###### tags: `one-offs` `convexity`
# Log-Convexity of the $C^\alpha$ Semi-Norm
**Overview**: In this note, I give a quick proof of a nice convexity result.
## From Moment-Generating Functions to Uniform Continuity Constants
A well-known fact in the context of exponential family theory (and / or statistical mechanics) is that for suitable functions $F$, the cumulant-generating function
\begin{align}
A:\theta\mapsto\log\left(\int\mu\left(\mathrm{d}x\right)\cdot\exp\left(\left\langle \theta,F\left(x\right)\right\rangle \right)\right)
\end{align}
is convex, modulo a few standard conditions. Less well-known (but perhaps indirectly related?) is that the logarithm of the $C^{\alpha}$ seminorm of a function is also convex! That is, fix a function $F$ and define
\begin{align}
H:\alpha\mapsto\sup\left\{ \frac{\mathsf{d}_{\mathcal{Y}}\left(F\left(x\right),F\left(x^{'}\right)\right)}{\mathsf{d}_{\mathcal{X}}\left(x,x^{'}\right)^{\alpha}}:x,x^{'}\in\mathcal{X},x\neq x^{'}\right\} .
\end{align}
It then holds that
\begin{align}
\log H\left(\alpha\right)=\sup\left\{ \log\mathsf{d}_{\mathcal{Y}}\left(F\left(x\right),F\left(x^{'}\right)\right)-\alpha\cdot\log\mathsf{d}_{\mathcal{X}}\left(x,x^{'}\right):x,x^{'}\in\mathcal{X},x\neq x^{'}\right\} ,
\end{align}
which is a supremum of functions which are affine in $\alpha$, and is hence convex in $\alpha$. I'm not sure about whether this has any serious applications, but it seems neat! It would also be rewarding to see whether there is a genuine connection to the result about cumulant-generating functions.

or

By clicking below, you agree to our terms of service.

Sign in via Facebook
Sign in via Twitter
Sign in via GitHub
Sign in via Dropbox
Sign in with Wallet

Wallet
(
)

Connect another wallet
New to HackMD? Sign up