The Logic of Testing Gravity

Testing gravity is an exercise in conditional reasoning. Field equations generate solutions only after sources and boundary data are supplied; solutions generate observables only after an operational protocol is specified; instruments generate data only through calibration and selection; and data support a conclusion only through a statistical model. This chapter builds that chain before any particular experiment is allowed to carry physical weight.

Helpful background. Weak Metric Perturbations and Gauge Freedom supplies the weak-field gravity expansion used by many tests. Probability and Inference for Gravity supplies probability and statistical inference. The Einstein Equation supplies the Einstein equations defining the reference theory. The Tests of Gravity overview supplies the volume-wide map of regimes and evidence.

Enter this chapter

The central object is not a residual or a fitted coefficient but a complete comparison,

$$\text{theory and conditions} \longrightarrow \text{observable} \longrightarrow \text{measured data} \longrightarrow \text{calibrated inference} \longrightarrow \text{bounded conclusion}.$$

Each arrow can fail in a different way. An invalid approximation is not repaired by a better detector. A missing calibration process is not repaired by a tighter theory prior. A statistically significant residual does not select a theory that was never used to predict the data. The pages in this chapter isolate these logical moves so later chapters can concentrate on their physical realizations.

The methodology and current-source statements use an evidence cutoff of 2026-08-19. That date fixes release identities and review boundaries; it is not a claim that a stable definition expires.

Route through the argument

1. Theory, Observable, and Measurement defines the forward map from gravitational dynamics through source, propagation, instrument, and noise models to data. Begin here to learn why a coordinate quantity or pipeline estimate is not automatically an observable.
2. Null Tests and Alternative-Model Tests separates an internal inconsistency statistic from a comparison between specified predictive hypotheses. It explains what a rejection permits—and what it does not.
3. Regimes, Scales, and Environmental Dependence identifies the dimensionless controls behind “weak,” “strong,” radiative, screened, laboratory, compact-body, and cosmological gravity.
4. Nuisance Parameters and Systematic Error places calibration, source physics, selection, correlated uncertainty, and model discrepancy inside the likelihood and its validation program.
5. Theory Priors and Parameter Measures shows how coordinates, support, stability, cutoffs, and hierarchical structure affect posterior and evidence statements.
6. From Constraint to Physical Conclusion assembles the full theory map and explains why a phenomenological interval is not yet a universal theory exclusion.

This is also the recommended reading order. The last page deliberately depends on distinctions established by the first five; reading it alone can make the caveats look optional when they are actually premises of the conclusion.

A reusable gravity-test checklist

For any later page, ask the following questions in order.

Layer	Required statement	Diagnostic check
Theory	Equations, parameters, branch, boundary data, and domain	Recover a controlled GR or Newtonian limit
Observable	Operational quantity and source/environment map	Verify invariance, dimensions, sign, and scaling
Measurement	Dataset release, calibration, selections, and estimator	Reprocess an injection through the pipeline
Statistics	Normalized likelihood, priors, nuisance covariance, and interval rule	Check coverage or simulation-based calibration
Interpretation	Exact theory mapping and scope of the claim	Predict an independent observable or regime

No row can be replaced by citing a confidence level. A $95\%$ interval may be correctly calculated for an incomplete model; a physically complete model may be weakly constrained; and two individually sound tests may share data or priors that prevent naive multiplication.

Three recurring distinctions

Sensitivity versus evidence. Expected sensitivity describes what an experiment could distinguish under specified signals and noise. Evidence describes what the realized data support. A null result from an insensitive experiment is not strong evidence for the null.

Uncertainty versus systematic error. A represented nuisance process contributes uncertainty that can be propagated. A misspecified process causes bias or miscoverage. Calling every difficulty “systematic uncertainty” conceals whether it is actually in the model.

Framework versus theory. PPN coefficients, waveform deformations, slip functions, and residual bases organize departures. A complete theory supplies correlations, environmental dependence, stability, and a domain. Framework constraints become theory constraints only through that additional structure.

Review the chapter

Consider a hypothetical clock network that finds an annual modulation in frequency ratios. Construct a three-stage program that distinguishes a new gravitational coupling from calibration and environmental explanations.

Solution

First, define a null statistic and calibrate it with the actual sampling, clock transfer functions, missing data, and the predeclared search over frequency and phase. Second, fit shared temperature, magnetic, reference-standard, and transfer-link nuisance processes jointly, then inject both null and modulated signals through the frozen pipeline to check coverage and power. Third, map one viable coupling to clock sensitivities and predict a different composition, site, or gravitational-potential dependence. The new-gravity interpretation becomes stronger only if that independent scaling is recovered while environmental monitors and alternative calibration models cannot reproduce it.

Further reading

• Cowan, G., Cranmer, K., Gross, E., and Vitells, O. (2011), “Asymptotic formulae for likelihood-based tests of new physics,” European Physical Journal C 71, 1554. Cowan et al. 2011, §§2–4.
• Mandel, I., Farr, W. M., and Gair, J. R. (2019), “Extracting distribution parameters from multiple uncertain observations with selection biases,” Monthly Notices of the Royal Astronomical Society 486, 1086–1093. Mandel, Farr, and Gair 2019, §§2–4.
• Particle Data Group (2024), “Statistics,” in Review of Particle Physics, Physical Review D 110, 030001. Particle Data Group 2024, §§40.1–40.4, PDF.
• Will, C. M. (2014), “The confrontation between general relativity and experiment,” Living Reviews in Relativity 17, 4. Will 2014, §§2–4.