Sheppard's crossing bound #
A threshold separates two unit Gaussian forms with correlation ρ with probability at
most arccos ρ / π. At threshold 0 this is Sheppard's formula for the probability that
two correlated Gaussians differ in sign; this file proves the inequality at every
threshold. It sharpens gaussPi_between_le, replacing the ratio
y = ‖β - α‖ / ‖α + β‖ in its bound (2/π) y with arctan y ≤ y.
gaussPi_abs_le_mul_abs: for independent standard GaussiansZ₁andZ₂, the event|Z₁| ≤ k |Z₂|has probability(2/π) arctan k.gaussPi_between_le_arctan: for coefficient vectors of equal norm with nonzero sum, a threshold separates the two forms with probability at most(2/π) arctan (‖β - α‖ / ‖α + β‖). The formsform (α + β)andform (β - α)are independent, and a crossing forces the first to lie within the absolute value of the second around twice the threshold. A centered Gaussian gives a symmetric interval at least the mass of any translate, so the threshold may be taken to be zero, leaving the planar angle law.two_mul_arctan_tanHalf:2 arctan (tanHalf x) = arccos x.gaussPi_between_le_arccos: the boundarccos ρ / πfor unit forms.
Sheppard's bound at every threshold. For coefficient vectors of equal norm with
nonzero sum, a threshold separates the two Gaussian forms with probability at most
(2/π) arctan (‖β - α‖ / ‖α + β‖).
2 arctan (tan (θ/2)) = θ in the form 2 arctan (tanHalf x) = arccos x.
Sheppard's bound for unit forms. A threshold separates two unit Gaussian forms
with probability at most arccos ρ / π, where ρ is their correlation.