Pith. sign in

REVIEW 2 major objections 4 minor 187 references

Warm-started multigrid can replace distributed FFTs for fixed-mesh gravity and is the solver that makes differentiable moving-mesh PM practical.

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 · grok-4.5

2026-07-14 07:51 UTC pith:HQWAY546

load-bearing objection Solid JAX methods paper: multigrid as the real enabler for differentiable moving-mesh PM, with fixed-mesh FFT competition that is real but hardware-tied. the 2 major comments →

arxiv 2607.10983 v1 pith:HQWAY546 submitted 2026-07-13 astro-ph.IM astro-ph.COastro-ph.GA

Fast(er)PM and Moving Mesh: JAX-native Geometric Multigrid Methods

Benjamin Horowitz This is my paper
classification astro-ph.IM astro-ph.COastro-ph.GA
keywords geometric multigridparticle-meshPoisson solverJAXmoving meshFastPMdifferentiable simulationcosmological structure formation
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.

Particle–mesh cosmology needs fast, differentiable Poisson solvers. Fourier methods dominate on fixed grids but force global all-to-all communication and break once the mesh is deformed. This paper builds a JAX-native geometric multigrid solver and shows it serves two roles. On static meshes, recycling the previous potential and applying Chebyshev-smoothed V-cycles turns multigrid into a cheap defect correction that reaches field-level accuracy with one or two cycles, cutting memory and often wall-clock or total GPU time relative to distributed FFTs. On moving meshes, the same solver handles the variable-coefficient curvilinear Poisson equation that ordinary FFT diagonalization cannot treat, so force resolution can be concentrated on filaments and halos while the code stays a regular, compilable, automatically differentiable array program. The result is a practical path from fast fixed-grid PM methods to adaptive-force simulations that remain usable inside gradient-based inference.

Core claim

Geometric multigrid for particle–mesh gravity plays two complementary roles: on fixed meshes, warm-started Chebyshev multigrid is a competitive, communication-avoiding alternative to distributed FFTs, and on moving meshes it becomes the enabling solver for a differentiable variable-coefficient curvilinear Poisson problem that FFT diagonalization cannot handle.

What carries the argument

Warm-started Chebyshev multigrid as defect correction: the previous time-step potential (optionally growth-scaled) is used as the initial guess so that one or two V-cycles correct only the residual; the same hierarchy, with geometry coefficients restricted level-by-level, solves the curvilinear Laplace–Beltrami operator on the deformed mesh.

Load-bearing premise

That the measured wall-clock and GPU-time gains over distributed FFTs on the tested GPU cluster will hold broadly enough to make multigrid a practical alternative, even though the paper itself notes that those gains depend on interconnect and library performance.

What would settle it

On a production-scale fixed mesh (for example 1024^3 or 2048^3), run matched FFT and warm one- or two-cycle multigrid force evaluations on the same allocation; if multigrid is systematically slower in wall-clock and higher in total GPU-seconds, or if two-cycle field-level correlation with the FFT reference collapses, the competitive-fixed-mesh claim fails. Separately, if the moving-mesh reconstruction cannot recover small-scale transfer above the static mesh at the same cell count once the geometry is strongly deformed, the adaptive claim fails.

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

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 / 4 minor

Summary. The paper presents a JAX-native geometric multigrid Poisson framework for particle–mesh cosmology and argues that multigrid plays two complementary roles: on fixed meshes, warm-started Chebyshev multigrid is a competitive, communication-avoiding alternative to distributed FFTs; on moving meshes, it is the enabling solver for a variable-coefficient curvilinear Poisson problem that ordinary FFT diagonalization cannot handle. For static FastPM, temporal coherence is exploited via recycled/growth-scaled predictors so that one or two warm V-cycles reach field-level accuracy relative to a matched discrete FFT operator, with reported wall-clock and GPU-time gains and a lower memory floor on Perlmutter A100 nodes. The same multigrid hierarchy is then used for a potential-flow moving-mesh PM scheme (after Pen), with a conservative finite-volume Laplace–Beltrami operator, mesh-motion and gravity solves, limiters, and an implicit adjoint that supports end-to-end differentiable reconstruction. Accuracy is tested against FFT references and CAMELS/AREPO CV0, and a reconstruction stress test shows recovery of small-scale amplitude at fixed nominal resolution.

Significance. If the dual-role claim holds, the work is a useful methods contribution for GPU-native, autodiff-compatible cosmological PM solvers. The enabling half is particularly valuable: a regular-array, JAX-compilable moving-mesh gravity path with shared multigrid infrastructure and a carefully constructed adjoint fills a gap between fixed-grid FastPM-style codes and adaptive force methods that are hard to differentiate. Strengths include matched discrete-operator comparisons, residual and field-level diagnostics (transfer/stochasticity), Appendix A error-accumulation tests, strong-scaling and memory-floor measurements, and a finite-difference check of the curvilinear adjoint under strong compression. Development code links are provided. The fixed-mesh performance half is more provisional because the measured speedups are machine- and interconnect-dependent, but the structural communication and memory arguments remain of practical interest even if wall-clock ratios do not fully generalize.

