Matthias Troyer wrote:
It is actually very important that uniform_real creates numbers in a half-open interval, and all random number libraries are actually implemented that way. To give just one example, the exponential distribution need to evaluate an expression like
std::log(1.-u)
where u is a uniform random number in the interval [0,1.). If instead u were from the interval [0,1], it could take on the value 1., leading to a zero argument to the log, and thus a problem. The half- open interval is thus the right thing to do.
Sorry, I was mistaken about this. I thought that the output of mt19937 was being fed directly into uniform_real, but what actually happens is that boost::variate_generator detects this situation and first uses boost::uniform_01 to correctly convert the distribution into a half-open one. This also explains a lot of the other things which were puzzling me. Daniel