Pith. sign in

REVIEW 2 major objections 5 minor 51 references

An adaptive fast multipole Poisson solver brings linear complexity to AMR grids and conserves momentum better than multigrid at refinement boundaries.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 12:35 UTC pith:ZAJM2325

load-bearing objection Solid AMR/ATS extension of the FMM Poisson solver with a clever force-symmetry trick and persuasive binary-orbit results; the ATS claim needs a harder test and the code should be public. the 2 major comments →

arxiv 2607.19489 v1 pith:ZAJM2325 submitted 2026-07-21 astro-ph.IM astro-ph.GAphysics.comp-ph

A Scalable Fast Multipole Method Poisson Solver for the RAMSES code: II. Adaptive Mesh Refinement and Adaptive Time Stepping

Jun-Young Lee , Romain Teyssier This is my paper
classification astro-ph.IM astro-ph.GAphysics.comp-ph
keywords fast multipole methodPoisson solveradaptive mesh refinementadaptive time steppingforce symmetrymomentum conservationmultigrid comparisonRAMSES
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper extends a linear-complexity fast multipole method (FMM) Poisson solver so it can run on the adaptive mesh refinement and adaptive time stepping of the RAMSES code. It introduces one FMM tree per refinement level, a merged tree that aggregates multipoles across active levels, and a new 'nearest field' region that makes coarse and fine cells interact mutually, preserving force symmetry across coarse-fine interfaces. On an NFW halo the solver agrees with the reference multigrid solver to 0.51% mean relative difference (1.55% maximum), and on a binary-orbit test multigrid spirals inward while FMM tracks the analytic orbit for roughly three orbits. The paper also reports better strong scaling for FMM as MPI rank counts grow, with the largest gains on well-populated AMR hierarchies. A sympathetic reader would care because it offers an alternative Poisson solver that sidesteps multigrid's coarse-fine boundary errors without sacrificing accuracy.

Core claim

The central claim is that FMM can be made fully compatible with cell-based AMR and sub-cycled time stepping by building a separate sparse FMM tree for each refinement level and merging them for neighbor searches, and that the resulting forces agree with multigrid to percent level while conserving momentum better. The key device is the nearest field: at the bottom of the hierarchy, a cell in direct contact with the target cell is treated as a source if it is at the same refinement level; if it is further refined, the direct force is computed from its finer children. Because RAMSES's graded octree guarantees neighboring cells differ by at most one level, this mutual treatment exactly preserves

What carries the argument

Multiple sparse FMM trees Tℓ, one per AMR level, carry multipoles for their own refinement level so that inactive (coarse) levels can keep stale multipoles while active (fine) levels are advanced during sub-cycling. A merged tree Ṯℓ+1 aggregates multipoles of all finer levels to speed up repeated neighbor searches. The nearest field is the set of AMR cells in direct contact with the target cell; if any neighbor is refined one level deeper, P2P forces are computed from those finer cells instead of the coarse cell, which enforces the mutual coarse-fine interaction needed for momentum conservation. The whole interaction geometry is inherited from the AMR grid, with no opening-angle criterion.

Load-bearing premise

The coarse-fine force-symmetry claim depends on RAMSES's graded-octree rule that neighboring cells differ by at most one refinement level; on a mesh that allowed deeper jumps, the nearest field would have to include deeper cells and the mutual-force construction would need reworking.

