# Night Block 2 referee pass — 10–11 November 2026 (PST)

Started 11:32 PM PST, November 10. Separate-agent referee of the Block 1
night proofs, written without relying on the night log's conclusions.
Research notes only: no site edits, no deployment, no messages, no
routine changes. Not an RH claim; no novelty or priority claim.

## Verdicts

| Item | Verdict |
|---|---|
| `order-solomon-column-pair-proof.md` (all-level Solomon numerator) | **PASS** — no mathematical error found |
| Q_4 = p²+2p+2 re-derived by a different method (full 4D HNF brute force) | **AGREES** |
| Script replay `night_solomon_column_pairs.py --mode pairs / --mode recursion` | **PASS**, Q_4/Q_5/Q_6 as expected |
| `order-solomon-chebyshev-short-proof.md` (unit-circle theorem, unconditional) | **PASS** — no mathematical error found |
| Script replay `night_solomon_chebyshev_check.py` | **PASS** (480 + 81 + 60 exact checks) |
| `night_solomon_verify_records.py` after replay | **PASS** (31 count records, 13 prefixes, 185 layers, 240 Sturm, 22 isolates) |

## 1. Column-pair count, checked line by line

[project, proved — refereed] Every step below was re-derived by hand.

- *Direct sum.* For X = [[a,b],[p^r c,d]], X·E_11 = [[a,0],[p^r c,0]] and
  X·E_22 = [[0,b],[0,d]]. E_11, E_22 lie in O_r and sum to 1, so a right
  ideal I satisfies I = I·E_11 ⊕ I·E_22, with I·E_11 = I ∩ (first column)
  and I·E_22 = I ∩ (second column). Correct.
- *Maps.* X·E_12 = [[0,a],[0,p^r c]], i.e. (b,d) = (a, p^r c), so
  F = diag(1,p^r). X·(p^r E_21) = [[p^r b,0],[p^r d,0]]; since the lower-left
  entry is p^r·c′, the encoded column is (a′,c′) = (p^r b, d), so
  G = diag(p^r,1). FG = GF = p^r. Correct. Since E_11, E_12, p^r E_21, E_22 is a
  Z_p-basis of O_r, conditions FU ⊂ V, GV ⊂ U are necessary and sufficient.
- *Four membership tests.* With U = ⟨(p^a,0),(t,p^b)⟩, V = ⟨(p^A,0),(T,p^B)⟩:
  F(p^a,0) ∈ V ⇔ A ≤ a; G(p^A,0) = (p^{A+r},0) ∈ U ⇔ a ≤ A+r;
  F(t,p^b) = (t,p^{b+r}) ∈ V ⇔ B ≤ b+r and t ≡ p^{r-e} T (mod p^A);
  G(T,p^B) = (p^r T,p^B) ∈ U ⇔ b ≤ B and p^r T ≡ p^e t (mod p^{A+d}),
  with d = a−A, e = B−b. These are exactly (5)–(6) of the note. Correct.
- *Residues and carries.* For each of p^A values of T, the t's form one
  coset of p^A Z modulo p^{A+d}: p^d canonical representatives. The second
  congruence reduces to p^{A+e} z ≡ 0 (mod p^{A+d}); replacing the raw
  expression by its canonical representative changes p^e t by a multiple of
  p^{A+d+e}, so the carry is harmless. Count p^A·p^{min(d,e)}. Correct,
  including A=0, d=0, e=0, p=2.
- *Index.* a+b+A+B = (A+d)+b+A+(b+e) = 2A+2b+d+e. Correct; the geometric
  sums give 1/((1−pv²)(1−v²)).
- *Scope.* The count is over all full-rank Z_p-sublattices of O_r closed under
  right multiplication by O_r — every embedded right submodule of finite
  index — not ideal classes and not only invertible ideals. (For example the
  two index-p ideals are noninvertible; invertible ideals have square index.)
  Correct.
- *Recursion completeness (r ≥ 1).* For r ≥ 1, f = E_12 and g = p^r E_21 span a
  square-zero ideal mod p (fg = p^r E_11 ≡ 0, gf = p^r E_22 ≡ 0), quotient
  F_p × F_p, so all simple right modules have p elements and every finite-index
  right ideal is reached by a chain of index-p steps. At r = 0 the simple
  modules of M_2(F_p) have p² elements and an index-p recursion fails — as the
  note says. Qualification correct and correctly applied (all recursion
  runs have r ≥ 1).
- *Palindrome.* (d,e) ↦ (r−d,r−e) gives min(r−d,r−e) = r−j+min(d,e) when
  d+e = j. Correct.
- *Agreement with the afternoon conjecture.* 1+Σ_{k=1}^j p^{⌊k/2⌋} equals the
  afternoon guess Σ_{m=0}^j p^{⌊m/2⌋}, and the upper half matches its
  palindrome. Correct.

