# Hypercomplex critical-line sprint: current questions

Pruned 11 November 2026, ~3 AM PST (Grant's request): only mathematical research questions are kept here. Removed items: /workspace/math/pruned-20261111/open-questions-PRUNED.md.

Updated November 10, 2026, 4:35 PM PT (research continuing; no deploy). Queue for the existing 4×/day routine; no schedule or saved-prompt change.

Latest page: https://zeta-zeros.grant-golland.workers.dev/critical-line-hypercomplex

Publication receipts are under `live/`; the latest page is linked above. This queue records research status, not a schedule change.

## Closed in the Nov 10, ~4:30 PM PT continuation (notes: q8-negative-twist-simplicity.md, inherited-variance-long.md, octonion-euler.md)

- [project — Arb enclosure, PARI coefficients beyond n=2000] Item 2/3: all 12 negative admissible twists (D=-7,-11,-19,-31,-43,-59) of the two sign -1 Q8 bases (2.2304.8t5.b.a, 2.42025.8t5.a.a) have root number -1 and a simple forced central zero (L'(1/2) from 2.6 to 8.4). The four earlier-unreliable 42025 rows are resolved with 43k-135k coefficients. Sign +1 base has no forced zero. Still open: audit of coefficients past n=2000 independent of PARI for these twists; larger |D|; universal simplicity not claimed.
- [numerical, uncertified] Item (b): own Riemann-Siegel windows at 1e6, 1e9 (120k zeros), 1e12: Sigma^2(10)=0.42, 0.49, 0.54, saturating near 0.5-0.56 at Berry scale, below GUE 0.83. Slow log growth, as expected. No certificate.
- [project, speculative framing] Item 5 octonion part: octavian Euler products (real s) keep the norm exactly but direction depends on order and, new vs associative control, on bracketing (<=0.8 deg at s=1.5, fading with s). No arithmetic octonionic Frobenius/L-function known; complex s unresolved.

## Closed in the Artin height-100 isolation run (Nov 10, 4:15 PM PT)

- [project — interval proof + literal Turing count] The 185-vs-224 mismatch is resolved: it was a 768-bit precision floor (Hardy ball radius ~1e-59 swallows the decaying kernel for t>~83), not a Turing, window, seed or sign-change problem. At 1536 bits all three Q8 Artin L-functions have every positive zero to height 100 isolated at width 1e-12: 179 / 180 / 224, equal to Turing N+(100), 0 failures, 0 overlaps. q8-artin-isolates-100.md; q8_artin_isolates_100_v2.py; figure figures/q8-artin-isolates-100.svg. No RH claim.
- [project — exact comparison] Independent a_p audit: all unramified p<=20000 (2260 per field), a_p from maximal-order residue degree vs PARI lfunan, 0 mismatches, all three fields. Ramified primes not independently checked. q8_frobenius_audit.py.
- [re-checked] Gelfond question: no change, stays closed (merged 2/3/5 gap law does not need it).

## Closed in the height10000 inheritance run (November 10, 2026, 01:48 PM PST)

- [project — interval proof + literal Turing average] Item6 requested10000 target complete: N_zeta(20000)=22491, order positive counts D2=47188,D3=48478,D30=55806; outer counts22491 each; middle2206/3496/10824; centres0/0/2. Full224-bit replay and all complementary exclusions pass. order-height10000.md and order-height/height-10000.json.
- [project — interval proof] Minimum inherited/local vertical gaps through10000: D2>0.000224748746,D3>0.000055637662,D30>0.000009241204; unique ordinate-pair witnesses certified,288-bit replay agrees. Both outer lines copy ordinary heights divided by2; each prime supplies its own evenly spaced local lattice. No interaction or uniform all-height gap is established.
- [project —4:30PM PST handoff] The requested30/100/1000/10000 chain is complete. Do not treat the older1000 “follow-up” wording below as an outstanding request. Preserve all certificates/IDs; next research should return to the remaining Artin/Turing or independent-coefficient questions, not silently launch a larger order window. Deploy only fully built/checked trees. No schedule or saved-prompt change, no external message.

## Closed in the height100/1000 run (Nov10, 1:26 PM PST)

- [project — interval proof + literal Turing average] Item6 finite targets100 and1000 complete. Full split, proof and resumable commands: order-height.md; data order-height/summary.json; higher-precision replay order-height/verification.json. These ordinary-zeta facts are already known far beyond our small windows; no RH progress.
- [project — handoff for the4:30PM PST sprint] Keep the completed30/100/1000 evidence and live IDs. Historical1000-run handoff, now superseded above: Height10000 was the requested FOLLOW-UP, not completed in that earlier run; source-pinned pipeline and measured estimate are ready. Do not confuse it with Artin height100 or claim success before its own Turing bound and replay. Deploy only a fully built/checked tree and record the new version. No schedule or saved-prompt change; no external message was sent.

## Closed in the off-axis interval run (Nov 10, 12:56 PM PST)

- [project — interval proof] Maximal quaternion-order ideal-index zetas D=2,3,30: complete nonreal zero maps to height30; positive counts32,36,57, every nonreal zero simple; centre orders0,0,2. Ordinary xi contour to60 gives26 total;13 disjoint on-line brackets plus conjugates exhaust it. Direct order contours give64,72,116, matching two inherited families plus exact local roots.33240 archived segments replayed as40702 validated512-bit leaves;78 inherited and47 local complementary-factor exclusions. Full certificate `order-off-axis-results.json` (gzip publication copy), `order-off-axis-verification.json`, `order-off-axis.md`; figure `figures/order-off-axis-map.svg`, mobile variant. Known factorization transported, not independent information or RH progress. Item6 finite off-axis target closed. Artin height100/Turing and coefficient audits beyond2000/5000 were not advanced.

## Closed in the Frobenius-quaternion run (Nov 10, 12:30 PM PT)

- [project — exact proof + exact rational checks; float numerics] Item 5 partly answered with a motivated system: actual Q8 Frobenius elements are quaternions. The ordered Euler product E_P(s), real s>1, has |E_P|² = L_P(rho,s) for every representative choice and order (norm multiplicativity); its direction depends on the choices (up to 53° at s=3/2 for 2.42025.8t5.a.a). frobenius-quaternions.md; figure figures/q8-frobenius-directions.svg.

## Closed in the twist-simplicity run (Nov 10, 7:30 AM PT)

- [project — interval proof] All 12 positive admissible quadratic twists of the two sign −1 bases (D = 13,17,29,37,53,61) have SIMPLE forced central zeros: L′(1/2) enclosures from 3.998… to 13.767…, all excluding 0, widths ≤ 4e-11. Same method reproduces both base slopes. Coefficients beyond the audited prefixes (2000/5000) are PARI's exact Artin coefficients, not separately audited. twist-simplicity.md/.py/results; figure figures/q8-twist-slopes.svg.

## Closed in the extended bundle

- [project — interval proof] All positive zeros through height 50 have disjoint brackets of width 1e-12. Positive counts at T=10/20/50 are 10/25/79 for 2.2304.8t5.b.a, 11/26/79 for 2.2304.8t5.a.a, and 15/34/101 for 2.42025.8t5.a.a. Symmetric multiplicity totals at 50: 159,158,203. Whole-contour winding counts match the located zeros plus partners/centres, proving no missing off-line or repeated zeros in these finite windows. q8-extended.py/results; extended.md.
- [known/project] Main terms M_N(50): 78.707641239,78.707641239,101.813898610. They are not exact predictions. Gamma-phase residuals are recorded, not an independent Turing averaged-argument certificate. The corrected accepted run took 235.04 seconds with Arb at 768 bits; every quadrature band and tail is bounded explicitly.
- [project] 5000 classic-Q8 coefficients independently audited by exact modular/maximal-order prime decomposition and cross-checked against PARI. Coefficient cutoffs for the accepted run: 1000,1000,5000. Earlier first-1000 arithmetic audit retained for the first two fields.
- [known/project] Two actual maximal orders verified by exact products and reduced discriminants: Hurwitz D=2, basis 1,i,j,(1+i+j+k)/2; D=3 in (-1,-3), basis 1,i,(1+j)/2,(i+k)/2. Ideal-index zeta is zeta(2s)zeta(2s-1) times local factors. Central values -log(2)/2 and -log(3)/2, not zeros. Local-factor zeros exactly at 1/2+i*pi*k/log(p), k nonzero; inherited selected zeros appear on quarter/three-quarter lines. All inherited zeros there would require RH. 48 mpmath root checks (not intervals), residual <8e-79. Voight §26 and quaternion-orders.py/results.
- [known consequence/project checked] D=30 order zeta has a double central zero, with coefficient -2 log(2)log(3)log(5)=-2.45117397417... . This local-factor cancellation differs from odd-sign Q8 Artin forcing. More generally r finite ramified primes give central order r-1; a definite rational algebra requires r odd.
- [project] New three-figure summary separates directional zeros, Artin centres, star brackets, order-zeta families, holes at infinity and lattice-centre normalization. No RH-progress claim.

## Earlier fresh bundle retained

- [project — interval proof] First positive critical zeros, each width 1e-12:
  - 2.2304.8t5.b.a: (1.612576900762, 1.612576900763).
  - 2.2304.8t5.a.a: (1.504548805445, 1.504548805446).
  - 2.42025.8t5.a.a: (1.264700058509, 1.264700058510).
- [project — interval completeness] Argument-principle counts through absolute height 2 are 3,2,3, multiplicity counted. The negative cases include their simple centre plus a conjugate pair; the positive case only the pair. Thus exactly one positive zero per base below 2; these are simple and on line. No additional off-line/repeated zeros in this FINITE Artin window. Not Turing's particular averaged-argument inequality, not a PARI mesh certificate, not ordinary-zeta/RH progress. Proof/data: fresh.md; q8-low-zeros.py/results.
- [project — exact topology] Xi sphere-complement infinity-end system: free tail groups with meridian inclusions, non-pro-trivial although ordinary inverse limit is trivial. Compactified zero set not locally connected at infinity, not a horned sphere. Our meridians survive H1; Alexander horned-sphere exterior pi1 is nontrivial with H1=0. Full proof/other-end caveat: infinity.md.
- [known/project — exact tests] Ghiloni–Perotti Remark 3 reproduced, associative control included. Three-direction STAR toy bracketings have roots (-e1-e2+e4+e7)/2 versus -e1; same imaginary-unit sphere, common normal (1+w²)^3. Finite POINTWISE product zeros CANNOT change by bracketing (norm multiplicativity). 431 rational checks; not an arithmetic L-function.
- [known/project] E8 shells exactly 240,2160,6720; integer contributions 112,1136,3136 and half-integer 128,1024,3584. Epstein mirror centre n/4: Gaussian .5, Z4 quaternion lattice 1, E8 2. Rescaling exponent to |v|^(-8u) renames E8 centre .5, not its zero lines; under RH its two nontrivial-zero families would be at Re u=1/8,7/8. Completed centre value -5/6, not zero. e8.py/results.

## Earlier closed results retained

- [project — interval proof] The two negative central zeros are simple: L'(1/2) in (3.32,3.49), (1.62,1.69). q8-simplicity.py/results; q8-sprint.md.
- [known/project] 18 positive admissible quadratic twists preserve signs, D=13,17,29,37,53,61 on three bases. SST Proposition 8.7; q8-sprint.py/results. Non-admissible -4 control flips the 2304 pair's signs; coprimality matters.
- [project] Actual Artin critical output maps: minus -> imaginary spheres, plus -> real points; direction reversal at t=2.5 in the 2304 negative base. q8-sprint-results.json.

## Next run: one auditable unit at a time

1. [open] Beyond height 50: target 100 only after increasing exact coefficient cutoffs, Gauss order and working precision enough to beat exponentially small completed values. Reuse the validated exponential-sum/Taylor method; prove all tail bounds before accepting signs. The second/third zero and height-4 questions are closed, not work to repeat. If pursuing literal Turing averaging, distinguish that new implementation from the existing complete argument-principle count.
2. [open] Negative admissible quadratic twists: PARI 2.17.2 reports unimplemented gamma factors. Find a supported independent field/representation construction or another documented implementation; do not manufacture unverified analytic data.
3. [mostly closed] Central-derivative certificates done for all 12 positive admissible sign −1 twists. Still open: negative-D twists (blocked with item 2), larger D families, and an independent audit of coefficients past n=2000/5000. Universal Q8 simplicity is not claimed and not a reachable computation.
4. [open] Actual xi norm-level tube topology at infinity. The bare zero-set end system is now calculated; do not repeat it as a new result. Seek what genuine level-set critical values or ends add, with no horn assertion unless proved.
5. [partly closed Nov 10, 12:30 PM PT: Frobenius-quaternion Euler products, real s>1 only — see frobenius-quaternions.md] Complex-s half closed Nov 11 (project; see frobenius-complex-s-20261111.md and /critical-line-hypercomplex#q8-frobenius-complex-s): quaternion-internal i is choice-dependent, central i (M₂(C)) gives det = Artin L exactly. Previously listed as open: any extension to complex s (p^{-s} and g_p then fail to commute, and the norm identity breaks), an octonionic analogue with real arithmetic meaning, and whether any choice-free invariant beyond the norm exists. Original item: a mathematically motivated multi-direction coefficient system beyond the constructed star toy. Distinguish finite pointwise products, coefficientwise star products, order, bracketing and convergence. The norm obstruction rules out finite pointwise bracket-sensitive zero sets.
6. [closed finite map through10000 Nov10 PST] Requested30/100/1000/10000 chain COMPLETE. At10000, positive counts D2=47188,D3=48478,D30=55806; each outer line22491; middle2206/3496/10824; centres0/0/2; all nonreal zeros simple. Literal Turing bound on[20000,20064] gives N_zeta(20000)=22491; full224-bit replay and complementary exclusions pass. Minimum vertical inherited/local gaps: D2>0.000224748746,D3>0.000055637662,D30>0.000009241204,288-bit replay agrees. See order-height10000.md and current certificates.706 ordinary/175 local chunks, source-pinned resume; six bounded replayable archive parts. Actual generation+required replay489.1sec,51.7MiB peak; the earlier15–30min forecast was conservative. Inheritance is exact, not interaction or new RH information. Preserve original30 and100/1000 proofs and every page ID. No larger window is silently queued.
7. [open] For E8, verify additional shell counts or zero-family examples only if they answer a specific question. Renaming the midpoint is not a new zero theorem. Distinguish lattice Epstein zeta from octavian multiplication and quaternion-order ideal zeta.