What would settle it

Take the binary-orbit setup with (ℓmin, ℓmax) = (6,8) and set nexpand=1, as in the paper's Appendix A: FMM should keep the particles on nearly circular orbits while multigrid spirals inward. A run that shows FMM also decaying, or multigrid remaining stable, would undercut the central conservation claim. A second check would disable the graded-octree condition (or use a mesh with a two-level jump) and test whether the FMM orbit remains stable, since the nearest-field argument no longer applies.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • With percent-level agreement on identical AMR meshes, FMM can serve as a drop-in alternative to multigrid for isolated-boundary simulations in RAMSES.
  • Because FMM evaluates the potential via Taylor expansions instead of interpolated Dirichlet boundary conditions, it remains stable across coarse-fine interfaces even when the refined region is barely expanded; multigrid requires mesh smoothing to avoid orbital decay.
  • The per-level tree structure keeps coarse multipoles stale but usable during sub-cycling, so adaptive time stepping incurs only O(N_active) merging overhead per solve rather than a full rebuild.
  • Strong-scaling measurements show FMM overtakes multigrid as MPI ranks increase, especially when each refinement level holds many octs, so the solver's advantage grows on larger parallel configurations.
  • The design's fixed geometry and precomputed Cartesian translation kernels make the method a candidate for GPU acceleration, as the authors state.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the nearest-field symmetry argument generalizes, other cell-based AMR codes with graded octrees could port the same construction without needing dual-tree walks or elliptic matching.
  • A natural testable extension is to add multipole drifting for inactive levels, which the authors list as future work; that would likely restore exact momentum conservation under sub-cycling where current FMM still shows a phase shift.
  • The paper's evidence suggests that the dominant AMR error in multigrid is not the solver's accuracy on a fixed mesh but its boundary treatment at coarse-fine interfaces; FMM's box-like expansion discontinuities appear less harmful dynamically, a distinction worth isolating in a controlled interface-crossing experiment.
  • For periodic domains, the paper suggests a hybrid PM-FMM split; one could test whether the short-range truncated FMM kernel retains the momentum-conservation advantage observed here.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper extends the unigrid FMM Poisson solver from Paper I to RAMSES AMR and ATS. It introduces per-refinement-level FMM trees, a merged tree for active levels, and a "nearest field" decomposition intended to preserve force symmetry across coarse-fine interfaces. The solver is compared with the reference multigrid solver on identical AMR meshes: NFW halo (0.51% mean relative difference), double-sphere potential, binary-orbit conservation diagnostics, hydrostatic NFW halo with AMR and ATS, a Boss-Bodenheimer protostellar collapse, and strong-scaling experiments. The paper claims percent-level agreement with MG, better momentum conservation across coarse-fine interfaces, and better parallel scalability at higher MPI ranks while MG remains faster on a single rank.

Significance. If the claims hold, the paper makes a useful contribution to self-gravitating AMR simulations: an O(N) FMM-based Poisson solver that works within RAMSES's existing AMR/ATS machinery, with no free parameters fitted to the reported accuracy and with validation against analytic solutions and an independent multigrid solver. The new algorithmic concepts (nearest field, merged tree) are clearly described. The main limitation is that the ATS-compatibility claim is not yet quantitatively supported: the temporal synchronization error from stale multipoles is acknowledged but not bounded or isolated in a dedicated test, and the code is not publicly deposited.

major comments (2)
  1. [§3.3, Algorithm 1; §4 ATS discussion] The abstract and conclusions claim the solver is "fully compatible with adaptive time stepping," but the manuscript itself states that "the coarse inactive multipoles are not drifted during each sub-cycle and are therefore subject to a temporal synchronization error." No analytical bound or dedicated numerical test isolates this error. The ATS binary-orbit test mentioned in §4 is not associated with any figure or table, so it cannot be inspected. The hydrostatic NFW test and the protostellar collapse are run with ATS and show global agreement with MG, but they do not separate the stale-multipole error from other effects. Please add either a quantitative estimate of the stale-multipole error or a designed experiment (e.g., a particle crossing a coarse-cell boundary during a sub-cycle, or a rapidly rearranging coarse mass distribution), or soften the "fully compatible" claim.
  2. [§4, Figures 5-6 and Appendix A] The claim that FMM shows better momentum conservation "across coarse-fine AMR interfaces" is supported only indirectly. In the binary-orbit tests, the charges are always forced to reside on the finest level, and with nexpand=2 the refined region is deliberately enlarged so the particles remain away from the interfaces. Appendix A reduces the buffer, but the particles still remain on the finest level and never cross a coarse-fine interface. Thus the tests demonstrate robustness to interface proximity, but not the behavior of the momentum-conservation properties when a particle actually crosses an interface. Please either add such a crossing test or qualify the abstract and conclusion to state that the improved conservation holds in the presence of refined regions with adequate buffers, rather than across interfaces in general.
minor comments (5)
  1. [Algorithm 3] The loop bound uses "ℓ_bound," which is not defined in the text. It should presumably be ℓ_min.
  2. [Algorithm 4] The else branch at lines 15-16 repeats the same expression "Φℓ += K_{j→ℓmax} M_{ℓmax;ℓmax−1}" as line 8. This is likely a typo; please clarify the intended operation.
  3. [Figure 7] The text says the figure shows the mass conservation error, but the four panels displayed are ΔE_pot, ΔE_kin, ΔE_int, and ΔE_tot. Either add a mass-conservation panel or remove the reference to mass conservation in the caption/text.
  4. [Data availability] The code is said to be "available from the authors upon reasonable request." For a methods paper, a public repository or DOI would substantially improve reproducibility and transparency.
  5. [§4, NFW comparison] The phrase "the 90th- and 99-th percentiles of the mean relative difference" is confusing; presumably it should read "percentiles of the relative difference distribution."

