eorcalc 0.1.1

GPU-accelerated reionization field calculator with power spectrum analysis for the Epoch of Reionization

EoRCaLC - Epoch of Reionization Calculator

EoRCaLC is a GPU-accelerated Python package for simulating reionization fields and calculating ionization evolution during the Epoch of Reionization (EoR). It combines high-performance GPU computing with sophisticated cosmological physics to model the ionization history of the universe.

Features

• 🚀 GPU Acceleration: High-performance calculations using CuPy (CUDA 12.x)
• 🌌 Reionization Field Simulation: Full 3D ionization field evolution with recombination
• 📊 Power Spectrum Analysis: Matter power spectrum calculations via CAMB
• 🔬 Ionization Physics: Advanced ionization/recombination modeling with mini-halos
• ⚡ Multi-scale Smoothing: Efficient top-hat filtering at multiple scales
• 📈 Optical Depth Calculation: Automatic Thomson scattering optical depth computation
• 💾 Data Management: Binary field I/O and CSV output for analysis

Installation

Prerequisites

• Python 3.8 or higher
• CUDA 12.x compatible GPU (for GPU acceleration)
• CUDA Toolkit 12.x installed

Basic Installation

pip install -e .

GPU Support (CUDA 12.x)

For GPU acceleration with CUDA 12.x:

pip install cupy-cuda12x

For CUDA 11.x users:

pip install -e ".[gpu-cuda11]"

Development Installation

pip install -e ".[dev]"

Dependencies

Core Dependencies

• numpy (≥1.20.0): Numerical computing
• scipy (≥1.6.0): Scientific computing and integration
• pandas (≥1.3.0): Data analysis and CSV output
• astropy (≥4.0): Astronomical calculations and units
• camb (≥1.3.0): Cosmological matter power spectrum
• massfunc (≥0.0.10): Halo mass function calculations
• cosfunc (≥0.1.0): Cosmological functions (n_H, dtdz, etc.)
• xxiop (≥0.1.0): Optical depth calculations
• cupy-cuda12x (≥12.0.0): GPU acceleration
• filelock (≥3.0.0): File locking for concurrent access

Module Overview

eorcalc.reion_field

Main module for reionization field simulation with GPU acceleration.

Main Function: reionization_calculator()

• Simulates 3D ionization field evolution across redshift
• Includes recombination effects and mini-halo feedback
• Multi-scale smoothing with top-hat filters
• Calculates optical depth and saves results

eorcalc.ioninti_gpu

GPU-accelerated ionization physics calculations.

Class: Ion

• Source term calculations (nion_interp, nion_st)
• Mini-halo absorption (nxi_interp, nxi_st)
• IGM recombination (dnrec_dz_path)
• GPU-optimized numerical integration

eorcalc.ioninti

CPU-based ionization calculations (alternative to GPU version).

eorcalc.iondiff

Ionization diffusion modeling with star formation rate density.

eorcalc.powerspec

Power spectrum and mass function calculations.

Class: MassFunctions

• Matter power spectrum via CAMB
• Halo mass functions (Sheth-Tormen, Press-Schechler, EPS)
• Custom transfer functions with scale-dependent modifications

eorcalc.special

Utility functions for data I/O and GPU interpolation.

• load_binary_data(): Load density fields from binary files
• TopHat_filter(): GPU-accelerated top-hat filtering
• xHII_field_update(): Ionization field updates
• interp1d_gpu(): GPU interpolation
• fstar(), xim(): Star formation efficiency and mini-halo feedback

Usage

Reionization Field Simulation

from eorcalc import reionization_calculator

# Run full reionization simulation
optical_depth = reionization_calculator(
    fesc=0.2,              # Escape fraction
    A2byA1=1.0,            # Transfer function amplitude ratio
    kMpc_trans=1e6,        # Transition scale [Mpc^-1]
    alpha=0.0,             # Power law index
    beta=0.0,              # Secondary index
    label='MH',            # Output label
    DIM=256,               # Grid dimension
    box_length=800,        # Box size [Mpc]
    save_on=True          # Save ionization fields
)

print(f"Optical depth: {optical_depth:.4f}")

Output:

• Ionization fields saved to reionf/{label}/rf_{z:.2f}.npy
• Ionization fraction history saved to csvfile/{label}.csv
• Prints redshift evolution and optical depth

GPU-Accelerated Ionization Physics

from eorcalc.ioninti_gpu import Ion
import cupy as cp

# Initialize at redshift z=7
ion = Ion(
    z=7.0,
    fesc=0.2,
    A2byA1=1.0,
    ktrans=1e6,
    alpha=0.0,
    beta=0.0
)

# Load density field (GPU array)
delta_field = cp.random.randn(256, 256, 256)

# Calculate source term
m_grid = ion.cosmo.rhom * (800/256)**3
source = ion.nion_interp(m_grid, delta_field)

# Calculate mini-halo absorption
minihalo = ion.nxi_interp(m_grid, delta_field)

# IGM neutral hydrogen
igm = ion.n_HI(delta_field)

print(f"Mean source: {cp.mean(source):.2e}")
print(f"Mean mini-halo: {cp.mean(minihalo):.2e}")

Power Spectrum and Mass Functions

from eorcalc.powerspec import MassFunctions

# Initialize with custom transfer function
cosmo = MassFunctions(
    A2byA1=1.0,      # Amplitude ratio
    kMpc_trans=1e6,  # Transition scale
    alpha=0.0,       # Power index
    beta=0.0         # Secondary index
)