## Publication guardrails

Identify the object. Known claims cite primary sources; project means checked here, not new mathematics; speculative means unproved. Do not upgrade floating-point checks to intervals. Preserve page IDs, plain explanations and figures. Build/check NO ERRORS before deploy; verify live page and assets. Leave literature/ read-only. Hamilton/Hurwitz order ideal zeta at 1/2 is -log(2)/2, not a forced Artin zero.


## Added 2026-11-10 (from downward/think-hard-transfer.md)
- [closed for the requested proper-subfield targets; project — interval proof, Nov 10 PST] Independent field-arithmetic/Gaussian-AFE pipeline now evaluates ζ_F5, ζ_F41, ζ_F205 and ζ_E without ordinary-ζ or Dirichlet/Artin-L value calls. Prime-ideal coefficients (n≤1000 in each field) match PARI exactly; bnfcertify plus interval unit-log determinants provide residues. At 1/2 all four orders are 0; at each of the first three ordinary zeros all are 1. Signed subfield projector recovers ordinary orders 0,1,1,1. Whole-strip positive counts at T=30 are 14,24,31,63; (14+24+31−63)/2=3 recovers N_ζ(30). This is methodologically independent of the old factored certificate and agrees with it. Report/scripts/certificate: ../downward/think-hard-independent.md and ../downward/results/independent/. Figure: ../downward/figures/independent-projector-match.svg. That computation made no deployment or saved-routine change; its results were subsequently published by the user-authorized Part A site update.
- [open, optional] Direct degree-eight ζ_K evaluator: not certified in the independent run; the proper-subfield projector does not require it. The chosen Gaussian tail bound becomes prohibitively expensive at gamma degree8. Do not relabel the old factored ζ_K calculation as independent.
- [open, analytic—not solved by finite checks] A uniform field/family inequality that excludes the compatible Heilbronn assignment Θ_G=m·1_G at a hypothetical off-line ordinary zero. The successful independent computation validates the finite return map, not an RH strategy. See ../downward/think-hard-heilbronn.md §3.


## Added 2026-11-10 (Grant): zero movement and kinds
- Inherited vs local zeros: tag them (our own labels, not standard notation) on figures and page; avoid "00", which reads as a double zero.
- Movement in the zero tree: deform local factors (1 - a_p p^{1-2s}) and interacting families; track whether zeros cross, repel, merge or change kind. Notes go in zero-movement.md.

## Added 2026-11-10, afternoon PT: Gelfond, textures follow-up, remaining queue

- [closed, correction — known argument + project check] The merged 2/3/5 gap law does **not** need Gelfond's theorem that exp(pi) is transcendental. Local zeros sit at k*pi/log p; the law samples residuals {k log q / log p} at those event times. Equidistribution on the square is unique factorisation in Z. Gelfond would be needed only for the unused integer-time sequence t |-> t log p / pi. The earlier zero-textures.md wording is corrected in its dated §9; figure figures/zero-textures-equidistribution.svg (and mobile). Script: zero_textures_equidistribution.py.
- [closed, this afternoon — project count certificate, 10 November 2026, America/Los_Angeles] (2) Q8 zero counts to height 100, with Turing averaging, per Artin factor of ζ_K for 2.2304.8t5.b.a. Isolated N_+(100): zeta 29 (existing height-20000 certificate; Trudgian's zeta bound is not applied, it needs t>168 pi), χ_8 61, χ_12 67, χ_24 79 (Rumely/Trudgian, t0=100>50; all three characters even), Artin 179 plus the exact central zero (Booker Theorem 4.5; N(t)=N_++1/2 isolates 179.5 in every window h=8..30). Distinct from the height-50 argument-principle counts. Hardy Z sign changes confirmed in the Turing windows. Note: q8-height100.md; figures: figures/q8-turing-counts and q8-turing-bounds (desktop and mobile). Scripts: q8_turing.py, q8_turing_figures.py. Not an RH claim. Other Q8 bases and interval isolation of individual Artin ordinates remain open.
- [closed, this afternoon — project coefficient audit, 10 November 2026, America/Los_Angeles] (3) Independent Artin coefficients for 2.2304.8t5.b.a by Hecke induction, not by factoring the octic. For each real quadratic subfield, the conductor-discriminant modulus (norms 288, 192, 96) plus both infinite places, selected by lfunparams [2304, 1, [1, 1]] and L(1/2)=0, produces the same series. All 5000 coefficients match PARI lfunartin of the octic, used only as a check. First nonzero prime trace a_23 = -2. Note: q8-hecke-coefficients.md; figure figures/q8-hecke-traces. The earlier q8-coefficients-results.json audits the other field 2.42025.8t5.a.a and is not in conflict. Not an RH claim.
- [closed for the reliable range, this afternoon — project numerical check, 10 November 2026, America/Los_Angeles] (4) Negative-D admissible twists. PARI 2.17.2 lfuntwist still refuses negative D. A documented lfuncreate of a_n chi_D(n), with gamma = [nu xor (D<0)] twice and the root number left unknown, was validated against lfuntwist for D=13 and then run for D=-7,-11,-19,-31,-43,-59 on all three Q8 bases. 14 of 18 rows have FE bits below -40; all 14 preserve the base sign, as SST Proposition 8.7 predicts. The four conductor-42025 rows with |D|>=19 are unreliable (too few coefficients) and are not used. Note: q8-negative-twists.md; figure figures/q8-negative-twists. Not interval certificates, not an RH claim. Still open: more coefficients for the large-42025 twists, and Arb derivative certificates for the new sign -1 twists.
- [closed, this afternoon — project computation on published lists, 10 November 2026, America/Los_Angeles] (5) Longer inherited number variance. Unfolded inherited Sigma^2 equals unfolded ordinary Sigma^2, so a longer inherited window is a longer ordinary window. Certified T=20000 stays flat at Sigma^2(10)=0.324, S rms=0.297. The first 100,000 published Odlyzko zeros barely move this (0.345). Two published high windows (n=1e12 and n=1e22, 10,000 zeros each, not our certificate) rise to 0.52 and 0.56 at L=10 and flatten near 0.45-0.60, still below GUE 0.83-1.02: a short stretch is too rigid, and S(T) grows only like sqrt(log log T). A certified Berry window is out of reach for this pipeline. Note: zero-textures-berry.md; figure figures/zero-textures-berry. Not an RH claim.
- [closed, this afternoon — literature distinction, 10 November 2026, America/Los_Angeles] (6) Octonion analogues. Three objects must be kept apart: (known) the E8 Epstein zeta Z_E8(s)=240*2^{-s} zeta(s) zeta(s-3), already computed in e8.py, a lattice zeta not an ideal zeta; (known) Coxeter's integral Cayley octaves, Duke Math. J. 13 (1946), whose underlying lattice is E8, with ideal-theoretic work by Mahler 1942, van der Blij-Springer 1959, and Rehm, but no standard ideal zeta built from that arithmetic; (speculative, not found) a Hey/Solomon/Voight-style zeta of a nonassociative octonion order. Voight is associative. This project does not invent that series, and does not relabel Z_E8 as that object. Note: octonion-analogues.md; figure figures/octonion-analogues. Ghiloni-Perotti checks remain function-theoretic, not an L-function.

## Added 2026-11-10, later afternoon PT: Block 2 closures

These close leftovers and follow-ups from the same afternoon's Block 1 notes. This is not a schedule change, not a saved-prompt change, and not a request to deploy.

- [closed, this afternoon — project count certificate] Turing counts to height 100 for the other two Q8 bases. Plus 2304: Artin N_+=180, certified, every h in {6..30} isolates 180 (h=6: [179.104, 180.904]); Dirichlet/zeta reused. 42025: zeta 29, chi_5 54, chi_41 87, chi_205 113 (Rumely, all even, all isolate); Artin N_+=224, Booker N=224.5, every h in {8..40} isolates 224.5. A 0.02 pad was unsafe because of zeros at 99.99341 and 100.00420; Hardy side-of-100 certificates and pad 1e-4 fix the isolation. Notes: q8-block2-leftovers.md, q8-turing-other-results.json; figures q8-turing-other-counts and q8-turing-other-bounds.
- [closed, this afternoon — project coefficient audit] Hecke induction for the plus 2304 and 42025 Artin factors: three constructions agree, 5000 coefficients match lfunartin (check only). Plus 2304 traces agree at p=1 mod 4 and flip at p=3 mod 4 (the non-admissible chi_{-4} relation). An order-2 character on Q(sqrt(2)) has the same conductor and gamma but a different series. Residue-degree Frobenius, read from the octic and not from the L-function code, agrees with lfunan at all 2260 unramified primes through 20000 on all three fields (0 mismatches). Notes: q8-hecke-other-results.json, q8-frobenius-audit-results.json.
- [closed for the previously unreliable rows, this afternoon — project numerical check] The four conductor-42025 negative twists with |D|>=19, rerun at PARI's preferred lengths (138,379 to 425,495), have FE bits about -202 and all preserve W=-1. Earlier W=+1 readings were short-series artefacts. Still a numerical check, consistent with SST Proposition 8.7, not an interval certificate of the root number.
- [closed, this afternoon — project interval isolation] Every positive Artin ordinate through height 100 has a disjoint width-1e-12 Hardy bracket, counts matching Turing N_+(100): 179, 180, 224. Accepted run: 1536-bit Arb, 256 Gauss nodes, height-dependent quadrature error, 20,000 coefficients for 42025. A second run reproduced every bracket. This is not a new winding-count completeness proof (the height-50 argument-principle certificate is unchanged), and it is not an RH claim. Note: q8-artin-isolates-100-results.json. The earlier open-questions item "beyond height 50: target 100" is closed for these three Artin factors in this finite window.
- [closed, this afternoon — project interval proof] Arb L'(1/2) excludes 0, with central multiplicity 1, for all twelve admissible odd twists of the two sign -1 bases, including the four large 42025 conductors. Enclosures from 2.64388 to 8.41216. The negative-D half of the earlier twist-simplicity open item is closed for these twelve twists. Note: q8-negative-twist-simplicity.md. Universal Q8 simplicity is still not claimed.
- [closed, this afternoon — project computation] Owner-labelled gaps through height 100. Smallest same-owner Artin gaps 0.0780, 0.0810, 0.0950. Closest cross-owner pair: chi_12 vs minus Artin at t=57.4627, gap 0.000128. Counts of pairs within 0.001 sit next to the independence prediction n m (2d/T). Note: q8-owner-gaps.md.
- [closed, this afternoon — project check of an exact law] Merged-comb gap law for a general number k of ramified primes, checked at D=6,10,15,21,30,42,210. Centre order k-1, continuous density of degree k-2, atom mass as recorded. D=30 and D=42 are definite (k=3). Even k is the same meromorphic product, not a definite-order zeta. Note: order-gap-law-k.md.
- [closed as a numerical experiment, this afternoon — project] Truncated Euler products and Dirichlet polynomials for the D=30 (and D=42) product through height 20, M up to 5000. Locals of H_M are exact at every cutoff; inherited zeros do not steadily emerge from a one-sided sum; a naive self-dual AFE for the product sits on the symmetry line, the wrong support for the inherited families. Note: order-truncations.md; comparison figure figures/order-truncations-afe.
- [closed later the same afternoon — project, numerical experiment] A smoothed product of two incomplete-gamma Riemann-Siegel approximants, one per zeta factor, times the exact ramified factors, at D=30 and D=42, height 20. Inherited zeros sit on the quarter and three-quarter lines from M=2 and are essentially exact by M=3 (mean distance 2e-8). The naive AFE does not do this. Note: order-rs-smoothed.md.

## Added 2026-11-10, later afternoon PT: Block 3 closures

These close follow-ups from Block 2. This is not a schedule change, not a saved-prompt change, and not a request to deploy.

- [closed, this afternoon — project, numerical experiment] Smoothed Riemann-Siegel product, as just recorded. Note: order-rs-smoothed.md. Figures: order-rs-smoothed, order-rs-smoothed-drift.
- [closed as far as certification allows, this afternoon — project, Turing count + numerical gaps] Owner-labelled zeros through height 500, all ten families Turing-complete at 300 and 500. Closest midpoint gap 9.66e-6 (chi_41 vs 42025 Artin at t=254.735), typical of a Poisson superposition. Certified Artin and zeta brackets through 100 are pairwise disjoint. PARI Hardy sign changes, not certified, are disjoint for the ten closest pairs through 500. Note: q8-owner-gaps-high.md.
- [closed as illustration, this afternoon] Animation frames of the merged comb for k=1 to 4 and the k=3 residual-torus flow. Note: order-animation.md.
- [closed as tightening, this afternoon] Plain-language guide grant-plain-20261110.md, and short closers on several of today's notes.

No writes to order-height/, no zeta-zeros-build edits, no deploy, no external message.

## Added 2026-11-10, evening PT: Block 4

- [closed, re-verified — known argument + project numerics] Gelfond: not needed for the merged 2/3/5 gap law; Weyl equidistribution needs only Q-independence of 1, log2/log5, log3/log5, i.e. unique factorisation. Numerics: atom 0.085509 at T=1e6 vs 0.085507; box discrepancy ~1/T. Open (not needed): a rate, which would need Baker-type linear-forms bounds. Note: block4-audit-and-gelfond.md.
- [closed — project audit + proof] Independent audit of Q8 Artin coefficients beyond 20,000 and without PARI classification: all primes <=300,000, three fields, agree with PARI lfunan; non-maximal primes 61 and 12919 (42025 field) fixed by Dedekind simple-factor rule; ramified a_p=0 proved (faithful rep, centre is the only involution). Still open: audit of twist coefficients (chi_D * a_n) for the large-conductor twists is by construction the same a_p times a character, not independently re-derived.

