gMicroMC: GPU Monte Carlo for Radiolysis Chemistry

• gMicroMC is a GPU-based Monte Carlo track-structure algorithm that simulates radiolysis chemistry in pure water across physical, physico-chemical, and chemical stages.
• It extends traditional models by incorporating multiple-pulse, ultra-high-dose-rate irradiation through a hybrid approach that combines SBS with a deterministic ODE solver.
• The method quantifies dose-rate dependent radical chemistry by tracking post-irradiation H2O2 yields, revealing a threshold behavior in radical accumulation.

Searching arXiv for papers on gMicroMC and closely related name confusions. gMicroMC is a GPU-based Monte Carlo track-structure algorithm for radiolysis chemistry that, in the cited work, is advanced to simulate multiple-pulse, ultra-high-dose-rate (UHDR / Flash-like) irradiations in pure water. In that formulation, it resolves the physical stage ($10^{-17}$–$10^{-15}\,\mathrm{s}$), the physico-chemical stage ($10^{-15}$–$10^{-12}\,\mathrm{s}$), and the chemical stage starting at about $10^{-12}\,\mathrm{s}$, while extending earlier single-pulse or short-time usage to pulse trains with inter-pulse chemical carry-over. The principal scientific use demonstrated for the code is the study of dose-rate-dependent radical chemistry, with the post-irradiation $G$-value of $\mathrm{H_2O_2}$ used as a surrogate for enhanced radical-radical recombination, especially $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$ (Molina-Hernandez et al., 3 Sep 2025).

1. Definition and scope

In the cited study, gMicroMC is presented as a radiolysis and radiation-chemistry code rather than as a generic particle-physics Monte Carlo, a dark-matter package, or a Bayesian sampling algorithm. Its computational domain is microscopic radiation track structure in water, where ionization and excitation events, secondary-electron transport, initial radiolytic-species generation, and subsequent diffusion-reaction chemistry are modeled in space and time. The work states that gMicroMC integrates physical, physico-chemical, and chemical modules adopted from PARTRAC and Geant4-DNA, and that the non-homogeneous chemical stage is handled with a step-by-step (SBS) diffusion-reaction method in which reactions are treated as diffusion-controlled (Molina-Hernandez et al., 3 Sep 2025).

The simulated medium in the main application is pure water. The principal species tracked are

$e_{\mathrm{aq}^{-},\ ^{\cdot}\mathrm{OH},\ \mathrm{H}^{\cdot},\ \mathrm{H}^{+},\ \mathrm{OH}^{-},\ \mathrm{H}_2,\ \mathrm{H}_2\mathrm{O}_2.$

The main irradiation studied is 70 MeV protons, with LET $=0.96\ \mathrm{keV/\mu m}$, chosen for computational convenience while still producing chemistry described as similar to low-LET clinical electrons. The cited work therefore places gMicroMC within radiation chemistry and Flash-radiotherapy mechanism studies, specifically as a microscopic simulator of pulsed water radiolysis rather than of biological response itself.

A recurrent source of confusion is nomenclature. The data also include GMCALC, which is a self-contained FORTRAN program for the Georgi–Machacek model (Hartling et al., 2014); micrOMEGAs 6.0, which is a dark-matter phenomenology code for $10^{-15}\,\mathrm{s}$0-component dark sectors (Alguero et al., 2023); and geodesic Monte Carlo (gMC), which is a manifold MCMC method for Bayesian inference (Holbrook, 2018). None of those is gMicroMC. In context, gMicroMC denotes a radiation-chemistry track-structure code.

2. Multiple-pulse extension and computational architecture

The central methodological development is the extension of gMicroMC to support multiple pulse simulations and Flash dose rates. The paper frames the limitation of conventional track-structure Monte Carlo as its emphasis on effectively independent tracks and chemistry up to about $10^{-15}\,\mathrm{s}$1, whereas multiple-pulse irradiation requires treatment of chemicals that persist and interact over much longer intervals. The new implementation therefore allows irradiation scenarios to be specified by total dose $10^{-15}\,\mathrm{s}$2, particle energy $10^{-15}\,\mathrm{s}$3, pulse dose rate $10^{-15}\,\mathrm{s}$4, average dose rate $10^{-15}\,\mathrm{s}$5, pulse full width at half maximum, and pulse frequency $10^{-15}\,\mathrm{s}$6 (Molina-Hernandez et al., 3 Sep 2025).

The pulse relations are given explicitly as

$10^{-15}\,\mathrm{s}$7

and

$10^{-15}\,\mathrm{s}$8

The number of primary particles needed to deliver a target dose in a square volume of interest of side length $10^{-15}\,\mathrm{s}$9 is computed from

$10^{-15}$0

The upgraded algorithm alternates between two transport regimes. During beam-on periods, the simulation uses SBS + periodic boundary conditions (PBCs) to resolve spatially non-homogeneous chemistry while new radiolysis products are injected according to the pulse schedule. During beam-off periods and after the last pulse, the simulation switches to a numerical ordinary differential equation (NumODE) solver that evolves homogenized concentrations at low computational cost. After the final pulse, the NumODE system is run out to 10 s so that a steady state in concentrations is reached before final $10^{-15}$1-values are recorded (Molina-Hernandez et al., 3 Sep 2025).