[project] Presentation remarks only (no error): the note could say in one line
that "all local Eichler orders of level p^r are conjugate to O_r" [known:
Voight, *Quaternion Algebras*, Def. 23.4.11 / 23.4.12ff], which is what makes
the specific O_r representative. An elementary reading:
p^{min(d,e)} = #Hom(Z/p^d, Z/p^e), so Q_r(v) = Σ_{d,e≤r} #Hom(Z/p^d,Z/p^e) v^{d+e}
[project observation; resemblance to Hall-algebra counts is analogy only].

## 2. Independent re-derivation of Q_4 (and more) by a different method

[project] New script `block2_bruteforce_hnf.py` (standard library, written this
block, imports no project code). It lists **every** row-Hermite-normal-form
sublattice of Z^4 of index p^k — the enumeration size was checked against the
known total count of sublattices (local ζ(s)ζ(s−1)ζ(s−2)ζ(s−3)): 6,347,715 at
(p,k)=(2,7), 25,095,280 at (3,5), 320,327,931 at (5,4), all equal — and keeps
those closed under right multiplication by the four Z-basis matrices of O_r,
using literal 2×2 integer matrix products. No column splitting, no index-p
recursion, no hand classification. Q is extracted afterwards by multiplying
by (1−v²)(1−pv²).

| (p,r,kmax) | A_0..A_kmax | Q extracted | closed form | match |
|---|---|---|---|---|
| (2,0,4) | 1,0,3,0,7 | 1,0,0,0,0 | same | yes |
| (2,1,4) | 1,2,5,6,13 | 1,2,2,0,0 | same | yes |
| (3,1,4) | 1,2,7,8,25 | 1,2,3,0,0 | same | yes |
| (2,3,8) | 1,2,7,12,27,40,75,96,171 | 1,2,4,6,8,8,8,0,0 | same (full palindrome + zero tail) | yes |
| (2,4,6) | 1,2,7,12,29,44,89 | 1,2,4,6,10,12,16 | same | yes |
| (2,6,7) | 1,2,7,12,29,46,95,142 | 1,2,4,6,10,14,22,28 | same | yes |
| (2,8,8) | 1,2,7,12,29,46,95,144,273 | 1,2,4,6,10,14,22,30,46 | same | yes |
| (3,4,4) | 1,2,9,16,50 | 1,2,5,8,17 | same | yes |
| (3,5,5) | 1,2,9,16,50,84 | 1,2,5,8,17,26 | same | yes |
| (5,4,4) | 1,2,13,24,110 | 1,2,7,12,37 | same | yes |

[project] So, by a genuinely different method: Q_4 = 10 (p=2), 17 (p=3), 37 (p=5)
= p²+2p+2; Q_5 = 14, 26 and Q_6 = 22 at p=2 agree; at p=2 the coefficients
Q_7 = 30, Q_8 = 46 also agree with the formula. The raw counts A_4 = 29, 50, 110,
A_5 = 46, 84, A_6 = 95 match the Block 1 table exactly. **Agreement, no
disagreement.** Records: `block2-bruteforce-*.json`, run logs
`block2-bruteforce-*-run.txt`.

## 3. Script replay

[project] Re-ran `python3 night_solomon_column_pairs.py --mode pairs` and
`--mode recursion` (11:33 PM PST): both `all_passed=True`. Extracted from the
fresh JSON records, in both modes: Q_4 = (10,17,37,65), Q_5 = (14,26,62,114),
Q_6 = (22,53,187,457) at p = (2,3,5,7). These are Q coefficients, not A counts.
The old `order_solomon_independent.py` main routine was NOT run.
The replays rewrote `night-solomon-pairs-20261110.json`,
`night-solomon-recursion-20261110.json`, `night-solomon-chebyshev-20261110.json`
(same content, new timestamps); `night_solomon_verify_records.py` was then
re-run and passed, refreshing the manifest's hashes (previous manifest kept
at /tmp/verif-before-block2.json for this session only).