major comments (2)
  1. [§3.3–3.5, Tables 2–3] §3.3–3.5 and Tables 2–3: the abstract and introduction frame warm Chebyshev multigrid as a competitive alternative with up to ~2× lower total GPU time. The measured 1.5–2.4× wall-clock and ~2× GPU-time gains are on Perlmutter A100 + Slingshot, which the text itself flags as likely suboptimal for all-to-all and machine-dependent. The structural advantages (halo vs all-to-all, real vs complex scratch, lower memory floor) are well supported, but the dual-role claim’s fixed-mesh half currently over-weights fabric-specific timings. Please rebalance the abstract/conclusions to lead with communication pattern and memory floor, present wall-clock ratios as platform-specific evidence, and (if feasible) add a brief comparison or discussion against a stronger distributed FFT baseline (e.g. jaxdecomp/cuDecomp on a fabric more favorable to collectives).
  2. [§4.7–4.8, Figs. 10–14] §4.7–4.8 and Figs. 10–14: the moving-mesh accuracy and reconstruction claims rest on a single small-box CAMELS CV0 setup (25 h^{-1} Mpc, 256^3 particles) and a deliberately minimal reconstruction experiment. That is acceptable for a methods demonstration, but the paper’s bridge claim toward “differentiable adaptive-force cosmological simulations” needs clearer scope limits: state that the tests do not yet establish cosmology- or redshift-general accuracy, quantify sensitivity to free limiter/κ parameters (Appendix C), and note that production field-level inference would require likelihood, priors, and bias validation beyond the MSE reconstruction shown.
minor comments (4)
  1. [§4.3] §4.3: note that Jacobi outperforms Chebyshev for the variable-coefficient Laplace–Beltrami problem, opposite to the static-mesh default. A short remark on why (spectrum of the preconditioned operator, anisotropy) would help readers choose smoothers.
  2. [Table 1, Figs. 4–6] Table 1 vs Figs. 4–6: cycle-count accuracy is shown at 256^3 while production scaling is at 1024^3/2048^3. A brief statement that the 1–2 warm-cycle operating point remains adequate at the larger sizes (or a small residual/transfer check) would tighten the accuracy–performance link.
  3. [§4.8.1–4.8.2] Eqs. (35)–(38): the straight-through limiter treatment and √g-conjugated adjoint are important; a one-sentence pointer in the main text to the finite-difference agreement under √g_min≈0.15 would make the reverse-mode claim easier to verify without hunting the prose.
  4. Minor presentation: a few missing spaces in compound words in the compiled text (e.g. “particle–mesh”, “all-to-all”); ensure figure captions for Figs. 2–3 and 8 fully define symbols used in the panels.

Circularity Check

1 steps flagged

Methods paper with external FFT/N-body benchmarks; warm-start is residual defect correction, not a tautological prediction, and self-citations supply infrastructure rather than the performance claims.

specific steps
  1. self citation load bearing [§2.3 Geometric multigrid; §3.4 Implementation (DiffHydro baseline)]
    "Following the notation of the diffhydro self-gravity solver, the flat-mesh PM Poisson problem is written as A_h φ_h = F_h ... The first three are the additions that give the present solver its speed over the originally published version in DiffHydro, and all are exposed as flags with the original Jacobi path preserved as a fallback."

    The flat multigrid hierarchy and Jacobi smoother are imported from the author's DiffHydro paper. This is infrastructure reuse, not a uniqueness or uniqueness-of-result claim: the present work's performance and accuracy conclusions rest on new Chebyshev fusion, warm-start timings vs FFT, and moving-mesh tests against external references, so the self-citation is not load-bearing for the dual-role claim.

full rationale

The paper is a numerical methods contribution whose central claims are (i) warm-started Chebyshev multigrid can match or beat distributed FFTs on fixed meshes under stated hardware conditions and (ii) the same multigrid hierarchy enables a differentiable moving-mesh PM solve of a variable-coefficient curvilinear Poisson equation that FFT diagonalization cannot handle. Both claims are checked against external references: the spectral FFT with the same discrete operator (Tables 1–3, Figs. 4–6, Appendix A) and the CAMELS CV0 AREPO N-body field (Figs. 10–14). Warm starting is standard defect correction of a residual (Eqs. 11–12, §3.1–3.2); the residual is not forced to zero by construction, and accuracy is measured against the FFT reference rather than redefined as success. Self-citations (DiffHydro, JaxPM, jaxdecomp) provide the prior Jacobi baseline, array framework, and distributed primitives; the production Chebyshev fusion, timings, moving-mesh limiters, and adjoint finite-difference checks are new content evaluated in this work. No uniqueness theorem, fitted parameter renamed as prediction, or self-definitional loop carries the dual-role claim. Minor self-citation of infrastructure is normal and non-load-bearing for the conclusions, so the circularity score is 1.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The work is numerical methods, not new physics. Load-bearing content is standard Poisson/PM cosmology plus classical multigrid, plus engineering choices (smoother degree, warm predictors, limiter thresholds, κ, cycle budgets) that control reported accuracy and stability. No new physical entities are introduced; free parameters are solver/mesh-control knobs.

