Validity of Stochastic and Mean-Field Gravity

Mean-field gravity predicts a geometry sourced by $\langle T_{ab}\rangle$; stochastic gravity adds the connected stress two-point function and its induced metric covariance. Neither framework comes with a single state-independent breakdown number. A defensible validity claim must name the observable, smearing scale, state, gauge-invariant response, cutoff, and approximation order.

Evidence cutoff. This frontier assessment includes evidence available through 2026-08-19. It distinguishes established leading-order results from proposed diagnostics and extrapolations.

Required background. Partial Differential Equations and Green Functions supplies retarded stability and distributional smearing.

Helpful background. Quantum Fields in Curved Spacetime supplies stress correlators and states, Einstein Equations supplies gauge and constraint structure, Black-Hole Geometry and Mechanics supplies horizon applications, and Semiclassical Gravity supplies the mean-field expansion being assessed.

What is being validated

There are at least three nested claims.

1. A background pair $(\bar g,\rho)$ solves the renormalized semiclassical equation within a stated curvature and frequency regime.
2. Gauge-invariant solutions of its homogeneous retarded linear-response equation remain bounded below the cutoff. This controls intrinsic metric perturbations.
3. The retarded propagation of the noise kernel produces acceptably small smeared induced metric fluctuations.

Passing the first test does not imply the second, and passing the second does not imply the third. Anderson, Molina-París, and Mottola motivate the homogeneous response criterion and state its domain in Anderson, Molina-París, and Mottola 2003, §§ III–V, PDF. Hu, Roura, and Verdaguer add the induced covariance and relate it to leading large-$N$ quantum correlations in Hu, Roura, and Verdaguer 2004, §§ II–VI, PDF.

The word “small” must refer to an observable. For a smeared gauge-invariant linear metric functional $H[F]$, one may compare $\operatorname{Var}H[F]$ with the squared background scale resolved by $F$. A local component at a mathematical point is neither gauge invariant nor generally finite. Different detectors can legitimately give different assessments because they probe different spectral and spacetime scales.

Results and limitations

The matrix records the literature available through the evidence cutoff 2026-08-19. A universal validity criterion remains unresolved: the controlled conclusions concern different observables, approximation orders, and backgrounds.

Evidence line	Controlled conclusion	What it supports	What it does not establish
Anderson, Molina-París, and Mottola 2003, §§ III–V, especially Eqs. (3.4) and (4.6), PDF	Gauge-invariant retarded response gives a necessary stability test; flat space is analyzed with a prescription for Planck-scale solutions.	Mean-field backgrounds must be tested dynamically, not only by the size of $\langle T_{ab}\rangle$.	Stability of arbitrary curved, nonstationary, or nonlinear solutions.
Hu, Roura, and Verdaguer 2004, § III, Eqs. (13)–(14), and Appendix C, Eqs. (C11)–(C18), PDF	The Einstein–Langevin two-point function reproduces the leading symmetrized quantum metric two-point function when intrinsic data are matched.	A controlled interpretation of Gaussian stochastic gravity at leading nontrivial order in $1/N$.	Higher metric correlators, graviton loops, or accuracy at finite $N$ without an error estimate.
Hu and Verdaguer 2008, §§ 3.2, 5.2.1, and 8.3.2	Stress covariance is positive after admissible smearing; coincidence and horizon limits can be singular.	Resolution-dependent induced-fluctuation tests and the need to state the detector.	A universal pointwise ratio of fluctuation to mean as a breakdown threshold.
Hu and Verdaguer 2008, §§ 4.1–4.3	The imaginary quadratic influence action defines Gaussian noise and the real causal part defines response.	A common state-consistent origin for noise and dissipation.	That decoherence has selected a unique classical spacetime, or that neglected higher influence cumulants are small.

The first two rows are the strongest positive results but have sharply limited scope. The third rules out an overly local diagnostic rather than supplying a universal replacement. The fourth explains why Gaussian stochastic equations arise, while leaving the quantum-to-classical interpretation dependent on coarse graining and histories. The review-level derivations and qualifications appear in Hu and Verdaguer 2008, §§ 3.2–3.3, 4.1–4.3, and 5.2.

Competing validity positions

Fluctuation-to-mean ratios versus smeared observables. One proposal diagnoses failure when a local stress variance is large relative to the squared mean. The objection is decisive in general: the denominator may vanish, the numerator is distributionally singular, and neither quantity specifies resolution. A smeared, dimensionless, gauge-invariant metric observable answers a better-posed question, but its admissible tolerance remains application dependent.

