A concentration-source-driven and GPU-accelerated standalone radiochemistry framework for continuous water radiolysis and pulse-structure simulation

Objective: Radiochemistry modeling plays a critical role in understanding the evolution of radiolytic species and their potential role in radiation-induced biological responses. Conventional Monte Carlo (MC) approaches can resolve early non-homogeneous chemistry, but their direct use in long-time, large-volume, or multi-pulse chemical-stage simulations is limited by high computational cost and boundary-condition constraints. This study aims to develop a GPU-accelerated, standalone radiochemistry framework driven by species concentration distribution sources (SCDS), enabling continuous water radiolysis modeling from the early non-homogeneous stage to the homogeneous stage and supporting flexible construction of pulse structures. Approach: SCDS were generated from Geant4-DNA simulations at tₛₜₐᵣₜ=5~ns and used as source terms for subsequent reaction--diffusion calculations. A concentration-dependent correction (CDC), fitted as a function of the local reactant concentration product, was incorporated to account for non-homogeneous reaction kinetics not captured by classical mass-action reaction–diffusion equations (RDEs). A 1~nm minimum reaction unit (MRU) and a dual-grid GPU implementation were used to preserve simulation accuracy while ensuring efficiency. The model was benchmarked against Geant4-DNA for 50 and 500 MeV protons, 100 MeV/u ⁴He ions and 200 MeV/u ¹²C ions, and was used to simulate inter-track effects and macro-pulse irradiation. Main Results: The fitted CDC functions captured the concentration-dependent kinetic discrepancy with R²>0.99 for representative reactions and became negligible in the low-concentration regime, thereby recovering classical RDE kinetics. Using a single set of CDC functions, RDEs-extended reproduced Geant4-DNA time-dependent G-values for 50 and 500 MeV protons, 100 MeV/u ⁴He ions and 200 MeV/u ¹²C ions, with most species showing deviations within ±5% at 1~. The GPU-accelerated implementation reduced computation time by $193.74$--1191.37× for the 1~ benchmark cases. Pulse-structure simulations showed stronger depletion of eaq⁻ and OH⁻ under shorter source separation and higher dose per pulse, indicating enhanced radical overlap and recombination. Significance: This work establishes an SCDS-driven, GPU-accelerated radiochemistry framework for continuous chemical-stage simulation of water radiolysis. The framework improves computational efficiency, reduces dependence on incident-particle tracking and boundary conditions, and supports pulse-structure simulations for studying radiolytic species evolution. The proposed model provides an efficient and extensible tool for simulating pulsed radiochemistry in realistic radiotherapy scenarios.