[brianpgordon]

I feel like this is over my head, but if the problem is that sorting a vector produces non-differentiable kinks in the output then why not just run a simple polynomial regression over it and differentiate that?