Pith. sign in

REVIEW 4 major objections 4 minor 19 references

High mass and halo resolution from fast low resolution simulations

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Calibrated low-resolution FastPM simulations reproduce full N-body halo and matter statistics at roughly one hundredth of the time steps.

desk verdict Useful methods paper with a genuinely new halo finder, but the accuracy claim rests on a single calibration simulation and needs an out-of-sample test before it can be trusted for survey mocks. read the letter →

arxiv 1908.05276 v1 pith:233VKAQE submitted 2019-08-14 astro-ph.CO

Biwei Dai , Yu Feng , Uros Seljak , Sukhdeep Singh This is my paper · ORCID
classification astro-ph.CO PACS 98.80.-k95.35.+d98.65.-r
keywords fastsimulationsquasi-N-bodyhalofinderfriends-of-friendspotentialgradientdescentbiasmatterpowerspectrumweaklensingconvergence
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that fast, low-resolution particle-mesh simulations can be calibrated to produce halo and matter statistics that rival high-resolution full N-body simulations of the same mass resolution, at about two orders of magnitude fewer time steps. For halos, the key is a modified friends-of-friends finder, relaxed-FoF, whose linking length grows for smaller and higher-redshift halos and which rejects spurious, unbound clumps by their velocity dispersion. For the matter field, the paper embeds a potential gradient descent (PGD) correction into every time step, adding a sub-grid displacement that restores power on nonlinear scales. The result is a FastPM run that matches high-resolution reference halo bias, mass function, real- and redshift-space clustering, halo-matter cross-power, and weak lensing convergence power spectra closely enough for survey-scale mock catalogs.

What carries the argument

Two calibrated corrections carry the argument. The first is relaxed-FoF, a halo finder that makes the linking length a function of halo particle number and redshift, $l(N_{p,i},z)$, and filters out fake halos using a velocity-dispersion threshold $r_0(z)$ relative to the expected mass–velocity-dispersion scaling; this reassembles fragmented low-resolution halos and removes unbound false detections. The second is potential gradient descent (PGD) embedded into every FastPM time step, which adds a particle displacement along the gradient of a filtered gravitational potential, with parameters $\alpha$, $k_l$, and $k_s$ chosen as simple functions of scale factor; this restores small-scale matter power and makes static snapshots and lightcone outputs consistently corrected.

What would settle it

Run FastPM with the published relaxed-FoF and PGD parameters on an independent simulation with a different cosmology or box size, and compare the halo mass function, halo bias, and matter power spectrum against a matching full N-body simulation; deviations larger than the few-percent agreement shown for the calibration run would show that the calibration does not transfer.

Watch

Extended reading notes

Core claim

On the halo side, the paper claims that the dominant failure of low-resolution FastPM is not force resolution but halo identification: small halos fragment under the standard linking length $l=0.2$, producing missing and overly clustered objects, while dense regions generate fake halos. Relaxed-FoF uses linking lengths $l(N_{p,i},z)$ that increase for smaller halos and at higher redshift, defined across six particle-number bins with linear interpolation (Eqs. 2.2–2.5), and removes candidates whose velocity dispersion exceeds $r_0(z)$ times the expected dispersion from the scaling relation (Eqs. 2.1, 2.6). With initial conditions generated on a mesh twice as fine as the particle grid, the missed-halo fraction at $10^{11}\,M_\odot$ drops below 10% at $z=2$, and halo bias, mass function, auto and redshift-space power spectra, cross-correlation with the reference catalog, and halo-matter cross-power all improve. On the matter side, embedding PGD after each FastPM step, with redshift-dependent parameters $\log(\alpha/\alpha_0)=Aa^2-Ba$ and $k_l=k_{l,0}a^\gamma$, restores the nonlinear matter power spectrum to roughly 1% agreement at fixed redshifts. A lightcone built by interpolating particle positions between steps then yields weak lensing convergence tomographic power spectra that agree substantially better with the theoretical nonlinear model.

Load-bearing premise

The linking lengths, fake-halo threshold, and PGD parameters are all fitted to one reference simulation, and the paper never tests whether those fixed functions remain accurate for different cosmologies, box sizes, or initial density fields before recommending them for survey mocks.