## Added 2026-11-10, late afternoon PT: Block 4 closures

These close follow-ups from Block 3. This is not a schedule change, not a saved-prompt change, and not a request to deploy.

- [closed, this afternoon — known mechanism + project recount] The factor-of-two deficit in inherited-local near-alignments is not a bookkeeping error. Units were already inherited heights, one per mirror pair. The deficit is the Landau / Ford-Zaharescu bias of {gamma log p / 2 pi} away from integers. The indicator form of the second main term is Conjecture 1 of Ford-Sound-Zaharescu, used as a predicted count; the twice-differentiable form is a theorem. Note: order-deficit-audit.md.
- [closed as a write-up, this afternoon — project] The merged-comb gap law, written as a self-contained proof sketch with every hypothesis stated, and checked on a second-pass independent enumeration. Unique factorisation only; not new deep math. Note: order-gap-law-k-proof.md.
- [closed as an interval check of an elementary identity, this afternoon — project] Centre order k-1, leading coefficient -2^{k-2} prod log p, enclosed at 512 bits for D=30, 42, and 210. Note: order-centre-leading.md.
- [closed as illustration, this afternoon] A ten-figure mobile-first gallery with plain-language captions. Note: gallery-20261110.md.
- Remaining older open items (complex-s Frobenius, Heilbronn uniform inequality, a direct degree-eight zeta_K evaluator, negative-D twist gamma factors, infinity level-set topology) were scanned and are not cheap-and-decisive; no new job was started from them.

No writes to order-height/, no zeta-zeros-build edits, no deploy, no external message.


## Night block 1, November 10, 2026 (PST): pending independent referee

[project, proved] A new column-lattice count confirms Solomon Q_4, proves
the full weighted-square numerator and palindrome, and computes Q_5/Q_6.
A shorter Chebyshev proof then removes the old closed-form condition
for this split local Eichler family. These are project proofs awaiting
a separate-agent/human referee pass, not mathematical-priority claims.
Read `NIGHT-LOG-20261110.md` for results, exact computation records, and
the precise next-block order. Preserve the afternoon statuses as history
until review. No live site edits or deployment were made by this block.

[speculative, mathematical] The log also preserves Grant's horned-sphere/Bost--Connes
question (a topology question about the Bost--Connes system) for a later literature check;
start from `grant-idea-horned-bost-connes.md`. It is not a proved topology switch.

## Night block 2, November 10, 2026, 11:45 PM PST: referee done

[project, proved — refereed] The night Block 1 column-lattice proof of the
full Solomon numerator Q_r(v)=sum_{d,e=0}^r p^{min(d,e)} v^{d+e} and the
short Chebyshev circle proof both passed a line-by-line referee pass; Q_4 was
re-derived a third way by full 4D Hermite-form brute force (p=2,3,5) in
agreement. The Solomon closed-form / palindrome / circle item is closed for
this split local Eichler family. Open: whether this formula is already in
the literature (none found in a short search; see
`BLOCK2-REFEREE-20261110.md` §5), a general FE for Gorenstein/Bass
quaternion orders (Bushnell–Reiner's self-duality remark is a lead), and
the octonion items. No site edit or deploy.

## Night block 3, 11 November 2026, 12:52 AM PST

- [closed — known + project, proved; double-computed] Octonion "ideal zeta" (the §6 octonion item and the octonion part of item 5). For the Coxeter order every one-sided ideal is mO (Petersson 2018 notes, Thm 1; vdB–Springer 1959; Allcock 1999). The left-multiplication Z-algebra is M_8(Z) (index 1, exact), so the counting zeta is ζ(8s): inherited zeros only, no new line. See octonion-ideal-count.md.
- [partly closed — project, double-computed; periodicity a conjecture] Gravesian order Z^8: 2-local factor Q(x)/(1−x⁴), x=2^{-s}, Q = 1+x+7x²+7x³+14x⁴+14x⁵+28x⁶+8x⁷+16x⁸ (two codes to index 2^12, one code to 2^18). Q(x)=(2x²)⁴Q(1/(2x)). 4 zeros on Re s=½, 4 off at Re s ≈ 1.183/−0.183 (Sturm-exact). Update Nov 11, 1:40 AM PST: periodicity now [project, proved — computer-assisted] (`gravesian-periodicity-proof.md`); G/2G ≅ F_2[(Z/2)³] derived directly, so the Albuquerque–Majid citation is no longer needed for it. Still open: a conceptual reason for period 4 rather than 8, Not RH-related. Second order checked (`octonion-intermediate-order.md`): R = Z<G,(1+e1+e2+e4)/2> has factor (1+3x²)/(1−x⁴) [project, proved — computer-assisted], zeros at Re s ≈ 0.7925 only, no functional equation. Open: Gorenstein test for R; the other intermediate orders.
- [reading, project] Bushnell–Reiner 1980 Thm 2 extends in outline to orders with trd-dual ξ^{-1}Λ, ξ normalizing Λ (Eichler O_r): the constant changes by a fixed power of p. The Tate-integral bookkeeping has not been written line by line. See night-br-atkin-lehner-20261111.md.
- Item 7 (E8): unchanged; no new shells were needed.

## Sprint run, 11 Nov 2026 7:30 AM PST

- Night block 3 item "Open: Gorenstein test for R; the other intermediate orders": **intermediate orders closed; Gorenstein test partly closed** (only an adapted definition exists). See `octonion-intermediate-orders-20261111.md`, with scripts and logs in `inter1111/`.
  - [project, computed exactly] Exactly 16 orders contain G = Z^8: G, one index-2 order S1 = G + Z(1+e1+…+e7)/2, seven index-4 orders (R and its conjugates), and seven maximal orders. Up to Aut(G) (1344 signed permutations) there are 4 classes. Correction: R ⊃ S1, so the chain is G ⊂ S1 ⊂ R ⊂ O. Only S1 is new.
  - [project, proved — computer-assisted, double-computed: PARI all-submodule search + pure-Python dual search to index 2^24, left and right; brute force at index 2 and 4] S1: 2-local factor (1+2x+15x²+35x³+14x⁴+14x⁵)/(1−x⁴), with conductor c = 2 bounding the primitive index by 2^16. No functional equation; gcd(Q, Q*) = 1, so no zero on Re s = ½ (exact). Zeros at Re s ≈ −0.567, 1.055, 1.943. Period 4 comes from L_{1+e_k}, as for G. G, R and O were re-derived and agree with Blocks 3–4.
  - [speculative adaptation of Voight Def. 24.2.1 / Lemma 24.2.5: codifferent locally principal, S^♯ = S·α, volume exponent N(α)^4] G yes, S1 no ([S1^♯:S1] = 2^6 is not a 4th power), R yes (R^♯ = R·(e1+e2)/2), maximal yes. The Block 4 guess "R fails Gorenstein" is refuted under this adaptation. Gorenstein-type does not give a ½-symmetric factor for octonion orders, and basic ⇏ Bass (S1 contains Z_2×Z_2).
  - Still open: an independent agent re-check; a standard (literature) Gorenstein notion for nonassociative orders; conceptual proofs of c(S1) = 2 and of the period-4 normalisation. Not RH-related.

## Queued 2026-11-11 (Grant asked, 3:30 AM PT): wild-sphere comparison for the zero set at infinity
- [closed for the bare zero-set comparison — project, proved; 11 Nov 2026 PST] See `fox-artin-comparison-20261111.md`. Finite sheets are smooth/locally flat; infinity is intrinsically non-locally-connected, not a manifold/sphere point. Complement 1-LCC fails at **all** sheet points as well as infinity: codimension two has ordinary meridians, so this is not a wildness criterion. The existing free-meridian tail system is not the Hawaiian-earring group. A slab-supported ambient-homeomorphism proof sends every bounded-strip rotational family escaping to infinity to the integer-radius model: no zeta-density or RH information survives in this unmarked topology. Fox–Artin Example 3.2 (p. 990), unlike our compactum, is an actual one-wild-point sphere with both domains open 3-cells; their local test is pp. 984–985. Primary Fox–Artin/Antoine 1921/Alexander inspected; Bing/Daverman opening definitions/theorem read via indexed excerpts, full PDFs not retrieved; Antoine 1920 announcement not retrieved.
- [open, narrowed] Specify a normalized norm-level family before investigating its compactification. Locally a small level about a zero is S¹×S^(d−2); fixed positive |ξ| does not isolate high zeros, since |ξ(σ+it)|→0 uniformly for 0≤σ≤1 (DLMF-based proof in note). No genuine Fox–Artin point or canonical joining/tubing construction has been proved. Figure and short section drafted for parent review; not published by this pass. No RH claim.

## Refinement run, 11 Nov 2026 9:22 AM PST
- [project · proved, computer-assisted, independently re-computed] "Independent re-check" of the octonion factors is now partly closed. A third, from-scratch program re-found the 16 orders, counted left and right ideals for every order (not one per class) to 2^16, and for G to 2^24. It also recomputed the conductor bound (c = 3, 2, 1, 0). Every count matches the G, S1, R-type and maximal factors. Still open: an outside referee; conceptual proofs of c(S1) = 2 and the period-4 normalisation. See `REFINEMENT-20261111.md`.

## Odd-prime figure, 11 Nov 2026 (PST)
- [closed — known; independently verified by two exact action-algebra computations] For every odd p and every one of the sixteen orders G⊆S⊆E₈, right-ideal index zeta is **1/(1−p⁻⁸ˢ)**. Only pᵃOₚ are ideals, with index p⁸ᵃ; no local zeros anywhere, hence none on Re s=½. Petersson 2021, Theorem 4.1(i), read directly, is stronger than the 2018 notes. Fano 64-word Gram certificate gives 8M₈(Z)⊆A uniformly; an independently coded Cayley–Dickson modular closure has rank 64 at 3,5,7. Counts through p³² agree (1 exactly at k=0,8,16,24,32); these are classification-derived counts, not full HNF sweeps. See `octonion-odd-primes-20261111.md`, `odd1111/`. No conjectural rational fit; no p=7 limitation; no RH claim.
- [open, narrowed] These orders cannot produce new odd-prime ideal-count zeros: their indices over G are powers of 2. An odd-prime phenomenon would require another order with odd-prime conductor (not computed) or a different explicitly defined counting object. The conceptual 2-local duality and period-4 normalization questions remain open.

## Global octonion ideal zetas, 11 Nov 2026 (PST)
- [closed — project, proved from established local factors] Z_G=ζ(8s)Q_G(x)(1+x⁴), Z_S1=ζ(8s)Q_S1(x)(1+x⁴), Z_R=ζ(8s)(1+3x²)(1+x⁴), Z_E8=ζ(8s), x=2⁻ˢ. Every local denominator pole cancels; only the simple pole s=1/8 remains. Nonmaximal orders gain a genuine zero column Re s=0, s=i(2n+1)π/(4log2). Only G has global zeros on ½. All kinds inherit zeros ρ/8; infinitely many on 1/16 are known, but **all** inherited zeros on 1/16 would require RH. New map transports 102 earlier certified ordinary zeros through t=30; no new RH result.
- [closed, carefully scoped — project, proved] E8 completion ξ(8s) reflects s↦1/8−s (center1/16). Usual completion ξ(8s)D_S(2⁻ˢ) fails that mirror for G,S1,R, even with a nonvanishing exponential conductor normalization: the bridge zero iπ/(4log2) has no reflected zero. Exact rational coefficient tests also rule out any single reciprocal center for each nonconstant D_S. This is **not** a theorem excluding every possible completion. Artificial symmetrization adds zeros; dividing out D erases the order correction.
- [open, narrowed] Seek an alternative completion only with a genuinely arithmetic definition and a stated allowable class of factors. The local G reciprocal center ½ is different from inherited1/16 and does not prove a global FE. Existing conceptual 2-local duality/normalizer questions remain open. Notes/data: `octonion-global-zetas-20261111.md`, `global1111/`.

## Mirror-line continuation, 11 Nov 2026 (PST)
- [closed — project, proved] Distinguish Q, F₂, and D=Q(1+x⁴). Q_G has center½; Q_R has shifted center log3/log4≈0.7925; Q_S1 none. Full F₂ has no monomial self-FE for G/S1/R because its pole line forces center0. D_G/D_S1/D_R have no single reciprocal center. E8 local pole mirror0 and global ξ(8s) center1/16 concern different objects.
- [closed — project, proved] Exactly four G roots are on the critical circle via Q_G=2x⁴P(x+1/(2x)), P=8u⁴+4u³−2u²+u−3. A rational Sturm chain has3−1=2 real roots, in(−5/4,−1) and(0,1); they give4 circle roots, while2 nonreal u roots give4 off. Reciprocity alone does not force circle confinement.
- [closed, new conceptual proof — project, proved] The formerly computational G/S1 normalizer/period-four question is closed: linearized alternativity gives T_iT_j+T_jT_i=−2δijI. For g=I+T_i, conjugation sends T_j to T_iT_j (j≠i), and sends T_omega to I+T_iT_omega; inverse formula also stays in A_S1. Thus g normalizes both operator algebras, detg16 and g²=2T_i, yielding L⊃gL⊃2L with index steps2⁴. This does not force a palindrome (S1 is a counterexample). A conceptual G-palindrome duality explanation is still open.
- [closed in a specified formal class — project, proved] Exact meromorphic gamma identity H(s)=D(2⁻ˢ)/D(2⁻ʳ⁽ˢ⁾)·H(r(s)), r(s)=1/8−s, is tautological and imposes no new zero restriction. The unique normalized reciprocal polynomial partner at that center has irrational coefficients: G at index2,2^(1/8)/2; S1 at index2,2^(1/8); R at index4,2^(1/4)/3. It is not an additive-index ideal-count zeta with the same finite-correction form. G trace dual(1/2)G is a fractional lattice, not an order; its multiplier order is G. Order conductor2O has multiplier O, so no new mirrored counted order emerges. No unrestricted no-completion theorem.
- [source checked / open] Bushnell–Reiner1980 pp156–158 re-read. Its associative simple-algebra/Fourier framework is not an octonion theorem; the earlier quaternion Atkin–Lehner extension remains an argument reading. See `octonion-mirror-lines-20261111.md`, `mirror1111/`. No RH claim.


