Attenuated Ray Transforms

• Attenuated ray transforms are defined as integrals along curves with an exponential attenuation factor that encodes physical decay or absorption.
• The method employs parallel transport, holomorphic integrating factors, and pseudodifferential operators to establish inversion formulas and stability in tomographic reconstruction.
• Applications include SPECT, PET, and tensor tomography, with the analysis focusing on injectivity, gauge obstructions, and stability under various geometric settings.

The attenuated ray transform refers to an integral transform that generalizes the classical (unattenuated) ray or geodesic X-ray transform by incorporating a multiplicative attenuation along the ray or curve of integration. Broadly, it seeks to recover information about functions, tensor fields, or sections over a manifold or domain from their integrals along curves, weighted by an attenuation factor encoding decay or absorption (e.g., physical absorption or parallel transport). This operator and its generalizations underpin foundational results in inverse problems, tomography, and geometric analysis.

1. Mathematical Definition and Geometric Context

On a smooth compact Riemannian manifold $(M,g)$ (possibly with boundary), the prototypical attenuated geodesic ray transform for a function $f: M \to \mathbb C$ and attenuation $a: M \to \mathbb C$ is given by

$$I_a[f](x,v) = \int_0^{\tau(x,v)} f(\gamma_{x,v}(t)) \exp\left(-\int_0^t a(\gamma_{x,v}(s)) ds\right) dt, \qquad (x,v) \in \partial_+ SM,$$

where $SM$ is the unit sphere bundle, $\partial_+ SM$ is the inflow boundary, $\gamma_{x,v}(t)$ is the geodesic through $x$ in direction $v$, and $\tau(x,v)$ is the first exit time from $M$ (Salo et al., 2010, Assylbekov et al., 2016).

More generally, for sections of vector bundles, one considers matrix-valued attenuation $A$ (a connection) and a Higgs field $\Phi$, leading to the transform

$$I_{A, \Phi}[f](x,v) = \int_0^{\tau(x,v)} U_{A,\Phi}(x,v, t)^{-1} f(\gamma_{x,v}(t)) dt,$$

where $U_{A, \Phi}$ is the parallel transport operator along $\gamma_{x,v}$ with attenuation by $(A, \Phi)$ (Paternain et al., 2011, Ainsworth, 2012, Paternain et al., 2016). For tensor tomography, the integrand is a symmetric tensor contracted along the direction $v$.

Extensions exist for magnetic flows, Gaussian thermostats, tensor fields, and generalized ray transforms of order $k$, with higher-order polynomial or angular moment weights (Derevtsov et al., 2019, Krishnan et al., 2018).

2. Injectivity, Gauge Obstructions, and Stability

The central analytic question is when the attenuated ray transform is injective: does $I_a[f]=0$ imply $f=0$ (or modulo natural obstructions)? For simple manifolds (strictly convex boundary, no conjugate points, nontrapping), canonical results show:

• For scalar attenuation, $I_a$ is injective on functions: if $I_a[f]=0$ for all boundary directions, then $f=0$ (Salo et al., 2010, Assylbekov, 2018).
• For vector or tensor valued fields, $I_{A, \Phi}$ is injective modulo "gauge" obstructions: if $I_{A,\Phi}[f]=0$, then $f$ is of the form $(A + \Phi)p$ for some section $p$ vanishing on the boundary (Paternain et al., 2011, Ainsworth, 2012, Guillarmou et al., 2015, Paternain et al., 2016).
• For tensor fields, the kernel consists of "potential tensors," i.e., symmetrized covariant derivatives of lower order tensors vanishing on the boundary (Ainsworth, 2012, Krishnan et al., 2018, Monard, 2017).

The kernel is thus controlled by natural geometric obstructions, which coincide with gauge equivalence classes. The presence of conjugate points (lack of simplicity) invalidates injectivity and stability, leading to intrinsic “microlocal” artifacts (Holman et al., 2017).

Stability is quantified via ellipticity of the normal operator $N_a = I_a^* I_a$ and often involves $L^2$ or Sobolev estimates relating the unknown function (or its solenoidal part) to the measured data (Salo et al., 2010, Assylbekov, 2018).

3. Attenuation Mechanisms: Connections, Higgs Fields, and Matrix Weights

In advanced settings, attenuation is dictated by connections $(A)$ and Higgs fields $(\Phi)$ on Hermitian or general vector bundles. Along each geodesic, attenuation is encoded by parallel transport, leading to the first-order ODE: $$\frac{d}{dt}U + (A(\dot\gamma) + \Phi(\gamma))U = 0, \qquad U(0) = \text{Id}$$ The attenuated transform then integrates the section against the inverse parallel transport (Paternain et al., 2011, Ainsworth, 2012, Paternain et al., 2016). Injectivity holds up to gauge transformations: two pairs $(A,\Phi)$ are indistinguishable if their data are related by a unitary gauge $Q$, with $(A', \Phi') = (Q^{-1} d_A Q, Q^{-1} \Phi Q)$ and $Q|_{\partial M} = \text{Id}$.

Similar constructions apply in the presence of magnetic fields, with the generator adapted to the magnetic flow (Ainsworth, 2012, Ainsworth et al., 2013), or for Gaussian thermostat flows with additional external fields (Assylbekov et al., 2021).