free parameters (5)
  • Chebyshev spectral interval parameter α
    Targets upper spectrum of Jacobi-preconditioned Laplacian on [2/α,2]; paper uses α≃8 as production choice affecting smoother strength.
  • Warm V-cycle budget (1–2 production cycles)
    Accuracy/speed operating point is chosen by hand against FFT field-level fidelity rather than derived uniquely.
  • Mesh-response gain κ
    Controls how strongly the deformation tracks mass; intermediate optimum selected by experiment (Fig. 9), not fixed by theory.
  • Limiter parameters c_max, s_max, source smoothing, repulsive strength η
    Numerical safeguards against folding/anisotropy; paper states they are numerical rather than physical and need systematic calibration.
  • Weighted-Jacobi ω=2/3 and pre/post sweep counts (ν1=ν2=2 moving mesh)
    Standard but still chosen multigrid hyperparameters that affect convergence cost.
axioms (4)
  • domain assumption Discrete periodic Poisson with seven-point Laplacian (static) or conservative finite-volume Laplace–Beltrami (moving) is an adequate gravity model for the PM applications considered.
    §2.1 and §4.3; comparisons are to FFT on the same operator or to N-body truth for clustering, not full force-law validation.
  • standard math Geometric multigrid V-cycles with local smoothers converge with mesh-independent factor for the model/variable-coefficient problems used.
    Invoked via classical multigrid theory and empirical residual curves (§2.3, §3.5, §4.3).
  • domain assumption Potential-flow mesh map x=ξ+∇ψ with limiters remains a useful quasi-Lagrangian adaptive force model without destroying large-scale phases.
    Core of §4 following Pen (1995); accuracy tests support but do not prove unbiased inference use.
  • ad hoc to paper Straight-through treatment of non-smooth limiters in reverse mode yields usable gradients for reconstruction.
    §4.8.1; engineering choice for AD stability, not derived optimality.

pith-pipeline@v1.1.0-grok45 · 30963 in / 3278 out tokens · 32372 ms · 2026-07-14T07:51:39.888277+00:00 · methodology

0 comments
read the original abstract

Efficient, differentiable Poisson solvers are a key component of modern particle--mesh simulations and field-level inference pipelines. FFT-based solvers are extremely effective on fixed Cartesian meshes, but they impose global all-to-all communication and rely on symmetries that are lost in adaptive or non-Cartesian coordinates. In this work, we present a JAX-native geometric multigrid framework for particle--mesh gravity and argue that multigrid plays two complementary roles: on fixed meshes it can be a competitive, communication-avoiding alternative to FFTs, while on moving meshes it becomes the enabling solver. For static FastPM evolution, warm-started Chebyshev multigrid acts as a defect-correction method, exploiting temporal coherence between time steps to reduce the number of V-cycles required for field-level accuracy. At large mesh sizes this reduces memory pressure and yields comparable or faster wall-clock performance than distributed FFTs, with up to a factor of two reduction in total GPU time at fixed final mesh size. We then embed the same solver in a differentiable moving-mesh particle--mesh method, where adaptive coordinate deformation produces a variable-coefficient curvilinear Poisson equation that cannot be solved by ordinary FFT diagonalization. The resulting method concentrates force resolution in nonlinear structures while retaining a regular, JAX-compilable, automatically differentiable array workflow. These results suggest geometric multigrid can be a practical bridge between fast fixed-grid PM methods and differentiable adaptive-force cosmological simulations.

Figures

Figures reproduced from arXiv: 2607.10983 by Benjamin Horowitz.

