# Two column lattices: an independent Solomon count for every level

10 November 2026, night session, America/Los_Angeles (PST).

## Plain-language result

[project, proved] The independent re-check agrees with today's coefficient
**Q_4 = p² + 2p + 2**. A simpler way to organize the count also gives the
whole polynomial, not just its first four coefficients. In this method a
right ideal is two ordinary two-dimensional lattices, one for each matrix
column. The order tells us how those lattices must fit together.

[project, proved] For the split local Eichler order

    O_r = { [[a,b],[p^r c,d]] : a,b,c,d in Z_p },   r >= 0,

let A_k count **all full-rank right ideals I contained in O_r** with
additive index [O_r:I] = p^k. We count actual embedded submodules, not ideal
classes, not just invertible ideals, and not arbitrary O_r-modules. Put
v = p^(-s). Then

    A(v) = sum_{k>=0} A_k v^k
         = Q_r(v) / ((1-v²)(1-pv²)),

    Q_r(v) = sum_{d=0}^r sum_{e=0}^r p^{min(d,e)} v^{d+e}.        (1)

[project, proved] This equals the previously conjectured closed form. It
proves Q_5 = 2(p²+p+1) for r>=5, Q_6 = p³+2p²+2p+2 for r>=6, and

    [v^(2r-j)] Q_r(v) = p^(r-j) [v^j] Q_r(v),   0 <= j <= r.     (2)

[project] This is an elementary project derivation, double-computed below.
It has not yet received an independent *human or separate-agent* referee
pass. No claim of priority or novelty in the mathematical literature is
made. It concerns this explicitly specified local order, not all
non-hereditary orders, not octonion orders, and not RH.

### Notation: coefficient versus whole polynomial

[project] Q_r(v) denotes the whole degree-2r polynomial at level p^r.
The agenda's Q_j means its coefficient [v^j]Q_r(v), which is independent
of r in the low range r>=j. Thus the requested Q_4=p^2+2p+2 is a
coefficient, NOT the whole polynomial Q_{r=4}(v). The same distinction
applies to the requested Q_5 and Q_6.

## 1. Split the right ideal using idempotents

[project, proved] Write a matrix in O_r using coordinates (a,b,c,d), so that
its first column is encoded by (a,c) and its second by (b,d). Right
multiplication by E_11 and E_22 projects onto these two column subspaces.
Both projections belong to O_r, and E_11+E_22=1. Thus every right ideal I
splits uniquely as

    I = U direct-sum V,

where U and V are full sublattices of Z_p², in the (a,c) and (b,d)
coordinates respectively. More explicitly, every combination of a column
from U and a column from V occurs in I: add their two projected matrices.
There is no remaining condition coupling the columns of an individual
element.

[project, proved] The other two basis elements of the order are
f=E_12 and g=p^r E_21. Direct multiplication gives

    [[a,b],[p^r c,d]] f = [[0,a],[0,p^r c]],
    [[a,b],[p^r c,d]] g = [[p^r b,0],[p^r d,0]].

In encoded column coordinates the resulting maps are

    F : (a,c) -> (a,p^r c),       F = diag(1,p^r),
    G : (b,d) -> (p^r b,d),       G = diag(p^r,1).

Consequently the necessary and sufficient conditions on U,V are

    F U subset V,       G V subset U.                            (3)

Sufficiency holds because E_11,f,g,E_22 are a Z_p-basis of O_r. The maps
satisfy GF=FG=p^r times the identity, a useful check on the coordinate
normalization.

## 2. Unique two-dimensional Hermite coordinates

[project, proved] Every full sublattice U of Z_p² has a unique basis

    (p^a,0), (t,p^b),

with a,b>=0 and t in a fixed complete residue system modulo p^a. Use
integer representatives 0<=t<p^a. Likewise write V as

    (p^A,0), (T,p^B),

where A,B>=0 and 0<=T<p^A. This follows directly by projecting a lattice
onto its second coordinate: its image is p^b Z_p and its kernel is
p^a Z_p in the first coordinate. Choose a lift of p^b and reduce its first
coordinate modulo that kernel. This also proves uniqueness, without
assuming a lattice-class parametrization.

