Multiversion of the Hausdorff–Young inequality

Paata Ivanisvili, Pavlos Kalantzopoulos · 2025

Read Multiversion of the Hausdorff–Young inequality by Paata Ivanisvili, Pavlos Kalantzopoulos (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-08494.txt.

Introduction

Ivanisvili and Kalantzopoulos prove multiversions of the Hausdorff-Young inequality in a 2025 mathematics paper. Their setting uses families of jointly Gaussian random vectors that may be correlated. Transforms of Mehler or noise type act on functions of these vectors.

The main inequality compares an expectation of a combined output with the combined expectation of the inputs. This framework unifies real and complex hypercontractivity for many functions at once. It yields multi-function forms of three classic results: the sharp Hausdorff-Young inequality, the log-Sobolev inequality, and a noisy Gaussian-Jensen inequality. One use is a new noise-based description of the Brascamp-Lieb inequality. Covariance plays a central role in that result.

Catalog record

CREATORPaata Ivanisvili, Pavlos Kalantzopoulos
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE105 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-08494

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Multiversion of the Hausdorff–Young inequality Paata Ivanisvili, Pavlos Kalantzopoulos Abstract . We consider a family of jointly Gaussian random vectors ξ j ∈ ℝ k j \xi_{j}\in\mathbb{R}^{k_{j}} , each standard normal but possibly correlated, and investigate when 𝔼 ​ F ​ ( B ⁡ ( | T z 1 ​ f 1 ​ ( ξ 1 ) | , … , | T z n ​ f n ​ ( ξ n ) | ) ) ≤ F ⁡ ( 𝔼 ​ B ​ ( | f 1 ​ ( ξ 1 ) | , … , | f n ​ ( ξ n ) | ) ) \mathbb{E}\,F\!\Bigl(B\bigl(|T_{z_{1}}f_{1}(\xi_{1})|,\dots,|T_{z_{n}}f_{n}(\xi_{n})|\bigr)\Bigr)\;\;\leq\;\;F\!\Bigl(\,\mathbb{E}\,B\bigl(|f_{1}(\xi_{1})|,\dots,|f_{n}(\xi_{n})|\bigr)\Bigr) holds, where T z T_{z} is either a Mehler transform ( z ∈ ℂ z\in\mathbb{C} ) or a noise operator ( z ∈ ℝ z\in\mathbb{R} ). This framework unifies and extends real and complex hypercontractivity to multi-function settings, yielding multiversions of the sharp Hausdorff–Young inequality, the log-Sobolev inequality, and a “noisy” Gaussian–Jensen inequality. Applications include a new covariance-based characterization of the Brascamp–Lieb inequality in the presence of noise. 2010 Mathematics Subject Classification 42B20, 42B35, 47A30, 42A38 1. Introduction A classical result in harmonic analysis is the Hausdorff–Young inequality with sharp constant . It states that for functions g in L^{p}(\mathbb{R}^{n}) and exponents 1<p\leq 2 , q=\frac{p}{p-1} , one has \bigl\|\widehat{g}\bigr\|_{L^{q}(\mathbb{R}^{n})}\;\leq\;C_{p,q}\;\bigl\|g\bigr\|_{L^{p}(\mathbb{R}^{n})}, where \widehat{g}(x)=\frac{1}{(2\pi)^{\frac{n}{2}}}\,\int_{\mathbb{R}^{n}}g(y)\,e^{-\,i\,x\cdot y}\,\mathrm{d}y is the Fourier transform, and the constant C_{p,q} is best possible. This celebrated theorem was first established by Beckner [ 11 ] . Beckner’s original argument began with an inequality on two points (a discrete \{-1,1\} hypercube setting), then used an induction and a central limit theorem to pass to the Gaussian setting, a result that has come to be called complex hypercontractivity . Afterward, a suitable change of variables yielded the Hausdorff–Young inequality in its full sharp form. Over the years, several other remarkable proofs of this sharp inequality have appeared, together with stability results [ 24 ] . Janson [ 41 ] used a Brownian-motion (stochastic-calculus) approach; Lieb [ 44 ] exploited in a very illuminating way the symmetry of the Gaussian extremizers (“Gaussian kernels have only Gaussian maximizers”); and