# Calculate halo mass function at z=7
z = 7.0
M = 1e10  # Solar masses

# Sheth-Tormen mass function
dndm_st = cosmo.dndmst(M, z)

# Press-Schechter mass function
dndm_ps = cosmo.dndmps(M, z)

# Excursion set peaks with environmental dependence
deltaV = 0.5
Mv = 1e12
dndm_eps = cosmo.dndmeps(M, Mv, deltaV, z)

print(f"ST dndm: {dndm_st:.2e} Mpc^-3")
print(f"PS dndm: {dndm_ps:.2e} Mpc^-3")

Data I/O

from eorcalc.special import load_binary_data, TopHat_filter
import cupy as cp

# Load density field from binary file
delta_field = load_binary_data(
    'ktrans1e2/updated_smoothed_deltax_z006.00_256_800Mpc',
    DIM=256
)

# Convert to GPU array
delta_gpu = cp.asarray(delta_field)

# FFT for filtering
delta_ffted = cp.fft.rfftn(delta_gpu, norm="forward")

# Apply top-hat filter at R=10 Mpc
delta_smoothed = TopHat_filter(
    delta_ffted, 
    R=10, 
    DIM=256, 
    box_length=800
)

print(f"Original variance: {cp.var(delta_gpu):.4f}")
print(f"Smoothed variance: {cp.var(delta_smoothed):.4f}")

Physics Implementation

Reionization Model

The package implements a semi-numerical reionization model:

1. Source Term: $$n_{\rm ion} = f_{\rm esc} \cdot Q_{\rm ion} \cdot f_* \cdot \frac{\rho_{\rm b}}{\rho_{\rm m}} \cdot \int_{M_{\rm min}}^{M_{\rm max}} \frac{dn}{dM} , M , dM$$

2. Mini-halo Absorption: $$n_{\xi} = \frac{\rho_{\rm b}}{\rho_{\rm m}} \cdot \int_{M_{\rm J}}^{M_{\rm min}} \xi_{\rm ion}(M) \cdot \frac{dn}{dM} \cdot M , dM$$

3. Recombination: $$\frac{dn_{\rm rec}}{dz} = C_{\rm HII} \cdot x_{\rm HE} \cdot \alpha_A \cdot n_{\rm HI} \cdot Q_{\rm HII} \cdot (1+z)^3 \cdot \frac{dt}{dz}$$

4. Ionization Criterion: $$\bar{f}{\rm esc} \cdot n{\rm ion}(R) > n_{\rm HI}(R) + \bar{n}{\xi} \cdot n{\xi}(R)$$

Multi-scale Smoothing

The code performs filtering at multiple scales from cell size to 50 Mpc:

• Logarithmically spaced smoothing radii
• Top-hat filters in Fourier space
• Ionization propagates from large to small scales

Directory Structure

EoRCaLC/
├── eorcalc/                 # Main package
│   ├── __init__.py         # Package initialization
│   ├── reion_field.py      # Main reionization simulator
│   ├── ioninti_gpu.py      # GPU ionization physics
│   ├── ioninti.py          # CPU ionization physics
│   ├── iondiff.py          # Ionization diffusion
│   ├── powerspec.py        # Power spectrum & mass functions
│   └── special.py          # Utility functions
├── pyproject.toml          # Package configuration
├── README.md               # This file
└── LICENSE                 # MIT License

Input Data Format

The package expects density field data in binary format:

• Filename pattern: updated_smoothed_deltax_z{z:06.2f}_{DIM}_{box_length}Mpc
• Binary format: Little-endian float32
• Shape: (DIM, DIM, DIM)
• Values: Overdensity δ = ρ/ρ̄ - 1

Output Data

Ionization Fields

• Location: reionf/{label}/rf_{z:.2f}.npy
• Format: NumPy array (CuPy saved)
• Shape: (DIM, DIM, DIM)
• Values: Ionized fraction (0 to 1)

Ionization History

• Location: csvfile/{label}.csv
• Columns: z (redshift), ionf (ionization fraction)
• Sorted by decreasing redshift

Performance Notes

• GPU Memory: Requires ~2-4 GB VRAM for 256³ grids
• Speed: ~10-100x faster than CPU for large grids
• Multi-GPU: Not currently supported
• Precision: Uses float32 for GPU, float64 for critical calculations

Examples

See the Jupyter notebook cece.ipynb for complete examples and visualization.

Citation

If you use this package in your research, please cite:

@software{eorcalc,
  author = {Hajime Hinata},
  title = {EoRCaLC: GPU-Accelerated Reionization Calculator},
  year = {2025},
  url = {https://github.com/SOYONAOC/EoRCaLC}
}

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• CAMB for cosmological calculations
• CuPy team for GPU acceleration tools
• The reionization modeling community

Contact

• Author: Hajime Hinata
• Email: [e-mail]
• GitHub: SOYONAOC/EoRCaLC

Troubleshooting

CUDA Issues

# Check CUDA version
nvcc --version

# Install matching CuPy version
pip install cupy-cuda11x  # for CUDA 11.x
pip install cupy-cuda12x  # for CUDA 12.x

Import Errors

# Install missing dependencies
pip install cosfunc xxiop massfunc

# Or install all dependencies
pip install -e .

Memory Errors

• Reduce grid dimension (DIM)
• Process fewer redshift snapshots at once
• Use CPU version for very large simulations