Figure 1
Figure 1. Figure 1: Geometric multigrid V-cycle for the Poisson solve. On the finest mesh, the potential satisfies 𝐴 ℎ 𝜑 ℎ = 𝐹 ℎ and the residual is 𝑟 ℎ = 𝐹 ℎ − 𝐴 ℎ 𝜑 ℎ . The residual is restricted to coarser meshes, where the error equations 𝐴 2ℎ𝑒 2ℎ = 𝑟 2ℎ and 𝐴 4ℎ𝑒 4ℎ = 𝑟 4ℎ are solved. The resulting corrections are prolongated back to the fine grid and added to the solution, 𝜑 ℎ ← 𝜑 ℎ + 𝑒 ℎ . The lower panels show the sam… view at source ↗
Figure 2
Figure 2. Figure 2: Warm-start Chebyshev multigrid as defect correction in a 2563 particle–mesh solve. From left to right: the logarithmic density contrast, the initial potential guess 𝜑 (0) recycled from the previous step, the multigrid correction Δ𝜑, and the refined potential 𝜑 = 𝜑 (0) + Δ𝜑. The correction is small compared with the full potential and is concentrated around nonlinear structure, showing why a warm-started Ch… view at source ↗
Figure 3
Figure 3. Figure 3: Residual convergence for cold and warm Chebyshev-multigrid solves at early (𝑎 = 0.2) and late (𝑎 = 0.7) times. Curves show the relative residual ∥𝐹 − 𝐴𝜑∥ /∥𝐹∥ after each V-cycle for a cold start, simple recycling, growth-scaled recycling, and polynomial extrapolation. Warm-start predictors reduce the initial residual by orders of magnitude and typically remove one to two V-cycles at fixed tolerance. arithm… view at source ↗
Figure 5
Figure 5. Figure 5: Per-step wall-clock time as a function of mesh size at fixed node count (32 GPUs), isolating the effect of the per-GPU problem size on the FFT/multigrid crossover. At small meshes the FFT is far cheaper (the warm solve is dominated by fixed halo-latency overhead), but the ratio climbs monotonically with mesh size crossing unity just above 5123 . The 20483 point has no FFT point because the spectral transfo… view at source ↗
Figure 4
Figure 4. Figure 4: Strong-scaling comparison of FFT and warm-start Chebyshev￾multigrid PM steps, measured in a single allocation on Perlmutter A100 (40 GB) nodes. The panels show wall-clock time per step for 5123 and 10243 meshes as the number of GPUs is varied. The FFT path is dominated by dis￾tributed transforms and global transposes, whereas warm Chebyshev multi￾grid uses one (wc1) or two (wc2) V-cycles plus local finite-… view at source ↗
Figure 6
Figure 6. Figure 6: Field-level accuracy of the warm-started Chebyshev-multigrid PM evolution relative to an FFT reference for a 10243 run. The top panel shows the power-spectrum transfer function 𝑃/𝑃FFT, and the bottom panel shows the stochasticity 1−𝑟 (𝑘). Solid curves use warm-started solves, while dashed curves show cold-start solves with the same number of V-cycles. Additional cycles systematically reduce both transfer-f… view at source ↗
Figure 7
Figure 7. Figure 7: The standard static particle mesh compared with the moving mesh force calculation, starting from the same particle distribution (left) and overlaid with the particles at every stage. In the static particle–mesh branch (top) the particles are painted onto a fixed grid, the density 𝜌 is Fourier transformed, ∇ 2𝜙 = 𝛿 is solved with an FFT, and the force −∇𝜙 is read back. The force resolution is set once by th… view at source ↗
Figure 8
Figure 8. Figure 8: Example moving-mesh geometry in a cosmological field. Left: the deformed mesh in a slice through the volume, with cells contracted along filaments and around collapsed structures and expanded in voids. Middle left: the scalar deformation potential 𝜓 whose gradient defines the coordinate map 𝒙(𝝃 ) = 𝝃+∇𝜓. Middle right: the corresponding logarithmic density field in real space. Right: the logarithmic density… view at source ↗
Figure 9
Figure 9. Figure 9: Effect of the mesh-response parameter 𝜅 on the moving-mesh deformation at 𝑧 = 0. Each panel shows a two-dimensional slice through the deformed computational grid for a simulation run from the same initial condition but varying 𝜅, with larger 𝜅 driving stronger contraction toward filaments and halo-like peaks. Moderate values increase force resolution in overdense regions while maintaining a smooth, inverti… view at source ↗
Figure 11
Figure 11. Figure 11: Cumulative mass fraction above density threshold for static PM meshes and the moving-mesh calculation. The moving mesh produces a higher-density tail than a static mesh of comparable nominal resolution, bring￾ing the one-point distribution closer to the CV0 truth in collapsed regions. This is the real-space counterpart of the improved small-scale correlation shown in [PITH_FULL_IMAGE:figures/full_fig_p01… view at source ↗
Figure 10
Figure 10. Figure 10: Clustering accuracy of static and moving-mesh PM calculations against the high-resolution reference at constant 2563 particle number. The upper panel shows the cross-correlation coefficient 𝑟 (𝑘), and the lower panel shows the transfer function 𝑇 (𝑘) = √︁ 𝑃/𝑃ref. Increasing the static mesh resolution improves agreement, while the moving-mesh calculation recovers substantially more small-scale correlation … view at source ↗
Figure 12
Figure 12. Figure 12: Cost–accuracy trade-off for static and moving-mesh PM calcula￾tions for fixed particle number 𝑁 = 2563 . Accuracy is summarized by the cross-correlation coefficient averaged over 𝑘 = 3–8 ℎ Mpc−1 relative to the CV0 truth, while the vertical axes show peak GPU memory and wall time per step. The moving mesh improves small-scale accuracy at fixed cell count by concentrating resolution into dense regions, but… view at source ↗
Figure 13
Figure 13. Figure 13: Differentiable moving-mesh reconstruction in a 25 ℎ −1Mpc slice, shown as a projected five-cell slab. Left: the optimized computational mesh, which contracts around the filaments and halo-like peaks that dominate the nonlinear density field; in the densest regions the cells are compressed to roughly a sixth of their reference volume (√ 𝑔min ≈ 0.15). Middle: the reconstructed logarithmic density obtained b… view at source ↗
Figure 14
Figure 14. Figure 14: Fourier-space diagnostics for static-mesh (i.e. JaxPM) and moving-mesh differentiable reconstructions of the same target field. Upper: the cross-correlation coefficient 𝑟 (𝑘) between the reconstructed and target 𝑧 = 0 density fields. Lower: the transfer function 𝑇 (𝑘) = √︁ 𝑃rec/𝑃true. The two reconstructions agree on large and intermediate scales and separate to￾ward the grid Nyquist frequency, where the … 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

