[Quantlib-dev] sobol issues

[Mark j. <mar...@gm...>]

Dear All,

I've been benchmarking the quantlib implementation against Frances Kuo's one.
This is downloadable from

http://web.maths.unsw.edu.au/~fkuo/sobol/

A number of points arise
1) I forgot to add the zeros at the end if the init lists for the new
numbers , this means a crash.
Obviously this is easy to fix. This is probably wrong in both the
release candidate and the current development version. Maybe for the
release, just revert back to pre my improvements...

2) Even with my recent fix to the polynomial ordering, Frances's
numbers disagree with the SobolLevitan and SobolLevitanLemieux despite
the fact that all 3 ought to be the same up to dimension 40. They seem
ok in dims 1 2 3 but dim 4 disagrees.

I'm still tracking down the cause of the disagreement.

Frances is a leading expert on low discrepancy numbers so we can
assume she is correct.

Mark


-- 
LIbor market model training courses in London and Sydney www.markjoshi.com

Assoc Prof Mark Joshi
Centre for Actuarial Studies
University of Melbourne
My website is www.markjoshi.com