4. Inversion Formulae and Computational Methods

On simple surfaces, explicit inversion schemes are available. In dimension two, holomorphic/antiholomorphic integrating factors (solutions to $Xw=-a$ odd in the fiber variable) are constructed using the fiberwise Hilbert transform, which reduces the attenuated transport equation to an unattenuated form (Salo et al., 2010, Assylbekov et al., 2016, Hoell et al., 2010). For instance,

$e^{-w}(X+a)e^{w} = X$

enables direct inversion via pseudodifferential operator theory and Poisson integral formulae, culminating in filtered back-projection expressions analogous to those in SPECT and PET (Hoell et al., 2010, Monard, 2017).

For tensor tomography and higher moment transforms, Bukhgeim’s $A$-analytic framework applies, transforming the inversion to a hierarchy of boundary value problems for Beltrami-type systems. Inversion proceeds via sequential Cauchy and Pompeiu-type integrals and triangular recurrence (Fujiwara et al., 2023, Bhardwaj et al., 4 May 2025). For partial data (e.g., boundary restriction to an arc), finite Hilbert transform techniques and analytic continuation enable local recovery (Fujiwara et al., 2017).

5. Range Characterizations and Projection Operators

The range of the attenuated ray transform can be characterized implicitly via boundary operators built from the fiberwise Hilbert transform, the scattering relation, and the parallel transport data (Paternain et al., 2013, Assylbekov et al., 2016, Ainsworth et al., 2013). For example, the operator

$P_a = B_a H Q_a$

maps preimages in the boundary data space to the range, with specific conditions for functions, one-forms, or tensors determined by Fourier mode constraints and solvability of associated transport equations. Decomposition into orthogonal subranges (functions, solenoidal one-forms, holomorphic/antiholomorphic modes) enables explicit inversion and data denoising (Assylbekov et al., 2016).

For transforms with connections and Higgs fields, the range is described by a sum of contributions from the boundary operator acting on smoothly extendable data and the attenuated transform of $A$-harmonic forms, reflecting the gauge-invariant subspace (Ainsworth et al., 2013).

6. Extensions: Polynomial and Angular Moments, Generalized ART

Generalized attenuated ray transforms (ART) incorporate higher polynomial weights or angular moments. The stationary $k$-th moment ART reads

$$u_k(x, \xi) = \int_0^\infty s^k \exp\left(- \int_0^s \alpha(x-\sigma \xi, \xi) d\sigma \right) f(x-s\xi, \xi) ds$$

with attenuation $\alpha(x,\xi)$ possibly complex-valued (Derevtsov et al., 2019). ARTs satisfy higher-order inhomogeneous differential equations,

$$\frac{1}{k!} (\mathcal{H} + \alpha)^{k+1} u_k(x, \xi) = f(x, \xi)$$

where $\mathcal{H}$ is the transport derivative. Moment identities and divergence relations allow recovery of tensor fields and study of associated tomography problems. Stability and uniqueness are proven via boundary value theory and analytic continuation (Derevtsov et al., 2019).

7. Applications and Significance

Attenuated ray transforms govern mathematical models for:

• Medical imaging: SPECT, PET, Doppler and polarization tomography (Salo et al., 2010, Monard, 2017, Paternain et al., 2016)
• Quantum state and polarization tomography (Paternain et al., 2016)
• Inverse boundary value problems with gauge fields (Ainsworth, 2012, Ainsworth et al., 2013)
• Tensor tomography for vector and higher rank field recovery (Krishnan et al., 2018, Assylbekov, 2018)
• Analysis of inverse problems with partial or incomplete data (Fujiwara et al., 2017, Fujiwara et al., 2023)
• Applications in physical optics, integral geometry, and wave propagation (Derevtsov et al., 2019)

The theoretical foundation guarantees both uniqueness (up to natural gauge obstructions) and stability in reconstruction, enables explicit inversion in favorable settings, and quantifies the effects of geometric complexity and boundary conditions. Generalizations support tensor field recovery and new classes of inverse problems in non-Euclidean and physically anisotropic media.

---

Table: Attenuated Ray Transform Settings and Main Results

Manifold Context	Attenuation Type	Injectivity Modulo	Key References
Simple surface	Scalar function	Trivial kernel	(Salo et al., 2010, Assylbekov et al., 2016)
Simple surface/bundle	Connection + Higgs field	Gauge transform	(Paternain et al., 2011, Ainsworth, 2012)
Magnetic/thermostat flows	Connection + Higgs field	Gauge transform	(Ainsworth, 2012, Assylbekov et al., 2021)
Higher-dimensional manifold	Matrix weights	Gauge transform	(Paternain et al., 2016)
Generalized ART (moments)	Polynomial/exponential	Boundary/initial	(Derevtsov et al., 2019, Fujiwara et al., 2023)

---

Editor’s term: Solenoidal-injectivity—injectivity modulo potential/gauge fields.

---

For rigorous mathematical proofs, explicit inversion algorithms, and implementation details, see (Salo et al., 2010, Paternain et al., 2011, Ainsworth, 2012, Assylbekov et al., 2016, Paternain et al., 2016, Krishnan et al., 2018, Derevtsov et al., 2019, Fujiwara et al., 2023), and the range characterization via boundary operators in (Paternain et al., 2013, Ainsworth et al., 2013).