Circularity Check

0 steps flagged

No significant circularity: AMR/ATS extension is validated against analytic solutions and an independent MG solver; only minor self-citations to Paper I for parameter choices.

full rationale

The paper's central numerical claims are supported by external, parameter-free benchmarks rather than by fitted inputs or by definition. For example, the NFW halo comparison reports a 'mean relative difference of 0.51% and a maximum difference of only 1.55%' against the independent multigrid solver (Section 4), the double-sphere test is compared with the 'analytical solution' (Figure 4), and the binary-orbit conservation claim is measured against 'the analytical solution' over roughly three orbits (Figures 5-6). No free parameter is fitted to these reported accuracies. The nearest-field construction is an algorithmic design input, not a derived prediction. The paper states: 'If any of these neighboring cells are further refined, we calculate the direct force from the finer cells. This treatment exactly preserves force symmetry across coarse-fine interfaces' (Section 3.2). This is a definitional design choice, and the paper immediately concedes that the overall method 'does not observe strict symmetry' and only enforces mutuality 'via the nearest field decomposition where the forces are most influential' (Section 4). The subsequent orbital test is empirical, not a logical consequence of the definition alone, and the paper does not claim global momentum conservation. The ATS limitation is explicitly acknowledged: 'the coarse inactive multipoles are not drifted during each sub-cycle and are therefore subject to a temporal synchronization error' (Section 3.3). This is an accuracy/completeness caveat and a sharp reviewer criticism, but it is not a case where a prediction reduces to its input. The same passage reports an ATS binary test, and the abstract's 'fully compatible' claim is a capability claim rather than a derived theorem; its weakness is a correctness risk, not circularity. The remaining dependence on Paper I is limited to parameter choices and the unigrid kernel: 'Following Paper I, we use p=3' (Section 2) and the footnote fixing 'Δℓ=1 for the AMR and ATS algorithm'. These self-citations are not load-bearing for the AMR/ATS extension, because the current paper re-validates the resulting solver against independent MG and analytic references. No uniqueness theorem, forbidden alternative, or ansatz is smuggled in via self-citation. Overall, the derivation chain is externally grounded; the only circularity-related observation is a minor, non-load-bearing reliance on the authors' prior paper.

Axiom & Free-Parameter Ledger

4 free parameters · 7 axioms · 0 invented entities

The central accuracy claims depend on hand-set algorithmic parameters (p=3, Δℓ=1, near-field size) and on the graded-octree geometry. These are not fitted to the test outputs, but they are not derived either. No new physical entities are introduced; 'nearest field' and 'merged tree' are algorithmic concepts, not entities with independent falsifiable handles.

