CORSIKA 8 Simulation Framework

Updated 8 July 2026

CORSIKA 8 is a modern simulation framework that models extensive air showers and particle cascades with a modular C++ architecture.
It replaces CORSIKA 7’s monolithic Fortran code by enabling configurable media, flexible geometry, and advanced detector observables like radio and optical emissions.
The framework supports both electromagnetic and hadronic interactions, validated against legacy codes, and leverages GPU and multi-threaded processing for high-performance computing.

CORSIKA 8 is a modern simulation framework for extensive air showers and, more broadly, for particle cascades in air and dense media. It was developed as a complete redesign of the long-established CORSIKA 7 code in order to preserve the physics capabilities relied upon in astroparticle physics while replacing a monolithic, hand-optimized Fortran implementation with a modular C++ architecture. In the literature, CORSIKA 8 is characterized not merely as a rewrite of an air-shower program, but as a flexible framework for configurable cascades, interaction media, geometries, and detector-related observables such as radio and optical emission (Huege et al., 2023, Alameddine et al., 2 Apr 2026).

1. Historical development and motivation

For more than two decades, CORSIKA 7 and its predecessors were the de facto standard Monte Carlo tools for simulating extensive air showers. Their scientific role spanned detector design, event reconstruction, and composition studies across astroparticle physics. That legacy code remained highly performant because it was hand-optimized, but it also accumulated a tightly intertwined source structure that became increasingly difficult to maintain, extend, and adapt to new computing paradigms. Several papers identify the retirement of original developers, the Fortran 77 basis, and the code’s monolithic structure as central reasons for the redesign (Sandrock, 2022, Reininghaus et al., 2019).

The CORSIKA 8 project began in 2018 as a complete rewrite in modern C++. The stated motivations were maintainability, extensibility, and support for modern supercomputing environments, including future multi-threading and GPU-oriented workflows. At the same time, the project was driven by new scientific requirements: full particle genealogy, flexible media definitions, cross-media showers, radio and Cherenkov modules, and simulation scenarios not naturally supported by an atmosphere-only architecture (Huege, 2022, Sandrock, 2022).

A recurrent theme in the project documentation is that CORSIKA 8 is a framework rather than a single-purpose application. This distinction is substantive. The framework model allows interchangeable physics modules, explicit environment handling, and separate validation of individual components. It also broadens the intended scope beyond air showers alone to particle cascades in ice, water, rock, ground, and mixed geometries (Reininghaus et al., 2019, Alameddine et al., 2021).

By 2023 and 2024, the collaboration described CORSIKA 8 as “physics-complete” in the sense that the major physical process chains relevant to shower development had been implemented, even though performance optimization, documentation, steering, and additional model integrations remained active areas of work (Huege et al., 2023, Gaudu, 2024).

2. Architecture, transport, and environment model

CORSIKA 8 is organized around a modular set of core components. The literature describes these as including a Cascade object, Tracking, Environment, a Process List, and an Output manager; earlier architectural descriptions also emphasize a particle stack, process sequence, environment, and transport as the core building blocks (Huege et al., 2023, Reininghaus et al., 2019).

The particle stack stores the particles currently being propagated. At minimum it contains four-momenta, four-positions, and particle codes, and it can be extended to include statistical weights for thinning and optional history information. The Process List or process sequence contains discrete and continuous processes on equal footing. Discrete processes include interactions, decays, and boundary crossings; continuous processes include energy loss, Cherenkov emission, and related along-track effects (Reininghaus et al., 2019).

Transport proceeds by propagating particles one by one until no particles remain. The general algorithm described for CORSIKA 8 is: propose a trajectory, limit the step size according to accuracy criteria and boundaries, sample discrete processes, perform continuous processes along the accepted segment, and update the particle stack. In the early architecture paper, trajectories are restricted to linear segments in order to keep boundary-intersection calculations with spheres and cuboids tractable (Reininghaus et al., 2019). A compact formal statement of the step selection logic later appears as

$\ell_{\text{step}} = \min\left( \ell_{\text{boundary}}, \ell_{\text{magnetic}}, \ell_{\text{decay}}, \ell_{\text{interaction}}, \ell_{\text{continuous},i} \right),$

with grammage-based interaction sampling and resampling when rates change during transport (Alameddine et al., 2 Apr 2026).

