When a system of electrons is cut into a region and the rest, its correlation matrix has a cross block, X_A = C_AB C_BA: the weight between the two sides of the cut. This paper proves what that weight can and cannot say, and then studies where opened (non-Hermitian) matter leaves the disk of its occupation chart. The weight between. X_A is positive and never larger than 1/4. Both sides of the cut see the same positive spectrum. It is stable under small changes, with a certified count. It sits below the local defect C_A − C_A², and for a pure Gaussian (Slater) state, where C is an orthogonal projector, the two are equal. Four times its trace bounds the Gaussian mutual information across the cut from below, with a sharp constant, and there is no upper constant without a spectral margin. Exact examples show that a one-body record does not in general determine correlations outside the quasi-free class. The signed boundary offset τ = log det(I − C_A)/det C_A moves no faster than τ̇² ≤ 4gS, and two preparations with the same census, offset, cross weight and coupling saturate this bound in opposite directions. The exit line. On the PT-symmetric Su–Schrieffer–Heeger chain, the band-averaged Petermann factor equals 2 exactly on the line E(0)E(π) = u², equivalently (w² − v²)² = 2u²(w² + v²). Exits predicted from that line before the run match the measured census exits of finite blocks to about 5e-8 in these runs. A linearly weighted operator, T = ½{X + ½, H}, commutes entrywise with the half-chain correlation block; the proof uses only H(k)² = E(k)²I, so it holds without Hermiticity. At 200 sampled gapped parameter points, its tested low-index edge eigenvectors exhibit an equally spaced entanglement ladder ε_j = jΔ + iθ, with a spacing Δ = 2πK′/K taken from a Hermitian band hidden inside the opened one and a shared phase θ given by an elliptic (Abel) formula. For this proposed ladder, the paper proves that it crosses the rim of the disk all at once, and that its phase reaches a quarter turn exactly on the Petermann line, in both directions. The dimer limit is proved exactly. Plain finite blocks give a census rotation equal to the same phase to within 7e-7. An analytic derivation of the ladder for the coupled chain remains open, and the paper states it as open. Related numerical work on the same model (Rottoli, Fossati and Calabrese, 2024) is cited and compared. Every claim carries a grade (proved, Lean-checked algebraic subclaim, verified numerically, known result, or open), and every number comes from a script included in the archive, with the reference-run environment recorded. Files: the paper (PDF); a web edition (HTML) with an interactive view of the proposed ladder model; and a zip holding the LaTeX source, the Python scripts and their outputs, the figure, the Lean proofs, and a README mapping each Lean grade to its theorem. Prepared with AI assistance (Claude, Anthropic; and OpenAI tools for parts of the formal development); the ideas and research direction are the author's.
When a system of electrons is cut into a region and the rest, its correlation matrix has a cross block, X_A = C_AB C_BA: the weight between the two sides of the cut. This paper proves what that weight can and cannot say, and then studies where opened (non-Hermitian) matter leaves the disk of its occupation chart. The weight between. X_A is positive and never larger than 1/4. Both sides of the cut see the same positive spectrum. It is stable under small changes, with a certified count. It sits below the local defect C_A − C_A², and for a pure Gaussian (Slater) state, where C is an orthogonal projector, the two are equal. Four times its trace bounds the Gaussian mutual information across the cut from below, with a sharp constant, and there is no upper constant without a spectral margin. Exact examples show that a one-body record does not in general determine correlations outside the quasi-free class. The signed boundary offset τ = log det(I − C_A)/det C_A moves no faster than τ̇² ≤ 4gS, and two preparations with the same census, offset, cross weight and coupling saturate this bound in opposite directions. The exit line. On the PT-symmetric Su–Schrieffer–Heeger chain, the band-averaged Petermann factor equals 2 exactly on the line E(0)E(π) = u², equivalently (w² − v²)² = 2u²(w² + v²). Exits predicted from that line before the run match the measured census exits of finite blocks to about 5e-8 in these runs. A linearly weighted operator, T = ½{X + ½, H}, commutes entrywise with the half-chain correlation block; the proof uses only H(k)² = E(k)²I, so it holds without Hermiticity. At 200 sampled gapped parameter points, its tested low-index edge eigenvectors exhibit an equally spaced entanglement ladder ε_j = jΔ + iθ, with a spacing Δ = 2πK′/K taken from a Hermitian band hidden inside the opened one and a shared phase θ given by an elliptic (Abel) formula. For this proposed ladder, the paper proves that it crosses the rim of the disk all at once, and that its phase reaches a quarter turn exactly on the Petermann line, in both directions. The dimer limit is proved exactly. Plain finite blocks give a census rotation equal to the same phase to within 7e-7. An analytic derivation of the ladder for the coupled chain remains open, and the paper states it as open. Related numerical work on the same model (Rottoli, Fossati and Calabrese, 2024) is cited and compared. Every claim carries a grade (proved, Lean-checked algebraic subclaim, verified numerically, known result, or open), and every number comes from a script included in the archive, with the reference-run environment recorded. Files: the paper (PDF); a web edition (HTML) with an interactive view of the proposed ladder model; and a zip holding the LaTeX source, the Python scripts and their outputs, the figure, the Lean proofs, and a README mapping each Lean grade to its theorem. Prepared with AI assistance (Claude, Anthropic; and OpenAI tools for parts of the formal development); the ideas and research direction are the author's.
