QuaTR: Quantum Radiative Transfer Toolkit

• QuaTR is a Python toolkit for quantum radiative transfer that uses quantum lattice Boltzmann methods to address collisionless transport in astrophysical simulations.
• The toolkit introduces a novel angular redistribution operator to reduce lattice-induced directional bias while preserving exponential memory compression via amplitude encoding.
• It integrates classical simulation workflows with Qiskit-based quantum circuit construction to validate isotropy through canonical tests such as isotropic source and opaque cloud shadow scenarios.

QuaTR, styled in the associated publication as QuaRT (Quantum Radiative Transfer), is a Python toolkit for the exploration of quantum algorithms—specifically quantum lattice Boltzmann methods (QLBM)—for radiative transfer problems in astrophysics and cosmology (Devkota et al., 15 Nov 2025). Its motivating premise is that radiation transport is often among the heaviest components of galaxy- and universe-scale simulations in both runtime and memory, especially in regimes with large dimensionality and long mean free paths. The toolkit targets non-scattering media typical of cosmological reionization and stellar radiation transport, and its central methodological contribution is a novel angular redistribution step within a QLBM that reduces lattice-induced directional bias while retaining the exponential memory compression associated with amplitude encoding.

1. Scope and problem class

QuaTR is designed for radiative transfer problems in which transport is largely collisionless and sources are isotropic, such as stellar and early-universe settings (Devkota et al., 15 Nov 2025). In these settings, classical discrete-angle methods can imprint the geometry of the underlying lattice onto the solution. The problem is especially acute when isotropic sources under stars must be accurately resolved in non-scattering media, because there is no physical scattering mechanism to smooth angular artifacts.

The toolkit therefore focuses on a restricted but astrophysically important regime. It does not attempt to be a general-purpose scattering solver. Instead, it addresses a class of radiative transfer workloads for which a lattice Boltzmann discretization of propagation directions is natural, and in which improved isotropy can materially affect the physical plausibility of shadows, spherical emission profiles, and beam superposition.

This specialization also determines the toolkit’s validation strategy. QuaTR includes canonical tests—an isotropic source, an opaque cloud shadow, and crossing beams—chosen to assess correctness, symmetry preservation, and the absence of spurious coupling in collisionless transport.

2. Governing equations and angular representation

At the level of continuum physics, QuaTR is organized around the radiative transfer equation. In its simplest one-dimensional form along a path-length coordinate $s$, neglecting scattering, the intensity satisfies

$$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$

In the more general angle- and position-dependent form, including scattering with phase function $\sigma_\nu$,

$$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$

QuaTR emphasizes the regime $\sigma_\nu \approx 0$, which is described as common in cosmological reionization and stellar radiation transport over large mean free paths (Devkota et al., 15 Nov 2025).

Methodologically, the toolkit does not use a spherical-harmonics expansion of the form

$$I(\mathbf{x}, \hat{\mathbf{n}}, \nu) = \sum_{\ell=0}^{\infty}\sum_{m=-\ell}^{\ell} a_{\ell m}(\mathbf{x}, \nu)\, Y_\ell^m(\hat{\mathbf{n}}).$$

Instead, it adopts a lattice Boltzmann discretization in which the angular dependence is represented by a finite set of discrete velocity directions $i$, with associated directions $\mathbf{e}_i$ and weights $w_i$. The radiation field is represented by angular populations $f_i$ at each lattice site. This choice aligns the transport problem with QLBM constructions and makes the direction space explicit as a quantum register.

3. Lattice Boltzmann formulation and angular redistribution

In QuaTR’s baseline lattice Boltzmann update, one time step $$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$0 is written as

$$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$1

Here the first term on the right is streaming, the second is relaxation toward a local equilibrium, and the third injects emission. For non-scattering, isotropic sources, the equilibrium is typically chosen as $$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$2, with $$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$3 set by local emission $$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$4, and a simple explicit source treatment is $$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$5.

The principal innovation of QuaTR is the insertion of an angular redistribution operator between streaming and relaxation (Devkota et al., 15 Nov 2025). In non-scattering radiative transport, there is no angular redistribution due to collisions, so anisotropy introduced by the discrete directional stencil can persist. QuaTR addresses this by mixing angular populations locally according to

$$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$6

with $$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$7 described abstractly as a nonnegative, typically doubly-stochastic matrix that preserves positivity, conserves total intensity $$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$8, and drives the local angular state toward isotropy.

The isotropic target state on a lattice with $$\frac{\mathrm{d}I_\nu}{\mathrm{d}s} = -\kappa_\nu I_\nu + j_\nu .$$9 discrete directions is

$\sigma_\nu$0

and the toolkit quantifies angular uniformity using normed deviations such as

$\sigma_\nu$1

QuaTR supports a “global” redistribution strategy that leverages knowledge of expected isotropy, for example around isotropic point sources in non-scattering media. Crucially, the authors report that this redistribution can be applied with no increase in computational complexity relative to the baseline QLBM per-step circuit construction. The reported outcomes are qualitative rather than tabulated numerically, but they include visibly reduced lattice imprint in isotropic-source tests and symmetric shadows without spurious directional artifacts.

