# The Solomon factor at level p^r: closed form, and the first coefficients

10 November 2026, America/Los_Angeles.

[project, partial; project conjecture]  The coefficients Q_1, Q_2, Q_3, and Q_4 are proved, for every r large enough that the index in question is at most the level.  The remaining coefficients Q_j for 5<=j<=r, and the palindrome (the local functional equation), remain a project conjecture.  Not an RH claim.

## The guessed formula

Let O be an Eichler order of level p^r at a split prime p, r>=1, and let Q_r^{Sol}(v), with v=p^{-s}, be the Solomon (all right O-modules) local factor relative to 1/((1-v^2)(1-p v^2)).  The guess, for every r>=1, is 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, Q_0=1, Q_1=2, Q_2=p+2 for r>=2, Q_3=2(p+1) for r>=3, Q_4=p^2+2p+2 for r>=4, and in general Q_{2s}=(p^{s+1}+p^s-2)/(p-1), Q_{2s+1}=2(p^{s+1}-1)/(p-1) for 2s<=r or 2s+1<=r as appropriate.  The upper half is determined by p^r v^{2r} Q(1/(p v))=Q(v).  (When r=1 the mid-range formula for Q_2 is replaced by the palindromic leading term Q_2=p, which is the same rule.)

An earlier wording that r=1 was exceptional confused this all-modules factor with the invertible factor 1+p u = 1+p v^2.  The all-modules factor at r=1 is 1+2v+p v^2, which is exactly the same rule.  This was obtained in two ways (index-p recursion through k=8, and a complete ambient HNF scan through k=6 at p=2 and k=5 at p=3), with Euler termination, at both p=2 and p=3.  Record: `order-solomon-r1-results.json`.

The same rule was already tested at r=2,3,4 (Block 10), each double-computed at p=2 and p=3.

## What is proved: Q_1=2 for every r>=1

Write A_k for the number of full-rank right O-submodules of index p^k, so that A(v)=Q(v)/((1-v^2)(1-p v^2)).  The split-maximal factor has only even powers, with M_0=1, M_1=0, M_2=1+p.  Hence A_1=Q_1, and Q_1 is the number of right O-submodules of index p.

Any such submodule I contains pO, so the count is the number of F_p-hyperplanes in O/pO that are right modules for the algebra O-bar=O/pO.  For every r>=1 this algebra is four-dimensional, with basis e=E11, f=E12, g=p^r E21, h=E22, and with g f =0 in O-bar.  Its radical is J=span{f,g}, J^2=0, and O-bar/J is F_p x F_p.  There are therefore exactly two simple right O-bar-modules, both of order p, and exactly two maximal right ideals:

    I_e = span{f,g,h},    I_h = span{e,f,g}.

Their preimages in O are the two index-p right modules, written M_e={v_p(a)>=1} and M_h={v_p(d)>=1} in the coordinates (a,b,c,d) of the matrix [[a,b],[p^r c, d]].  Thus Q_1=A_1=2 for every r>=1.  (By contrast, for the maximal order r=0 one has O-bar=M_2(F_p), whose unique simple is two-dimensional, and there is no right ideal of index p.)

## What is proved: Q_2=p+2 for every r>=2

From the Euler relation, A_2=(p+1)+Q_2, so it is enough to prove A_2=2p+3.

The only simple right O-modules of finite p-power order have order p, because a simple factors through O-bar/J isomorphic to F_p x F_p.  In particular there is no simple of order p^2.  Hence if I is a right O-module of index p^2, the quotient O/I is not simple, so it has a submodule of order p, whose preimage is an index-p overmodule of I.  Every index-p^2 module is therefore contained in M_e or in M_h.  Their intersection K=M_e cap M_h is the preimage of I_e cap I_h = span{f,g}, of index p^2, and is the unique module contained in both.

It remains to count the index-p right submodules of M_e, and of M_h.  These correspond to O-bar-stable hyperplanes in the four-dimensional spaces M_e/p M_e and M_h/p M_h.  For r>=2 the right action, in coordinates (a,b,c,d) of [[a,b],[p^r c, d]], is as follows.

