mathematical-checks.md

back to table · edit · history · where entries came from · files · download

8043 bytes, as of the version from 2026-11-20 10:08 (current). Recorded here, not run.

# T774: definition, bounds and checks

This is build evidence, not an independent scientific verdict. The natural domain is every positive integer r. All ten entries r=1,...,10 are newly computed; no earlier entries are retained. The archived skill manifest is ../20261120T083619Z-build-skills/manifest.json, bundle SHA-256 b5c3c7985b7c9e178b4899d083a9e3bc4284989dff76f05e5b0e52835d47eed2. Both archived bundles and the fresh live resource manifest match.

## Model and branch

The projective quartic is sum(X_j^4)-4*t*prod(X_j)=0. The mirror surface is the smooth minimal resolution of its quotient by {a in mu_4^4: prod(a)=1}/mu_4; it is not the original quartic or its singular quotient model. Hartmann arXiv:1101.4601v1 Section 4.1 constructs this surface, Theorem 4.8 gives the integral marking and primitive T_0 embedding, equation (10) gives [Omega]=[P*h-e+2*P^2*f], and Example 4.27 and Theorem 4.29 fix the Frobenius germ P=W_2/(2*pi*i*W_1). On t>1 take the real logarithm in z=t^-4 and the positive imaginary period tending to i*infinity as t tends to positive infinity. There is no modular reduction. The isolated inconsistent vertex in Theorem 4.37 is not used.

Dembele et al. arXiv:1906.07384v3 (2.5) gives rho=256*R/(R+64)^2 with R=(eta(tau)/eta(2*tau))^24. Thus t^4=(R+128+4096/R)/256 on the positive branch. Yang arXiv:2004.00820v1 (5.3), (5.15)-(5.17) gives the independent Legendre pullback. Yang's t=psi^-4 is the present z=t^-4, not this table's coefficient. Its theta nome in (4.29) is exp(pi*i*tau); the eta nome is exp(2*pi*i*tau). The two must not be interchanged.

At tau=i*sqrt(r), eta is positive. R=q^-1*prod(1+q^n)^-24 increases as r increases, since every factor and q^-1 increase when positive q decreases. The canonical level-two relation at j(i)=1728 is (m-8)^2*(m+64)=0, m=4096/R>0, so R_1=512. Consequently R_r>=512, and t_r>1 for every r>=1. The map R+128+4096/R is increasing for R>64. In the selected region lambda in (0,1/2], Yang's z=16*(1-lambda)*lambda^2/(2-lambda)^4 is positive and at most 32/81<1; continuation from lambda=0 meets no branch point. This fixes P(t_r)=i*sqrt(r), rather than merely an exponential or modular equivalence class. The AGM check verifies this lift numerically at every stored value.

For smoothness, a singular point of the quartic cannot have a zero coordinate: the partial derivatives would force all coordinates to zero. Multiplication of the four equations X_j^3=t*prod_{k!=j}(X_k) then gives t^4=1. Every stored interval lies strictly above one; infinity is absent. The quotient resolution and its polarization are those in Hartmann, not an assumption that the quotient model itself is smooth.

## Enclosures and independence

The producer at 600 bits sums S_N=sum_{n=1}^N log(1+q^n). The omitted positive sum is at most q^(N+1)/(1-q), because log(1+x)<=x. It stops when that bound is less than 2^(-624), expands S_N symmetrically by the tail bound, then propagates it through exp(-24*S)/q and two positive square roots. No fixed-precision result is wrapped afterward. All inputs, pi, sqrt(r), exp, log and arithmetic are Arb enclosures. Measured output accuracy is at least 590 bits, with 21 to 68 product terms.

The alternate implementation uses 900-bit positive theta series, 24 terms, and geometric tail bounds. Theta_3=1+2*sum(q^(n^2)); the omitted tail is <=2*q^((N+1)^2)/(1-q^(2*N+3)). Theta_2=2*q^(1/4)*sum_{n=0}^N q^(n*(n+1)); the omitted tail is <=2*q^(1/4)*q^((N+1)*(N+2))/(1-q^(2*N+4)). Their quotient yields lambda and then t through Yang's pullback. The minimum theta coefficient accuracy is 891 bits.

