Compton Data Space in Scattering Analysis

• Compton Data Space is a coordinate system that transforms raw detector hits into reduced representations tailored to specific inverse problems in scattering research.
• It enables the reconstruction of gamma-ray images and momentum densities by mapping energies, positions, and interaction geometries into tractable measurement spaces.
• Calibration, sequencing, and background rejection techniques refine this data space, enhancing imaging resolution and analytical accuracy across diverse applications.

Compton Data Space denotes the coordinate system in which measurements produced by Compton scattering are represented for reconstruction or inference. In gamma-ray astronomy, raw detector hits are reduced to quantities such as scattered-photon direction and Compton scatter angle; in electronic-structure studies, the corresponding data are directional Compton or magnetic Compton profiles, i.e. one-dimensional projections of a three-dimensional momentum density; in Compton scattering tomography, the measured object is a spectrum indexed by detected energy and source–detector geometry (Zoglauer et al., 2021, Kierans et al., 2022, Ernsting et al., 2014, Gödeke et al., 2022). Taken together, these uses suggest that Compton Data Space is not a single universal formalism but a family of reduced measurement representations tailored to a particular inverse problem.

1. Scope and domain-specific usage

Across the cited literature, Compton Data Space appears either as an explicit term or as an operational equivalent. In each case, the common feature is a mapping from detector observables to a representation in which the forward response is geometrically or statistically tractable.

Domain	Primary measured space	Operational Compton data space
Gamma-ray telescopes	Interaction energies and positions	Scattered-photon direction plus Compton angle
Detector calibration	Channel vectors, coincidence tuples, timing streams	Corrected deposited-energy or 3D interaction-coordinate space
Condensed-matter Compton scattering	Momentum density or spin-resolved momentum density	Directional Compton or magnetic Compton profiles
Compton scattering tomography	Energy-resolved source–detector spectra	Sampled spectral arrays indexed by energy and geometry

In COSI, for example, the raw Compton data space is the full interaction record

$E_1, x_1, y_1, z_1, \ldots, E_N, x_N, y_N, z_N,$

whereas the minimum Compton data space for imaging is a 3-dimensional space with coordinates given by the celestial longitude and latitude of the scattered gamma-ray direction and the reconstructed Compton scatter angle (Zoglauer et al., 2021). In the Compton-telescope formalism summarized by Boggs and collaborators, detector space is reduced to Compton Data Space coordinates $(\chi,\psi,\varphi)$, often with total energy treated as a fourth dimension, and imaging is then the inversion of a response mapping from sky coordinates to this reduced space (Kierans et al., 2022).

A related but distinct usage appears in condensed matter. There, the measured object is not a sky-cone constraint but a directional projection of the electron momentum density,

$$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$

or, in the magnetic channel,

$$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$

so the data space is native to momentum space rather than source localization (Ernsting et al., 2014).

2. Gamma-ray telescope Compton data space

For a gamma-ray Compton telescope, the fundamental detector observables are interaction energies and interaction locations. A modern event is therefore initially a set of measured hits,

$$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$

sometimes augmented by timing or recoil-electron-track information (Kierans et al., 2022). After sequencing, the total deposited energy estimates the incident photon energy, the first two ordered interaction positions define the scattered-photon direction, and the partition of energy between the first interaction and the remainder determines the Compton scatter angle.

For a two-site event, the practical reconstruction relation is

$$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$

with $E_0=E_1+E_2$ and $E_{\rm scat}=E_2$ (Kierans et al., 2022). In ComPair, the same event model is written operationally in terms of the first scatter position, the scattered-photon absorption position, the recoil-electron energy, the scattered-photon energy, and, when available, the recoil-electron direction. Without electron tracking, these observables yield an event circle on the sky; with tracking, the circle is reduced to an event arc (Moiseev et al., 2015).

The reduced coordinate system is historically associated with COMPTEL-style analyses. In the INSPIRE study, the three-dimensional data space is defined explicitly by the scattering direction $(\chi,\psi)$ and the scattering angle $\theta$, and the paper states that the gamma-ray pattern from a celestial coordinate $(\chi,\psi,\varphi)$0 lies on a surface in $(\chi,\psi,\varphi)$1 space, where the apex of the cone is located at $(\chi,\psi,\varphi)$2 and the cone’s semi-angle is $(\chi,\psi,\varphi)$3; incompletely absorbed events populate the interior of the cone (Kataoka et al., 18 Sep 2025). This makes precise the distinction between a point in Compton Data Space and a cone, circle, or arc in source space.

