We study the computational behavior and the interpolation accuracy of
functions defined by a sum of Gaussians (SOG) in surface fitting. We c
ompare the interpolation accuracies of SOG and multiquadrics (MQ) in d
ata interpolation, and describe an algorithm for approximation of scat
tered data by SOG with minimal least-squares error. We also characteri
ze the computational behaviors of SOG and MQ in data interpolation as
functions of their parameters sigma and r, respectively.