This hybrid structure is the distinctive architectural feature of the extended code. It allows microscopic track-structure chemistry during irradiation while making long inter-pulse and post-irradiation evolution computationally tractable.

3. Transport, boundary conditions, and reaction network

The multiple-pulse implementation uses a micrometric voxel as a representative element of bulk water by imposing two periodic boundary conditions during the non-homogeneous SBS stage. The first remaps species that diffuse out of the voxel: $10^{-15}$2 The second applies the minimum-image convention for separations used in reaction testing: $10^{-15}$3 To avoid self-reaction with periodic images, the reaction-radius constraint

$10^{-15}$4

must hold (Molina-Hernandez et al., 3 Sep 2025).

Once the species distribution is taken to be homogeneous, the code advances the chemistry through deterministic ODEs. The paper gives, for solvated electrons,

$10^{-15}$5

with

$10^{-15}$6

All corresponding ODEs for the reaction network are numerically integrated.

The chemistry list used in the paper is the following: $10^{-15}$7 with tabulated rate constants (Molina-Hernandez et al., 3 Sep 2025).

For self-recombination, the supplement derives diffusion-controlled expressions. For hydroxyl radicals,

$10^{-15}$8

and

$10^{-15}$9

These relations are directly relevant because the paper’s key observable, $10^{-12}\,\mathrm{s}$0, is controlled by reaction R6.

4. Irradiation setup, observables, and validation

The main multiple-pulse study uses pure water, 70 MeV protons, LET $10^{-12}\,\mathrm{s}$1, $10^{-12}\,\mathrm{s}$2, total dose up to 20 Gy, and a volume of interest with $10^{-12}\,\mathrm{s}$3. The pulse dose rates are $10^{-12}\,\mathrm{s}$4 and $10^{-12}\,\mathrm{s}$5, while the average dose rate ranges from 0.01 to 100000 Gy/s through variation of pulse frequency (Molina-Hernandez et al., 3 Sep 2025).

The principal readout is the post-irradiation $10^{-12}\,\mathrm{s}$6 yield, obtained as the slope of the dose–concentration distributions from plots of $10^{-12}\,\mathrm{s}$7 versus delivered dose. The paper states conceptually that $10^{-12}\,\mathrm{s}$8-values are the number of chemical species produced normalized to deposited energy, usually in $10^{-12}\,\mathrm{s}$9 or species/100 eV, but it does not give a formal equation for the $10^{-12}\,\mathrm{s}$0-value in the main text. Accordingly, the cited work uses $10^{-12}\,\mathrm{s}$1 operationally as a surrogate for the extent of radical-radical reactions rather than as a starting point for a formal derivation (Molina-Hernandez et al., 3 Sep 2025).

Validation proceeds in several layers. For the periodic-boundary implementation, concentration evolution was tested in voxels of

$10^{-12}\,\mathrm{s}$2

with initial concentrations scaled from a single-pulse 70 MeV proton simulation using $10^{-12}\,\mathrm{s}$3 and $10^{-12}\,\mathrm{s}$4. All three sizes yielded essentially the same concentration evolution, indicating volume independence under PBCs. Against Kinetiscope, the SBS+PBC model reproduced the same qualitative trends, though at steady state near $10^{-12}\,\mathrm{s}$5 the maximum concentration differences were approximately 20% for $10^{-12}\,\mathrm{s}$6 and 12% for $10^{-12}\,\mathrm{s}$7.

By contrast, the NumODE stage matched Kinetiscope to within 1% for all species at any time under the same homogeneous-mixture conditions. The computational gain was also large: the SBS + PBC treatment at $10^{-12}\,\mathrm{s}$8 took about 1 day, whereas NumODEs and Kinetiscope completed the same homogeneous problem in less than a second, corresponding to a speed-up of about

$10^{-12}\,\mathrm{s}$9

5. Mechanistic results at Flash-relevant dose rates

The principal physical result is a threshold behavior in average dose rate. For each of the two pulse dose rates tested, the simulations identify a transition between a regime in which radicals deplete between pulses and a regime in which reactive species persist and accumulate from pulse to pulse. Below threshold, the paper states that complete radical depletion occurs within the pulse intervals, leading to constant $G$0-values. Above threshold, reactive species accumulate throughout the irradiation, increasing radical-radical reactions and thus the $G$1-values of $G$2 (Molina-Hernandez et al., 3 Sep 2025).

The reported thresholds are on the order of $G$3 for $G$4 and $G$5 for $G$6. The work further states that the threshold depends on pulse dose rate: lower dose per pulse produces fewer species per pulse and requires more time for depletion, whereas higher $G$7 shifts the threshold upward by about a factor of ten in the reported simulations.