[project, proved] The additive index of I is

    [O_r:I] = [Z_p²:U][Z_p²:V] = p^(a+b+A+B).                   (4)

A vector (x,y) belongs to V precisely when

    y is divisible by p^B,
    x - T (y/p^B) is divisible by p^A.

The analogous test for U exchanges (A,B,T) with (a,b,t).

## 3. Four membership tests

[project, proved] Apply (3) to each of the two basis vectors. The tests for
the first vectors are

    A <= a,         a <= A+r.

The second-coordinate tests for the second vectors are

    B <= b+r,       b <= B.

Define

    d = a-A,        e = B-b.

The diagonal tests say exactly

    0 <= d <= r,    0 <= e <= r,

while A,b can be any nonnegative integers. The remaining tests are

    t - p^(r-e) T == 0 modulo p^A,                             (5)
    p^r T - p^e t == 0 modulo p^(A+d).                        (6)

There are no omitted conditions: these are the first-coordinate tests
for F(t,p^b) and G(T,p^B), respectively.

## 4. Count the residue parameters, including the carry

[project, proved] Fix A,b,d,e as above and fix any of the p^A possible
values of T. By (5), the possible residues t modulo p^(A+d) form the
coset

    t = p^(r-e) T + p^A z modulo p^(A+d),

where z runs through Z/p^d Z. As z varies, this is a bijection with the
allowed canonical representatives 0<=t<p^(A+d). The displayed expression
need not itself be the canonical integer representative; reducing it is
essential and does not change the count.

Substitute in (6). Since p^e p^(r-e)T = p^r T, the condition becomes

    p^(A+e) z == 0 modulo p^(A+d).

If e>=d, all p^d residues z work. If e<d, z must be divisible by p^(d-e),
so exactly p^e residues work. In either case the number of pairs (t,T) is

    p^A p^{min(d,e)}.                                         (7)

This argument includes A=0, d=0, e=0, r=0, and p=2. It uses no division
by 2, no missing nonzero-residue restriction, and no coprimality assumption
on t or T.

## 5. Sum the count

[project, proved] Equations (4) and (7), with a=A+d and B=b+e, give

    A(v) = sum_{A,b>=0} sum_{d,e=0}^r
               p^(A+min(d,e)) v^(2A+2b+d+e)
         = (sum_{A>=0}(pv²)^A)(sum_{b>=0}v^(2b))
               * sum_{d,e=0}^r p^{min(d,e)} v^(d+e).

This is a valid identity of formal power series: every coefficient has
only finitely many contributions. It proves (1), the denominator,
polynomial termination, and the exact degree 2r all at once. In
particular Q_0=1 and the leading coefficient is p^r.

[project, proved] For 0<=j<=r, all pairs d+e=j are allowed. Therefore

    Q_j = sum_{d=0}^j p^{min(d,j-d)}.

Splitting at the midpoint yields

    Q_(2h)   = p^h + 2 sum_{m=0}^{h-1} p^m,
    Q_(2h+1) = 2 sum_{m=0}^h p^m.

These equal 1+sum_{k=1}^j p^{floor(k/2)}. At j=4,5,6 this gives

    Q_4 = p²+2p+2                 (r>=4),
    Q_5 = 2p²+2p+2                (r>=5),
    Q_6 = p³+2p²+2p+2             (r>=6).

[project, proved] For the upper half, use the actual bijection of the
finite square

    (d,e) -> (r-d,r-e).

If d+e=j, its image has sum 2r-j and

    min(r-d,r-e) = r-max(d,e) = r-j+min(d,e).

Thus its weight is p^(r-j) times the old weight. This proves (2), or

    p^r v^(2r) Q_r(1/(pv)) = Q_r(v).

The raw counts A_k do NOT obey a palindrome. The bijection is on the
finite polynomial's square of excess exponents, not on all embedded
right ideals of complementary index. This avoids the failed trace-dual
argument recorded in today's earlier notes.

