Matrix-Weighted X-Ray Transform

• Matrix-weighted X-ray transform is a generalization of the classical X-ray transform, integrating vector- or matrix-valued functions along geodesics with matrix attenuation.
• It provides rigorous injectivity and stability results in two-dimensional and higher-dimensional settings, using methods like foliation and analytic microlocal analysis.
• The framework extends to non-Abelian settings, underpinning applications in geometric, quantum, and polarization tomography, with practical inversion and statistical reconstruction techniques.

The matrix-weighted X-ray transform generalizes the classical scalar geodesic ray transform by introducing a matrix-valued weight, thereby enabling the integration of vector- or matrix-valued functions along geodesics with multiplicative matrix attenuation or more general coupling. This framework underpins a range of inverse problems involving connections, Higgs fields, and non-Abelian scattering data, with significant implications for geometric tomography, quantum and polarization tomography, and microlocal analysis.

1. Definition and Basic Framework

Let $(M,g)$ be a compact, oriented Riemannian manifold of dimension $n\geq 2$ with smooth, strictly convex boundary. The unit tangent bundle is denoted $SM = \{(x,v)\in TM : |v|_g=1\}$. For $(x,v)\in SM$, let $\gamma_{x,v}(t)$ be the maximal unit-speed geodesic starting at $x$ in the direction $v$, with $\tau_+(x,v)$ its first exit time from $M$.

Given a continuous, typically smooth (or real-analytic in recent developments) map $W: SM \rightarrow \mathrm{GL}(N, \mathbb{C})$, called a matrix weight, and a function $n\geq 2$0, the matrix-weighted geodesic X-ray transform $n\geq 2$1 is defined as:

$n\geq 2$2

where $n\geq 2$3 and $n\geq 2$4 is the inward unit normal. In the non-Abelian (scattering) variant, one is given a matrix field $n\geq 2$5 and considers the fundamental matrix solution $n\geq 2$6 of the ODE:

$n\geq 2$7

with $n\geq 2$8 forming the inverse problem’s data $n\geq 2$9 (Monard et al., 2019, Paternain et al., 2016, Chihara et al., 30 Jun 2025).

2. Relation to Classical and Attenuated X-Ray Transforms

The matrix-weighted transform strictly generalizes the scalar attenuated X-ray transform. When $SM = \{(x,v)\in TM : |v|_g=1\}$0 is scalar, or $SM = \{(x,v)\in TM : |v|_g=1\}$1, the ODE becomes separable, yielding:

$SM = \{(x,v)\in TM : |v|_g=1\}$2

Thus, $SM = \{(x,v)\in TM : |v|_g=1\}$3 recovers the classical scalar attenuated X-ray transform of $SM = \{(x,v)\in TM : |v|_g=1\}$4. In the general case, $SM = \{(x,v)\in TM : |v|_g=1\}$5 accounts for the effects of parallel transport in a vector bundle with connection, Higgs field, or more general matrix-valued attenuation or weight (Monard et al., 2019, Chihara et al., 30 Jun 2025, Paternain et al., 2016).

3. Injectivity and Uniqueness Results

3.1 Two-Dimensional and Real-Analytic Settings

On simple surfaces ($SM = \{(x,v)\in TM : |v|_g=1\}$6, no conjugate points, strictly convex boundary), the transform $SM = \{(x,v)\in TM : |v|_g=1\}$7 is injective for smooth matrix weights $SM = \{(x,v)\in TM : |v|_g=1\}$8 and for the scattering case $SM = \{(x,v)\in TM : |v|_g=1\}$9 with $(x,v)\in SM$0 valued in $(x,v)\in SM$1 or $(x,v)\in SM$2 (Monard et al., 2019, Paternain et al., 2020). For real-analytic metrics and weights, local injectivity near a strictly convex boundary point, as well as global injectivity on analytic surfaces, is established for any real-analytic $(x,v)\in SM$3-valued weight (Chihara et al., 30 Jun 2025). In the non-Abelian (connection/Higgs field) setting, the scattering data locally determines the Higgs field near strictly convex boundary, and global uniqueness holds by analytic continuation (Chihara et al., 30 Jun 2025).

3.2 Higher Dimensions and Foliation

For $(x,v)\in SM$4, global injectivity holds for smooth matrix weights under the “foliation condition”: existence of a smooth strictly convex exhaustion function, whose sublevel sets form strictly convex hypersurfaces throughout $(x,v)\in SM$5 (Paternain et al., 2016, Bohr, 2020, Ilmavirta et al., 2019). This enables layer-stripping arguments that successively recover $(x,v)\in SM$6 across layers from the boundary inward.