## Octonion strangeness at 2 — 11 November 2026, PST (Grant pt194u)

- [closed for these four kinds; project, computer-assisted] Independent exact tensor constructions and a cyclic-closure algorithm versus a full 417,199-subspace sieve agree: nilpotent two-sided ideal dimensions G/S1/R/O = 7/6/4/0; quotient algebras F2, F2×F2, M2(F2), split Cayley. Associator-span dimensions mod2 = 0/3/4/7. G is commutative and associative mod2; the others are neither. All remain alternative/Moufang. Integer Smith invariants quantify mod4/mod8 spans without calling Z/4 a field. Notes: octonion-strangeness-at-2-20261111.md; data/code two1111/.
- [closed; project, proved] Polar ranks 0/2/4/8 follow from [S:G]=1/2/4/16: discriminant groups have exponent2 and polar determinant 2^8/[S:G]^2. This is a norm-rank theorem, not a zero-count theorem. O is nonsingular at2; split does not mean degenerate. Integral inclusions do not become embeddings of the reductions.
- [closed; project, computer-assisted] Mod2 right-ideal counts by codimension are G (1,1,7,7,15,7,7,1,1), S1 (1,2,15,35,15,8,7,7,1), R (1,0,3,0,1,0,3,0,1), O (1,0,0,0,0,0,0,0,1). They count only ideals containing2S; G and S1 first miss full-count lifts at index32 (7 vs15 and8 vs16). These are not the Q numerators.
- [closed; project, proved and checked] Same mod2 algebra need not determine cyclic lift data: T_m=Z*1+2^m O, m1,2, both reduce to F2 plus a seven-dimensional square-zero ideal, but have0 and128 cyclic index8 right quotients. Do not upgrade this quotient-type comparison to an uncomputed total index8 coefficient.
- [open, conceptual] Derive G's particular reciprocal numerator/quartic from arithmetic of the integral order, rather than inferring it from the radical, index or associator rank. Only G has half-line local zeros in the four-row comparison; no invariant-to-zero forcing theorem is established. Existing quartic/Sturm proof already gives exactly four roots on the circle; do not repeat that as open.
- [source-access caveat] Petersson pp2–3 and the low-characteristic Cayley paper were read. Coxeter1946 body not accessible; §4 pointer comes from Petersson. Conway–Smith ch9 pp99–105 and §§11.2/11.4 pp135/140 verified only in publisher contents; chapter bodies not read. A permitted copy would allow direct historical comparison, but no mathematical conclusion here depends on pretending those bodies were read.
- [published and verified, 11 November 2026, 11:19 AM PST] New section/desktop/mobile figure: /critical-line-hypercomplex#octonion-strangeness-at-2. check.py NO ERRORS; nine current/prior acceptance tests pass; ten live resources HTTP200 with exact bytes, all354 prior IDs preserved. Current version b63dc630-0feb-4a33-ac83-31037e90f6f8. DEPLOY-LOG.md and two1111/live-receipt.json record the deployment and verification. No external message, saved-routine or schedule change.

## Complex-s Frobenius products, 11 Nov 2026, ~12:35 PM PST (sprint)

- [closed for item 5's complex-s half — project, exact identity + float numerics] See `frobenius-complex-s-20261111.md`. With the complex unit I inside H, nrd(1 − g p^{−s}) = 1 − 2r(cos θ cos φ + sin θ sin φ (u·I)) + r²: only the Frobenius axis's component along I matters. The aligned class keeps one eigenvalue factor (sign-choice dependent), orthogonal classes lose all t-dependence (1 + r²). The norm stays order-free but becomes choice-dependent and ≠ |L(rho,s)| (log deviations up to 0.85, choice spreads up to 1.54 at σ = 3/2, field 42025). With a central complex unit (M2(C)), det E = L(rho,s) for every choice/order (≤ 1e-14), the trace is not choice-free. No zeros or critical-strip claim.
- [open, narrowed] Item 5 remainder: an octonionic analogue with real arithmetic meaning; whether any choice-free invariant beyond det exists for a richer construction (e.g. keeping track of the chosen primes as extra data, or a Hecke-character model over the aligned quadratic field). [speculative] The aligned-class product looks like half of a Hecke L-function of the order-4 character over that quadratic field, with the half picked by choosing primes; not checked.
- [done, staged Nov 11 ~1:00 PM PST] Section #q8-frobenius-complex-s added to /critical-line-hypercomplex after #q8-frobenius-quaternions (inside the Artin-examples disclosure), with site-style desktop/mobile SVGs (`figures/q8-frobenius-complex-s-site{,-mobile}.svg`, ~27 KB each). Built, check.py NO ERRORS, tests pass. Live deploy awaiting owner approval; the 827 KB results JSON is deliberately not published.

## Zeros-and-primes bridge — 11 November 2026, PST (Grant pt199u)

- [closed — project, proved from established factors; independently computed] The proposed zero-wave decomposition of the ordinary all-ideal summatory count is the wrong observable. For Z_K(s)=zeta(8s)C_K(2^-s), A_K(x)=sum_j c_j floor((x/2^j)^(1/8)); -C_K(1)<A_K(x)-C_K(2^(-1/8))x^(1/8)<=0 for every x. Zeros are not poles of its Perron integrand. Counts at2^128 are G6,845,563; S1 6,870,661; R394,101; E8 65,536. Two exact factor decompositions, separate odd-m enumeration and every prefix through2^18 agree. This is factor-derived counting, not a new lattice search.
- [closed — known classical formula + project adaptation] The zero-sensitive object is instead the **signed** logarithmic-derivative diagnostic B. It splits exactly into8psi_0(x^(1/8)) plus prime2-supported signed weights. Inherited checked pairs contribute x^(1/16) carriers at gamma/8; local columns yield an exact Fourier/geometric formula. Negative b4 values -39,-446,-14,0 rule out interpreting these coefficients as positive octonion prime-ideal counts. Local terms can dominate the inherited part; they are not generally a small or purely log-periodic wobble. No RH assumption or new odd-prime information.
- [closed as numerical illustration, not a certified tail bound] First100 previously certified pairs reused; numpy versus110-digit mpmath truncations agree below1.7e-12. Normalized grid RMS5/20/100 pairs:0.149/0.109/0.066. Exact signed local coefficients to128 agree by recurrence and integer matrix traces. See `octonion-explicit-formula-20261111.md`, `explicit1111/`.
- [open, quantitative] Specify an Abel/Fejer or other smoothing and prove a controlled local-column truncation error, uniform on a stated domain with jump conventions fixed. Combine this with a classical inherited-zero remainder without pretending finite numerical reconstructions are certificates.
- [open, arithmetic interpretation] A generic companion matrix reproduces the local polynomial's power sums but explains no order arithmetic. Find a natural arithmetic operator/duality if one exists; the earlier conceptual G-reciprocity question and outside-referee status remain open.
- [closed for six order shapes at p=3, Nov 11 ~2 PM PST — project, double-computed; see octonion-odd-conductor-20261111.md and /critical-line-hypercomplex#octonion-odd-conductor] Orders non-maximal at 3 (preimages of subalgebras of O/3O: Q, P6, P5, N4, U, and Z₃+3O) give local factors N(3^-s)/(1-3^-4s or -8s) with zeros but none on Re s=1/2 (exact gcd test), no mirror, and an odd-prime explicit-formula signal. Z₃+3O is degenerate (Gaussian-binomial counts). Was: To obtain genuinely new odd-prime factors, study an order with odd conductor or another explicitly defined counting object. Replotting this sixteen-order family's rescaled ordinary zeros cannot create an independent odd-prime signal. No RH-progress claim and no new schedule/routine.
- [partly closed Nov 11 ~5 PM PST — project; see octonion-odd-conductor-5-7-20261111.md and /critical-line-hypercomplex#octonion-odd-conductor-5-7] Same shapes at p=5,7 (N4, P5 also p=11): two programs agree for U, N4, P5 at 5 and 7 and P6 at 5; no zero on Re s=1/2 (exact gcd test); no mirror (P6 trivially on log(p+1)/(2 log p)); odd-prime explicit-formula jumps persist. U at p=3 is now double-computed too. Still open from the old item: diagonal order D; other ideal notions (two-sided only, invertible/locally principal) that might restore a functional equation. No RH claim.
- [partly closed Nov 11 ~5:10 PM PST — project, proved; see octonion-odd-conductor-general-p-20261111.md and /critical-line-hypercomplex#octonion-odd-conductor-proofs] General-p formulas: P6 = (1+(p+1)x²)/(1−x⁴) and Q = N_Q(x)/(1−x⁸) proved for every odd p (four-block reduction to two plane lattices); no zero on ½ for every p for both; coefficient of x = #ring maps Λ→F_p (2 for U, N4, P5; 0 for P6, Q). Still open: uniform classification of stable lattices for U, N4, P5 (nilpotent chain maps, no block splitting) — their fitted formulas remain [project conjecture] (N4, P5 confirmed at p=11); if true, no zero on ½ for any prime (resultants). Analytic derivation of N4's ½ ± 1/(4√2 p^1.5 log p). Independent exhaustive P6/Q at p=11 (P6 run started). Was: [open, new — project conjecture] Prove the polynomial-in-p formulas for the local polynomials (N4, P5 confirmed at a fourth prime p=11; U fitted from 3,5,7 only; P6 = 1+(p+1)x²), e.g. by a uniform cell decomposition of the stable-lattice poset. Finish Q at p=7 and the method-B check of P6 at p=7 (and Q at p=5 if still single). Z_p+pO at p≥5: structure proved for all odd p, but the totally-singular t-distribution and deep-lattice count (81 at p=3; p⁴ is a guess) are uncomputed.
- [N4, P5 closed Nov 11 ~8:25 PM PST — project, proved (computer-assisted, p-independent); see octonion-n4-p5-proof-20261111.md and /critical-line-hypercomplex#octonion-odd-conductor-proofs] Hull/core/glue reduction (Theorem 5) classifies the stable lattices for every odd p: N4 256 cells, P5 126, all one-step; exact glue counts give the fitted formulas, so "no zero on ½" is unconditional for N4 and P5, and the N4 pinch 1/(4√2 p^1.5 log p) is derived. Checked cell by cell at p = 3, 5, 7, 11.
- [closed Nov 11 ~9:05 PM PST — project, proved (computer-assisted); see octonion-u-proof-20261111.md and /critical-line-hypercomplex#octonion-odd-conductor-proofs] U: the 52 two-step cells are counted via the F_p[t]/t² look-alike (n4proof/tmodel.py) plus a carry check showing Teichmüller lifts never carry (carry2.py); all 802 cells match at p = 3, 5, 7; U's fitted formula is proved and the resultant rules out zeros on ½ for every odd p. Was: [open — project] U: 750 of 802 cells proved; the 52 cells with hull–core gap 2 (glue in a (Z/p²)-module; counts at p = 3, 5, 7 in n4proof/delta2-U.json, not p × the mod-p count) still need a uniform count. Proving it would make U's fitted formula and its "no zero on ½" unconditional.
- [open — project] Overnight checks: Q at p = 7 by method B and P6 at p = 11 by brute force were still running at ~9:05 PM PST (no longer needed for the proofs; they are independent cross-checks).

- [open — project, new] Two-step transfer in general: the carry check is a per-case computation for U. A structural reason (e.g. a torus grading forcing one-term carry sums) would explain it and might extend to deeper (δ ≥ 3) orders.
- [speculative — new, pt216u] Half L-function (q8-half-hecke-20261111.md, /critical-line-hypercomplex#q8-half-l-function): does the aligned product F have a natural boundary (Re s = 0 or earlier)? Estermann/Dahlquist/Kurokawa-type theorems suggest yes; not checked, sources not re-read.

## Q8 Artin explicit formula, 11 Nov 2026 (PST)

- [closed — known theorem (Weil 1952; Iwaniec–Kowalski Thm 5.12), project numerical check, double-computed] The Weil/Guinand explicit formula for 2.2304.8t5.b.a (N=2304, Γ_R(s+1)², ε=−1), 2.42025.8t5.a.a (N=42025, Γ_R(s)², ε=−1) and control 2.2304.8t5.a.a balances. It uses the certified zeros to height 100 and exact p^k ≤ 20000. With Gaussian h(r)=exp(−(r/T)²), T=1..20, |zero − (conductor+gamma+prime)| ≤ 1.7·10⁻¹⁰, always ≤ 0.36 of the error budget. With the cos-shifted family (T=8, x₀ ≤ 8.5), the residual is ≤ 1.4·10⁻¹¹ while the prime side is O(1–6). mpmath vs numpy/scipy codes agree to 2·10⁻¹⁴. The residual at T≈20 equals the predicted contribution of zeros above 100. The central zero supplies 87% (T=1) to 2% (T=20) of the zero side for b.a; omitting it breaks the identity by exactly 1. See `q8-explicit-formula-20261111.md`, `explicit-q8-1111/`.
- [closed — project, proved] Quaternion Frobenius in the explicit formula. Re(Frob^k) = ½χ(Frob^k), so the real part reproduces half the trace side exactly. The pure i,j,k part (size 0.26–0.44 at T=1, larger than the visible prime side 0.097) flips sign under a change of prime above p, so it is not a function of L(ρ)'s zeros. Its class average is 0 (the conjugation average lies in the centre R). Under the fixed "all +" convention, its k=1 part is ¼(ζ + χ_u − χ_v − χ_w) prime sums with ρ-coefficient 0. This was checked against PARI zeros of ζ and the quadratic L-functions to 10⁻¹⁰. The explicit formula sees only the reduced trace.
- [open] Replace the heuristic zero-count margin (E=10 above height 100) by a proved explicit N(T) bound for degree-2 Artin L-functions (e.g. a Trudgian-type bound for Dedekind zeta of K), making the budget fully rigorous; and certify the ζ / quadratic-L zeros used in 3(e).
- [open, speculative] For other finite G ⊂ H^× spanning H (binary tetrahedral/octahedral/icosahedral Artin representations, with quaternionic 2-dim ρ), is the reduced trace the only choice-free linear statistic of quaternion Frobenii, and does the fixed-convention pure part always decompose into L-functions of proper subquotients with zero ρ-weight? (Q8 done here.)
- [open] Fold a short "explicit formula" subsection (panels a, c, d) into /critical-line-hypercomplex when the build is idle and the owner approves deploy.

