Stretched Horizon Boundary Condition

• Stretched horizon boundary condition is a regulator that replaces a degenerate null horizon with a finite timelike hypersurface, thereby establishing well-posed variational formulations.
• It is implemented in various forms—Schwarzian, Carrollian, and conformal—each adapting the gravitational boundary data to specific dynamical and thermodynamic requirements.
• The approach underpins analyses in quantum extremal surfaces, island constructions, and membrane paradigms, influencing black hole microstate localization and near-horizon scattering studies.

The stretched horizon boundary condition is a family of boundary prescriptions imposed on a timelike hypersurface placed a finite proper distance from a null horizon, or on a conformal analogue of such a surface, in order to regulate near-horizon dynamics, define a well-posed variational principle, or encode effective horizon degrees of freedom. In the literature, it appears in several distinct but related forms: as a free or mixed gravitational boundary condition generating a Schwarzian boundary action, as a Carrollian or membrane-paradigm boundary condition with a finite stress tensor and hydrodynamic equations, as a reflecting or partially reflecting wall for matter perturbations, and as a critical surface in quantum extremal surface and island constructions where information becomes effectively localized on the boundary itself (Carlip, 2022, Freidel et al., 2022, Matsuo, 2020).

1. Geometric setting and meaning of the stretched horizon

Near a nonextremal horizon, the relevant two-dimensional geometry is Rindler-like. In Carlip’s Euclidean treatment of a stationary nonextremal black hole, the near-horizon metric in the radial-time plane becomes

$ds^2 = d\rho^2 + \rho^2 dT^2,$

with the horizon at $\rho=0$ and $T$ periodic (Carlip, 2022). In asymptotically flat Schwarzschild coordinates, the Lorentzian near-horizon region can be written in terms of the proper radial coordinate

$x \simeq 2\sqrt{r_h(r-r_h)},$

so that

$$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$

and a stretched horizon may be placed at

$r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$

namely one proper Planck length outside the classical horizon (Terashima, 25 Jun 2025).

A complementary geometric formulation uses a timelike hypersurface $H$ at finite distance from a null boundary $N$, together with a rigged structure $(k,\mathbf n)$ satisfying $\iota_k\mathbf n=1$ and a null rigging choice $\rho=0$0. In that formulation, the stretched horizon is a regular timelike representative of the null boundary, and the null limit is taken by letting the normal norm $\rho=0$1 while keeping the rigged metric and connection regular (Freidel et al., 2022).

These constructions share a common purpose. The true horizon is null and therefore degenerate from the standpoint of ordinary Hamiltonian, Brown–York, or boundary-value formulations. Replacing it by a nearby timelike surface produces nondegenerate boundary data, finite redshifted observables, and a controllable place on which to impose either geometric or matter boundary conditions. This suggests that “stretched horizon boundary condition” is not a single universal condition but a class of prescriptions adapted to different questions: thermodynamics, variational principles, effective hydrodynamics, entanglement, or scattering.

2. Variational principles and boundary data

One important realization arises in Euclidean near-horizon dilaton gravity. After dimensional reduction, the boundary term for fixed induced metric takes the form

$\rho=0$2

Carlip then considers a free-boundary problem in which $\rho=0$3 and $\rho=0$4 are fixed, while the intrinsic and extrinsic geometry of the stretched horizon curve $\rho=0$5 are left dynamical. To make the variational principle well-posed, one adds

$\rho=0$6

To quadratic order in the near-horizon expansion, the resulting boundary dynamics is governed by a Schwarzian action,

$\rho=0$7

with $\rho=0$8 (Carlip, 2022). The stretched horizon boundary condition here is therefore a mixed gravitational condition that promotes horizon reparametrizations to physical edge modes.

A different implementation is given by the rigged-Carrollian formalism. There the natural Dirichlet data on the stretched horizon are the Carrollian geometric fields

$\rho=0$9

supplemented by the gauge condition

$T$0

The boundary functional

$T$1

renders the Einstein–Hilbert variational principle well-posed in this polarization, and the pre-symplectic structure identifies $T$2 as the canonical boundary data (Freidel et al., 2022). In this formulation, the boundary condition is neither purely Dirichlet nor Brown–York in the conventional sense; it is a finite-distance Carrollian boundary condition regular in the null limit.

A third form fixes the conformal class of the induced metric and the trace $T$3 of the extrinsic curvature on a timelike boundary $T$4 near a cosmological horizon. The resulting conformal Brown–York stress tensor is

$T$5

and the boundary momentum constraint implies covariant conservation when $T$6 is constant (Anninos et al., 18 Dec 2025). This is again a mixed condition: not fixed metric, not fixed full extrinsic curvature, but fixed conformal class and fixed mean curvature.