## 4. Chebyshev short proof, checked line by line

[project, proved — refereed]

- (1): pairs with d−e = ±h contribute ρ^h Σ_{e=0}^{r−h} z^{2e−(r−h)} = ρ^h U_{r−h}
  each, so z^{−r}P_r = U_r + 2Σ_{h≥1} ρ^h U_{r−h}; degree r, leading 2^r. Correct.
- Telescoping: the coefficient of ρ^m in (1−2ρx+ρ²)S_r is U_{r−m} − 2xU_{r−m+1}
  + U_{r−m+2} (terms present). m = 0: U_r; m = 1: U_{r−1} − 2xU_r = −U_{r+1};
  2 ≤ m ≤ r: 0 by the recurrence; m = r+1: U_1 − 2xU_0 = 0; m = r+2: 1. So
  D·S_r = U_r − ρU_{r+1} + ρ^{r+2}. Correct.
- U_{n+1} = xU_n + T_{n+1}: true at n = 0,1 and both sides obey the same
  recurrence. Then D·C_r = (1−ρ²)U_r − 2ρT_{r+1} + 2ρ^{r+2}. Correct.
- At x_m = cos(mπ/(r+1)): U_r = 0, T_{r+1} = (−1)^m, so D·C_r = 2ρ(ρ^{r+1} − (−1)^m):
  positive for odd m, negative for even m (uses 0 < ρ < 1). D ≥ (1−ρ)² > 0 on [−1,1].
- x = −1: U_r(−1) = (−1)^r(r+1), T_{r+1}(−1) = (−1)^{r+1}, giving
  (−1)^r D C_r(−1) = (1−ρ²)(r+1) + 2ρ + 2(−1)^r ρ^{r+2} > 0 (odd r: 2ρ(1−ρ^{r+1}) > 0).
  Sign at x_r is (−1)^{r+1}, opposite to (−1)^r at −1. Correct, including r = 1.
- Hence r sign changes on x_1 > … > x_r > −1: r distinct roots in (−1,1), all of
  them since deg C_r = r. Each gives two distinct non-real z on |z| = 1, distinct
  pairs for distinct x: 2r distinct roots of the degree-2r P_r (leading
  coefficient p^r ρ^{2r} = 1). So all roots of Q_r are simple with |v| = p^{−1/2},
  with the angle brackets mπ/(r+1) < θ_m < (m+1)π/(r+1). Correct.
- Hypotheses p > 1 real, r ≥ 1: correct and sufficient.

[project] Replay of `night_solomon_chebyshev_check.py`: all_passed = True;
480 exact rational identities, 81 bivariate symbolic identities, 60 Laurent
comparisons.

## 5. Literature check (agenda item 3), about 15 minutes

What was actually read, and what it shows:

