REVIEW 2 major objections 3 minor 14 references
Collisional Fragmentation Support in TRACE
T0 review · 2 major / 3 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read TRACE now supports collisional fragmentation by accepting timesteps mid-collision and shepherding new particles through close encounters, matching slower integrators in a protoplanetary disk test while running up to 71x faster.
desk verdict A useful, honest software Note with a real new capability, but the new-particle bookkeeping has an untested edge case that should be acknowledged before the 'any collision prescription' claim is taken at face value. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The key mechanism is the 'irreversible timestep acceptance' rule combined with a 'new-particle close-encounter expansion'. When a collision changes the particle set mid-timestep, TRACE abandons the strict time-reversibility check and accepts the step; then it integrates all new particles and all previously close-encountering particles with BS or IAS15. This is what allows collision routines that add particles to run inside the hybrid scheme without breaking the integrator.
What would settle it
Run the same fragmentation prescription in a much denser disk (or with a Leinhardt-Stewart style model producing many fragments with wide velocity dispersion) and compare the energy error and final body distribution against IAS15; if TRACE's accepted irreversible timesteps accumulate secular errors or the survivor distribution diverges qualitatively, the claim that TRACE handles arbitrary collision routines in general is falsified.
Extended reading notes
Core claim
The central claim is that TRACE's collision handling can be made robust for arbitrary collision routines, including fragmentation that adds particles, by changing two parts of the algorithm. First, when a collision is detected in a BS or IAS15 substep during a close encounter, the timestep is flagged as irreversible and automatically accepted, bypassing the time-reversibility step that would otherwise reject the step because the particle set has changed. Second, every newly generated particle is assumed to be in a close encounter with every particle already flagged for close encounter and with every other new particle, so all of them are integrated with the high-accuracy close-encounter integrator for the remainder of the step. The paper shows that in a 30-embryo disk integrated for $10^{5}$ years, TRACE produces survivor counts and collision counts that agree qualitatively with MERCURIUS, IAS15, and BS, while achieving median speedups of 36x/47x/52x over MERCURIUS, 21x/22x/22x over BS, and 56x/71x/65x over IAS15 for the merger/fragmentation/hard-sphere prescriptions.
Load-bearing premise
The claim that TRACE robustly supports any collision routine rests on a single test with 30 embryos, one narrow radial range, and three simple collision prescriptions, none of which is a fully physical fragmentation model.
Editorial extensions
If this is right
- Large-N planet formation simulations that include fragmentation can run with TRACE instead of slower integrators like IAS15 or BS, giving speedups on the order of tens to over 70x.
- Users of REBOUND can now write arbitrary user-defined collision routines that add or remove particles mid-timestep and expect TRACE to handle them without the integrator crashing or producing obviously unphysical behavior.
- The qualitative outcomes of fragmentation simulations, such as the number of surviving bodies and the timing of collisions, should match those from MERCURIUS and direct integrators in similar low-to-moderate-N settings.
- The paper's result implies that the time-reversibility sacrifice made when a collision occurs does not, in practice, degrade the statistical evolution of the system in these tests.
Reading between the lines
- The validation here is limited to simple collision prescriptions and a single disk configuration; a natural extension would be to benchmark TRACE against IAS15 with a physically calibrated fragmentation model such as Leinhardt-Stewart, measuring not just survivor counts but mass and angular momentum distributions.
- The 'irreversible timestep acceptance' mechanism may introduce a subtle bias in systems with very frequent collisions, since each accepted irreversible step breaks symplecticity; one could test this by measuring energy error per collision in long integrations.
- If the speed advantage persists for higher particle numbers, TRACE could enable direct statistical studies of planet formation across many initial conditions that were previously computationally prohibitive, potentially changing how parameter surveys are designed.
- An independent check could compare TRACE's accepted-timestep collision handling against the same collision routine run purely with IAS15 to see whether the position of the collision within a step affects fragment trajectories.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The Note describes new collision-handling logic in TRACE, REBOUND's hybrid integrator. It explains that during a close encounter, collisions are checked at every BS or IAS15 substep; when a collision adds or removes particles mid-timestep, the step is flagged irreversible and accepted, and any new particles are assumed to be in close encounter with previously flagged particles and with each other. The authors validate the implementation on a 30-embryo protoplanetary disk at 0.1-0.5 AU integrated for 10^5 years with three simple collision prescriptions (perfect mergers, hard-sphere, and a simple fragmentation scheme), comparing survival curves, collision counts, and walltimes against MERCURIUS, IAS15, and BS. They report qualitative agreement and median speedups up to 71x (IAS15, fragmentation), and state that all updates are in the public REBOUND release.