4. Quantum encoding and circuit realization

QuaTR maps the lattice Boltzmann radiative-transfer state into a quantum state by amplitude encoding the joint position–direction populations (Devkota et al., 15 Nov 2025). Conceptually,

$\sigma_\nu$2

with amplitudes $\sigma_\nu$3 proportional to $\sigma_\nu$4, subject to normalization.

The streaming step is implemented as a conditional shift on the position register,

$\sigma_\nu$5

Absorption, relaxation, and emission are local operations on the angular register at each site. The summary does not specify the exact technique used to implement the nonunitary relaxation map, but it notes that such a map may be approximated through block-encoding and linear-combination-of-unitaries constructions, by embedding into a larger state space with ancillas, or by variational circuits calibrated to reproduce the target linear map over one step. In the released toolkit, the qlbm_circuits module provides circuit constructors for collision or relaxation and source injection in one, two, and three dimensions.

The angular redistribution is realized quantum mechanically as an operation on the direction register,

$\sigma_\nu$6

applied uniformly or problem-adaptively across the domain.

The principal scaling claim of this design is memory compression. A classical grid with $\sigma_\nu$7 cells and $\sigma_\nu$8 directions requires $\sigma_\nu$9 storage, whereas the amplitude-encoded quantum registers require, in principle, $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$0 qubits, plus ancillas. The paper gives the example of a $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$1 grid, approximately $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$2 sites, with $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$3 directions: about $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$4 qubits for position and $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$5 qubits for direction, for a total of about $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$6 data qubits before ancillas. At the same time, the paper explicitly notes that classical simulators such as Qiskit Aer do not realize these asymptotic memory savings, because runtime still scales with the state dimension.

5. Software structure and workflow

QuaTR is implemented as a Python library using Qiskit for circuit construction and Qiskit Aer for simulation (Devkota et al., 15 Nov 2025). Its organization follows the structure of the QLBM workflow.

Module	Role
qlbm_rt	High-level simulate method that constructs and executes the full quantum circuit for each time step and returns lattice data
qlbm_circuits	Circuit constructors for streaming, collision or relaxation, source injection, and angular redistribution in 1D, 2D, and 3D
lbm_utils and qlbm_utils	Classical and quantum utilities for stencils, weights, equilibria, initialization, and analysis glue
analysis	Post-processing for isotropy metrics, profiles, and diagnostics
test	Canonical validation problems
Demo notebooks	Include a fully classical LBM for side-by-side comparison

A typical workflow begins by specifying dimensionality, domain size and resolution, a stencil such as D3Q27, opacity $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$7, emissivity $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$8, boundary conditions, and a time step $$\frac{\mathrm{d}I(\mathbf{x}, \hat{\mathbf{n}}, \nu)}{\mathrm{d}s} = -\kappa_\nu(\mathbf{x})\, I(\mathbf{x}, \hat{\mathbf{n}}, \nu) + j_\nu(\mathbf{x}) + \int_{4\pi} \sigma_\nu(\hat{\mathbf{n}}', \hat{\mathbf{n}})\, I(\mathbf{x}, \hat{\mathbf{n}}', \nu)\, \mathrm{d}\Omega' .$$9 compatible with lattice Courant constraints. A backend such as the Aer simulator is then selected, and the simulation is run with or without angular redistribution enabled. Post-processing extracts observables such as

$\sigma_\nu \approx 0$0

directional fluxes, and isotropy diagnostics from the returned lattice data.

The toolkit is distributed via pip as PyQuaRT on PyPI, with source code on GitHub and public documentation. Its use is directed toward both algorithmic experimentation and side-by-side comparison of classical and quantum formulations of the same lattice Boltzmann radiative-transfer step.

6. Validation, limitations, and significance

QuaTR’s validation suite consists of three standard tests whose roles are sharply differentiated (Devkota et al., 15 Nov 2025). In the isotropic source test, a point or distributed emitter in a uniform non-scattering medium should produce a spherically symmetric intensity profile; the reported result is that angular redistribution reduces lattice-direction imprint and improves spherical symmetry. In the opaque cloud shadow test, the shape of the shadow is sensitive to angular resolution; the reported result is cleaner, symmetric shadow boundaries. In the crossing-beams test, the goal is to verify noninteracting superposition in a collisionless setting; the reported result is correct crossing without spurious coupling.

The current formulation also has explicit limitations. General scattering, represented by the angular integral operator in the full RTE, is not yet implemented. The equilibrium and relaxation maps are linear rather than strictly unitary, so rigorous quantum realizations may require block-encoding or variational approximations, increasing ancilla counts and calibration complexity. The summary does not present detailed runtime or memory benchmarks, and it does not discuss error-mitigation strategies. Practical execution is therefore currently centered on simulators rather than on large-scale hardware deployments.

These limitations delimit the present contribution. QuaTR is not presented as a complete quantum radiative-transfer solver for all astrophysical regimes, nor as an experimentally benchmarked quantum speedup. Its significance lies in the coherence of the framework: the core RTE is mapped into streaming, relaxation, source injection, and angular redistribution steps; these steps are exposed both algorithmically and as Qiskit circuits; and the non-scattering regime most susceptible to lattice anisotropy is addressed by a dedicated redistribution operator that reportedly improves isotropy without increasing per-step circuit-construction complexity. Within that scope, QuaTR provides a concrete research platform for studying QLBM-based radiative transfer and for testing how quantum representations might eventually address one of the most expensive components of cosmological simulation.