# Refinement run, Wednesday 11 November 2026 (9:22 AM PST start)

Daily refinement run for Zeta Zeros. Scripts, logs and JSON for the hyper-critical test are in `refine1111/` beside this note. No external messages.

## 1. Hyper-critical test: independent re-count of all 16 octonion orders containing Z^8

**Status: [project · proved, computer-assisted, independently re-computed].** Distinct from the 7:30 AM sprint, which computed one order of each kind with two programs (methods 1 and 2 in `inter1111/`). This run uses a third program written from scratch (`refine1111/recount.py`). It shares only the multiplication-table convention (Fano lines {i, i+1, i+3} mod 7, e_i e_{i+1} = e_{i+3}) and none of the enumeration or counting code.

- **Orders, by a different route.** The program closes G = Z^8 plus each half-vector under multiplication (rejecting anything that leaves ½G), then repeats with joins until nothing new appears. It does not sweep all 309,193 subspaces. Result: 16 orders with index over G 1, 2, 4×7, 16×7, matching the sprint.
- **Ideal counts for every order, not one per class.** Left and right S-stable sublattices of index 2^k, k ≤ 16, for all 16 orders (32 counts). Algorithm: a composition-series BFS. Children of a stable L are its maximal stable sublattices, i.e. kernels L → L/N for minimal submodules of the F_2-dual of L/2L (cyclic-closure test). Duplicates are removed with my own integer HNF. Logs: `run-K16-G-S1.txt`, `run-K16-rest.txt`; JSON `recount-K16-*.json`.
  - G: 1, 1, 7, 7, 15, 15, 35, 15, 31, then period 4 (15, 35, 15, 31) to k = 16, left = right.
  - S1: 1, 2, 15, 35, 15, 16, then period 4 (15, 35, 15, 16) to k = 16, left = right.
  - All seven R-type orders: 1, 0, 3, 0 repeating, left = right.
  - All seven maximal orders: 1 at k ≡ 0 mod 8, else 0, left = right.
- **Bound that makes the counts decisive** (`conductor_indep.py`, PARI HNF/SNF of the Z-algebra spanned by words in the multiplication matrices, own closure loop): [M_8(Z) : A_S] = 2^96, 2^68, 2^40, 1 and c = 3, 2, 1, 0 for G, S1, R-type, maximal (all 16 orders, left and right). A primitive stable lattice contains 2^c S, so its index is ≤ 2^{8c}. With counts to 2^16, the factor is therefore determined exactly for S1 (8c = 16), R-type (8) and maximal (0). `factors.py` / `factors-check.json`: every one of the 32 count sequences equals the series of the claimed factor.
- **Containment pattern** (`extras.py`, `extras-run.txt`): each R-type order lies in exactly 3 maximal orders and each maximal order contains exactly 3 R-type orders (Fano-plane-like incidence), as the sprint observed.
- **Zeros** (`extras.py`): gcd(Q_S1, Q_S1*) = 1 and gcd(1+3x², its reflection) = 1 again (sympy, exact), so there are no zeros on Re s = ½ for S1 or R-type orders. S1 zero real parts −0.567 (×2), 1.0551, 1.9431 (×2) reproduced.
- G to 2^24: see section 3 (the run status is recorded there).

What this adds: the sprint's "R and conjugates / maximal orders" claims relied on symmetry plus one representative. Now every order has been counted directly by independent code, and the conductor bound has been recomputed independently. What it does not add: an outside referee, a conceptual proof of c(S1) = 2, or any RH content (these are finite Euler factors at the single prime 2).

## 2. New main-site visual: zeros of the partial sums ζ_N

`zeta-zeros-build/partialsums/` (compute.py, svg.py, partialsums.json) → `img/partial-sum-zeros.svg`, section `/zeros-on-the-critical-line#partial-sum-zeros` (figure `#partial-sum-zeros-figure`). It is also on the home-page gallery ("Forty-eight figures").
- Every zero of ζ_N(s) = Σ_{n≤N} n^{-s} with 0 < t ≤ 60, for N = 10, 30, 100: 22, 33, 44 zeros. Method A: numpy grid and Newton. Method B: mpmath 40-digit findroot from A, with relative residual < 1e-30; A and B agree to 5e-14. The argument principle on [−N−2, 1.8]×[0, 60] gives exactly 22, 33, 44. (−N−2 is safe because all zeros have σ > −N, Gonek–Ledoan Thm 1.)
- Observed: each of ζ's 13 zeros below 60 has a ζ_N zero within 0.40, with real parts 0.24–0.74, so off the line. The remaining 9/20/31 zeros form a band that moves right toward σ = 1 as N grows. Totals match (T/2π) log N (21.99, 32.48, 43.98).
- Cited [known]: Turán 1948; Montgomery 1983; Montgomery–Vaughan 2001 (constant 4/π − 1); Platt–Trudgian 2016 (no zeros with σ > 1 exactly for N ≤ 18 and N = 20, 21, 28); Gonek–Ledoan 2010 (N_X(T) = (T/2π) log X + O(X)). The Euler–Maclaurin "correction nudges zeros" sentence is labelled as our reading.
- PNG for Grant: `/workspace/math/partial-sum-zeros-20261111.png` (static render, animation removed).

## 3. G to 2^24
Done: `python3 recount.py 24 0` (28 s), log `run-K24-G.txt`, JSON `recount-K24-0.json`, check `g24-check.txt`. Left and right counts for G to index 2^24 equal the series of Q_G(x)/(1−x⁴) exactly. With c(G) = 3, 2^24 is the full primitive range, so this is an independent confirmation of the Block 3/4 Gravesian factor. The page line "not yet independently re-checked" for G and R is replaced by a pointer to `#octonion-orders-recount`.

## 4. Build / deploy
- Edited: `build.py` (one exec line; backup `build.py.bak-partialsums`), new `partialsums_page.py`, `hyper_critical_page.py` (backup `hyper_critical_page.py.bak-refine1111`).
- check.py: NO ERRORS (13 pages, 1864 local refs).
- Deployed version 706a5dd4-315a-480e-8938-7652a729cdc8 (9:50 AM PST); live 200 with #partial-sum-zeros-figure and #octonion-orders-recount.

## Open / not done
- No outside referee for any of the octonion factors; conceptual proofs of c(S1) = 2 and the period-4 normalisation are still open.
- The "Gorenstein-type" adaptation was not touched in this run (it remains [speculative]).