## Aligned Frobenius product as half an Artin L-function — 11 Nov 2026, ~9:00 PM PST (sprint)

- [closed — project, exact identity + float numerics; see `q8-half-hecke-20261111.md`, `halfhecke1111/`, `figures/q8-half-hecke.png`] The item-5 [speculative] "half of a Hecke L-function" line. With the complex unit of s equal to u and the canonical all-plus choice, F(s) = L(ρ,s)^{1/2}·[ζL(χ_u)/(L(χ_v)L(χ_w))]^{i/4}·[ζ(2s)/L(χ_u,2s)]^{1/4}·G(s), G zero-free holomorphic on Re s>1/3; L(ρ) = F·F⁻·∏_inert(1+p^{−2s})^{−1}. Verified for three Q8 fields × three units to 2.6e-7 (σ=2) and 1.4e-4 (σ=3/2), errors shrinking with the prime cutoff.
- [closed — project] Consequences: F is multivalued past Re s=1 (|F| jumps by e^{∓π/2} round zeros of ζ, L(χ_u) / L(χ_v), L(χ_w); sign flip round zeros of L(ρ)); measured box monodromy 5π/2 + 15πi/2 as predicted. The forced central Artin zero (root number −1) becomes a (s−½)^{1/4} branch point; root number +1 gives (s−½)^{−1/4} (slopes ±0.24997). The inert-prime factor O(σ) ~ C(2σ−1)^{1/2} → 0 as σ→½⁺ at every height: in Model A the line ½ is a square-root barrier of the norm, caused by lost phases, not zeros. No RH information.
- [open, speculative] Natural boundary of F (Estermann–Dahlquist–Kurokawa type) — read the primary sources and decide whether Re s = 0 is a natural boundary for F.
- [open] Same computation for binary tetrahedral/octahedral/icosahedral groups in H^× (non-cyclic Gal(K/F)?): which exponents appear, and does the line ½ always become a branch line of the aligned norm?
- [open] Fold a short subsection after #q8-frobenius-complex-s when the build is idle (closed form, monodromy, central exponent, figure).

## Binary tetrahedral Frobenius (2T = SL(2,3)) — 12 Nov 2026, ~8:30 AM PST (sprint)

- [closed — project, proved; see `tetrahedral-frobenius-20261112.md` §2, `tetra1112/`] "Is the reduced trace the only choice-free linear statistic?" Yes, for **every** finite non-abelian G ⊂ H^× (binary dihedral incl. Q8, 2T, 2O, 2I): the span of G is H and G/{±1} ⊂ SO(3) fixes no axis (Theorem 1). Numerically for 2T: constraint singular values 16.97×3, 0.
- [closed, answer partly negative — project, proved + exhaustive] "Does the fixed-convention pure part decompose into proper subquotients with zero ρ-weight?" For 2T: ρ-weight zero is possible (168 of 1536 conventions, incl. the aligned one) but **never** only proper subquotients: the faithful cubic twists ρ⊗ω, ρ⊗ω̄ always appear (Theorem 2: pure parts of the classes 6a and 3a = −6a are disjoint). Aligned convention: Im g_p^k = Im χ_{ρ⊗ω}(Frob^k)·n + ε_k[4a, k odd]·u, 1_{4a} = ¼(1+ω+ω̄−ψ3); coefficients a_ρ = 0, a_{ρ⊗ω} = −(i/2)n, a_{ρ⊗ω̄} = (i/2)n, a_1 = a_ω = a_ω̄ = u/4, a_ψ3 = −u/4. Fields LMFDB 2.26569.24t7.b.a / .a.a (N = 163², ε = −1, indicator −1); Re(Frob^k) = ½χ_ρ to 9e-16 vs PARI a_n; explicit-formula prime sums quaternion vs PARI agree to 1.2e-16.
- [closed — project, exact local identity + float numerics] The half-L item for 2T: with I = order-6 axis, F₆ = L(ρ)^{1/2}[L(ρ⊗ω)/L(ρ⊗ω̄)]^{1/2}[ζL(ω)L(ω̄)/L(ψ3)](2s)^{1/8}·H, H zero-free on Re s > ¼ (checked to 2.4e-9 at σ = 2, 1.3e-6 at σ = 3/2). All exponents real, so |F₆| is single-valued; central point (s−½)^{3/8} (slopes 0.37495 / 0.37501; monodromy 3π/4). The order-4 primes give the Model A norm factor O₄ ~ (2σ−1)^{1/6} (t ≠ 0; ¼ at t = 0): the line ½ is still a barrier of the norm, but a sixth root. General law β = Σ δ_C sin²θ_C sin²α_C (Q8: ½); 2T minimum over axes (9−√33)/24.
- [closed — project] Overnight cross-checks: P6 at p = 11 brute force agrees with the proof (P_O, P_Λ, 26 = 2p+4 lattices). Q at p = 7 method B stopped unfinished at 10:59 PM PST Nov 11 after 3 of 5 levels (99 of 211 lattices), matching method A on every completed level; it is a partial check only.
- [open — project, new] Zero side for the 2T pure part: rebuild the n-direction prime sum from PARI zeros of L(ρ⊗ω) (conductor 163 for b.a) and the u-direction from zeros of ζ, L(ω), L(ω̄), L(ψ3), as done for Q8 in 3(e).
- [open — project, new] 2O (binary octahedral, LMFDB 2.?.48t? / stem 8T23-type) and 2I (binary icosahedral: both 2-dim reps quaternionic with real characters, Galois-conjugate over Q(√5)): which subquotients and twists carry the fixed-convention pure part, and what is min_I β? [speculative] β_min > 0 always for non-abelian G.
- [closed 12 Nov 2026, ~4:30 PM PT — known, LMFDB API] The conductor-163 twists of 2.26569.24t7.b.a are LMFDB 2.163.8t12.a.a and 2.163.8t12.a.b (orbit 2.163.8t12.a; same stem field 8.0.705911761.1; smallest conductor of any 2-dimensional representation with projective image A4, as recorded by LMFDB). For 2.26569.24t7.a.a the twists are 2.26569.8t12.a.a / .a.b. JSON: tetra1112/sources/. The HTML search page is still behind a reCAPTCHA from this host; the API is not.
- [closed 12 Nov 2026, ~4:30 PM PT — project, proved (computer-assisted), plus known non-vanishing on Re s=1] 1/6 barrier (1/4 at height 0), general law β=Σ δ_C sin²θ_C sin²α_C, and min_I β=(9−√33)/24, by exact local identities and an explicit K majorant |log K|≤9.305 r³ (r≤1/2). The 3/8 central exponent is the same identity together with the simple pole of ζ(2s) and the simple zero of L(ρ), provided L(ρ⊗ω,½)≠0. That non-vanishing is now an interval certificate for both 2T fields (§6 of the note; |L(½)| > 1.8283 for 2.163.8t12.a.a / .a.b and |L(½)| > 0.72674 for 2.26569.8t12.a.a / .a.b). Scripts: tetra1112/certify_exponents.py, certify_lhalf.py; page #binary-tetrahedral-exponents.
- [closed 12 Nov 2026, ~5:25 PM PST — project, computer-assisted, interval arithmetic] L(ρ⊗ω, ½) ≠ 0 for both 2T fields, by Arb/FLINT enclosures in tetra1112/certify_lhalf.py: |L(½)| > 1.8283 for the conductor-163 twists 2.163.8t12.a.a / .a.b, and |L(½)| > 0.72674 for the conductor-26569 twists 2.26569.8t12.a.a / .a.b. This removes that hypothesis from the 3/8 exponent. Still open: an interval enclosure of L'(ρ, ½) ≠ 0.
- [open — project, new] Q at p = 7 method B: finish or re-run (needs ~10+ h at the observed rate) if a full independent check is still wanted.

## Barrier exponent for all non-abelian G ⊂ H^× — 12 Nov 2026, ~5:00 PM PST (sprint; see `binary-polyhedral-barrier-20261112.md`, `beta1112/`)

- [closed — project, proved] "[speculative] β_min > 0 always for non-abelian G": true, by Jordan's lemma (a proper cyclic subgroup cannot meet every class). Also c = Σδ_C sin²θ_C = 3/4 for all such G.
- [closed — project, proved] Binary dihedral: β_min = ¼ (n odd), ½ − ¼cos(π/n) (n even).
- [closed — project, computer-assisted, interval arithmetic] 2O: min β = 0.14871 (root of 192β³−288β²+126β−13); 2I: min β = 0.06977 (degree 6 over Q). The values are exact at a critical point and are certified global minima by an Arb/FLINT branch-and-bound on the three positive cube faces (Lipschitz bound 3/4 and, on cells with a unique active set, β ≥ 3/4 − λ_max(M)). Script: beta1112/certify_global.py; note §6. The [speculative] "2O smaller than 2T" is false (0.1487 > 0.1356). The universal floor β ≥ β_{2I} for every finite non-abelian G ⊂ H^× is now unconditional (classification of finite subgroups of H^×).
- [closed 12 Nov 2026, ~5:35 PM PST — project, computer-assisted, interval arithmetic] Global optimality for 2O/2I, by the interval branch-and-bound just above. No leftover boxes; enclosure width < 10^{-63}.
- [closed 12 Nov 2026, ~10:15 PM PST — project, float numerics] Field-level check for 2O and 2I: galpol real Galois fields (LMFDB HTML/API blocked from this host; 12 LMFDB 24T201 records confirm the 2I group only). 2O is [48,28], not GL(2,3). Aligned Model A last-decade slopes 0.1509 vs 0.1487 (2O, height 11.3, X=20 000) and 0.0688 vs 0.0698 (2I, height 23.5). 2O has a simple forced zero of ρ (PARI slopes → 1) and L(ρ⊗ε,½)≈4.735; 2I L(ρ,½) was not evaluated (character field ℚ(√5), galoisinit fails). Note `polyhedral-fields-20261112.md`; page #binary-polyhedral-fields.


## Function-field / Weil analogue of the quaternion half-product — 12 November 2026, ~11:00 PM PST (Grant pt228u)

- [closed for the four exact questions on Q8, with one item left open — project, exact series + point counts; known input: Weil 1948] Q8 covers of P^1 were built over F_3, F_5, F_7 and F_9 (Bolza y^2=s^5-s when q≡1 mod 4; the sextic y^2=s^6+s^4+s^2+1 when q≡3 mod 4). L(ρ,u) is a degree-2 polynomial, double-checked by point counts and PARI; all roots lie on |u|=q^{-1/2}. Only the F_5 Bolza cover has a central zero (root number −1). The commuting-variable Euler product over the F_5 cover equals 1−5u^2 coefficient-wise through degree 5 (827 unramified places; Chebotarev χ²=0.37 on 4 df; the two-halves-plus-inert identity holds to the same degree). The barrier exponent is exactly ½, matching the number field, and the Model A barrier is the whole critical circle. The main term of the half-product is algebraic in u; whether the remainder G is algebraic was not decided (series only through degree 5). A 2T cover was not computed. Note `function-field-half-product-20261112.md`; figure `figures/function-field-half-product-site.svg`. Not folded onto the page in this sprint.
- [closed — folded 12 Nov 2026, ~4:50 PM PST] β table and figure now live on /critical-line-hypercomplex#binary-polyhedral-beta (just after #binary-tetrahedral-exponents).


## 2T central simplicity — 12 Nov 2026, ~8:50 PM PST (sprint; see `tetra-central-simplicity-20261112.md`)

- [closed — project, computer-assisted, interval arithmetic] "Still open: an interval enclosure of L'(ρ, ½) ≠ 0" for the 2T fields: L′(ρ,½) ∈ (2.62, 2.85) for 2.26569.24t7.b.a and (2.97, 3.07) for 2.26569.24t7.a.a; both forced central zeros are simple. The 3/8 exponent is now hypothesis-free for both fields. On page (#binary-tetrahedral-exponents), version 2ed2cec9.
- [closed 12 Nov 2026, ~10:15 PM PST — project, float numerics] Field-level 2O/2I check, as just above (galpol fields; LMFDB blocked). Slopes agree with 0.14871 / 0.06977 to a few thousandths at X = 20 000. Page #binary-polyhedral-fields.
- [open — project] Zero side for the 2T pure part (rebuild the n-direction prime sum from zeros of L(ρ⊗ω), conductor 163).
- [open, speculative] Natural boundary of the aligned products F (Q8) and F₆ (2T).
- [open — project] Q at p = 7 method B (independent cross-check only; ~10 h).


