Hi r00kie, If you want to calculate the error given by an MC integration, you run it several times with different seeds (and the same number of paths) and calculate the standard deviation. If it's higher than your tolerance, you increase the number of seeds. With QMC you don't have a seed. However, as Douglas01 mentioned, there randomized quasi-random numbers.Here's roughly how randomized quasi-random numbers work: in a Halton sequence you start with 1/2, then you take 1/4 and 3/4, and then you take 1/8, 5/8, 3/8, 7/8. Now, for reasons of symmetry, you can take first 3/4 and then 1/4; after that you can take 5/8 and then 1/8, or you can take 3/8 and 7/8. You can get infinitely many sequences that have the same discrepancy at all times. Chosing such a sequence amounts to choosing a random number. But now you have another problem: for this particular sequence, the first 1023 numbers are the same, no matter how you pick them. That's only a problem in very low dimension though.Quartz, I'm not sure pseudo-random numbers are easier to parallelize. The Halton sequence is extremly easy to parallelize, in any dimension.Your point about "pathological" integrands is very interesting. I didn't know about that. Best,V.