back to table · edit · history · where entries came from · files · download
7966 bytes, as of the version from 2026-11-17 11:31 (current). Recorded here, not run.
# Mathematical and computational evidence for T731
This is build evidence. It is not the fresh critic's revision-bound verdict or argument-selection sidecar.
## Sources and normalization actually inspected
Hartmann, arXiv:1101.4601v1, Sections 4.1–4.3 defines X_t as the simultaneous minimal resolution of the diagonal fourth-root quotient of the all-plus Fermat pencil. Theorem 4.8 fixes the lattice marking; equation (10) and Remark 4.32 give [Omega]=[P h-e+2P^2 f] for the ordered basis (h,e,f). Its Gram matrix is [[4,0,0],[0,0,1],[0,1,0]]. Therefore the cycle pairing ratio is -integral_h(Omega)/(4 integral_f(Omega)); this is a scale-independent consequence, not an assertion that a normalized PF solution is a single integral period.
Example 4.27 gives F=sum A_n z^n and L=F log(z/256)+4 sum A_n(H_4n-H_n) z^n, with A_n=(4n)!/(256^n(n!)^4). Theorem 4.29 identifies P=L/(2*pi*i*F). Theorem 1.1 and proof of Theorem 4.37 specify triangle vertices infinity, i/sqrt(2), (1+i)/2. The isolated i/2 in the statement of 4.37 conflicts with its proof and the 4.28 limit. We use the germ and independent elliptic comparison, not a guessed correction to the local U_1,U_2 expression. The smooth origin is exactly (1+i)/2; the singular point t=1 is a control only.
Yang, arXiv:2004.00820v1, equations (5.3), (5.15)–(5.17), identifies the ratio with the Legendre period ratio under z=16(1-lambda)lambda^2/(lambda-2)^4. Yang's psi is our t; Yang's t is our z. Equations (4.25)–(4.33) fix the elliptic parameter and theta nome. K(m) uses parameter m, never modulus m.
Dembélé–Panchishkin–Voight–Zudilin, arXiv:1906.07384v3, (2.4)–(2.5), gives the rank-three eta modular identity and the CM control rho((1+3i)/2)=-1/48. The higher-rank Asai assertions are conjectural and are not used.
HTML and PDFs are archived, with URLs and SHA-256 in sources.json. The three full sources are research artifacts, not table attachments or copied prose. Current related-table documents/metadata are archived separately. T679/T680 are rank-four Calabi–Yau quantities. T25/T425 supply different elliptic quantities; T722 has a different marking. None is independent numerical validation of this table.
## Production branch and certified arithmetic
Set u=lambda/(2-lambda). Exact substitution gives z=4u^2(1-u^2). Thus s=sqrt(1-z), u=sqrt((1-s)/2), lambda=2u/(1+u). For z>1 on the upper bank, s=-i*sqrt(z-1); u has positive real and imaginary parts. For 0<z<1 both roots are positive real. This continues lambda~sqrt(z)>0. At z>1, lambda has positive imaginary part and 1-lambda has negative imaginary part: principal elliptic K follows the prescribed path without crossing its cuts. No root is reselected by proximity and no modular reduction is performed.
All rational inputs are enclosed before roots or special functions. Arb carries their uncertainty through the elliptic ratio, and the denominator is explicitly separated from zero. Each nonzero component passes a radius check stronger than 120 decimal digits before rounding to 100 digits. Production measurements and actual denominator bounds are in production.json. Reality for t>1 follows from real 0<lambda<1, and the real part is exactly zero. The origin is an exact dyadic complex ball.
## Exact t=3/4 component, independently of decimal patterns
Write d=sqrt(-7) with Im(d)>0. Exact algebra gives lambda=(31+3d)/32, r=(3-d)/8, r^2=1-lambda, and ((1-r)/(1+r))^2=1-1/lambda. It also gives z=256/81. The algebra is checked in cm_check.py over Q(sqrt(-7)).