- [known] Voight, *Quaternion Algebras* (jvoight.github.io/quat-book.pdf,
  downloaded and text-searched this block): (25.3.1)/(25.3.2) define ζ_O as a
  sum over **invertible** right ideals, with N(I) = nrd(I)²; the Eichler mass
  formula for Eichler orders (Thm 25.3.18, Main Thm 25.3.19, Brzezinski 1990,
  Körner 1987) goes through unit indices λ(O,p). A Chapter 26 exercise (parts (a)–(e), near printed p. 442) with the
  "a_{p^e}(O_n)" recursion is for the maximal order M_n(R). In the sections
  searched I did **not** find an explicit invertible local factor at
  non-squarefree Eichler level, nor any all-module (Solomon) count.
- [known] Lynch, arXiv:2607.12302 (abstract/introduction, Theorem 1.1, read via
  search excerpt; also read in Block 1): an effective general formula
  ζ_Λ(L,M;s)/ζ_{Λ′}(L′;s) as a finite Möbius-weighted sum over submodules with
  IL′ ⊂ Λ′X, I the conductor-type ideal; explicit applications are to
  Z_p[Z/pZ]. In principle the Eichler numerator is computable from it (here
  I = p^r M_2(Z_p)), but no Eichler formula is displayed there.
- [known] Bushnell–Reiner, Bull. AMS 2 (1980) announcement (read via the search
  index's text of the AMS PDF; direct PDF download was blocked by the site):
  Theorem 4 gives a local functional equation for integral group rings, and
  the authors remark that the group-ring hypothesis "is only needed to ensure
  that Λ is self-dual with respect to some suitable pairing." Eichler orders
  are Gorenstein (Voight 24.1.1–24.1.2), so a classical FE covering our
  palindrome may exist in Bushnell–Reiner's full paper; this block did not
  verify that it applies (see the next paragraph).
  **Confirmed from the full paper (11:47 PM PST).** [known: Bushnell–Reiner,
  Math. Z. 173 (1980), printed p.156 Theorem 2 and printed p.158 Remarks 1
  and 3, read as page images from the GDZ scan in `night-sources-20261110/`]
  Theorem 2 for group rings RG reads ζ_Λ(s)/ζ_Λ(1−s) = (Λ′:Λ)^(1−2s)
  ζ_Λ′(s)/ζ_Λ′(1−s). Remark 1: "One can obviously generalize this argument to any
  order Λ which is self-dual under an appropriate trace form." That gives
  (Λ′:Λ)^(1−2s) ζ_Λ′(s)/ζ_Λ′(1−s) = ζ_Λ(s)/ζ^r_Λ(1−s), with ζ^r for right
  lattices; the authors "do not know whether ζ_Λ = ζ^r_Λ always". Remark 3:
  with φ = ζ_Λ/ζ_Λ′ this is φ(s) = (Λ′:Λ)^(1−2s) φ(1−s). They add that it forces
  the top term p^(n(1−2s)) and "is consistent with a 'Riemann hypothesis' for
  ζ_Λ(s)".
  [project, proved] For O_r, Λ′ = M_2(Z_p), (Λ′:Λ) = p^r and v′ = 1/(pv), so
  φ(s) = (Λ′:Λ)^(1−2s) φ(1−s) is exactly our palindrome
  Q(v) = p^r v^(2r) Q(1/(pv)). Our left and right zetas coincide (adjugate
  involution, Block 1 §9). The column-pair count therefore proves the
  Bushnell–Reiner-shaped functional equation for this order directly.
  [project, checked by hand] Remark 1 does not literally apply as stated:
  under trd, O_r^# = [[Z_p, p^(−r)Z_p],[Z_p, Z_p]] = ξO_r with
  ξ = [[0,p^(−r)],[1,0]] (the Atkin–Lehner-type normalizer). That is invertible
  (Gorenstein) but not c·O_r for a scalar c, so O_r is not self-dual under any
  scalar multiple of trd. Whether the Bushnell–Reiner argument extends to
  "dual = normalizing element times Λ" is plausible but was NOT checked.
  [known, same Remark 3] Bushnell–Reiner already flag an RH-type expectation
  for these φ. Our circle theorem fits that remark for this one local family.
  The idea is not new, and the result is not RH.
  Wording note for wrap-up: earlier session text says there is "no general FE
  theorem for non-hereditary quaternion Solomon zetas". That should now cite
  Bushnell–Reiner 1980 Remark 1 (an FE for orders self-dual under a trace form).
  That remark does not literally cover O_r, so the earlier statement is
  incomplete, not wrong.
- [known] Huang–Jiang (talk slides, UBC/UCI 7 Oct 2023, read in full): for
  commutative Gorenstein orders R, ζ_{R^d}(s) has a functional equation s ↦ d−s.
  Commutative only; not a source for our noncommutative case.
- [known] Denert, Math. Comp. 54 (1990) (abstract read): explicit genus zeta
  functions for local **hereditary** orders only; Eichler level p^r with r ≥ 2
  is not hereditary, so not covered.
- Kim–Jun 2012: see Block 1 (parity obstruction stands; not re-read).
- Not found / not read this block: Brzezinski's papers on Bass orders,
  Lakatos 2002 full text, arXiv:2310.15545 (semilocal Solomon products; title
  and abstract excerpt only).

[project] Conclusion: the searches found no source that displays
Q_r(v) = Σ p^{min(d,e)} v^{d+e}, the invertible count Q_r^inv, or the two
unit-circle statements for Eichler level p^r. This is NOT evidence of novelty:
the formula is elementary (tiled-order / pairs-of-lattices count), is
computable from Lynch's or Bushnell–Reiner's general theory, and could easily
be classical (e.g. in work on tiled/graduated orders or Bass orders). Keep
"original proof" meaning "proved here", never "new to mathematics".

## 6. Does anything overturn the live Zeta Zeros site?

[project] **No.** Everything checked agrees with what is live:
Q_4 = p²+2p+2 is confirmed, now by two further independent counts (column
pairs, and the full 4D HNF brute force of this block). The live caveats are
now *stronger* than necessary: "Q_4 not yet independently re-derived" can be
retired; the closed form and palindrome can move from conjecture to
[project, proved]; the circle theorem can be stated unconditionally for these
local numerators. Those are wording upgrades for the authorized wrap-up, not
corrections of errors. Do not swap the Solomon Q into places that show the
invertible factor. No site file was touched in this block.
