costAlphaLog_fourth_deriv_at_zero
plain-language theorem explainer
The fourth derivative at zero of the alpha-parameterized cost in logarithmic coordinates equals alpha squared for nonzero alpha. Workers on branch selection in Recognition Science cite this identity when applying higher-order calibration to isolate the canonical reciprocal cost J. The argument is a direct term application of the fourth-derivative existence result at the evaluation point, followed by algebraic simplification.
Claim. Let $G_α(t) := (1/α²)(cosh(α t) - 1)$ for $α ∈ ℝ ∖ {0}$. Then the fourth derivative of $G_α$ at $t=0$ satisfies $G_α^{(4)}(0) = α²$.
background
In the Alpha Coordinate Fixation module, higher-derivative calibration is developed as one route to select the parameter α in the bilinear family of costs. The relevant function is the log-reparametrized cost G_α(t) defined by (1/α²)(cosh(α t) - 1). Its Taylor expansion at the origin begins with the constant term zero, linear term zero, quadratic term t²/2 (hence unit second derivative), cubic term zero, and quartic term (α² t⁴)/24.
proof idea
The proof applies the upstream lemma establishing the general fourth-derivative formula at the point t=0 to obtain the expression α² cosh(α · 0). It then rewrites the derivative value and simplifies the hyperbolic cosine at zero together with the zero-multiplication rule.
why it matters
The result is invoked directly by the high-calibration equivalence theorem, which states that the α-cost is high-calibrated precisely when α² equals one. This equivalence is packaged into the alphaCoordinateFixationCert definition that collects the fixation certificates. It realizes the higher-derivative calibration option from section 5 of the branch-selection paper and thereby contributes to forcing the coordinate α to one, which recovers the J-cost via the uniqueness result.
Switch to Lean above to see the machine-checked source, dependencies, and usage graph.
papers checked against this theorem (showing 30 of 115)
-
Quantum gravity breaks acceleration-thermal equivalence for atoms
"⟨dH_A/dτ⟩_VF contains ... (1 + 5a²/ω² + 4a⁴/ω⁴) multiplied by the Unruh factor"
-
Continued fraction blocks match transformers at half the size
"continuants K0=1, K1=ad, Kk=ad−k+1 Kk−1 + Kk−2 ... ∂f̃/∂ak = (−1)^k (Kd−k/Kd)^2"
-
Proximal DC algorithm solves signed Fréchet regression on manifolds
"curvature-dependent strong convexity of the proximal subproblems... δD ≜ δ+(D), ζD ≜ ζ−(D)"
-
Self-dual solutions derived for linked-ring polymer model
"energy... splitted... into a self-dual term and a term describing short-range interactions... minimized by the self-duality conditions (Du,d a,1 + i Du,d a,2) Ψu,d a = 0... cosh-Gordon equation"
-
P1-KAN outperforms other KANs on irregular functions
"P1 finite element method... Ψl,j_p(x) piecewise linear hat functions on uniform or adapted grid"
-
New index cuts false negatives in voltage recovery checks
"The monotonically decreasing upper envelope U(t) and monotonically increasing lower envelope L(t) are constructed from the decomposed components"
-
K-FAC regularizer disentangles task vectors without data
"the Jacobian Gram matrix is an instance of the generalized Gauss-Newton (GGN) matrix"
-
Decaying hints lift non-English reasoning without drift
"Cosine decay schedule... pl_t = 1/2 (1 + cos(π t / T))"
-
Tuned lattice swaps atomic and photonic character of moving excitations
"polaritonic basis... Eαn/ℏ = nω + Δ/2 ± (Δ/2)√(1+4ng²/Δ²)"
-
Soft-edge gap corrections are multilinear in leading derivatives
"Gβ,j(s, τ; ξ) = sum Pβ,j,k(s, τ) · ∂^k_s Fβ(s; ξ)"
-
Super-resonance enables strong self-interactions for 100 GeV dark matter
"S(v) = πα_A / v * sinh(12 m_χ v / π m_A) / [cosh(...) - cos(...)]"
