Searcharxiv⌕ Search

arXiv subjects

Assaf Naor

Publications and source records attributed to Assaf Naor.

At least 55 records · Page 3Linked to original sources
arXiv2015

Bourgain's discretization theorem

Bourgain's discretization theorem asserts that there exists a universal constant $C\in (0,\infty)$ with the following property. Let $X,Y$ be Banach spaces with $\dim X=n$. Fix $D\in (1,\infty)$ and set $δ= e^{-n^{Cn}}$. Assume that $\mathcal N$ is a $δ$-net in the unit ball of $X$ and that $\mathcal N$ admits a bi-Lipschitz embedding into $Y$ with distortion at most $D$. Then the entire space $X$ admits a bi-Lipschitz embedding into $Y$ with distortion at most $CD$. This mostly expository article is devoted to a detailed presentation of a proof of Bourgain's theorem. We also obtain an improvement of Bourgain's theorem in the important case when $Y=L_p$ for some $p\in [1,\infty)$: in this case it suffices to take $δ= C^{-1}n^{-5/2}$ for the same conclusion to hold true. The case $p=1$ of this improved discretization result has the following consequence. For arbitrarily large $n\in \mathbb{N}$ there exists a family $\mathscr Y$ of $n$-point subsets of ${1,...,n}^2\subseteq \mathbb{R}^2$ such that if we write $|\mathscr Y|= N$ then any $L_1$ embedding of $\mathscr Y$, equipped with the Earthmover metric (a.k.a. transportation cost metric or minimumum weight matching metric) incurs distortion at least a constant multiple of $\sqrt{\log\log N}$; the previously best known lower bound for this problem was a constant multiple of $\sqrt{\log\log \log N}$.

math.FA↗
arXiv2015

Pythagorean powers of hypercubes

For $n\in \mathbb{N}$ consider the $n$-dimensional hypercube as equal to the vector space $\mathbb{F}_2^n$, where $\mathbb{F}_2$ is the field of size two. Endow $\mathbb{F}_2^n$ with the Hamming metric, i.e., with the metric induced by the $\ell_1^n$ norm when one identifies $\mathbb{F}_2^n$ with $\{0,1\}^n\subseteq \mathbb{R}^n$. Denote by $\ell_2^n(\mathbb{F}_2^n)$ the $n$-fold Pythagorean product of $\mathbb{F}_2^n$, i.e., the space of all $x=(x_1,\ldots,x_n)\in \prod_{j=1}^n \mathbb{F}_2^n$, equipped with the metric $$ \forall\, x,y\in \prod_{j=1}^n \mathbb{F}_2^n,\qquad d_{\ell_2^n(\mathbb{F}_2^n)}(x,y)= \sqrt{ \|x_1-y_1\|_1^2+\ldots+\|x_n-y_n\|_1^2}. $$ It is shown here that the bi-Lipschitz distortion of any embedding of $\ell_2^n(\mathbb{F}_2^n)$ into $L_1$ is at least a constant multiple of $\sqrt{n}$. This is achieved through the following new bi-Lipschitz invariant, which is a metric version of (a slight variant of) a linear inequality of Kwapie{ń} and Schütt (1989). Letting $\{e_{jk}\}_{j,k\in \{1,\ldots,n\}}$ denote the standard basis of the space of all $n$ by $n$ matrices $M_n(\mathbb{F}_2)$, say that a metric space $(X,d_X)$ is a KS space if there exists $C=C(X)>0$ such that for every $n\in 2\mathbb{N}$, every mapping $f:M_n(\mathbb{F}_2)\to X$ satisfies \begin{equation*}\label{eq:metric KS abstract} \frac{1}{n}\sum_{j=1}^n\mathbb{E}\left[d_X\Big(f\Big(x+\sum_{k=1}^ne_{jk}\Big),f(x)\Big)\right]\le C \mathbb{E}\left[d_X\Big(f\Big(x+\sum_{j=1}^ne_{jk_j}\Big),f(x)\Big)\right], \end{equation*} where the expectations above are with respect to $x\in M_n(\mathbb{F}_2)$ and $k=(k_1,\ldots,k_n)\in \{1,\ldots,n\}^n$ chosen uniformly at random. It is shown here that $L_1$ is a KS space (with $C= 2e^2/(e^2-1)$, which is best possible), implying the above nonembeddability statement. Links to the Ribe program are discussed, as well as related open problems.

