# Referee notes on the Eichler local-factor proofs

10 November 2026, America/Los_Angeles.

A line-by-line re-derivation of the four arguments scheduled to go on the site: the invertible unit-circle theorem, the invertible count a_e, the Solomon sign lemma for the closed-form family, and the coefficient Q_3=2(p+1).  Closed values were recomputed in sympy; every strict inequality was checked for all q>=sqrt(2), including the edge p=2; sign-change counts were confirmed; the volume sums and the passage from a_e to Q_r^{inv} were expanded independently.  Nothing here is a claim about the Riemann hypothesis.

Verdict in brief: all four statements stand.  No mathematical error was found.  Three presentation gaps in the original writeups were real enough, for a public note, that the missing algebra has been filled in.  Q_4 was not proved.

## 1. Invertible unit-circle theorem

Source: `order-eichler-unitcircle-proof.md`.  Label: [project, proved], elementary.

The reduction of P_r to S(x)=2 T_r(x)+c U_{r-2}(x), with x=cos(theta/2) running once through (-1,1) as theta runs through (0,2 pi), is correct.  The two displayed sine forms of g(theta) are identical, by the prosthaphaeresis identity -2 cos(r theta/2) sin(theta/2)+sin((r+1)theta/2)=sin((r-1)theta/2), checked symbolically.  At the extrema x_k=cos(pi k/r) of T_r, the value U_{r-2}(x_k)=(-1)^{k+1} is the elementary identity sin(k pi - alpha)=(-1)^{k+1} sin(alpha) with alpha=pi k/r, not a hidden Chebyshev fact; it holds for every r>=2 and k=1,...,r-1, including r=2 where U_0=1.  The resulting signs at the r+1 ordered points -1<x_{r-1}<...<x_1<1 are (-1)^r, (-1)^{r-1}, ..., -1, +1, hence exactly r adjacent opposite pairs, hence exactly r roots of a degree-r polynomial in (-1,1).  Those roots are automatically simple.  The side remark P_r'(-1)=r-c(r-1)/2>0, for odd r, is correct but unnecessary once simplicity is read off the degree.  The factor 2-c is at least 1, and equals 3/2 at p=2, so it never vanishes on the family.  w=1 is never a root.

No mathematical gap.  No patch was required.  The test-point count is exactly r sign changes, confirmed for r=1..30 by enumerating the predicted signs.

## 2. Invertible count a_e, and the polynomial Q_r^{inv}

Source: `order-eichler-qinv-formula.md`.  Label: [project, proved], elementary.

The orbit-volume reduction a_e=p^{2e}/(1-1/p)^2 * V_e is correct: the action of O^x is free on generators of reduced norm p^e, vol(O^x)=(1-1/p)^2, and right multiplication by alpha has Jacobian p^{-2e}.  The low slices 0<=k=v_p(a)<r are correctly evaluated, each contributing (1-1/p)^2 p^{-e}, and give a_e=(e+1)p^e for e<r.

The original writeup stated the high-slice volumes Y(s,n) and the value of V_high without a derivation.  That is a genuine presentation gap for a site-bound proof (Gap A).  The formulae themselves are correct, including at p=2 and including the boundary s=n (the two closed forms for Y disagree at s=n, and the s<=n line is the right one).  An independent computation, organising on i=v_p(b) rather than on v_p(b c) so as to avoid a unit-cancellation peculiar to p=2, recovers

    Y(s,n)=(s(p-1)+p)(p-1)p^{-n-2}     if s<=n,
    Y(s,n)=(n+1)(p-1)^2 p^{-n-2}       if s>n.

The two geometric sums for V_high then evaluate, under p>1, to (1-p^{-2})(p^{-e}-p^{r-1-2e}), and the high contribution is (p+1)/(p-1)(p^e-p^{r-1}).  This was checked symbolically on the p>1 branch and numerically on a grid of p,r,e.  Adding the r p^e low ideals gives the displayed formula for a_e, e>=r.  The passage from a_e to the coefficients b_e of (1-u)(1-p u)A(u) was checked symbolically for every r=1..7 and every e through r+5: b_0=1, b_1=p-1 (or p when r=1), b_e=(p-1)p^{e-1} for 2<=e<=r-1, b_r=p^r, and b_e=0 for e>r.  The specialisations Q(p^{-2})=1+1/p and the r=1 hereditary count a_e=(2p^{e+1}-p-1)/(p-1) both hold.

Fix applied: the note now includes the split on v_p(b) and the evaluation of the two sums.  No change to the statement or the label.

## 3. Solomon sign lemma, for the closed-form family

Source: `order-solomon-unitcircle-proof.md`.  Label: [project, proved], conditional on the closed form; elementary.

The reduction of the palindromic Q to the cosine polynomial C, the kernel formula C=sum_t alpha_t D_{r-t}, and the closed form of D_N were re-derived and checked (the D_N identity is elementary, though a raw sympy subtraction does not cancel until expand_trig is applied).  The original wording "summing by parts" for the double-sum rearrangement was inaccurate (Gap B, wording only) and has been corrected.

