My question is a follow-up to Abdelmalek Abdesselam's recent post
What makes Gaussian distributions special? Local field version?
asking about various characterizations of (real-valued) Gaussian distributions which remain valid for other analogues of Gaussian distributions/functions (e.g. in the p-adic context). One interesting characterization arises with Babenko-Beckner's refinement of the Hausdorff inequality (see https://en.wikipedia.org/wiki/Babenko–Beckner_inequality). For real numbers $s, t$ with ${1 \over s} + {1 \over t} = 1$ and $1 < s \leq 2$ it is known that the Fourier transform $f \mapsto \hat{f}$ maps $L^s(\Bbb{R}^n)$ to $L^t(\Bbb{R}^n)$ and satisfies the inequality
\begin{equation} \displaystyle \| \hat{f} \, \|_t \ \leq \ s^{n \over {2s}} \, t^{-{n \over {2t}}} \, \| f \|_s \quad \left( {\scriptstyle \begin{array}{l} \text{Babenko} \\ \text{Beckner} \\ \text{inequality} \end{array}} \right) \end{equation}
When $s = t = 2$ this inequality becomes an equality which is valid for all $f \in L^2(\Bbb{R}^n)$. For $s < 2$ equality is achieved if and only if $f$ is a Gaussian function. If the factor $s^{n \over {2s}} \, t^{-{n \over {2t}}} $ is omitted the inequality is still valid although it ceases to be sharp; this formulation is known as the Hausdorff-Young inequality.
My question concerns, initially, an analogue of the Hausdorff-Young inequality for (quadratic extensions of) finite fields: Let $q$ be a power of a prime $p$ and let $\Bbb{F}_q$ be the finite field with $q$ elements. Choose a non-square $\delta \in \Bbb{F}_q$ and form the quadratic extension $\Bbb{F}_q\big( \sqrt{\delta} \big)$. We view elements of $\Bbb{F}_q\big( \sqrt{\delta} \big)$ as linear combinations of the form $z = x + \sqrt{\delta} y$ with $x, y \in \Bbb{F}_q$ subject to the usual rules of addition and multiplication. Conjugation and norm are expressed, respectively, as $\bar{z} := z^q = x - \sqrt{\delta} y$ and $\mathrm{N}(z) := z\bar{z} = x^2 - \delta y^2$. Furthermore define $\mathrm{Tr}(z) := z + \bar{z}$. Choose any non-trivial additive character $\psi: \Bbb{F}_q \longrightarrow \Bbb{C}^*$ and define the $\Bbb{F}_q\big( \sqrt{\delta} \big)$-Fourier transform $\widehat{f}$ of a complex-valued function $f: \Bbb{F}_q\big( \sqrt{\delta} \big) \longrightarrow \Bbb{C}$ by
\begin{equation} \widehat{f}(z) \ := \ {1 \over q} \, \sum_{w \in \Bbb{F}_q\big( \sqrt{\delta} \big)} \, f(w) \, \psi \Big(-\mathrm{Tr}(zw) \Big) \end{equation}
If we endow the function space $\Bbb{C}\big[ \Bbb{F}_q\big( \sqrt{\delta} \big) \big]$ with the hermitian inner product
\begin{equation} \langle f , g \rangle \ := \ \sum_{w \in \Bbb{F}_q\big( \sqrt{\delta} \big)} \, f(w) \, \overline{g(w)} \end{equation}
then Plancherel holds, i.e. $\| \widehat{f} \, \|_2 = \| f \|_2$ and the Hausdorff-Young inequality takes the form
\begin{equation} (\dagger) \quad \| \widehat{f} \, \|_t \ \leq \ \|f \, \|_s \end{equation}
for any pair of real numbers $s,t$ with ${1 \over s} + {1 \over t} = 1$ and $1 < s \leq 2$. This is a finite field adaptation of a more general version of the Hausdorff-Young inequality that holds for finite abelian groups (see for example https://www.e-periodica.ch/cntmng?pid=ens-001:2000:46::190).
Question 1: Is inequality ($\dagger$) sharp for values $1<s<2$? If it is, which functions saturate the inequality?