On M_e={a in p Z_p} / p M_e, with coordinates (alpha, beta, gamma, delta) for a=p alpha, the action of [[x,y],[p^r z, w]] is (alpha, beta, gamma, delta) |-> (alpha x,  beta w,  gamma x + delta z,  delta w), independent of y.  The stable hyperplanes are {alpha=0} together with the p+1 hyperplanes {B beta + D delta=0}, for a total of p+2.  On M_h={d in p Z_p} / p M_h, with coordinates (alpha, beta, gamma, delta) for d=p delta, the action is (alpha, beta, gamma, delta) |-> (alpha x,  alpha y + beta w,  gamma x,  delta w).  The stable hyperplanes are {delta=0} together with the p+1 hyperplanes {A alpha + C gamma=0}, again p+2.

Therefore A_2=2(p+2)-1=2p+3, and Q_2=p+2, for every r>=2.  (For r=1 the action on M/pM changes, because g f = p h is nonzero in O/p^2 O; the same overmodule argument gives A_2=2(p+1)-1=2p+1, hence Q_2=p, matching 1+2v+p v^2.)

The same charts classify the 2p+3 modules of index p^2, for later use.  The image of pO in M_e/p M_e is the line span{e_alpha}, so the unique child of M_e that does not contain pO is N_e={alpha=0}={v_p(a)>=2}; the remaining p+1 children contain pO, namely K={delta=0} and the p modules L_{e, lambda}={v_p(a)>=1, b+lambda d \equiv 0 mod p} for lambda in F_p.  Symmetrically, the unique child of M_h that does not contain pO is N_h={v_p(d)>=2}, and the others are K together with p modules L_{h, nu}.  Thus the types of index-p^2 modules, for r>=2, are: one copy of K (two parents, contains pO); 2p modules of type L (one parent, contain pO); and two modules N_e, N_h (one parent, do not contain pO).

## What is proved: Q_3=2(p+1) for every r>=3

From the Euler relation, A_3=Q_3 + Q_1 M_2=Q_3+2(p+1), so it is enough to prove A_3=4(p+1).  The argument is again a count of stable hyperplanes, now on each type of index-p^2 module, followed by an explicit listing of the children so that overlaps can be removed.  The hypothesis r>=3 is used once, to make the action on N_e/p N_e coincide with the r>=2 action on M_e/p M_e already computed (the lower-left generator contributes only terms of valuation at least 3, hence vanishes in N_e/p N_e).

The same method as for Q_2 gives the number of children of each type, for r>=3 (in fact the counts for K and for type L already hold for r>=2):

- N_e, and likewise N_h, has the same action as M_e (respectively M_h), hence p+2 children.
- On K/pK, for r>=2, both off-diagonal generators act as zero, and the action is (alpha, beta, gamma, delta) |-> (alpha x, beta w, gamma x, delta w).  The stable hyperplanes are the p+1 functionals of the form (A,0,C,0) together with the p+1 of the form (0,B,0,D), a total of 2p+2, with no overlap.
- On L_{e, lambda}/p L_{e, lambda}, in coordinates (alpha, mu, gamma, delta) with b=-lambda d + p mu, the action is (alpha, mu, gamma, delta) |-> (alpha x,  alpha y + mu w,  gamma x + delta z,  delta w).  The only stable functionals are (1,0,0,0) and (0,0,0,1), so each of the 2p modules of type L has exactly two children.

These child counts already imply that the number of incidences (an index-p^3 module counted once for each of its index-p^2 overmodules) is (2p+2)+2p*2+2(p+2)=8p+6.  To pass from incidences to modules one lists the children in coordinates and compares.

The p+2 children of N_e are A3_e={v_p(a)>=3}, K_{2,1}={v_p(a)>=2, v_p(d)>=1}, and P_lambda={v_p(a)>=2, b+lambda d \equiv 0 mod p} for lambda in F_p.  Symmetrically, the children of N_h are A3_h={v_p(d)>=3}, K_{1,2}={v_p(a)>=1, v_p(d)>=2}, and p further modules Q_nu={v_p(d)>=2} cut by a linear condition in (a,c).  Each P_lambda is also the {alpha=0} child of L_{e, lambda}, and all p of the L_{e, lambda} share their second child, the preimage G_min of span{g}, namely {v_p(a), v_p(b), v_p(d)>=1}.  Likewise all L_{h, nu} share the preimage F_min of span{f}, namely {v_p(a), v_p(c), v_p(d)>=1}, and their other children are the Q_nu already listed.  Finally, the 2p+2 children of K comprise F_min, G_min, K_{2,1}, K_{1,2}, and 2p-2 further modules that do not appear in any previous list (the remaining hyperplanes {A alpha + C gamma=0} and {B beta + D delta=0}, other than the four just named).

