Fluctuating and Smeared Horizons

When metric fluctuations are non-negligible, a coordinate equation such as $r=2GM$ does not become a literal randomly moving membrane. The useful quantities are operational, gauge-invariant observables: for example a smeared arrival time, redshift, expansion, or escape event defined by emitters and detectors. Their fluctuations depend on the quantum state, retarded response, detector resolution, and spacetime smearing. A full probability distribution requires more information than a two-point function unless a controlled Gaussian approximation applies.

Required background. Quantum Information supplies variances, reduced states, and measurement distributions.

Helpful background. Quantum Fields in Curved Spacetime supplies stress correlations; Einstein Equations supplies metric response; Black-Hole Geometry and Mechanics supplies horizon definitions; Semiclassical Gravity supplies stochastic gravity; and Causality and Global Structure supplies operational causal boundaries.

From a mean metric to a stochastic response

Write the metric as $g_{ab}=\bar g_{ab}+h_{ab}$. The mean $\bar g_{ab}$ solves the semiclassical equation. Stress fluctuations are encoded by the noise kernel

$$N_{ab\,cd}(x,y) =\frac12 \left\langle \left\{\hat t_{ab}(x),\hat t_{cd}(y)\right\} \right\rangle, \qquad \hat t_{ab}=\hat T_{ab}-\langle\hat T_{ab}\rangle.$$

At leading large-$N$ or stochastic order, an Einstein–Langevin equation has the schematic form

$$\mathcal L^{ab}{}_{cd}h^{cd}=8\pi G\,\xi^{ab}, \qquad \langle\xi_{ab}(x)\xi_{cd}(y)\rangle_{\rm st}=N_{ab\,cd}(x,y),$$

with retarded boundary conditions. Thus

$$\langle h h\rangle_{\rm ind} =(8\pi G)^2\,G_{\rm ret}NG_{\rm ret}^{T}.$$

This is the induced covariance. The full answer also contains intrinsic fluctuations propagated from the initial metric state and, if initial metric and matter fluctuations are correlated, cross terms. The formula is meaningful only after gauge fixing or projection onto a gauge-invariant observable and after smearing the distributional kernel. Hu and Verdaguer derive the noise and response structure in Hu and Verdaguer 2008, §§ 3.2–3.3 and 4.1–4.3, PDF. A Gaussian stochastic source reproduces the retained symmetrized two-point information; it does not establish every higher moment or an exact quantum distribution.

Operational horizon observables

Choose an emitter worldline, detector worldline, wavepacket, and clock protocol. A null signal has perturbed world function $\sigma=\sigma_0+\sigma_1+\cdots$. On the background ray $\sigma_0=0$; the variance of the linear shift, smeared over the source and detector profiles, produces an arrival-time width schematically

$$(\Delta t)^2 \propto \int d\lambda\,d\lambda'\, k^ak^bk^ck^d \langle h_{ab}(x(\lambda))h_{cd}(x(\lambda'))\rangle_f.$$

Ford develops this lightcone-fluctuation observable for specified graviton states and backgrounds Ford 1995, §§ II–IV, PDF. It answers a measurable question: how broadly distributed are detection times relative to the mean light cone?

Near a mean horizon, repeat the protocol over emission events and define $P_{\rm esc}(x)$, the probability that a wavepacket reaches the detector within a declared observation time. Contours such as $P_{\rm esc}=1/2$ give a central operational boundary. If the transition is a Gaussian cumulative distribution in a specified relational location parameter, the $0.16$ and $0.84$ contours approximately bracket its central $68\%$ interval: their separation is $2\sigma$, and the half-width is one standard deviation. A general non-Gaussian transition has no such identification. These finite-time boundaries are protocol dependent and do not determine a global event horizon, whose definition depends on the entire future.

Gauge and smearing checks

Under $h_{ab}\mapsto h_{ab}+2\nabla_{(a}\zeta_{b)}$, the coordinate position of a surface $F[g,x]=0$ changes. A complete round-trip proper time, frequency ratio between specified tetrads, or causal success event with transformed endpoints is gauge invariant. Computing only $\delta r_H=-\delta F/F'$ without relational anchors can therefore manufacture a horizon fluctuation from a coordinate transformation.