The classical lambda duplication identity is lambda(2*tau)=((1-sqrt(1-lambda(tau)))/(1+sqrt(1-lambda(tau))))^2 on the selected branch; it follows from theta reduction/Landen identities, DLMF 20.7(iv), (vi), and lambda=theta_2^4/theta_3^4, DLMF 23.15. The TS transformation gives lambda(1-1/tau)=1-1/lambda(tau), DLMF 23.18.1–2. The square root r is the branch with positive real part, continued from the germ; it remains so on this path.
Consequently lambda(2*tau)=lambda(1-1/tau) exactly at this input. The rigorous elliptic enclosure puts both arguments inside 0.49<Re<0.51, 1.3<Im<1.4. Lambda is injective on that rectangle: equal lambda values differ by Gamma(2), and if a Gamma(2) matrix has c!=0 then |c|>=2 and Im(gamma*w)=Im(w)/|cw+d|^2<=1/(4 Im(w))<0.2 for Im(w)>1.3; this contradicts Im(gamma*w)>1.3. If c=0, the transformation is an even integer translation, impossible between two points in this rectangle unless it is the identity. Here we use Yang Section 4.3's Hauptmodul/uniformization property. Thus 2*tau=1-1/tau, and 2*tau^2-tau+1=0. The upper half-plane root is (1+i*sqrt(7))/4. The real component 1/4 is stored exactly; the imaginary radical is enclosed to 100 digits in type C. The raw elliptic output before this exact-component preservation remains in elliptic-raw.json.
## Numerical checker and deliberately wrong controls
pf_check.py independently implements the third-order scalar ODE, its harmonic-number Frobenius initial data, and Taylor transport. It does not use the elliptic pullback, elliptic K, or Arb. run_pf.py runs mpmath at 145 and 180 decimal digits and preserves full checkpoints in pf.log and pf-results.json. All 22 inexact fibres are checked. The common assumptions are the published PF germ and continuation path. PF truncation is a numerical convergence check, not a rigorous error certificate; production's bound is separately supplied by Arb.
check_evidence.py parses each stored decimal into its claimed exact rational rectangle, rejects nonfinite/large enclosures, and compares both PF precision runs with that entire rectangle. A 1e-125 numerical checker allowance is explicitly recorded, far smaller than stored uncertainty. The exact origin and exact real components use the same guard only for cross-check rounding.
A second check evaluates x=eta(2*tau)^24/eta(tau)^24 = q*product(1+q^n)^24 and rho=256*x/(1+64*x)^2 with q=exp(2*pi*i*tau). For N factors and r=|q|, the omitted logarithm is bounded by E=24*r^(N+1)/((1-r)*(1-r^(N+1))); the resulting multiplicative error is bounded by exp(E)-1. A differentiated geometric tail bounds the derivative as well. Nonzero derivative lower bounds measure local inverse conditioning; they do not alone select the global lift. The triangle constraints and independent PF continuation check that lift. All nonzero fibres are covered, using the actual stored rectangles, not midpoint copies.
Known controls: exact origin, source CM image -1/48, and the singular limiting value i/sqrt(2). Wrong controls include swapped K factors, a missing 256, squaring the elliptic parameter, lower-bank continuation, integer shifts, NaN and a vacuous box. Integer-shift controls deliberately retain the correct eta image while failing the lift and PF tests. Results are machine-generated in check-results.json and stored-check-results.json, not hand-authored attestations.
## Reproduction and limitations
From an external Sage environment with numberdb and mpmath installed, run `sage -python run_pf.py > pf.log`, extract the final PF_RESULT_JSON object to pf-results.json, then run `sage -python check_evidence.py` with production.json beside it. `sage -python generate.py --verify` checks all live entries with the owner's configured API key. In this environment every computation used agents/sage.sh; its mounted files are named explicitly in the run logs.
The scalar table supplies neither a reconstruction of cycles nor a new proof of a Hodge relation. Six is a noncanonical exploratory height bound. Scientific critique and the independent argument-selection verdict remain the controller's next stages. No human reviews are accepted and no draft is offered here.
Canonical instructions: ../20261117T024513Z-build-skills/manifest.json, core SHA-256 981b5f1d6f597ce7396556b3c1067a3b6220cf91beed5c2d9f45d610ce994a2c; matching bundle 984965681e2f8925c511a121c118efcde59581c7b0d906cac8b641bf6925e55a. Bundle hashes were revalidated before computation.