For a region of a free-fermion ground state the entanglement Hamiltonian K_A = log[(1 - C_A) C_A^-1] splits into a part proportional to the identity and a trace-free remainder. The identity part shifts every level rigidly, so every observable built from level differences is blind to it. This paper is about that part. It is not a bulk quantity. The block carries one ladder per entangling cut, the trace is the difference of the two cut-phase charges, and on the exactly dimerised chain with a single cut carrying the whole charge it has the closed form 2 artanh(v / sqrt(v^2 + t^2)), derived in one line and measured to 10^-11. In exact arithmetic the trace is a definite number: zero to thirty-three digits when the two cut phases agree, and 0.786153199019314523 when they do not. The apparent divergence reported in earlier work was double-precision truncation of a finite sum, and is retired here. Along the two lines through the Rice-Mele plane where the half-line entanglement Hamiltonian is known in closed form, it is proportional to the Bisognano-Wichmann deformation T = (1/2){X, H} with no additive constant, and the ladder spacing and phase are fixed by the literature. Between those lines the commuting operator survives, measured at 10^-13 against controls that fail at 1.5 and 29, but the proportionality does not: the level map acquires an intercept b that is zero to 3 x 10^-9 on both lines and of order 10^-1 in the interior, and is identical whether the region has one entangling cut or two. That intercept is the measured failure of exact Bisognano-Wichmann proportionality. At first order in the staggering v the intercept is derived rather than fitted. The symbol's determinant is det(z - P) = z(z - 1) independently of momentum, so the factorisation is canonical everywhere; its logarithm is affine in an involution R = H/E with R^2 = I; the abelian Szego-type boundary functional is identically even in v and therefore cannot carry an odd intercept; and the first-order piece follows from the chiral zero mode of the dimerised chain, whose amplitudes are (-r)^j C(2j,j) 4^-j in closed form. The result is b/v = 2(E' - K')/(1 - r^2) + K', K' = K(1 - r^2), E' = E(1 - r^2), r = t1/t2, whose expansion in the modulus u is, in exact rationals, h(u)/(pi/8) = 1 - (5/8) u^2 - (9/64) u^4 - (65/1024) u^6 - (595/16384) u^8 - ... So pi/8, -5/8 and -9/64 are consequences rather than recognitions, and the natural elliptic candidate [9E - K]/16 - pi/8 is excluded by 7.7 percent on the first coefficient it does not get to choose. The closed form agrees with direct measurement extrapolated to v -> 0 by two to three orders better than the fitted series it replaces; a previously reported u^6 coefficient of -0.0648 is corrected here to -65/1024 and identified as truncation. The trace-free half is treated in parallel. A Clausius relation with a single temperature holds exactly when that half is proportional to the region's own Hamiltonian. At a ground state it never is, and the measured temperature is zero, which is correct rather than a failure. In a thermal state the same instrument returns 1/beta to five significant figures, and the boundary of the region where it does so collapses onto two dimensionless ratios. Two further results are included. The offset vanishes when the two cuts sit at the same phase in the unit cell, without requiring chirality, time reversal, inversion or cleanliness, and a mirror-symmetrised disordered chain gives a larger offset than its unsymmetrised control, which rules reflection out as the mechanism. And in two dimensions the offset is exactly additive along the boundary for a transverse coupling proportional to the identity in sublattice space, for a reason short enough to state, with a derived crossing past which additivity fails and a sigma_z coupling that breaks it smoothly. Every claim carries one of four grades: [T] proven here, [V] computationally verified with stated controls and a stated failure mode, [C] imported from the cited literature, [H] interpretation. The two exact lines, the commuting deformation, the boundary-locality of the entanglement spectrum, and the first law of entanglement entropy are not this work's and are used as given. Sixteen retired claims, including two withdrawn from earlier versions of this manuscript, are listed with the measurement that killed each. Files in this deposit: the_offset_belongs_to_the_boundary_v8_submission.pdf — 35 pages, gate output verbatim, script listings supplied separately. the_offset_belongs_to_the_boundary_v8_complete.pdf — 59 pages, identical body with all scripts printed inline. offset_v8_scripts.zip — 23 files. Every number quoted in the paper is printed by one of them; the two gate files reproduce their stored output byte for byte. KEYWORDS entanglement Hamiltonian; entanglement spectrum; modular Hamiltonian; Bisognano-Wichmann; Rice-Mele model; SSH model; free fermions; boundary charge; cut-phase charge; corner transfer matrix; Toeplitz operators; Wiener-Hopf factorisation; Szego-Widom asymptotics; algebraic Heun operator; complete elliptic integrals; entanglement first law; area law; Landau-Widom counting function ADDITIONAL NOTES Supersedes earlier versions of this manuscript. Two claims present in versions 1 to 3 are withdrawn and recorded in the kill list: that the incommensurate offset has no value, and that the offset is one quantum. Both were double-precision truncation of a finite sum. The u^6 coefficient of the shape function reported in the companion results ledger is likewise superseded by the derived value. Computational verification, gating and drafting assistance by an AI collaborator (Claude, Anthropic). Every gate is independently re-runnable from the ancillary scripts.
The Ledger Outlives the Metric: Wall Species, the Reciprocity Obstruction, and a Measured Crossing Behind an Operator-Side Reading of Entropic Gravity We identify the operator-side object that licenses two recent programmes deriving gravity from relative entropy, and we measure where that license expires and what survives its expiration. For a family of projectors Π(θ), a metric–curvature inequality holds if and only if a quadratic form Q built from the round-trip transition amplitudes is positive semidefinite — guaranteed when Π is self-adjoint, and losing that guarantee, not the inequality itself, otherwise. We prove this is one of three distinct wall species of differing codimension, with a licence locus 4 det g ≥ Ω² nested strictly inside the metric-definiteness locus det g ≥ 0. We identify a reciprocity obstruction: for B-symmetric generators under B-symmetric probes, the round-trip form is forced to a perfect square, and det g vanishes identically at every parameter value and every degree of non-Hermiticity — not approximately, not on a measure-zero set. Measured on a hyperboloidal Pöschl–Teller generator, the gauge-invariant witness W sits at the instrument floor (≤1.1×10⁻⁷) across 28 operating points, seven orders of magnitude below a non-reciprocal control. Breaking reciprocity lifts W linearly in the perturbation, log-log slope 1.000 against a pre-registered prediction of 1.00 ± 0.15. The paper's central measurement crosses the metric wall directly. On a six-dimensional generator with a tunable dial, det g changes sign at κ* = 0.627127751476while the contour eigenvalue count and the regularised trace-log log-determinant — the ledger — hold at 4.4×10⁻¹⁶ and 6.7×10⁻¹⁶ respectively, identically on both sides of the wall and at it. The metric fails; the ledger does not notice. We extend the record layer to two further checkpoints, each posed with its bar stated before the measurement. The ledger's response kernel on a gapless three-dimensional lattice generator extrapolates to 4πrG(r) → 1.001, the Newtonian potential, with its Hessian the inverse-square law; gapping the same generator gives a Yukawa kernel with the predicted mass. The trace-log of a region scales as an area with a logarithmic correction at the gapless point (Widom form, fit to 0.2–1.1%) while its count scales as the volume, splitting the record's two halves across the two plateaus classically expected of an entropy and a charge. A companion result, proved independently in normalised Hamming space, shows why this robustness cannot be accidental: a regularised trace-log that is Lipschitz in a metric forgiving local surgery is thereby disqualified from marking any wall in that metric — the same property, read from two sides, as both the survival and the metric's failure. Twenty-two further theorems and measured results are established, including a positive semidefinite reading of the metric as a symmetrised pairing of two families (self-adjointness recovering the Gram structure exactly), an eta-invariant argument placing the surviving quantity inside established Atiyah–Patodi–Singer