Cookie Consent by Free Privacy Policy Generator Editorial review — “Geometry of Seeing” & “Geometry of the Cosmic Web” | Igor Moiseev

Editorial review — “Geometry of Seeing” & “Geometry of the Cosmic Web”

Editorial review — “Geometry of Seeing” & “Geometry of the Cosmic Web”

Referee-style review of both blog series as submissions to a joint mathematics/astrophysics venue. Compiled 2026-07-08 from three specialist passes (cosmic-web Parts 1–2; appendices B1–B5; appendices A1–A4) plus a line-level pass over Seeing Parts 1–4, A5, and the spacetime capstone.

Editorial summary and verdicts

Geometry of Seeing (P1–P4, A1–A5). A rare object: a pedagogically serious, internally cross-checked exposition of sub-Riemannian geometry on SE(2) written by a co-author of the source literature, with runnable self-checks and correct sign discipline almost everywhere. Two problems block publication as-is: (1) a series-level conflation of the SE(2) sub-Riemannian problem with Euler’s elastica problem — appendix A3 contains the mathematically correct account, and the rest of the series contradicts it; (2) the “Open Problem” framing in Part 1 misstates the state of the art relative to Sachkov’s 2010–2011 theorems and now also contradicts Part 4. Verdict: major revision (narrative re-anchoring; no result collapses).

Geometry of the Cosmic Web, empirical core (P1–P2, B1–B5). Methodologically above field norms — pre-registered kill criteria, a self-caught benchmark artifact honestly reported, jackknife-deflated significances, physically-null controls, an oracle-bound technique worth imitating. The physics is consistent with standard ΛCDM structure formation except where noted. Blocking issues: the H3 verdict as printed contradicts its own kill criterion, the 9σ headline needs re-framing against the published tSZ-filament literature, one unit error, and a required pre-registration deviations table. Verdict: minor-to-moderate revision.

Spacetime capstone (cosmic-web P3). The Klein→Cartan→gauge story and the tidal dictionary are textbook-correct; the filament-theory section matches standard Zel’dovich/caustic theory. Needs hedging on three physics-status claims (torsion bounds, GW-polarization exclusions, Einstein–Cartan bounce attribution). Verdict: minor revision.

Part A — Findings that touch established theory

(the “errors and mishaps with the current theory” the review was asked to prioritise)

A1. [MAJOR — Seeing, series-wide] SR geodesics on SE(2) are not Euler elastica, and the series says they are — except A3, which gets it right. Part 2 (§”From the pendulum to the curve”) claims the SR geodesic’s plane projection has curvature κ(s) = 2k·cn(s|k²) after the ds = |h₁|dt reparametrisation, “carried out in full in Appendix A3.” A3 actually derives the opposite and is correct: the SR projection has κ_SR = cot(φ/2) — unbounded, with cusps where the forward velocity h₁ vanishes (which happens on every inflectional libration) — and the smooth 2k·cn / 2·sech / 2·dn profiles belong to the sister problem with u₁ ≡ 1 (Euler’s elastica). Independent checks: cot(φ/2) satisfies no Duffing equation κ’’ + κ³/2 − μκ = 0; and the κ = 2k·cn ⇒ θ = 2 arcsin(k·sn) ⇒ y = 2k(1−cn) triple verified numerically in this repo is exactly the elastica curve, self-consistently. This matches the published distinction (Duits–Boscain–Rossi– Sachkov: SE(2) SR geodesics are generically cuspidal; elastica are a different family) — and A5’s own aside about “cuspidal trajectories” already knows it. Inherited by: P2 §pendulum-to-curve and its figures (“every geodesic of SE(2) projects to one of these curve types”); P1 Fig 4 caption; A1:399–403, 520–521; A2:329–331; A4:282–288; A5’s boxed “plane projection of the SR geodesic” (its own hedge notwithstanding); P3/P4’s labeling of the y = 2k(1−cn) family as “the SR geodesic”; the capstone’s synthesis-table cell “κ = 2k cn”. Fix: adopt A3’s “shared vertical subsystem, two horizontal problems” formulation series-wide; state precisely which problem each closed form solves, quoting Sachkov (2011) eqs. (24)–(28) for the SR side. The Maxwell/cut analysis of P3–P4 stands for the elastica family as computed; whether each “first Maxwell time = 4K(k²)/ω₀” claim is the SR-problem statement, the elastica statement, or an upper branch of a min over strata must be verified against Moiseev–Sachkov (2010) — see A5 below.