These families are pairwise distinct as index-p^3 lattices, by their defining valuation conditions.  In brief: A3_e has v_p(a)>=3, which none of the others do; A3_h has v_p(d)>=3, likewise unique; each P_lambda has v_p(a)>=2 and d unrestricted at valuation 0 (so not contained in K), and the p values of lambda are distinct hyperplanes; the Q_nu are the symmetric family with v_p(d)>=2 and a unrestricted; K_{2,1} has v_p(a)>=2 and v_p(d)>=1, which is strictly smaller than a P_lambda and strictly larger than A3_e; K_{1,2} is the symmetric intermediate; F_min and G_min are the unique two modules containing pO (equivalently, the preimages of the two one-dimensional right ideals of O/pO); and the remaining 2p-2 children of K all satisfy v_p(a)=v_p(d)=1 together with one extra linear congruence in the chart of K, so they are contained in K but in neither N_e nor N_h.  An apparent overlap between any two of these families would impose an extra congruence and drop the index to at least p^4.  The count of distinct children is therefore

    (p+2) + (p+2) + 1 + 1 + (2p-2) = 4p+4 = 4(p+1).

Thus A_3=4(p+1) and Q_3=2(p+1) for every r>=3.  As a check, this listing also recovers the parent-type counts later seen in the enumerator: there are exactly two index-p^3 modules containing pO, namely F_min and G_min, matching the elementary count of one-dimensional right ideals of O/pO (only span{f} and span{g}).

The same A_3 was seen, independently of r for r>=3, in the index-p recursion at (p,r)=(2,3),(2,4),(2,5),(3,3),(3,4), each giving A_3=4(p+1).

## What is proved: Q_4=p^2+2p+2 for every r>=4

From the Euler relation, A_4=Q_4 + Q_2 M_2 + Q_0 M_4=Q_4+(p+2)(p+1)+(1+p+p^2), so it is enough to prove A_4=3p^2+6p+5.  The argument is the same child-count as for Q_3, now on each of the five types of index-p^3 modules already listed there.  The hypothesis r>=4 is used once, to make the action on A3_e/p A3_e (and symmetrically on A3_h) coincide with the r>=2 action on M_e already computed: the lower-left generator contributes only terms of valuation at least 4, hence vanishes in A3_e/p A3_e.  The actions on the remaining types already hold for r>=3, and in several cases for r>=2.

The five types and their child counts, for r>=4, are as follows.

- A3_e={v_p(a)>=3}, and likewise A3_h, has the same action as M_e (respectively M_h), hence p+2 children.  In the chart a=p^3 alpha of A3_e, those children are A4_e={v_p(a)>=4}, K_{3,1}={v_p(a)>=3, v_p(d)>=1}, and P_lambda^{(3)}={v_p(a)>=3, b+lambda d \equiv 0 mod p} for lambda in F_p.  Symmetrically, the children of A3_h are A4_h={v_p(d)>=4}, K_{1,3}={v_p(a)>=1, v_p(d)>=3}, and the p modules Q_nu^{(3)} obtained by cutting {v_p(d)>=3} by a linear condition in (a,c).