Hu [ 33 ] essentially replaced the stochastic part of Janson’s proof by writing everything in terms of heat flows. Later in [ 37 ] it was noted that the continuous heat-flow argument is, in fact, a limiting version of Beckner’s original discrete inductive scheme. This (continuous) flow is sometimes called the Beckner–Janson flow [ 37 ] , although it is perhaps less familiar to many experts. It turns out that if one tries to use standard heat-flow methods in the usual manner, the monotonic interpolation property for the Hausdorff–Young inequality fails , as shown by Bennett, Bez, and Carbery [ 4 ] . In contrast, the subtle Beckner–Janson flow does give a monotonic approach and is the key new ingredient in these proofs. Multifunction extensions. In this paper, we carry forward the Beckner–Janson flow ideas to a multifunction setting. In doing so, we revisit several classical sharp estimates in harmonic analysis (such as the aforementioned Hausdorff–Young inequality, the Gaussian hypercontractive inequalities, etc.) and obtain their precise sharp analogues for multiple functions. To get a feel for one of our main results on the sharp multiversion of the Hausdorff–Young inequality, we refer the reader to Theorem 7 , i.e., the “baby” two-function case with equal exponents; in that setting, taking f=g and \rho=0 recovers the sharp Beckner–Babenko inequality. Beyond these specific corollaries, our main technical contribution lies in an FB theorem , i.e., Theorem 13 , which provides a fairly general framework for proving sharp multifunction estimates via the Beckner–Janson flow. We show that the classical Brascamp–Lieb inequalities and our new multifunction Hausdorff–Young inequalities can all be viewed as special cases of this single abstract FB theorem. While we mostly focus here on how this theorem extends sharp known inequalities to their natural “multi” analogues, we believe it may well have broader applicability for other families of inequalities, by allowing one to choose different pairs of (F,B) . 2. Main results Notations. In this paper \xi:=(\xi_{1},\ldots,\xi_{n}) denotes a normal random vector in \mathbb{R}^{k_{1}+\ldots+k_{n}} , where each \xi_{j} is a standard normal Gaussian vector in \mathbb{R}^{k_{j}} for j=1,\ldots,n . We remark that \xi_{1},\ldots,\xi_{n}, are not necessarily independent of each other. Throughout the paper, unless stated otherwise, the exponents p_{1},\ldots,p_{n} are assumed to be positive, the matrix \mathrm{cov}(\xi) has full rank, \lambda_{\min} and \lambda_{\max} will denote the smallest and the largest eigenvalues of the covariance matrix \mathrm{cov}(\xi) . The symbol \mathscr{P}(\mathbb{R}^{n}) denotes the space of all polynomials h:\mathbb{R}^{n}\mapsto\mathbb{C} , and for any function f we let \mathscr{P}(\mathbb{R}^{n})f:=\{hf,\;h\in\mathscr{P}(\mathbb{R}^{n})\} . Our first main result is the extension of complex hypercontractivity to several functions. Theorem 1 ( n -function complex hypercontractivity) . Let \alpha\geq 1 and (z_{1},\ldots,z_{n})\in\mathbb{C}^{n} . The inequality (2.1) \displaystyle\|\prod_{j=1}^{n}|T_{z_{j}}f_{j}(\xi_{j})|^{p_{j}}\|_{\alpha}\leq\|\prod_{j=1}^{n}|f_{j}(\xi_{j})|^{p_{j}}\|_{1} holds for all polynomials f_{j}:\mathbb{R}^{k_{j}}\mapsto\mathbb{C} if and only if (2.2) \displaystyle\sum_{i,j}(\Re w_{i})\mathrm{cov}(\xi_{i},\xi_{j})(\Re w_{j})^{T}-\alpha(\Re z_{i}w_{i})\mathrm{cov}(\xi_{i},\xi_{j})(\Re z_{j}w_{j})^{T}\geq\Re\sum_{j}\frac{(1-z_{j}^{2})w_{j}w_{j}^{T}}{p_{j}} holds for all w_{j}\in\mathbb{C}^{k_{j}} where j=1,\ldots,n . In the theorem ||X||_{p}=(\mathbb{E}|X|^{p})^{1/p} denotes the p ’th moment of a random variable X . The Mahler