A Comparative Study of the Streaming Instability: Unstratified Models with Marginally Coupled Grains
Pith reviewed 2026-05-15 15:52 UTC · model grok-4.3
The pith
Seven codes reproduce the streaming instability sequence of growth, filaments, and saturation, with dust model as main source of quantitative variation at moderate resolution.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
All seven codes reproduce the characteristic sequence of exponential growth, filament formation, and turbulent saturation. Quantitatively, the dust model remains the dominant source of variation at moderate resolution: particle-based simulations reach higher peak densities and exhibit broader high-density tails than fluid-based models at 512 squared resolution, although increasing the number of particles brings their initial maximum density evolution into close agreement with that of dust-fluid models. At 1024 squared, these differences diminish substantially, indicating better agreement of the saturated-state statistics across dust treatments.
What carries the argument
The unstratified streaming instability with dimensionless stopping time of unity, evolved across finite-volume and finite-difference codes that treat dust either as Lagrangian particles or as a pressureless fluid.
If this is right
- The core sequence of the streaming instability is robust across different numerical methods.
- Quantitative density statistics depend on dust treatment at moderate resolution.
- Raising resolution reduces differences between particle and fluid dust models.
- Particle implementations commonly encounter load imbalance in parallel execution.
- GPU runs are two to three times more energy efficient and scale better at high resolution.
Where Pith is reading between the lines
- Planetesimal formation predictions that rely on peak densities should use resolutions of at least 1024 squared to limit sensitivity to dust modeling choice.
- The energy advantage of GPU implementations may shift large parameter studies of disk instabilities toward GPU-based codes.
- Future work could test whether the same convergence pattern holds when vertical stratification or a range of stopping times is included.
Load-bearing premise
The unstratified setup with fixed stopping time of unity together with the chosen statistical diagnostics is enough to judge robustness across codes without missing resolution-dependent or model-specific artifacts.
What would settle it
If a higher-resolution run or additional code shows persistent large differences in maximum density distributions between particle and fluid dust treatments even at 2048 squared, the reported convergence would be falsified.
read the original abstract
The streaming instability is a leading mechanism for concentrating solids and initiating planetesimal formation in protoplanetary disks. Although numerous studies have explored its linear growth, nonlinear evolution, and implications for planet formation, the diversity of numerical methods and dust treatments used across the literature has made it difficult to assess which features of the instability are physically robust and which arise from code-dependent choices. We present the first systematic comparison of seven hydrodynamic codes--spanning finite-volume and finite-difference schemes and modeling dust either as Lagrangian particles or as a pressureless fluid--applied to the unstratified streaming instability with a dimensionless stopping time of unity. All codes reproduce the characteristic sequence of exponential growth, filament formation, and turbulent saturation, demonstrating broad agreement in the qualitative behavior of the instability. Quantitatively, however, the dust model remains the dominant source of variation at moderate resolution: particle-based simulations reach higher peak densities and exhibit broader high-density tails than fluid-based models at $512^2$ resolution, although increasing the number of particles brings their initial maximum density evolution into close agreement with that of dust-fluid models. At $1024^2$, these differences diminish substantially, indicating better agreement of the saturated-state statistics across dust treatments. In terms of computational performance, most particle implementations suffer from imbalanced parallelized loads, while execution on a GPU is at least two to three times more energy efficient and scales better at higher resolutions than on CPUs. Given the intrinsic stochasticity of this nonlinear system, only statistical diagnostics remain meaningful across codes.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript presents the first systematic comparison of seven hydrodynamic codes (finite-volume and finite-difference, with dust modeled as Lagrangian particles or pressureless fluid) applied to the unstratified streaming instability at fixed stopping time τ_s=1. All codes reproduce the qualitative sequence of exponential growth, filament formation, and turbulent saturation. Quantitatively, dust treatment dominates variations at 512² resolution (particle models show higher peak densities and broader high-density tails), but these differences diminish substantially at 1024²; performance comparisons note load imbalance in particle codes and GPU efficiency advantages.