free parameters (4)
  • FMM multipole/local expansion orders (n=2, p=3) = n=2 (P2M/M2M), p=3 (local expansion)
    Chosen in Paper I and reused here; controls truncation of the multipole and Taylor expansions and therefore force accuracy. Not fitted to test data, but the reported percent-level agreement is conditional on it (§2).
  • FMM-AMR level offset Δℓ = 1
    Fixes the near-field kernel size at the bottom of the downward pass. Footnote 1 states Δℓ=2 gives nearly identical accuracy but worsens scaling, so Δℓ=1 is fixed for the AMR/ATS implementation.
  • MG reference tolerance ε = 1e-6
    The reference multigrid solver uses a tolerance of 1e-6, typically about 6 V-cycles. All FMM-vs-MG comparisons inherit this choice (§4).
  • Mesh smoothing parameter nexpand = 2 (binary orbit tests)
    Test setup parameter used to keep particles away from coarse-fine interfaces. The paper shows MG results degrade substantially with nexpand=1 (Appendix A), so this parameter affects the reported momentum-conservation comparison.
axioms (7)
  • standard math Multipole expansions of the 1/r Green's function converge for well-separated cells (Greengard & Rokhlin 1987).
    The FMM operations P2M, M2M, M2L, and L2L in Section 2 rely on this classical result.
  • domain assumption Truncation at multipole order n=2 and local order p=3 yields percent-level accuracy for the tested AMR geometries.
    This is validated empirically in Section 4 but is not derived; the accuracy claim depends on this numerical approximation being adequate for all intended AMR configurations.
  • domain assumption RAMSES AMR grids are graded octrees: neighboring cells differ by at most one refinement level.
    Invoked in Section 3.2 to restrict the nearest field to cells at level ℓ or ℓ+1 and to simplify the interaction lists.
  • domain assumption Particle masses are deposited onto AMR cells (CIC/TSC) before the Poisson solve; the FMM solves the grid-deposited density.
    The cell-centered PM-FMM approach in Section 2 inherits the mass deposition error of the original PM scheme; the solver does not act directly on point particles.
  • domain assumption Isolated boundary conditions are used; the solver is not extended to periodic domains.
    The paper explicitly restricts the implementation to isolated boundaries in Section 6.
  • domain assumption ATS is implemented as sub-cycling with two half-steps per coarse step; inactive levels' multipoles are kept stale.
    The tree life cycle in Section 3.3 is built around this specific temporal refinement scheme.
  • domain assumption The leap-frog integrator remains symplectic when the time step is fixed.
    Used in the binary orbit tests to isolate solver errors from integrator errors; the paper notes that ATS breaks symplecticity and is tested separately.

pith-pipeline@v1.3.0-alltime-deepseek · 18047 in / 14389 out tokens · 142529 ms · 2026-08-01T12:35:04.378926+00:00 · methodology

0 comments
read the original abstract

We present an extended implementation of a scalable, O(N) Poisson solver based on the fast multipole method (FMM), fully compatible with adaptive mesh refinement (AMR) and adaptive time stepping (ATS) within the RAMSES framework. Building on the unigrid algorithm in Lee & Teyssier (2026), we introduce several novel elements, including the use of multiple FMM trees, one for each AMR level, a merged FMM tree for all active levels to optimize neighbor searches, and the introduction of the concept of "nearest-field" to enforce force symmetry across refinement levels. Across a broad set of test problems, we find excellent, percent-level agreement with our reference traditional multigrid (MG) solver. We show, however, that FMM exhibits better momentum conservation properties across coarse-fine AMR interfaces. Finally, despite the overhead introduced by the spatio-temporal adaptivity, FMM shows better scalability than MG across various AMR configurations, with the largest gains obtained for the largest configurations.

Figures

Figures reproduced from arXiv: 2607.19489 by Jun-Young Lee, Romain Teyssier.