The environment model is one of the most consequential departures from CORSIKA 7. Instead of assuming an atmosphere-only world, CORSIKA 8 uses a volume tree in which geometric objects such as spheres and cuboids are assigned medium properties and arranged by containment. The root volume is the Universe. This allows fast determination of the smallest containing volume at a point and therefore of the active medium during propagation (Reininghaus et al., 2019, Ammerman-Yebra et al., 2023).

Medium properties can include mass density, elemental or isotope composition, magnetic field, refractive index, and related quantities, with the framework designed so that modules request only the properties they need. Later architectural descriptions formalize this with composable medium interfaces and explicit coordinate and unit handling, including a compile-time unit system based on PhysUnits and geometry represented through points, vectors, coordinate systems, and $\mathrm{SE}(3)$ transformations (Alameddine et al., 2 Apr 2026).

This environment model enables a class of simulations that earlier CORSIKA generations did not support naturally: showers crossing boundaries between different media in a single run. Examples discussed in the literature include air-to-water and air-to-ice cascades, as well as more specialized environments involving mountains, Antarctic ice, and detector-specific geometries (Sandrock, 2022, Ammerman-Yebra et al., 2023).

3. Physics content and interaction models

CORSIKA 8 supports both hadronic and electromagnetic cascades and is therefore capable of simulating complete extensive air showers. In the hadronic sector, the framework interfaces to several state-of-the-art high-energy interaction models, including QGSJet-II.04, EPOS-LHC, Sibyll 2.3d, and Pythia 8.3; low-energy interactions are handled by UrQMD in earlier status papers and by FLUKA in later ones, while decays can be treated by Sibyll 2.3d and Pythia 8 (Sandrock, 2022, Huege et al., 2023).

The hadronic model interface is designed to be generic. Later architectural work describes a getCrossSection(...) / doInteraction(...) abstraction that requires only projectile and target identification plus four-momenta, with frame transformations handled inside the framework. This facilitates both shower simulation and validation against external event-generator data (Alameddine et al., 2 Apr 2026).

The electromagnetic cascade is handled differently from CORSIKA 7. In the legacy code, the EM part was based on a customized and deeply integrated version of EGS4. In CORSIKA 8, electromagnetic transport is provided by PROPOSAL, specifically PROPOSAL 7.6.2 in the validation literature, a modular C++14 propagation library with Python bindings. It is used for electrons, positrons, photons, muons, and tau leptons (Sandrock et al., 2023, Huege et al., 2023).

The dominant EM shower processes remain bremsstrahlung and $e^+e^-$ pair production. CORSIKA 8 also implements high-energy effects and rare channels that are essential for physics-complete electromagnetic showers: the Landau-Pomeranchuk-Migdal effect, muon-pair production by photons, and photohadronic interactions. In the CORSIKA 8 implementation, the LPM suppression is applied through a Neumann rejection method: an interaction is sampled, a random number is drawn, and the interaction is rejected if that random number is larger than the ratio of the cross-section with LPM suppression to the cross-section without suppression (Sandrock et al., 2023).

The electromagnetic comparison literature explicitly identifies several process-level differences between CORSIKA 7 and 8. Rayleigh scattering is not implemented in CORSIKA 8; triplet production,

$e^\pm \rightarrow e^\pm e^+ e^-,$

is not implemented in CORSIKA 7; the parametrization of the photoelectric effect differs; and the calculation of secondaries in photohadronic interactions differs. The resulting cross-sections differ by at most a few percent, and only at the lowest or extremely high energies, or where a process contributes very little (Sandrock et al., 2023).

Photohadronic interactions are particularly important for the muon content of electromagnetic showers. Their cross-section is subdominant, about $1\%$ of pair production, but the hadronic products generate muons, and the resulting hadron decays are identified as the dominant source of the muon content in electromagnetic showers. In CORSIKA 8, the low-energy part is handled by SOPHIA, while at higher energies the chosen high-energy hadronic model is invoked (Sandrock et al., 2023).