Significance. If the central algorithmic claim is correct, this is a practically useful contribution: TRACE is the recommended REBOUND integrator for chaotic close-encounter systems, and adding fragmentation support with speedups of tens to seventy times would benefit large-N planet formation simulations. The manuscript is appropriately modest in several respects: it labels the fragmentation prescription as 'not a fully physical collision outcome,' it describes agreement only qualitatively, and it provides a public code link and random-realization comparisons against three independent integrators. These are real strengths. However, the advertised generality—that TRACE now supports 'any collision prescription'—rests on a specific new-particle handling rule in Section 2 that needs scrutiny before the significance claim is fully justified.
major comments (2)
- [Section 2] The new-particle close-encounter rule as written leaves a gap in collision detection. After an irreversible collision step, each new particle is assumed to be in close encounter only with (a) particles already flagged for close encounter in Step 1 and (b) other new particles. The text does not describe any check of new particles against non-flagged particles during the remainder of the accepted timestep, and collision detection is said to occur only during BS or IAS15 substeps. A high-speed fragment can therefore cross the physical radius of a previously non-flagged embryo within the same 5-day global timestep without a collision check. This case is not exotic for the authors' own fragmentation prescription, where small fragments are launched opposite the center-of-mass velocity. Please either specify an additional global collision check that covers new-particle/non-flagged-particle pairs in the accepted irreversible step, or restrict the 'any collision prescription' claim and add a test that deliberately forces this configuration.
- [Section 3] The benchmark does not quantitatively exercise the case that matters for the generality claim. The paper states that 'differences between integrators are not statistically significant,' and the single test architecture (30 embryos at 0.1-0.5 AU, 10^5 yr, three simple prescriptions, 10 realizations) is too coarse to reveal a rare missed fragment--embryo collision of the type described above. A targeted experiment with a high-ejection-speed fragmentation prescription, or with a much smaller critical switching radius, would show whether the Section 2 rule silently drops collisions. Without such a test, the qualitative agreement in Figure 1 supports the narrow claim that TRACE behaves comparably for this disk setup, but not the broader claim that arbitrary collision routines that add particles are supported robustly.
minor comments (3)
- [Section 3, Figure 1] One IAS15 run 'proved too computationally expensive to run to completion'; please state explicitly whether the median and speedup statistics are computed over 10 runs or over the 9 completed runs, and how the incomplete run is treated in the plotted curves.
- [Section 3, speedup sentence] The sentence 'The median speedup TRACE offers is 36x/47x/52x (MERCURIUS), 21x/22x/22x (BS) and 56x/71x/65x (IAS15) for the merger/fragmentation/hard-sphere prescriptions, respectively' lists the prescription order only at the end; adding the labels before each parenthetical or using a table would remove ambiguity.
- [Section 4] The code link points to a git branch named 'TRACE frag paper'; for a published Note, consider also pointing to a tagged release or an archived version so the exact code used for Figure 1 remains reproducible.
Circularity Check
No significant circularity: the collision logic and performance benchmarks are self-contained and validated against independent integrators.
full rationale
The paper's central claims are (1) that TRACE now implements a back-end collision framework that can remove and add particles mid-timestep, and (2) that on a 30-embryo test system TRACE matches the qualitative behavior of MERCURIUS, IAS15, and BS while running faster. Neither claim reduces to its own inputs. The collision handling described in Section 2 is an algorithmic procedure, not a derived prediction: collisions are detected during substepping, the step is flagged irreversible, and new particles are assigned to the close-encounter set by an explicit stated assumption. That assumption is a modeling limitation, not a circular definition of success. The performance test uses three independently defined collision prescriptions, including a simple fragmentation rule that is clearly stated to be 'not a fully physical collision outcome,' and compares TRACE against pre-existing, independently implemented integrators in the REBOUND ecosystem. The agreement is qualitative, and the speedups are measured runtimes; no free parameter is fitted to reproduce the comparison curves. The paper cites earlier work by overlapping authors (L24 for TRACE, Rein et al. 2019 for MERCURIUS, Rein & Spiegel 2015 for IAS15, Rein & Tamayo 2015 for WHFast), but those citations provide the algorithms and background rather than a forced uniqueness or a self-validating premise. No equation is defined in terms of the result it is claimed to predict, and no fitted quantity is renamed as a prediction. The reader's concern about missed encounters is a correctness or robustness issue, not a circularity issue, and is outside the scope of this pass.
