; the fast mode is integrable throughout this range and beyond. At $\nu=1$ the slow-mode norm diverges logarithmically. At $\nu=0$ the two power solutions coalesce and the logarithmic branch must be analyzed separately. This is the usual bulk inner product; enlargements using renormalized boundary dynamics are different phase spaces.

Reflecting conditions

A closed evolution commonly requires vanishing symplectic flux through the timelike boundary and conservation of an appropriate total energy. These are related but distinct conditions. Restrict first to the standard two-mode window $0\lt\nu\lt1$, with smooth coefficients and no leading logarithmic branch. The finite boundary symplectic flux of two scalar variations is proportional to

$$\mathcal F_{\partial} \propto 2\nu\ell^{d-1}\int_{\partial M}d^dx\sqrt{-g_{(0)}}\, \left(\delta_1\alpha\,\delta_2\beta -\delta_2\alpha\,\delta_1\beta\right).$$

For $\nu\geq1$, arbitrary variation of the slow mode need not have finite unrenormalized flux: subleading slow-mode terms can contribute divergences. Logarithmic branches, including the BF endpoint $\nu=0$, also require a separately specified expansion and any necessary counterterms or corner terms. The following two-mode calculation is not an automatic extension to those cases.

It vanishes when one fixes $\alpha$ (Dirichlet-type), fixes $\beta$ where that mode is admissible (Neumann-type), or imposes an integrable mixed relation

$$\beta=\frac{\delta W[\alpha]}{\delta\alpha}$$

for an appropriate boundary functional $W$, with the functional derivative defined using the boundary volume measure. Symmetry of its second variation makes the integrated antisymmetric flux vanish, provided any boundary-of-boundary terms are also controlled. A linear Robin condition is the special case $\beta=\kappa\alpha$.

Fixing a source makes its allowed variation zero; it does not make its time derivative zero. A prescribed time-dependent Dirichlet source can therefore inject physical energy despite vanishing symplectic flux between allowed variations. For time-independent homogeneous data, or time-independent mixed conditions with the necessary boundary energy included, time-translation symmetry can give a conserved energy. With Robin conditions, energy may move between the bulk and that boundary contribution even though their sum is conserved.

“Reflecting” refers to zero net flux for the chosen closed system. It does not mean every wave returns with a universal phase or that all field components satisfy the same algebraic condition. Gauge fields and metric perturbations have constraints and gauge degeneracies; their admissible boundary data must be expressed in gauge-invariant or properly reduced variables.

The boundary condition must be preserved by time evolution and by the selected asymptotic symmetries. A condition that makes the linear operator self-adjoint but gives energy unbounded below can still be unstable. Positivity of the self-adjoint extension is a separate requirement in the Ishibashi–Wald construction.

Transparent and coupled boundaries

A transparent condition allows flux to leave the bulk instead of returning it. On global AdS, such a condition cannot describe a closed bulk theory with a self-adjoint Hamiltonian unless additional degrees of freedom receive the flux. One may couple the bulk field to a boundary system, glue the AdS region to another spacetime, or use an absorbing prescription in an effective calculation.

The combined bulk-plus-boundary system can be unitary and energy conserving even though bulk energy alone changes. Its phase space, action, and causal problem include the new boundary variables. Calling a nonlocal outgoing prescription “transparent” without naming that completion conceals where information and energy go.

Transparent conditions can be useful for numerical or scattering questions designed to suppress repeated reflections, but they change the problem from standard global AdS. Stability conclusions obtained with absorbing boundaries do not answer the closed reflective problem, and conversely.

Mixed and field-dependent conditions

Mixed conditions relate the two asymptotic modes rather than fixing one. If the relation derives from a boundary potential $W$, the variational principle can be made differentiable by adding the corresponding finite boundary term. Nonlinear $W$ produces field-dependent conditions and can preserve only a subgroup of the original boundary conformal symmetries.

Three tests are mandatory:

1. Flux: the symplectic flux vanishes or is balanced by explicit boundary degrees of freedom.
2. Well-posedness: the spatial evolution operator has an appropriate self-adjoint or hyperbolic domain, and nonlinear compatibility conditions hold at the corner where initial and boundary data meet.
3. Stability: the conserved energy is bounded below or another coercive estimate controls evolution.

A condition can pass the flux test and fail stability. Certain mixed scalar conditions admit negative-energy states or nonlinear runaway behavior even when the BF bound holds. The BF bound is a local spectral threshold, not a guarantee for every boundary functional.

For gravity, fixing the boundary conformal metric is a Dirichlet-type condition. Allowing it to vary can produce boundary gravitational dynamics and a different symplectic structure. Counterterms needed to render the action and charges finite must be consistent with the same choice; Gibbons–Hawking–York Boundary Term gives the basic variational logic before asymptotic renormalization.

Comparison by physical question

Boundary choice	Symplectic flux	Evolution picture	Main qualification
Dirichlet-type	zero between variations with fixed source	closed for suitable stationary data; driven for time-dependent sources	slower mode fixed; gauge/gravity data need constraints
Neumann-type	zero where both modes are admissible	alternate evolution domain	stationary sources and the correct energy are still needed for conservation
Mixed or Robin	zero if generated by an integrable relation	boundary contribution can exchange energy with bulk	positivity and symmetry can change
Transparent or absorbing	generally nonzero	open bulk or coupled boundary system	recipient degrees of freedom and nonlocality must be stated

No row is universally preferred. A holographic source problem, a classical stability theorem, a numerical absorbing boundary, and a conserved-energy spectral problem ask for different structures. Comparisons are meaningful only after holding the field and asymptotic phase space fixed.

Exercise

For scalar asymptotic data $(\alpha,\beta)$, show that the linear Robin condition $\beta=\kappa\alpha$ with fixed real $\kappa$ makes the bilinear symplectic flux above vanish. Does that prove stability?

Solution

Allowed variations obey $\delta\beta=\kappa\delta\alpha$. Therefore

$$\delta_1\alpha\,\delta_2\beta -\delta_2\alpha\,\delta_1\beta =\kappa \left(\delta_1\alpha\,\delta_2\alpha -\delta_2\alpha\,\delta_1\alpha\right)=0.$$

The condition is reflecting in this symplectic sense. Stability does not follow. One must still show that the associated self-adjoint extension has energy bounded below and that nonlinear evolution is controlled. Some values or nonlinear generalizations can admit negative-energy modes despite zero boundary flux.

Asymptotic Expansions near the AdS Boundary next derives which metric coefficients are fixed by the boundary representative and which remain state dependent.

References

• 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.
• Wald, Robert M., General Relativity, Appendix E, University of Chicago Press, 1984.