Figure 1
Figure 1. Figure 1: Visualization of the construction of a typical FMM tree over the chosen region. The green and blue cells in the AMR grid correspond to refinement levels ℓ and ℓ + 1, respectively. A grid Tℓ;ℓ ′ is defined as a tree that starts from cells with refinement level ℓ and consists of FMM cells at level ℓ ′ , with the cells colored according to the refinement level from which they originate. Gray areas mark locati… view at source ↗
Figure 2
Figure 2. Figure 2: Anatomy of the force-field on an AMR grid example, featuring the intermediate field, near field, and the nearest field, centered around the target cell with refinement level of ℓ colored in yellow. Firstly, the intermediate field is calculated from (i) gray AMR cells at level ℓ − 1 using P2P (brown dots), M2P from either (ii) Tℓ (blue dots) or the (iii) merged tree T˜ ℓ+1 (green dots) at FMM level ℓ − 1. S… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of the isolated NFW halo’s graviational potential map solved on a four-level AMR grid between MG and FMM. The rightmost panel exhibits that both the 90th- and 99-th percentiles of the mean relative difference are sub-percent. Apart from the largest source of error at the central peak, the remaining error is mostly concentrated near the coarse FMM cell boundaries. The different AMR cell sizes are… view at source ↗
Figure 4
Figure 4. Figure 4: A double uniform spheres test case shown in Paper I, now with geometrically refined AMR grid. We display the profile of gravitational potentials obtained using FMM (blue) and MG (orange) in comparison with the analytical solution (black). The values are calculated at the cell-centers closest to the axis piercing through the spheres’ centers. Dotted vertical lines guide the interiors of the two spheres, and… view at source ↗
Figure 5
Figure 5. Figure 5: A test problem consisting of two isolated, equal-mass point charges orbiting one another. The charges, with 𝐺𝑀 = 0.01, are separated by 0.15 and placed inside a box of length 1 to reduce boundary-induced biases. The system is evolved up to 𝑡 = 8, which corresponds to roughly three orbits. In each panel, the analytical solution is plotted with hollow circles, while the numerical results are indicated by fil… view at source ↗
Figure 6
Figure 6. Figure 6: Diagnostics of conservation properties for the binary orbit test case in [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Mass and energy conservation diagnostics together with relative changes in each energy budget for an isolated and hydrostatic NFW halo (𝑣200 = 150 km/s and 𝑐 = 10) with gas (mass fraction of 𝑓gas = 15%) for FMM (blue) and MG (orange) solved with AMR and ATS. Both solvers demonstrate stable solutions at similar levels for a long period (𝑡unit ≈ 0.98 Gyr.) The small fluctuations from the relative difference … view at source ↗
Figure 8
Figure 8. Figure 8: Comparison between MG (top row) and FMM (bottom row) solutions for the Boss & Bodenheimer (1979) test, which describes the protostellar collapse from a rotating cloud. The collapse begins from a spherical cloud of uniform density with a mass of 1 𝑀⊙ and a radius of about 1000 AU, with a 10% perturbation of the 𝑚 = 2 mode. Initial rotation is set so that the rotational energy is 1% of the gravitational pote… view at source ↗
Figure 9
Figure 9. Figure 9: Strong scaling results comparing FMM (blue) and MG (dark orange) run with adaptive time stepping (solid lines) and with single time stepping (dashed lines). We evolve an NFW halo using grid sizes of (ℓmin, ℓmax ) = (7, 10) (left panel), (8, 11) (middle panel), and (7, 13) (right panel) up to a fixed final simulation time. For each grid configuration, we solve the problem with increasing number of MPI ranks… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

51 extracted references · 8 canonical work pages · 1 internal anchor

  1. [1]

    Unigrid Algorithm

    A Scalable Fast Multipole Method Poisson Solver for the RAMSES code: I. Unigrid Algorithm. , keywords =. doi:10.1093/mnras/stag1241 , archivePrefix =. 2606.13793 , primaryClass =

    arXiv

  2. [2]

    , keywords =

    Smoothed particle hydrodynamics: theory and application to non-spherical stars. , keywords =. doi:10.1093/mnras/181.3.375 , adsurl =

    doi:10.1093/mnras/181.3.375

  3. [3]

    Journal of Computational Physics , year = 1984, month = mar, volume =

    Adaptive Mesh Refinement for Hyperbolic Partial Differential Equations. Journal of Computational Physics , year = 1984, month = mar, volume =. doi:10.1016/0021-9991(84)90073-1 , adsurl =

    doi:10.1016/0021-9991(84)90073-1 1984

  4. [4]

    Astronomische Nachrichten , keywords =

    Larger, faster, better: Current trends in cosmological simulations. Astronomische Nachrichten , keywords =. doi:10.1002/asna.201211701 , adsurl =

    work page doi:10.1002/asna.201211701

  5. [5]

    1977 , url=

    Multi-level adaptive solutions to boundary-value problems math comptr , author=. 1977 , url=

    1977

  6. [6]

    Fedorenko , doi =

    R.P. Fedorenko , doi =. A relaxation method for solving elliptic difference equations , url =. USSR Computational Mathematics and Mathematical Physics , number =. 1962 , bdsk-url-1 =

    1962

  7. [7]

    Bakhvalov , doi =

    N.S. Bakhvalov , doi =. On the convergence of a relaxation method with natural constraints on the elliptic operator , url =. USSR Computational Mathematics and Mathematical Physics , number =. 1966 , bdsk-url-1 =

    1966

  8. [8]

