# The Solomon closed-form family, on the unit circle

10 November 2026, America/Los_Angeles.

[project, proved], conditional on the closed form; elementary.  Every root of the closed-form Solomon polynomial lies on |v|=p^{-1/2}.  The argument is a sign lemma for a cosine polynomial, proved for every r>=1 and every prime p by evaluating a Dirichlet-type kernel at the angles m pi/(r+1) and checking a single remaining value at theta=pi.  This is not an RH claim, and it is not new deep mathematics.  It is still conditional on the closed-form coefficients themselves: Q_1, Q_2, and Q_3 are proved in the companion note, but the remaining coefficients and the palindrome are a project conjecture.

## The family

For an integer r>=1 and a prime p, let Q(v) be the palindromic polynomial of degree 2r given by

    Q_j = sum_{m=0}^j p^{floor(m/2)}     for 0 <= j <= r,
    Q_{2r-j} = p^{r-j} Q_j               for 0 <= j <= r.

Equivalently, writing q=sqrt(p), one has the elementary evaluations Q_{2s}=(p^{s+1}+p^s-2)/(p-1) and Q_{2s+1}=2(p^{s+1}-1)/(p-1).  The functional equation p^r v^{2r} Q(1/(p v))=Q(v) is built into the palindrome.

## Substitution

Set v = z / sqrt(p) and F(z) = p^{r/2} Q(z/sqrt(p)).  The palindrome of Q makes F self-reciprocal with real coefficients: z^{2r} F(1/z)=F(z).  On the unit circle z=e^{i theta}, the function

    C(theta) = e^{-i r theta} F(e^{i theta})
             = Q_r + sum_{k=1}^r 2 Q_{r-k} p^{k/2} cos(k theta)

is real, and the roots of Q on |v|=p^{-1/2} are exactly the zeros of C on (0, 2 pi).  As a cosine polynomial of order r, C has at most 2r zeros in a period unless it vanishes identically.  It therefore suffices to exhibit 2r genuine sign changes.

## The sign lemma

Lemma.  For every prime p and every r>=1, and for each m=1,2,...,2r+1,

    C( m pi / (r+1) )  has the sign of  (-1)^{m-1},

and in particular is nonzero.  Also C(0)=F(1)>0, since all coefficients of F are positive.

If the lemma holds, the values of C at 0, at the 2r+1 interior points, and at 2 pi, have signs

    +,  +, -, +, -,  ...,  -, +,  +

(the first and last interior signs are both +, and the middle value, at theta=pi, has sign (-1)^r).  The intermediate-value theorem therefore supplies a root in each of the 2r intervals between consecutive interior test points, and none is forced in (0, pi/(r+1)) or in ( (2r+1)pi/(r+1), 2 pi ).  These 2r roots are simple, and they are all the roots of C.  Hence every root of the closed-form Q lies on |v|=p^{-1/2}.

The remainder of this note proves the lemma.

## A kernel formula

Write alpha_t = p^{floor(t/2)} and Q_j = sum_{t=0}^j alpha_t for j<=r.  Rearranging the double sum (Q_{r-k} = sum_{t=0}^{r-k} alpha_t) gives

    C(theta) = sum_{t=0}^r alpha_t D_{r-t}(theta),

where D_N(theta) = 1 + 2 sum_{k=1}^N q^k cos(k theta).  The standard summation formula

    sum_{k=0}^N q^k cos(k theta) = [1 - q cos(theta) - q^{N+1} cos((N+1)theta) + q^{N+2} cos(N theta)] / Delta(theta),

with Delta(theta) = 1 - 2 q cos(theta) + q^2, rearranges to

    D_N(theta) = [1-q^2 + 2 q^{N+1} (q cos(N theta) - cos((N+1)theta))] / Delta(theta).

The denominator is (q - cos(theta))^2 + sin^2(theta) >= (q-1)^2 > 0, so it never vanishes and never changes sign.  Consequently

    C(theta) = (1/Delta(theta)) * [ (1-q^2) Q_r + 2 sum_{t=0}^r alpha_t q^{r-t+1} (q cos((r-t)theta) - cos((r-t+1)theta)) ].

