bug report on random number generators

Brian Gough bjg@network-theory.co.uk
Fri Feb 21 20:33:00 GMT 2003


Heiko Bauke writes:
 > I have found some bugs in the initialization routine for generators
 > gsl_rng_transputer, gsl_rng_randu, gsl_rng_borosh13 and some other
 > serious bugs.
 > 

Thanks for the bug report and patches. 

Sorry for not replying to your previous message.  I forwarded it to
the original author of the Knuth routines, Carlo Perassi, and he is
working on fixing them.  

Ideally these bugs should have been caught before they were included
by the test of computing the 10,000th value analytically but I must
have let them slip through without checking that.

I will correct the period in the comments from 2^32 to 2^30.  Thanks
for pointing that out.  

I don't use (s|1) since it produces identical sequences for even and
odd seeds.  Although this allows seeds with sub-maximal period this is
the intention -- the 'randu' type generators are provided as a way to
produce poor-quality sequences.

-- 
Brian



More information about the Gsl-discuss mailing list