Dethinning Extensive Air Shower Simulations

We describe a method for restoring information lost during statistical thinning in extensive air shower simulations. By converting weighted particles from thinned simulations to swarms of particles with similar characteristics, we obtain a result that is essentially identical to the thinned shower, and which is very similar to non-thinned simulations of showers. We call this method dethinning. Using non-thinned showers on a large scale is impossible because of unrealistic CPU time requirements, but with thinned showers that have been dethinned, it is possible to carry out large-scale simulation studies of the detector response for ultra-high energy cosmic ray surface arrays. The dethinning method is described in detail and comparisons are presented with parent thinned showers and with non-thinned showers.

💡 Research Summary

The paper addresses a fundamental limitation of extensive air‑shower (EAS) simulations used for ultra‑high‑energy cosmic‑ray (UHECR) studies. Standard Monte‑Carlo codes such as CORSIKA and AIRES employ a statistical “thinning” technique to reduce the enormous number of secondary particles that would otherwise have to be tracked. In thinning, particles below a chosen energy threshold ε_th are randomly discarded, and the surviving particles are assigned a weight w = 1/p, where p is the survival probability. While this preserves the average particle density in the dense core of the shower, it severely degrades the statistical quality of the particle distribution at large lateral distances (kilometers from the core), which is precisely the region sampled by surface detector (SD) arrays. Consequently, simulations that rely on thinned showers cannot accurately reproduce the fluctuations and timing information needed for SD response modeling.

The authors propose a “dethinning” algorithm that reconstructs the lost information by converting each weighted particle into a swarm of w − 1 new particles, thereby restoring unit weight for all particles. The procedure consists of four steps: (1) select a vertex point along the original particle’s trajectory, constrained by a maximum distance D_max derived from the requirement that the reconstructed particle’s arrival time does not precede the shower front; (2) generate a new direction by sampling a two‑dimensional Gaussian distribution (σ of a few degrees) centered on the original trajectory – this defines a “Gaussian cone”; (3) project the new particle to ground level, assigning it a time and energy consistent with the original particle’s kinematics and with a small random perturbation (±10 % Gaussian for energy); (4) repeat steps 2–3 w − 1 times, adding an extra particle with probability equal to the fractional part of w when w is non‑integer. The algorithm also includes several tunable parameters: cone opening angle β (3° km⁻¹ for electromagnetic particles, 1° km⁻¹ for muons), a minimum lateral distance r_min ≥ 100 m (to avoid the saturated core), an acceptance probability P = exp(−Δχ/ε) for particles whose slant‑depth difference Δχ exceeds a threshold (ε = 50 g cm⁻²), and a refined vertex height choice (the smaller of D_max and a depth‑based estimate D′) to keep spatial spread realistic.

A key aspect of the study is the choice of ε_th. The authors find that ε_th = 10⁻⁷ yields a sufficiently dense set of original weighted particles such that dethinning works without further adjustments. For ε_th = 10⁻⁶, which reduces CPU time and storage by a factor of ten relative to 10⁻⁷, the algorithm still succeeds after careful tuning of the parameters listed above. In contrast, ε_th = 10⁻⁵ provides too few original particles, and the reconstructed showers deviate noticeably from non‑thinned references.

The validation proceeds in two stages. First, the authors compare a thinned shower (ε_th = 10⁻⁶) with its own dethinned version. They divide the shower footprint into eight radial rings (500–4500 m) and six azimuthal wedges, then histogram particle fluxes in ten bins of incident angle and three particle types (photons, electrons, muons), yielding 1 440 distinct energy spectra. The agreement across all spectra demonstrates that dethinning preserves the angular and energy distributions of the parent thinned shower. Second, they compare dethinned showers with a library of >100 fully non‑thinned CORSIKA simulations generated in parallel. Because the non‑thinned and thinned runs cannot be identical, the authors first normalize the total secondary flux of the non‑thinned set to match the thinned set for each wedge and particle type. After this normalization, they examine 6 × 6 m² tiles across the same radial range, extracting the times at which 10 % (t₁/₁₀) and 50 % (t₁/₂) of the total flux arrive, as well as the integrated photon, electron, and muon fluxes. The dethinned and non‑thinned results agree within statistical uncertainties, confirming that the dethinning procedure reproduces the full‑shower physics relevant to surface detector observations.

The practical impact is substantial. Dethinned showers retain the fidelity of fully simulated showers while requiring only one‑tenth of the CPU time and storage needed for non‑thinned runs. This makes large‑scale Monte‑Carlo studies of surface detector arrays—such as the Telescope Array (TA) experiment—feasible, enabling accurate aperture calculations even in the energy regime where the trigger efficiency is below 100 %. The method also opens the door to systematic studies of detector response, energy reconstruction, and composition analyses that were previously limited by the coarse nature of thinned simulations.

In summary, the paper introduces a robust, physics‑based dethinning algorithm that restores the detailed particle content lost during statistical thinning. By carefully handling temporal consistency, angular spread, energy perturbations, and lateral distance cuts, the authors demonstrate that dethinned showers are virtually indistinguishable from fully simulated ones. This advancement dramatically reduces computational demands while preserving the accuracy needed for modern UHECR surface detector analyses.