Assumptions & free parameters
free parameters (2)
- critical switching radius =
3.63 Hill radii
- global timestep =
5 days
assumptions (4)
- domain assumption TRACE's hybrid scheme (WHFast for non-encounters, BS/IAS15 for close encounters) is accurate per L24.
- domain assumption Close encounter criterion based on mutual Hill radii is appropriate for this disk.
- domain assumption Collisions break time-reversibility and symplecticity, but these errors do not qualitatively affect system evolution.
- domain assumption The simple fragmentation prescription (one large, two small fragments along the collision axis) is sufficient to exercise the new collision machinery.
Cite this review
Pith. "Pith review of Collisional Fragmentation Support in TRACE." pith.science (2026). https://pith.science/paper/FV5CBSTM
@misc{pith2026250504399,
author = {Pith},
title = {Pith review of: Collisional Fragmentation Support in TRACE},
year = {2026},
howpublished = {\url{https://pith.science/paper/FV5CBSTM}},
note = {Machine review of arXiv:2505.04399}
}
read the original abstract
We present improved collision support for TRACE, a state-of-the-art hybrid integrator in REBOUND. TRACE now supports collisional fragmentation and can handle both removing and adding particles mid-timestep. We describe the back-end logic implemented for robust collision support, and compare TRACE's performance to other integrators including MERCURIUS on a large-N protoplanetary disk simulation with various collision prescriptions, a system which TRACE previously could not handle. TRACE matches the behavior of these integrators, while offering potentially vast speedups of over 70x. All updates described in this Note are available with the most recent public release of REBOUND.
Figures
Reference graph
Works this paper leans on
-
[1]
Chambers, J. E. 1999, MNRAS, 304, 793, doi: 10.1046/j.1365-8711.1999.02379.x
arXiv 1999
-
[2]
Childs, A. C., & Steffen, J. H. 2022, MNRAS, 511, 1848, doi: 10.1093/mnras/stac158
-
[3]
2006, Oberwolfach Reports, 3, 805
Hairer, E., Hochbruck, M., Iserles, A., & Lubich, C. 2006, Oberwolfach Reports, 3, 805
work page 2006
-
[4]
Hernandez, D. M., & Dehnen, W. 2023, MNRAS, 522, 4639, doi: 10.1093/mnras/stad657
-
[5]
Hunter, J. D. 2007, Computing in Science & Engineering, 9, 90, doi: 10.1109/MCSE.2007.55
-
[6]
Leinhardt, Z. M., & Stewart, S. T. 2012, ApJ, 745, 79, doi: 10.1088/0004-637X/745/1/79
-
[7]
Lu, T., Hernandez, D. M., & Rein, H. 2024, MNRAS, 533, 3708, doi: 10.1093/mnras/stae1982
-
[8]
2023, The Astrophysical Journal, 948, 41, doi: 10.3847/1538-4357/acc06d
Lu, T., Rein, H., Tamayo, D., et al. 2023, The Astrophysical Journal, 948, 41, doi: 10.3847/1538-4357/acc06d
Show all 14 references
-
[9]
Pham, D., Rein, H., & Spiegel, D. S. 2024, The Open Journal of Astrophysics, 7, 1, doi: 10.21105/astro.2401.02849
2024 arXiv
-
[10]
Flannery, B. P. 2002, Numerical recipes in C++ : the art of scientific computing
2002
-
[11]
Rein, H., & Liu, S. F. 2012, A&A, 537, A128, doi: 10.1051/0004-6361/201118085
2012 doi
-
[12]
Rein, H., & Spiegel, D. S. 2015, MNRAS, 446, 1424, doi: 10.1093/mnras/stu2164
2015 doi
-
[13]
2015, MNRAS, 452, 376, doi: 10.1093/mnras/stv1257 4
Rein, H., & Tamayo, D. 2015, MNRAS, 452, 376, doi: 10.1093/mnras/stv1257 4
2015 doi
-
[14]
M., Tamayo, D., et al
Rein, H., Hernandez, D. M., Tamayo, D., et al. 2019, MNRAS, 485, 5490, doi: 10.1093/mnras/stz769
2019 doi
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.