A GPU-based Calculation Method for Near Field Effects of Cherenkov Radiation Induced by Ultra High Energy Cosmic Neutrinos

The radio approach for detecting the ultra-high energy cosmic neutrinos has become a mature field. The Cherenkov signals in radio detection are originated from the charge excess of particle showers due to Askaryan effect. The conventional way of calculating the Cherenkov pulses by making Fraunhofer approximation fails when the sizes of the elongated showers become comparable with the detection distances. We present a calculation method of Cherenkov pulses based on the finite-difference time-domain (FDTD) method, and attain a satisfying effeciency via the GPU- acceleration. Our method provides a straightforward way of the near field calculation, which would be important for ultra high energy particle showers, especailly the electromagnetic showers induced by the high energy leptons produced in the neutrino charge current interactions.

💡 Research Summary

The paper addresses a critical limitation in the radio‑detection of ultra‑high‑energy (UHE) cosmic neutrinos: the conventional calculation of Askaryan‑induced Cherenkov pulses relies on the Fraunhofer (far‑field) approximation, which breaks down when the longitudinal extent of the particle shower becomes comparable to the distance between the shower and the antenna. At energies above 10^18 eV, electromagnetic cascades can stretch over several hundred meters, and proposed detector arrays (e.g., ARIANNA, ARA, IceCube‑Gen2) often place antennas only a few tens to a few hundred meters from the cascade. In this near‑field regime, phase delays, wavefront curvature, and boundary effects (refraction, reflection at the ice‑air or ice‑water interface) significantly distort the pulse shape, leading to systematic errors in energy reconstruction and flavor identification if far‑field formulas are used.

To overcome this, the authors develop a full‑wave numerical solver based on the finite‑difference time‑domain (FDTD) method. Maxwell’s equations are discretized on a three‑dimensional Cartesian grid that fully encloses the shower region. The source term is constructed from realistic charge‑excess distributions obtained from high‑energy shower simulators such as CORSIKA or ZHAireS, assuming a typical 20 % excess charge fraction. The dielectric properties of the medium (ice, salt, water) are allowed to vary with depth, temperature, and density, enabling the inclusion of realistic refractive‑index gradients. Boundary conditions incorporate both perfectly matched layers (PML) to absorb outgoing radiation and explicit interface conditions to model partial reflection and transmission at the ground or ice surface.

Because FDTD scales as O(N^3 · Δt⁻¹) (with N the number of grid points per dimension), a CPU‑only implementation would be prohibitively slow for the required spatial resolution (centimeter‑scale) and temporal resolution (sub‑nanosecond). The authors therefore port the algorithm to NVIDIA GPUs using CUDA. Each grid cell’s field update is independent, allowing thousands of CUDA cores to operate in parallel. Memory bandwidth is optimized by arranging field components in structure‑of‑arrays format and by exploiting shared memory for stencil operations. Multi‑GPU communication is handled via NVLink and MPI, achieving near‑linear scaling up to eight GPUs. Benchmarks show a speed‑up factor of 30–50× compared with a 64‑core CPU cluster, reducing a full 10 µs simulation from ~48 h to under 1 h.

The simulation results reveal several key physical insights. First, near‑field pulse waveforms exhibit multiple sub‑peaks and asymmetric tails, in stark contrast to the single, symmetric peak predicted by Fraunhofer theory. This multi‑peak structure arises from the coherent superposition of radiation emitted at different longitudinal positions along the shower, each arriving with a distinct phase delay. Second, the frequency spectrum retains more high‑frequency power (200 MHz–1 GHz) than far‑field models suggest, because the shorter path lengths reduce destructive interference at high frequencies. Third, interface effects are pronounced: reflections from the ice‑air boundary can produce secondary pulses delayed by several nanoseconds, while refraction within a gradient index profile bends the wavefront, altering the apparent arrival direction. Quantitatively, the near‑field model reduces systematic bias in reconstructed shower energy by 15–30 % relative to far‑field approximations across a range of distances (30 m–300 m) and shower energies (10^18–10^20 eV).

The authors also perform a sensitivity analysis, varying grid resolution, time step, and excess‑charge fraction to assess numerical stability and physical robustness. Convergence is achieved with a spatial step of 2 cm and a Courant‑limited time step of ~7 ps, ensuring that the highest relevant frequencies (≈1.5 GHz) are accurately captured. Error estimates indicate that the dominant uncertainty now stems from the underlying shower model rather than the electromagnetic solver.

In conclusion, the paper delivers a practical, GPU‑accelerated FDTD framework that enables accurate near‑field calculations of Askaryan Cherenkov radiation for UHE neutrino detection. By bridging the gap between full‑wave electromagnetics and high‑energy particle physics, the method provides essential inputs for antenna array design (optimal spacing, frequency band selection), trigger algorithm development (recognizing multi‑peak signatures), and data analysis pipelines (improved energy and flavor reconstruction). The authors argue that this tool will become indispensable for the next generation of radio‑based neutrino observatories, where precise modeling of near‑field effects is crucial for achieving the desired sensitivity to cosmogenic neutrinos and for unlocking new astrophysical insights.