jzmaddock

@dschmitz89 : the smoothness issue in the left tail should be fixed now: values returned may be truncated to zero in order to avoid returning total garbage, but none the...

I was about to say that we're fine, but actually we're not: when you zoom in enough there are little pockets that have underflowed to zero with correct values either...

This should ultimately be fixed in https://github.com/boostorg/math/pull/1134 which will be Boost-1.86 now.

> will look into version bump and using Boost for noncentral_t.pdf in the coming weeks. Note that fixes for the non-central-t are only in Boost _develop_ at present, it'll be...

> but maybe you get at least a gut feeling looking at the formula? Not really without implementing it, but my guess is it's derived from the 1F1 equivalence and...

Question: what is the use case for these very small results? At present the Boost PDF function here is nice and smooth and then truncates to zero (this is the...

Ah... let me take that back, zooming right in close, things don't look quite so good:  We could always truncate sooner I guess...

OK, improvements now merged to Boost develop. Here's what the v=8, mu=16 case looks like now, with samples every 0.0025:  I'm pretty please with that, though I suspect we...