costAlphaLog_high_calibrated_iff
plain-language theorem explainer
The theorem equates high-calibration of the α-cost (fourth derivative of G at zero equals one) with the algebraic condition α² = 1. Branch-selection work in Recognition Science cites it to isolate the canonical J-cost inside the bilinear family. The proof is a one-line wrapper that unfolds the calibration predicate and substitutes the fourth-derivative evaluation lemma.
Claim. Let $G_α(t) = α^{-2}(cosh(α t) - 1)$. Then $G_α^{(4)}(0) = 1$ if and only if $α^2 = 1$, for $α ≠ 0$.
background
The module AlphaCoordinateFixation formalizes Option 2 of the three candidate α-fixations listed in the companion branch-selection paper. It works inside the one-parameter family obtained from the bilinear branch theorem: $G_α(t) = α^{-2}(cosh(α t) - 1)$ for α ≥ 1. The second derivative at zero is identically 1 for every α, so unit-curvature calibration does not distinguish members of the family. Higher-derivative calibration is defined by the predicate IsHighCalibratedLog G, which requires the fourth derivative of G at zero to equal 1.
proof idea
The proof is a one-line wrapper. It unfolds the definition of IsHighCalibratedLog and rewrites the fourth-derivative evaluation supplied by the upstream lemma costAlphaLog_fourth_deriv_at_zero, which states that the fourth derivative of CostAlphaLog α at zero equals α².
why it matters
This result supplies the algebraic half of the α-pin theorem (alpha_pin_under_high_calibration) and is recorded inside the certificate alphaCoordinateFixationCert. It closes the higher-derivative route to J-uniqueness inside the Recognition framework, complementing T5 (J-uniqueness) and the eight-tick octave structure. The other two fixation routes (generator calibration and action minimization) remain open.
Switch to Lean above to see the machine-checked source, dependencies, and usage graph.
papers checked against this theorem (showing 30 of 223)
-
Noise smoothing on truncated means yields exponential tails from second moments
"ψ(u) = u−u³/6 on [−√2,√2] ... −log(1−u+u²/2)≤ψ(u)≤log(1+u+u²/2)"
-
Timed light raises detected molecules and lowers errors in molecular links
"photolysis reaction ... S + hυ → P with rate J"
-
Approximate models enable efficient estimates for damped sinusoids
"Proposition 2 ... β0 ∈ (β, β + γ (tN + tN−1)]"
-
Diffusion models traverse full RDP surface without retraining
"score-scaled probability flow ODE decoder"
-
Black hole thermodynamics links swampland to dark matter
"V(ϕ) = -e^{-2aϕ}/ℓ_p² ... P = -3V(ϕ)/8π ... f(r) = 0 as state equation"
-
Coordinated rotations of antennas and IRS panel raise MIMO sum rates
"Ge(ε) = G0 cos^{2p}(ε)"
-
Nonlinear correction fixes RNA-seq sample biases
"log2(bXj + c) ≈ log2(b) + log2(Xj) + c/(bXj) (multiplicative bias becomes additive shift after log)"
-
Max-entropy relaxation solves multiway k-cut on graphs
"free energy L … temperature 1/β … as β → ∞, L → D … annealing … phase transitions"
-
Strong b-process responses raise addiction risk most in model
"Wk(t|…) = sum of exponential terms … W(θ) = Γa∆/α (1 − Γb/β)"
-
Scale-dependent eddy density matches wall turbulence mean and variance
"exact closed-form expressions for the mean influence function … mean-variance duality: horizontal head determines I₁ while inclined legs dominate I_φ"
-
Rényi divergence keeps spectral estimates stable under frequency outliers
"gradient G_α(θ;eIn) and Hessian bound L = U1² + U2/(1-α) independent of contamination z (Remark 1)"
-
Closed-form newsvendor quantity handles misspecification too
"Transform function φ_α(v) = α/p v² (or piecewise) yielding T_φ[G]"
-
Scattering transforms limit over-smoothing in graph message passing
"EG(X) = EG(W0,J X) + Σ EG(Wr,l X) … tightness of Framelet system requires … bα(λℓ/2J)^2 + Σ db(r)(λℓ/2J)^2 = 1"
-
Information projections match between geometry and probability theory
"Legendre-Fenchel transform ... free energy F(μ|p)=log∑pieμi"
-
Training-free method restores high-res diffusion output
"E[x²_t] = sum x² / (C H W) ... as ω increases, the energy exhibits a gradually increasing trend"
-
Moments estimator recovers edge weights in reinforced random walks
"Ev[√U_e]=a_e/2·Γ(½(a_i+1−…))…; DKL expressed via ψ and lnΓ"
-
Log filtering yields accurate 2-bit KV cache compression
"LogQuant ... log-distributed high-attention pattern ... log2 sparsity selection"
-
FLARE 3B beats larger models using only 630 vision tokens
"Text-Guided Unified Vision Encoding ... joint attention between mapped text and visual embeddings inside encoder layers"
-
Dark photons add mass-suppressed corrections to Schwarzschild black holes
"ρ(r) = 1/4π ΔV(r) ... rf'(r) + f(r) − 1 / r² = −8πρ(r) ... f(r) = 1−2M/r − (gD² mA'/2π) e^{-mA'r} − ..."
-
Convex neural nets solve stochastic programs as linear programs
"Q(x, ξs) is the pointwise maximum of affine functions... Thus Q(x) is convex in x"
-
Frequency theory tracks fast and slow dispersion in porous media
"ansatz … c̃ = a + b·∇⟨c⟩ … closure problems for a and b"
-
Dual² algorithms cut complexity in decentralized optimization
"F_ρ(y) = H^*(−√C y) ... Condition 1: F_ρ is convex and L_Fρ-smooth"
-
Smoothed support function turns conic DC problems into single-inequality subproblems
"hμ(y) := μ log(∑ exp(λi(y)/μ)) + 10^{-5}μ (for λmax on Sm)"
-
Harmonic sampling raises PIMC acceptance rates six- to sixteenfold
"harmonic Trotter splitting e^{-τ Ĥ} ≈ e^{-τ/2 V_anh} e^{-τ Ĥ_ho} e^{-τ/2 V_anh} removes harmonic Trotter error exactly"
-
Attractor networks self-orthogonalize via free energy minimization
"Langevin function ... continuous Bernoulli ... J† symmetric part governs stationary distribution"
-
Task vectors transfer between models of different widths without retraining
"min ∥Hin,A τA⊤ Hout,A⊤ − Hin,B τB⊤ Hout,B⊤∥F"
-
Low-z Little Red Dots match high-z versions in spectra and shape
"We parameterize their UV-optical continua using a simple broken power-law model ... fλ = fλv (λ/λv)^kblue ..."
-
Cosmological wavefunction coefficients simplify via Grassmannian
"AΔ,J4,s = N(Δ,J) E^{-1} (2S)^J P_J((U-T)/S) 3F2[1,J+1,J+1;Δ+J,3-Δ+J | 2S] + homogeneous 2F1 terms"
-
Clipped Linear Lottery stabilizes randomized picks with near-optimal regret
"L-smoothness: ||p(X) - p(X')||_1 <= L ||X - X'||_{1,1}"
-
Entmax turns KV cache truncation into exact support recovery
"α-entmax(s)i = [(α−1)si − τ]_{+}^{1/(α−1)} ... support S = {i : (α−1)si > τ}"