Source-linked AI summary
Multi-physics simulations using a hierarchical interchangeable software interface
Simon Portegies Zwart, Stephen McMillan, Inti Pelupessy, Arjen van Elteren
TL;DR
Growing astrophysical simulations are difficult to maintain and extend when multi-physics components are built into monolithic codes. This paper uses AMUSE to combine independent solvers for embedded star clusters, finding that star formation efficiency is a poor predictor of cluster survival.
Problem
Monolithic multi-physics simulation codes create maintenance and expansion challenges as astrophysical simulations incorporate more physical processes.
Source: §1 · p. 1Method
Results
Star formation efficiency is a poor predictor of a cluster’s final state; the simulations identify early stellar winds and the first supernova as major episodes of gas loss.
Source: §5 · p. 4Takeaways & Limitations
Cluster survival may depend on the masses and orbits of a few massive stars, with small differences in the most massive stars potentially crucial.
Source: §5 · p. 6Radiative feedback remains limited in the models, and the authors identify radiative-transfer modeling as a next step.
Source: §5 · p. 6
Abstract
from arXiv · showhide
We introduce a general-purpose framework for interconnecting scientific simulation programs using a homogeneous, unified software interface. Our framework is intrinsically parallel, and conveniently separates all components in memory. It performs unit conversion between different modules automatically and defines common data structures to communicate across different codes. We use the framework to simulate embedded star clusters. For this purpose we couple solvers for gravitational dynamics, stellar evolution and hydrodynamics to self consistently resolve the dynamical evolution simultaneousy with the internal nuclear evolution of the stars and the hydrodynamic response of the gas. We find, in contrast to earlier studies, that the survival of a young star cluster depends only weakly on the efficiency of star formation. The main reason for this weak dependency is the asymmetric expulsion of the embedding gas from the cluster.
1. Introduction
Growing demand for more detailed astrophysical simulations has made maintaining and extending monolithic multi-physics codes increasingly difficult. AMUSE offers an alternative built from independently maintained solvers, and this paper applies it to the survival of embedded star clusters.
1. Introduction
- Monolithic multi-physics codes make maintenance and expansion difficult as simulations grow more detailed.Legacy design choices can also hamper development, limit scaling on parallel computers, and make maintenance almost impossible.Source: §1 · p. 1
- AMUSE combines well-tested, independently maintained community solvers into applications with homogeneous interfaces.Its software suite can support individual tasks or be reassembled to combine solvers across astrophysical domains.Source: §1 · p. 2
- The paper combines three AMUSE components to examine how star formation efficiency relates to the survival of embedded star clusters.Source: §1 · p. 2
2. AMUSE
AMUSE hides independently written community codes behind physically motivated interfaces, allowing users to combine them into numerical experiments. A Python script manages the modules, while operator splitting couples the gas and gravitational dynamics.
2. AMUSE
- Physically motivated interfaces hide the complexity and numerical implementation of independently written community codes.The interfaces expose standard building blocks that users can combine into applications and numerical experiments.Source: §2 · p. 2
- A Python user script selects and manages community modules, while each module provides MPI communication, unit handling, and an object-oriented data model.The high-level script prioritizes algorithmic flexibility and ease of programming; computational cost is concentrated in component codes.Source: §2 · p. 2
- The cluster model combines direct N-body dynamics, stellar evolution, and SPH hydrodynamics to model stars and intracluster gas.The gas includes both primordial material and gas released by stellar winds and supernovae.Source: §2 · p. 2
- BRIDGE uses semi-symplectic operator splitting to couple gas and gravitational evolution across separate codes.The paper adopts operator splitting within AMUSE rather than implementing the coupling in one monolithic code.Source: §2 · p. 3
3. Initial conditions
The simulations use 1,000-star clusters with gas, shared Plummer distributions, and a Salpeter initial mass function. Additional realizations and gas-particle counts test sensitivity to stochastic stellar masses and gas-dynamics resolution.
3. Initial conditions
- Each simulated cluster contains gas and N = 1000 stars in Plummer distributions with the same characteristic radius.Stellar masses follow a Salpeter IMF from 0.1 to 100 M⊙, constrained to a maximum mass of approximately 22 M⊙ and assigned independently of stellar positions.Source: §3 · p. 3
- The study presents models A2 and A5 and tests IMF-realization and gas-dynamics-resolution effects with additional simulations.Multiple random IMF realizations examine variation in high-mass-star counts; varying gas-particle numbers tests resolution dependence.Source: §3 · p. 3
4. Results
Models A2 and A5 show different cluster-binding histories despite differing feedback efficiencies. Their gas-expulsion timing differs, while the simulations use Π—the age-to-crossing-time ratio—to distinguish bound from ballistic expansion.
4. Results
- The models use feedback efficiencies ffb = 0.1 for A2 and ffb = 0.01 for A5, defining ffb as the fraction of stellar wind and supernova energy deposited as ISM thermal energy.Source: §4 · pp. 3–4
- In A2, stellar winds inflated buoyant bubbles into a free-flowing wind by 4.37 Myr, and the cluster ISM was ejected at approximately 9.5 Myr.The gas visible then originated from the strong AGB wind of the most massive progenitor, about 21 M⊙.Source: §4 · p. 4
- A5 retained most cluster gas until its first supernova at 9.54 Myr, whereas A2’s stronger feedback had expelled most gas earlier.Just before the supernova, A5 was much more compact than A2; the simulations used the same initial IMF realization and stellar positions.Source: §4 · p. 4
- A5 reached a bound state within a few Myr and remained marginally bound to about 30 Myr, while A2 did not reach a bound state until after about 30 Myr.Here Π < 1 indicates ballistic expansion and Π > 1 indicates a bound state.Source: §4 · p. 4
5. Conclusions
The simulations find that star formation efficiency poorly predicts cluster survival, while early gas loss and subsequent cluster relaxation help explain this weak dependence. Outcomes also vary with the most massive stars, and surviving clusters are strongly mass segregated.
5. Conclusions
- Star formation efficiency is a poor predictor of a cluster’s final state: early Wolf-Rayet winds expel much gas, and the interval before the first supernova lets clusters relax and become more resilient to primordial-gas loss.The first supernova blows away most remaining gas, but the preceding winds have already allowed much of it to escape with little damage.Source: §5 · p. 4
- Statistical variation in the initial mass function can strongly affect early evolution, and differences of only a few M⊙ in the most massive stars may determine whether a cluster survives.The paper notes that clusters with a slight enhancement of massive stars may dissolve, while others may fare better if their most massive stars are more separated in mass.Source: §5 · p. 6
- Surviving clusters are strongly mass segregated, with a degree of segregation matching that required to explain observations of the Pleiades.Source: §5 · p. 6
- Radiative feedback remains limited in these models; the authors identify adopting a radiative-transfer code as the next improvement.Source: §5 · p. 6