pub fn std_normal_cdf(x: f64) -> f64Expand description
Standard-normal CDF Φ(x).
Evaluates ½·erfc(-x/√2) through libm::erfc — the SunOS msun
implementation, which is accurate to within a ULP across the entire real
line. The same ½·erfc(-x/√2) identity backs the device
std_normal_cdf JIT under gpu::polya_gamma (CUDA’s erfcf/erfc is
itself derived from the msun implementation), so CPU and GPU posterior
callers see bit-identical tail masses.
statrs::distribution::Normal::cdf and statrs::function::erf::erfc
share a rational-approximation core with a ~10⁻¹¹ precision floor in
the bulk; routing the saddle-point tail-mass evaluator
exponential_tail_mass through that floor would have spoiled
Φ(η ≈ ±1) digits the GPU oracle considers correct and broken the
GPU/CPU parity gate.