## 2T pure part: zero side — 13 Nov 2026, ~8:00 AM PST (sprint; see `tetra-pure-zero-side-20261113.md`, `tetra-zero1113/`, `figures/tetra-pure-zero-side.png`)

- [closed — known theorem (Weil 1952; Iwaniec–Kowalski Thm 5.12, non-self-dual form), project numerical check, double-computed, float not interval] "Zero side for the 2T pure part" for 2.26569.24t7.b.a and .a.a. The u-direction (¼(ζ + L(ω) + L(ω̄) − L(ψ3))) and the n-direction (zeros of L(ρ⊗ω), L(ρ⊗ω̄)) are rebuilt from PARI zeros to height 110. Gaussian T = 1…20 and cos-/sin-shifted families: |zero − (arch + pole + prime)| ≤ 1.0·10⁻¹⁰ at T=1 (prime-truncation budget 6·10⁻⁹) and ≤ 7·10⁻¹⁵ (u), ≤ 3·10⁻¹⁶ (n) for T ≥ 1.5; shifted families ≤ 5·10⁻¹⁴. mpmath vs scipy agree to ≤ 8·10⁻¹⁴; the zero-cutoff dependence at T = 16, 20 matches the density estimate.
- [closed — project, exact + numerics] The n-direction needs an **odd** test function: it is (zeros of L(ρ⊗ω) with Im>0) − (zeros with Im<0), the asymmetry of a non-self-dual L-function; an even Gaussian sees only Re Λ and is blind to it. Example (b.a): T=1 upper +0.9397, lower −1.0218, sum −0.0821; T=8 +5.8395 vs −5.8387. Coefficients −i/2, +i/2 combine to the real number (H_π − H_π̄)/2 = H_π. Λ_{ρ⊗ω}(163^k) is real, so no ramified correction.
- [closed — project] The central zero of L(ρ) contributes nothing to the pure part (a_ρ = 0 exactly by character inner products; the pure part balances without any ρ zeros), but supplies 49%/95% (T=1, b.a/a.a) → 1.7% (T=20) of the real part, whose identity breaks by exactly 1 without it.
- [open — project, small] The k ≥ 2 / ramified correction Δ_u of the u-direction (4.8·10⁻⁴ at T=1, 0.405 at a window on p = 163) is computed from primes only, not from zeros; an L(2s)-type zero-side interpretation is not attempted.
- [open — project] Certify the zeros used (PARI is uncertified) and replace the E = 10 margin by a proved explicit N(T) bound; both fields only to height 110, T ≤ 20 (odd sgn·Gaussian family only T ≤ 8).
- [open — project, new] Same bookkeeping for 2O and 2I (fields from `polyhedral-fields-20261112.md`), where the character fields are Q(√2), Q(√5).
- Current short open list: Q at p = 7 method B (independent check, ~10 h); [speculative] natural boundary of F (Q8) and F₆ (2T); fold the 2T explicit-formula subsection on the page when the owner approves; this note's zero-side subsection awaits the same decision.
- [closed — site, 13 Nov 2026 ~8:10 AM PST, Grant request] Folded onto the live page: /critical-line-hypercomplex#binary-tetrahedral-pure-zero-side ("Where the extra quaternion direction lives"), after #binary-polyhedral-beta, with the coefficient table, the T = 1 example (+0.940 / −1.022 / −0.082), the residual summary and a site-style figure (panels b and c). Labels [known]/[project, float numerics]; PARI zeros are uncertified floats; no RH claim. Still open: the same method for 2O and 2I; odd-family runs at higher T; zeros beyond height 110; a certified zero list.


## 2O central L-values — 13 Nov 2026, ~10:15 AM PST (routine sprint; see `octa-central-certificate-20261113.md`, `octa1113/`)

- [closed — project, computer-assisted, interval arithmetic] "Arb enclosure of L′(ρ, ½)" for the 2O galpol field (conductor 924 800, Γ_ℝ(s)²): **L′(ρ, ½) ∈ 1.94965615008486 ± 4.6·10⁻¹⁵**, so the forced central zero is simple. The sign twist has **L(ρ⊗ε, ½) ∈ 4.734975816857929 ± 5.1·10⁻¹⁶**, so it does not vanish. Root numbers −1 / +1 are measured from θ(½)/(2θ(2)) = ∓1 ± 2·10⁻¹⁵. Method: exact termwise Mellin integrals via chained `acb.integral` (the 2T Riemann-sum hulls are too wide at this conductor). Two runs (160-bit, c·n ≤ 120 and 224-bit, c·n ≤ 130) agree with each other and with PARI floats. Coefficients are audited against PARI aₙ (n ≤ 40 000), residue degrees (4 200 primes) and (·/5). Inputs: Langlands–Tunnell (Λ entire), PARI conductor and local factors.
- [closed 13 Nov 2026, ~12:15 PM PST — project, proved (computer-assisted identity) plus interval certificates, with one numerical hypothesis] The 2O analogue of the 2T 3/8 identity: along an 8-fold axis, F₈ = L(ρ)^{1/3} L(ρ⊗ε)^{−1/6} L(σ₄)^{1/6} · (S₄ L-functions to imaginary and 2s-powers, including ζ(2s)^{1/8−i/8}) · H, with H holomorphic and zero-free on Re s > 0. The simple zero of L(ρ) and the pole of ζ(2s) give the exponent 5/24, provided L(σ₄, ½) ≠ 0. That non-vanishing is a PARI float (L(σ₄, ½) ≈ 1.201, root number +1 from an L(2) comparison), not an interval certificate. Measured slope 0.208329 against 0.208333. Note `octahedral-exponent-20261113.md`; scripts `octa-exp1113/`. Page #binary-octahedral-exponent.
- [closed — project, computer-assisted, interval arithmetic, 13 Nov 2026 ~8:55 PM PST] Arb enclosure of L(σ₄, ½) and a rigorous root number for σ₄ (conductor 2 959 360 000). With the correct Euler factor 1/(1+17^{−2s}) at 17 (the older product a_n(ρ)a_n(χ₂) is wrong there; the float 1.201 is withdrawn): **W(σ₄)=+1** because S=∑ a_n I_{1/2}(z_n) lies in 6100±91 and excludes 0; **L(σ₄, ½) ∈ [0.9368394678458571, 0.9510108707472682]**, which excludes 0. Script `cert1113/certify_sig4.py`; note `sig4-cert-2I-search-20261113.md`. The 5/24 exponent is now on this certificate (L(σ₄) entire unconditionally). No RH claim.
- [open] 2I central values (character field ℚ(√5); coefficients would come from the cached automorphisms, since `galoisinit` fails).
- [closed — site, 13 Nov 2026 ~10:20 AM PST] Folded into /critical-line-hypercomplex#binary-polyhedral-fields (one sentence, tag [project, computer-assisted, interval arithmetic], no 2O exponent claimed), version c6aa1494.
- [closed — site, 13 Nov 2026 ~12:15 PM PST] Folded as /critical-line-hypercomplex#binary-octahedral-exponent, with the identity, the 5/24 exponent, the L(σ₄, ½) caveat, the slope, and a site figure.


## 2I central values — 13 Nov 2026, ~1:20 PM PST (routine sprint; see `icosa-central-20261113.md`, `icosa1113/`)

- [closed — project, float numerics] "2I central values" for the galpol degree-120 field (ramified 2, 11, 23). Frobenius quaternions now known at all 33 860 unramified primes < 400 000 (17 automorphism-denominator primes and 12 further index divisors done by a p-maximal-order test). Both faithful 2-dim characters ρ, ρ′ (conjugate over ℚ(√5)) have conductor 2⁸·11²·23² = 16 386 304 and root number **+1** (lfuncheckfeq −55/−51, nothing else in a = 2…12, ε = ±1 fits). L(ρ, ½) ≈ 5.8516, L(ρ′, ½) ≈ 0.5133: **no forced central zero in this field.** L(ρ, 2) matches the Euler product to 7·10⁻¹⁰.
- [closed — project] The forced central zero is a property of the field/character, not of the group: the same 2O field already had ρ: −1, ρ⊗ε: +1 [known in general for quaternion fields, Fröhlich 1972].
- [open] Arb certificates for L(ρ, ½), L(ρ′, ½) ≠ 0 and the 2I root numbers.
- [open] The 2I aligned-product identity (analogue of 2T 3/8, 2O 5/24) and its central exponent; with W = +1 it gets no forced-zero term [speculative until written].
- [open — project, local prediction not numerically confirmed] Find a 2I field with root number −1. Candidate: LMFDB 2.3806401.120, conductor 1951² = 3 806 401 (only ramified prime 1951 ≡ 11 (mod 20)). Every faithful order-10 character of F_{1951}^× has χ(−1)=−1; under the Tate identification W_p(ρ)=χ(−1) for this tame f=1 setting, W(ρ)=−1. Control: conductor 2141² (2141 ≡ 1 (mod 20)) has χ(−1)=+1. A theta test is not yet trusted, because (disc/p) is identically +1 (A₅, disc a square) and does not split the two order-5 / order-10 classes of 2I. Note `sig4-cert-2I-search-20261113.md`, script `cert1113/search_2I_oddW.py`.
- Still open from before: Arb enclosure of L(σ₄, ½) (2O); zero side for 2O/2I; Q at p = 7 method B; [speculative] natural boundary of F (Q8) and F₆ (2T).


## 2I aligned product and central exponent — 13 Nov 2026, ~5:05 PM PST (sprint; see `icosa-exponent-20261113.md`, `icosa-exp1113/`)

- [closed — project, proved (exact ℚ(ζ₆₀) computer algebra)] "The 2I aligned-product identity": along a 10-fold axis (the maximal cyclic C₁₀, which contains −1 and meets 6 of 9 classes), F₁₀ = L(ρ)^{1/3} L(ρ′)^{−1/6} L(Sym³ρ)^{1/6} · (A₅ L-functions to imaginary powers) · ζ(2s)^{1/8−0.154i} L(3,2s)^{−1/8−…} L(3′,2s)^{−1/8−…} L(5,2s)^{1/8} · (faithful L(·,2s) to imaginary powers) · H. Same real k=1 weights as 2O. Local factors match to ≤ 7·10⁻¹⁵, global products (33 857 primes < 400 000) to ≤ 6·10⁻¹⁴ at σ = 2, 1.5, 1.2. H is holomorphic and zero-free on **Re s > 1/3** (r_k periodic, poles of d log H at roots of unity).
- [correction — project] The 2O note/page say H is holomorphic for Re s > 0. The proof only gives Re s > 1/3; for 2I the class mean of r₃ is −0.235i ≠ 0, so H has a log branch point at s = 1/3 (and r₄ gives one at ¼). Not yet changed in the 2O note text or on the page (beyond this entry); harmless for s = ½.
- [closed — project, local constants + numerics] W(Sym³ρ) = −1 for this field (D₂ = Q₈, D₁₁ = D₂₃ = Dic₃; tame Gauss sums give W₁₁(ρ) = W₂₃(ρ) = −1; W₁₁(4′) = +1, W₂₃(4′) = −1, W₂(4′) = 1). Conductor 2¹⁶·11⁴·23², Euler factor (1+23^{−2s})^{−1} at 23. Confirmed numerically: θ(1/y)/(yθ(y)) = −1 to 3·10⁻⁹; L(2) matches only for W = −1. So L(Sym³ρ, s) has odd order at ½ (unconditional, Brauer FE).
- [closed — project, interval arithmetic, conditional on Artin's conjecture for ρ, ρ′] "Arb certificates for L(ρ, ½), L(ρ′, ½) ≠ 0": L(ρ,½) ∈ 5.851618798 ± 1.2·10⁻¹⁰, L(ρ′,½) ∈ 0.5132584851 ± 8.8·10⁻¹¹ (`certify_icosa.py`, 160 bits, 77 312 terms). Even icosahedral Artin conjecture is open [known, Booker 2006], so these are conditional.
- [closed — project, float + hypotheses] "2I central exponent": e = −1/8 + m_ρ/3 − m_ρ′/6 + m_{Sym³}/6 (proved), m_{Sym³} odd (proved). With Artin's conjecture for ρ, ρ′, Sym³ρ and L′(Sym³ρ, ½) ≈ 14.974 ≠ 0 (float): **e = 1/24**. The prediction −1/8 (from W(ρ) = W(ρ′) = +1 alone) is refuted by the forced zero of Sym³ρ. Closed-form slope 0.041647 vs 0.041667 (last interval 5·10⁻⁵ → 10⁻⁵).
- [open] Interval certificate of L′(Sym³ρ, ½) (conductor 5.1·10¹¹, Γ_ℝ⁴ kernel); an unconditional handle on ord_{½} L(ρ) without Artin's conjecture; L(3,½), L(3′,½), L(4,½) (imaginary part of the exponent); W₂(ρ_{Q₈}) locally; fix the "Re s > 0" wording for 2O/2T H.
- [open, from before] Interval certificate of L′(Sym³ρ, ½); 2I field with W(ρ)=−1 (1951² candidate, not numerically confirmed); zero side for 2O/2I; Q at p = 7 method B; [speculative] natural boundary of F (Q8) and F₆ (2T).
- Current short open list: L′(Sym³ρ,½) certificate · 2I field with W(ρ)=−1 (1951² candidate) · Re s > 1/3 wording fix for 2O/2T · A₅ central values for 2I · zero side for 2O/2I · Q at p = 7 method B · [speculative] natural boundaries.