A2. [MAJOR — Seeing P1] The “Open Problem” box misstates the state of the art and now contradicts Part 4. P1 presents t_cut = t_MAX as “conjectured — and verified numerically,” with “a complete proof in all degenerate cases remains open.” Sachkov (ESAIM COCV 2010, 2011) proved the cut time and full optimal synthesis for the SE(2) SR problem; Part 4 says so (“SE(2) itself is solved”) and relocates the open problem to the general Maxwell-equals-cut question. Re-word P1’s box to match P4: either state precisely which degenerate boundary statement remains unproved (with citation) or frame the open problem as the general theorem beyond SE(2).

A3. [MAJOR — Cosmic P1/P2] “Matter falls across filaments, not flowing along them” flattens a standard nuance — and the series’ own data shows the along-flow. Standard picture: filament growth is transverse (the posts get this right), but flow along filaments feeding nodes is real and significant (filaments as transport highways). The series’ own measurements show it: ⟨|v̂·e₃|⟩ ≈ 0.54–0.57 > 0.5 in every bin, P2-statistic 0.364 > 1/3 inside spines. Correct statement: “filament growth is transverse; along-spine drainage toward nodes exists and is detected here at weak amplitude” — with the caveat that 128³ PM at 1 h⁻¹Mpc under-resolves intra-filament streaming (needs external check against Tempel+2014-class alignment amplitudes).

A4. [MAJOR — Cosmic P1] H3’s ✓ contradicts its own pre-registered kill criterion, and the 9σ headline misreads against the literature. Pre-registration made H3 comparative (lift spines must at least match the standard skeleton’s stack significance); measured: Hessian 8.99σ vs lift 6.86σ, and after halo masking 1.95σ vs 0.07σ — by its own criterion H3 is killed or at best confirmed only in the weakened “both webs trace real gas” sense. Separately, P1’s bare “up to 9σ” implies a 9σ filament detection, which would conflict with the field; the series’ own decomposition shows the 9σ is dominated by the tracer galaxies’ halo gas, with the inter-halo bridge at ≈2σ per instrument and y ~ 1.2–1.4×10⁻⁸ — which matches published bridge detections (de Graaff+2019, Tanimura+2019). Re-grade the H3 row; reframe the headline as “halo-dominated 9σ; bridge component ≈2σ, reproducing published amplitudes.”

A5. [MAJOR-verify — Seeing P3, project page, A5] Symmetry group and “first” Maxwell time vs the source paper. The series commits to ℤ₂×ℤ₂ (time reversal, mirror, composite) and to t¹_MAX = 4K(k²)/ω₀ for the inflectional family. Moiseev–Sachkov (2010) work with the full reflection group (ε¹…ε⁷ ≅ ℤ₂×ℤ₂×ℤ₂ in the companion elastica paper) and express the first Maxwell time as a minimum over strata, at least one branch of which is a transcendental-equation root rather than 4K. Verify against the source: (a) group order as attributed; (b) whether 4K is the binding stratum for all k or an upper branch. If the latter, P3’s box and P4’s “cut time = 4K” require the min-form, and the “same elliptic period” punchline needs a qualifier.

A6. [Cosmic P2/B1/B4] “Adhesion (1989)” misattribution. The ladder row “stick at walls — adhesion (1989)” scoring 5.34 (worse than ZA) is the series’ isotropic-sticking proxy, not the published adhesion model (Burgers/Hopf–Lax), which is generally regarded as improving on ZA. Relabel; do not let the proxy’s failure read as a failure of Gurbatov–Saichev–Shandarin adhesion.

A7. [Capstone] Three physics-status claims need hedging. (i) “Torsion has never been detected; no experiment is yet precise enough to bound it meaningfully” — too strong: laboratory/astrophysical bounds on specific torsion couplings exist (spin-polarized-matter experiments; Kostelecký-framework constraint tables). Say “bounds exist on specific couplings; no detection.” (ii) “absence of extra GW polarisations rules most propagating-torsion versions out” — fair qualitatively but needs a citation and “constrains” rather than “rules out.” (iii) The Einstein–Cartan bounce should be attributed (Popławski). Also, Newton–Cartan arises from gauging the Bargmann (centrally extended Galilei) group — footnote-level correction.

A8. [Seeing P2/A3] “Euler spiral” mislabel of the separatrix. The E = 1 elastica κ = 2 sech s is the borderline/solitary elastica (one loop, Δθ = 2π, asymptotic to a line) — not the Euler–Cornu clothoid (κ ∝ s, two eyes), whose phenomenology (“Fresnel”, “winding around two limiting points”) P2 wrongly attributes to it. Rename “critical/borderline elastica”; note P2’s own Δθ = 2π line is correct and inconsistent with the clothoid reading. (The k→1 inflectional curves do wind into two eyes; the mislabel is specifically about the separatrix itself.)

