Exactly speaking You cannot convert your sqtr(n) D value to p-value without K-S distribution. And this way you cannot reject or accept a hypothesis with alpha level. As far as I remember, a framework for series summation exists in math.special so well-known series representation can be used to calculate at least asymptotic(n->inf) cdf which was obtained by Kolmogorov. I think this implementation is sufficient for many particular uses but implementation distributions for low n is also of interest. John Maddock wrote:

...

Isn't there KS-distribution in boost.math? I didn't find this but maybe I missed something.

As far as I understand there is no way to perform KS-test using boost.math framework due to absence of needed statistic.

Point taken, it's not there because no-one has asked for it yet, and because it's dammed hard to calculate ;-)

I'll add it to the TODO list.

Regards, John.

Great! I think the code should be moved to the library from tests. Not only boost developers want to test distributions :) Steven Watanabe wrote:

There's an approximation to it in libs/random/test/statistic_tests.hpp. I'm using it heavily to validate the distributions. I have no idea how accurate it is, though.

In Christ, Steven Watanabe