## 2O pure part: zero side — 13 Nov 2026, ~9 PM PST (sprint; see `octa-pure-zero-side-20261113.md`, `octa-zero1113/`, `figures/octa-pure-zero-side.png`)

- [closed — known theorem (Weil 1952; Iwaniec–Kowalski Thm 5.12), project numerical check, double-computed, float not interval] "Zero side for 2O" (galpol [48,28], ramified 2, 5, 17). The Model A real part and all three pure directions (u: 4-fold, n: 3-fold, m: 2-fold axis) were rebuilt from PARI zeros to height 110 (L(χ₂): 100) of ζ, L(ε), L(χ₂), L(ψ₃), L(ψ₃⊗ε), L(ρ) (and L(ρ⊗ε) for the contrast convention). Gaussian T = 1…16 and cos-shifted T = 8, x₀ = 1…8: residuals ≤ 3.9·10⁻¹⁰ at T = 1 (prime-truncation budget 4·10⁻⁹), ≤ 6·10⁻¹⁶ for Gaussian T ≥ 1.5, ≤ 1.1·10⁻¹⁴ for the shifted windows. mpmath vs scipy (explicit quaternion products) agree to ≤ 4·10⁻¹⁴. Zero counts match the main-term density to within 1.
- [closed — project, proved (exact character inner products in ℚ(√2, √3))] Which L-functions carry the pure part. Model A, k = 1: only the S₄ L-functions. u = ((1+√2)ζ + (1−√2)ε + 2χ₂ − (1+√2)ψ₃ + (√2−1)ψ₃⊗ε)/8, which equals the imaginary F₈ exponents; n = (√3/6)(ζ + ε − χ₂); m = ¼(ζ − ε + ψ₃ − ψ₃⊗ε). L(ρ), L(ρ⊗ε), L(σ₄) get weight 0, so the central zero of L(ρ) feeds only the real part (share 0.40 at T = 1 → 0.016 at T = 16; the identity breaks by exactly 1 without it). Faithful L-functions enter the pure part only at k = 2 (u: (√2/4)(ρ − ρ⊗ε); n: (√3/6)(ρ + ρ⊗ε − σ₄)), i.e. through the L(·, 2s) factors. Proof: central parity sin k(π−θ) = (−1)^{k+1} sin kθ.
- [closed — project, proved] **No asymmetry channel in 2O, in any convention.** All characters are real, so every coefficient is real and every L is self-dual with zeros symmetric under γ ↦ −γ (no RH needed). Odd test functions give 0 = 0, and the 2T mechanism (n-direction = up/down zero asymmetry of the non-self-dual L(ρ⊗ω)) cannot occur. The same proof covers Q8 and 2I, so 2T is the only one of the four with that channel.
- [closed — project] Convention dependence: with "−g" representatives (C8− → −e^{πu/4}, C3 → −g₆) the n-direction becomes purely faithful, (√3/6)(ρ + ρ⊗ε − σ₄), and u gets ±¼ on ρ, ρ⊗ε (the central zero then enters u with weight ¼; u balances to ≤ 4.6·10⁻¹¹ at T = 1). The C4a/C4b parts are always S₄-only, so the pure part is never purely faithful. The u sign flips (−0.334 vs +0.267 at T = 1).
- [open — project, small] Δ (ramified p = 2, 5, 17 inside the S₄ L-functions, plus k ≥ 2 terms) is large here (Δ_u = 0.270, Δ_m = 0.313 at T = 1, mostly p = 2) and is computed from primes only. A zero-side reading of the k = 2 layer via L(·, 2s) was not attempted.
- [open — project] σ₄ zeros (conductor 2.96·10⁹), needed for convention B's n-direction; heights > 110; certified zero lists and a proved N(T) bound; the 2I numerics (the "no odd channel" proof already applies).
- Current short open list: L′(Sym³ρ,½) certificate · 2I field with W(ρ)=−1 (1951² candidate) · Re s > 1/3 wording fix for 2O/2T · A₅ central values for 2I · zero side for 2I (numerics only; 2O done, "no odd channel" proved for Q8/2O/2I) · σ₄ zeros for the convention-B check (2O) · Q at p = 7 method B · [speculative] natural boundaries.


## 2I fields with root number −1 — 14 Nov 2026, ~7:50 AM PST (sprint; see `icosa-oddW-20261114.md`)

- [closed — project, float numerics + local proof; conditional on Artin holomorphy] "Find a 2I field with W(ρ) = −1": **both** 2.3806401.120.a (1951²) and 2.4583881.120.a (2141²) have W = −1 (lfuncheckfeq 2⁻⁵⁵ / 2⁻⁵⁴; the other sign and the other Γ-factor fail). Mechanism: W_∞ = −1 iff complex conjugation is −1 ∈ 2I (2141); W_p = −1 for abelian D_p iff rec_p(−1) = −1 ∈ 2I (1951: inertia C₁₀, (p−1)/10 odd).
- [correction] 13 Nov "control 2141 has W = +1" and "1951 stem disc p¹⁶" were wrong (see note §4).
- [open] Arb certificates of L′(½) ≠ 0 for both (Γ_ℝ² / Γ_ℝ(s+1)² kernels, conductors ~4·10⁶); p = 2 blind spot of the oriented-discriminant test; aligned-product central exponent for these fields (ρ and ρ′ both W = −1); page fold after #binary-icosahedral-exponent.
- Current short open list: page fold of the 2I W = −1 result · Arb L′(½) for 1951/2141 · 2I aligned exponent for 1951/2141 · L′(Sym³ρ,½) certificate (galpol 2I) · A₅ central values for 2I · zero side for 2I (numerics) · σ₄ zeros (2O convention B) · Q at p = 7 method B · [speculative] natural boundaries.



## 2I forced central zeros: interval certificates — 14 Nov 2026, ~12:45 PM PST (routine sprint; see `icosa-oddW-certificate-20261114.md`, `icosa-cert1114/`, `figures/icosahedral-odd-sign-site.svg`)

- [closed — project, computer-assisted, interval arithmetic; conditional on Artin holomorphy] "Arb L′(½) for 1951/2141": L′(ρ, ½) ∈ 14.6404198415908494271272808 ± 4.9·10⁻²⁶ (2.3806401.120.a, Γ_ℝ(s)²) and 8.38821569820444695438668613 ± 3.7·10⁻²⁷ (2.4583881.120.a, Γ_ℝ(s+1)²). Both forced central zeros are simple. Two independent runs (cutoff 60 / 300 bits, 80 / 400 bits) agree; PARI floats within 8·10⁻¹⁶.
- [closed — project, same status] The √5-conjugates ρ′ also have W = −1 (PARI checkfeq 2⁻⁵⁰ / 2⁻⁴⁶; interval theta ratio) and L′(ρ′, ½) ∈ 3.7050867341305652169151417 ± 5.2·10⁻²⁶ (1951), 6.188781635851670992970410311 ± 8.9·10⁻²⁸ (2141). So each field has two simple forced central zeros (ρ and ρ′).
- [closed — project, interval arithmetic, under the same hypothesis] Root numbers −1 certified by θ(½)/(2θ(2)) = −1 ± 10⁻³³ (all four). Independently, the morning's local proof (Tate's ε(χ)ε(χ⁻¹) = χ(−1)) gives W = −1 without the hypothesis.
- [closed — project] Coefficient audit against the LMFDB projective quintic: cycle type mod p matches |a_p| at all 2 258 unramified p < 20 000 per field (0 mismatches). Ramified Euler factors are 1 (inertia C₁₀ / C₅ has no fixed vector).
- [correction/lesson — project] acb `bessel_k` loses ≈ 2.9·x bits near the cutoff x, so 160-bit runs with cutoff 120 (13 Nov 2O/2I enclosures) are valid but ~10⁸ times wider than needed. Use bits ≳ 2.9·cutoff + target.
- [open] 2I aligned-product exponent for 1951/2141 (now ρ and ρ′ both have certified simple forced zeros under H; Sym³ρ still to do); an independent audit of the sign within each ±-pair of a_p; p = 2 blind spot (lift to ℤ₂ mod 8); anything unconditional on ord_{½} L(ρ).
- [closed — site, 14 Nov 2026 ~12:45 PM PST] Page fold of the 2I W = −1 result: /critical-line-hypercomplex#binary-icosahedral-odd-sign (mechanism, interval table, figure), version bd8aba3e.
- Current short open list: 2I aligned exponent for 1951/2141 · L′(Sym³ρ,½) certificate (galpol 2I) · A₅ central values for 2I · zero side for 2I (numerics) · σ₄ zeros (2O convention B) · Q at p = 7 method B · [speculative] natural boundaries.


## Uniform edge bound for every finite G ⊂ H^× — 14 Nov 2026, ~2:00 PM PST (Grant pt235u; see `uniform-edge-20261114.md`, `edge1114/`, `figures/uniform-edge-site.svg`)

- [closed — project, proved; unconditional] "Does the edge stay within a fixed distance of ½ for every finite group of unit quaternions?" Along the standard axis (longest cyclic direction): yes, σ ∈ {1/4, 1/3}, so ½ − σ ∈ [1/6, 1/4], with ¼ attained by Dic₃ and 2T and 1/6 by Q8, Dic_n (n ≥ 4), 2O, 2I. σ_hol = σ_abs in every standard case (c₁ ≠ 0, non-integer).
- [closed — project, proved] General law: σ_abs = 1/k₀ (Chebotarev), H ≡ 1 iff G ⊂ C_I, otherwise 3 ≤ k₀ ≤ max(4, P(|G|)) (Jordan's lemma + a prime-power step). Over all axes the distance is < ½, sup ½ approached by Dic_p on a j-axis (σ = 1/p), never attained.
- [closed — project] The 2O/2T "Re s > 0" wording item: 2O's edge is 1/3, and it is a true branch point (c₁ = −i/12); 2T's is 1/4 (c₁ = −1/16). The page already says 1/3 for 2O; the 2O note keeps its erratum.
- [open] σ_hol when c₁ = 0 (Dic_p j-axis, generic axes of 2T/2O/2I): needs zero information for L(χ, k₀ s), χ ≠ 1; not determined. Natural boundaries of H remain [speculative].
- [closed — site, 14 Nov 2026 ~1:45 PM PST] Page fold: /critical-line-hypercomplex#uniform-edge (table, bound, figure #hyper-uniform-edge), version 555094f5.
- Current short open list: σ_hol when c₁ = 0 · 2I aligned exponent for 1951/2141 · L′(Sym³ρ,½) certificate (galpol 2I) · A₅ central values for 2I · zero side for 2I (numerics) · σ₄ zeros (2O convention B) · Q at p = 7 method B · [speculative] natural boundaries.


## 2I aligned exponent for the root-number −1 fields — 14 Nov 2026, ~9 PM PST (sprint; see `icosa-oddW-aligned-exponent-20261114.md`, `sym3-1114/`)

- [closed — project, proved from known local facts (Tate ε(χ)ε(χ⁻¹) = χ(−1)), unconditional] W(Sym³ρ) = +1 for 2.3806401.120.a and 2.4583881.120.a (D at p₀ abelian, so Sym³ρ|_D = ψ³+ψ+ψ⁻¹+ψ⁻³, local sign ψ(−1)⁴ = 1; W_∞ = 1 for both).
- [closed — project, double-precision floats, not interval; presupposes Artin holomorphy] L(Sym³ρ, ½) = 1.95738387069 (1951) and 0.0381348369208 (2141), both nonzero, converged over cutoffs, theta ratio 1 ± 2·10⁻⁹, L(2) matches Euler product. Also W = +1 and nonzero central values for ρ⊗ρ′ (3.3536269316 / 0.9346170744) and Sym⁴ρ (2.2524781612 / 3.9408979633).
- [correction] The 4:20 PM sprint's PARI numbers for 2141 (L(½) ≈ 0.0318–0.0382) were not converged; correct value 0.0381348369. 1951's 1.9576 stands.
- [closed — project, conditional on Artin holomorphy for ρ, ρ′, Sym³ρ and float L(Sym³ρ,½) ≠ 0] Aligned F₁₀ exponent for both fields: e = −1/8 + 1/3 − 1/6 + 0 = **1/24**, same as the galpol 2I field but from ρ, ρ′ (W = −1) instead of Sym³ρ (W = −1 there). Root-number patterns (−,−,+) here versus (+,+,−) there.
- [open — speculative] Is W(ρ)·W(Sym³ρ) = −1 forced for 2I fields?
- [open — project] Arb certificate of L(Sym³ρ, ½) ≠ 0 (heavy-tailed Γ_ℝ⁴ kernel, ~4·10⁷ terms); L(3, ½), L(3′, ½) (need oriented classes beyond p < 20 000); direct numerical check of F₁₀ for these fields.
- Current short open list: σ_hol when c₁ = 0 · L′(Sym³ρ,½) certificate (galpol 2I) and Arb L(Sym³ρ,½) for 1951/2141 · A₅ central values L(3,½), L(3′,½) · zero side for 2I (numerics) · σ₄ zeros (2O convention B) · Q at p = 7 method B · whether W(ρ)W(Sym³ρ) = −1 is forced [speculative] · [speculative] natural boundaries.


## W(ρ)·W(Sym³ρ) test on the 701 family — 15 Nov 2026 (sprint; see `w701-20261115.md`, `w701-1115/`)

- [closed — project, floats (lfuncheckfeq ≤ 2⁻⁴⁹, not interval) + local proof] "Is W(ρ)·W(Sym³ρ) = −1 on a fourth independent-looking family?" All eight 2I representations 2.1965604.120.a/b (2²·701²), 2.7862416.120.a/b (2⁴·701²), 2.31449664.120.a–d (2⁶·701²) have product **−1**: W(ρ) = +1 at 2² (W(Sym³ρ) = −1, float, θ ratio −1 ± 7·10⁻⁹), W(ρ) = −1 at 2⁴ and 2⁶ (W(Sym³ρ) = +1: float for 2⁴, θ ratio 1 ± 3·10⁻¹⁰; local proof only for 2⁶). The conjecture is **not refuted**; it is more evidence, not a general proof.
- [closed — project, local proof sketch for this family] Mechanism: the 8 records are quadratic twists of one (a_p ratios ±1); D₇₀₁ = C₁₀ abelian with ψ(−1) = +1; D₂ ≅ Dic₃ for the base, Sym³ρ|D₂ = θ₊ ⊕ θ₋ ⊕ ρ|D₂ (θ± unramified, Frobenius ↦ ±i), so W₂(Sym³ρ) = W₂(ρ); wild χ twists multiply W₂(Sym³ρ)/W₂(ρ) by χ₂(−1) (Tate), and the product formula for χ gives P = −1.
- [correction] Γ-type: ρ(c) = −1 ⇔ 0 real roots of the stem polynomial; 4 of the 8 records are even (Γ_ℝ(s)²). Using Γ_ℝ(s+1)² for all gave "neither sign passes" for those four; fixed.
- [correction] W(ρ) is **not** constant along a twist family when the twist shares primes with N: +1 at 2² but −1 at 2⁴, 2⁶; the product P is constant.
- [project, Euler factors] Sym³ρ for the base twist has a nontrivial Euler factor 1/(1 + 2⁻²ˢ) at 2 (inertia C₃ fixes a 2-dim subspace) and conductor exponent 2 there; wild twists: exponents 8, 12.
- [open] W(Sym³ρ) numerically for 2.31449664.120.a–d (N = 2¹²·701⁴, ~3·10⁸ coefficients) · uniform proof of P = −1 [speculative: reciprocity for −1 and the double cover 2I → A₅; observed ∏_v ρ(rec_v(−1)) = −1 ∈ 2I in the 1951 and 2141 cases] · a nonabelian-decomposition check on the galpol 2I field by local theory · exponent for the 701 family (expected 1/24 by the same pattern; not computed) · Arb certificates for the 701 forced zeros.
- Current short open list: σ_hol when c₁ = 0 · L′(Sym³ρ,½) certificate (galpol 2I) and Arb L(Sym³ρ,½) for 1951/2141 · A₅ central values L(3,½), L(3′,½) · zero side for 2I (numerics) · σ₄ zeros (2O convention B) · Q at p = 7 method B · uniform proof that W(ρ)W(Sym³ρ) = −1 for all 2I [speculative; proved for the 701 family by a local sketch, float-checked on 4 of 8] · W(Sym³ρ) for 701 at 2¹² · [speculative] natural boundaries.