## 6. Direct enumeration and independent recursion

[project] `night_solomon_column_pairs.py --mode pairs` scans the rank-two
Hermite pairs and checks (3) by integer vector membership, without using
(7) or the conjectured Q to create any lattices. It extracts Q only after
counting. Nineteen (p,r,kmax) cases passed, including:

- p=2, r=0..7, through k=2r+2;
- p=3, r=0..5, through k=2r+2, and r=6 through k=12;
- p=5 at (r,kmax)=(4,4),(5,5),(6,6);
- p=7 at (r,kmax)=(6,6).

[project] `--mode recursion` separately invokes the previous general
four-dimensional index-p submodule recursion. It does not invoke today's
hand-classified families. Ten cases passed: (p,r,kmax)=(2,4,10), (3,4,6),
(5,4,4), (7,4,4), (2,5,12), (2,6,14), (3,5,10), (3,6,6), (5,6,6),
(7,6,6). All common count prefixes agree exactly. Complete embedded
4-dimensional Hermite-key sets also agree through k=4 in every case,
and through k=8 in the p=2 cases; their SHA-256 digests are recorded.

[project] A separate unfiltered pair scan at five small parameter choices
verified that the diagonal membership prefilter excludes no genuine
ideals. No roots or floating-point tolerances enter either count.

| p | A_4 (r>=4) | Q_4 | A_5 (r>=5) | Q_5 | A_6 (r>=6) | Q_6 |
|---|---:|---:|---:|---:|---:|---:|
| 2 | 29 | 10 | 46 | 14 | 95 | 22 |
| 3 | 50 | 17 | 84 | 26 | 226 | 53 |
| 5 | 110 | 37 | 196 | 62 | 782 | 187 |
| 7 | 194 | 65 | 356 | 114 | 1890 | 457 |

[project] For example, at p=2, r=6, the full polynomial is

    Q = [1,2,4,6,10,14,22,28,40,48,64,64,64].

Both counts give two zero coefficients after it, at degrees 13 and 14.
At p=3,r=6 the directly scanned polynomial is

    Q = [1,2,5,8,17,26,53,78,153,216,405,486,729].

[project] A finite zero tail is only a check; it does not prove termination
or the all-r formula. Sections 1-5 provide that proof.

Records:

- `night-solomon-pairs-20261110.json`
- `night-solomon-recursion-20261110.json`
- `night-solomon-pairs-run-20261110.txt`
- `night-solomon-recursion-run-20261110.txt`

## 7. Consequence for the critical line, and publication caution

[project, proved] Combining (1) with the previously proved polynomial
sign lemma in `order-solomon-unitcircle-proof.md` removes that lemma's
closed-form hypothesis for these local Eichler orders. All 2r roots of
Q_r(v) are simple and have |v|=p^(-1/2). The proof by strict alternating
signs there supplies 2r distinct roots, hence also simplicity. Since
|p^(-s)|=p^(-Re(s)), each zero of this *local numerator* lifts to
Re(s)=1/2. For r=0, Q=1 and there are no such roots.

[project, proved] This is not a claim that a global quaternion-order zeta
has only half-line zeros. Its inherited ordinary-zeta families remain
separate. Nor is simplicity of the local polynomial a statement that
zeros from different primes or other factors cannot coincide.

[project] No site files were edited or deployed. The live Q_4 formula is
confirmed, not contradicted. Its "not independently re-derived" caveat
now has an independent method available. The site's open labels for Q_5,
Q_6, the full formula and the palindrome could be revised only after the
next block independently referees this proof. Preserve the old status as
the historical record until then. No RH-progress or joint-field claim.

## 8. Literature check (not a priority claim)

[known] Solomon zeta functions count finite-index submodules, and their
ratio to a maximal-overorder zeta is polynomial in q^(-s), by
Bushnell-Reiner's theorem. A newly opened source is Sean B. Lynch,
*Solomon zeta functions over arithmetic orders*, arXiv:2607.12302,
Introduction and Theorem 1.1 / Corollary 4.6:
https://arxiv.org/html/2607.12302 . The source supplies an effective general
finite-sum formula; its displayed applications concern group-ring orders,
not the specific Eichler minimum-weight square (1).

[project] This brief search did not establish whether (1) already appears
elsewhere. It must NOT be described as new to mathematics.

