Table of Contents
Fetching ...

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

Bart F. A. van Baal, Anders Jerkstrand

TL;DR

ExTraSS advances nebular-phase spectral modelling by delivering a fully 3D NLTE radiative-transfer code with domain decomposition to overcome extreme memory demands. It integrates gamma-ray transport, NLTE level populations with non-thermal excitation, and full RT through ray-tracing, iterating between the radiation field and atomic populations. Validation against 1D SUMO benchmarks shows compatible photoionization and photoexcitation rates, with reasonable level-population distributions, while memory-saving strategies and scalable MPI decomposition enable 3D application to large ejecta. The work lays a practical foundation for detailed, angle-dependent spectra of core-collapse and other explosive transients, with clear paths for future enhancements such as dust, molecules, time dependence, and additional microphysics.

Abstract

In the nebular phase, supernovae are powered by radioactive decay and continuously fade, while their densities have decreased enough such that the expanding nebula becomes (largely) optically thin and the entire structure can be studied. Models for the nebular phase need to take Non-Local Thermodynamic Equilibrium (NLTE) effects into account, while at the same time radiative transfer effects often cannot be ignored. To account for the asymmetric morphologies of SNe, 3D input ejecta models must be used. In this work, we present the \texttt{ExTraSS} (EXplosive TRAnsient Spectral Simulator) code, which has been upgraded to be fully capable of 3D NLTE radiative transfer calculations in order to generate synthetic spectra for explosive transients in the nebular phase, with a focus on supernovae. We solve the long-standing difficulty of 3D NLTE radiative transfer -- to manage generation and storage of millions of photoexcitation rates over $\sim10^{5}$ of cells -- by developing a new Domain Decomposition algorithm. We describe this new methodology and general code operations in detail, and verify convergence and accuracy.

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

TL;DR

ExTraSS advances nebular-phase spectral modelling by delivering a fully 3D NLTE radiative-transfer code with domain decomposition to overcome extreme memory demands. It integrates gamma-ray transport, NLTE level populations with non-thermal excitation, and full RT through ray-tracing, iterating between the radiation field and atomic populations. Validation against 1D SUMO benchmarks shows compatible photoionization and photoexcitation rates, with reasonable level-population distributions, while memory-saving strategies and scalable MPI decomposition enable 3D application to large ejecta. The work lays a practical foundation for detailed, angle-dependent spectra of core-collapse and other explosive transients, with clear paths for future enhancements such as dust, molecules, time dependence, and additional microphysics.

Abstract

In the nebular phase, supernovae are powered by radioactive decay and continuously fade, while their densities have decreased enough such that the expanding nebula becomes (largely) optically thin and the entire structure can be studied. Models for the nebular phase need to take Non-Local Thermodynamic Equilibrium (NLTE) effects into account, while at the same time radiative transfer effects often cannot be ignored. To account for the asymmetric morphologies of SNe, 3D input ejecta models must be used. In this work, we present the \texttt{ExTraSS} (EXplosive TRAnsient Spectral Simulator) code, which has been upgraded to be fully capable of 3D NLTE radiative transfer calculations in order to generate synthetic spectra for explosive transients in the nebular phase, with a focus on supernovae. We solve the long-standing difficulty of 3D NLTE radiative transfer -- to manage generation and storage of millions of photoexcitation rates over of cells -- by developing a new Domain Decomposition algorithm. We describe this new methodology and general code operations in detail, and verify convergence and accuracy.

Paper Structure

This paper contains 19 sections, 6 equations, 10 figures, 4 tables.

Figures (10)

  • Figure 1: A pseudo-code schematic for the NLTE_solver and the EXCION_solver subroutines. The EXCION_solver is responsible for determining the level populations of all elements present inside a cell, while the NLTE_solver combines this with finding the temperature $T_\text{cell}$ through the T_solver.
  • Figure 2: A pseudo-code schematic for the full program flow of ExTraSS, and how the new RayTraceGrid interacts with the previous NLTE_solver. If there is no previous solution present, the NLTE_solver has to run first, to be able to generate the rays which are transported by the RayTraceGrid module.
  • Figure 3: An equatorial slice of the 3D grid, to show how this is broken into four slices by the domain decomposition scheme, with the main communicating cores on each node marked, and the communication chains represented by the arrows. In this example, the nodes each have 128 cores, but this number will vary for different clusters.
  • Figure 4: A schematic overview of all the communication streams for the four-domain setup from Figure \ref{['fig:DoDec_sketch']}. The grid of small blocks represents the 127 workers (cores 1-127 on each node) per node, the small arrows communication between each worker and their manager, and the bigger arrows between the different managers. Each node has its own colour, with the small arrows indicating both the "origin" of the data (base) as well as the "target" (head). Each arrow represents a different MPI communication stream -- the small arrows represent all 127 streams between the workers and their manager.
  • Figure 5: A pseudo-code schematic of how the non-blocking MPI communication is implemented. This shows the communication steps taken by the manager towards the workers on the same node (lines 2-5), as well as cross-node to other managers (lines 6-14). The same logic is used for the "downstream" side. The data received from another node is copied to a local buffer, such that the non-blocking IRECV request to the other manager can be made immediately.
  • ...and 5 more figures