## Evaluation at the test angles

Let theta_m = m pi/(r+1) and sigma = (-1)^m, so that cos((r+1) theta_m)=sigma.  The elementary reductions

    cos((r-t) theta_m) = sigma cos((t+1) theta_m),
    cos((r-t+1) theta_m) = sigma cos(t theta_m)

then give

    C(theta_m) = (1/Delta) * [ (1-q^2) Q_r + 2 sigma q^{r+1} Psi(theta_m) ].

Delta is positive, so the sign of C(theta_m) is the sign of that numerator.  The lemma asks for the sign of (-1)^{m-1} = -sigma.  Equivalently, one needs the strict inequality

    sigma (q^2-1) Q_r - 2 q^{r+1} Psi(theta_m)  > 0.

where

    Psi(theta) = sum_{t=0}^r alpha_t q^{-t} (q cos((t+1)theta) - cos(t theta))
               = Re[ (q e^{i theta} - 1) G(e^{i theta}/q) ],

and G(w) = sum_{t=0}^r alpha_t w^t is the truncated generating function of the steps.  Splitting even and odd indices, one has the polynomial identities

    G(w) = (1+w) (1 - q^{r+1} w^{r+1}) / (1 - q^2 w^2)                 if r is odd,
    G(w) = [1+w - q^r w^{r+1} (1 + q^2 w)] / (1 - q^2 w^2)             if r is even.

At a test angle, q^{r+1} (e^{i theta}/q)^{r+1} = sigma, and 1 - q^2 (e^{i theta}/q)^2 = 1 - e^{2 i theta}, which vanishes only at theta=0 (not a test point) and at theta=pi (the middle test point m=r+1, treated separately below).

## The values of Psi, away from theta=pi

At a test angle other than theta=pi one has 1-q^2 w^2 = 1-e^{2 i theta} != 0, and Psi = Re[(q e^{i theta}-1) G(e^{i theta}/q)].  Write zeta = e^{i theta}.

If r is odd and m is even, then sigma=1 and G(w)=(1+w)(1-sigma)/(1-zeta^2)=0, so Psi=0.  The sign condition reduces to (q^2-1) Q_r > 0, which is immediate.

If r is odd and m is odd, then sigma=-1 and G(w)=(1+w)*2/(1-zeta^2).  Then

    (q zeta-1)(1+zeta/q) = [q(zeta^2-1) + (q^2-1) zeta]/q,

so Psi = Re[ 2/q * ( -q + (q^2-1) zeta/(1-zeta^2) ) ] = -2 + Re[ 2(q^2-1)/q * zeta/(1-zeta^2) ].  The remaining term is purely imaginary: zeta/(1-zeta^2) = i/(2 sin theta).  Hence Psi=-2, independently of m and of p (and the identity continues to hold for theta in (pi, 2 pi), where sin theta is negative).  The sign condition becomes 4 q^{r+1} > Q_r (q^2-1).  For odd r=2 ell+1 one has Q_r=2(q^{r+1}-1)/(q^2-1), so the inequality is 2 q^{r+1} > q^{r+1}-1, or equivalently q^{r+1}+1>0, which is strict.

If r is even and m is even, the even formula for G becomes [1+w-(1/q)(1+q^2 w)]/(1-zeta^2).  Multiplying by (q zeta-1) and cancelling a factor (1-zeta) (lawful because zeta != 1) yields Psi = Re[ (q-1)/q * (q zeta-1)/(1+zeta) ].  Writing zeta=e^{i theta} and dividing numerator and denominator by e^{i theta/2}, the real part of (q zeta-1)/(1+zeta) is (q-1)/2, including for theta in (pi, 2 pi) where cos(theta/2) is negative but cancels.  Thus Psi=(q-1)^2/(2q).  The sign condition becomes Q_r (q^2-1) > q^r (q-1)^2.  For even r=2 ell one has Q_r=(q^{r+2}+q^r-2)/(q^2-1), and the difference of the two sides is 2(q^{r+1}-1), which is strict for q>1.