At the test angles, the companion sum Psi=Re[(q e^{i theta}-1)G(e^{i theta}/q)] takes one of four closed values, according to the parities of r and of m, away from theta=pi.  The original writeup stated these values as "a direct expansion" or "simplifies to a constant," without the algebra (Gap C).  They have now been expanded in the note, and were independently recomputed in sympy as identities on the unit circle (the difference from the predicted constant, after substituting zeta=e^{i t} and conjugating, simplifies to 0 in all four cases):

    r odd,  m even, theta != pi:   Psi=0,
    r odd,  m odd:                 Psi=-2,
    r even, m even, theta != pi:   Psi=(q-1)^2/(2q),
    r even, m odd,  theta != pi:   Psi=-(q+1)^2/(2q).

The half-angle reductions remain valid on (pi, 2 pi): for the even-r even-m case, cos(theta/2) is negative but cancels; for the even-r odd-m case, sin(theta/2)>0 throughout (0,2 pi).  The middle test point theta=pi is always the unique test angle at which 1-e^{2 i theta}=0, and is always of the parity excluded from the four lines above, so there is no overlap with the special evaluation of C(pi).

The sign condition sigma(q^2-1)Q_r - 2 q^{r+1} Psi>0, which the original note used without a derivation from the wanted sign (-1)^{m-1}=-sigma (Gap D, one line), is now displayed.  The four resulting inequalities reduce to q^{r+1}+1>0, 2(q^{r+1}-1)>0, or 2(q^{r+1}+1)>0, all strict for every q>=sqrt(2), including p=2.  The evaluation of C(pi) from D_N(pi) and G(-1/q) matches the definition of C, and the positivity arguments (a sum of two positive terms when r is odd; the difference (r+1)q^r(q^2-1)+2(q^{r+1}+1) when r is even) hold at p=2.

The test-point count is exactly 2r sign changes: 2r+1 interior points with signs (-1)^{m-1}, the first and last both +, and no forced root in the two end caps adjacent to C(0)=C(2 pi)>0.  Confirmed for r=1..39.  As a cosine polynomial of order r, C has at most 2r zeros in a period, and C is not identically zero, so these 2r zeros are all of them and are simple.

No mathematical gap.  Fix applied: the four expansions of Psi, the sign-condition line, and the wording of the double-sum rearrangement are now in the note.  The label is unchanged.  The statement remains conditional on the closed-form coefficients.

## 4. The coefficient Q_3=2(p+1), for r>=3

Source: `order-solomon-closed-form.md`.  Label: [project, proved] for Q_1, Q_2, and Q_3; the rest a project conjecture.

The Euler identity A_3=Q_3+2(p+1) is correct.  The classification of the 2p+3 modules of index p^2 (one K, 2p of type L, two of type N) follows from the already-proved Q_2 charts, once one records which stable hyperplanes contain the image of pO: that image is the alpha-axis in M_e/p M_e, so N_e={alpha=0} is the unique child of M_e not containing pO.  The child counts p+2 for N_e and N_h (r>=3), 2p+2 for K (r>=2), and 2 for each type-L module (r>=2) are correct; the original writeup summarised the stable-functional computations for K and L rather than writing the eigenvector condition in full (Gap E, presentation).  Those computations were re-derived: on K/pK the off-diagonal generators act as zero for r>=2 and the stable functionals split cleanly into two disjoint P^1-families; on L_{e,lambda} the only solutions of the eigenvector equation are (1,0,0,0) and (0,0,0,1).  The listing of children in coordinates is correct, and the original disjointness clause ("an apparent overlap would drop the index to p^4") was true but thin (Gap F).  A short comparison of the defining valuation conditions has been added.  The resulting count is 4(p+1), so Q_3=2(p+1).  Independently, the index-p enumerator gives A_3=4(p+1) at the five pairs (p,r)=(2,3),(2,4),(2,5),(3,3),(3,4).

No mathematical gap.  Fix applied: the disjointness comparison is now in the note.  The label is unchanged.

## 5. Q_4, unfinished

The same child-counting produces stable child-numbers for the five types of index-p^3 modules when r>=4: the 2p modules of type P or Q, the two of type F_min or G_min, the 2p-2 remaining children of K, and the two of type A3, each have p+2 children; the two modules K_{2,1} and K_{1,2} have 2p+2 children.  The incidences total 4p^2+14p+8.  The enumerator gives A_4=3p^2+6p+5 at (2,4),(2,5),(3,4), matching the closed-form coefficient Q_4=p^2+2p+2.  The overlap analysis among the index-p^4 children was not completed (twelve parent-type classes already appear at p=2 and at p=3), so Q_4 is not proved and remains a project conjecture.

## 6. Labels, literature, and what was not claimed

The invertible arguments are correctly labelled [project, proved], elementary, and not new deep mathematics.  The Solomon unit-circle theorem is correctly labelled as conditional on the closed form.  Prior-art remarks (Lakatos covering only r<=p+1; no explicit Q_r^{inv} in the checked sections of Voight, Denert, or Brzezinski; no off-the-shelf local FE for a non-hereditary Eichler Solomon zeta) were not re-litigated here, and nothing in the re-derivation requires them to be strengthened or withdrawn.  No note claims progress on the Riemann hypothesis.

## Actions

The missing algebra for Y(s,n) and V_high was written into `order-eichler-qinv-formula.md`.  The four expansions of Psi, the sign-condition line, and the double-sum wording were written into `order-solomon-unitcircle-proof.md`.  The Q_3 disjointness comparison, and a one-paragraph record of the unfinished Q_4 count, were written into `order-solomon-closed-form.md`.  The status table and the plain-language note were updated to record this referee pass.  No statement was withdrawn or downgraded.
