We prove a local amplitude-ratio bound for positive ground states and an explicit reverse comparison for fidelity under finite-region erasure. An exact lattice calculation for a connected critical Ising chain gives amplitude exponent
1/8, preserved by arbitrary optimal control on any fixed spin set and, at the level of logarithmic exponents, on
o(logN) spins. We derive the finite Cauchy product and its uniform strip asymptotics, and document exact two-spin optimization and an interferometric protocol for measuring the recovery. A separate finite-mediator theorem quantifies local compression and a decoder while retaining the rest of the environment exactly. We also give a ground-preparation Ramsey bound, counterexamples to stronger projected-state inferences, and independent full-Hamiltonian calculations.
Bostelmann, Cadamuro and Minz noted that they presented no proof of convergence of finite-dimensional modular data to the continuum, and that this requires further work; the closest constructive framework, the operator-algebraic scaling-limit programme, likewise leaves modular-data convergence untreated. For half-line free fermions we prove such a theorem for compact-local wavelet schemes, and one delimiting it. Positively, for gauge-invariant quasi-free CAR states, strong exhaustion, strong embedded covariance convergence, and endpoint faithfulness imply norm convergence of local state restrictions and strong-resolvent convergence, after faithful block-completion, of the log-odds modular generators; Trotter–Kato gives local modular-automorphism convergence. We verify these inputs for admissible schemes, including the pinned Daubechies embedding along a dyadic refinement sequence and, under a multiresolution-exhaustion hypothesis, the fixed-order Cohen–Daubechies–Vial boundary model; in the pinned model, local state restrictions on smooth windows away from the endpoint converge at rate O(a^{s}), 0<s ≤ p fixed, those reaching it at rate O(a^{1/2}), the exact order of the driving defect for p ≥ 2 with nonzero endpoint value. Negatively, the locality of this convergence is sharp within a subclass (compactly supported orthonormal-translate interior generator, boundary-free pullback modulo compacts, alias-free, and satisfying the hypotheses of the abstract theorem) for the raw (un-renormalized, infrared-unmodified) fixed-cutoff comparison: a non-exact passband at an interior Fermi-window point makes both Powers–Størmer–Araki square-root defects noncompact, so that comparison supplies no same-folium global normality, whereas exact passband with trace-class boundary error does. Compact support forbids exact passband: the locality yielding local modular convergence obstructs the raw global comparison.
Neural renormalization-group (RG) methods and machine-learned holographic maps train a network on lattice samples or on a lattice action and read physics off the result: coarse couplings, critical exponents, or an emergent bulk geometry. We ask which of these outputs the training objective and the readout themselves determine. For a bijection trained by a divergence to a factorized target with a free coarse coupling, every admissible coupling attains zero loss through a smooth optimal-transport map that respects every symmetry acting identically on field and latents, so neither more training nor an equivariant architecture selects the coupling. Restricting the class can: in the Gaussian chain a zero-loss linear map that is local below the coarse correlation length has the decimation coupling, the threshold is sharp, and on the square lattice no finite-range linear map reaches zero loss. A Gaussian kernel on a network graph yields a distance bounded above and below by the graph distance whatever the data, and exact wavelet latents of a power-law Gaussian give an isotropic hyperbolic geometry except along chains through the wavelet's jumps, which a free-fermion computation ties to the anisotropy reported for the exact holographic mapping. We then re-examine three published pipelines. A hyperbolicity test of neural-RG holography is passed by kernels built without boundary data. In RGFlow at the reported sizes, general linear maps reach zero loss at a continuum of couplings, whereas in a small, unconverged φ⁴ computation RGFlow's own RealNVP class selects a coupling through its Jacobian structure, close to the decimation value in the Gaussian sector. In inverse RG of the three-dimensional spin glass, with a ten-parameter stand-in upscaler on small lattices, the one clean upscaler that closes under blocking reproduced the forward Monte Carlo RG fixed point of its training pair, non-closing ones drifted, and the lower training loss went to a non-closing upscaler in both training pairs. The sharp results are Gaussian. We close with diagnostics, each with a recipe.