If r is even and m is odd, the same reduction, now with sigma=-1, yields Psi = Re[ (q+1)/q * (q zeta-1)/(1-zeta) ].  The real part of (q zeta-1)/(1-zeta) is -(q+1)/2 (here theta/2 lies in (0,pi), so sin(theta/2)>0 throughout the open period).  Thus Psi=-(q+1)^2/(2q).  The sign condition becomes q^r (q+1)^2 > Q_r (q^2-1), whose difference is 2(q^{r+1}+1), again strict.

In all four cases the needed inequality is strict for every prime p (the bound is weakest at p=2, and remains strictly positive there), so C(theta_m) is nonzero and has the claimed sign, for every test point other than theta=pi.  For even r the middle test point is an odd m, and for odd r it is an even m; in both parities it is exactly the excluded angle theta=pi, so there is no overlap between this section and the next.

## The remaining point, theta=pi

At theta=pi the kernel specialises to the geometric sum D_N(pi)=[1-q-2(-q)^{N+1}]/(1+q).  Hence

    C(pi) = 1/(1+q) * [ (1-q) Q_r - 2 (-q)^{r+1} G(-1/q) ].

The value G(-1/q) is read off the series, with no division by zero: it equals (r/2 + 1 - r/(2q)) if r is even, and ((r+1)/2)(1-1/q) if r is odd.  The needed sign is that of (-1)^r, equivalently (-1)^r C(pi)>0.

If r is odd, this quantity is 1/(1+q) times (q-1) Q_r + 2 q^{r+1} G(-1/q), a sum of two positive terms.

If r is even, the same computation reduces to the comparison (r+2) q^{r+1} - r q^r > (q^{r+2}+q^r-2)/(q+1).  Clearing the positive factor q+1, the difference of the two sides is (r+1) q^r (q^2-1) + 2(q^{r+1}+1), which is positive.

This completes the proof of the sign lemma, and therefore of the unit-circle statement for the closed-form family.

## Remarks

A remainder bound against the top term 2 p^{r/2} cos(r theta) cannot be the mechanism at the first test angle: there the top term has the wrong sign, and the lower terms have to win.  The test angles theta=m pi/r are also unsuitable in general; already at r=2, p=2 the cosine polynomial vanishes at pi/2.  The angles m pi/(r+1) are the right ones, as they were for the invertible factor, but the cosine polynomial is not a two-term Chebyshev combination, and the argument above uses the kernel closed form instead.

The numerically tightest test value is |C|=2(sqrt(p)-1) at r=1, p=2 (and again at the first test angle for r=2, p=2), equal to about 0.828.  A high-precision check of the four closed values of Psi, of the formula for C(pi), and of the sign pattern, for every r=1..20 and p in {2,3,5,11} (1760 test points), found no failures.  Combined with the earlier floating-point sweep through r<=40 and p<=47, this is consistent with the proof, not a substitute for it.

The unit-circle statement for the Solomon factor of an Eichler order of level p^r is therefore proved whenever the closed form is known.  That is presently for the coefficients Q_1, Q_2, and Q_3 (the last for r>=3), and in the double-computed cases r=1,2,3,4 as a whole.  For general r it remains conditional on the remaining coefficients and on the palindrome.

Figure: `order-solomon-sign-lemma`.  Records: `order-solomon-unitcircle-sign-check.json`, `order-solomon-sign-lemma-check.json`.

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

[project, proved — refereed] The condition "whenever the closed form is
known" is now met for every r>=1: the closed form is proved in
`order-solomon-column-pair-proof.md`. Independently,
`order-solomon-chebyshev-short-proof.md` gives a shorter proof that every
root of Q_r(v)=sum_{d,e=0}^r p^{min(d,e)} v^{d+e} is simple with
|v|=p^(-1/2). Both were refereed in night Block 2
(`BLOCK2-REFEREE-20261110.md`); no error found. The circle statement for
these local Solomon numerators is therefore unconditional. The text above
saying it "remains conditional" is the dated afternoon record, kept as
history. This is a statement about one local factor, not a global zeta
function, not RH.