The noise kernel is a distribution and generally cannot be restricted sharply to a null hypersurface. Smear in time, transverse angle, and—when required—a timelike or spacelike thickness set by the detector. Hu and Verdaguer’s horizon discussion emphasizes that smearing only along the horizon can leave divergent induced fluctuations and that an operational probe must resolve a finite neighborhood Hu and Verdaguer 2008, § 8.4, PDF.

The smearing scale $\ell_f$ must satisfy two opposing inequalities:

$$\ell_{\rm UV}\ll\ell_f\ll L_{\rm background}.$$

The first keeps the observable below the EFT cutoff and suppresses distributional coincidence singularities; the second preserves horizon-scale resolution. If no window exists, semiclassical theory cannot assign the requested sharpness.

Results and open questions through 2026-08-19

Evidence through 2026-08-19 supports stochastic metric correlations and specific lightcone-broadening observables. Whether every quantum horizon admits a state-independent universal width remains unresolved and conjectural; no universal verdict follows.

Evidence and exact locator	Result under stated assumptions	Limitation or competing evidence
Anderson, Molina-París, and Mottola 2003, §§ III–V, PDF	Retarded gauge-invariant response supplies a boundedness test for the mean solution	Linear response alone does not determine intrinsic or induced fluctuation probabilities
Hu and Verdaguer 2008, §§ 3.2–4.3, PDF	Noise kernels and retarded Green functions determine induced metric two-point functions at stochastic order	Requires a state, renormalization, gauge-invariant observable, smearing, and mean-field hierarchy
Ford 1995, §§ II–IV, PDF	Geodesic-interval fluctuations yield arrival-time broadening in explicit graviton-state examples	Background and state specific; does not define a universal black-hole horizon location
Hu and Verdaguer 2008, § 8.4, PDF	Near-horizon analyses show sensitivity to regularity, flux correlations, and smearing dimension	Conflicting estimates can reflect singular states or insufficient smearing rather than physical macroscopic fuzziness

The conservative position retains a sharp horizon of the mean metric and assigns fluctuations only to declared observables. The competing geometric position summarizes a family of outcomes by a stochastic “horizon tube.” The second is acceptable shorthand only when the observer, horizon type, gauge-invariant locator, state, smearing kernel, and confidence level are specified.

Validity and open questions

Check the flat-vacuum limit against the chosen reference; its stress fluctuations need not vanish when its renormalized mean stress does. Check positivity of the full variance $\langle(Q_f-\langle Q_f\rangle)^2\rangle$ for every Hermitian smeared observable $Q_f$. A difference of variances obtained by subtracting a reference state need not be positive and cannot itself be interpreted as an absolute probability width. Refine the smearing profile without taking an inadmissible point or null limit, and compare induced fluctuations with intrinsic graviton fluctuations. A width comparable to the curvature radius invalidates linearization; a width at the cutoff is not resolved by the EFT.

The conclusions could change after a gauge-invariant detector calculation changes a horizon-width estimate, an independently reproduced noise-kernel computation resolves a smearing divergence, a new state-regularity theorem changes the admissible horizon class, or nonlinear stochastic evolution changes the probability distribution; an observational timing bound or a corrected intrinsic-fluctuation calculation could also change the conclusion.

Exercise

An operational arrival time is Gaussian with mean $10.0$ and standard deviation $0.2$. What is the probability of arrival before $9.8$? What interval contains the central $68\%$ of outcomes, and why is neither endpoint a universal horizon?

Solution

The standardized value is $(9.8-10.0)/0.2=-1$, so the Gaussian cumulative probability is approximately $0.159$. The central $68\%$ interval is approximately $[9.8,10.2]$. Its endpoints depend on the state, emitter, detector, wavepacket, observation time, and smearing prescription. They define this experiment’s causal transition region, not an observer-independent geometric surface.

References

• Anderson, P. R., Molina-París, C., and Mottola, E. (2003). “Linear Response, Validity of Semiclassical Gravity, and the Stability of Flat Space.” Physical Review D 67, 024026. DOI. Open PDF.
• Ford, L. H. (1995). “Gravitons and Light Cone Fluctuations.” Physical Review D 51, 1692–1700. DOI. Open PDF.
• Hu, B. L., and Verdaguer, E. (2008). “Stochastic Gravity: Theory and Applications.” Living Reviews in Relativity 11, 3. DOI. Open PDF.