Active Wave Measurements

Active measurements replace passive light sightings with controlled sources and recorded wave responses. The measurement may be a source-to-solution map in an interior observation region or a Dirichlet-to-Neumann map on a timelike boundary. Because singularities propagate along null bicharacteristics, the response contains light observation data; amplitudes and lower-order terms can carry additional metric information.

Required background. Analysis for Mathematical Relativity supplies unique continuation and Sobolev control; PDEs and Green Functions for Gravitation supplies hyperbolic boundary maps; and Lorentzian metrics and signature supplies causal domains and wave operators.

Helpful background. The Einstein equation supplies the coupled nonlinear application, and Causality and Global Structure supplies domains of dependence and chronological diamonds.

Two response operators

For a globally hyperbolic spacetime without boundary and an observation region $V$, solve the retarded problem

$$(\Box_g+Q)u=f, \qquad u=0\ \text{before }\operatorname{supp}f,$$

and define

$$L_{g,Q}:f\longmapsto u\vert_V, \qquad \operatorname{supp}f\subset V.$$

For a manifold with timelike boundary, prescribe $u\vert_{\partial M}=h$ and measure its outward normal derivative,

$$\Lambda_{g,Q}:h\longmapsto\partial_\nu u\vert_{\partial M}.$$

The exact time interval, observed boundary portion, initial condition, and function spaces are part of either definition. A diffeomorphism fixed on the measurement set leaves the map unchanged. Depending on the equation and data, a conformal transformation can also be invisible or can be accompanied by a transformation of $Q$.

Away from coincident source and receiver points, wavefront singularities of the response propagate along null bicharacteristics, with additional reflected or glancing branches when the boundary problem permits them. The kernel may also have a singularity on the diagonal. Its propagating canonical relation records endpoint and direction data; its principal symbol records geometric spreading, boundary transmission, and coefficients. This is how a wave response contains more than a list of travel times.

A linear reconstruction theorem

Eskin studies formally self-adjoint second-order hyperbolic operators on admissible domains that have been carried to cylinders $D_0\times\mathbb R$. The coefficients are smooth, the Lorentzian principal part obeys the stated uniform hyperbolicity and spatial ellipticity conditions, the real vector potential and the other coefficients are analytic in time, and $\Gamma_0\times\mathbb R$ is a nonempty open part of the timelike boundary. The Bardos–Lebeau–Rauch exact-controllability condition must hold on a sufficient interval beginning at every initial time. If the two partial DN maps agree on that boundary portion for every compactly supported smooth Dirichlet input, the complete operators are related by a proper boundary-fixed change of variables and a unit-modulus gauge fixed on the observed boundary Eskin 2017, Theorem 1.2 and §§1–7.

The proof architecture is instructive:

1. geometric-optics solutions extract integral identities and boundary jets;
2. a characteristic-surface version of the boundary-control method gives a local identification near the observed boundary;
3. controllability and domain-of-influence arguments move the reconstruction inward;
4. successive local identifications are glued modulo the declared gauge.

The theorem’s observation and control hypotheses ensure sufficient boundary visibility; they are not automatic for a trapped geometry or an arbitrary short time interval. Other microlocal arguments instead extract light observation sets directly from the singular support of a response kernel.

Unique continuation and visibility

To turn a response operator into interior information, one needs a mechanism preventing a nonzero hidden solution from vanishing on the observed set. Unique continuation across a suitable hypersurface supplies that mechanism locally; iterating it expands the recoverable region. Geometric control or convex foliation hypotheses can ensure that relevant rays meet the boundary.

Alexakis, Feizmohammadi, and Oksanen prove recovery of a smooth zeroth-order coefficient $Q$ on a known metric background from the full boundary Cauchy data set of $\Box_g+Q$. Their connected smooth Lorentzian manifold has timelike boundary and must satisfy all five geometric hypotheses: a proper surjective temporal function; the stated curvature bound and, when its constant is positive, spatial-diameter bound; a global exponential-map and causal-path condition outside each double null cone; null nontrapping; and finite-order contact of null geodesics with the boundary. Equality of Cauchy data then implies equality of the coefficients Alexakis, Feizmohammadi, and Oksanen 2022, Theorem 1.2 and §§1–2. Their Corollary 1.3 exhibits a nonempty open class of smooth, genuinely time-dependent metrics satisfying these conditions. The result is not unrestricted simultaneous metric and coefficient reconstruction.

Trapped null rays or an observation time shorter than the return time violate visibility. A perturbation supported in a causally inaccessible region cannot affect the measured response during that interval. This counterexample shows why “full boundary data” still needs a sufficiently long time and a declared accessible domain.

Linearization and its boundary

For a nonlinear equation

$$\Box_gu+H(x,u)=f,$$

assume $H(x,0)=0$ and use vanishing past data. Then $u=0$ is the background solution, and the first derivative of its source-to-solution map at $f=0$ is the linear response for $\Box_g+\partial_uH(x,0)$. It transports each injected singularity independently. If the corresponding linear inverse problem has invisible directions or gauge, first-order data cannot remove them.

Higher Fréchet derivatives contain products of linearized waves. Those interactions generate new singularities in the interior and are the subject of the next page. Kurylev, Lassas, and Uhlmann show that this nonlinear information can reconstruct the conformal geometry of the causally reachable and returnable diamond even where the associated linear inverse problem is open Kurylev, Lassas, and Uhlmann 2018, Theorem 1.5 and §§3–5.

The limitation is conceptual as well as technical: an exact infinite- dimensional response operator represents arbitrarily many noiseless experiments. A finite, noisy acquisition is not equivalent to knowing that operator, and qualitative uniqueness does not supply a condition number or a convergent discretization.

Check causal invisibility

Let $K$ be a compact metric perturbation whose causal future does not meet the observation set $V$ before time $T$. Can the restricted map $L:f\mapsto u\vert_{V\cap\{t\le T\}}$ detect $K$?

Solution

No. Finite propagation speed confines the influence of both the sources and the perturbation to their causal futures. If no causal curve from the source region passes through $K$ and then reaches the observed spacetime slab, the retarded solutions agree there. This is an exact nonuniqueness, not merely poor stability. Enlarging the time interval helps only if a source–perturbation– receiver causal path then exists.

The causal separation must hold for both compared forward problems (or for a common cone bound on the admissible perturbations). Changing an unknown metric changes its cones; invisibility established only in one background does not justify an arbitrary metric perturbation that opens a new causal route to the receiver.

References

• Alexakis, Spyros, Ali Feizmohammadi, and Lauri Oksanen. “Lorentzian Calderón Problem under Curvature Bounds.” Inventiones Mathematicae 229 (2022): 87–138. DOI.
• Eskin, Gregory. “Inverse Problems for General Second Order Hyperbolic Equations with Time-Dependent Coefficients.” Bulletin of Mathematical Sciences 7 (2017): 247–307. 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.