[known] Full text was opened for In-Suk Kim and Sung-Tae Jun,
*Zeta functions on a cetain orders in a quaternion algebra*, 19(3)
(2012), pp.297-304, DOI 10.7468/jksmeb.2012.19.3.297:
https://ocean.kisti.re.kr/downfile/crosscheck/ksmed/JAKO201228449925766.pdf .
Its p.302 definition of a left ideal, in its later quadratic-extension-order
setup, imposes local principality, I_p=O_p a_p with a_p invertible.
The earlier Eichler discussion writes weights n(M)^(-2s) using ideal norms,
not an explicit additive index for arbitrary noninvertible submodules.
For those submodules, the square of the reduced norm need not equal the
additive index (our index-p submodules already show why this distinction
matters). The paper's earlier and later conventions must therefore be
reconciled before using its formula as a comparison. This block has NOT
established an equivalence or a contradiction with that paper. Its
"functional equalities" are not presently a verified source for the
specific all-additive-index formula (1).
Journal issue metadata: https://accesson.kr/pam/v.19/3/2012 .

[project] Earlier DOI/KoreaScience fetches timed out or returned empty
replies; the public KISTI full-text URL above succeeded through WebFetch.
Do not interpret the earlier failures as absence of the article. No
priority conclusion follows from this scope check. Section 11 below
subsequently settles the same-variable parity obstruction; it does not
refute the paper or settle its intended ideal family.

## 9. Definition and left/right convention audit

[known] Bushnell and Reiner, *Zeta Functions of Arithmetic Orders and
Solomon's Conjectures*, Math. Z. 173 (1980), printed p.135, equation (1),
define the series by summing (L:N)^(-s) over **all** full left
Lambda-lattices N contained in L. Printed p.136, equations (3)-(4),
then separately decompose that series into genera. Thus the defining
series is not already restricted to the genus of the order. The article
was downloaded from the public GDZ article range and its first two
printed pages were read as images:

https://gdz.sub.uni-goettingen.de/download/pdf/PPN266833020_0173/LOG_0017.pdf

Archive: `night-sources-20261110/bushnell-reiner-1980.pdf`; definition-page
images: `br1980-page-02.png`, `br1980-page-03.png` in that directory.
The archive includes a repository cover sheet, so PDF page 2 is printed
p.135. The scan has no article text layer. No exact Eichler formula
from the rest of this article has been checked.

[project, proved] The right-ideal convention in this note gives the same
series as left ideals for this order. The adjugate involution

    J([[a,b],[p^r c,d]]) = [[d,-b],[-p^r c,a]]

preserves O_r, is additive and bijective, and reverses products:
J(xy)=J(y)J(x). In encoded coordinates it is (a,b,c,d)->(d,-b,-c,a),
a unimodular Z_p-linear map. It therefore sends embedded full-rank
right ideals bijectively to embedded full-rank left ideals, preserving
additive index. This reconciles the handedness without restricting to
principal ideals, ideal classes, or a single genus.

## 10. A positive sum of squares (additional algebraic audit)

[project, proved] Put S_n(v)=1+v+...+v^n. Another exact expression is

    Q_r(v)=S_r(v)^2
           +(p-1)*sum_{k=1}^r p^(k-1)*v^(2k)*S_{r-k}(v)^2.

Expand p^min(d,e)=1+(p-1)*sum_{k=1}^{min(d,e)}p^(k-1) and interchange
the finite sums. This proves the identity without the root-location
argument. In particular Q_r(v)>0 for every real v when p>1: at v=0
the first term is 1, and at v!=0 the k=r term is strictly positive
(for r=0, Q_0=1). Thus no numerator root is real. This also independently
checks noncancellation at the four real denominator roots.

[project] `night_solomon_auxiliary_check.py` verifies this coefficientwise
and verifies the independent three-step level recurrence from
`order-solomon-level-stability.md`, for p=2,3,5,7 and r=0..80. These are
algebraic consistency checks of the proved square, not additional
independent lattice enumerations.

