Computer Generation of Normal Random Variables
A. J. Kinderman, John G. Ramage · Journal of the American Statistical Association · 1976
This paper suggests that user accessibility as well as speed and accuracy should be used as a criterion to evaluate algorithms for generating random variables. FORTRAN implementations of several of the best current normal algorithms are studied. Two of the algorithms were also implemented in assembler for making comparisons with other studies and for assessing the cost of accessibility. A new mixing algorithm which combines features of the convenient and fast methods of Marsaglia et al. is introduced.