Taken together, these results show that the phrase “boundary condition” is literal but context-dependent. The boundary may carry Schwarzian soft modes, Carrollian canonical data, or a traceless conformal Brown–York tensor, depending on which degrees of freedom are treated as sources and which are treated as responses.

3. Quantum extremal surfaces, islands, and information localization

In entanglement-based formulations, the stretched horizon enters through the generalized entropy

$T$7

with quantum extremality condition

$T$8

For the entropy of the region outside a surface $T$9 in eternal Schwarzschild, near-horizon variables

$x \simeq 2\sqrt{r_h(r-r_h)},$0

lead in the short-distance regime to the extremality equation

$x \simeq 2\sqrt{r_h(r-r_h)},$1

The stable and unstable saddles merge at

$x \simeq 2\sqrt{r_h(r-r_h)},$2

after which there is no separate interior island endpoint. At $x \simeq 2\sqrt{r_h(r-r_h)},$3, the entropy of the outside region drops to $x \simeq 2\sqrt{r_h(r-r_h)},$4, and the analysis interprets the critical surface $x \simeq 2\sqrt{r_h(r-r_h)},$5 as a stretched horizon on which black-hole information is localized (Matsuo, 2020).

That paper further proposes an entropy-level boundary condition: the dominant replica saddle is anchored at the critical surface, the branch cut merges with the boundary, and the effective boundary dynamics may be summarized by

$x \simeq 2\sqrt{r_h(r-r_h)},$6

together with the matching relation

$x \simeq 2\sqrt{r_h(r-r_h)},$7

This does not derive explicit Neumann or Dirichlet conditions for all bulk fields; rather, it motivates a boundary interpretation in which the stretched horizon carries the microstate data (Matsuo, 2020).

Fuzzball-inspired reflecting-wall models complicate this picture. In a two-dimensional BCFT model with a perfectly reflecting boundary at $x \simeq 2\sqrt{r_h(r-r_h)},$8, the matter boundary condition is vanishing energy flux,

$x \simeq 2\sqrt{r_h(r-r_h)},$9

The boundary modifies the replica computation through image contributions to the entropy, and the resulting island saddle can appear, disappear, and reappear in time, producing the “blinking island” phenomenon. In higher-dimensional estimates, Dirichlet-like boundary conditions tend to suppress stable islands by turning candidate stationary points into local maxima, whereas Neumann-like conditions can support local minima near the wall (Ageev et al., 8 May 2026). A common misconception is therefore that the presence of a stretched horizon automatically reproduces the standard island mechanism. The existing results indicate instead that island existence is highly sensitive to the boundary condition and to the position of the stretched horizon.

4. Carrollian membrane dynamics and horizon fluids

The membrane-paradigm version of the stretched horizon is formulated in terms of a rigged stress tensor

$$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$0

constructed from the Weingarten tensor along the normal. In the gauge $$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$1, the projected Einstein equations become

$$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$2

so in vacuum the stretched-horizon stress tensor is exactly conserved (Freidel et al., 2022). The same formalism identifies Carrollian fluid variables on the stretched horizon, including

$$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$3

and gives a smooth $$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$4 limit to a null horizon without the singularities of the conventional Brown–York description.

This regularization is central to the modern Carrollian reading of the membrane paradigm. The stretched horizon is not merely a regulator; it is a finite-distance hypersurface on which gravitational data reorganize into energy density, momentum density, pressure, viscous stress, and heat current. The null limit then becomes a Carrollian limit rather than a singular degeneration (Freidel et al., 2022).

A related but distinct realization appears for a timelike boundary near a cosmological horizon in four-dimensional de Sitter space. There, linearized perturbations in the cosmic patch exhibit shear and sound modes whose dispersion relations match those of a relativistic conformal fluid, with

$$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$5

The same setup admits a nonlinear treatment, and in the large-$$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$6 limit the stretched-horizon dynamics reduces to a universal Rindler regime (Anninos et al., 18 Dec 2025). This refines the standard membrane picture by replacing the usual viscous membrane with a traceless conformal fluid tied to fixed conformal class and fixed mean curvature.

The notion extends even further in asymptotically flat gravity. In even dimensions, conformal null infinity can be treated as a stretched horizon in the conformally compactified spacetime. In that setting, the boundary carries a Carrollian stress tensor, the radial derivative of the asymptotic stress tensor defines the asymptotic Weyl tensor, and the renormalized symplectic flux becomes finite once Penrose boundary conditions and a further order of the equations of motion are imposed (Freidel et al., 2024). A plausible implication is that the stretched-horizon framework is not restricted to black-hole horizons; it also organizes the geometry and flux algebra of null infinity.

