Quantum-Gravity Observables and Test Taxonomy

A quantum-gravity test is an inference chain from a defined microscopic or effective mechanism to a calibrated data distribution. The chain is only as strong as its least controlled link. The useful unit of comparison is therefore not an experiment “for” a named program, but a tuple of operator, regime, source, propagation, response, nuisance model, statistical test, and attainable conclusion.

Required background. Observable and Regime Matrix for Quantum Gravity supplies the distinction between microscopic, semiclassical, and asymptotic observables. From Measured Intensity to Many-Body Claim supplies the measurement-to-claim logic. Helpful background. Model Selection and Parameter Inference develops likelihood comparison, while Analogue and Phenomenological Evidence Ceilings explains why reproducing a mechanism is not identifying a microscopic theory.

The mechanism-to-data map

Let $\theta$ denote the coefficient or state parameter of interest, $\lambda$ conventional physics parameters, and $\nu$ calibration, background, and population nuisances. A forward model has the schematic composition

$$\theta \longrightarrow \mathcal{L}_{\mathrm{eff}}(\theta,\lambda) \longrightarrow s_{\mathrm{src}} \longrightarrow s_{\mathrm{prop}} \longrightarrow \mu(\theta,\lambda,\nu) \longrightarrow p(d\mid\theta,\lambda,\nu).$$

Here $\mu$ is the expected detector-level signal and $d$ is the recorded datum. Each arrow carries assumptions. Matching a UV construction to $\mathcal{L}_{\mathrm{eff}}$ may be unavailable; the EFT may require a preferred frame; the source may have unknown intrinsic lags; propagation can introduce plasma or lensing effects; the response can drift; and the likelihood can neglect tails. A valid result names these assumptions rather than hiding them inside a single “quantum-gravity scale.”

For independent data segments $d_i$ with a shared parameter $\theta$ and segment-specific nuisances $\nu_i$,

$$p(\theta\mid d)\propto \pi(\theta) \int d\lambda\,\pi(\lambda) \prod_i\left[ \int d\nu_i\, p(d_i\mid\theta,\lambda,\nu_i)\pi(\nu_i) \right].$$

Combining many events is powerful only if selection effects and correlations are included. Reusing the same calibration, waveform family, foreground map, or EFT assumption does not create independent confirmation.

First application: a Planck-suppressed timing signal

Consider a photon group velocity

$$v(E,z)\simeq 1-\xi\frac{E(1+z)}{M_{\mathrm{QG}}},$$

with an observed delay between energy channels

$$\Delta t_{\mathrm{obs}} =\xi\frac{\Delta E}{M_{\mathrm{QG}}} \int_0^{z_s}\frac{1+z}{H(z)}\,dz +\Delta t_{\mathrm{int}} +\Delta t_{\mathrm{inst}}.$$

The first application is to build the full chain for this signal. The coefficient is $\xi/M_{\mathrm{QG}}$ in a specified photon-sector operator basis. The source model supplies the intrinsic lag $\Delta t_{\mathrm{int}}$ and its population distribution. Cosmology supplies $H(z)$. The instrument supplies energy dispersion, dead time, and clock calibration. The likelihood must fit propagation and intrinsic lags jointly. A bound is reproducible only after the energy definition, redshift posterior, event selection, nuisance priors, and confidence construction are stated.

The strongest ordinary conclusion is then: within this dispersion model and source-lag family, the coefficient lies in the reported interval. It is not a measurement of the Planck scale, and it is not evidence for or against every UV program. Quantum gravity treated as an EFT is predictive at low energy, but generic Planck suppression alone does not determine Wilson coefficients Donoghue 1994, §§ II–III.

Evidence classes and claim ceilings

Five labels prevent category errors:

• Forecast: simulated sensitivity under declared noise, duty cycle, and population assumptions.
• Exclusion: a parameter region disfavored within a forward model.
• Anomaly: an unexplained residual whose instrumental and conventional alternatives remain live.
• Mediator witness: data incompatible with a stated class of classical mediation models.
• Detection: a reproducible signal with controlled systematics and a physical interpretation that survives serious alternatives.

Bayes factors, likelihood ratios, and posterior intervals answer different questions. A Bayes factor compares specified models and depends on prior volume; a confidence interval describes repeated-sampling coverage under a model; neither supplies a missing physical alternative. Look-elsewhere corrections matter when the statistic or signal morphology was selected after inspecting the data.

Adversarial control: manufacture the same residual

Inject an energy-dependent clock drift or an intrinsic source lag drawn from a correlated population. Pass it through the same selection and fitting pipeline used for the proposed propagation effect. If the recovered posterior favors $\xi\ne0$, the analysis has not identified quantum-gravity propagation. The remedy may be an independent clock channel, a hierarchical source model, a redshift scaling test, or a second messenger with different systematics.

The control should end with a downgrade rule. If calibration injection reproduces the effect, the claim is instrumental. If intrinsic-lag flexibility absorbs it, the result is a source-model constraint. If the propagation coefficient is stable across detectors, sources, energies, and alternative lag models, it becomes an anomaly in that operator channel—not yet a microscopic detection.

Common pitfalls

Calling a large suppression scale a detection. A lower bound on $M_{\mathrm{QG}}/\lvert\xi\rvert$ is an exclusion in a coefficient model. The numerical proximity of that bound to $M_{\mathrm{Pl}}$ has no evidential force by itself.

Comparing programs through generic effects. Modified dispersion, decoherence, or short-distance corrections can arise in many models and in program-neutral EFTs. Program discrimination requires a derived coefficient pattern or correlated signature that was fixed before the data comparison.

The chapter overview contains the structure diagram and validity and failure diagram. They are embedded there once so that their shared chapter-level context is not repeated on every article.

For the chapter-wide comparison of assumptions, counterevidence, falsifiers, and claim ceilings, see the claim-domain table.

References

• Amelino-Camelia, G. “Quantum-Spacetime Phenomenology.” Living Reviews in Relativity 16, 5 (2013). DOI.
• Donoghue, J. F. “General Relativity as an Effective Field Theory: The Leading Quantum Corrections.” Physical Review D 50, 3874–3888 (1994). DOI.
• Liberati, S. “Tests of Lorentz Invariance: A 2013 Update.” Classical and Quantum Gravity 30, 133001 (2013). DOI.