A9. [Seeing A2] Chow–Rashevskii proof sketch is false as stated. The composition-of-flows map has rank ≤ 2 at the origin (bracket directions appear at second order — that is the whole subject); the inverse-function-theorem argument fails at t = 0. Fix via the endpoint map at nonzero parameters or cite the Orbit Theorem.

Part B — Factual/internal errors (fix-list)

Cosmic web.

  1. P2 fig-ladder caption: “grid cells of about 1.4 million light-years” — 1 h⁻¹Mpc ≈ 4.7 Mly (the posts’ own footnotes get it right); “1.4” is the Mpc value mislabeled.
  2. B4 transfer table mixes units against its own source: the coarse row is in 2 h⁻¹Mpc voxels (≈4.18/4.08 Mpc/h physical) under a “physical h⁻¹Mpc” header — as printed, coarsening appears to halve the physical error, contradicting B4’s “error scale is set by physics, not the grid.”
  3. B1:235 quotes 0.364 as a velocity streaming statistic; it is E2’s chord-deviation direction statistic P2 (null 1/3). The velocity statistic is P1 ≈ 0.56 (null 1/2). Name both.
  4. Pre-registration deviations to table (mandatory for the framing): DisPerSE/NEXUS never run (in-house Hessian substituted); E0 testbed changed to Voronoi toys; M1 tolerance 0.5 → 2 vox, junctions → 3 vox; orientations 60–160 → 42; M5 controls ≥1000 → 200; H4’s M4 replaced by the direction-statistic protocol (state the substitution; the inference itself is sound).
  5. “Tidal tensor of the model’s own density” (B1/B4): undefined object as phrased — say “computed from the model’s density via Poisson.” Add one sentence noting the density-Hessian detector (axis = top eigenvector) vs potential-Hessian tidal frame (axis = e₃) opposite-convention trap.
  6. B4 damping recipe: state the frame (damping toward box rest is not Galilean-invariant; defend the bulk-velocity≪infall assumption), and discuss why both-perpendicular damping beats pancake-ordered damping when B2’s own ordering predicts otherwise.
  7. Number drift to reconcile: ZA 4.97/4.98/5.00; E5 4.91±0.14 vs “4.93”; “cutting error by 3–17%” vs table span −2…−22%; B3 text Δ=−0.045…−0.066 vs figure max −0.057; “70+ control stacks” vs 200; 1.2/0.5 printed as ≈2.2σ (it is 2.4σ); “89% of the gap” recomputes to 90.2%; “512 h⁻¹Mpc cubes” are radially ~250 h⁻¹Mpc slabs; P2:560 “4.4 h⁻¹Mpc residual” is the along-axis component (3D RMS ≈8); fig-e1b caption “leads at both densities” contradicts the text’s tie at σ∥=3.
  8. ZA displacement written with the raw peculiar potential is dimensionally inconsistent in P1 (B1’s Φ⁽¹⁾ and the capstone’s “suitably scaled” phrasing are fine); define ψ₀ = −Φ₀ and separate the two ψ’s in B1 (velocity potential vs displacement divergence).
  9. B5: add the n_e² (X-ray) vs n_e (tSZ) scaling and the WHIM 10⁵–10⁷ K range; qualify “WHIM sits near y ~ 10⁻⁸” as the stacked-bridge value (prominent individual bridges reach 10⁻⁷–10⁻⁶); per-bin 10.1σ > overall 8.99σ needs the correlated-bins clause; ACT large-scale-filtering diagnosis needs an external check (DR6 co-adds Planck at large scales).
  10. Citation hygiene: E7b report cited but lives inside E8; B3 figure cites E0/E0a absent from its reference list.

Seeing.

  1. A3:360 generator typo: ξ = u₁E₁ + u₂E₃ (not E₂) — as written the reconstruction is a pure-translation flow contradicting the chart equations two lines below.
  2. P1’s ω₀(k) (“linearised oscillation frequency”, k-dependent) vs A5’s cylinder-coordinate ω₀ vs P3’s ω₀ = 1 normalization — unify the definition once, in one place.
  3. Abnormal extremals: P2’s “abnormal ⟺ h₁ ≡ 0” should be h₁ ≡ h₂ ≡ 0 (contact structure ⇒ abnormals are trivial/constant); the “Sachkov 2004 (generalized Dido)” citation is the wrong problem for this claim.
  4. A3 PMP box omits the free-terminal-time condition ℋ ≡ 0 (what fixes the unit level); transversality unstated.
  5. A4 Fig A4.2 JS bug: slider handler resets the AGM iteration before the new m is read — iteration converges against the previous m while the reference line shows the new one. Reorder.
  6. Smaller: A1 Fig A1.1 “pure a₂ → circle through origin” (pure rotation fixes the origin; the circle is the screw case); Part 2 “K(k²) is the exact half-period of sn” — quarter-period (its own code comment is right); A1 §6 asserts rather than derives the coadjoint cylinders that P2 promises it derives; A3 Fig A3.3 caption promises a √c-rescale the code ignores; A5’s conjugate_time_inflectional docstring says “closed form” over an approximation; A3’s Part-2 “quotation” is a paraphrase inside quotation marks; verify A1 ref “ESAIM COCV 17(4)” issue number and P1’s “Fig. 34” pointer.