- Each of the 2p modules of type P or Q has p+2 children.  On P_lambda={v_p(a)>=2, b+lambda d \equiv 0 mod p}, in coordinates (alpha, mu, gamma, delta) with a=p^2 alpha and b=-lambda d + p mu, both off-axis generators other than the residual z-term in (gamma, delta) vanish for r>=3, and the action is (alpha, mu, gamma, delta) |-> (alpha x, mu w, gamma x + delta z, delta w).  The stable hyperplanes are {alpha=0} together with the p+1 hyperplanes {B mu + D delta=0}.  The children are therefore P_lambda^{(3)} (already listed under A3_e); the single module G^{(2)}={v_p(a)>=2, v_p(b)>=1, v_p(d)>=1}, independent of lambda, which is shared by every P_lambda; and p further modules S_{lambda, lambda'}={v_p(a)>=2, b+(lambda + lambda' p) d \equiv 0 mod p^2} for lambda' in F_p, each with unique parent P_lambda.  Symmetrically, writing Q_nu={v_p(d)>=2, a+nu c \equiv 0 mod p} for nu in F_p^*, and Q_infty={v_p(d)>=2, v_p(c)>=1}, for the p modules of type Q other than K_{1,2}, the children are Q_nu^{(3)} (already listed under A3_h); the single module F^{(1,2)}={v_p(a)>=1, v_p(c)>=1, v_p(d)>=2}, shared by every Q_nu; and p unique-parent lifts S'_{nu, tau}.  For nu finite these are {v_p(d)>=2, a+(nu + tau p) c \equiv 0 mod p^2}; for Q_infty they are {v_p(c)>=2, v_p(d)>=2} together with {v_p(c)>=1, v_p(d)>=2, p a + kappa c \equiv 0 mod p^2} for kappa in F_p^*.

- K_{2,1}={v_p(a)>=2, v_p(d)>=1}, and likewise K_{1,2}, has the same action as K, hence 2p+2 children.  In the chart a=p^2 alpha, d=p delta of K_{2,1}, both off-diagonal generators vanish for r>=3, and the action is (alpha, beta, gamma, delta) |-> (alpha x, beta w, gamma x, delta w).  The children are K_{3,1} (already listed under A3_e); F^{(2)}={v_p(a)>=2, v_p(c)>=1, v_p(d)>=1}; G^{(2)} (already listed under the P_lambda); K_{2,2}={v_p(a)>=2, v_p(d)>=2}; the p-1 modules T_lambda={v_p(a)>=2, v_p(d)>=1, p b + lambda d \equiv 0 mod p^2} for lambda in F_p^*; and p-1 further modules {v_p(a)>=2, v_p(d)>=1, a + mu p^2 c \equiv 0 mod p^3} for mu in F_p^*, each with unique parent K_{2,1}.  Symmetrically, the children of K_{1,2} are K_{1,3}, F^{(1,2)}, G^{(1,2)}={v_p(a)>=1, v_p(b)>=1, v_p(d)>=2}, the same K_{2,2}, the p-1 modules T'_mu={v_p(a)>=1, v_p(d)>=2, a + mu p c \equiv 0 mod p^2} for mu in F_p^*, and p-1 unique-parent modules {v_p(a)>=1, v_p(d)>=2, d + lambda p^2 b \equiv 0 mod p^3}.

- F_min={v_p(a), v_p(c), v_p(d)>=1} has the same action as M_e, hence p+2 children, namely F^{(2)}, F^{(1,2)}, the module pO itself, and the p-1 modules W_lambda={v_p(a), v_p(c), v_p(d)>=1, p b + lambda d \equiv 0 mod p^2} for lambda in F_p^*.  Symmetrically, G_min has the action of M_h, with children G^{(2)}, G^{(1,2)}, pO, and the p-1 modules W'_mu={v_p(a), v_p(b), v_p(d)>=1, a + mu p c \equiv 0 mod p^2}.

- The remaining 2p-2 children of K, written R_lambda^d={v_p(a), v_p(d)>=1, p b + lambda d \equiv 0 mod p^2} and R_mu^a={v_p(a), v_p(d)>=1, a + mu p c \equiv 0 mod p^2} for lambda, mu in F_p^*, each have p+2 children, not the 2p+2 that a naive comparison with K would suggest.  The distinction is in the residual off-diagonal action, which does not vanish in R/pR.

  On R_lambda^d, in coordinates (alpha, mu, gamma, delta) with a=p alpha, d=p delta, and b=-lambda delta + p mu, the generator E12 acts nontrivially: the action is (alpha, mu, gamma, delta) |-> (alpha x, alpha y + mu w, gamma x, delta w), the same as on M_h.  (The image of E12 is (0, a, 0, p^r c), which lies in R_lambda^d for r>=2 and is (0, alpha, 0, 0) in the chart.)  The stable hyperplanes are {delta=0} together with {A alpha + C gamma=0}.  The children are G^{(1,2)}, independent of lambda and already listed under K_{1,2} and G_min; T_lambda, already listed under K_{2,1}; W_lambda, already listed under F_min; and the p-1 intersections U_{lambda, mu}=R_lambda^d cap R_mu^a={v_p(a), v_p(d)>=1, p b + lambda d \equiv 0, a + mu p c \equiv 0 mod p^2} for mu in F_p^*.

