🔭 astrophysics

ExTraSS: a Domain Decomposed 3D NLTE Radiative Transfer spectral synthesis code for nebular phase transients

This paper presents ExTraSS, a 3D NLTE radiative transfer code that utilizes a novel domain decomposition algorithm to overcome storage challenges and generate synthetic spectra for asymmetric supernovae in the nebular phase.

Original authors: Bart F. A. van Baal, Anders Jerkstrand

Published 2026-05-05

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Bart F. A. van Baal, Anders Jerkstrand

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a supernova as a massive, expanding firework that has just gone off. In the first few weeks, it's so bright and dense that we can only see the glowing outer shell, like looking at the surface of a cloud. But as time passes, the debris spreads out, becomes thinner, and eventually, the "fog" clears. This is the nebular phase. Suddenly, we can see deep inside the explosion, and the light we see comes from the entire structure, not just the surface.

The paper introduces a new computer program called ExTraSS (EXplosive TRAnsient Spectral Simulator) designed to recreate what these dying stars look like during this "clearing fog" phase. Here is how it works, explained simply:

1. The Challenge: A Room Full of People Trying to Talk

To understand a supernova, scientists need to track how light (photons) bounces around inside the expanding gas. In the early days, the gas is so thick that light gets trapped. But in the nebular phase, the gas is thin enough that light can travel far, but it's still complex.

The problem is that the explosion isn't a perfect sphere; it's lumpy and messy, like a crumpled piece of paper. To simulate this, scientists use a 3D grid (like a giant digital honeycomb) with hundreds of thousands of tiny cells.

In the past, trying to calculate how light moves between all these cells was like trying to hold a conversation in a stadium where everyone is shouting at once. The computer needed to remember millions of "conversation rates" (how often light hits an atom and excites it) for every single cell. This required so much memory that even the world's most powerful supercomputers would choke. It was like trying to carry a library of books in a backpack that can only hold a single notebook.

2. The Solution: The "Neighborhood" Strategy (Domain Decomposition)

The authors solved this memory problem with a clever trick called Domain Decomposition.

Imagine the giant stadium of shouting people is too big for one person to manage. Instead of one person trying to listen to everyone, they divide the stadium into four smaller neighborhoods.

The Workers: In each neighborhood, a team of workers (computer cores) handles the calculations for just that small area.
The Managers: Each neighborhood has one "Manager" core. The workers don't talk to the workers in other neighborhoods directly. Instead, they tell their Manager, "Hey, I have a message for the next neighborhood." The Managers then talk to each other to pass the messages along.

This way, no single computer has to hold the entire library of books in its backpack. It only needs to hold the books for its own neighborhood. This allows the scientists to run incredibly detailed simulations that were previously impossible.

3. How the Simulation Works

The program runs in a loop, like a chef tasting a soup and adjusting the seasoning:

The Explosion Model: They start with a 3D map of the exploding star (the "soup").
Gamma Rays: First, they calculate the energy from radioactive decay (the heat source) that powers the light.
The "Taste Test" (NLTE Solver): The computer calculates the temperature and how atoms are behaving in each cell. It asks: "Is this atom excited? Is it ionized?"
The Light Travel (Radiative Transfer): Then, it shoots virtual light rays through the 3D grid to see how they interact with the atoms.
The Loop: The results from the light travel change the temperature and atom states, so the computer goes back to step 3 and does it again. It keeps repeating this until the simulation stabilizes and matches the laws of physics.

4. Did It Work? (The Taste Test)

The authors tested ExTraSS by comparing it to other known programs and simpler 1D models (which are like looking at the star from just one angle).

The Results: The new program produced results that matched the older, trusted programs very closely.
The Difference: While the older programs were like looking at a flat map, ExTraSS is like looking at a 3D hologram. It confirmed that even though the star is messy and 3D, the physics holds up, and the new method is accurate.

Summary

ExTraSS is a new, super-efficient tool that allows astronomers to simulate the complex, 3D "afterglow" of exploding stars. By breaking the massive calculation into smaller, manageable neighborhoods and having them communicate through managers, the code can now handle the huge amount of data required to see the universe in high definition. This helps scientists figure out exactly what kind of stars exploded and how much energy they released, simply by looking at the light they left behind.

Technical Summary of ExTraSS: A Domain Decomposed 3D NLTE Radiative Transfer Spectral Synthesis Code

Problem Statement
Modeling the nebular phase of supernovae (SNe) presents a significant computational challenge. In this phase, ejecta densities have decreased sufficiently for the structure to become largely optically thin, meaning the entire volume contributes to emission. However, accurate modeling requires Non-Local Thermodynamic Equilibrium (NLTE) physics to account for complex microphysics, while radiative transfer (RT) effects cannot be ignored, particularly for hydrogen-rich core-collapse SNe. Furthermore, the asymmetric morphologies of SNe necessitate 3D input models.

A primary bottleneck in 3D NLTE RT is the generation and storage of photoexcitation rates. For a grid with $\sim 10^5$ cells, storing these rates for millions of transitions can require upwards of 1 TB of data, exceeding the memory capacity of single compute nodes on modern supercomputers. Previous 3D NLTE efforts have either been limited to 1D, used optically thin approximations, or employed "on-the-spot" treatments that neglect global radiation field effects. There was a need for a code capable of performing full 3D NLTE radiative transfer with global coupling between the radiation field and atomic level populations.

