Paperlane

no. 16 · Tuesday 17 November 2026

Efficient Recursive Evaluation of Gaussian, Gamma, and Inverse Gaussian Tail Probabilities

Joshua Joseph George and Hassan Safouhi

math.NA

Abstract

Accurate computation of tail probabilities motivates efficient high-order implementations of the $G_n^{(1)}$ extrapolation transformation. We derive finite recurrence relations in the transformation order for Gaussian, gamma, and inverse Gaussian tail approximations. The Gaussian and inverse Gaussian numerator and denominator sequences satisfy four-term recurrences, whereas the gamma sequences satisfy a three-term recurrence. The derivation exploits the structure of the underlying differential operator and shows that the reduced numerator and denominator obey recurrences with identical coefficients. This construction avoids the repeated evaluation of higher-order derivatives and nested sums required by the explicit formulations. As a result, generating the approximation sequence through order $n$ requires $O(n)$ arithmetic operations, compared with $O(n^3)$ for the Gaussian and gamma distributions and $O(n^4)$ for the inverse Gaussian distribution. The recurrence algorithms also require only a fixed number of previously computed quantities at each order. Numerical results for the three distributions are in excellent agreement with values reported in the literature

fields
Numerical Analysis, Statistics Theory
pdf ↗ page arXiv:2611.19782 ↗