5. Reflecting walls, normal modes, and partially reflecting horizons

In a different line of work, the stretched horizon is a literal wall for matter perturbations. For a free scalar field on BTZ, the wall is placed at

$$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$7

with $$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$8 fixed by a proper Planckian distance, and the boundary condition is Dirichlet,

$$ds^2 \simeq -\frac{x^2}{4r_h^2}dt^2 + dx^2 + r_h^2 d\Omega^2,$$9

This replaces the near-horizon continuum by a discrete set of standing waves. In the nonrotating BTZ case, the mode spacing is

$r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$0

and highly excited typical pure states built on this stretched-horizon vacuum reproduce the Hartle–Hawking thermal correlator in the small-$r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$1 limit, with variance corrections suppressed as $r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$2 (Burman et al., 2023). Closely related normal-mode analyses interpret the same Dirichlet wall as the bulk mechanism underlying black-hole microstate level spacing: the spectrum is linear in a radial quantum number but only logarithmic or weak in angular quantum numbers, producing the quasi-degeneracy responsible for area-law entropy rather than volume-law entropy (Krishnan et al., 2023).

For gravitational-wave echo phenomenology, the wall is generally taken to be partially reflecting rather than perfectly reflecting. The near-wall field is written as

$r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$3

and the boundary condition is encoded in

$r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$4

For low-frequency gravitational perturbations, the reflection probability takes the Boltzmann-like form

$r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$5

with $r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$6, while the dissipation is independent of the Planck mass at leading order (Terashima, 25 Jun 2025). In this language, the stretched horizon boundary condition is neither pure Dirichlet nor pure absorption; it is a lossy interface whose reflectivity is fixed by near-horizon scattering with blue-shifted Hawking radiation.

An operator-algebraic CFT realization reaches the same structure from the boundary side. Quantization of the vacuum modular Hamiltonian on an interval of $r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$7 requires excising disks of radius $r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$8 around the fixed points of the modular flow and imposing a conformal no-flux condition

$r_0=r_h+\frac{l_p^2}{4r_h},\qquad x_0=l_p,$9

The regulated Hilbert space then has discrete normalizable states, a finite Virasoro central extension factor, and thermal two-point functions that emerge in the $H$0 limit. In the bulk dual, the excised fixed points correspond to the horizons of AdS-Rindler or planar BTZ, and the regulator is interpreted as the stretched horizon (Das, 2024). This suggests that the reflecting-wall picture and the boundary modular-quantization picture are two realizations of the same regulated near-horizon discreteness.

6. Finite-cutoff holography, de Sitter horizons, and unresolved issues

Finite-cutoff holography provides several further realizations. In $H$1, a constant-$H$2 world tube just inside the true cosmological horizon supports a mixed boundary condition governed by the Hamilton–Jacobi equation of a $H$3-deformed CFT. The radial position is encoded by

$H$4

and the stretched horizon is reached at

$H$5

At this tuning, the deformed CFT is said to inhabit the stretched horizon, and the entropy is

$H$6

so the horizon area law acquires a calculable logarithmic correction (Shyam, 2021).

In two-dimensional dilaton-gravity models interpolating between $H$7 and near-$H$8, the stretched horizon is implemented as a finite radial Dirichlet wall. The canonical ensemble fixes the Tolman inverse temperature $H$9 and the boundary dilaton value $N$0, and the quasilocal thermodynamic variables are

$N$1

Depending on the potential and on wall position, the heat capacity can be positive or negative, and the models exhibit Hawking–Page-like first-order transitions in certain regimes (Anninos et al., 2022). By contrast, the $N$2 deformation of the double-scaled SYK model places the boundary on a stretched horizon in a positive-curvature region of a $N$3 dilaton-gravity dual, but in that model the stretched-horizon configuration is always thermodynamically unstable (Aguilar-Gutierrez, 2024).

These examples make the limitations of the concept explicit. A perfectly reflecting wall near the horizon may require an external support force density that diverges as the wall approaches $N$4, which suggests that such a wall is dynamically unnatural in the simplest fuzzball-inspired models (Ageev et al., 8 May 2026). In Carlip’s quantization of boundary reparametrizations on the stretched horizon, the one-loop Gaussian integral is not finite with the original sign choice; after analytic continuation it becomes finite but $N$5-independent, so the boundary degrees of freedom do not contribute to thermodynamics at one loop (Ali et al., 2022). This suggests that the existence of a stretched-horizon action does not by itself determine its thermodynamic role.

A recurring misconception is that the stretched horizon is a uniquely defined physical surface with a universal boundary theory. The literature instead supports a more limited statement: the stretched horizon is an effective timelike regulator whose boundary condition depends on the observable being computed. It may be a free gravitational boundary supporting a Schwarzian mode, a Carrollian hypersurface carrying a conserved stress tensor, a mixed finite-cutoff holographic boundary, a reflecting or partially reflecting wall for perturbations, or a critical QES surface on which information becomes localized. The unifying feature is not uniqueness of boundary data, but the replacement of a null degeneracy by finite-distance boundary dynamics.