A new error estimate of the fast gauss transform

The fast Gauss transform of L. Greengard and J. Strain [SIAM J. Sci. Statist. Comput., 12 (1991), pp. 79--94] reduces the computational complexity of the evaluation of the sum of N Gaussians at M points in d-dimensional space from ${\cal O}(MN)$ to ${\cal O}(M+N)$ floating-point operations. In this note, we provide numerical evidence that the error estimate of Lemma 2.1 in [SIAM J. Sci. Statist. Comput., 12 (1991), pp. 79--94] is erroneous and then proceed to calculate a replacement error estimate for the fast Gauss transform, incorporating an improved upper bound for Hermite functions