Exact Hermitian Formula and Rigorous Asymptotics of Remote Bogoliubov-Kubo-Mori Distinguishability after Local Gaussian Filtering in a Massive Harmonic Chain
Alexandru Rora
This record contains the preprint, its Supplemental Material, and the complete computational package behind them. VERSION 2 NOTE. This version corrects the physical interpretation of the decay rate. No previously reported scientific result, coefficient, numerical value, or computational formula was altered, and the code archive is carried over unchanged from v1. Eq. (4) was algebraically expanded to state kappa explicitly; an explanatory section, numbered S0 so as to leave the version-1 numbering intact, was added to the Supplemental Material; and the vertical range of Fig. 2 was rescaled to match its data, with no plotted value changed. On a spatial lattice the inverse equal-time correlation length kappa and the Hamiltonian energy gap m_L are distinct quantities, coinciding only in the continuum limit. Version 1 described 2 kappa as a two-particle spectral threshold, which reads as a statement about the energy spectrum; the two-particle energy threshold is 2 m_L. The result concerns the spatial decay of an equal-time correlator and is governed throughout by kappa. RESULT. For a massive harmonic chain, a local Gaussian filter applied at one site changes the reduced state of a distant block of b sites by an amount whose Bogoliubov-Kubo-Mori distinguishability decays as delta_B(r) ~ r^{-1} exp(-2 kappa r), at exactly twice the one-particle inverse equal-time correlation length kappa = 2 arcsinh(m_L/2) = arccosh(1 + m_L^2/2). The doubled rate is the lowest two-particle SPATIAL exponent of the even (:phi^2:) channel: the equal-time :phi^2: correlator is the square of the one-particle propagator, so its nearest complex-momentum singularity sits at twice the one-particle branch point. On the lattice kappa is distinct from the energy gap m_L = min_k omega(k), from which it differs at third order in the mass; the sweep resolves the difference directly, the static-source benchmark returning lambda_A/kappa = 1.00000000 at all six masses while lambda_A/m_L falls to 0.99496 at m_L = 0.35. The same branch point fixes the algebraic prefactor. Smearing the filter with a form factor having a zero of order p at the branch point suppresses the one-particle kernel amplitude by r^{-p}; because the covariance tangent is quadratic in that kernel, the BKM distinguishability acquires an additional factor r^{-2p}, giving delta_B^(p)(r) ~ r^{-(2p+1)} exp(-2 kappa r) for arbitrary fixed p. THREE LOAD-BEARING RESULTS, EACH WITH AN INDEPENDENT COMPUTATIONAL ROUTE. (1) An exact Hermitian formula for the restricted BKM susceptibility, chi_b(X_Q) = -sum_ij H_ij Ytilde_ij^2, in which positivity is manifest and eigenvalue degeneracies are handled by the divided-difference limit. Checked against a from-scratch Gaussian relative-entropy implementation by finite-differencing D(V + eps X || V), and against a Daleckii-Krein route at 2136 sampled points with median relative difference 4.3e-15, the worst-conditioned point re-checked at 100 digits. (2) Rigorous asymptotics obtained by collapsing the Fourier contour onto the branch cut and using an exact Beta representation. This gives a Poincare expansion to arbitrary order with a uniform Taylor-Lagrange remainder, and the source hierarchy for arbitrary fixed p rather than only the first few orders. The third-order coefficient a_3 = (80 coth kappa - 75 coth^3 kappa)/1024 follows from the same route. (3) A 24-cell finite-lattice sweep and an independent arbitrary-precision infinite-chain calculation, the latter free of finite-size and periodic-image artifacts, confirming the coefficients through third order across all cells. -------------------------------------------------------------- This record contains the preprint, its Supplemental Material, and the complete computational package behind them. VERSION 1 NOTE RESULT. For a massive harmonic chain, a local Gaussian filter applied at one site changes the reduced state of a distant block of b sites by an amount whose Bogoliubov-Kubo-Mori distinguishability decays as delta_B(r) ~ r^{-1} exp(-2 kappa r), at exactly twice the one-particle inverse correlation length kappa. The doubled rate is the lowest two-particle threshold of the even (:phi^2:) channel; the branch point of the lattice propagator fixes the algebraic prefactor. Smearing the filter with a form factor having a zero of order p at the branch point suppresses the tail by a further factor r^{-p}, for arbitrary fixed p. THREE LOAD-BEARING RESULTS, EACH WITH AN INDEPENDENT COMPUTATIONAL ROUTE. (1) An exact Hermitian formula for the restricted BKM susceptibility, chi_b(X_Q) = -sum_ij H_ij Ytilde_ij^2, in which positivity is manifest and eigenvalue degeneracies are handled by the divided-difference limit. Checked against a from-scratch Gaussian relative-entropy implementation by finite-differencing D(V + eps X || V), and against a Daleckii-Krein route at 2136 sampled points with median relative difference 4.3e-15, the worst-conditioned point re-checked at 100 digits. (2) Rigorous asymptotics obtained by collapsing the Fourier contour onto the branch cut and using an exact Beta representation. This gives a Poincare expansion to arbitrary order with a uniform Taylor-Lagrange remainder, and the source hierarchy for arbitrary fixed p rather than only the first few orders. The third-order coefficient a_3 = (80 coth kappa - 75 coth^3 kappa)/1024 follows from the same route. (3) A third-order accounting of the finite-window fit deficit across all 24 (m_L, b) cells of the lattice sweep, including the scalar closed form c_3 = (22 coth kappa - 21 coth^3 kappa)/128 at b = 1, which cross-checks the whole propagation chain. WHAT IS IN THE ARCHIVE. Seven production modules for the finite-lattice sweep; six independent verification scripts; 87 unit tests and 12 integration tests; the persisted raw sweep data for all 24 cells; and the implementation and mutation-audit records. Every number quoted in the paper is produced by a script included here. The full sweep takes about nine minutes but is not required: its output is included, with its SHA-256 recorded. ON THE AUDIT RECORDS. Two files on this record document failures rather than successes, deliberately. VERIFICATION_LOG.txt lists what was found wrong and fixed while assembling the package: ten stale equation cross-references, three proof gaps, a LaTeX typo that compiled without any warning, an arithmetic overstatement. EXTERNAL_AUDIT_LOG.md records four adversarial review rounds run by two independent systems given each other's output and instructed to disagree. One of those rounds produced a coverage result rather than a fix: deliberate mutation testing showed that two normalization constants were, at the time, verified only self-consistently by the test suite. The shipped code was and is correct, and the affected published numbers were reproduced independently from outside the codebase, but "the tests pass" was not by itself sufficient evidence for those two constants. The oracle tests in test_normalization_oracles.py close that gap by measuring both from the lattice operator directly. LICENSING. The two manuscript PDFs are All Rights Reserved. Within the archive, source code is MIT and data and documentation are CC BY 4.0; see LICENSE. AI-USE DISCLOSURE. Claude Sonnet 5 (Anthropic) was used for derivation, implementation, numerical verification, literature and citation checking, and drafting. Independent adversarial review used OpenAI ChatGPT (web) in Codex/Work Mode, reported by that system as Codex based on GPT-5.6 Sol High, across multiple rounds between approximately 20 July and 9 August 2026. A further adversarial audit, the branch-cut/Beta derivation through a_3 and c_3, and the mutation-testing round used Claude Opus 5 (Anthropic), 10-11 August 2026. Every issue raised by any reviewer was independently re-derived or re-computed before any change was accepted. The author reviewed and is responsible for all content, all code, and all claims.
•Jul 27, 2026•0 citations