## Zero side for 2I (numerics) — 15 Nov 2026, ~1:25 PM PST (routine sprint; see `icosa-pure-zero-side-20261115.md`, `icosa-zero1115/`, `figures/icosa-pure-zero-side.png`)

- [closed — project, proved (exact character inner products in ℚ(√5))] Which L-functions carry the 2I Model A pure part. k = 1: only the A₅ L-functions ζ, L(3), L(3′), L(4), L(5). 5-fold axis v: ((s₁+s₂)/5)ζ + ((s₂−s₁)/5)L(3) + (2s₁/5)L(3′) − ((s₁+s₂)/5)L(4), s_j = sin(jπ/5) (equals the imaginary F₁₀ exponents); 3-fold n: (√3/6)(ζ + L(4) − L(5)); 2-fold m: ¼(ζ − L(3) − L(3′) + L(5)). Real part ½L(ρ) exactly. Faithful L(ρ), L(ρ′), L(4′), L(6) have weight exactly 0 at k = 1 and return at k = 2 (v: (2s₂/5)ρ − ((s₁+s₂)/5)ρ′ + ((s₂−s₁)/5)(4′ − 6); n: (√3/6)(ρ + ρ′ − 4′)). All characters real ⇒ **no odd channel** (the 2O/Q8/2I proof, now with numerics); 2T remains the only one with an asymmetry channel.
- [closed — known theorem (Weil 1952; Iwaniec–Kowalski Thm 5.12), project numerical check, double-computed, float not interval; presupposes Artin holomorphy] Zero side for the 2I real part and the 5-fold direction. PARI zeros (exact ℤ[φ] coefficients at 288 bits) to height 70 (ζ: 110) of ζ, L(ρ), L(3), L(3′), L(4); Gaussian T = 1…12 and shifted T = 8, x₀ = 1…8: per-L residuals ≤ 3.9·10⁻¹⁰ at T = 1 (budget 10⁻⁸), ≤ 6.9·10⁻¹⁶ for Gaussian T = 1.5…8, 1.3·10⁻¹⁴ at T = 12, ≤ 1.1·10⁻¹⁴ shifted; real part 5.5·10⁻¹² (T = 1), v 1.1·10⁻¹⁰ (T = 1), both ≤ 6.2·10⁻¹⁵ afterwards. mpmath vs scipy (explicit quaternion products) ≤ 3.4·10⁻¹⁴. Zero counts match the main term within 1. Δ_v = 6.8·10⁻⁷ (T = 1), ramified part exactly 0 (weights sum to 0).
- [closed — project, floats (lfuncheckfeq 2⁻⁵⁰…2⁻⁵⁵)] New A₅ data for the galpol 2I field: L(3), L(3′), L(4) have conductor 2⁸·11²·23², L(5) has 2⁸·11⁴·23⁴ = 1.05·10¹²; all root number +1, no forced zero; Euler factors at 2: 1, 1, (1−2^{−s})^{−1}, (1−2^{−s})^{−2}; at 11, 23: (1+p^{−s})^{−1} (3, 3′), (1−p^{−2s})^{−1} (4), (1−p^{−s})^{−1} (5).
- [correction/lesson — project] PARI `lfunzeros` with double-precision a_n silently drops zeros above height ≈ 20 (88 of 183 for L(ρ) at 60); exact coefficients (ℤ[φ] pairs) and `set_real_precision_bits(≥192)` restore the full count.
- [open — project] Zero side of the 3-fold and 2-fold directions: needs L(5) zeros (conductor 1.05·10¹², PARI needs 3.1·10⁷ coefficients; only 5.4·10⁶ unramified p are classified). Δ_n = 0.024 and Δ_m = 0.45 at T = 1 (prime side; m dominated by p = 2). Also: independent proof of the 2-part of the conductors/Euler factors of 3, 3′, 4, 5; Artin holomorphy for even icosahedral ρ; k = 2 layer as a zero sum; why Q₂ + Q₃₊ ≈ PP in the 5-fold direction (7·10⁻⁷).
- Current short open list: σ_hol when c₁ = 0 · L′(Sym³ρ,½) certificate (galpol 2I) and Arb L(Sym³ρ,½) for 1951/2141 · A₅ central values L(3,½), L(3′,½) · L(5) zeros / 2I zero side for the 3-fold and 2-fold directions (needs 3·10⁷ Frobenius classes) · σ₄ zeros (2O convention B) · Q at p = 7 method B · uniform proof that W(ρ)W(Sym³ρ) = −1 for all 2I [speculative; proved for the 701 family by a local sketch, float-checked on 4 of 8] · W(Sym³ρ) for 701 at 2¹² · [speculative] natural boundaries.


## A₅ central values for 2I — 15 Nov 2026, ~5:30 PM PST (routine sprint; see `a5-central-values-20261115b.md`, `a5central1115/`, `figures/a5-central-values.png`)

- [closed — project, floats (PARI 160 bits with exact ℤ[φ] coefficients, checkfeq 2⁻¹⁶⁸…2⁻¹⁷⁷; not interval); conditional on Artin holomorphy] "A₅ central values L(3,½), L(3′,½)": all nonzero and positive. galpol (2⁸·11²·23²): L(3,½) = 2.664383358954442196, L(3′,½) = 1.001953204174685429, L(4,½) = 0.906342895582119996. 1951²: L(3,½) = 1.858292355500529175, L(3′,½) = 0.715002437311023169. 2141²: L(3,½) = 0.719557917198393369, L(3′,½) = 3.125663386896181999. W = +1 for all (local proof for 1951/2141: Sym²ρ|D = ψ²⊕1⊕ψ⁻², conductor p₀², Euler factor (1−p₀^{−s})^{−1}, Γ_ℝ(s)³).
- [closed — project] Double computation: 64-bit vs 160-bit agree to ≤ 2.3·10⁻¹⁶; independent Meijer-G theta sum (no PARI L-code) agrees to ≤ 3.1·10⁻⁹ (quadrature-limited); galpol L(4,½) = ζ_{K₅}(½)/ζ(½) for LMFDB 5.5.16386304.1 (a_p audit 0/2 259 mismatches) agrees to all 48 digits.
- [closed — project, conditional as in the 1/24 theorem] The proviso "provided L(3,½), L(3′,½), L(4,½) ≠ 0" is met for all three 2I fields: F₁₀ ∼ c(s−½)^{1/24 + i(s₁+s₂)/10}, c ≠ 0, ∞. The A₅ factors at s are positive on [½, 2] (no sign change at step 0.01) and only rotate c: φ_A₅ = +0.102, −0.406, +0.265 rad (orientation-dependent; swapping ρ↔ρ′ gives 0.261, −0.251, 0.026).
- [closed — project] p = 2 blind spot for 2141: the flipped a₂ also fails the functional equation of L(3), L(3′) (2⁻⁷), confirming a₂ = 0.618 by a second (A₅) L-function.
- [correction] 14 Nov note §4 "L(3), L(3′) vanish together since they are Galois conjugates" is not a known theorem; should have been [speculative]; moot now. Also: the oriented traces already reached p < 1.2·10⁵ (not 2·10⁴), enough for these degree-3 L-functions.
- [open] Arb certificates of these nine values (Γ_ℝ(s)³ kernel needs a Meijer-G enclosure, ~10⁵ terms); L(5,½) for galpol (conductor 1.05·10¹²) — not needed for F₁₀; direct numerical check of F₁₀ (including the phase of c) for the 1951/2141 fields.
- [closed — site, 15 Nov 2026 ~5:05 PM PST] Page fold: /critical-line-hypercomplex#binary-icosahedral-a5-central (table, figure #hyper-a5-central-values), after #binary-icosahedral-sign-product; version 942fbf00.
- Current short open list: σ_hol when c₁ = 0 · L′(Sym³ρ,½) certificate (galpol 2I) and Arb L(Sym³ρ,½) for 1951/2141 · Arb certificates for the A₅ central values (Γ_ℝ³ kernel) · L(5) zeros / 2I zero side for the 3-fold and 2-fold directions (needs 3·10⁷ Frobenius classes) · σ₄ zeros (2O convention B) · Q at p = 7 method B · uniform proof that W(ρ)W(Sym³ρ) = −1 for all 2I [speculative; proved for the 701 family by a local sketch, float-checked on 4 of 8] · W(Sym³ρ) for 701 at 2¹² · direct numerical check of F₁₀ for 1951/2141 · [speculative] natural boundaries.


## Direct numerical check of F₁₀ for 1951/2141 — 15 Nov 2026, ~9:30 PM PST (routine sprint; see `F10-direct-check-20261115.md`, `f10direct1115/`, `figures/f10-direct-check.png`)

- [closed — project, floats (PARI 160 bits, exact ℤ[φ] coefficients; double theta sums for conductor p₀⁴; not interval); conditional on Artin holomorphy] "Direct numerical check of F₁₀ for the 1951/2141 2I fields, including the phase of c". Identity (global L-values via functional equations + H) vs the direct Euler product of F₁₀: |Δ log F₁₀| = 1.1·10⁻⁵ / 2.2·10⁻⁵ (σ = 1.2), 2.7·10⁻⁷ / 7.4·10⁻⁷ (1.5), 6.0·10⁻¹⁰ / 2.2·10⁻⁹ (2), each at the orientation-averaging noise level. Along s = ½ + ε, ε = 10⁻¹…10⁻⁶: fitted modulus slope 0.0416666645 (1951) / 0.0416663923 (2141) vs 1/24; phase slope 0.1538841938 / 0.1538842070 vs (s₁+s₂)/10 = 0.1538841769; last local slopes within 2·10⁻⁶ / 5·10⁻⁶ (modulus), 4·10⁻⁶ / 6·10⁻⁶ (phase), shrinking linearly in ε.
- [closed — project, floats] Constant c (file orientation, symmetric branch): |c| = 2.13255132, arg c = +0.0092020 rad (1951); |c| = 0.78493939, arg c = +0.2233186 rad (2141). Closed-form limit and Richardson extrapolation agree to ≤ 1.7·10⁻¹⁰. Swapped orientation: |c| = 1.07281, arg c = −0.15023 (1951); 0.67422, −0.22010 (2141). Estimated accuracy ~2·10⁻⁶ in |c|, ~10⁻⁵ rad in arg c.
- [clarification — project, elementary] F₁₀ has a logarithmic branch point at s = 1 (ζ(s)^{i(s₁+s₂)/5}); on (½, 1) |c| is fixed only up to e^{±π(s₁+s₂)/5} = 2.630^{±1} (upper / lower path: 5.608 / 0.811 for 1951, 2.064 / 0.298 for 2141). arg c and the exponent 1/24 + iγ are branch-independent.
- [clarification] φ_A₅ (−0.406, +0.265) is one of six contributions to arg c, not arg c itself: + ζ(2s) 0.1067, + ζ(s) 0.1165, + 2s-pieces, + arg H(½) (−0.031 / −0.177). The 15 Nov sentence "the A₅ pieces only turn c by a fixed angle" is right for L(3), L(3′), L(4) only.
- [closed — project] New local datum: D = I = C₁₀ at 1951 and D = I = C₅ at 2141 (degree-24 stem field), so L(Sym⁵ρ) has Euler factor 1 at 1951 and (1 − 2141^{−s})^{−2} at 2141. log L(Sym⁵ρ, 1) = −0.90732 / +0.44286 (truncated Euler product, phase effect ≤ 6·10⁻⁶ rad).
- [open] Interval version of any of this (needs Arb for Γ_ℝ(s)³, Γ_ℝ(s)⁴, Γ_ℝ(s)⁵ kernels and a rigorous H tail); L(Sym⁵ρ, 1) by a functional equation (conductor 1951⁶ ≈ 5.5·10¹⁹ out of reach; 2141⁴ feasible); oriented Frobenius classes beyond 1.2·10⁵ (would remove the orientation averaging); the same direct check for the galpol 2I field (its Sym³ρ has the forced zero, so the closed form uses L′(Sym³ρ, ½)).
- [closed — site, 15 Nov 2026 ~9:35 PM PST] Page fold: /critical-line-hypercomplex#binary-icosahedral-f10-direct (table, figure #hyper-f10-direct-check), after #binary-icosahedral-a5-central; version id in /workspace/math/DEPLOY-LOG.md.
- Current short open list: σ_hol when c₁ = 0 · L′(Sym³ρ,½) certificate (galpol 2I) and Arb L(Sym³ρ,½) for 1951/2141 · Arb certificates for the A₅ central values (Γ_ℝ³ kernel) · direct F₁₀ check for the galpol 2I field · L(5) zeros / 2I zero side for the 3-fold and 2-fold directions (needs 3·10⁷ Frobenius classes) · σ₄ zeros (2O convention B) · Q at p = 7 method B · uniform proof that W(ρ)W(Sym³ρ) = −1 for all 2I [speculative; proved for the 701 family by a local sketch, float-checked on 4 of 8] · W(Sym³ρ) for 701 at 2¹² · [speculative] natural boundaries.
