Gated Gaussian Lorentzians

Peak-Matched Bounded-Gate Approximation, Baffled Ray Optics, 1/x² Wing Asymptotics & Robust Estimation

L. Van Warren, PhD & Google Gemini 4 Carbon High · Drylab v1.0

Math Treatise →

Preamble: Replacing a Ravenous Beast With a Housecat

In the wild, the Lorentzian (Cauchy) distribution is an uncontrollable, ravenous beast: its heavy algebraic wings devour every statistical moment, leaving its mean undefined, its variance infinite, and any unguarded estimator ambushed by extreme outliers. Yet indoors—within a finite gate of a few standard deviations around its crest—that ferocious silhouette looks remarkably polite. By pinning a Gaussian to the exact peak height of the Lorentzian and gating our operating window to a chosen span of standard deviations, we swap an untamable predator for a domesticated housecat: analytical, finite-moment, and effortless to integrate, while keeping honest watch over the claws still lurking in the wings. Meow.

1. Interactive Gated Approximation & Baffled Ray Bench (Indoors: The Housecat)

STAT-RAY 99.6 · CSS3-UX 99.6 · DRYLAB-QA 99.8
Auto-Tune Gaussian SD:
Peak-Pinned Gated Gaussian vs. Lorentzian (In-Gate & Wing Error Shaded) Hover plot to inspect pointwise error
Physical Ray Emitter & Aperture Baffles (Why Gating Tames the Beast) Accepted vs. Baffled Wing Rays
Shared Peak Crest—
Lorentzian Gate Mass—
Gaussian Gate Mass—
In-Gate Abs Error (L₁)—
In-Gate Net Signed Error—
In-Gate RMS Error—
Uncaptured Wing Mass—
Rays Inside Gate—
Ungated → Gated Ray SD—
Curious Cat’s Derivation Drawer: Symbolic Integration First, Then Numerical Specifics MATH-SYMB 99.8 · Native Offline MathML

Peak-Pinned Housecat Surrogate & Gate Mass

Pinning a Gaussian crest to the Lorentzian peak L(μ) = (πγ)⁻¹ sets amplitude A = (πγ)⁻¹ with tunable width σ over gate G = [μ − Nσ, μ + Nσ]:

ML(Nσ) = 2π arctan(Nσγ) , MG(Nσ) = σ2ππγ erf(N2)

Modulus Cusp & Sample-Mean Invariance

Upper-half-plane residue calculus around z_+ = μ + iγ yields φ_X(t) = exp(iμt − γ|t|); homogeneity of |t/n| · n = |t| makes X̄ₙ invariant:

φX¯n(t) = [exp(iμtn−γ|tn|)]n = exp(iμt−γ|t|)

Fisher Bound vs. Copas Unimodal Profile

Outside the gate, joint Cauchy MLE achieves Var(x̂₀) ∼ 2γ²/n on Copas's unimodal envelope ℓ(x₀, γ̂(x₀)), while median & half-IQR achieve 8/π² ≈ 81.06% efficiency.

2. Wild Far-Wing Log₁₀ Asymptotics (±25γ), Voigt Convolution & Unbaffled Lighthouse

MATH-PROB 99.6 · JS-SIM 99.7
Probability Density Comparison (Hover cursor to inspect) Move pointer over plot
Unbaffled Origin: Uniform Angle Θ ∈ (−π/2, π/2) → X = x₀ + γ tan Θ

3. Wild-Wing Law of Large Numbers Breakdown: Running Mean vs. Running Median

STAT-NUM 99.8 · CSS3-UX 99.6
Running Estimator Trajectory vs. Sample Count n (True x₀ = 0) Hover trajectory to inspect n
Draws n0
Gaussian Mean X̄ₙ—
Gaussian Sample SD—
Cauchy Mean X̄ₙ—
Cauchy Sample SD—
Cauchy Median X̃ₙ—
Max |Cauchy Draw|—

4. Wild-Wing Peak Estimation Showdown: Mean/SD vs. Median/Half-IQR vs. Cauchy MLE

STAT-NUM 99.8 · MATH-PROB 99.8
Relative Log-Likelihood ℓ(x₀) − ℓ_max Hover curve to inspect ℓ(x₀)
500-Trial Location RMSE vs. Cramér–Rao Bound (√(2/n) γ)
Estimator Method Location x̂₀ Scale γ̂ (or SD) |x̂₀ − x₀| Error 500-Trial RMSE