Fuchs–Meiri–Sarnak: audit of the two headline thinness families

Result. The cited reflection-quotient method can be completed for Theorem 1(i) in every odd dimension n >= 5 and for Theorem 1(ii) in every odd dimension n >= 7. It fails for the dimension-5 member of (ii): the actual orbit lattice is integrally isometric to the reflective exception <-16> ⊕ D4. Thus the published claim excluding all the relevant lattices from Nikulin's list is false. Thinness of that particular hypergeometric group remains unresolved in this audit; no arithmeticity conclusion follows from the ambient lattice being reflective.

This is a curated source-and-arithmetic adjudication, with no additional model calls or independent paid repair verdict. The new dimension-5 finding is separate from the already established omission in the parameter classification, Theorem 2.6. The degree-13 classification counterexample is not needed anywhere below.

1. Frozen source and external proof inputs

The reviewed journal edition is Fuchs–Meiri–Sarnak, Hyperbolic monodromy groups for the hypergeometric equation and Cartan involutions, JEMS 16 (2014), 1617–1671: publisher PDF. Relevant locations: Theorem 1 p.1619, the reflection-lattice discussion and Table 1 pp.1634–1635, Proposition 4.4 pp.1639–1640, and Lemma 4.6 / Theorem 4.7 pp.1643–1646. The external proof-input cutoff is the printed received date, 2013-08-11. Native arXiv v2 formulas were checked against the published statements; the frozen manuscript files have not been edited.

The external mathematical inputs used here are:

• Nikulin 1979, On factor groups of the automorphism groups of hyperbolic forms modulo subgroups generated by 2-reflections, original Russian Doklady 248(6), 1307–1309. Primary article and PDF link. The definitions and Theorems 1, 4, 5 classify the relevant even lattices; Theorem 4 specifically includes <16> ⊕ negative D4, the opposite-sign convention to ours. The original scanned article was inspected through the MathNet text extraction, retained as nikulin-1979-web-extract.txt. Direct PDF downloads from this host timed out; no local original-PDF hash is claimed. The result's 1979 publication and received date are recorded by MathNet. Its proof is accepted as a classical external theorem, not re-proved here.
• Beukers–Heckman 1989, Monodromy for the hypergeometric function nFn−1, Invent. Math. 95, 325–354. Archived primary paper. Theorems 3.5 (Levelt realization), 5.8 (the remaining imprimitivity criterion) and 6.5 (Zariski closure) supply the standard analytic identification and algebraic-group conclusions. Exact hypotheses are checked in section 7 below; the classical proofs themselves are not re-audited.

Both sources predate the cutoff by decades. Nikulin's arXiv:0810.2945v1 was also cached, but is background only. Vinberg's high-dimensional theorem is not required for these two even-lattice families, since the cited Nikulin classification already applies in every relevant rank. No later theorem is used to justify the original proof.

2. Companion pairs and the correct finite generator

Use standard companion matrices A and B for P and Q; put C=A^{-1}B and u=(C−I)e_n. In both cases C is a rank-one involution, u_n=−2, and

The complete cyclotomic polynomials for the two families are

Here T specifies an oriented root orbit, not a change in the actual group H=<A,B>. In family I, A has finite order dividing n+1. In family II, A has finite order dividing 2n−2, whereas the original B has a triple root at 1 and is of infinite order. Since conjugation by −A is identical to conjugation by A, T works for generating the same normal reflection subgroup. This removes the need to identify the original infinite-order B with a finite-order generator.

The spectra are disjoint for every odd n >= 5. For I use gcd(n,n+1)=1 and the omitted roots 1 and −1. For II a common root other than ±1 would satisfy z^(n−2)=1 and z^(n−1)=−1, forcing z=−1, a contradiction; ±1 are separately disjoint. P is squarefree in both families, proving the stated finite-order assertion for A.

Let u_j=T^j u for 0 <= j < n and L=span_Z(u_0,...,u_{n−1}). These vectors are independent. In the companion module Q[x]/P, u is represented by −x^{-1}Q(x); coprimality makes it a cyclic vector for A and hence T.