A further methodological feature is electromagnetic thinning. CORSIKA 8 introduces an improved thinning algorithm in which, below a thinning threshold $E_{\mathrm{th}}$ , only one secondary is retained with probability proportional to its energy fraction, and statistical thinning is used if weight limitation becomes relevant. The framework paper gives

$p_i = \max\left(\frac{E_i}{E_1+E_2}, \frac{w_0}{w_\mathrm{max}}\right), \qquad w_i = \frac{w_0}{p_i} \leq w_\mathrm{max},$

with the stated aim of reducing artificial fluctuations and narrowing high-weight peaks relative to CORSIKA 7 (Huege et al., 2023).

4. Validation against CORSIKA 7 and other codes

Validation is central to the published characterization of CORSIKA 8. The typical strategy is not to require bitwise identity with older software, but to compare observables against CORSIKA 7 and other established codes while accounting for independent implementations and known physics differences (Sandrock, 2022, Alameddine et al., 2 Apr 2026).

For electromagnetic showers, an early milestone was the agreement among CORSIKA 7, CORSIKA 8, AIRES, and ZHS for the longitudinal profile and charge excess of $1\ \mathrm{TeV}$ electromagnetic showers (Sandrock, 2022). A more complete validation at higher energy was later performed for showers initiated by electrons in the US standard atmosphere, with FLUKA as low-energy hadronic model and Sibyll 2.3d as high-energy model. The benchmark observables were longitudinal particle-number profiles, charge excess, lateral distributions, and energy spectra near shower maximum (Sandrock et al., 2023).

For 1000 showers initiated by a $100\ \mathrm{TeV}\ e^-$ with a $1\ \mathrm{MeV}$ particle cut, the charged-particle and photon longitudinal profiles agreed to better than $\mathrm{SE}(3)$ 0, charge excess to better than $\mathrm{SE}(3)$ 1, and lateral distributions within $\mathrm{SE}(3)$ 2 except very close to the shower core, within about $\mathrm{SE}(3)$ 3. Differences in the lowest and highest energy bins of spectra near shower maximum were no more than $\mathrm{SE}(3)$ 4. Muon and hadron longitudinal profiles differed more strongly, showing about $\mathrm{SE}(3)$ 5 more particles in one code versus the other, and shower maxima were earlier by about $\mathrm{SE}(3)$ 6 (Sandrock et al., 2023).

At ultrahigh energy, the same study examined 5000 showers of $\mathrm{SE}(3)$ 7 with a $\mathrm{SE}(3)$ 8 cut to isolate the LPM-affected regime. Showers with LPM suppression developed more slowly and reached maximum later, while agreement between CORSIKA 7 and 8 became somewhat worse than at $\mathrm{SE}(3)$ 9 but remained within about $e^+e^-$ 0– $e^+e^-$ 1. The authors noted that the somewhat larger number of charged particles in CORSIKA 8 may be due to the increasing importance of triplet pair production, which is not accounted for in CORSIKA 7 (Sandrock et al., 2023).

The general conclusion of that validation program was that electromagnetic shower simulation in CORSIKA 8 is in good agreement with the latest Fortran version of CORSIKA 7: charge excess and longitudinal profiles agree to within $e^+e^-$ 2– $e^+e^-$ 3, energy spectra and lateral distributions near shower maximum agree well with only small deviations near the core, and LPM-modified showers agree within about $e^+e^-$ 4 (Sandrock et al., 2023).

Hadronic-shower validation follows the same pattern. Status papers report good agreement between CORSIKA 8, CORSIKA 7, and MCEq for hadronic particle spectra at observation level, with an explicit example of a vertical proton shower of $e^+e^-$ 5 observed at $e^+e^-$ 6 a.s.l. (Sandrock, 2022). The 2023 framework paper reports that for 300 vertical proton showers at $e^+e^-$ 7, with thinning and energy cuts, the agreement between CORSIKA 8 and CORSIKA 7 is generally within $e^+e^-$ 8 for electrons/positrons, photons, muons, and hadrons, although a systematic excess of muons and hadrons in CORSIKA 8 remained under investigation (Huege et al., 2023).

More recent framework-level validation summarizes the outcome as agreement at the few-percent level for key observables, with a small but significant $e^+e^-$ 9 for proton-induced showers with QGSJet-II.04 that is attributed to different treatment of $e^\pm \rightarrow e^\pm e^+ e^-,$ 0 hyperons (Alameddine et al., 2 Apr 2026).