index theory, and a clock exponent — 1 for semidefinite generators, 1/2 for indefinite ones — matching the two known horizon lapse laws structurally rather than by fit. The junction with entropic gravity is stated at the grade it has earned. What is proved is that a relative entropy and the operator ledger are both built from a trace of a matrix logarithm, that neither requires a definite pairing, and that both remain finite exactly where a Gram determinant changes sign — which is, at the level the mathematics supports, why the recent horizon-entropy programmes' substitution of an entropy for a metric works where it does. No horizon is derived here, and no claim is made that any of the generators tested is a spacetime. That identification is stated as a reading, graded accordingly, and left as the paper's open problem. 63 pages, 8 figures, 22 theorems/propositions, 30 numerically gated results with captured scripts and output, complete reproduction package included.
I study the modular data of a free harmonic field on a periodic lattice with a positive mass regulator, restricted to ball-shaped regions. The entropy response of a region to a weak band-limited probe is a sum of first-order shifts of the symplectic eigenvalues, weighted by a logarithmic factor that diverges at the band edge. Because a ball has many symplectic eigenvalues extremely close to one half, this weighted response is not controlled by any uniform-continuity argument. The paper asks what can be said rigorously about it, and what the answer implies for the entanglement-equilibrium programme.The main results are: (1) an exact factorization of the region data through a boundary Schur complement, which writes the deviation of the symplectic eigenvalues from one half as half the squared singular values of an explicit boundary operator; (2) a trace bound on the number of symplectic gaps above a threshold, with an area-scaling form based on exponential locality, so that both parts of the band-edge counting hypothesis of the companion analysis become theorems at every fixed positive mass; (3) a bound on the weighted first-order response by the operator norm of an explicit positive operator of rank at most 4b times the entropy weights of the 4b smallest gaps, proved by a majorization argument that uses neither locality nor translation invariance; (4) a finite-strength version based on the Lidskii-Wielandt inequality, and a gap-independent bound of order tau times the logarithm of the ratio of the region size to tau on the exact entropy increase; (5) a scheme-independence theorem: the exact statements hold for every positive matrix on every finite set and depend on the discretization only through the restricted covariance data, the ranks, and the region size; (6) explicit power-law bounds in the radius on the band-edge part and on the resolved part of the first-order response, valid for a class defined by three stated hypotheses; (7) a dichotomy for the rank-based strategy: a vanishing envelope forces the whole response ratio to vanish and is therefore incompatible with the conformal identification, so any proof of the identification must resolve individual modes; (8) a quantitative saturation bound at fixed probe strength showing that the first-order divergence at the edge is a small-strength artefact and that, along the top of the trace-small window, the two orders of limits do not commute.Numerical results are given for a two-dimensional model with mass 0.01, obtained in extended precision with a companion program that uses an exact reflection-symmetry block decomposition. The smallest symplectic gaps fall below 10 to the minus 29 for balls of radius up to 12. The response ratio to the area term does not approach the conformal prediction in the accessible window for small band centers, the edge share of the response lies between about 0.2 and 1.1 without a visible trend, and the locality constants of the analytic part are not informative in this regime. The limits of the results are stated explicitly: no proof that the model belongs to any of the classes for which the conditional statements hold, no lower bound on the smallest gap, no massless counting theorem, and no dynamical statement of any kind.All numerical data are produced by two self-contained Python programs, published together with the paper, that reproduce every table and print all internal consistency checks.