Editorial extensions

If this is right

  • A low-resolution FastPM run with relaxed-FoF produces halo catalogs whose missed-halo fraction, bias, mass function, and real- and redshift-space power spectra are comparable to those of a full N-body run at the same particle mass.
  • For halos above $3\times10^{11}\,M_\odot$, the calibrated simulation keeps deviations in redshift-space halo power and halo-matter cross-power within about 6% out to $k=2\,h\,\mathrm{Mpc}^{-1}$ and $z\le2$.
  • Embedding PGD at every time step brings static-snapshot matter power to roughly 1% agreement with the reference and makes lightcone outputs consistently corrected, so weak lensing convergence tomographic power spectra approach the theoretical prediction.
  • Because the corrected FastPM needs only about 20 time steps for a lightcone, survey-scale mock generation over cubic-gigaparsec volumes becomes feasible at the $10^{11}\,M_\odot$ halo resolution required by future surveys.
  • The relaxed-FoF catalog also has better large-scale auto power in real and redshift space, and better halo-matter cross-power, than a full N-body simulation at the same mass resolution.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The calibration functions—linking length bins, fake-halo threshold, and PGD parameters—are fitted to a single reference simulation, and the paper does not demonstrate that they transfer to another cosmology, box size, or initial density field; testing transferability on an independent simulation is the natural next step.
  • Relaxed-FoF changes how mass is assigned to individual halos, so statistics that depend on internal structure, such as concentration or assembly bias, may shift in ways not captured by the power-spectrum tests presented.
  • Because PGD alters particle positions in every step, it also changes particle velocities; the method's small-scale velocity statistics beyond the tested redshift-space halo power remain a useful further check.
  • A direct extension would be to fit the same functional forms to several independent reference simulations and ask whether the fitted parameters drift; stable parameters would strengthen confidence in the method's transferability.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. This manuscript describes two modifications to the FastPM quasi-N-body code aimed at improving low-resolution mock catalogs. First, a 'relaxed-FoF' halo finder uses halo-mass- and redshift-dependent linking lengths plus a velocity-dispersion-based fake-halo rejection, with parameters calibrated so that the FastPM halo mass function and bias match TNG300-2-Dark. Second, a potential gradient descent (PGD) correction is inserted after each FastPM time step, with redshift-dependent parameters fitted to the TNG300 matter power spectrum. The authors compare halo bias, mass function, auto/cross power spectra, redshift-space power spectra, and catalog matching against TNG300-1/2/3-Dark, and show that a PGD-enabled FastPM lightcone improves the lensing convergence tomographic power spectrum relative to halofit. The concluding claim is that calibrated FastPM is comparable to a high-resolution N-body simulation at the same mass resolution with two orders of magnitude fewer time steps.

Significance. If the calibration functions transfer beyond the single 205 h^-1 Mpc box and the TNG300-2-Dark initial density field, the paper provides a practical recipe for producing large-volume mocks for DESI/LSST at substantially reduced cost. The paper's strengths are its direct comparisons with several TNG resolutions sharing the same initial field, its frank discussion of the resolution-limited behavior of small halos, and the embedding of PGD into the time integration so that lightcone outputs are consistently corrected. However, because the central statistics that define the calibration are matched by construction and no out-of-sample test is reported, the headline claim is currently conditional on a transferability assumption that the paper does not establish.

major comments (4)
  1. [Sec. 2.1 (Eqs. 2.2–2.6) and Figs. 3–4] The relaxed-FoF parameters are introduced as 'simple functions we choose to produce correct halo mass function and halo bias.' Consequently, the agreement shown for the mass function (Fig. 4) and halo bias (Fig. 3) is a statement about the calibration target, not an independent test. The paper should either provide an out-of-sample validation (e.g., a different realization, box size, or cosmology, or at least a split-sample procedure within the same box) or explicitly reword the claims so that these halo comparisons are presented as calibration checks. The abstract's 'comparable to high resolution N-body' claim needs this evidence.
  2. [Sec. 3.1 (Eqs. 3.5–3.6) and Fig. 13] The PGD parameters are fitted to match the matter power spectrum of the reference simulation, so the improved P(k) in Fig. 11 is by construction at the fitted redshifts. The out-of-sample lightcone test in Fig. 13 compares the measured convergence power spectrum to the halofit analytic prediction rather than to a full N-body lightcone, and no error bars or significance levels are given; this does not by itself validate the matter field accuracy for lensing. Please add a comparison to a full N-body (or at least a validated emulator) lightcone, or soften the lensing claim accordingly.
  3. [Secs. 2.1 and 3.1] The fitted values of the free parameters l1, l6, A1, A2, B1, B2, alpha0, A, B, kl,0, gamma, and ks,0 are never reported. Without these values, the 'calibration' cannot be reproduced or tested, and the reader cannot assess whether the functional forms are stable or overfit. A table of the best-fit parameters and, ideally, a description of the fitting procedure and the covariance of the fits should be included.
  4. [Sec. 3.1] The text says the mass resolution of the FastPM run used for the PGD fit is 125 times lower than TNG300-2-Dark, which would correspond to 250^3 particles, but the halo section uses 625^3 particles (8 times lower). The particle number and box for the PGD calibration runs are not stated; please clarify this inconsistency and specify which simulations produced Fig. 11.
minor comments (4)
  1. [Algorithm 1] In line 18 of the algorithm, the loop variable is declared as i but the body references halo[j]; this appears to be a typo for halo[i].
  2. [Eqs. (3.3)–(3.4)] The filter scales k_l and k_s are introduced via k^2/k_l^2 and k^4/k_s^4, but their units and the fitted ranges are not stated; please define them and report the fitted values.
  3. [Fig. 11 caption and Sec. 3.1] The labels 'FastPM + PGD fit TNGDark' and 'FastPM + PGD fit TNG' are not explained in the body; specify which reference simulation was used for each fit and whether the same functional form is used for both.
  4. [Page 15] The phrase 'The shadow region shows the 1% deviation' should read 'shaded region'; the same figure captions should consistently use 'shaded'.