5. Radio and optical emission modules

One of the defining features of CORSIKA 8 is that radio and optical observables are implemented as modular processes rather than as tightly coupled appendices to the shower code. This is especially important for experiments in which detector response depends sensitively on geometry, refractive-index structure, and cross-media propagation (Huege, 2022, Huege et al., 2023).

The radio module is consistently described as having four top-level components: a track filter, a formalism, a propagator, and an antenna or observer object. The track filter selects which tracks contribute; the formalism computes the emitted field; the propagator handles transport to observer locations; and the antenna or observer stores the time-domain signal. This separation allows different propagators and formalisms to be combined independently, and it permits multiple radio processes to be run simultaneously on the same shower (Karastathis et al., 2023, Alameddine et al., 2024).

Two microscopic time-domain formalisms are implemented: the Endpoint formalism, as in CoREAS, and the ZHS algorithm, as in ZHAireS. Their co-existence inside one framework is scientifically important because it enables same-shower comparisons without shower-to-shower fluctuations (Karastathis et al., 2023, Gottowik, 19 Sep 2025). A 2025 validation study reports that, for the same underlying showers in CORSIKA 8 with optimized step sizes, Endpoint and ZHS converge to the same radiation energy within $e^\pm \rightarrow e^\pm e^+ e^-,$ 1 in the $e^\pm \rightarrow e^\pm e^+ e^-,$ 2– $e^\pm \rightarrow e^\pm e^+ e^-,$ 3 band and within $e^\pm \rightarrow e^\pm e^+ e^-,$ 4 in the $e^\pm \rightarrow e^\pm e^+ e^-,$ 5– $e^\pm \rightarrow e^\pm e^+ e^-,$ 6 band; CORSIKA 8 and CORSIKA 7 agree on radiation energy to better than $e^\pm \rightarrow e^\pm e^+ e^-,$ 7 in $e^\pm \rightarrow e^\pm e^+ e^-,$ 8– $e^\pm \rightarrow e^\pm e^+ e^-,$ 9 and better than $1\%$ 0 in $1\%$ 1– $1\%$ 2 (Gottowik, 19 Sep 2025).

The radio literature also emphasizes the importance of particle-track step size. Coarser tracking can bias fluence predictions by about $1\%$ 3, whereas finer step sizes substantially improve agreement between CORSIKA 8, CORSIKA 7, and ZHAireS. This is presented as a numerical effect rather than a fundamental physics discrepancy (Gottowik, 19 Sep 2025, Alameddine et al., 2024).

Parallel processing has been introduced for the radio process through the Gyges scheduler. The radio workload is parallelized over bundles of antennas rather than single antennas, because a single antenna’s calculation is often too small to occupy a thread efficiently. Benchmarks with up to 48 worker threads show that, for favorable detector sizes, speedups reach a factor of 10 or more, while different thread counts produce identical pulses in all three polarizations for both CoREAS and ZHS (Jr et al., 2023).

Optical emission is similarly modularized. CORSIKA 8 includes Cherenkov and fluorescence handling, with both CPU-based and GPU-oriented developments. By 2022, the framework already contained a Cherenkov module with two implementations, one vectorized and one using GPU parallelization, and both were reported to agree with each other and with CORSIKA 7 for the ground-level Cherenkov-light distribution from a $1\%$ 4 shower (Sandrock, 2022).

Later work developed a GPU-accelerated optical-photon propagation pipeline for atmospheric fluorescence and Cherenkov light. In the setup described in 2023, the GPU code was still external, synchronized through shared memory, with a ZeroMQ-based synchronization layer planned for later merging (Baack et al., 2023). That work treats fluorescence and Cherenkov emission separately, using an AIRFLY-based parameterization for fluorescence,

$1\%$ 5

and a Frank–Tamm-based calculation for Cherenkov emission,

$1\%$ 6

The reported validation found that the radial Cherenkov-light density from 100 gamma-induced showers of $1\%$ 7 matches exactly between CORSIKA 8 replay mode and the CORSIKA 8 Cherenkov module, and differs by less than $1\%$ 8 between full CORSIKA 8 and CORSIKA 7, while mean arrival times agree nearly $1\%$ 9 out to about $E_{\mathrm{th}}$ 0 from the core (Baack et al., 2023).