Methodology
The authors present ExTraSS (EXplosive TRAnsient Spectral Simulator), a code upgraded to perform full 3D NLTE radiative transfer. The code operates in a "snap-shot" mode, following photon transport independently of time, suitable for late phases where optical depths are low.

Grid and Geometry: ExTraSS utilizes a spherical coordinate grid, allowing for high resolution in the inner, slower-moving ejecta and larger cells at the outer radii. The angular variation of the radiation field is discretized using $N_{\theta, VA} \times N_{\phi, VA}$ viewers (default 20 $\times$ 20).
$\gamma$ -ray Transport: The code tracks radioactive decay (primarily $^{56}$ Ni and $^{56}$ Co) using a Monte Carlo approach with Compton scattering. Positrons are assumed to be locally trapped. The energy deposition from $\gamma$ -rays drives non-thermal heating, ionization, and excitation via the Spencer-Fano formalism.
NLTE Solver: The solver iteratively determines local temperature and level populations. It couples thermal and non-thermal physics, balancing heating (from $\gamma$ -rays and photoionization) against cooling (bound-bound, free-bound, and two-photon emission). The solver uses a Newton-Raphson scheme with adaptive damping to converge level populations.
Radiative Transfer (RT): The code implements a full ray-tracing technique to calculate the global radiation field. Rays are emitted from cell centers towards all viewing angles. The code calculates optical depths for photoionization (using hydrogenic approximations for excited states and Verner et al. data for ground states) and photoexcitation (using the Sobolev approximation for line interactions).
Domain Decomposition: To overcome memory limitations, ExTraSS employs a Domain Decomposition algorithm. The computational domain is split into $\phi$ $ϕ$ -based slices (azimuthal sectors), with each node handling a specific sector.
- Communication Architecture: Each node designates one "manager" core to handle inter-node communication, while the remaining "worker" cores perform ray tracing and local calculations.
- Optimization: The system uses non-blocking MPI communication to minimize worker idle time. Data is transferred in bundles (rays) rather than individually. The manager cores coordinate the transfer of rays between domains, ensuring that workers spend maximum time on computation rather than waiting for data.

Key Contributions

First 3D NLTE RT for SNe: This work represents, to the authors' knowledge, the first implementation of domain decomposition specifically for a supernova radiative transfer code, enabling full 3D NLTE calculations that were previously infeasible due to memory constraints.
Memory Optimization: The paper details the implementation of shared memory, "Jagged Arrays" (allocating memory only for existing species/levels), and the domain decomposition scheme, which scales memory usage inversely with the number of nodes used.
Algorithmic Refinements: The authors optimized the frequency of calls to the computationally expensive Spencer-Fano subroutine (using a 4+10 scheme: called on iterations 1, 4, and every 10 thereafter) and refined the damping schemes for the Newton-Raphson level population solver to balance convergence speed and stability.

Results and Validation
The authors validated ExTraSS through several benchmarks:

Convergence and Efficiency: Tests on the Dardel supercomputer (using 128-core nodes) showed that the code achieves a computational efficiency of $\sim 0.8$ for the radiative transfer module when load balancing is optimized. Global convergence (ratio of bolometric luminosity to $\gamma$ -ray deposition) is reached within $\sim 17$ iterations, typically stabilizing around a ratio of 0.96–1.0.
Comparison with SUMO (1D): ExTraSS was compared against the 1D code SUMO using a "1D-turned-3D" model (J18).
- Physical Conditions: Temperatures and electron fractions agreed well, though ExTraSS showed slightly higher temperatures in the inner ejecta due to lower neutral fractions.
- Rates: Photoionization and photoexcitation rates were generally compatible, though ExTraSS showed slightly higher rates in the innermost regions.
- Spectra: The synthetic spectra showed good agreement, with minor discrepancies in Fe II emission and H $\alpha$ strength, attributed to differences in the number of included atomic levels (ExTraSS caps at 100 levels per ion vs. hundreds in SUMO).
Comparison with Multi-Code Benchmark (Blondin et al. 2022): ExTraSS was tested against the 'toy06' model alongside ARTIS-nebular, CMFGEN, and SUMO.
- Physical Conditions: ExTraSS computed slightly lower temperatures in the inner ejecta and lower Cobalt ionization fractions compared to some codes, but results were generally consistent with the ensemble.
- Spectra: The spectral output agreed well with SUMO and the other codes, particularly in the 5100–6400 Å and >8800 Å ranges. Discrepancies were noted in the Fe emission peaks around 4600–5000 Å, where ExTraSS predicted lower luminosities, potentially due to the level population truncation or ionization differences.

Significance
The paper claims that ExTraSS is now a fully functional, domain-decomposed 3D NLTE radiative transfer spectral synthesis code. Its significance lies in its ability to handle the massive memory requirements of 3D NLTE physics, opening the path for highly accurate modeling of asymmetric supernova explosions in the nebular phase. While currently applied to core-collapse supernova models (specifically Type II), the authors state the code is designed to be applicable to a wide variety of input models and explosive transients, provided additional physical components (such as time-dependence, dust, or molecules) are incorporated in future updates. The work establishes a foundation for better constraining explosion properties, ejecta masses, and compositions through detailed spectral signatures.