# A shorter circle proof from the column-pair square

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

[project, proved] This is an independent, shorter proof of the previously
proved polynomial unit-circle statement, now combined with the exact
all-level local factor in `order-solomon-column-pair-proof.md`. It does
not invoke the earlier parity-dependent kernel evaluations. It is
not peer-reviewed; no novelty or RH-progress claim is made.

## Statement and notation

[project, proved] For p>1 real and r>=1 integer, set

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

Every root of Q_r is simple and has modulus p^(-1/2).
For prime p, the companion lattice count identifies Q_r with the
Solomon numerator of O_r=[[Z_p,Z_p],[p^r Z_p,Z_p]]. This theorem is only
about that local numerator, not the inherited global zeros.

## Convert the square into a cosine polynomial

[project, proved] Put rho=1/sqrt(p), so 0<rho<1, and

    P_r(z) = Q_r(rho z)
           = sum_{d,e=0}^r rho^{|d-e|} z^{d+e}.

Let U_n and T_n denote the Chebyshev polynomials determined by

    U_0=1, U_1=2x, U_(n+1)=2x U_n-U_(n-1),
    T_0=1, T_1=x,  T_(n+1)=2x T_n-T_(n-1).

For z!=0 set x=(z+z^(-1))/2. The elementary identity

    U_n(x) = sum_{j=0}^n z^{2j-n}

follows from this recurrence. Group the square by h=|d-e| and multiply
by z^(-r). The diagonal contributes U_r, and each h>=1 contributes
2rho^h U_(r-h). Thus

    z^(-r) P_r(z) = C_r(x),
    C_r(x) = U_r(x) + 2 sum_{h=1}^r rho^h U_(r-h)(x).            (1)

In particular C_r is a real polynomial of degree r with leading
coefficient 2^r.

## One telescoping identity

[project, proved] Write S_r=sum_{h=0}^r rho^h U_(r-h), so C_r=2S_r-U_r.
Multiplication by D(x)=1-2rho x+rho² and the U recurrence give

    D S_r = U_r-rho U_(r+1)+rho^(r+2).

For completeness, multiply the finite sum term by term. The terms of
rho-degree 2,...,r cancel by U_n-2x U_(n-1)+U_(n-2)=0. The two
boundary terms at the upper end use U_1=2x U_0. The surviving terms
are U_r-rho U_(r+1)+rho^(r+2). For r=1 the same identity follows
directly from S_1=2x+rho.

Use U_(r+1)=x U_r+T_(r+1), which also follows from the defining
recurrences. Then

    D(x) C_r(x)
      = (1-rho²) U_r(x) - 2rho T_(r+1)(x) + 2rho^(r+2).         (2)

For -1<=x<=1, D(x)>0, since D(x)>=(1-rho)².

## Exactly r roots in (-1,1)

[project, proved] Let x_m=cos(m*pi/(r+1)) for m=1,...,r.
The elementary trigonometric forms of the same recurrences give

    U_r(x_m)=0,
    T_(r+1)(x_m)=(-1)^m.

So equation (2) becomes

    D(x_m) C_r(x_m) = 2rho [rho^(r+1)-(-1)^m].                 (3)

Since 0<rho^(r+1)<1, this is strictly positive for odd m and strictly
negative for even m. At the remaining endpoint x=-1, the recurrence
gives U_r(-1)=(-1)^r(r+1) and T_(r+1)(-1)=(-1)^(r+1). Hence

    (-1)^r D(-1) C_r(-1)
      = (1-rho²)(r+1)+2rho+2(-1)^r rho^(r+2) > 0.              (4)

For even r every summand is positive. For odd r, the last two terms
combine to 2rho[1-rho^(r+1)]>0. This includes r=1.

The points x_1>...>x_r>-1 therefore have alternating signs, and the
sign at -1 is opposite to that at x_r. There is one root in each of
the r intervals

    (x_(m+1),x_m),  m=1,...,r,    with x_(r+1)=-1.

Since C_r has degree r, these r roots exhaust it and are simple.
All lie strictly between -1 and 1.

## Return to the unit circle

[project, proved] If x_j is one of those roots, the equation

    (z+z^(-1))/2=x_j

has the two distinct roots z=x_j +/- i sqrt(1-x_j²), both on |z|=1.
Different x_j give different pairs. Equation (1) therefore supplies
2r distinct roots of P_r, which has degree 2r. They exhaust its roots.
The substitution v=rho z proves the statement.