I'm inclined to the view that $(\dagger)$ is not sharp. Consider functions of the form
\begin{equation} \begin{array}{ll} \displaystyle \mathrm{G}_x(z) &\displaystyle := \ \psi \big( x\, \mathrm{N}(z) \big) \\ \displaystyle \mathrm{H}_v(z) &\displaystyle := \ \psi \big( \mathrm{Tr} \, (vz)^2 \big) \end{array} \end{equation}
where $x \in \Bbb{F}_q$ and $v \in \Bbb{F}_q\big( \sqrt{\delta}\big)$ are parameters. I would argue that $\mathrm{G}_x$ should be regarded as $\Bbb{F}_q\big( \sqrt{\delta} \big)$ analogue of the real-valued Gaussian $z \mapsto \exp ( -x |z|^2)$ on the complex plane for reasons connected to the cuspidal representation theory of $\mathrm{SL}_2\big( \Bbb{F}_q \big)$ which are briefly outlined in my response to Abdesselam's post. Both $\mathrm{G}_x$ and $\mathrm{H}_v$ are near eigenfunctions of the $\Bbb{F}_q\big( \sqrt{\delta} \big)$-Fourier transform. Let's check this to be sure:
\begin{equation} \begin{array}{ll} \displaystyle \widehat{\mathrm{G}}_x(z) &\displaystyle = \ {1 \over q} \sum_{w \in \Bbb{F}_q\big( \sqrt{\delta} \big)} \psi \big(x \mathrm{N}(w) \big) \, \psi \big(-\mathrm{Tr}(zw) \big) \\ &\displaystyle = \ {1 \over q} \sum_{w \in \Bbb{F}_q\big( \sqrt{\delta} \big)} \psi \big(x w \bar{w} - zw - \bar{z} \bar{w} \big) \\ &\displaystyle = \ {1 \over q} \sum_{w \in \Bbb{F}_q\big( \sqrt{\delta} \big)} \psi \Big( x^{-1} \mathrm{N}\big( xw - \bar{z} \big) - x^{-1} \mathrm{N}(z) \Big) \\ &\displaystyle = \ {1 \over q} \ \psi \big( -x^{-1} \mathrm{N}(z) \big) \sum_{w \in \Bbb{F}_q\big( \sqrt{\delta} \big)} \psi \Big( x^{-1} \mathrm{N}\big( xw - \bar{z} \big) \Big) \\ &\displaystyle = \ {1 \over q} \ \mathrm{G}_{-x^{-1}}(z) \sum_{u \in \Bbb{F}_q\big( \sqrt{\delta} \big)} \psi \big( x^{-1} \mathrm{N}(u) \big) \\ &\displaystyle = \ {1 \over q} \ \mathrm{G}_{-x^{-1}}(z)\, \Bigg( 1 \, + \, (q+1) \, \sum_{y \in \Bbb{F}_q^*} \psi \big( x^{-1} y \big) \Bigg) \\ &\displaystyle = \ {1 \over q} \ \mathrm{G}_{-x^{-1}}(z) \, \big( -q \big) \quad \text{($\psi$ is non-trivial!)} \\ &\displaystyle = \ -\mathrm{G}_{-x^{-1}}(z) \end{array} \end{equation}
Indeed, $\mathrm{G}_x$ will be an eigenfunction of the $\Bbb{F}_q\big( \sqrt{\delta} \big)$-Fourier transform if and only if $x^2 = -1$ in $\Bbb{F}_q$. A similar computation confirms that
\begin{equation} \widehat{\mathrm{H}}_v \ = \ \widehat{\mathrm{H}}_v(0) \, \overline{\mathrm{H}}_{(2v)^{-1}} \end{equation}
where $\widehat{\mathrm{H}}_v(0)$ is a (normalized) Gauss sum with unit complex modulus. Clearly $\mathrm{H}_v$ will be a eigenfunction if and only if $v^4 = -{1 \over 4}$ in $\Bbb{F}_q \big( \sqrt{\delta} \big)$.
Note that $\| \mathrm{G}_x \|_s = \| \mathrm{H}_v \|_s = q^{2 \over s}$.
Question 2: Is the inequality
\begin{equation} (\dagger \dagger) \quad \displaystyle \| \widehat{f} \|_t \ \leq \ q^{{2 \over t} - {2 \over s}} \, \| f \|_s \end{equation}
valid within the range $1 < s <2$? Is it sharp and which functions saturate the inequality if it is? Clearly these would include functions of the form $f(z) = c \, \mathrm{G}_x(z-w)$ or $f(z) = c \, \mathrm{H}_v(z-w)$ for some choice of parameters $x \in \Bbb{F}_q$ or $v \in \Bbb{F}_q\big( \sqrt{\delta}\big)$, some shift $w \in \Bbb{F}_q\big( \sqrt{\delta} \big)$, and some overall scaling factor $c \in \Bbb{C}$. But are there others?
thanks, ines.