-
N-norm bound links scaled gradient to Laplacian in all dimensions
"Proof ... radial term via T_{2,1} and angular term via 1D Hardy (16)"
-
Warm start reduces DFT+DMFT iterations by two to four times
"physics-constrained representation tailored to the analytic structure of the self-energy ... high-frequency expansion Σ(iω_n) = Σ(∞) + O(1/iω_n)"
-
Dynamic anchors make preference tuning robust to noise
"Theorem 5.1 (Anchor Gap as Local Sharpness) … Γi(θ) ≤ ρ∥∇θMi(θ)∥2 − ½(ϵ*i)⊤∇²θMi(θ)ϵ*i + O(ρ³)"
-
Floquet analysis reveals low-loss regimes for large-momentum atom transfer
"acceleration profile a(η)L(t) involving cosh(η/2)/sinh(η/2) and exponential terms; anti-resonances suppressing Γ0"
-
Riemannian method matches supervised VPR accuracy with zero training
"PEM distance with power α=0.5 … matrix square root … Newton-Schulz iterations"
-
Group RL for LLMs reframed as explicit projection on response simplex
"max_w∈Δ^{K-1} Ĵ(w) = ∑ w_k R_k - τ D_KL(w∥P_t) ... w^*_k = softmax(R_k/τ + s_t,k)"
-
ZEBRA keeps 94% of quality on half an LLM budget
"fi(x)=ai(1−e−bix) … f′i(xi)=λ"
-
Hyperbolic operator adds L1 bias to stable sparse transformer training
"cosh-entropy ... fails inverse μ-coercivity ... hyperbolic entropy ... L1-bias"
-
Photon subtraction plus Kerr approaches 1/N² phase estimation scaling
"F2 = 4⟨Δ² n̂²_a⟩ ... strictly positive term f involves higher-order moments"
-
Bessel integral bound holds uniformly for all gamma in (0,1)
"Lemma 2.1 ... r_α(x) > x/(x+2α+2)"
-
New formula gives sample sizes for causal survival studies
"Proposition 1: eV_RCT / eV_Freed = 2 cosh(τ_0)[cosh(τ_0)−1]/τ_0² ≥1"
-
Norm growth law forecasts grokking delays under AdamW
"Vt contracts exponentially … log-linear fit … rate-preserving observable"
-
Closed-form particle flow update matches Kalman filter exactly
"D = R^{-1/2} H P H^T R^{-1/2} ... D = V Λ V^T ... Φ(λ,0) = I + E Ω(λ) F^T with ω_i(λ) = (1 + λ α_i)^{-1/2} - 1/α_i"
-
Real wormholes emerge from imaginary scalars in supergravity
"The superpotential... W(ϕ1,ϕ2) = ±2/L e^{-ϕ2/√6} [2 + e^{√(3/2)ϕ2} cosh(ϕ1/√2)] ... analytic extension ϕ1 → iη1"
-
Hybrid KAN-MLP raises F1 scores 5.33% in IMU activity recognition
"LarctanKAN... arctan(kx)"
-
Only Born-Infeld models stay natural across uncertainty and gravity
"f(G) = λ (sqrt(1 + 2G/λ) − 1) ... F(u) = λ/(2κ) (sqrt(1 + κ² u² /(3λ)) − 1)"
-
Mehler formula extends to oscillator in Aharonov-Bohm field
"phase function φ(t,x,y,η) = S(t,y,η) + (x−xt)·ξt + iB|x−xt|²/2"
-
Pink noise in economies traces to synchronized circulations
"sin[(ω + λ)t] + sin[(ω − λ)t] = 2 cos(λt) sin(ωt) ... [A(t) sin(ωt)]² = ½ A²(t) − ½ A²(t) cos(2ωt)"
-
Directional derivative measures faithfulness in zero-shot image explanations
"gc(zf + ϵ v̂t(x)) − gc(zf) via first-order Taylor"