    A new high resolution code called RAMSES

    Cosmological hydrodynamics with adaptive mesh refinement. A new high resolution code called RAMSES. , keywords =. doi:10.1051/0004-6361:20011817 , archivePrefix =. astro-ph/0111367 , primaryClass =

    Pith/arXiv arXiv

  9. [9]

    Journal of Computational Physics , keywords =

    A simple multigrid scheme for solving the Poisson equation with arbitrary domain boundaries. Journal of Computational Physics , keywords =. doi:10.1016/j.jcp.2011.02.044 , archivePrefix =. 1104.1703 , primaryClass =

    Pith/arXiv arXiv 2011

  10. [10]

    Computational Astrophysics and Cosmology , keywords =

    PKDGRAV3: beyond trillion particle cosmological simulations for the next era of galaxy surveys. Computational Astrophysics and Cosmology , keywords =. doi:10.1186/s40668-017-0021-1 , archivePrefix =. 1609.08621 , primaryClass =

    Pith/arXiv arXiv

  11. [11]

    Monthly Notices of the Royal Astronomical Society , volume =

    Springel, Volker and Pakmor, Rüdiger and Zier, Oliver and Reinecke, Martin , title = ". Monthly Notices of the Royal Astronomical Society , volume =. 2021 , month =. doi:10.1093/mnras/stab1855 , url =

    doi:10.1093/mnras/stab1855 2021

  12. [12]

    Journal of Computational Physics , keywords =

    A Hierarchical <E10>O</E10>(N) Force Calculation Algorithm. Journal of Computational Physics , keywords =. doi:10.1006/jcph.2002.7026 , archivePrefix =. astro-ph/0202512 , primaryClass =

    arXiv 2002

  13. [13]

    , keywords =

    A Very Fast and Momentum-conserving Tree Code. , keywords =. doi:10.1086/312724 , archivePrefix =. astro-ph/0003209 , primaryClass =

    Pith/arXiv arXiv

  14. [14]

    Journal of Computational Physics , keywords =

    A Fast Algorithm for Particle Simulations. Journal of Computational Physics , keywords =. doi:10.1016/0021-9991(87)90140-9 , adsurl =

    doi:10.1016/0021-9991(87)90140-9

  15. [15]

    Acta Numerica , year = 1997, month = jan, volume =

    A new version of the Fast Multipole Method for the Laplace equation in three dimensions. Acta Numerica , year = 1997, month = jan, volume =. doi:10.1017/S0962492900002725 , adsurl =

    doi:10.1017/s0962492900002725 1997

  16. [16]

    Journal of Computational Physics , year = 1999, month = nov, volume =

    A Fast Adaptive Multipole Algorithm in Three Dimensions. Journal of Computational Physics , year = 1999, month = nov, volume =. doi:10.1006/jcph.1999.6355 , adsurl =

    arXiv 1999

  17. [17]

    An Improved Fast Multipole Algorithm for Potential Fields , url =

    Hrycak, Tomasz and Rokhlin, Vladimir , doi =. An Improved Fast Multipole Algorithm for Potential Fields , url =. 1998 , bdsk-url-1 =. https://doi.org/10.1137/S106482759630989X , journal =

    work page doi:10.1137/s106482759630989x 1998

  18. [18]

    , keywords =

    Hierarchical Particle Mesh: An FFT-accelerated Fast Multipole Method. , keywords =. doi:10.3847/1538-4365/ab2d24 , archivePrefix =. 1906.10734 , primaryClass =

    Pith/arXiv arXiv 1906

  19. [19]

    , keywords =

    Natively Periodic Fast Multipole Method: Approximating the Optimal Green's Function. , keywords =. doi:10.3847/1538-4357/abd9c2 , archivePrefix =. 2011.07099 , primaryClass =

    Pith/arXiv arXiv 2011

  20. [20]

    Computational Astrophysics and Cosmology , keywords =

    A fast multipole method for stellar dynamics. Computational Astrophysics and Cosmology , keywords =. doi:10.1186/s40668-014-0001-7 , archivePrefix =. 1405.2255 , primaryClass =

    Pith/arXiv arXiv

  21. [21]