Circularity Check

2 steps flagged · score 6.0 of 10

Relaxed-FOF and PGD parameters are fit to TNG300-2-Dark's halo mass function, halo bias, and matter power spectrum, and those same statistics are then presented as the validation; independent halo clustering statistics provide partial but not complete out-of-fit support.

  1. fitted input called prediction [Sec. 2.1 (Relaxed-FoF), after Eq. (2.6); Sec. 2.2, Figs. 3-4]
    "The function l(Np,i,z) and r0(z) are simple functions we choose to produce correct halo mass function and halo bias."

    The free parameters l1, l6, A1, A2, B1, B2 in Eqs. (2.2)-(2.6) are explicitly chosen so that relaxed-FoF reproduces TNG300-2-Dark's halo mass function and halo bias. Figures 3 and 4 then present those same two statistics, measured in the same 205 h^-1 Mpc box with the same initial density field, as evidence of improvement. The mass-function and bias agreement is therefore a restatement of the calibration target, not an independent prediction. The halo auto power spectrum, cross-correlation coefficient, and RSD spectra shown in Figs. 5-10 are not direct fit targets and provide some independent support, which keeps the circularity partial rather than total.

  2. fitted input called prediction [Sec. 3.1 (PGD Embedded in FastPM), after Eq. (3.6); Fig. 11]
    "These parameters are fitted by matching the matter power spectrum at all redshift simultaneously."

    The PGD parameters alpha0, A, B, kl,0, gamma, and ks,0 are fitted to the matter power spectrum of TNG300-2-Dark (and TNG300-2) over all redshifts. Figure 11 then shows that FastPM+PGD matches the reference P(k), and the paper counts this as an improvement. Since the plotted statistic is exactly the fitting target, the matter-power match is enforced by construction. The weak-lensing convergence spectra in Fig. 13 are line-of-sight projections of this same fitted P(k) and are compared to halofit rather than to an independent N-body lightcone, so they inherit, rather than independently confirm, the P(k) fit.

full rationale

Two calibration steps are explicitly fitted to the reference simulation: relaxed-FoF linking lengths and the fake-halo threshold are chosen to reproduce TNG300-2-Dark's halo mass function and bias (Sec. 2.1), and the PGD parameters are fitted to the matter power spectrum (Sec. 3.1). The paper then shows those same statistics as evidence of improvement (Figs. 3, 4, and 11), so for these statistics the agreement is partly by construction rather than an out-of-sample test. The remaining halo statistics---real-space and redshift-space auto power spectra, cross-correlation coefficient, and halo-matter cross power---are not direct fit targets and provide some independent support. The lensing convergence spectra are integrals of the fitted P(k) and are validated against halofit, not against an independent full N-body lightcone, so they do not fully escape the fit. The paper is self-contained against the TNG300 suite, but it does not test whether the fitted calibration functions transfer to a different cosmology, box size, or initial density field; that is a correctness risk rather than circularity per se. The self-citation to the authors' earlier PGD paper [10] supplies the correction ansatz, but because the parameters are re-fitted here, it is not load-bearing circularity. Overall this is partial circularity: the headline halo mass function, bias, and matter P(k) claims reduce to calibration targets, while the clustering and lensing results carry independent content.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim depends on a small number of fitted free parameters that are tuned to the reference simulation. No new physical entities are introduced; the new ingredients are algorithmic (relaxed-FOF) and a modified gravity-style correction (PGD). The axioms are mostly standard assumptions in cosmological simulations, with the ad hoc anchor being the choice that large-halo linking length must approach 0.2.

