Exploring the Boundaries of Differentiable Radiation Transport and Detector Simulation

TL;DR

This paper applies automatic differentiation to particle transport simulations, identifies instability issues at material boundaries, and introduces a mitigation strategy to enable stable gradients for detector design optimization.

Contribution

It presents a novel approach to differentiate through Geant4-like simulations by mitigating boundary-crossing instabilities, facilitating optimization tasks.

Findings

Gradient explosions occur at material boundaries due to extreme sensitivities.

Stopping gradient propagation at boundaries stabilizes derivatives.

The method enables optimization in detector design problems.

Abstract

We present an application of automatic differentiation for particle transport through matter using a Geant4-like radiation transport simulation with a full electromagnetic physics model. When differentiating this step-based transport, we observe exploding gradients driven by rare but extreme sensitivities at material boundaries, which propagate through subsequent transport and shower development. To obtain usable derivatives for optimization, we introduce a targeted mitigation strategy that stops gradient propagation through boundary-crossing operations under identifiable unstable conditions while leaving the forward (primal) simulation unchanged. We demonstrate that this enables stable, optimization-ready gradients in a detector-design problem.