Light Deflection and Astrometric Tests

Astrometric gravity tests measure how a source direction changes as the light ray passes a gravitating body. The relativistic signal is achromatic and has a calculable elongation dependence, but the measured angle also contains source motion and structure, observer motion, plasma refraction, catalog-frame rotation, and instrument attitude. A constraint on light bending must fit those layers together.

Required background. Weak-Field Gravity supplies the two metric potentials and PPN interpretation; Geodesics and Jacobi Fields supplies null trajectories and angular separation. Helpful background. Probability and Statistical Inference supplies correlated astrometric likelihoods; Tests of Gravity supplies the separation of observable, calibration, and conclusion.

Metric light bending

For a ray with impact parameter $b$ passing an isolated spherical mass $M$, the leading asymptotic PPN deflection is

$$\delta\theta = 2(1+\gamma)\frac{GM}{b}.$$

General relativity has $\gamma=1$ and therefore gives $4GM/b$. The expression is dimensionless, falls as $b^{-1}$, and vanishes with $M$. The observed deflection between finite source and observer positions also contains endpoint geometry, and a Solar-System reduction must include all gravitating bodies, their multipoles, and the observer ephemeris. Will 2014, §4.1.1 and Eqs. (59)–(61) distinguishes the coordinate trajectory from the angle measured relative to a local reference direction.

The factor $1+\gamma$ can be understood from the weak metric: time curvature and spatial curvature both contribute to the null ray’s transverse acceleration. Integrating along an unperturbed path with coordinate $z$ and closest approach $b$ gives, with $c$ restored,

$$\delta\theta =\frac{(1+\gamma)GM}{c^2} \int_{-\infty}^{\infty}\frac{b\,dz}{(b^2+z^2)^{3/2}} =\frac{2(1+\gamma)GM}{bc^2}.$$

This first-order straight-path calculation requires $GM/(bc^2)\ll1$. Using $GM_\odot/c^2=1.477$ km and $b=6.96\times10^5$ km gives about $1.75$ arcseconds in GR for a grazing solar ray. For an effectively infinite-distance source observed at finite solar distance $r$, the elongation $\chi$ gives $\delta\theta=(1+\gamma)GM\cot(\chi/2)/(c^2r)$. At $\chi=90^\circ$ and $r=1$ au this is about $4.1$ milliarcseconds. Substituting $b=r$ into the two-infinite-endpoint formula would incorrectly double this value.

At radio frequency $\nu$, cold-plasma bending scales approximately as $\nu^{-2}$ times the transverse gradient of electron column, whereas metric bending is achromatic. A uniform column delays the radio signal without bending it. Multi-frequency very-long-baseline interferometry can therefore separate much of the solar-corona term. Optical observations avoid that radio delay but introduce chromatic image calibration, scan-angle coverage, point-spread-function, source-color, and attitude effects. Quasars provide nearly fixed distant directions, yet source structure and photocenter variability can still shift their apparent positions.

Astrometric forward model

For source $q$ at epoch $t$ and band $\nu$, a useful model is

$$\mathbf n^{\rm obs}_{qt\nu} = \mathcal A_t \left[ \mathbf n_{q0} +\boldsymbol\mu_q(t-t_0) +\boldsymbol\pi_q(t) +\sum_A\delta\mathbf n_A(\gamma,J_{2A},\ldots) +\delta\mathbf n_{\rm plasma}(\nu) \right] +\mathbf c_{qt\nu} +\boldsymbol\epsilon.$$

The operator $\mathcal A_t$ maps the barycentric direction through aberration and instrument attitude; $\boldsymbol\mu_q$ and $\boldsymbol\pi_q$ are proper motion and parallax; $\delta\mathbf n_A$ is gravitational deflection; and $\mathbf c$ contains calibration terms. Catalog frame rotation and spin must be solved using a global set of sources, not absorbed into one conjunction.

The evidence basis is current through 2026-08-19. Any result must pin the epoch-astrometry dataset, catalog version, source list, observer ephemeris, calibration and attitude solution, solar-elongation and quality selection, covariance model, software version, and retrieval date. Gaia analyses should name an actually available DR3 or focused-product data product rather than an anticipated later catalog. The DR3 content and validation boundary are documented in Gaia Collaboration 2023, §§2–7 and Appendices A–C.

Selection includes elongation, magnitude, color, source morphology, scan-angle coverage, radio structure quality, frequency availability, saturation, and rejected epochs. Completeness is with respect to that selected astrometric sample. Systematic uncertainty from calibration variables shared by many sources induces covariance; treating catalog entries as independent can make the uncertainty on $\gamma$ unrealistically small. The fit should separate source-structure evolution, chromatic centroid shifts, attitude and basic-angle drift, catalog-frame rotation, plasma-model error, and ephemeris error, then propagate their cross-source and cross-epoch correlations. Alternate source subsets and calibration bases test whether the gravity parameter is absorbing one of those structured effects.

Likelihood and discriminating controls

With stacked tangent-plane residuals $\mathbf r$,

$$-2\ln\mathcal L = \mathbf r^{\mathsf T}C^{-1}\mathbf r +\ln\det C +\text{constant}.$$

The parameter vector should include $\gamma$, source positions and motions, frame rotation, attitude and basic-angle terms, plasma coefficients, solar multipoles, and relevant calibration hyperparameters. Priors or auxiliary constraints and the confidence or credible construction must be stated. Residuals should be checked against elongation, observing frequency, source color and structure, scan direction, and epoch.

Three controls make the interpretation harder to fake:

1. compare sources observed on opposite sides of the gravitating body, for which the deflection direction reverses;
2. use multiple radio frequencies to isolate dispersive plasma, or multiple optical colors to test chromatic calibration; and
3. inject a shift in $\gamma$ while refitting source and attitude parameters, then verify recovery on withheld sources and epochs.

The reference frame, time scale, observer motion, and data construction used by planetary ephemerides and astrometry are described together in Fienga and Minazzoli 2024, §§3.2–3.5 and §5. A deflection residual from a fixed catalog is not by itself a gravity result if source positions or attitude were determined under the same bending model. Reprocessing or a joint fit is required.

Reproduction requires the official data product and calibration, source and epoch masks, observer ephemeris, light-propagation code, covariance construction, and a machine-readable residual table. The pipeline should recover the GR elongation law in a synthetic catalog, return null on a no-mass injection, and remain stable under alternate plasma and frame models.

Exercises

Two radio observations of the same quasar at frequencies $\nu_1$ and $\nu_2$ contain an achromatic gravitational shift $g$ and a plasma shift $p/\nu^2$. Solve for $g$ and explain what assumption limits the construction.

Solution

Writing $d_i=g+p/\nu_i^2$, multiply by $\nu_i^2$ and subtract to eliminate $p$. This gives $g=(\nu_1^2d_1-\nu_2^2d_2)/(\nu_1^2-\nu_2^2)$. The result follows because the gravitational term is common while the leading plasma term is exactly dispersive. It fails if source structure, instrumental phase, or higher-order plasma effects differ between frequencies and are not calibrated. Their covariance must be included, and additional bands or off-conjunction observations are needed to test the assumed $\nu^{-2}$ law.

References

• Fienga, A., and Minazzoli, O., “Testing Theories of Gravity with Planetary Ephemerides,” Living Reviews in Relativity 27 (2024) article 1, DOI.
• Gaia Collaboration, “Gaia Data Release 3: Summary of the Content and Survey Properties,” Astronomy & Astrophysics 674 (2023) A1, DOI, official archive.
• Will, C. M., “The Confrontation between General Relativity and Experiment,” Living Reviews in Relativity 17 (2014) article 4, DOI.