Ray Space Transformations Overview

• Ray space transformations are operators that integrate functions or tensor fields along curves or geodesics, generalizing classical Radon and X-ray transforms.
• They incorporate various approaches including weighted, momentum, and matrix-valued transforms using back-projection, Fourier methods, and PDE techniques for inversion.
• The rigorous study of their kernel, range, and stability underpins applications in tomography, optics, and inverse problems across diverse geometric settings.

A ray space transformation is any operator that maps functions or tensor fields defined on a base space (such as Euclidean, Riemannian, or pseudo-Riemannian manifolds) into data defined on the space of rays—typically, lines or geodesics parameterized by direction and base point—via integration along those rays. These transformations generalize the classical Radon and X-ray transforms to a wide array of curve families, geometric backgrounds, weights (scalar, matrix, or polynomial), and field types (scalars, tensors).

1. Formal Definitions and Principal Examples

A prototypical ray transform is the geodesic X-ray transform $I_m$ on a Riemannian manifold $(M,g)$, defined for smooth symmetric $m$-tensor fields $f$ by

$$I_m[f](\gamma)=\int_\gamma f(\dot\gamma(t),\dots,\dot\gamma(t))\,dt,$$

where $\gamma$ ranges over a family of rays, commonly geodesics or lines in flat space (Guillarmou et al., 2019). For $m=0$, this reduces to the scalar Radon transform. The manifold of rays is often identified with a suitable quotient of the tangent bundle (e.g., $TS^{n-1}$ in Euclidean $\mathbb{R}^n$).

Key generalizations include:

• Momentum Ray Transforms $I^k$ on symmetric tensors: integrating with monomial weights $t^k$,

$$(I^k f)(x,\xi) = \int_{-\infty}^\infty t^k \langle f(x + t\xi),\,\xi^m\rangle\,dt,\qquad (x,\xi)\in TS^{n-1}$$

synthesizes all moments up to order $m$ and parameterizes the full data on ray space (Krishnan et al., 2018, Krishnan et al., 2019).

• Weighted and Mixing Ray Transforms: Integration with direction-dependent weights (scalar, matrix, or intertwining automorphisms), crucial for applications in inverse problems with anisotropy, attenuation, or elasticity (Ilmavirta et al., 2019, Ilmavirta et al., 2020).
• Light Ray Transforms (LRT): Integration along null geodesics in Lorentzian or pseudo-Euclidean spaces, central in tomography in spacetime and hyperbolic inverse problems (Bhattacharyya et al., 21 Oct 2025, Agrawal et al., 19 Feb 2025).
• Windowed Ray Transforms: Incorporate a “window” function $h$ into the integration, generalizing analytic-signal extensions and facilitating invertibility for broad function classes (Moon, 2013).

2. Geometric and Analytical Structures

Ray space transformations inherit profound geometric structure:

• The space of rays is a homogeneous manifold, often realized as $TS^{n-1}$ or a manifold of geodesics (Krishnan et al., 2018, Guillarmou et al., 2019).
• In Riemannian settings, fiber bundles and contact structures naturally arise, and on orientable surfaces, the unit-sphere bundle $SM$ is paramount (Hoop et al., 2018, Ilmavirta et al., 2020).
• For pseudo-Euclidean metrics, the null directions parametrizing the LRT form quadratically defined submanifolds (light cones), complicating the microlocal analysis of the transforms (Agrawal et al., 19 Feb 2025).

Crucial operator-theoretic properties:

• Range conditions: The range of a ray transform is characterized by highly nontrivial symmetry and moment constraints, with parity properties and John-type PDE systems in dimensions $n\geq 3$, and by generalized Gelfand-Helgason-Ludwig integral conditions in the plane (Krishnan et al., 2019).
• Kernel structure and gauge freedoms: The kernel often consists of potential (exact, or gauge-type) tensors—i.e., images of symmetrized covariant derivatives ($d^s$), contraction operators ($\lambda$), or more complex algebraic constructs depending on the background geometry and field rank (Hoop et al., 2018, Ilmavirta et al., 2020).
• Invariance and algebraic relations: Weighted and mixed transforms correspond to compositions with invertible automorphisms on the field space, yielding algebraic reductions of kernel and stability properties across wide transform classes (Ilmavirta et al., 2020).

3. Inversion Theory and Stability Estimates

Many ray transforms admit explicit or constructive inversion formulas under suitable conditions:

• For the classical and momentum ray transforms in Euclidean space, one employs back-projection, Fourier-analytic, or filtered inversion methods, fully reconstructing tensor fields from all moments up to order $m$ (Krishnan et al., 2018).
• In Sobolev spaces, the Reshetnyak formula yields explicit isometries between function spaces before and after transformation, proving stability and allowing norm control and error propagation analysis (Krishnan et al., 2021, Krishnan et al., 2018).
• For weighted and matrix-valued transforms, inversion typically reduces to layered or foliation-based arguments in the underlying manifold, provided the weight is injective; convexity and foliation conditions control the propagation of uniqueness and stability (Ilmavirta et al., 2019).
• For transforms on asymptotically conic or noncompact manifolds, advanced microlocal techniques (e.g., semiclassical and 1-cusp pseudodifferential analysis) establish invertibility and stable reconstruction in weighted Sobolev scales, especially near infinity (Vasy et al., 2022).

