let X~N(mu,sigma) then z=((x-mu)/sigma)~N(0,1) this is basic normal dist stuff now lets do it backwards ie let z~N(0,1) then X=(sigma*z + mu)~N(mu,sigma) so there ya go. with a N(0,1) generator to make it N(mu, sigma) transform it by sigma*X + mu.
you can prove this without too much trouble by writing E[X] as integral(x*p(x)) (for continuous, obviously sum for discrete) and showing E[aX+b] = aE[X] + b and then considering variance as E[X^2] - E^2[X]
also your code is wrong, it works by writing the normal distribution integral in 2d space and picking a point(v1,v2) in the unit circle which is why you reject if v1^2 + v2^2 > 1. so v1,v2 should lie between -1 and 1 which is why the code calls UnitRV() which should return a value between 0 and 1 however your UnitRV returns a value between -1 and 1 putting v1 and v2 in the range -3 to 1 which shouldnt effect your numbers (i think) but will require more numbers to be generated per each (x1,x2)
ive just looked at the maths for generating a N(mu,sigma) distribution closely instead of guessing(as i did originally), you have to do a translated elliptical substitution (x-a=q*cos t, y-b=r*sin t rather than the much simpler polar subsition x=r*cos t,y=r*sin t) into the double integral and its far too much work when you can just multiply by variance and add mean to the resulting N(0,1) numbers.
**i deleted the paragraph that was here cos it was a load of nonsense
my masters project is based on random number generation so i do know quite a bit about it and have probably assumed too much knowledge on your part, if theres anything you dont get then please ask.
Last edited by kev82; 08-10-2004 at 07:17 PM.