transform T_{z_{j}} acts on a polynomial f_{j} written in Fourier-Hermite basis f_{j}=\sum_{|\beta|\leq\mathrm{deg}(f_{j})}c_{\beta}H_{\beta} as follows T_{z_{j}}f_{j}=\sum_{|\beta|\leq\mathrm{deg}(f_{j})}z_{j}^{|\beta|}c_{\beta}H_{\beta}, where \beta=(\beta_{1},\ldots,\beta_{k_{j}})\in(\mathbb{Z}_{+})^{k_{j}} is the multiindex, |\beta|=\beta_{1}+\ldots+\beta_{k_{j}} , c_{\beta}\in\mathbb{C} and H_{\beta}(x)=\mathbb{E}(x+i\xi_{j})^{\beta} denotes probabilists’ Hermite polynomial, where for a complex vector w_{j}=(w_{j1},\ldots,w_{jk_{j}})\in\mathbb{C}^{k_{j}} we set w_{j}^{\beta}=w_{j1}^{\beta_{1}}\cdots w_{jk_{j}}^{\beta_{j}} , z_{j}w_{j}=(z_{j}w_{j1},\ldots,z_{j}w_{jk_{j}}) , and \Re w_{j}=(\Re w_{j1},\ldots,\Re w_{jk_{j}}) for all j=1,\ldots,n . 2.1. Case of imaginary z_{j} ’s If n=1 then the condition ( 2.2 ) simplifies to |\Re w_{1}|^{2}-\alpha|\Re z_{1}w_{1}|^{2}\geq\Re\frac{(1-z_{1}^{2})w_{1}^{2}}{p_{1}} which after setting p_{1}=p and choosing \alpha=q/p for any q\geq p becomes the same condition as in the classical complex hypercontractivity in Gauss space due to Beckner [ 11 ] for z=i\sqrt{p-1} , and q=p/(p-1) ; Weissler [ 51 ] (for all exponents p,q , 1<p\leq q<\infty except 2<p\leq q\leq 3 and its dual), Janson [ 41 ] , Lieb [ 44 ] , and Epperson [ 26 ] . The complex hypercontractivity has found numerous applications in heat-smoothing conjecture [ 28 , 45 ] , Markov–Bernstein inequalities and its reverse forms [ 29 ] , moment comparison inequalities [ 36 , 29 ] , and it provides L_{p} bounds in a systematic way for Fourier multipliers acting on functions having prescribed frequencies [ 35 , 29 ] . Perhaps the first application of complex hypercontractivity goes back to Beckner. Theorem A (Beckner [ 11 ] ) . Let X\sim N(0,\mathrm{Id_{n}}) . The inequality (2.3) \displaystyle\|T_{i\sqrt{p-1}}f(X)\|_{q}\leq\|f(X)\|_{p}, holds for all polynomials f\in\mathscr{P}(\mathbb{R}^{n}) if \displaystyle\frac{1}{p}+\frac{1}{q}=1,\quad 1\leq p\leq 2. The inequality ( 2.3 ) is equivalent to Hausdorff–Young inequality with a sharp constant, namely (2.4) \displaystyle\left[\int_{\mathbb{R}^{n}}\left|\widehat{g}\left(x\right)\right|^{q}\mathrm{d}x\right]^{1/q}\leq C\left[\int_{\mathbb{R}^{n}}\left|g\left(x\right)\right|^{p}\mathrm{d}x\right]^{1/p} holds for all g=\mathscr{P}(\mathbb{R}^{n})e^{-|x|^{2}/2p} with the best possible constant \displaystyle C=\frac{p^{n/2p}}{q^{n/2q}}(2\pi)^{\frac{n}{2q}-\frac{n}{2p}},\quad\text{and}\quad\widehat{g}(x):=\frac{1}{(2\pi)^{n/2}}\int_{\mathbb{R}^{n}}g(y)e^{-ix\cdot y}dy. The connection of T_{z}f for imaginary z and Fourier transform \widehat{f} can be seen through Mehler’s formula \displaystyle T_{z}f(x)=\int_{\mathbb{R}^{n}}f(y)\frac{1}{(\sqrt{1-z^{2}})^{n}}\exp\left(-\frac{|x|^{2}+|y|^{2}}{2}\left(\frac{z^{2}}{1-z^{2}}\right)+x\cdot y\frac{z}{1-z^{2}}\right)\mathrm{d}\gamma(y), where d\gamma(y)=\frac{e^{-|y|^{2}/2}}{(2\pi)^{n/2}}dy is the standard Gaussian measure on \mathbb{R}^{n} . Thus, for t\geq 1 (2.5) \displaystyle T_{i\sqrt{t-1}}f(x) \displaystyle=\frac{e^{\frac{|x|^{2}(t-1)}{2t}}}{t^{n/2}}\widehat{fe^{-\frac{|y|^{2}}{2t}}}\left(-\frac{x\sqrt{t-1}}{t}\right). Using the well-known scaling property of the Fourier transform, one can verify that the function e_{t}(x):=e^{-|x|^{2}/2t} has Fourier transform \widehat{e_{t}}(\omega)=t^{n/2}e^{-|\omega|^{2}t/2} . Thus formula ( 2.5 ) rewrites (2.6) \displaystyle T_{i\sqrt{t-1}}f(x)=\frac{\widehat{fe_{t}}}{\widehat{e_{t}}}\left(-\frac{x\sqrt{t-1}}{t}\right). Therefore, Beckner’s inequality ( 2.3 ) can be seen as another way of rewriting Hausdorff–Young inequality ( 2.4 ) in a form where the sharp constant

Read the complete text