Significance. If the results hold, the work provides a valuable benchmark establishing that the core qualitative behavior of the streaming instability is robust across numerical methods and dust treatments, while quantifying how resolution and particle number control convergence of saturated-state statistics. This directly aids interpretation of planetesimal-formation simulations and highlights practical trade-offs in code choice and hardware.
major comments (2)
- [Results (quantitative comparison at 512² vs 1024²)] The central claim of convergence at 1024² relies on statistical diagnostics, but the manuscript does not report formal error bars or convergence metrics (e.g., L2 norms on density PDFs or growth-rate uncertainties) that would make the quantitative agreement falsifiable; this weakens the assertion that differences 'diminish substantially' without a clear threshold.
- [Methods and setup] The unstratified, fixed-τ_s=1 setup is appropriate for the stated scope, yet the paper does not test sensitivity to small variations in stopping time or box size; given that the instability is known to be sensitive to these parameters in the literature, a brief robustness check would strengthen the claim that the reported sequence is representative.
minor comments (2)
- [Figures] Figure captions should explicitly state the number of particles used in each particle-based run and the precise definition of 'peak density' (e.g., maximum cell value or smoothed).
- [Performance comparison] The performance section would benefit from a table summarizing wall-clock time per timestep and energy consumption across all seven codes at both resolutions.
Simulated Author's Rebuttal
We thank the referee for the positive assessment and recommendation for minor revision. We address each major comment below with targeted revisions to strengthen the quantitative claims and discussion of scope.
read point-by-point responses
-
Referee: [Results (quantitative comparison at 512² vs 1024²)] The central claim of convergence at 1024² relies on statistical diagnostics, but the manuscript does not report formal error bars or convergence metrics (e.g., L2 norms on density PDFs or growth-rate uncertainties) that would make the quantitative agreement falsifiable; this weakens the assertion that differences 'diminish substantially' without a clear threshold.
Authors: We agree that formal metrics would make the convergence statement more rigorous. In the revised manuscript we will add L2 norms between the time-averaged density PDFs at 512² and 1024² for each dust treatment, together with standard deviations computed from multiple independent realizations to furnish error estimates on peak densities and growth rates. These additions will provide a clear, falsifiable threshold for the reduction in differences. revision: yes
-
Referee: [Methods and setup] The unstratified, fixed-τ_s=1 setup is appropriate for the stated scope, yet the paper does not test sensitivity to small variations in stopping time or box size; given that the instability is known to be sensitive to these parameters in the literature, a brief robustness check would strengthen the claim that the reported sequence is representative.
Authors: We recognize the known sensitivity of the streaming instability to τ_s and domain size. Running the full seven-code suite across additional parameter values lies outside the intended scope of this focused code-comparison study and would require prohibitive resources. In revision we will insert a concise paragraph in the discussion that cites the relevant literature on parameter dependence and justifies the choice of the standard τ_s=1, unstratified configuration as the appropriate benchmark case; we will also note that the qualitative sequence remains robust across the modest variations already explored in the broader literature. revision: partial
Circularity Check
No significant circularity identified
full rationale
The paper is a comparative numerical study of the streaming instability across seven hydrodynamic codes using direct simulations of the unstratified, fixed stopping-time case. No derivation chain, first-principles prediction, or fitted parameter is claimed; all reported outcomes (exponential growth, filament formation, saturation statistics) are generated by running the codes themselves. The central claim of broad qualitative agreement with dust-model variation at moderate resolution is therefore an empirical observation, not a reduction to self-defined inputs or self-citations. No load-bearing step reduces by construction to the paper's own assumptions.
Axiom & Free-Parameter Ledger
free parameters (2)
- stopping time =
1
- grid resolution
axioms (2)
- standard math Standard hydrodynamic equations govern the gas-dust interaction in an unstratified disk
- domain assumption Dust can be modeled either as discrete Lagrangian particles or as a pressureless fluid
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.