Reionization Emulator

This project explores how the universe's first galaxies reionized the intergalactic medium by modeling the kinetic Sunyaev-Zel'dovich (kSZ) effect during the Epoch of Reionization (Eor). I'm working under Dr. Paul La Plante at UNLV's LEADS Lab, using the Bridges-2 supercomputer at the Pittsburgh Supercomputing Center (PSC) to run large-scale cosmological simulation with the Zerion model.

The main goal of this project is to build a machine learning emulator that predicts the kSZ angular power spectrum (Cl) directly from reionization model parameters. This would allow for rapid exploration of cosmological parameter space without running new simulations.

Scientific Background

The kSZ effect measures distortions in the Cosmic Microwave Background (CMB) caused by free electrons moving during reionization. By simulating this process and computing the angular power spectrum, we can extract statistical information about when and how quickly reionization occurred.

Workflow Overview

1. Run Zerion simulations on Bridges-2 HPC with varying reionization parameters using Latin Hypercube Sampling (z_mean, alpha_zre, kb_zre, b0_zre).
2. Build dataset containing model parameters and select simulation outputs formated as HDF5 files
   • Model Parameters
     • alpha_zre - Controls how long reionization lasts. Bounded: [0.10 - 0.90]
     • kb_zre Determines how uneven the ionized regions are (higher values more clustered). Bounded: [0.10 - 2.0]
     • z_mean Sets the midpoint of reionization. Bounded: [7.0 - 9.0]
     • b0_zre Adjusts the overall strength or amplitude of the ionization field. Constant:
   • Simulation Output
     • tau - The optical depth to reionization; measures how many CMB photons were scattered by free electrons.
     • ksz_map - A 2D map of temperature fluctuations caused by the kinematic Sunyaev–Zel’dovich effect.
     • pk_tt - The 3D matter power spectrum P(k); shows how matter is distributed across different spatial scales within the simulation volume.
     • xmval_list - The average ionized fraction of the universe at different times; shows how reionization progressed.
     • zval_list - The redshift values that correspond to each point in the ionization history.
3. Compute angular power spectrum from each kSZ map using a 2D Fourier transform and add data into processed dataset.
   • ell- The angular frequency
   • cl_ksz- The raw angular power spectrum in uK^2 (variance per angular scale)
   • dcl- Basic uncertainty per bin
   • dcl_ksz- The rescaled spectrum
4. Train emulator to predict angular power spectrum given new parameters using a neural network.

Directory Structure

• src - Directory for the emulator architecture, training logic, model evaluation and data preprocessing.
• data - Directory containing the direct simulation output (raw), processed simulation data (processed), and test sets.
• scripts - Python scripts for running simulations, processing data, computing angular power spectrum.
• notebooks - Jupyter notebooks for analysis and visualization.
• results - Plots and charts documenting progress.

Acknowledgments

This work is part of ongoing research with Dr. Paul La Plante in the LEADS Lab at the University of Nevada, Las Vegas, using computing resources with the Pittsburgh Supercomputing Center.