    Journal of Computational Physics , keywords =

    Flexibly imposing periodicity in kernel independent FMM: A multipole-to-local operator approach. Journal of Computational Physics , keywords =. doi:10.1016/j.jcp.2017.11.012 , archivePrefix =. 1705.02043 , primaryClass =

    Pith/arXiv arXiv 2017

  22. [22]

    2020 , publisher =

    Bramas, Berenger , title =. 2020 , publisher =. doi:10.21105/joss.02444 , url =

    work page doi:10.21105/joss.02444 2020

  23. [23]

    International Journal for Numerical Methods in Engineering , year = 2011, month = jan, volume =

    PetFMM A dynamically load-balancing parallel fast multipole library. International Journal for Numerical Methods in Engineering , year = 2011, month = jan, volume =. doi:10.1002/nme.2972 , adsurl =

    work page doi:10.1002/nme.2972 2011

  24. [24]

    Journal of Magnetism and Magnetic Materials , year = 2010, month = jan, volume =

    Simple recursive implementation of fast multipole method. Journal of Magnetism and Magnetic Materials , year = 2010, month = jan, volume =. doi:10.1016/j.jmmm.2009.09.033 , adsurl =

    work page doi:10.1016/j.jmmm.2009.09.033 2010

  25. [25]

    , keywords =

    A hierarchical O(N log N) force-calculation algorithm. , keywords =. doi:10.1038/324446a0 , adsurl =

    doi:10.1038/324446a0

  26. [26]

    Computer simulation using particles

  27. [27]

    Annalen der Physik , year = 1921, month = jan, volume =

    Die Berechnung optischer und elektrostatischer Gitterpotentiale. Annalen der Physik , year = 1921, month = jan, volume =. doi:10.1002/andp.19213690304 , adsurl =

    doi:10.1002/andp.19213690304 1921

  28. [28]

    , keywords =

    The Athena++ Adaptive Mesh Refinement Framework: Multigrid Solvers for Self-gravity. , keywords =. doi:10.3847/1538-4365/acc2c0 , archivePrefix =. 2302.13903 , primaryClass =

    Pith/arXiv arXiv

  29. [29]

    Portability of Fortran's `do concurrent' on GPUs

    Portability of Fortran's `do concurrent' on GPUs. arXiv e-prints , keywords =. doi:10.48550/arXiv.2408.07843 , archivePrefix =. 2408.07843 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2408.07843

  30. [30]

    SIAM Journal on Scientific Computing , year = 2001, month = jan, volume =

    A New Fast-Multipole Accelerated Poisson Solver in Two Dimensions. SIAM Journal on Scientific Computing , year = 2001, month = jan, volume =. doi:10.1137/S1064827500369967 , adsurl =

    work page doi:10.1137/s1064827500369967 2001

  31. [31]

    Static tests

    Treecode2: the power of pluralism ─ I. Static tests. , keywords =. doi:10.1093/mnras/stag272 , archivePrefix =. 2602.06295 , primaryClass =

    arXiv

  32. [32]

    SIAM Journal on Scientific Computing , keywords =

    FFT, FMM, or Multigrid? A comparative Study of State-Of-the-Art Poisson Solvers for Uniform and Nonuniform Grids in the Unit Cube. SIAM Journal on Scientific Computing , keywords =. doi:10.1137/15M1010798 , archivePrefix =. 1408.6497 , primaryClass =

    Pith/arXiv arXiv

  33. [33]

    DICE: Disk Initial Conditions Environment

  34. [34]

    , keywords =

    SWIFT: A modern highly-parallel gravity and smoothed particle hydrodynamics solver for astrophysical and cosmological applications. , keywords =. doi:10.1093/mnras/stae922 , archivePrefix =. 2305.13380 , primaryClass =

    Pith/arXiv arXiv

  35. [35]

    , keywords =

    Adaptive Refinement Tree: A New High-Resolution N-Body Code for Cosmological Simulations. , keywords =. doi:10.1086/313015 , archivePrefix =. astro-ph/9701195 , primaryClass =

    Pith/arXiv arXiv

  36. [36]

    CASTRO: A New Compressible Astrophysical Solver. I. Hydrodynamics and Self-gravity. , archivePrefix = "arXiv", eprint =. doi:10.1088/0004-637X/715/2/1221 , adsurl =

    doi:10.1088/0004-637x/715/2/1221

  37. [37]