We prove the ground-state energy of the periodic, half-filled 4×4 repulsive Hubbard model with t = 1 to at least 23 decimal places at U = 1, 2, 3, 4, 5, 6, 7, 8, 10, 12 and 16: every enclosure has width below 10⁻²⁴. The double occupancy, the spin correlations and the staggered structure factor have certified interval widths below 10⁻¹⁰, and the kinetic energy below 10⁻⁹. Hypercube, spin and pseudospin symmetries reduce the problem to a 19,186-dimensional block that contains the ground state by Lieb's theorem and a character argument. A binary64 Cholesky factorization, verified using a rigorous rounding-error bound evaluated in exact rational arithmetic, certifies spectral separation within this block; exact Rayleigh quotients and residuals then give Temple enclosures of the energy and bounds on the angle to the ground state, which certify the observables. Separation certificates at both ends and a perturbative trial family enclose E₀(U) for every U in [3.9, 4.1], with uniform width at most 1.6 × 10⁻⁹. Two independent implementations replay the certificates, the interlacing and Temple steps are proved in Lean 4, and all computations run on one laptop. At U = 4 and 8, exact-arithmetic brackets that use neither Lieb's theorem nor floating point are also given, and the enclosures bound their deficits. At U = 4 the certified energy rounds to −13.62185 and excludes the published value −13.62192 [Physical Review B 94, 085103]. Contents of the bundle paper/ — the manuscript (Markdown and PDF), its figures, and supplementary/ with five videos generated from the certificate files. rayleigh/symmetry/ — the enclosure pipeline (block dimensions, exact basis, certificates, observables, the interval certificate) and recheck/, the independent Rust reference checker; certificates/ holds every certificate, trial vector, run log, checker log and the clean-replay report. sos/, rayleigh/demo/, cluster/ — the exact-arithmetic bracket certificates and the expected outputs of their checkers (golden_ files); the bracket checkers belong to the authors' certification software and are not included. formal/ — the Lean 4 statements (rank-one interlacing, Temple's inequality, the sum-of-squares remainder bound). tests/ — the generators of every table, figure and video of the manuscript. replay.sh, REPLAY.md — the automated replay and the step-by-step commands. AUDIT_AXIOMS.txt — #print axioms of the formal statements. SHA256SUMS — checksums of every file. Verification PYTHON=<python with numpy and scipy> ./replay.sh rebuilds the symmetry block from the Hamiltonian, reproduces every certificate from the stored integer vectors, runs the Rust reference checker, recomputes the observables and checks the manuscript's tables, and writes REPLAY_REPORT.md. It needs about 12 GB of memory and 16 GB of free disk, and takes about half an hour on the hardware above; its work directory is removed at the end. REPLAY.md lists the commands step by step; rayleigh/symmetry/certificates/REPLAY_REPORT_2026-10-14.md is a clean replay of this material. Not included: the two 662,547,600-byte full-sector vectors of the upper bracket certificates (their SHA-256 digests are in rayleigh/demo/). Integrity: shasum -a 256 -c SHA256SUMS inside the bundle; the SHA-256 of the tarball is in the .sha256 file beside it. Environment Python: 3.13.7 with numpy 2.5.4, scipy 1.18.1, matplotlib 3.11.2 (producer-requirements.txt) Rust: 1.90.0 (reference checker; cargo build --release in rayleigh/symmetry/recheck) Lean: leanprover/lean4:v4.33.1 with Mathlib v4.33.1 (formal statements) Hardware: Apple M3 Pro, 18 GiB (all production and replay) License: CC BY-NC 4.0. Commercial use requires written permission from Origin Neural AI (labs@smartledger.solutions).