Part C — Statistics & methodology assessment

The B3 protocol (matched realized length, held-out seeds, competitor-favoring calibration, reference-free mass criterion, jackknife over exchangeable patches, provenance hashing) is sound and better than most published web-finder benchmarks; the corrected-benchmark story (a p≈10⁻¹⁴ “win” reversed by a 19% length mismatch) is exemplary and publishable on its own. An editor would still require: (i) an explicit look-elsewhere / multiple-testing policy for the E3→E3e estimator sequence (pre-registration appears only in B5’s E3c gate); (ii) honesty about n = 3 held-out seeds (variance-of-variance; report spread or add seeds); (iii) “statistical tie” claims (p = 0.36, 0.55) upgraded to equivalence statements (CI on Δ or TOST); (iv) error propagation for the null subtraction, and the “9σ” acknowledged as a Gaussian-tail extrapolation of a 200-draw empirical null; (v) a completeness-vs-length curve rather than one matched operating point; (vi) E2’s far-bin (16–64 vox) “transverse infall” reinterpreted — the nearest-spine frame is not meaningful 30+ h⁻¹Mpc away.

Part D — Strengths (what an editor would highlight)

Part E — Promising directions (“perspective aspects”)

  1. Derive β. The capstone makes it well-posed: is β≈0.6 the leading post-Zel’dovich tidal term evaluated in the eigenframe? A derivation (or a clean failure) upgrades the damping model from empirical to physical. Related: a MUSCLE×damping field-level hybrid.
  2. The anisotropy-descriptor win is real and portable: tidal-alignment 0.73–0.76 vs 0.67 (sims) and 0.677 vs 0.617 (BOSS) suggests applications to intrinsic alignments / spin–filament studies — the lift’s one clean survivor deserves its own paper.
  3. Caustic-skeleton bridge: the capstone’s catastrophe dictionary connects the series to an active program (Feldbrugge–van de Weygaert); the tidal-eigenframe detectors here are natural inputs to it.
  4. Sub-Lorentzian SE(2) analogues of the Maxwell/cut results (P4’s open direction) — genuinely uncharted, and this author is unusually placed to chart it.
  5. Small-scale bridge estimators on ACT-class beams, with the leakage-null machinery already built, could turn the ≈2σ bridge into a competitive measurement.

Part F — Prioritized recommendations

Before anything ships: A1 (re-anchor SR vs elastica series-wide, from A3’s correct version), A2 (rewrite P1’s Open-Problem box to match P4), A4 (re-grade H3; reframe 9σ), B-list items 1–4 (unit error, B4 table units, mislabeled statistic, deviations table). Second pass: A3/A5/A6/A7/A8 wording and verifications against the source papers; Part C’s statistics upgrades; the remaining B-list. Then: the series is publishable — the cosmic-web methodology/damping-model material arguably in a journal, not only a blog.

Addendum (2026-07-08) — fixes applied

All text-applicable findings above were applied in commits f50633b (Seeing) and fa44cbc (cosmic web), with one upgrade: the source papers were fetched and read (arXiv:0807.4731; 0903.0727 §§2–3), which sharpened three items beyond the review’s “verify” requests — the reflection group is (ℤ₂)³; the SR cut time on the inflectional family is 2K(k²) (half the series’ previous 4K claim, which is the elastica mirror-tie value); and Sachkov Thm 2.1 shows the inflectional/critical SR families have no conjugate points at all (finite conjugate bounds exist only for the rotating family, switching branches at k₀ ≈ 0.909). P1’s box, P2’s preview, P3, P4 (text and both figures), A5, and the project page were rewritten around the verified statements.

Not applied (analysis-level, not text fixes): error propagation for the B5 null subtraction; a completeness-vs-length curve for B3; verifying the “Fig. 34” pointer in P1’s caption; independent re-derivation of the CAMELS twin claim. These remain open in Part C / B-list.