This is the same integral orbit lattice as that generated using the original B. Indeed B=AC and C z=z+z_n u; induction expresses B^j u as A^j u plus an integral combination of earlier A-orbit vectors. The transition is triangular with diagonal 1. Alternating signs for T=−A are also unimodular. The exceptional lattice below is therefore not an artifact of changing to a different lattice.

3. Uniform Gram matrices and connectivity

Normalize (u,u)=−2. In the basis u_0,...,u_{n−1}, the form K is symmetric Toeplitz. Its first row is

These are all-dimensional formulas, not an inference from the finite examples. One derivation uses the determinant lemma on B=A(I+u e_n^t):

For I, the ratio is (1+z)^2(1−z^n)/[(1−z)(1−z^(n+1))]. Its coefficients through degree n−1 give exactly the first row above. For II replace z by −z: the ratio becomes (1+z)^2(1+z^(n−2))/[(1−z)(1+z^(n−1))]. The final two coefficients are 5 and 6; the preceding ones are 1,3,4,...,4.

In the orbit basis, T is the companion of its characteristic polynomial, and C=I+e_0 K_{0,*}. Checking the shifted columns and last column gives T^tKT=C^tKC=K. For I the last companion column is (1,−1,1,...,1); for II it is (1,−1,0,...,0,1). These identities prove invariance uniformly. Thus L is integral and H-stable; its diagonal is −2, with adjacent pairings −3, so it is even, with primitive bilinear form. Successive u_j are joined by actual edges of X_f.

The minus sign in T=−A for family II is essential to this literal graph statement: (u,Au)=+3, but (u,−Au)=−3. Cartan reflections are unchanged when their root vector is negated. This corrects the published positive-pairing use without changing the group or its integral lattice.

4. Signature, discriminant and rational isotropy for all dimensions

For I, with J the all-ones matrix and E the path adjacency matrix,

The positive-definite matrix D=2I+E has leading minors j+1. Its inverse takes the all-ones vector to (1/2,0,1/2,...,1/2). The determinant lemma gives det K=−n(n+1). A rank-one subtraction from a positive form, together with K_{00}<0, gives exactly one negative direction. Since n is odd and greater than 1, the determinant has an odd divisor; L is not 2-elementary.

For II put N=n−1 (even), y_0=x_0−x_{n−1}, y_j=x_j (1 <= j <= n−2), and s=sum x_i. In R^N define columns r_0=e_0−e_{N−1} and r_j=e_{j−1}+e_j for 1 <= j < N, and let R have these columns. Then

The change x -> (y,s) has determinant of absolute value 2, and |det R|=2 for even N. This proves signature (n−1,1) and det K=−64 in every odd n >= 5. With w=e_0+e_{n−1}, one has Kw=−81 and (w,w)=−16. Hence w/8 is in L and has order 8 modulo L. This excludes every 2-elementary lattice without requiring an all-n Smith-normal-form conjecture.

There are explicit rational null vectors as well: −e_0−e_2+e_3 works for I in every rank and for II when n >= 7; −e_0−2e_2+e_4 works for II at n=5. Thus the rational-isotropy prerequisite does not require an unexamined appeal to a general theorem.

5. Reproducing the finite exceptions: the dimension-5 failure

The exact script reconstructs the root-lattice Gram matrices and their Smith factors instead of trusting Table 1. The candidate non-2-elementary exceptional discriminant orders in odd ranks are:

The script additionally retains the 2-elementary U+A1+E8 printed in FMS as a harmless extra candidate, and retains U+A1+E6 from the primary classification. Neither affects our exclusions. The basic D4 discriminant group is (Z/2)^2, not Z/4; the printed factors 4·4, 4·8, 4·16 in the rank-5 table do not record the correct groups for <-2^k>⊕D4. Their correct nonunit Smith factors are (2,2,4), (2,2,8), (2,2,16).