  On R_mu^a, in coordinates (kappa, beta, gamma, delta) with a=-mu p gamma + p^2 kappa, d=p delta, and b=beta, the generator E12 does vanish for r>=2, but the lower-left generator does not.  The image of g is (p^r beta, 0, p delta, 0); writing this back in the chart of R_mu^a gives kappa' = p^{r-2} beta + mu delta, so for r>=3 the induced action is (kappa, beta, gamma, delta) |-> (kappa x + mu delta z, beta w, gamma x, delta w).  A stable functional (A,B,C,D) cannot have A nonzero, because the z-term would move only the last coordinate; the remaining stable hyperplanes are {gamma=0} and {B beta + D delta=0}, a total of p+2.  The children are F^{(2)}, independent of mu and already listed under K_{2,1} and F_min; T'_mu, already listed under K_{1,2}; W'_mu, already listed under G_min; and the same intersections U_{lambda, mu} for lambda in F_p^*.

These child counts imply that the number of incidences is (4p+2)(p+2)+2(2p+2)=4p^2+14p+8.  To pass from incidences to modules one compares the coordinate lists just given.  The distinct index-p^4 modules, with their counts, are:

- A4_e and A4_h: 2;
- K_{3,1} and K_{1,3}: 2;
- the P_lambda^{(3)} and the Q_nu^{(3)}: p+p;
- the unique-parent lifts S_{lambda, lambda'} and S'_{nu, tau}: p^2+p^2;
- G^{(2)} and F^{(1,2)}: 2;
- F^{(2)} and G^{(1,2)}: 2;
- the unique-parent children of K_{2,1} and of K_{1,2}: 2(p-1);
- K_{2,2}: 1;
- the T_lambda and the T'_mu: (p-1)+(p-1);
- pO: 1;
- the W_lambda and the W'_mu: (p-1)+(p-1);
- the intersections U_{lambda, mu}: (p-1)^2.

These families are pairwise distinct as index-p^4 lattices, by their defining valuation and congruence conditions.  In brief: A4_e is the unique module with v_p(a)>=4, and A4_h is the unique module with v_p(d)>=4; each P_lambda^{(3)} has v_p(a)>=3 and d unrestricted at valuation 0, and the p slopes are distinct, while K_{3,1} is the intermediate with v_p(a)>=3 and v_p(d)>=1; the Q_nu^{(3)} and K_{1,3} are the symmetric family; each S_{lambda, lambda'} has v_p(a)=2 for a generic element and d unrestricted, so is contained in neither K_{2,1} nor a P^{(3)}, and the p^2 slopes modulo p^2 are distinct; the S' are the symmetric family with v_p(d)=2 generically and a unrestricted; G^{(2)} is characterised by v_p(a)>=2, v_p(b)>=1, v_p(d)>=1 with c unrestricted; F^{(2)} by v_p(a)>=2, v_p(c)>=1, v_p(d)>=1 with b unrestricted; and F^{(1,2)}, G^{(1,2)} are the two remaining depth-two "min" modules, with v_p(d)>=2 rather than v_p(a)>=2; K_{2,2} is the unique module with v_p(a)>=2 and v_p(d)>=2 and no further linear condition; pO is the unique module in which all four coordinates have valuation at least 1; each T_lambda has v_p(a)>=2, v_p(d)>=1 and the congruence p b+lambda d \equiv 0, and a generic element has v_p(b)=0 (so T_lambda is not contained in G^{(2)}); the T'_mu are the symmetric family; the unique children of K_{2,1} carry the deeper congruence a+mu p^2 c \equiv 0 mod p^3, which is not the R_mu^a condition a+mu p c \equiv 0 mod p^2, and a generic element has v_p(c)=0 (so these are not contained in F^{(2)}); the unique children of K_{1,2} are the symmetric family with d+lambda p^2 b \equiv 0 mod p^3; each W_lambda has v_p(c)>=1 together with the R_lambda^d congruence, and a generic element has v_p(b)=0 (so W_lambda is not pO); the W'_mu are symmetric; and each U_{lambda, mu} is cut out by both residual R-conditions at once, with a generic element of valuations v_p(a)=v_p(d)=1 and v_p(b)=v_p(c)=0.  An apparent overlap between any two of these families would impose an extra congruence or a higher valuation and drop the index to at least p^5.  The count of distinct children is therefore

