[R] Sobol Sequences - Convergence Issues

[Animesh Saxena]

Hi,

My question is regarding the R package "randtoolbox". I was testing it for few 
days for generating sobol sequences. As per my findings the numbers for higher 
dimensions > 30 are prone to clustering. This might be due to bad choice of 
initialization numbers or in some lingo called free direction numbers (no I am 
not talking about the seed value). 

Generally in most bookish algo's they are chosen as 2^(b-p)
where b = the number of bits
p = p'th direction number
where p runs from 1 to g (g = polynomial used in sobol sequence which is of 
degree g).

I hope the notation is clear. If this direction number had been chosen (as 
mentioned above) then sobol sequence creates a problem. Can this be modified?
I want to specify my own direction numbers. I checked the code, and found that 
it points to a fortran code file. 

Any suggestions?



-- 
Animesh Saxena


www.quantanalysis.in