3.3 Piecewise Constant Functions and Minimal Regularity

For piecewise constant $(x,v)\in SM$7 and continuous injective matrix weights $(x,v)\in SM$8, injectivity holds in all dimensions $(x,v)\in SM$9 under nontrapping and strictly convex boundary (plus foliation for $\gamma_{x,v}(t)$0), without derivative bounds or assumption regarding conjugate points (Ilmavirta et al., 2019).

3.4 Generic and Twisted Matrix Weights

Generic injectivity for the X-ray transform twisted by a unitary connection on a Hermitian vector bundle (i.e., parallel transport matrix weights) has been proven in dimension $\gamma_{x,v}(t)$1 on Anosov or strictly convex no-conjugate-point manifolds (Cekić et al., 2021). The microlocal-perturbative method used extends to matrix weights of finite Fourier degree in velocity.

4. Stability, Reconstruction, and Statistical Inversion

Quantitative stability results are established both in two and higher dimensions. In the two-dimensional non-Abelian case, there exist constants depending on $\gamma_{x,v}(t)$2 and $\gamma_{x,v}(t)$3-norms of $\gamma_{x,v}(t)$4 such that:

$\gamma_{x,v}(t)$5

with an explicit form for $\gamma_{x,v}(t)$6. This enables statistical inversion: using a Bayesian framework with Gaussian process priors for $\gamma_{x,v}(t)$7, noisy measurements of $\gamma_{x,v}(t)$8 are inverted by (infinite-dimensional) pCN-MCMC, with contraction rates for both the posterior and the mean estimator. For smooth $\gamma_{x,v}(t)$9, the $x$0-error in recovery decays with the optimal $x$1 rate in the number $x$2 of measurements (Monard et al., 2019).

In higher dimensions, H\"older-type stability estimates relate the $x$3-norm of the error in $x$4 on compact sets to a H\"older power of the $x$5-difference of the data $x$6 on suitable geodesic classes, with explicit control of exponents and constants in terms of $x$7-norms of the weights (Bohr, 2020).

5. Microlocal and Analytic Techniques

Recent advances rely on analytic microlocal analysis and the double fibration framework: the matrix-weighted transform is realized as a Fourier integral operator associated to the twisted conormal bundle of the incidence variety in $x$8. Under the Bolker condition (no conjugate points), analytic or smooth singularities of $x$9 are propagated to those of $v$0. In the real-analytic setting, propagation of analytic wavefront sets via the matrix-weighted transform allows for unique analytic continuation, and hence global injectivity, even from partial data near a strictly convex boundary point (Chihara et al., 30 Jun 2025).

6. Applications, Obstructions, and Extensions

Typical applications include quantum state tomography, where a matrix-valued Hamiltonian is probed via its propagator, and polarization tomography, involving tensor weights and recovery of perturbations. In the setting involving both connection and Higgs field, the kernel of the matrix-weighted X-ray transform consists of gauge transformations, i.e., $v$1 for suitable $v$2 vanishing at the boundary. Uniqueness is then modulo this natural gauge (Paternain et al., 2016).

The methodology accommodates Banach-space valued $v$3 and operator-valued, merely continuous weights, provided injectivity in the fiber (Ilmavirta et al., 2019). Inverse problems with minimal regularity, e.g., piecewise-constant targets and continuous weights, are covered. However, for smooth weights and functions, non-injectivity can occur on simple 2-manifolds (cf. Boman counterexample), indicating the sharpness of regularity conditions.

7. Overview Table: Notions and Key Results

Setting	Hypotheses	Result Type
2D, real-analytic	Strictly convex, analytic boundary	Local/global injectivity for analytic weights (Chihara et al., 30 Jun 2025)
2D, smooth/simple surface	Strictly convex, no conjugate points	Injectivity for $v$4 / $v$5 weights (Monard et al., 2019, Paternain et al., 2020)
$v$6, smooth	Foliation, strictly convex boundary	Global injectivity for smooth weights (Paternain et al., 2016, Ilmavirta et al., 2019)
Any $v$7, piecewise constant $v$8	Strictly convex boundary, foliation ($v$9)	Injectivity for continuous weights (Ilmavirta et al., 2019)
Non-Abelian scattering	Above + gauge constraint	Unique recovery up to gauge
Generic matrix weights	Anosov or strictly convex, no conjugate points	Generic injectivity (Cekić et al., 2021)

The matrix-weighted X-ray transform thus provides the analytic core for a broad spectrum of inverse boundary and tomography problems, with the interplay of geometric, analytic, microlocal, and statistical techniques yielding the current landscape of injectivity, stability, and practical inversion methodologies.