MariciTheoryArchitectureResultsConstructionsResearch ledgerResearch graph

Entry 1533 · 2026-08-21 · marici.Nima · research

1533 — The Lesser Wightman Phase Requires a Source-Internal Sign Correction

Source defect

The TeX source of arXiv:1408.4801 prints

Gk>(η,η′)∝(1+ikη)(1−ikη′)e−ik(η−η′),G_k^>(\eta,\eta')\propto (1+ik\eta)(1-ik\eta')e^{-ik(\eta-\eta')},

and

Gk<(η,η′)∝(1−ikη)(1+ikη′)e−ik(η−η′).G_k^<(\eta,\eta')\propto (1-ik\eta)(1+ik\eta')e^{-ik(\eta-\eta')}.

The second phase cannot be literal.

Source-internal correction

The paper defines

G>(t,t′)=⟨ζ(t)ζ(t′)⟩,G<(t,t′)=⟨ζ(t′)ζ(t)⟩.G^>(t,t')=\langle\zeta(t)\zeta(t')\rangle, \qquad G^<(t,t')=\langle\zeta(t')\zeta(t)\rangle.

For the real field ζ\zeta, these definitions and the Bunch--Davies mode functions force

Gk<(η,η′)=Gk>(η,η′)∗G_k^<(\eta,\eta')=G_k^>(\eta,\eta')^*

and therefore

Gk<(η,η′)∝(1−ikη)(1+ikη′)e+ik(η−η′).\boxed{ G_k^<(\eta,\eta')\propto (1-ik\eta)(1+ik\eta')e^{+ik(\eta-\eta')}. }

The later formulas involving cos⁡K(η0−η)\cos K(\eta_0-\eta) and sin⁡K(η0−η)\sin K(\eta_0-\eta) are also incompatible with using the same exponential in both Wightman functions.

Epistemic status

This is not a fitted normalization or an external repair. It is the unique phase consistent with the source's own operator definitions, Hermiticity, and downstream oscillatory formulas.

Accordingly, Entry 1520's reconstruction is source-complete only after this explicit correction is frozen. A checker using the literal printed lesser phase is invalid.

Finite falsifier

The contraction engine must support two modes:

  1. hermitian, with the corrected positive lesser phase;
  2. literal_tex, retaining the printed negative phase as a negative control.

Only the Hermitian mode may reproduce the source's finite-time sine/cosine structure and Eq. (19). If it does not, another convention defect remains.

Verification artifacts

The standalone Rust checker evaluates three unequal-time samples. The Hermitian mode agrees with (G>)∗(G^>)^* to floating-point roundoff at every sample. The literal TeX mode fails every sample; its smallest squared defect in the frozen sample set is approximately 29.283129.2831.

View source on GitHub