    , archivePrefix = "arXiv", eprint =

    ENZO: An Adaptive Mesh Refinement Code for Astrophysics. , archivePrefix = "arXiv", eprint =. doi:10.1088/0067-0049/211/2/19 , adsurl =

    doi:10.1088/0067-0049/211/2/19

  38. [38]

    , keywords =

    GAMER: A Graphic Processing Unit Accelerated Adaptive-Mesh-Refinement Code for Astrophysics. , keywords =. doi:10.1088/0067-0049/186/2/457 , archivePrefix =. 0907.3390 , primaryClass =

    Pith/arXiv arXiv

  39. [39]

    , keywords =

    GAMER-2: a GPU-accelerated adaptive mesh refinement code - accuracy, performance, and scalability. , keywords =. doi:10.1093/mnras/sty2586 , archivePrefix =. 1712.07070 , primaryClass =

    Pith/arXiv arXiv

  40. [40]

    , keywords =

    OCTO-TIGER: a new, 3D hydrodynamic code for stellar mergers that uses HPX parallelization. , keywords =. doi:10.1093/mnras/stab937 , archivePrefix =. 2101.08226 , primaryClass =

    Pith/arXiv arXiv

  41. [41]

    Martin, D. F. and Cartwright, K. L. , title =. 1996 , month = oct, url =

    1996

  42. [42]

    Berger and P

    M.J. Berger and P. Colella , doi =. Local adaptive mesh refinement for shock hydrodynamics , url =. Journal of Computational Physics , number =. 1989 , bdsk-url-1 =

    1989

  43. [43]

    , keywords =

    An Extension of the Athena++ Framework for Fully Conservative Self-gravitating Hydrodynamics. , keywords =. doi:10.3847/1538-4365/abcfbd , archivePrefix =. 2012.01340 , primaryClass =

    Pith/arXiv arXiv 2012

  44. [44]

    , keywords =

    The Jeans Condition: A New Constraint on Spatial Resolution in Simulations of Isothermal Self-gravitational Hydrodynamics. , keywords =. doi:10.1086/310975 , adsurl =

    doi:10.1086/310975

  45. [45]

    Research in Astronomy and Astrophysics , keywords =

    A hybrid Fast Multipole Method for cosmological N-body simulations. Research in Astronomy and Astrophysics , keywords =. doi:10.1088/1674-4527/21/1/3 , archivePrefix =. 2006.14952 , primaryClass =

    Pith/arXiv arXiv 2006

  46. [46]

    Parallel dual tree traversal for the symmetric fast multipole method

    Momentum-conserving self-gravity in the phantom smoothed particle hydrodynamics code. Parallel dual tree traversal for the symmetric fast multipole method. arXiv e-prints , keywords =. doi:10.48550/arXiv.2602.05804 , archivePrefix =. 2602.05804 , primaryClass =

    work page doi:10.48550/arxiv.2602.05804

  47. [47]

    , keywords =

    Fragmentation in a rotating protostar: a comparison of two three-dimensional computer codes. , keywords =. doi:10.1086/157497 , adsurl =

    doi:10.1086/157497

  48. [48]

    , keywords =

    A new multifluid method for dusty astrophysical flows: Application to turbulent protostellar collapses. , keywords =. doi:10.1051/0004-6361/202554662 , archivePrefix =. 2508.19127 , primaryClass =

    Pith/arXiv arXiv

  49. [49]

    , keywords =

    Second Core Formation and High-Speed Jets: Resistive Magnetohydrodynamic Nested Grid Simulations. , keywords =. doi:10.1086/507179 , archivePrefix =. astro-ph/0603456 , primaryClass =

    Pith/arXiv arXiv

  50. [50]

    , keywords =

    Chemical solver to compute molecule and grain abundances and non-ideal MHD resistivities in prestellar core-collapse calculations. , keywords =. doi:10.1051/0004-6361/201526780 , archivePrefix =. 1604.05613 , primaryClass =

    Pith/arXiv arXiv

  51. [51]

    , year = 1969, month = jan, volume =

    Numerical calculations of the dynamics of collapsing proto-star. , year = 1969, month = jan, volume =. doi:10.1093/mnras/145.3.271 , adsurl =

    doi:10.1093/mnras/145.3.271 1969