Batched Wavefront Sweeps for Linear Transport Uncertainty Quantification

Abstract

We develop a graph-compatible batched wavefront formulation for upwind discontinuous Galerkin discretisations of linear transport. For sweepable discretisations, the cell equations form a block lower-triangular system under a topological ordering of the directed upwind dependency graph. This causal structure can be shared across families of fixed-source problems whose local operators, sources and inflow data vary. We exploit this observation by grouping channels with a common dependency graph into graph-compatible sweep classes and carrying samples, right-hand sides, energy groups and sign-compatible angular ordinates as tensor dimensions within a common wavefront schedule. We prove that the resulting batched wavefront algorithm is algebraically equivalent to independent classical DG sweeps for every channel in a class, while exposing parallelism simultaneously across wavefront cells and channel dimensions. We realise the formulation as a GPU tensor program based on batched cell-local solves and quantify its work, depth and storage requirements, including the throughput--memory tradeoff introduced by sample microbatching. We further characterise the discrete parameter-to-observable map on fixed face-sign branches and show that reverse-mode differentiation through the wavefront algorithm reproduces the corresponding discrete adjoint action. Numerical experiments verify the expected DG convergence and adjoint consistency, demonstrate substantial GPU execution gains from sample batching, and illustrate the method in a material-shadowing uncertainty-quantification problem and a coupled multigroup C5G7/KAIST-inspired power-iteration workload.