    2 + 2 + 2p + 2p^2 + 2 + 2 + 2(p-1) + 1 + 2(p-1) + 1 + 2(p-1) + (p-1)^2
    = 3p^2 + 6p + 5.

Thus A_4=3p^2+6p+5 and Q_4=p^2+2p+2 for every r>=4.  As a check, the same listing recovers the parent-type counts later seen in the enumerator (for instance: G^{(2)} has p+2 parents, namely the p modules P_lambda together with K_{2,1} and G_min; F^{(2)} has p+1 parents, namely the p-1 modules R_mu^a together with K_{2,1} and F_min; each U_{lambda, mu} has exactly the two parents R_lambda^d and R_mu^a; and pO has exactly the two parents F_min and G_min).

The same A_4 was seen, independently of r for r>=4, in the index-p recursion at (p,r)=(2,4),(2,5),(3,4), each giving A_4=3p^2+6p+5.  The 3p^2+6p+5 coordinate lattices listed above were also constructed directly from their defining conditions, reduced to Hermite form, and compared with the enumerator's complete fourth layer at those three pairs: the two sets agree, with no missing or extra keys.  Record: `order-solomon-q4-check.json`.

## The palindrome, still a conjecture

The remaining coefficients Q_j for 5<=j<=r, and the palindrome Q_{2r-j}=p^{r-j} Q_j, have not been proved.  A natural attempt is to produce the palindrome from a duality on right modules.  The reduced-trace (or matrix-trace) dual I |-> I-hash, scaled by p^r and transported by the standard involution, does send O to the right O-module {v_p(a)>=r, v_p(d)>=r} of index p^{2r}, and it sends some modules of index p^k to modules of index p^{2r-k}.  This is not a bijection between all right submodules of those two indices: already A_k is not equal to A_{2r-k} (for r=3, p=2 one has A_1=2 and A_5=40).  The desired palindrome is a relation on the Solomon polynomial Q, equivalently on the ratio of the local zeta of O to the split-maximal factor, and not a palindrome of the raw counts A_k.  A literature functional equation for the full Solomon zeta of a non-hereditary Eichler order, divided by the known equation for M_2(Z_p), would give the palindrome, but no such theorem was found in a form that applies here (Solomon 1977; Bushnell-Reiner 1980/86; Voll 2010; Chassin 2023 treats the Lipschitz order).  The local functional equation therefore remains a project conjecture, supported by the double-computed cases r=1,2,3,4.

Records: `order-solomon-r1-results.json`, `order-solomon-r3-results.json`, `order-solomon-q3-check.json`, `order-solomon-q4-check.json`.

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

[project, proved — refereed] The closed form above is now proved for every
r>=0, not only for Q_1..Q_4: the column-lattice count in
`order-solomon-column-pair-proof.md` gives
Q_r(v)=sum_{d,e=0}^r p^{min(d,e)} v^{d+e}, whose low coefficients equal the
guessed sum_{m=0}^j p^{floor(m/2)} and whose upper half is the palindrome
Q_{2r-j}=p^{r-j}Q_j. That proof was refereed line by line in night Block 2
(`BLOCK2-REFEREE-20261110.md`) with no error found, and Q_4 was re-derived a
third way (full 4D Hermite-form brute force, p=2,3,5) in agreement.
The earlier sections of this note, which call Q_5 onward and the palindrome
a "project conjecture", are kept unchanged as the dated afternoon record
(10 November, Blocks 11-14); they are superseded by this update, not
contradicted. Scope: this split local Eichler order only; counts all
embedded right ideals (not the invertible factor). No novelty or RH claim.
