# The quality of an RNG depends on the application

submited by
Style Pass
2022-06-23 17:00:03

A random number generator can be good for some purposes and not for others. This isn’t surprising given the fundamentally impossible task such generators are supposed to perform. Technically a random number generator is a pseudo random number generator because it cannot produce random numbers. But random is as random does, and for many purposes the output of a pseudorandom number generator is random for practical purposes. But this brings us back to purposes.

for n = 0, 1, 2, … That is, we start at 0 and repeatedly add β, taking the fractional part each time. This gives us a sequence of points in the unit interval. If you think of bending the interval into a circle by joining the 1 end to 0, then our sequence goes around the circle, each time moving β/360°.

If your purpose is Monte Carlo integration, then yes it is. Our sequence has low discrepancy. You can approximate the integral of a function f over [0, 1] by taking the average of f(xn) over the first N elements of the sequence. Doing Monte Carlo integration with this particular RNG amounts to quasi Monte Carlo (QMC) integration, which is often more efficient than Monte Carlo integration.