Quasi random numbers in high dimension

I am trying to generate quasi random numbers with the Halton sequence. However, the dimensionality appears to give problems. I want to simulate 13 variables with 50 timesteps. This means I have to generate 13*50 650 quasi-random numbers per simulation.

I found this explanation on values for 'Leap'. However, using the 651'th prime number does not give the results I need. See the following example:

   p = haltonset(13*50,'Skip',1e7,'Leap',4861)
   p = scramble(p,'RR2')
   X = net(p,1000);
   boxplot(X(:,1:10))

Is it possible to generate quasi-random numbers with the Halton sequence and such a high dimensionality? Or are there alternatives available that do work?

Answer 1

[Tom Lane]

The link gives ambiguous advice. The text suggests using P-1 even though the example uses P. If I try 4860, things look okay.