187 extracted references · 42 canonical work pages · 26 internal anchors

  1. [1]

    Nonlinear Reconstruction of the Velocity Field

    Nonlinear Reconstruction of the Velocity Field. , keywords =. doi:10.3847/1538-4357/ab5580 , archivePrefix =. 1908.08217 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-4357/ab5580 1908

  2. [2]

    Hydrodynamic methods and sub-resolution models for cosmological simulations

    Hydrodynamic methods and sub-resolution models for cosmological simulations. arXiv e-prints , keywords =. doi:10.48550/arXiv.2502.06954 , archivePrefix =. 2502.06954 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2502.06954

  3. [3]

    , keywords =

    The Santa Barbara Cluster Comparison Project: A Comparison of Cosmological Hydrodynamics Solutions. , keywords =. doi:10.1086/307908 , archivePrefix =. astro-ph/9906160 , primaryClass =

    Pith/arXiv arXiv doi:10.1086/307908

  4. [4]

    , keywords =

    The Sunyaev-Zel'dovich angular power spectrum as a probe of cosmological parameters. , keywords =. doi:10.1046/j.1365-8711.2002.05889.x , archivePrefix =. astro-ph/0205468 , primaryClass =

    doi:10.1046/j.1365-8711.2002.05889.x 2002

  5. [5]

    Differentiable Stochastic Halo Occupation Distribution

    Differentiable stochastic halo occupation distribution. , keywords =. doi:10.1093/mnras/stae350 , archivePrefix =. 2211.03852 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stae350

  6. [6]

    Halo nonlinear reconstruction

    Halo Nonlinear Reconstruction. , keywords =. doi:10.3847/1538-4357/aa89e7 , archivePrefix =. 1703.08301 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-4357/aa89e7

  7. [7]

    Nonlinear reconstruction of redshift space distortions

    Nonlinear reconstruction of redshift space distortions. , keywords =. doi:10.1103/PhysRevD.97.043502 , archivePrefix =. 1711.03218 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1103/physrevd.97.043502

  8. [8]

    Journal of Computational Physics , volume=

    Shaping the force law in two-dimensional particle-mesh models , author=. Journal of Computational Physics , volume=. 1974 , publisher=

    1974

  9. [9]

    Computer Simulation Using Particles

  10. [10]

    , keywords =

    Numerical techniques for large cosmological N-body simulations. , keywords =. doi:10.1086/191003 , adsurl =

    doi:10.1086/191003

  11. [11]

    arXiv e-prints , keywords =

    Particle-Mesh code for cosmological simulations. arXiv e-prints , keywords =. doi:10.48550/arXiv.astro-ph/9712217 , archivePrefix =. astro-ph/9712217 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.astro-ph/9712217

  12. [12]

    , keywords =

    The Athena++ Adaptive Mesh Refinement Framework: Design and Magnetohydrodynamic Solvers. , keywords =. doi:10.3847/1538-4365/ab929b , archivePrefix =. 2005.06651 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4365/ab929b 2005

  13. [13]

    Towards optimal parallel PM N-body codes: PMFAST

    Towards optimal parallel PM N-body codes: PMFAST. , keywords =. doi:10.1016/j.newast.2005.02.001 , archivePrefix =. astro-ph/0402443 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1016/j.newast.2005.02.001 2005

  14. [14]

    Mathematics of computation , volume=

    Multi-level adaptive solutions to boundary-value problems , author=. Mathematics of computation , volume=

  15. [15]

    USSR Computational Mathematics and Mathematical Physics , volume=

    A relaxation method for solving elliptic difference equations , author=. USSR Computational Mathematics and Mathematical Physics , volume=. 1962 , publisher=

    1962

  16. [16]

    , 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 doi:10.1086/313015

  17. [17]

    , keywords =

    A Linear Moving Adaptive Particle-Mesh N-Body Algorithm. , keywords =. doi:10.1086/192219 , adsurl =

    work page doi:10.1086/192219

  18. [18]

    Proceedings of the 1998 IEEE International Conference on Acoustics , year = 1998, volume =

    FFTW: an adaptive software architecture for the FFT. Proceedings of the 1998 IEEE International Conference on Acoustics , year = 1998, volume =. doi:10.1109/ICASSP.1998.681704 , adsurl =

    doi:10.1109/icassp.1998.681704 1998

  19. [19]

    IEEE Proceedings , keywords =

    The Design and Implementation of FFTW3. IEEE Proceedings , keywords =. doi:10.1109/JPROC.2004.840301 , adsurl =

    doi:10.1109/jproc.2004.840301 2004

  20. [20]

    , year = 2026, month = jun, doi =

    jFoF: GPU Friends-of-Friends Halo Finding with Gradient Propagation. , year = 2026, month = jun, doi =

    2026

  21. [21]