[project, proved] There is also a useful exact phase bracket. In the
upper semicircle there is exactly one root e^(i theta_m) for each
m=1,...,r, with

    m*pi/(r+1) < theta_m < (m+1)*pi/(r+1).                     (5)

Their lower-semicircle conjugates provide the other r roots. No root
is real. For v=p^(-s), these give corresponding equally replicated
vertical phase bands on Re(s)=1/2. This is a local-factor conclusion;
it is not a claim of universal spacing or noncoincidence for a global
zeta function.

## Independent exact checks

[project] `night_solomon_chebyshev_check.py` uses Python Fraction arithmetic:

- 480 exact polynomial identities (2), r=1..80, at rho in
  {1/10,1/2,2/3,7/10,9/10,99/100};
- strict endpoint signs at x=-1 and x=1 for all those cases;
- 60 exact Laurent-coefficient comparisons between (1) and the
  original rho^{|d-e|} square, r=1..20 at rho=1/2,2/3,7/10.

All passed. These checks support, rather than replace, the recurrence
proof and the strict inequalities. They contain no approximate root
locations. Record: `night-solomon-chebyshev-20261110.json`.

[project] Next block should independently referee both the counting proof
and this shorter circle argument before changing live claim labels.

## Exact finite certificates independent of the angular sign proof

[project] An additional check uses Sturm's theorem with exact integer and
rational arithmetic, not the alternating-angle lemma. Write

    y = sqrt(p) x,
    B_r(y) = p^(r/2) C_r(y/sqrt(p)).

[project, proved] This is an integer polynomial. One construction is
V_0=1, V_1=2y, V_(n+1)=2y V_n-p V_(n-1), followed by

    B_r = V_r + 2(V_0+...+V_(r-1)).

Equivalently it is obtained from the actual Q coefficients using
W_0=2, W_1=2y, W_(k+1)=2y W_k-p W_(k-1), by

    B_r = Q_r + sum_{k=1}^r Q_(r-k) W_k.

[project, proved] The normalization has the exact Laurent identity

    Q(v) = v^r B_r((p v+v^(-1))/2).

A root y with -sqrt(p)<y<sqrt(p) produces the two distinct roots

    v = (y +/- i sqrt(p-y²))/p,

whose squared modulus is exactly 1/p. A degree-r B with r distinct roots
in that real interval therefore proves the local numerator's circle
statement in that finite case. This is stronger than enclosing a complex
root in a small ball that merely intersects the desired circle.

[project] `night_solomon_sturm_check.py --rmax 60` verified all 240 cases
p in {2,3,5,7}, r=1..60. For each case it:

1. constructed B independently by both displayed recurrences;
2. recovered every coefficient of Q by the exact Laurent identity;
3. computed a rational Sturm chain, with positive primitive rescaling;
4. chose a rational bound b<sqrt(p) using an 80-bit integer square root;
5. found exactly r simple real roots in (-b,b).

[project] At p=2,3 and r=5,6, coefficients are taken from the new embedded
lattice enumeration, and each of the r real roots has a separate dyadic
isolating interval of width at most 2^(-50). At higher levels the input
is the proved square formula, not a second lattice count. All 240 cases
passed. Data and Sturm-chain digests:
`night-solomon-sturm-20261110.json`. The script includes a tested
termination correction for roots exactly at a dyadic midpoint; that
implementation detail does not change any mathematical formula.

[project] A separate floating-point bisection checks 36 polynomials
(p=2,3,5,7; r=1,2,4,5,6,12,24,48,96), 792 upper-semicircle angles,
by direct complex evaluation of Q. Maximum residual divided by the sum
of absolute term magnitudes was 3.91e-14. This is NOT an interval
certificate. Angle data for later illustrations:
`night-solomon-phase-20261110.json`; script
`night_solomon_phase_check.py`.

[project] The final exact checker additionally treats rho as a formal
variable: 81 bivariate identities (r=0..80) pass coefficientwise, besides
the 480 rational-parameter identities and 60 Laurent comparisons. See
`night-solomon-chebyshev-20261110.json`. Sharper half-band bounds and an
exact phase equation are derived separately in
`order-solomon-level-phase-consequences.md`; they are not needed for
this shorter root-location proof.

## 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.
