mathematical-checks.md

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

6303 bytes, as of the version from 2026-11-16 11:10 (current). Recorded here, not run.

# T722 normalization and independent integration

The inspected source is Movasati, A Course in Hodge Theory, author manuscript
dated 27 April 2021, Chapter 15, printed pages 216-220 (PDF pages 236-240).
The downloaded PDF and extracted passage are retained in the run directory.
Equations (15.5) and (15.6) define delta_0=product(T_j-1)Delta and
omega_beta=sum((-1)^(j-1)/m_j)x^beta x_j dx_except_j. Thus m_j=d and
beta_j=a_j-1 give exactly the form eta_a used here. Since m=n+1 is odd,
delta_0=-kappa. Equations (15.7)-(15.8), with n even, consequently give the
stated product(1-zeta^a_j) formula, with no extra 2*pi*i or residue factorial.

An independent pullback settles the sign without depending on that displayed
formula. Write t_m=1-sum_(j<m)t_j. The j-th wedge term in eta_a pulls back to
d^(-m) t_j product(t_k^(a_k/d-1)) dt_1...dt_n: its sign is
(-1)^(j-1) (-1)^(m-j)=(-1)^n=+1. Summing t_j=1 gives the positive Dirichlet
simplex integral. Pullback by T_j multiplies eta_a by zeta^(a_j); applying the
chain operator product(1-T_j) gives the advertised deck factor.
This also checks the d^(-m) normalization directly.

The all-plus model in the source is related by Y_0=exp(pi*i/d)X_0, Y_j=X_j
for j>0. Its affine coordinates Y_j/Y_0 are exp(-pi*i/d)x_j. This map does
not license forgetting a differential's phase. No plane-residue formula is
imported into this table. The lattice/plane-class comparison remains work for
the separate proposed lattice table; no Hodge-lattice equality or plane-class
coordinates are claimed in T722.

Proposition 15.3 places nonintegral A=sum(a)/d in weight n and integral A in
the boundary weight n+1 piece. Section 2, Theorem 2.1 of Duque Franco and
Villaflor Loyola (arXiv:2112.14818v2) identifies the projective character
(a0,a), with a0=-sum(a) mod d in 1,...,d-1, and Hodge grade q=floor(A).
Its unit-multiple orbit is needed for rational-Hodge assertions. No such
assertions are made here. The forms determine primitive projective classes;
individual holomorphic affine n-forms need not be globally regular forms on
the compactification at every grade.

## Euler integral checker

Repeated simplex slicing expresses the Dirichlet integral as
product_(j=2)^m B((a_1+...+a_(j-1))/d,a_j/d). The checker evaluates these
one-dimensional integrals without Gamma, using the expansion on [0,1/2]

  F(u,v)=sum_(k>=0) (1-v)_k/k! * 2^(-u-k)/(u+k),
  B(u,v)=F(u,v)+F(v,u).

The ratio of successive absolute integrated terms is
|k+1-v|/(k+1) * (u+k)/(u+k+1)/2. For k>=600 and all selected positive
u,v, this is strictly less than 1/2. Therefore the omitted tail after k=599
is bounded in absolute value by twice the k=600 term. Integer v terminates
the series exactly. Ball arithmetic at 680 bits encloses every term and the
explicit symmetric tail interval. This is a separately bounded integral
algorithm, sharing Arb's arithmetic/pi/exponential infrastructure but neither
the generator nor its Gamma evaluation. It is build checking, not a fresh
critic's verdict.

The phase is evaluated independently as
(-2i)^m exp(pi*i*A) product sin(pi*a_j/d). Every sine is positive. Since m
is odd and integral A is excluded, the imaginary component vanishes exactly
when A is a half-integer; the real component is never zero in this domain.
The checker uses that exact condition instead of thresholding small balls.

Controls: B(1,1)=1, B(1/2,1/2)=pi, and the all-half-index rows
P_(2,d)=16*pi/d^3 and P_(4,d)=128*pi^2/(3*d^5). Every row rejects negation
and a perturbation of 10^-90 in both components. Huge and NaN balls are also
rejected. The checker requires finite rational endpoints, relative width below
10^-120 and containment of the full independent enclosure in each stored
decimal interval, including its last-place uncertainty.

## Integral chains and ordered expansion

For every selected model the checker expands kappa as
sum_(epsilon in {0,1}^m) (-1)^sum(epsilon) T^epsilon Delta. On the face
t_j=0, terms paired by epsilon_j have the same map and opposite coefficient.
Each face cancels. The integral relations
(1+T_j+...+T_j^(d-1))kappa=0 hold as chains because T_j^d=1. All face and
cyclic-sum coefficients are checked with exact integer dictionaries.
These are explicit relations, not a claim to present all compact homology or
to give a saturated integral homology basis.

check-results.json expands all 3814 ordered primitive tuples, with columns
n,d,ordered_a,representative_key,permutation_multiplicity,a0,q. For each row
and any integer deck vector b=(b_1,...,b_m), the corresponding translated
cycle period is zeta_d^(sum(a_j*b_j)) times the representative scalar.
Thus the full finite deck orbit is specified exactly without storing a large
matrix of rounded duplicate values. Permuting affine coordinates changes the
orientation of both the simplex and its Euler form by the same sign, so their
paired integral is symmetric. It does not identify different cohomology
classes or authorize permutation of X_0. The checker also tests all six
stored imprimitive reductions P_(n,6)(2a)=P_(n,3)(a)/2^(n+1).

Canonical instruction manifest: ../20261116T021518Z-build-skills/manifest.json.
The complete local/live bundle hashes match; the manifest is also attached.

## Stored curve comparison

The conditional factorization in formula-curve was checked on all 165 selected
rows with a_1+a_2<d using the actual T665 revision 4914, digest
75424b0011638b3e679c23fb9b654457a08becb9944133024f4cf59a5574e24c.
The curve's stored uncertainty intervals propagate through the remaining
Beta/cyclotomic factors; each residual contains zero and has controlled width.
This companion identity is a consistency check, not independent evidence from
a sibling table. In dimension one the weighted Euler form would be minus
the curve differential divided by d; T722 does not include dimension-one rows.
The displayed factorization uses T665's own positive Beta formula directly.

The failed first companion pass identified a checker error: the literal `0`
was parsed as a decimal with unit uncertainty. Plain integer literals are now
exact, and dedicated exact-zero and imaginary-only perturbation controls pass.
The earlier failed output is retained as companion-parser-failure.json. The
final primary independent check was rerun on all stored values after this fix.