@altavir hipparchus is a forked continuation of apache commons math, check it out.
I have not yet looked deeply into kmath, but I hope it has a bit less of an oo approach than Java. Sometimes it's just easier to pass in a function argument instead of an object with an interface... I'm also really missing complex numbers and the continuation of all functions to complex arguments - e. g. I need Erfi(z) =-i Erf(i z) which means the error function has to accept complex arguments. Yes, you can divide all that into the real and imag part with each having real-valued integral (FresnelC, FresnelS), but it would be so much easier to just use complex numbers. Also complex numbers in matrices should be there!
And while we're at it, why not extend to quaternions ... But I guess that is not so in demand.
Maybe I'll contribute to the kmath project some of my stuff when I have the time š