[Rd] normal random number generator in R

Tibshirani tibshirani at gmail.com
Tue May 2 18:44:25 CEST 2006


I am trying to figure out how normal random number generator works in R. As I 
look at .../src/nmath/snorm.c file, I find the default algorithm is inverse 
CDF method. In more detail, instead of directly using uniform value by 
unif_rand(), snorm function will first get a sum by adding unif_rand() and 
2^27*unif_rand(), then divide it by 2^27 and transfer it to qnorm5() function 
for inverting. Only a short comment for this operation is available in the 
/* unif_rand() alone is not of high enough precision */

Just curious why this operation is needed? Is it a general algorithm for 
inverse CDF method, or simply unif_rand() in R returns float precision?


More information about the R-devel mailing list