Page 1 of 1

Low Discrepancy Sequences

Posted: July 26th, 2004, 12:33 pm
by hunting
Hello,I am trying to generate pseudo-random numbers for a simulation problem with high dimensionality (>100). Two questions:1. Can anyone recommend a specific type of quasi-random sequence? Many indicate Sobol is the best but I am not sure.2. Can anyone provide VBA code for implementation?Best rgds

Low Discrepancy Sequences

Posted: July 26th, 2004, 6:20 pm
by Thetis
sobol is fresh!many matlab codes for sobol on the net; i think vba is too slow

Low Discrepancy Sequences

Posted: July 27th, 2004, 3:39 am
by Ouyang

Low Discrepancy Sequences

Posted: July 27th, 2004, 8:00 am
by grabben
Ouyang, the Mersenne Twister is not a Low Discrepancy Sequence, i.e. it's not a quasi random sequence, but rather a pseudo random sequence, i.e. a "standard" random number sequence. It is however a really good sequence.Sorry for high-jacking this thread, but most people suggest that Sobol is best, however I've heard that this is based on asymtotic behaviour, and that for the Sobol sequence it takes a really long time before the asymtotic behaviour starts kicking in. I've heard that for most applications the generalized Faure sequence is preferable.Any comments on this? I'll see if I can dig up the reference.

Low Discrepancy Sequences

Posted: July 29th, 2004, 4:59 pm
by tlaro
if you compare the measure of discrepancy for various low discrepancy sequences to that of random uniforms, the lds "disperse better" than random uniforms only in low dimensions (say 2-5). I suspect that for your problem with 100 dimensions, you would get comparable "dispersion" with random uniforms.

Low Discrepancy Sequences

Posted: August 7th, 2004, 8:39 pm
by Dostoevsky
Some of the issues related to Quasi Monte Carlo ( and Sobol sequences ) have been already covered here:http://www.wilmott.com/messageview.cfm? ... readid=149 http://www.wilmott.com/messageview.cfm? ... eadid=4298