A summary of classically appearing inversion schemes is given below:

Transform Type	Inversion Approach	Reference
Radon/X-ray (Euclidean, $m=0$)	Back-projection, filtered Fourier	(Krishnan et al., 2018)
Momentum Ray ($m>0$)	Iterative moment and PDE approach	(Krishnan et al., 2019)
Weighted geodesic	Layer-stripping, tangent-cone trick	(Ilmavirta et al., 2019)
LRT/Momentum Light Ray	Fourier-slice, moment hierarchy	(Bhattacharyya et al., 21 Oct 2025)
Asymptotically conic manifolds	1-cusp pseudodifferential parametrix	(Vasy et al., 2022)

4. Kernel, Range Characterization, and Injectivity

The structure of the kernel and range is fundamental in both the uniqueness theory and in algorithmic inversion:

• In Euclidean space, the kernel of the $m$-tensor ray transform consists of symmetrized derivatives (“potential tensors”) vanishing at infinity, and injectivity up to this gauge is classical (Krishnan et al., 2018, Krishnan et al., 2021).
• For momentum ray transforms, injectivity of the joint map $(I^0,\dots,I^m)$ holds on Schwartz $m$-tensor fields, with the range described by systems of John equations (PDEs of order $2(m+1)$), and in the plane by explicit moment conditions (Krishnan et al., 2019).
• For mixed, transverse, and weighted ray transforms, kernel equivalence follows from bundle endomorphism algebra: all such transforms are reducible to the base geodesic transform via invertible mixing, so injectivity and stability propagate between transform classes (Ilmavirta et al., 2020).

In the setting of Lorentzian or pseudo-Euclidean signature, the light ray transform has an elliptic normal operator whose symbol is singular on the light cone but elliptic elsewhere, yielding robust inversion and stability off the degenerate locus (Agrawal et al., 19 Feb 2025).

5. Modern Applications: Tomography, Optics, and Physics

Ray space transformations are central in modern integral geometry and tomography:

• Computed Tomography, Geophysics: Recovery of spatial media properties from line or geodesic integrals is a cornerstone of medical imaging and seismic inversion (Krishnan et al., 2018, Guillarmou et al., 2019).
• Optics and Beamline Modeling: Precise ray tracing for complex conic surfaces in synchrotron X-ray optics employs analytic transforms between global coordinate systems and surface-anchored local frames, necessitating exact transformations of ray data and normals (Rio et al., 2024).
• Inverse Problems in Pseudo-Riemannian and Lorentzian Frameworks: Light ray and momentum transforms model measurements in spacetime, with applications in inverse boundary problems for wave propagation and in gravitational lensing studies (Bhattacharyya et al., 21 Oct 2025, Agrawal et al., 19 Feb 2025).
• Quantum and Statistical Physics: Ray (or projective) space representations of canonical transformations implement symmetries in infinite-dimensional fermionic Fock spaces, with foundational implications in BCS theory and quantum field theory (Kupsch, 2013).

6. Notable Theoretical Advances and Open Problems

Major theoretical advances include:

• Complete range characterization for momentum transforms on the Schwartz space (Krishnan et al., 2019).
• Injectivity with minimal regularity and matrix-valued weights for piecewise-constant data on foliated manifolds (Ilmavirta et al., 2019).
• Unification of mixing-type transforms by algebraic conjugation, illuminating kernel and range equivalence in a variety of geometric contexts (Ilmavirta et al., 2020).
• Sharp Sobolev stability and higher-order Reshetnyak-type isometries, essential for quantifying error and stability in high-order tensor tomography (Krishnan et al., 2021).
• Algorithms for analytic continuation and local reconstruction under partial data conditions, especially relevant in practical limited angle tomography (Bhattacharyya et al., 21 Oct 2025).

Open directions involve:

• Extension of kernel and injectivity results for higher-rank tensors in dimensions $n\geq 3$ (particularly for mixed transforms).
• Analysis of stability and inversion in the presence of conjugate points, trapping, or nontrivial topology.
• Adaptation of Fourier integral operator theory and microlocal analysis to singular and degenerate settings (such as the light cone in pseudo-Euclidean space).

7. Summary Table of Transform Properties

Transform	Target Space	Injectivity	Stability	Key References
Scalar X-ray	Functions on rays	Up to potentials	Sobolev isometry	(Krishnan et al., 2018, Krishnan et al., 2021)
Momentum Ray	Vectors of moments	Full (joint) injectivity	Range by John PDEs	(Krishnan et al., 2018, Krishnan et al., 2019)
Matrix-weighted	Matrix-valued data	Piecewise constants	Convex foliation needed	(Ilmavirta et al., 2019)
Light Ray	Data on null lines	Elliptic away from cone	$H^1$/$H^1_{\log}$ equivalence	(Agrawal et al., 19 Feb 2025)
Mixed/Mixing	Automorphism images	Equivalence via algebra	Kernel/range transfer	(Ilmavirta et al., 2020, Hoop et al., 2018)
Asymptotically conic	Geodesic data at infinity	Gaussian decay	1-cusp pseudodifferential	(Guillarmou et al., 2019, Vasy et al., 2022)

Ray space transformations constitute a unifying analytic language for integrating, reconstructing, and manipulating geometric and physical data along families of curves, critical in both theoretical mathematics and applied sciences such as imaging, optics, and inverse problems. Their continued study reveals deep connections among geometry, analysis, and modern computational methods.