When the average sign of a fermionic quantum Monte Carlo simulation reads zero, the zero can mean threedifferent things: the run was too short (a data zero), the chosen Hubbard-Stratonovich decoupling created it (a modelzero), or no decoupling in reach removes it (a principle zero). We carry the tagged-zero vocabulary of an earlier noteover to the sign problem and test it in three seals on the two-dimensional Hubbard model (U = 4, 4x4 periodic lattice,plus 4- and 6-site clusters where the sign is known exactly by enumerating every auxiliary field), each run once onpoints unused during development. Model zero: with identical partition functions (relative difference below 2e-11), thespin decoupling keeps above 0.998 while the charge decoupling drops the average phase to 0.0003. Data zero:with the exact sign 0.035 known, the share of 100 chains that read zero fell from 0.94 to 0.52 at the predicted pricetag n* = 1603 sweeps and to 0.00 beyond it, as the error law says. On the 4x4 lattice a sign extrapolated from beta4-7 to beta 8 ([0.041, 0.145]) was met (0.106); at beta 10 the run read a data zero, as forecast. Principle zero: acrossa family of decouplings (discrete and Gaussian, spin and charge, and a continuous mixing parameter alpha built fromthe identity (n-1)^2 + m^2 = 1, the best member is pure spin, where the sign sits on a cusp: moving alpha either waycosts sign, more steeply on the imaginary side (slopes 7.9 and 5.0 per unit alpha at beta 6), and the decay with betaremains (0.48 to 0.18 from beta 6 to 8). Pure spin is thus empirically optimal among the tested members, and the tagreads "principle, relative to this family", not a complexity-theoretic statement. The half-filling sign, often calledprotected, is protected only by symmetric decouplings: alpha = -0.5 drops it to 0.41. Of 19 sealed forecasts 17 cameout R, one U and one F (cusp slopes not scaling linearly with beta). Checks made after AI comments on a draft showtwo verdicts to be fragile: with conservative binning errors one of four Q1 comparisons falls to 1.9 standard errors, andthe F of O4 becomes U once the shared reference point is treated as correlated; the cusp also appears in theGaussian family. The work is a check on how zeros are read, not a new algorithm; it is meant as a small, reproducibleentry point to the sign problem for readers of the tagged-zero series. (1) Text: CC-BY-4.0 | Code: GNU Affero General Public License v3.0 or later (AGPL-3.0-or-later) (2) Data and code. Seal documents with SHA-256 files, result files, development logs, the interruption log and allanalysis and figure scripts (GNU Affero General Public License v3.0 or later) are inBridge_Paper_Tagging_the_Zeros_of_the_Sign_Problem_scripts_v1.0.0.zip. No external data are used. (3) AI use. Designs, code, analyses and text were produced by Kurado (Claude Opus 5.5, Anthropic) working with M. Watabe, who chose every direction. Draft comments by ChatGPT and Gemini, relayed by M. Watabe, prompted the checksin Section 7; they are AI commentary, not peer review.
Shared splittings are equal signed frequency differences in the labelled transition families of two nuclei.We prove a trace-hierarchy theorem for 𝑁 ≥ 2 coupled spin-½ nuclei: the traces of the total-𝐼 𝑧 blocks determine exactly 𝑁 -1 linearly independent shared-splitting functionals of the energy levels.Their energy-level weights are the Krawtchouk polynomials 𝐾 𝑟 (𝑘; 𝑁) , 𝑟 = 2, … , 𝑁 .For the pairwise scalar-coupling Hamiltonian, the member 𝑟 = 2 gives the spectator-averaged signed sum rule ∑ 𝑆 𝑖𝑗 av 𝑖<𝑗 = ∑ 𝐽 𝑖𝑗 𝑖<𝑗 , and every member 𝑟 ≥ 3 vanishes.The highest member gives complete spectator independence for three spins; for 𝑁 ≥ 4, the hierarchy includes 𝑁 -3 intermediate relations implicit in the classical block-trace formula.Completeness follows by intersecting the shared-splitting weight space with the space of weights constant on magnetization blocks.The underlying four-level shared-splitting equality is algebraic at any coupling strength.Its observation requires non-zero transition moments, sufficient signal and resolved line positions.Its value equals the coupling constant for an isolated pair, but need not do so in larger systems.The trace functionals are invariant under relabelling within each block and respond selectively to added longitudinal product terms of rank three or higher.We specify a continuation-based convention for individual labels, subject to non-degeneracy, and give worked AB, ABC and ABCD examples.We provide high-precision numerical evidence of global non-uniqueness in parameter recovery from shared splittings alone.On noisy simulated spectra with interleaved multiplets, we test candidate assignments using fitted frequencies and their covariance.Completeness is relative to the block traces; stronger affine-relation and generic-rank claims remain open.