The COSI pipeline adopts the same structural idea in a different notation. Its minimum imaging CDS is $(\chi,\psi,\varphi)$4, where $(\chi,\psi,\varphi)$5 and $(\chi,\psi,\varphi)$6 are the celestial longitude and latitude of the scattered gamma-ray direction and $(\chi,\psi,\varphi)$7 is the reconstructed Compton scatter angle (Zoglauer et al., 2021). This coordinate reduction is central because the detector does not form an image directly; rather, it accumulates a distribution in CDS that must be forward-modeled or inverted.

3. Sequencing, resolution, and event selection

Once a Compton event has been mapped into a reduced space, the main technical questions are sequencing, angular consistency, and background rejection. For compact multi-site instruments, ordering is not directly measured and is commonly inferred by comparing kinematic and geometric Compton angles. The quality factor used in Compton Kinematic Discrimination is

$(\chi,\psi,\varphi)$8

and the sequence with the lowest $(\chi,\psi,\varphi)$9 is taken as the most probable (Kierans et al., 2022).

The standard one-dimensional source-consistency coordinate is the Angular Resolution Measure. In the INSPIRE analysis,

$$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$0

and the adaptive source-selection rule is

$$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$1

with $$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$2 (Kataoka et al., 18 Sep 2025). The paper presents this not as a replacement for full response modeling in $$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$3 space, but as a direct and intuitive source-centered slice through that space for known point-like sources.

Electron tracking enlarges the event space beyond the conventional cone variables. In the Electron Tracking Compton Camera, the recoil-electron track, its $$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$4–$$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$5 signature, its containment, and the projection angle $$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$6 of the initial recoil vector onto the event-circle plane become additional event coordinates. The paper argues that this changes the topology of the reconstruction problem from an annulus on the sky to a localized patch and reports that, after all event screenings, background contamination is reduced by two orders of magnitude (Hamaguchi et al., 2019).

A more radical modification of event space appears in the multi-column Compton camera of stacked Si pixel sensors. There, standard Compton reconstruction is supplemented by a directional shadow modulation of first-scatter positions, represented by the template equation

$$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$7

where $$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$8 is the model histogram for injection-angle identification number $$J(p_z)=\iint \left[\rho_\uparrow(\mathbf p)+\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$9 and $$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$0 is the measured histogram. In simulation, an MCCC with 1 m column height, 0.5 mm pixel size, 100 layers, and 10 columns for the 1-D direction can distinguish two sources separated by $$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$1 with $$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$2M Compton-reconstructed events (Fukazawa, 5 Mar 2025). This suggests a hybrid data space in which the ordinary Compton cone is refined by a detector-geometry-dependent first-scatter response.

4. Detector-coordinate calibration and measurement-space correction

Compton Data Space is often determined as much by calibration as by kinematics. In POLAR, the detector’s measurement object is a module-level 64-component energy vector, and the forward model is written explicitly as

$$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$3

where the response matrix $$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$4 combines scintillation/light collection, crosstalk, and gain/non-uniformity (Xiao et al., 2015). In this formulation, detector data space is the recorded channel vector, while corrected Compton analysis is performed in the reconstructed deposited-energy space.

The associated in-flight calibration of POLAR uses four weak $$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$5Na sources and exploits the $$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$6 Compton edge from the collinear $$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$7 annihilation photons. The calibration proceeds through geometrically selected two-hit events, response-convolved Monte Carlo fitting, and temperature- and high-voltage-dependent conversion factors; the paper reports a mean relative temperature coefficient of $$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$8 per degree (Xiao et al., 2017). This is a detector-specific realization of Compton data space in which raw ADC amplitudes and trigger coincidences are mapped to calibrated per-bar energy deposits and hit topologies.

In COSI, the analogous issue is not channel crosstalk but missing depth. The detector geometry directly provides two coordinates from orthogonal strip crossings, while the third interaction coordinate must be inferred from the charge collection time difference

$$J_{\rm mag}(p_z)=\iint \left[\rho_\uparrow(\mathbf p)-\rho_\downarrow(\mathbf p)\right]\,dp_x\,dp_y,$$9

The calibration is operationally per-pixel affine-corrected rather than a single universal global map, and the paper reports that more than $$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$0 of pixels have depth resolution below $$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$1 FWHM at $$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$2 and below $$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$3 FWHM at $$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$4 (Rogers et al., 2 Feb 2026). Although that work does not introduce a formalism for Compton Data Space itself, it provides one of the coordinates required to construct it accurately.