Family I has discriminant n(n+1), different from every same-rank exception. Family II has discriminant 64, excluding every exception in odd ranks >=7, but not the rank-5 exception <-16>⊕D4.

At n=5 the second-family Gram matrix has first row (−2,−3,−4,−5,−6). Define

Then det M=−1 and

Nikulin's Theorem 4 includes this exact lattice (with reversed sign), so O(L)/R_2(L) is finite. Thus the infinite-quotient hypothesis used by the certificate is actually false, not just missing from an explicitly verified calculation. Merely tightening the wording of Lemma 4.6 cannot fix this application. A separate thinness argument, or an arithmeticity demonstration, is needed for this one group.

The actual parameters here are alpha=(1/8,3/8,1/2,5/8,7/8), beta=(0,0,0,1/3,2/3). This is a genuine member of the stated n>=5 family; it is irreducible, primitive and hyperbolic. The arithmetic below does not show that H itself has finite index in O(L).

6. Completing the valid thinness deductions

Let N be generated by the reflections A^j C A^{-j}; these are the same reflections in the oriented vectors T^j u. Since H=<A,C>, N is normal and H/N is cyclic of order dividing ord(A). Thus N has finite index in H for both families, including the cases where the original B is infinite-order.

All the relevant roots lie in one component of X_f by their −3 adjacent pairings. The checked rank-two identity writes each neighboring Cartan-reflection product as a product of integral norm-2 hyperbolic reflections. Products telescope along paths; all generators of N therefore have the same involutive image in O(L)/R_2(L). Its image has order at most two. Whenever that quotient is infinite, N and hence H have infinite index in O(L).

Nikulin's classification and the exclusions in sections 4–5 establish this for I in all odd n>=5 and II in all odd n>=7. This is an explicit completion of the intended argument, retaining its supplied classical ingredients. It is not a rescue of the exceptional dimension-5 application.

7. Zariski density and the meaning of thinness

To distinguish thinness from just infinite index, the usual hypergeometric Zariski-closure input must be applied. It is not necessary to trust the disputed seven-family classification for this.

The normal reflection subgroup N acts irreducibly over C. An N-invariant subspace that is affected by one rank-one reflection contains that root; the nonzero adjacent pairings propagate containment through the full orbit basis. If every reflection fixes the subspace pointwise, it is orthogonal to the basis, hence zero. By Beukers–Heckman Theorem 5.8, an imprimitive H with irreducible reflection subgroup would require one spectrum to be a scaled regular n-gon, i.e. a characteristic polynomial x^n−a. Neither of our P,Q has that form: their x^(n−1) coefficients are nonzero. Thus H is primitive.

Two adjacent reflections have a product whose restriction to their span has eigenvalues (7+3sqrt(5))/2 and (7−3sqrt(5))/2; it has infinite projective order. Hence H is infinite and is not a scalar shift of a finite group. The complete cyclotomic spectra are reciprocal, and the distinguished reflection has determinant −1. Beukers–Heckman Theorem 6.5 therefore gives full complex orthogonal Zariski closure. The explicit real form has signature (n−1,1), as established above. These checks also show that dimension 5 is not excluded by primitivity or by a failure of the standing hyperbolicity assumption.

8. Reproduction, scope and next obligation

Run python3 check_headline_families.py. The run checks all 26 odd-rank cases from 5 through 29, both companion spectra, finite-order bounds, cyclic bases, integral unimodular transition from the original B-orbit, invariant Gram identities, exact signatures, rational null vectors and independent exceptional-lattice Smith computations. It verifies the integral D4 isometry exactly. The all-dimensional arguments in sections 2–7 are separately supplied; finite numerical success alone would not establish them.

Public reproduction and remaining work

The exact script, results, external-input records, one-page PDF supplement and portable bundle accompany this analysis. The public panel view contains the decorated decision graph. The dimension-five subgroup index remains unresolved after the later bounded computations; the other thinness families and appendix remain separate audit obligations. No additional paid inference was used.