External review, round 2 · reviews v3 · verdict: major errors

# External verification: 09-anchor-formation

- **Subproject:** 08-ilang-propagation
- **Package:** 09-anchor-formation
- **Verified version:** v3
- **External round:** 2 of 3
- **Date:** 2026-10-26T17:36:53+01:00
- **Focus points:** none

---
VERDICT: major errors

## Summary

The exact finite-volume response, its uniform/place-dependent split, and the recurrence argument are sound, and the supplied code supports these formulas over the tested windows. However, the proofs of uniform-in-λ formation bounds mishandle the poles associated with localized modes: removing a pole from an anchor transfer kernel does not remove only the localized spectral contribution of the resulting field. This undermines the all-later-λ arguments in both dimensions. The derivation also leaves the required critical anchor strength unresolved.

## Issues

### I1. Localized-mode poles are incorrectly removed from convolution kernels
- **Location:** Step 6, derivation of (9.10) and the image bound in (9.18); Step 8, “Every later λ”
- **Severity:** major
- **Problem:** The decomposition of the anchor transfer kernel into an integrable part and an undamped pole term is not compatible with subsequently treating the entire pole-term convolution as an exponentially localized contribution. For example, if an above-band pole contributes
  \[
  \chi_b(s)=\frac{B_b}{s^2+\nu_b},
  \]
  its contribution to the image strength has transform
  \[
  \widehat{\sigma_b}(y,s)
  =\frac{B_b\gamma_s(y)}{s(s^2+\nu_b)}.
  \]
  Its λ-mean is \(B_b g_\infty(y)/\nu_b\), which decays algebraically, not exponentially. Only the bound-state spectral residue of the *complete response* is exponentially localized; convolution with the pole part of the transfer kernel also generates continuous-spectrum and static contributions. Consequently, the Young inequality used for (9.18) omits terms when \(\kappa_a>\kappa_h/\gamma_3\).

  The same problem occurs on the line. The stated transform
  \[
  \widehat K_1(s)=\frac1{1+\kappa_a\gamma_s(0)}
  \]
  has imaginary-axis poles for every positive \(\kappa_a\), at the above-band eigenvalue
  \[
  \nu_b=2\kappa_h+\sqrt{4\kappa_h^2+\kappa_a^2}.
  \]
  Thus \(K_1\) is not an integrable kernel as asserted. If those poles are subtracted, its transform is no longer the one displayed, and its integral is generally no longer zero. The subsequent convolution argument therefore does not establish (9.13) uniformly for every later λ.
- **Suggested fix:** Separate localized and continuous contributions at the level of the complete anchored response, or explicitly retain and bound all contributions generated by the transfer-kernel poles. Re-establish the uniform image bound and the line convolution estimate before using them to prove formation for every later λ. The finite-window numerical checks do not replace these estimates.

### I2. The critical anchor strength remains outside the requested determination
- **Location:** Setup, “Excluded members”; Step 6; Open issues; Results 4–6 and 10
- **Severity:** major
- **Problem:** The question specifies every \(\kappa_h,\kappa_a>0\), but the derivation excludes \(\kappa_a=\kappa_h/\gamma_3\) from the image asymptotics and all positive cubic formation statements. This is not a vanishing-reference or short-distance cancellation of the kind the question permits excluding: ordinary source and observation placements still have nonzero long-range references at this anchor strength. Identifying the slower kernel decay does not determine whether formation occurs, its speed, or its dependence on the image geometry there. Hence item 2 is not settled over the stated class.
- **Suggested fix:** Analyze the threshold contribution sufficiently to establish or disprove the required formation bounds, and state the corresponding criterion and speed. Until then, explicitly characterize the package as incomplete for the class requested in the question.

### I3. The continuum ring error estimate is asserted without justification
- **Location:** Step 9, “Ring, order (O2), κₐL ≫ κₕ”, and (9.15)
- **Severity:** minor
- **Problem:** The passage from the discrete anchored ring to a fixed-end continuum string is asserted, followed by an \(O(1)\) error estimate said to grow at most linearly in τ. The reflection-amplitude limit and the standard sine-mode formulas establish the candidate leading expression, but they do not by themselves establish this error bound after summing modes or specify its uniformity in the source fraction α. The code tests finite τ and a single α, so it does not justify the general error claim. Under M4, the continuum string description must also follow from the stated discrete operator rather than serve as a physical input.
- **Suggested fix:** Obtain the low-label quantization and mode weights directly from the discrete eigenvalue equation, state the allowed range of α and τ, and bound the mode-sum error. Alternatively, weaken the error claim to the convergence actually established.

## Focus points

None given.

Author: Ivan Honis · ivan.honis@ndot.io · License: CC BY 4.0 · Human–AI note