A compact laboratory CZT imager makes the same point at smaller scale. Its practical list-mode space is the coincidence tuple

$$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$5

which is calibrated to

$$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$6

From these quantities, the event is transformed into a cone, or equivalently an event circle on the sky (Arya et al., 2024). This makes explicit that detector-level Compton data space is often a calibrated coincidence space before it is an image-space manifold.

5. Momentum-space Compton data space in condensed matter

In condensed-matter physics, Compton Data Space refers to directional projections of the electron momentum density rather than event cones. The central object is the electron momentum density

$$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$7

evaluated in extended momentum space with $$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$8 (Ernsting et al., 2014). The measured Compton profile is the double projection of this density along the scattering direction, and the magnetic Compton profile isolates the spin difference.

In ferromagnetic Fe and Ni, this reduced space is used to analyze not only structural anisotropy but also correlation-induced redistribution of momentum-space weight. The correlation diagnostic is the directional difference

$$\{(E_1,\vec r_1),(E_2,\vec r_2),\dots,(E_N,\vec r_N)\},$$9

and its second moment

$$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$0

The paper’s key conclusion is that local dynamical correlations can manifest as anisotropic effects in momentum space because they act on an underlying band structure and bonding geometry that are highly directional (Chioncel et al., 2014).

A common misconception is that a nonzero magnetic Compton profile necessarily implies net magnetization. The PbTiO$$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$1 study shows otherwise. There, the magnetic Compton profile is

$$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$2

and time reversal requires

$$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$3

so the integrated moment vanishes even though the directional profile is nonzero and antisymmetric (Bhowal et al., 2021). The effect is attributed to a broken-inversion, time-reversal-symmetric momentum-space spin asymmetry associated with a $$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$4-space magnetoelectric toroidal moment.

Sparse directional sampling has also led to an explicitly inverse-problem formulation of this data space. In the compressed-sensing reconstruction of the bcc-Li Fermi surface, the object space is the three-dimensional momentum density $$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$5, while the measurement space is a set of directional profiles $$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$6; after Fourier transformation, the discretized relation is

$$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$7

The reconstruction uses an $$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$8 penalty on discrete momentum gradients, together with non-negativity and the electron-count sum rule, and the paper reports recovery of the Fermi surface from as few as 14 directions (Otsuki et al., 2022). This is a particularly clear example of Compton Data Space as a tomographic measurement space.

6. Spectral-tomographic and computational formulations

In Compton scattering tomography, the measured data are energy-resolved source–detector spectra,

$$\cos\varphi = 1 - \frac{m_ec^2}{E_2} + \frac{m_ec^2}{E_1 + E_2},$$9

with $E_0=E_1+E_2$0 ballistic, $E_0=E_1+E_2$1 first-order scattered, and higher $E_0=E_1+E_2$2 corresponding to multiple scattering (Gödeke et al., 2022). For first-order scattering, the contributing object points lie on the locus

$E_0=E_1+E_2$3

which is a pair of circular arcs in 2D and a spindle torus in 3D. After sampling in energy and source–detector geometry, the semi-discrete data space is a finite array in $E_0=E_1+E_2$4, and the paper develops both RESESOP and Deep Image Prior reconstructions to accommodate forward-model uncertainty (Gödeke et al., 2022).

A closely related computational viewpoint appears in real-time gamma-ray Compton imaging. For monolithic-crystal cameras, the raw space is a time-ordered detector-hit stream. Event identification is reformulated through local time differences

$E_0=E_1+E_2$5

yielding a vectorizable and parallelizable clustering rule (Gameiro et al., 2024). After position and energy reconstruction, each accepted event becomes a list-mode Compton-cone datum, and imaging is implemented as an $E_0=E_1+E_2$6 backprojection with precomputed geometry and angular pruning (Gameiro et al., 2024).

Taken together, these formulations suggest that Compton Data Space is best understood as an intermediate inverse-problem space. Its purpose is neither merely to store raw detector outputs nor merely to display final images, but to place the measurements in coordinates where the response function, admissible event manifold, and uncertainty structure become tractable. That role is common to gamma-ray astronomy, Compton tomography, detector calibration, and momentum-density reconstruction, even though the specific coordinates differ substantially across those fields.