## 11. A concrete obstruction to identifying norm-weighted conventions

[known] Kim--Jun 2012, printed pp.298-299 (URL in section 8), define
the earlier Eichler series using n(M)^(-2s), with local ideal norms
powers of p. Their displayed local formulas consequently use only even
powers of v=p^(-s). This statement concerns the variable convention, not
the validity of their formulas or their precise choice of ideal family.

[project, proved] For r>=1, take the embedded right ideal

    I_a={ [[a,b],[p^r c,d]] in O_r : a divisible by p }.

It has additive index p. It is a right ideal: in a product xy, the
upper-left entry is a_x*a_y+p^r*b_x*c_y, divisible by p when x is in I_a.
Every determinant in I_a is divisible by p, while diag(p,1) belongs to
I_a and has determinant p. Thus its reduced-norm ideal is p*Z_p.
Its additive-index weight is p^(-s), but its norm-squared weight would
be p^(-2s). The analogous ideal with d divisible by p gives the second
index-p ideal. Both are noninvertible, since a locally principal ideal
has additive index equal to the square of its reduced norm.

[project, proved] In particular our all-additive-index local series has
coefficient A_1=2, whereas a series whose weights are integral ideal
norms raised to -2s has no v term at all. Under the same variable s,
these series cannot be identified. This resolves the specific parity
obstruction without extending the paper's later local-principality
definition backwards into its earlier section. Which precise norm-ideal
family is intended there remains a separate bibliographic question.
No accusation of a false paper, and no priority conclusion, follows.

## 12. Why index-p recursion is complete here (and not for every order)

[project, proved] The independent index-p recursion used in section 6
is exhaustive for r>=1, but that fact should not be silently generalized
to arbitrary orders. In the abstract four-dimensional algebra O_r/pO_r,
the images of f=E_12 and g=p^r E_21 span a square-zero ideal: fg and
gf are p^r times diagonal idempotents, hence zero modulo p. The quotient
by that ideal is F_p x F_p. Its simple modules all have p elements.
Every simple constituent of any finite p-power order O_r-module is
annihilated by p and by the square-zero radical, so it has p elements
as well. Therefore every finite quotient O_r/I admits an O_r-module
composition series with successive additive index p. Lifting this
series proves that repeatedly enumerating all stable index-p children
reaches every embedded finite-index right ideal.

[project, proved] At r=0 this argument fails: O_0/pO_0=M_2(F_p), whose
simple right modules have p^2 elements. A recursion restricted to index p
would have no first children and would miss nontrivial index-p^2 ideals.
Accordingly all recursion runs in this session use r>=1; r=0 is checked
by direct column-pair enumeration only. The complete-HNF pair comparisons
are independent finite checks, not a replacement for this completeness
argument.

### Further finite checks completed in this block

[project] The four-dimensional recursion was extended independently to
(p,r,kmax)=(3,6,14) and (5,6,12). At p=3 it matches the complete
column-pair polynomial and adds two exact zero tail coefficients;
A_14=35647. At p=5 it gives the complete polynomial

    [1,2,7,12,37,62,187,310,925,1500,4375,6250,15625],

with A_12=175492 and the full palindrome. Records:
`night-solomon-extra-p3-recursion-20261110.json` and
`night-solomon-extra-p5-recursion-20261110.json`.

[project] A seeded structural audit constructs 1200 canonical ideal
parameters, including 482 noncanonical raw-t expressions requiring
reduction. All 19200 actual 2x2 matrix-action membership tests pass;
91 small residue-coset sets agree exactly with direct congruence scans.
The actual matrix multiplications also cross-check the action matrices
used by the recursion. At r=0, p=2,3,5, the negative-scope test correctly
finds zero index-p children but p+1 index-p^2 ideals. Record:
`night-solomon-structural-audit-20261110.json`.

## Referee status — 11:45 PM PST, November 10, 2026 (night Block 2)

[project] Refereed line by line by a separate agent in night Block 2:
PASS, no mathematical error. See `BLOCK2-REFEREE-20261110.md`. Earlier
sentences in this note saying it "has not yet received" a referee pass are
kept as the Block 1 record. Still not peer-reviewed; no novelty or RH claim.
