Skip to main content
10 events
when toggle format what by license comment
Aug 16, 2023 at 6:21 comment added japalmer Not true for completely Bernstein functions (operator monotone functions) in general either. Must depend as well on the relative magnitudes of the first and second derivatives or something. I need to find an integral representation that applies to the listed functions e.g., and has a kernel with MLRP.
Aug 16, 2023 at 5:37 comment added japalmer Actually it's not true in general even for Bernstein $g$. There must be some way to characterize the functions this is true for other than the property itself.
Aug 16, 2023 at 3:48 comment added japalmer I use Maple, but I can easily check derivatives and plots.
Aug 16, 2023 at 3:26 comment added japalmer I think it will be possible using the Levy-Khinchine representation of Bernstein functions. The $m$th integral of the Levy-Khinchine kernel can be calculated as well. Proving that these kernels have monotone likelihood ratio, and then that the scale mixing density also has MLRP should do it.
Aug 15, 2023 at 20:35 comment added Iosif Pinelis @japalmer : This will probably be hard to prove for a general class of functions. You may want to try different ones using my Mathematica notebook.
Aug 15, 2023 at 17:05 comment added japalmer Last resort conjecture: $g$ is a Bernstein function in the concave case, or $g$ is an $m$th integral of a Bernstein function, or a scale mixture of $\cosh(x)$ in the convex case.
Aug 15, 2023 at 16:21 comment added japalmer Actually not absolutely monotonic, as $|x|^p$ works for $p>2$. Maybe just univalent derivatives of all orders.
Aug 15, 2023 at 16:19 vote accept japalmer
Aug 15, 2023 at 16:19 comment added japalmer Damn. That function apparently with appropriate exponent seems to work for any finite number of derivative conditions. I guess it requires absolutely monotonic or Bernstein functions. I'll work on it before addressing it here again.
Aug 15, 2023 at 15:32 history answered Iosif Pinelis CC BY-SA 4.0