Homogeneous stability versus full stochastic stability. A bounded homogeneous response is necessary because a growing physical mode invalidates the background even without noise. The broader position also requires the induced covariance to stay controlled. These are complementary, not rival calculations: one uses poles of the retarded operator; the other convolves its propagator with the noise kernel.

Leading large-$N$ control versus finite-$N$ gravity. The strong positive position is that stochastic gravity is not merely phenomenological: at leading order it reproduces a well-defined quantum two-point function. The caution is that black-hole applications often have finite field content, nonlinear secular evolution, and horizon-sensitive smearings. No theorem turns the leading two-point agreement into a bound on every higher cumulant or on microscopic information recovery.

Decoherence versus classicality. Environmental decoherence can suppress interference between selected coarse-grained metric histories. That helps explain when stochastic trajectories are useful, but it does not by itself choose the coarse graining, prove small backreaction, or convert every stochastic realization into an observable spacetime.

A practical hierarchy of checks

Start with a renormalized self-consistent background and declare the EFT cutoff. Construct the causal response from the same state and renormalization prescription. Project onto gauge-invariant, smeared observables and check homogeneous poles, branch cuts, and secular terms. Next propagate both an admissible initial metric covariance and the stress noise kernel. Vary the smearing scale within the EFT window and verify that the conclusion is not created by contact singularities or cutoff-scale roots.

If the induced variance is large, Gaussian stochastic gravity diagnoses the failure of the mean description for that observable; it does not automatically provide the correct nonlinear replacement. If all two-point tests pass, the conclusion is still restricted to the perturbative order tested. Estimate higher stress cumulants, graviton-loop counting, and the time over which small corrections can accumulate.

Limitations and open questions

There is no background-independent numerical threshold separating “semiclassical” from “quantum gravitational.” Existing controlled equivalences concern linearized two-point functions, usually for free or effectively Gaussian matter and a specified large-$N$ scaling. General nonstationary horizons lack the equilibrium simplifications of a KMS state. Singular observables can fail even when all physically smeared observables in a finite-resolution experiment remain controlled.

Open questions. The conclusions could change when a theorem or counterexample changes the leading large-$N$/Einstein–Langevin correspondence; a gauge-invariant induced-fluctuation calculation extends control to a new nonstationary background; a finite-$N$ calculation bounds higher cumulants or graviton loops over secular times; or a cited result is corrected or withdrawn. A single valid counterexample can invalidate a general claim, whereas independent agreement strengthens confidence in a new positive result.

Exercise

Take $N$ independent identical matter fields with one-field stress mean $\mu$ and connected variance $\sigma^2$ in a fixed smeared channel. Under the large-$N$ scaling $G=\bar G/N$, determine the scaling of the total mean source, relative stress fluctuation, induced metric-fluctuation amplitude, and induced variance. Then state what changes if the fields have pairwise correlations of order $\sigma^2$.

Solution

Independence gives $\langle T_{\mathrm{tot}}\rangle=N\mu$ and $\operatorname{Var}T_{\mathrm{tot}}=N\sigma^2$. The mean gravitational source is

$$G\langle T_{\mathrm{tot}}\rangle =\frac{\bar G}{N}N\mu =\bar G\mu,$$

so it remains finite. The relative stress fluctuation scales as $\sqrt{N}\sigma/(N\lvert\mu\rvert)=O(N^{-1/2})$ when $\mu\ne0$. The stochastic metric amplitude scales as $G\sqrt{\operatorname{Var}T_{\mathrm{tot}}}=O(N^{-1/2})$, and its variance scales as $O(N^{-1})$.

If every pair carries an $O(\sigma^2)$ covariance, the $N(N-1)$ cross terms can make the total variance $O(N^2)$. The induced metric amplitude can then remain $O(1)$ rather than decrease. Large field number suppresses fluctuations only with the correlation assumptions used in the expansion; it is not a purely combinatorial guarantee.

References

• Anderson, P. R., Molina-París, C., and Mottola, E. 2003. “Linear Response, Validity of Semi-Classical Gravity, and the Stability of Flat Space.” Physical Review D 67, 024026. DOI. Open PDF.
• Hu, B. L., and Verdaguer, E. 2008. “Stochastic Gravity: Theory and Applications.” Living Reviews in Relativity 11, 3. DOI.
• Hu, B. L., Roura, A., and Verdaguer, E. 2004. “Induced Quantum Metric Fluctuations and the Validity of Semiclassical Gravity.” Physical Review D 70, 044002. DOI. Open PDF.