Lens and Light Observation Data

Light observation data record where and in which direction null signals are seen. They probe causal and conformal geometry without requiring clocks to measure timelike proper time. Two related data types are useful: a lens or scattering relation follows rays from an entry boundary to an exit boundary, while a light observation set collects the receivers reached from one unknown source event. Their information content and failure modes are different.

Required background. Analysis for Mathematical Relativity supplies singular and set-valued reconstruction; PDEs and Green Functions for Gravitation supplies propagation of wavefronts; and Lorentzian metrics and signature supplies null geodesics and conformal structure.

Helpful background. The Einstein equation explains when field equations can fix a conformal factor, and Causality and Global Structure supplies chronological diamonds and caustics.

Lens and observation data

Let $\partial M$ have an observed timelike portion $U$. A bare null lens relation schematically maps an inward projective future-null direction to its first outward return,

$$\mathcal S:(x,\xi)\longmapsto(y,\eta),$$

possibly together with a travel parameter. The domain excludes rays that are trapped, hit an unobserved boundary, or fail to return. If $c$ is a positive smooth function and only unparametrized rays and projective directions are measured, replacing $g$ by $c^2g$ leaves the relation unchanged. Vector or covector representatives require a declared boundary normalization, and an affine parameter requires one at entry.

There is a useful local scalar representation. Suppose two small timelike or spacelike hypersurfaces meet a unique connecting null geodesic transversally and its endpoints are not conjugate. The pairs joined by nearby null geodesics form a codimension-one submanifold; any defining function for that submanifold generates a reduced covector scattering relation. Its metric linearization is, up to a smooth nonzero factor, the light ray transform of the metric perturbation Stefanov 2024, Theorem 2.1 and §2.3.

For passive observations, let $V$ be a neighborhood of an observer’s timelike worldline and $q$ a source event in an unknown region $W$. Define the future light cone and its observation set by

$$\mathcal L^+(q)=\exp_q(L_q^+M)\cup\{q\}, \qquad P_V(q)=V\cap\mathcal L^+(q).$$

This is not generally $V\cap\partial J^+(q)$: beyond the cut locus the exponential image can enter $I^+(q)$ and can contain several arrivals. The earliest subset is instead

$$E_V(q)=\left\{x\in P_V(q): \begin{array}{l} \text{there are no }y\in P_V(q)\text{ and future-pointing timelike}\\ \alpha:[0,1]\to V\text{ with }\alpha(0)=y\text{ and }\alpha(1)=x \end{array} \right\}.$$

Recording observed null directions enriches this to a set in the projectivized null bundle over $V$. The family $\{E_V(q):q\in W\}$ is unlabeled in the passive problem: the reconstruction must identify which observed set represents which abstract interior point.

Reconstruction from earliest arrivals

Under global hyperbolicity and a causal accessibility condition, distinct source events in the relevant diamond have distinct earliest observation sets. This makes

$$q\longmapsto E_V(q)$$

an embedding into a space of closed subsets of $V$. The proof then proceeds geometrically:

1. inclusion and convergence relations among observation sets recover the topology of $W$;
2. arrival points along a family of receiver curves supply coordinate functions and recover its smooth structure;
3. tangent hyperplanes to smooth light fronts recover null covectors at each point;
4. the null cone determines the Lorentzian metric up to a positive conformal factor.

More precisely, Kurylev, Lassas, and Uhlmann compare smooth globally hyperbolic Lorentzian manifolds of dimension $n\geq3$. The observation regions are neighborhoods of known smooth timelike paths, the conformal geometry on those regions is identified, and each relatively compact source set lies in $J^-(p^+)\setminus I^-(p^-)$. Equality of the unindexed families of earliest light observation sets then yields a diffeomorphism of the source regions and equality of their conformal classes Kurylev, Lassas, and Uhlmann 2018, Theorem 1.2 and §§2, 5. Corollary 1.3 upgrades this to an isometry only for Ricci-flat source regions, when the apparatus identification is already isometric and every topological component meets the observation region. In general, metric scale needs additional amplitude, time-separation, volume, or field-equation information.

The method is constructive in a mathematical sense: it supplies procedures for building the recovered atlas and cone field from exact set data. It is not automatically a numerically stable algorithm. Small perturbations near a caustic can rearrange which branch arrives first.

Caustics and multiple arrivals

Away from the vertex and before a conjugate point, each light-front branch is locally a smooth null hypersurface. Different branches can already cross at a cut point without conjugacy, so smoothness of one branch does not make the whole front single-valued. At a caustic the exponential map loses rank, and beyond it several rays may reach one receiver. The full light observation set can self-intersect and need not be the graph of one arrival-time function. Keeping only the earliest branch avoids some multiplicity but discards later-ray information.

Caustics are therefore not simply noise. Their singular stratification encodes the null exponential map, but extracting it demands microlocal branch tracking. A theorem that assumes no conjugate points is describing a simpler cell of the problem; a theorem allowing them must say how branches are separated.

For manifolds with boundary and reflected null rays, Hintz and Uhlmann assume a smooth connected Lorentzian manifold with timelike boundary and a proper surjective temporal function. In the introductory theorem the boundary is strictly null-convex; the full theorem replaces this by tameness of the broken null flow. The source regions are open with compact closure, the observation portions satisfy $U^0\Subset U$, distinct sources have distinct observations on $U^0$, and no observed boundary point on a relevant broken null geodesic is conjugate to its source. Under those hypotheses the boundary light observation sets determine the topological, differentiable, and conformal structure of the source region. The constructive proof still allows interior conjugate points, multiple reflections, and self-intersecting light cones Hintz and Uhlmann 2019, Theorem 1.2 and §§2.4, 3. Reflection obeys the specified Lorentzian Snell law; an absorbing or partially transmitting boundary would define different data.

Visibility and nonuniqueness

Partial boundary coverage produces a visibility region: points from which a sufficient family of null geodesics reaches the observed set with usable directions. A trapped set, shadow region, event horizon, or missing receiver interval can leave geometry undetermined. One may sometimes extend the result using field equations or analyticity, but light data alone do not justify it.

The unavoidable gauges are also easy to check. A diffeomorphism fixed on $V$ relabels the interior without changing the observations, and a conformal rescaling preserves null curves. The limitation of a light-set theorem is therefore both regional and conformal. Calling the output “the spacetime metric” without these qualifications overstates the result.

Flat-space diagnostic

In a convex domain of Minkowski space, a future null ray enters at $(x,\xi)$ and exits at $(y,\eta)$. What pieces of data change under the constant rescaling $g\mapsto c^2g$?

Solution

The null cone, unparametrized ray, entry and exit points, and tangent directions as projective null directions do not change. If tangent vectors are normalized using an observer on the boundary, their numerical representative may rescale with that convention. An affine length is not intrinsically fixed for a null curve; a boundary normalization can make its parameter observable, but under metric rescaling the normalization changes. Thus bare lens data determine the conformal geometry, while calibrated amplitude or clock information is needed to recover scale.

References

• Hintz, Peter, and Gunther Uhlmann. “Reconstruction of Lorentzian Manifolds from Boundary Light Observation Sets.” International Mathematics Research Notices 2019 (2019): 6949–6987. DOI.
• Kurylev, Yaroslav, Matti Lassas, and Gunther Uhlmann. “Inverse Problems for Lorentzian Manifolds and Non-linear Hyperbolic Equations.” Inventiones Mathematicae 212 (2018): 781–857. DOI.
• Stefanov, Plamen. “The Lorentzian Scattering Rigidity Problem and Rigidity of Stationary Metrics.” arXiv:2212.13213v3 (2024). arXiv.