We study boundary criticality at the Nishimori multicritical point of the two-dimensional (2D) random-bond Ising model. Using tensor-network methods, we construct a family of microscopic boundary conditions that incorporates both boundary-spin rotation and boundary disorder. We identify three conformal boundary fixed points, corresponding to free, fixed, and random boundary conditions, and map out the boundary renormalization group flows among them. We extract the corresponding boundary conformal data, including the boundary entropies and the scaling dimensions of boundary primary operators, which characterize the boundary universality class. At the free boundary fixed point, we uncover the multifractal scaling of boundary-spin fields. We further complement the numerical results with a controlled renormalization group analysis. Finally, we connect the boundary conformal data to quantum error-correcting codes, establishing a bridge between boundary universality class and boundary decoding threshold.
Local information can be lost rapidly while a collective mode survives for a long time. We study this distinction on a physical–observation joint state space. The second variation of relative entropy under a local record map gives a positive Fisher-loss operator; the least loss over normalized noninvariant scores is the spectral gap of the associated posterior-resampling process. The local strength and the compatibility of surviving directions enter separately. A fixed-width interacting ladder has a positive gap at every fixed finite coupling, as follows from an explicit block comparison, although strong correlation makes the bound small. We then reverse the construction. For a given irreducible real Hamiltonian with nonpositive off-diagonal entries, its ground amplitude determines pair-record probabilities and energy-valued loss weights, reproducing the physical excitation gap exactly. A positive trial amplitude gives an error interval controlled by the oscillation of its local energy, without requiring the exact ground state. When signs remain, the energy form is a difference of positive information forms and an explicit cancellation constant controls the gap. Sparse diagonalization of open transverse-field chains up to 65 536 configurations agrees with both the reconstructed information spectra and an independent free-fermion calculation. The critical law n∆n/J → π, the gapped regime, and the ordered finite-chain splitting are distinguished. The result is an explicit connection for specified channels and Hamiltonians, with the requirements of positivity, locality, and energy normalization kept separate.
We test the Gluing Method on the square-lattice two-dimensional Ising model by reconstructing global free-energy generators from complementary local relations. The emphasis is kept explicit throughout: the inputs used as \emph{gluing material} are separated from quantities obtained \emph{after gluing} and from data reserved for validation. At zero magnetic field, Kramers--Wannier duality is absorbed into the coordinate
τ=logsinh2K, so that the reduced free energy decomposes as
Φ(τ)=τ/2+G(∣τ∣). The gluing materials are then only a critical logarithmic local expansion and a low-order high-temperature outer expansion; the low-temperature sector is generated by duality rather than fitted independently. Positive flat weights preserve the prescribed critical asymptotics to every finite algebraic order. With one crossover parameter fixed from three scalar anchors, the resulting bulk free energy agrees with the exact Onsager benchmark at a maximum relative error of
1.915×10−5 on the tested interval. Internal energy and specific heat are derived from the same glued generator. Zero-field finite-size corrections are organized separately as one-variable scaling functions multiplying
L−2,L−4,…. At nonzero field in the thermodynamic limit, the gluing materials are the established high-temperature, critical, and low-temperature local expansions of the universal Ising magnetic scaling free energy
Φmag(η),
η=m/∣h∣8/15. The low-temperature input contains the
∣h∣ cusp associated with spontaneous symmetry breaking. Mapping the glued scaling free energy back to the square lattice through nonlinear scaling fields yields a single generator for the magnetic free-energy increment, magnetization, and susceptibility. The construction reproduces high-temperature linear response, the critical isotherm
M∝∣H∣1/15, and the two one-sided limits
M(H→0±)=±M0 without separately fitting magnetization branches. Finally, finite size and finite field are combined in the scaling variables
xt=Lm and
xh=L15/8h. The two protected one-dimensional axes, a central mixed-response expansion, a low-temperature spectral-rounding relation, and a massive outer description are glued into a two-variable scaling surface
Yglue(xt,xh). The resulting lattice free energy closes in the compact form
ΦLglue(K,H)=Φlatglue(K,H)+L2Yglue(Lm,L15/8h)−Ybulk(Lm,L15/8h)+O(L−4) where
m=m(r,H) and
h=h(r,H) are the nonlinear scaling fields and
Ybulk is the thermodynamic singular scaling surface expressed in the same variables. Thus all finite-size/finite-field information is carried by the excess
Yglue−Ybulk, while the thermodynamic lattice free energy remains the previously constructed
Φlatglue. On eleven transfer-matrix points withheld from the construction, the blind free-energy test gives RMS absolute error
3.90×10−4 and maximum absolute error
9.55×10−4 in
Y. Differentiation of the same free energy produces finite-size magnetization and susceptibility; magnetization is especially stable on the low-temperature side, while the second derivative is more sensitive in crossover regions.