Numeric and exploratory research workbench for the Golden Physics Project's shadow holography framework — the standing discovery repo behind GPPVerify, which carries the formalized, Lean 4-checked results.
Shadow holography. One involution appears in two places. It is the celestial shadow map
Δ ↦ 2 − Δ on conformal dimensions, and the zeta functional equation s ↦ 1 − s. Under
Δ = 2s they coincide, and the reflection τ(s) = 1 − s̄ fixes exactly the critical line
(celestial Re Δ = 1). The framework, developed in Daniel Toupin's On the Nature of
Nature, reads this as a nested holographic dictionary. Each level is a Mellin or
Pontryagin duality, so each is exact:
- Line ↔ sphere. The multiplicative line
ℝ₊, split atx = 1, is the boundary. The spectral sphere, split by the critical line (the equator), is its hologram. The inversionx ↦ 1/xbecomess ↦ 1 − s̄. - Integers ↔ functional equation. The integers are the Fourier modes of a circle, and Poisson self-duality becomes the functional equation.
- Primes ↔ torus. The primes span a compact torus
∏ S¹, and the critical line embeds in it densely: a hologram inside a hologram.
In this language the functional equation says the hologram is symmetric. RH is one open
question inside the framework: whether the hologram is also positive (reflection-positive
across the equator). Celestial scattering meets the same equator in its scale sector. That
is a structural parallel, not a claim that ξ is a physical correlator. The full write-up,
with what is a theorem, what is open, and what has been killed, is
discovery/arithmetic_holography/ARITHMETIC_HOLOGRAPHY_RIGOROUS.md.
Nothing in this repo is proved. It holds numeric evidence, hand derivations, and
literature checks. Once a result is solid enough to state as a theorem, it is formalized in
GPPVerify (no sorry, no axiom
asserting an open claim, CI-verified). CLAUDE.md documents that workflow and the
branch-hygiene rules.
discovery/arithmetic_holography/: making the shadow-holography dictionary rigorous. This covers the three levels above, the Osterwalder–Schrader form of the one open positivity statement, and the constraints any route must respect (Beurling / Davenport–Heilbronn, de Bruijn–Newman, the Fisher no-go). It includes a numerical falsifier (riemann_split_hb.py) that killed one proposed hemisphere-dominance route.weil_decay/(root scripts +discovery/weil_decay/) — the truncated Weil quadratic form. Connes–van Suijlekom and Connes–Consani–Moscovici build, for a prime cutoffcand bandN, a finite Galerkin matrixQ(c)whose zeros provably sit on the critical line for every finitec; convergence asN → ∞is the open question this thread measures. See "The Weil-decay question" below for the current numbers.discovery/celestial_box/— extending the audited scalar-box cut/dispersion construction toward pure Einstein gravity; tracking exactly where D-dimensional unitarity subtleties (rational terms invisible to 4D cuts) start to matter.discovery/wiener_hopf/— the exact bridge between the principal-series cut weight, the Wiener–Hopf Fourier window, and the conical prefactor ats = 1/2 + it.discovery/positive_reals_cft/— isolating the precise representation-theoretic structure behind treating(R⁺, ×)as a one-dimensional principal-series system.discovery/number_thermodynamics/— the canonical Gibbs distributionP_β(n) = n⁻β/ζ(β)on the positive integers and its thermodynamic reading.
λ_min(c), the smallest even-sector eigenvalue of Q(c), is non-negative for every
finite c. (Weil positivity, and hence RH, is this holding in the limit. This is one numerical
probe of the positivity question, not a route the project is built around.) We are not testing
whether it's positive — we're measuring how fast it decays:
Is
log λ_minlinear inlog c, and if so, what is the constant?
Measured so far (unconverged, N=20, T=100): 3.67e-27 at c=7, 8.09e-18 at c=5. A
two-point natural-log slope of about -64, against -2γ₁ = -28.27 and
-4γ₁ = -56.55. If the constant is arithmetic, this is a quantity that sees the
zeros, which is more than can be said for most reformulations of RH.
N-convergence comes first. Going from dim 33 to dim 41 at c=7 moved λ_min from
2.4e-25 to 3.67e-27, a factor of 65 — N=20 is not converged and any slope fitted to
it is meaningless. The scan does c=7 at N = 20, 28, 36 before anything else.
Caveat on small c. At c=3 (N=20, T=100) we saw λ_min = -0.332 — almost certainly
marginal basis resolution, per the reference implementation's own account at c=23, 29.
The c=29 case has since been resolved at higher precision: stably positive
(≈1.59e-62, 120-digit residual 1.64e-121) at 90–150 digits, though N and T are still
unconverged there too, so this doesn't yet rescue a c-slope fit. See
discovery/weil_decay/C29_PRECISION_AUDIT.md.
python point.py --c 7 --N 36 --T 300 --dps 90 for one point, or push / use
workflow_dispatch to run the whole matrix in parallel on Actions. Results land in
results.jsonl and RESULTS.md, committed back automatically.
Discovery only, across every thread above. See CLAUDE.md for the process rules
(branch hygiene, when and how a result graduates to GPPVerify).