    , keywords =

    Differentiable cosmological hydrodynamics for field level inference and high dimensional parameter constraints. , keywords =. doi:10.1093/mnras/staf1785 , archivePrefix =. 2502.02294 , primaryClass =

    work page doi:10.1093/mnras/staf1785

  22. [22]

    , keywords =

    Differentiable fuzzy cosmic web for field-level inference. , keywords =. doi:10.1051/0004-6361/202555817 , archivePrefix =. 2506.03969 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/202555817

  23. [23]

    arXiv e-prints , keywords =

    Field-Level Inference with Microcanonical Langevin Monte Carlo. arXiv e-prints , keywords =. doi:10.48550/arXiv.2307.09504 , archivePrefix =. 2307.09504 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.2307.09504

  24. [24]

    , keywords =

    Benchmarking field-level cosmological inference from galaxy redshift surveys. , keywords =. doi:10.1088/1475-7516/2025/12/039 , archivePrefix =. 2504.20130 , primaryClass =

    work page doi:10.1088/1475-7516/2025/12/039 2025

  25. [25]

    Mujic{\Lambda}: Reconstructing Initial Conditions from Incomplete Redshift Surveys with Projected Optimization

    Mujic : Reconstructing Initial Conditions from Incomplete Redshift Surveys with Projected Optimization. arXiv e-prints , keywords =. doi:10.48550/arXiv.2604.26823 , archivePrefix =. 2604.26823 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2604.26823

  26. [26]

    , keywords =

    The Manticore Project I: a digital twin of our cosmic neighbourhood from Bayesian field-level analysis. , keywords =. doi:10.1093/mnras/staf767 , archivePrefix =. 2505.10682 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/staf767

  27. [27]

    Journal of Parallel and Distributed Computing , volume=

    FFT, FMM, and multigrid on the road to exascale: Performance challenges and opportunities , author=. Journal of Parallel and Distributed Computing , volume=. 2020 , publisher=

    2020

  28. [28]

    arXiv e-prints , keywords =

    diffhydro: Inverse Multiphysics Modeling and Embedded Machine Learning in Astrophysical Flows. arXiv e-prints , keywords =. doi:10.48550/arXiv.2512.13403 , archivePrefix =. 2512.13403 , primaryClass =

    work page doi:10.48550/arxiv.2512.13403

  29. [29]

    Journal of Open Source Software , volume=

    jaxDecomp: JAX Library for 3D Domain Decomposition and Parallel FFTs , author=. Journal of Open Source Software , volume=

  30. [30]

    Proceedings of the Platform for Advanced Scientific Computing Conference , pages=

    Distributed-memory simulations of turbulent flows on modern GPU systems using an adaptive pencil decomposition library , author=. Proceedings of the Platform for Advanced Scientific Computing Conference , pages=

  31. [31]

    , keywords =

    Differentiable Cosmological Simulation with the Adjoint Method. , keywords =. doi:10.3847/1538-4365/ad0ce7 , archivePrefix =. 2211.09815 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4365/ad0ce7

  32. [32]

    Journal of Imaging , volume=

    Python non-uniform fast Fourier transform (PyNUFFT): An accelerated non-Cartesian MRI package on a heterogeneous platform (CPU/GPU) , author=. Journal of Imaging , volume=. 2018 , publisher=

    2018

  33. [33]

    Poisson, Sim. M. 1826 , publisher=

  34. [34]

    Increasing Fisher Information by Moving-Mesh Reconstruction

    Increasing Fisher information by Potential Isobaric Reconstruction. , keywords =. doi:10.1093/mnras/stx774 , archivePrefix =. 1611.10013 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1093/mnras/stx774

  35. [35]

    Isobaric Reconstruction of the Baryonic Acoustic Oscillation

    Isobaric Reconstruction of the Baryonic Acoustic Oscillation. , keywords =. doi:10.3847/2041-8213/aa738c , archivePrefix =. 1703.09742 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/2041-8213/aa738c 2041

  36. [36]

    A High-resolution Adaptive Moving Mesh Hydrodynamic Algorithm

    A High-Resolution Adaptive Moving Mesh Hydrodynamic Algorithm. , keywords =. doi:10.1086/313074 , archivePrefix =. astro-ph/9704258 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1086/313074

  37. [37]

    Iterative Poisson Solvers for Self-gravity with the GPU Code Astaroth

    Iterative Poisson Solvers for Self-gravity with the GPU Code Astaroth. arXiv e-prints , keywords =. doi:10.48550/arXiv.2605.03563 , archivePrefix =. 2605.03563 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.03563

  38. [38]

    A simple and efficient solver for self-gravity in the DISPATCH astrophysical simulation framework

    A simple and efficient solver for self-gravity in the DISPATCH astrophysical simulation framework. Journal of Physics Conference Series , year = 2018, series =. doi:10.1088/1742-6596/1031/1/012021 , archivePrefix =. 1806.10098 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/1742-6596/1031/1/012021 2018

  39. [39]

    A Performance-Portable, Massively Parallel Distributed Nonuniform FFT

    A Performance-Portable, Massively Parallel Distributed Nonuniform FFT. arXiv e-prints , keywords =. doi:10.48550/arXiv.2605.10678 , archivePrefix =. 2605.10678 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.10678

