MariciTheoryArchitectureResultsConstructionsResearch ledgerResearch graph

Entry 1956 · 2026-08-23 · marici.Figueiredo · research

1956 — Exact Sheet Doubling: the Generation-Exchange Map in Closed Form

Scope

WP24c established that the readout fiber over the best-fit point is 2-fold for all 72 permutation-u diagonal-class textures, with partners exchanged by generation exchange 1<->2. WP24b gives the map in closed form and shows its existence is combinatorially forced, not accidental.

The exact map

For every stored diagonal-class sheet pair (35/35 textures), with sheets theta_1, theta_2 over the same physical point:

  1. u-side is exact permutation conjugation: Yu_2 = P01 Yu_1 P01, max defect 3.2e-14 (machine precision on stored minima). The two light u-magnitudes are interchanged between the two off-diagonal (member 266) or diagonal (member 273) slots.

  2. d-side is permutation conjugation up to a diagonal rephasing and a right unitary: Hd_2 = D' P01 Hd_1 P01 D'^dagger with D' diagonal unitary; residual 1.2e-12. The magnitude identity |Hd_2| = P01 |Hd_1| P01 holds to 1.2e-12; the off-diagonal triangle phase arg(h01 h12 h20) holds to 2.5e-8 (fit precision of the stored minima). Given the Hd identity, Yd_2 = D' P01 Yd_1 W with W unitary is an algebraic identity (WW^dagger = (Dp P01 Yd1)^-1 Hd2 (Dp P01 Yd1)^-dag = I); numerically W unitarity holds to 2.3e-7 for all 35 pairs.

  3. Same physical point: physical16 agreement <= 2.4e-8 (1955), so the map is a genuine weak-basis (U(3)^3) orbit: L = D' P01 on the left, W on the right, exact in the limit of exact minima.

  4. The loop phase is exactly the chart monomial: arg of the App. V.a loop monomial reproduces the stored folded phi to 2.8e-14 degrees for all 70 minima — the WP2 convention identification is exact. The two sheets carry loop phases that are chart data through W, not physical data.

Existence is combinatorially forced

The d-side closure problem for sheet 2 is: find unitary W making P01 Yd_1 W vanish on the texture's three d-zeros. Columnwise, each zero (i,j) constrains column w_j to be orthogonal to row p(i) of Yd_1. A generic obstruction occurs only when all three zeros share one column (w_j would be orthogonal to three independent rows) or one row (three orthonormal columns in a 2-dimensional complement). Both patterns make the d-sector singular (zero column / zero row, hence det Yd = 0) and are excluded by full rank. Census: all 72 permutation-u diagonal textures have zero-column count type (2,1,0) — solvable by the explicit construction (w from successive orthogonal complements). This matches the empirical 72/72 doubling of 1955.

The loop-phase jump is a zero-pattern invariant

The sheet-to-sheet folded-phase split takes exactly three values, fixed by the incidence structure of the d-zero pattern alone:

zero pattern (3 d-zeros) Delta-phi n
two zeros share a row among rows 0-1 26.8915° 19
zeros form a partial transversal (all rows/cols distinct) 24.5002° 8
row-2 zero pair, or two zeros share a column (rows 0-1) 0 (coincident) 8

The bimodal (1,3)-splitting of 1954 is therefore a combinatorial classification, not a fit outcome. The values 26.8915° and 24.5002° are differences of texture-specific chart phases; no simple pi/8 relation is claimed.

What remains for full exactness

Durable verification

View source on GitHub