free parameters (4)
  • l1, l6, A1, A2 (linking length parameters) = not stated
    Appear in l(Np,i,z) = ((6-i)Np,1 + (i-1)Np,6)/5 with l1, l6, A1, A2; chosen to reproduce the reference halo mass function and bias (Section 2.1, Eq. 2.3-2.5).
  • B1, B2 (velocity dispersion threshold) = not stated
    r0(z) = B1 - B2 log(1+z) sets the fake-halo rejection threshold; fitted to halo bias (Eq. 2.6).
  • α0, A, B (PGD amplitude) = not stated
    log(α/α0) = A a^2 - B a; fitted to match the matter power spectrum at all redshifts simultaneously (Eq. 3.5).
  • kl,0, γ, ks,0 (PGD filter scales) = not stated
    kl = kl,0 a^γ and ks = ks,0; fitted to match the matter power spectrum (Eq. 3.3-3.6).
assumptions (4)
  • domain assumption 2LPT initial conditions with the same random seed and linear power spectrum reproduce the same large-scale structure as the TNG simulations.
    Required so that FastPM and TNG halos can be matched one-to-one (Section 2).
  • domain assumption The Evrard et al. velocity dispersion-mass scaling relation holds for halos down to ~10^11 M_sun with ~20 particles.
    Used in Eq. 2.1 to define the fake-halo rejection criterion; low-resolution halos may not follow this relation.
  • ad hoc to paper Standard FoF linking length 0.2 is correct for large halos at z=0, so the calibrated l(Np,z) is forced to approach 0.2 for large mass.
    This anchors the relaxed-FOF calibration to the prior FoF convention (Section 2.1).
  • domain assumption TNG300-2-Dark is an accurate reference for halo and matter statistics at the mass range of interest.
    The paper relies on TNG's accuracy and checks against TNG300-1-Dark, but all calibration and validation use the same reference simulation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of High mass and halo resolution from fast low resolution simulations." pith.science (2026). https://pith.science/paper/233VKAQE

@misc{pith2026190805276,
  author       = {Pith},
  title        = {Pith review of: High mass and halo resolution from fast low resolution simulations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/233VKAQE}},
  note         = {Machine review of arXiv:1908.05276}
}
read the original abstract

Generating mocks for future sky surveys requires large volumes and high resolutions, which is computationally expensive even for fast simulations. In this work we try to develop numerical schemes to calibrate various halo and matter statistics in fast low resolution simulations compared to high resolution N-body and hydrodynamic simulations. For the halos, we improve the initial condition accuracy and develop a halo finder "relaxed-FOF", where we allow different linking length for different halo mass and velocity dispersions. We show that our relaxed-FoF halo finder improves the common statistics, such as halo bias, halo mass function, halo auto power spectrum in real space and in redshift space, cross correlation coefficient with the reference halo catalog, and halo-matter cross power spectrum. We also incorporate the potential gradient descent (PGD) method into fast simulations to improve the matter distribution at nonlinear scale. By building a lightcone output, we show that the PGD method significantly improves the weak lensing convergence tomographic power spectrum. With these improvements FastPM is comparable to the high resolution full N-body simulation of the same mass resolution, with two orders of magnitude fewer time steps. These techniques can be used to improve the halo and matter statistics of FastPM simulations for mock catalogs of future surveys such as DESI and LSST.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 3 canonical work pages

  1. [1]

    Kitaura, S

    F.-S. Kitaura, S. Rodr´ ıguez-Torres, C.-H. Chuang, C. Zhao, F. Prada, H. Gil-Mar´ ın, H. Guo, G. Yepes, A. Klypin, C. G. Sc´ occola,et al. , MNRAS 456, 4156 (Mar. 2016), 1509.06400

    arXiv 2016

  2. [2]

    Joudaki, C

    S. Joudaki, C. Blake, A. Johnson, A. Amon, M. Asgari, A. Choi, T. Erben, K. Glazebrook, J. Harnois-D´ eraps, C. Heymans,et al. , MNRAS 474, 4894 (Mar. 2018), 1707.06627

    arXiv 2018

  3. [3]

    DES Y1 Results: Validating cosmological parameter estimation using simulated Dark Energy Surveys