For the inverse period, z=t^-4, u=sqrt((1-sqrt(1-z))/2), lambda=2*u/(1+u), all with positive roots. The imaginary coordinate is AGM(1,sqrt(1-lambda))/AGM(1,sqrt(lambda)). Eighteen arithmetic-geometric iterations bracket each AGM between its geometric and arithmetic iterates, so the union of their final balls is a rigorous enclosure, not a convergence guess. DLMF 19.8.1-19.8.5 supports this relation (DLMF uses modulus sqrt(lambda); this table uses elliptic parameter lambda).

Every producer enclosure and every theta enclosure lies wholly inside the corresponding stored 100-significant-digit decimal interval. The complete stored interval is propagated through the inverse period, with finite nonvacuous results that contain sqrt(r). This numerical consistency is distinct from the source identity establishing the universal period relation. Generator verification reports 10/10 matched, but is only reproducibility.

The checker has known answers t_1^4=81/32, lambda(i)=1/2 and m(i)=8. Controls reject 255 in place of 256, a shift by one, a negative fourth root, a square root in place of a fourth root, the wrong theta nome, the wrong lattice sign, NaN and a huge interval. The alternate methods share Sage/Arb, arithmetic indexing and source identities with the producer. They are independent formulas within this build session; the fresh critic must supply actor independence.

## Exact algebraicity and root selection

PARI polclass(-4*r) returns the Hilbert class polynomial H for j. Its documentation specifies the default inv=0 as the classical j-invariant, with nonfundamental discriminants retained. The canonical modular equation from PARI ellmodulareqn(2) has type flag 0 (canonical) and polynomial (m+16)^3-m*j. The canonical normalization is m=2^12*(eta(2*tau)/eta(tau))^24=4096/R. The flag 1 of polmodular is an invariant selector, not this canonical-equation flag; the initial erroneous probe was refused and was not evidence.

Exact rational arithmetic forms m^h*H((m+16)^3/m), factors it, and retains the unique irreducible factor compatible with the certified positive eta enclosure. It eliminates m from this factor and 256*t^4*m-m^2-128*m-4096, then selects the unique compatible irreducible factor in Q[t]. Root association uses the modular identities and the exact class-polynomial theorem; numerical recognition alone is not an algebraicity proof. Opposite exact signs at the two stored endpoints and a derivative enclosure excluding zero certify one simple root inside that interval. All resulting degrees exceed one (4 through 16), so no proved rational output has been rounded into a decimal. The ten exact certificates are attached. At r=1 the polynomial is x^4-81/32.

The existing T634 2A values and T657 polynomials were read live and compared at every selected index. All ten affine fourth-power identities and polynomial residuals agree, using the stored uncertainty. These are dependent corpus consistency checks, additional to the theta and AGM checks; neither existing table is an independent numerical oracle.

## Lattice certificate and limits

In T_0=<4>+U take G=[[4,0,0],[0,0,1],[0,1,0]] and gamma_r=(0,1,-2r). Its exact norm is -4r, its entries have gcd one, and its pairing with Omega=(P,-1,2P^2) is 2*(P^2+r). At P=i*sqrt(r) this vanishes exactly. Also Omega^2=0 and Omega*conj(Omega)=8r>0. T_0 is primitively embedded in the K3 lattice by Hartmann's Theorem 4.8 and proof (Theorem 4.11), and gamma is outside the generic algebraic rank-19 rational subspace. Hence Lefschetz (1,1) gives an additional integral divisor class, with negative norm between 4 and 40 in absolute value.

For an explicit ambient embedding in 2E8(-1)+U_0+U_1+U_2, write h=a_0+2*b_0, e=a_1, f=b_1, where a_j*b_j=1 in each hyperbolic plane. The generic <-4> generator is a_0-2*b_0. This describes the specified sublattices, not a claim that their direct sum is the whole unimodular K3 lattice. The file lattice-certificate.json records the exact selected vectors and pairings.

No effective line, conic or curve equations, degree bound with respect to a polarization, saturated special Neron-Severi lattice, entire Noether-Lefschetz locus or reconstructed correspondence is claimed. The two-table family is only a known-identity CM control and partial mathematical coverage of the originating request. The polynomial companion does not exist yet; its live link must be added during family consistency review, not guessed now.