The corresponding increase in post-irradiation $G$8 yield between low-dose-rate and maximum-UHDR conditions is quantitative. For $G$9, the paper reports

• SDR: $\mathrm{H_2O_2}$0, $\mathrm{H_2O_2}$1 species/100 eV
• UHDR: $\mathrm{H_2O_2}$2, $\mathrm{H_2O_2}$3 species/100 eV

For $\mathrm{H_2O_2}$4, it reports

• SDR: $\mathrm{H_2O_2}$5, $\mathrm{H_2O_2}$6 species/100 eV
• UHDR: $\mathrm{H_2O_2}$7, $\mathrm{H_2O_2}$8 species/100 eV

These correspond to increases of 30% and 20%, respectively. The paper also states that increasing pulse dose rate itself amplified the build-up effect, with $\mathrm{H_2O_2}$9 production increasing by 26% between the two tested $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$0 values.

Within the pure-water reaction network, the mechanistic interpretation is direct. The enhancement in $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$1 is attributed to increased carry-over of $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$2 between pulses, which enhances

$$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$3

The paper also notes a root-like tendency of $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$4 versus dose because growing $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$5 concentrations increase its own consumption through

$$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$6

6. Interpretation, misconceptions, and limitations

A central interpretive distinction in the cited work is between the inter-track mechanism and reactive species build-up. The inter-track hypothesis attributes enhanced radical-radical chemistry to simultaneous overlap of nearby tracks during the early non-homogeneous stage and is therefore driven primarily by pulse dose rate $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$7. By contrast, reactive species build-up is a pulse-train effect controlled by average dose rate $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$8, because it depends on radicals surviving the interval between pulses. The paper concludes that reactive species build-up is more likely than the inter-track mechanism to promote radical-radical reactions under clinically relevant UHDR conditions (Molina-Hernandez et al., 3 Sep 2025).

This conclusion is carefully delimited. The simulations are for pure water only. They do not include oxygen, biomolecular scavengers, pH buffering, proteins, membranes, DNA, cellular ultrastructure, or biological endpoints such as DNA damage, cell survival, tissue oxygenation, immune response, or normal-tissue sparing. The chemistry list is explicitly restricted, and the authors note that oxygen and pH reactions are absent. They further remark that this omission may underlie disagreement with experiments in aerated systems, where recent measurements often report a decrease in $$^{\cdot}\mathrm{OH}+^{\cdot}\mathrm{OH}\rightarrow \mathrm{H_2O_2}$$9 at UHDR, opposite to the pure-water prediction.

The hybrid nature of the algorithm is another important limitation. The multiple-pulse framework alternates a particle-based SBS treatment with a deterministic ODE treatment. Because SBS+PBC and Kinetiscope are not numerically identical in homogeneous validation, the interface between non-homogeneous and homogeneous stages may introduce artifacts. The paper presents this as a necessary practical compromise rather than a complete solution.

A further potential misconception concerns the relationship between gMicroMC and other similarly named software. It is not GMCALC, which is a Georgi–Machacek-model spectrum, constraints, and decay calculator with a MadGraph5 interface (Hartling et al., 2014). It is not micrOMEGAs, which computes dark-matter relic density and detection observables in generic BSM models (Alguero et al., 2023). It is not geodesic Monte Carlo, a manifold-based Hamiltonian sampling algorithm for Bayesian inference (Holbrook, 2018). The shared “MC” substring reflects only the broad fact that all are Monte Carlo-related in some sense; their scientific domains are unrelated.

7. Position in the software landscape and research significance

Within the scope of the cited work, gMicroMC occupies a specialized niche: it is a microscopic radiolysis code designed to connect track-structure transport to pulse-structured chemistry over timescales ranging from femtoseconds to seconds. Its practical innovation is the ability to simulate multiple pulses, apply periodic boundary conditions so that a micrometric voxel represents bulk water during the non-homogeneous stage, and then switch to a fast ODE-based homogeneous chemistry solver for long beam-off intervals and post-irradiation evolution (Molina-Hernandez et al., 3 Sep 2025).

This makes the code suitable for studies of dose-rate dependence in water radiolysis, pulse-structure effects, and radical carry-over between pulses. A plausible implication is that the code can serve as a mechanistic bridge between independent-track radiolysis simulations and macroscopic pulse-train irradiation conditions relevant to Flash-radiotherapy discussions. The cited work, however, does not claim that gMicroMC already constitutes a full biological Flash simulator.

The principal significance of the reported extension is therefore twofold. Methodologically, it overcomes the earlier inability of track-structure chemistry simulations to treat realistic pulse trains and average-dose-rate effects over long timescales. Scientifically, it supports the proposition that, in pure water, brief intervals between pulses led to a reactive species build-up, which enhanced radical-radical reactions, and that this build-up is more likely to promote radical-radical reactions than the inter-track mechanism. Within the limits of the model, gMicroMC is thus a dedicated computational framework for investigating dose-rate-dependent radiation chemistry under UHDR and Flash-like conditions.