    N. MacCrann, J. DeRose, R. H. Wechsler, J. Blazek, E. Gaztanaga, M. Crocce, E. S. Rykoff, M. R. Becker, B. Jain, E. Krause, et al. , MNRAS 480, 4614 (Nov. 2018), 1803.09795

    work page Pith review arXiv 2018

  4. [4]

    Feng, M.-Y

    Y. Feng, M.-Y. Chu, U. Seljak, and P. McDonald, MNRAS 463, 2273 (Dec. 2016), 1603.00476

    arXiv 2016

  5. [5]

    Tassev, M

    S. Tassev, M. Zaldarriaga, and D. J. Eisenstein, J. Cosmology Astropart. Phys. 6, 036, 036 (Jun. 2013), 1301.0322

    arXiv 2013

  6. [6]

    M. Levi, C. Bebek, T. Beers, R. Blum, R. Cahn, D. Eisenstein, B. Flaugher, K. Honscheid, R. Kron, O. Lahav, et al. , arXiv e-prints (Aug. 2013), 1308.0847

    arXiv 2013

  7. [7]

    LSST Science Collaboration, P. A. Abell, J. Allison, S. F. Anderson, J. R. Andrew, J. R. P. Angel, L. Armus, D. Arnett, S. J. Asztalos, T. S. Axelrod, et al. , arXiv e-prints (Dec. 2009), 0912.0201

    arXiv 2009

  8. [8]

    Aghamousa, J

    DESI Collaboration, A. Aghamousa, J. Aguilar, S. Ahlen, S. Alam, L. E. Allen, C. Allende Prieto, J. Annis, S. Bailey, C. Balland, et al. , arXiv e-prints (Oct. 2016), 1611.00036

    arXiv 2016

Show all 19 references
  1. [9]

    DeRose, R

    J. DeRose, R. H. Wechsler, J. L. Tinker, M. R. Becker, Y.-Y. Mao, T. McClintock, S. McLaughlin, E. Rozo, and Z. Zhai, ApJ 875, 69, 69 (Apr. 2019), 1804.05865

    2019 arXiv

  2. [10]

    B. Dai, Y. Feng, and U. Seljak, J. Cosmology Astropart. Phys. 11, 009, 009 (Nov. 2018), 1804.00671

    2018 arXiv

  3. [11]

    Springel, R

    V. Springel, R. Pakmor, A. Pillepich, R. Weinberger, D. Nelson, L. Hernquist, M. Vogelsberger, S. Genel, P. Torrey, F. Marinacci, et al. , MNRAS 475, 676 (Mar. 2018), 1707.03397

    2018 arXiv

  4. [12]

    A. E. Evrard, J. Bialek, M. Busha, M. White, S. Habib, K. Heitmann, M. Warren, E. Rasia, G. Tormen, L. Moscardini, et al. , ApJ 672, 122 (Jan. 2008), astro-ph/0702241

    2008 arXiv

  5. [13]

    N. Hand, Y. Feng, and C. Modi, bccp/nbodykit: nbodykit v0.2.9 (Nov. 2017), https://doi.org/10.5281/zenodo.1051244

    2017 doi

  6. [14]

    Hoekstra and B

    H. Hoekstra and B. Jain, Annual Review of Nuclear and Particle Science 58, 99 (Nov. 2008), 0805.0139

    2008 arXiv

  7. [15]

    Kilbinger, Reports on Progress in Physics 78(8), 086901, 086901 (Jul

    M. Kilbinger, Reports on Progress in Physics 78(8), 086901, 086901 (Jul. 2015), 1411.0115

    2015 arXiv

  8. [16]

    Zonca, L

    A. Zonca, L. Singer, D. Lenz, M. Reinecke, C. Rosset, E. Hivon, and K. Gorski, The Journal of Open Source Software 4, 1298 (Mar. 2019)

    2019

  9. [17]

    K. M. G´ orski, E. Hivon, A. J. Banday, B. D. Wandelt, F. K. Hansen, M. Reinecke, and M. Bartelmann, ApJ 622, 759 (Apr. 2005), arXiv:astro-ph/0409513. – 18 –

    2005 arXiv

  10. [18]

    Takahashi, M

    R. Takahashi, M. Sato, T. Nishimichi, A. Taruya, and M. Oguri, ApJ 761, 152, 152 (Dec. 2012), 1208.2701

    2012 arXiv

  11. [19]

    C. Modi, E. Castorina, Y. Feng, and M. White, arXiv e-prints (Apr. 2019), 1904.11923. – 19 –

    2019 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.