Probabilistic foundations of Gaussian Distribution & erf.
The continuous Gaussian (normal) distribution is the cornerstone of classical probability theory and statistical inference. Characterized by its symmetric bell curve, location-scale invariance, and maximum-entropy properties, it describes the limiting behavior of aggregated random phenomena via the Central Limit Theorem. In this laboratory, we examine the analytical properties of the Gaussian density function, evaluate the Gauss error function erf(z), explore the 68-95-99.7% empirical rule, and confirm the universal Full Width at Half Maximum scaling factor m = 2.355. In mathematical statistics, empirical data science, and modern computational engineering, understanding the asymptotic sampling properties and theoretical behavior of Continuous Gaussian Distribution, Z-Scores, and Error Function Numerical Integration provides crucial guarantees. Rigorous experimental validation connects foundational probability laws with observable empirical phenomena, ensuring that statistical inference remains valid under finite sample sizes and real-world data collection constraints.
Viva voce questions & analytical theory
What is the mathematical formulation of the continuous Gaussian probability density function?
A continuous random variable X follows a normal distribution N(μ, σ^2) if its probability density function is f(x; μ, σ) = (1 / (σ √(2π))) exp(- (x - μ)^2 / (2σ^2)) for x ∈ (-∞, ∞). The parameter μ ∈ ℝ is the location parameter (mean, median, and mode), and σ > 0 is the scale parameter (standard deviation). The normalization constant 1/(σ√(2π)) is derived from the Poisson-Gauss integral ∫_{-∞}^∞ e^(-t^2) dt = √π.
How is the Gauss error function erf(z) defined and related to the standard normal CDF Φ(z)?
The Gauss error function is an entire special function defined as erf(z) = (2 / √π) ∫_0^z e^(-t^2) dt. By substituting t = u / √2, the integral can be connected directly to the standard normal cumulative distribution function Φ(z) = (1 / √(2π)) ∫_{-∞}^z e^(-u^2 / 2) du. Splitting the integral at zero yields Φ(z) = 1/2 + (1/2) erf(z / √2). The complementary error function is erfc(z) = 1 - erf(z) = (2 / √π) ∫_z^∞ e^(-t^2) dt.
How is the Full Width at Half Maximum (FWHM) of a Gaussian profile derived analytically?
For a centered Gaussian f(x) = (1/(σ√(2π))) exp(-x^2 / (2σ^2)), the maximum amplitude occurs at x = 0 with value f(0) = 1/(σ√(2π)). Half-maximum occurs at points x_{1/2} where f(x_{1/2}) = f(0) / 2. Setting exp(-x_{1/2}^2 / (2σ^2)) = 1/2 and taking natural logarithms yields -x_{1/2}^2 / (2σ^2) = -ln(2), which simplifies to x_{1/2} = ± σ √(2 ln 2). The total width is FWHM = x_{1/2, +} - x_{1/2, -} = 2 √(2 ln 2) σ ≈ 2.354820045 σ.
What are the inflection points of the Gaussian density curve, and what is their geometric significance?
The inflection points of f(x) occur where the second derivative f''(x) = 0. Differentiating f(x) twice with respect to x gives f''(x) = f(x) [(x - μ)^2 - σ^2] / σ^4. Setting f''(x) = 0 reveals that inflection points occur precisely at x = μ ± σ. Geometrically, the curve transitions from concave downward on (μ - σ, μ + σ) to concave upward on (-∞, μ - σ) and (μ + σ, ∞), directly linking standard deviation to curve curvature.
Why is the normal distribution the maximum entropy distribution for a fixed mean and variance?
Under the Shannon differential entropy functional H(f) = - ∫ f(x) ln f(x) dx subject to normalization ∫ f(x) dx = 1 and fixed second moment ∫ (x - μ)^2 f(x) dx = σ^2, applying the calculus of variations with Lagrange multipliers yields the Euler-Lagrange equation -ln f(x) - 1 - λ_0 - λ_1 (x - μ)^2 = 0. Solving for f(x) yields the exponential quadratic form f(x) ∝ exp(-λ_1 (x - μ)^2), proving the normal distribution maximizes informational entropy.
How does the Box-Muller transform generate exact normal variates from uniform variates?
Let U_1, U_2 ~ i.i.d. Uniform(0, 1). In Cartesian coordinates, the joint density of two independent standard normals is f(z_1, z_2) = (1/(2π)) exp(-(z_1^2 + z_2^2)/2). Transforming to polar coordinates z_1 = R cos θ, z_2 = R sin θ gives Jacobian R, yielding joint density (1/(2π)) R exp(-R^2 / 2). The angle θ is uniform on [0, 2π) (θ = 2π U_2), and R^2 follows Exponential(1/2), so R = √(-2 ln U_1). This produces Z_1 = √(-2 ln U_1) cos(2π U_2) and Z_2 = √(-2 ln U_1) sin(2π U_2).
What is the 68-95-99.7 empirical rule, and how are its percentages calculated via erf?
The empirical rule states that for a normal distribution, approximately 68.27% of observations fall within μ ± 1σ, 95.45% within μ ± 2σ, and 99.73% within μ ± 3σ. These probabilities correspond to P(|X - μ| ≤ kσ) = P(|Z| ≤ k) = Φ(k) - Φ(-k) = 2Φ(k) - 1 = erf(k / √2). Evaluating erf(1/√2) ≈ 0.682689, erf(2/√2) ≈ 0.954500, and erf(3/√2) ≈ 0.997300 confirms the exact mathematical percentages.
Experimental precautions and statistical checks
- Ensure numerical integration grids extend to at least ±5σ to capture more than 99.9999% of distribution area.
- Verify that the Box-Muller uniform variate inputs strictly exclude zero (U_1 > 0) to avoid logarithmic singularity.
- Maintain double-precision floating-point arithmetic when computing tail probabilities erf(z) for large z.
- Check that baseline peak amplitude is measured at exactly x = μ before evaluating half-maximum thresholds.
Theoretical assumptions & limitations
- Real physical measurements have bounded support, whereas the mathematical Gaussian distribution has infinite tails.
- Empirical datasets often display heavier tails or slight skewness that violate strict Gaussian assumptions.