  40. [40]

    , keywords =

    Towards optimal extraction of cosmological information from nonlinear data. , keywords =. doi:10.1088/1475-7516/2017/12/009 , archivePrefix =. 1706.06645 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/1475-7516/2017/12/009 2017

  41. [41]

    Astronomy and Computing , keywords =

    FlowPM: Distributed TensorFlow implementation of the FastPM cosmological N-body solver. Astronomy and Computing , keywords =. doi:10.1016/j.ascom.2021.100505 , archivePrefix =. 2010.11847 , primaryClass =

    doi:10.1016/j.ascom.2021.100505 2021

  42. [42]

    , keywords =

    Solving large scale structure in ten easy steps with COLA. , keywords =. doi:10.1088/1475-7516/2013/06/036 , archivePrefix =. 1301.0322 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/1475-7516/2013/06/036 2013

  43. [43]

    Second Data Release of the COSMOS Lyman-alpha Mapping and Tomographic Observation: The First 3D Maps of the Detailed Cosmic Web at 2.05<z<2.55

    Second Data Release of the COSMOS Ly Mapping and Tomography Observations: The First 3D Maps of the Detailed Cosmic Web at 2.05 < z < 2.55. , keywords =. doi:10.3847/1538-4365/ac982d , archivePrefix =. 2109.09660 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/1538-4365/ac982d

  44. [44]

    , keywords =

    Planting a Lyman alpha forest on ABACUSSUMMIT. , keywords =. doi:10.1093/mnras/stad1920 , archivePrefix =. 2305.08899 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stad1920

  45. [45]

    Radiative Transfer in a Clumpy Universe. IV. New Synthesis Models of the Cosmic UV/X-Ray Background. , archivePrefix = "arXiv", eprint =. doi:10.1088/0004-637X/746/2/125 , adsurl =

    doi:10.1088/0004-637x/746/2/125

  46. [46]

    arXiv e-prints , keywords =

    Strong bound on canonical ultra-light axion dark matter from the Lyman-alpha forest. arXiv e-prints , keywords =

  47. [47]

    , keywords =

    Revealing Reionization with the Thermal History of the Intergalactic Medium: New Constraints from the Ly Flux Power Spectrum. , keywords =. doi:10.3847/1538-4357/aafee4 , archivePrefix =. 1809.06980 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4357/aafee4

  48. [48]

    , keywords =

    Hints, neutrino bounds, and WDM constraints from SDSS DR14 Lyman- and Planck full-survey data. , keywords =. doi:10.1088/1475-7516/2020/04/038 , archivePrefix =

    doi:10.1088/1475-7516/2020/04/038 2020

  49. [49]

    , keywords =

    The Cosmic Baryon Budget. , keywords =. doi:10.1086/306025 , archivePrefix =. astro-ph/9712020 , primaryClass =

    Pith/arXiv arXiv doi:10.1086/306025

  50. [50]

    , keywords =

    Simulating intergalactic gas for DESI-like small scale Lyman forest observations. , keywords =. doi:10.1088/1475-7516/2021/04/059 , archivePrefix =. 2012.04008 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/1475-7516/2021/04/059 2021

  51. [51]

    , keywords =

    New Constraints on IGM Thermal Evolution from the Ly Forest Power Spectrum. , keywords =. doi:10.3847/1538-4357/aafad1 , archivePrefix =. 1808.04367 , primaryClass =

    Pith/arXiv arXiv doi:10.3847/1538-4357/aafad1

  52. [52]

    arXiv e-prints , keywords =

    Standard Big-Bang Nucleosynthesis after Planck. arXiv e-prints , keywords =

  53. [54]

    Annual Review of Astronomy and Astrophysics , volume =

    McQuinn, Matthew , title =. Annual Review of Astronomy and Astrophysics , volume =. 2016 , doi =

    2016

  54. [55]

    Astronomy and Computing , keywords =

    TensorFlow Hydrodynamics Analysis for Ly- Simulations. Astronomy and Computing , keywords =. doi:10.1016/j.ascom.2024.100858 , archivePrefix =. 2407.16009 , primaryClass =

    work page doi:10.1016/j.ascom.2024.100858 2024

  55. [56]

    New synthesis models of the cosmic UV/X-ray background

    Radiative transfer in a clumpy universe: IV. New synthesis models of the cosmic UV/X-ray background. arXiv e-prints , keywords =. doi:10.48550/arXiv.1105.2039 , archivePrefix =. 1105.2039 , primaryClass =

    Pith/arXiv arXiv doi:10.48550/arxiv.1105.2039 2039

  56. [57]

    Planck 2018 results. VI. Cosmological parameters. , keywords =. doi:10.1051/0004-6361/201833910 , archivePrefix =. 1807.06209 , primaryClass =

    Pith/arXiv arXiv doi:10.1051/0004-6361/201833910 2018

  57. [58]

    , year = 1965, month = nov, volume =

    On the Density of Neutral Hydrogen in Intergalactic Space. , year = 1965, month = nov, volume =. doi:10.1086/148444 , adsurl =

    doi:10.1086/148444 1965

  58. [59]

    , keywords =

    Formation and Evolution of X-Ray Clusters: A Hydrodynamic Simulation of the Intracluster Medium. , keywords =. doi:10.1086/169350 , adsurl =

    work page doi:10.1086/169350

  59. [60]

    , keywords =

    A Hydrodynamic Approach to Cosmology: Methodology. , keywords =. doi:10.1086/191630 , adsurl =

    doi:10.1086/191630

  60. [61]

    , keywords =

    Cosmological Simulations with TreeSPH. , keywords =. doi:10.1086/192305 , archivePrefix =. astro-ph/9509107 , primaryClass =

    Pith/arXiv arXiv doi:10.1086/192305

  61. [62]

    , keywords =

    The one-dimensional power spectrum from the SDSS DR14 Ly forests. , keywords =. doi:10.1088/1475-7516/2019/07/017 , archivePrefix =. 1812.03554 , primaryClass =

    Pith/arXiv arXiv doi:10.1088/1475-7516/2019/07/017 2019

  62. [63]

    doi:10.1093/mnras/stt449 , Eprint =

    , Keywords =. doi:10.1093/mnras/stt449 , Eprint =

    doi:10.1093/mnras/stt449

  63. [64]

    , keywords =

    The cosmological simulation code GADGET-2. , keywords =. doi:10.1111/j.1365-2966.2005.09655.x , archivePrefix =. astro-ph/0505010 , primaryClass =

    doi:10.1111/j.1365-2966.2005.09655.x 2005

  64. [65]

    2019 4th International Conference on Pattern Recognition and Image Analysis (IPRIA) , pages=

    Deep Generative Models: Deterministic Prediction with an Application in Inverse Rendering , author=. 2019 4th International Conference on Pattern Recognition and Image Analysis (IPRIA) , pages=. 2019 , organization=

    2019

  65. [66]

    2016 international joint conference on neural networks (IJCNN) , pages=

    Training deep neural networks on imbalanced data sets , author=. 2016 international joint conference on neural networks (IJCNN) , pages=. 2016 , organization=

    2016

  66. [67]

    Sixth International Conference on Machine Learning and Applications (ICMLA 2007) , pages=

    Learning with limited minority class data , author=. Sixth International Conference on Machine Learning and Applications (ICMLA 2007) , pages=. 2007 , organization=

    2007

  67. [68]

    arXiv e-prints , keywords =

    Data Augmentation Using GANs. arXiv e-prints , keywords =

  68. [69]

    arXiv e-prints , keywords =

    Adversarial Training of Variational Auto-encoders for High Fidelity Image Generation. arXiv e-prints , keywords =

  69. [70]

    arXiv e-prints , keywords =

    Deep Generative Models for Galaxy Image Simulations. arXiv e-prints , keywords =

  70. [71]

    Deep Neural Network Classifier for Variable Stars with Novelty Detection Capability

    Deep Neural Network Classifier for Variable Stars with Novelty Detection Capability. , keywords =. doi:10.3847/2041-8213/ab212c , archivePrefix =. 1905.05767 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3847/2041-8213/ab212c 2041

  71. [72]

    arXiv e-prints , keywords =

    HInet: Generating neutral hydrogen from dark matter with neural networks. arXiv e-prints , keywords =

  72. [73]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

    A variational u-net for conditional appearance and shape generation , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

  73. [74]

    Pattern Recognition , volume=

    Recent advances in convolutional neural networks , author=. Pattern Recognition , volume=. 2018 , publisher=

    2018

  74. [75]

    Advances in neural information processing systems , volume=

    Learning structured output representation using deep conditional generative models , author=. Advances in neural information processing systems , volume=

  75. [76]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Image-to-image translation with conditional adversarial networks , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  76. [77]

    European conference on computer vision , pages=

    Perceptual losses for real-time style transfer and super-resolution , author=. European conference on computer vision , pages=. 2016 , organization=

    2016

  77. [78]

    , archivePrefix = "arXiv", eprint =

    Painting with baryons: augmenting N-body simulations with gas using deep generative models. , archivePrefix = "arXiv", eprint =. doi:10.1093/mnrasl/slz075 , adsurl =

    doi:10.1093/mnrasl/slz075

  78. [79]

    , keywords =

    On the evolution of the H I column density distribution in cosmological simulations. , keywords =. doi:10.1093/mnras/stt066 , archivePrefix =. 1210.7808 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/stt066

  79. [80]

    , keywords =

    Equation of state of the photoionized intergalactic medium. , keywords =. doi:10.1093/mnras/292.1.27 , archivePrefix =. astro-ph/9612232 , primaryClass =

    Pith/arXiv arXiv doi:10.1093/mnras/292.1.27

  80. [81]

    Effect of the Metallicity on the X-ray Emission from the Warm-Hot Intergalactic Medium

    Effect of Metallicity on X-ray Emission from the Warm-hot Intergalactic Medium. , keywords =. doi:10.1088/0004-637X/721/1/46 , archivePrefix =. 1007.3274 , primaryClass =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1088/0004-637x/721/1/46

Showing first 80 references.