math.FA↗
arXiv2014

Expanders with respect to Hadamard spaces and random graphs

It is shown that there exists a sequence of 3-regular graphs $\{G_n\}_{n=1}^\infty$ and a Hadamard space $X$ such that $\{G_n\}_{n=1}^\infty$ forms an expander sequence with respect to $X$, yet random regular graphs are not expanders with respect to $X$. This answers a question of \cite{NS11}. $\{G_n\}_{n=1}^\infty$ are also shown to be expanders with respect to random regular graphs, yielding a deterministic sublinear time constant factor approximation algorithm for computing the average squared distance in subsets of a random graph. The proof uses the Euclidean cone over a random graph, an auxiliary continuous geometric object that allows for the implementation of martingale methods.

math.MG↗
arXiv2014

Solution of the propeller conjecture in $\mathbb{R}^3$

It is shown that every measurable partition ${A_1,..., A_k}$ of $\mathbb{R}^3$ satisfies $$\sum_{i=1}^k||\int_{A_i} xe^{-\frac12||x||_2^2}dx||_2^2\le 9π^2.\qquad(*)$$ Let ${P_1,P_2,P_3}$ be the partition of $\mathbb{R}^2$ into $120^\circ$ sectors centered at the origin. The bound is sharp, with equality holding if $A_i=P_i\times \mathbb{R}$ for $i\in {1,2,3}$ and $A_i=\emptyset$ for $i\in \{4,...,k\}$ (up to measure zero corrections, orthogonal transformations and renumbering of the sets $\{A_1,...,A_k\}$). This settles positively the 3-dimensional Propeller Conjecture of Khot and Naor (FOCS 2008). The proof of reduces the problem to a finite set of numerical inequalities which are then verified with full rigor in a computer-assisted fashion. The main consequence (and motivation) of $(*)$ is complexity-theoretic: the Unique Games hardness threshold of the Kernel Clustering problem with $4 \times 4$ centered and spherical hypothesis matrix equals $\frac{2π}{3}$.

cs.CC↗
arXiv2013

Comparison of metric spectral gaps

Let $A=(a_{ij})\in M_n(\R)$ be an $n$ by $n$ symmetric stochastic matrix. For $p\in [1,\infty)$ and a metric space $(X,d_X)$, let $γ(A,d_X^p)$ be the infimum over those $γ\in (0,\infty]$ for which every $x_1,...,x_n\in X$ satisfy $$ \frac{1}{n^2} \sum_{i=1}^n\sum_{j=1}^n d_X(x_i,x_j)^p\le \fracγ{n}\sum_{i=1}^n\sum_{j=1}^n a_{ij} d_X(x_i,x_j)^p. $$ Thus $γ(A,d_X^p)$ measures the magnitude of the {\em nonlinear spectral gap} of the matrix $A$ with respect to the kernel $d_X^p:X\times X\to [0,\infty)$. We study pairs of metric spaces $(X,d_X)$ and $(Y,d_Y)$ for which there exists $Ψ:(0,\infty)\to (0,\infty)$ such that $γ(A,d_X^p)\le Ψ(γ(A,d_Y^p))$ for every symmetric stochastic $A\in M_n(\R)$ with $γ(A,d_Y^p)<\infty$. When $Ψ$ is linear a complete geometric characterization is obtained. Our estimates on nonlinear spectral gaps yield new embeddability results as well as new nonembeddability results. For example, it is shown that if $n\in \N$ and $p\in (2,\infty)$ then for every $f_1,...,f_n\in L_p$ there exist $x_1,...,x_n\in L_2$ such that {equation}\label{eq:p factor} \forall\, i,j\in \{1,...,n\},\quad \|x_i-x_j\|_2\lesssim p\|f_i-f_j\|_p, {equation} and $$ \sum_{i=1}^n\sum_{j=1}^n \|x_i-x_j\|_2^2=\sum_{i=1}^n\sum_{j=1}^n \|f_i-f_j\|_p^2. $$ This statement is impossible for $p\in [1,2)$, and the asymptotic dependence on $p$ in \eqref{eq:p factor} is sharp. We also obtain the best known lower bound on the $L_p$ distortion of Ramanujan graphs, improving over the work of Matoušek. Links to Bourgain--Milman--Wolfson type and a conjectural nonlinear Maurey--Pisier theorem are studied.

math.MG↗
arXiv2013

Spectral calculus and Lipschitz extension for barycentric metric spaces

The metric Markov cotype of barycentric metric spaces is computed, yielding the first class of metric spaces that are not Banach spaces for which this bi-Lipschitz invariant is understood. It is shown that this leads to new nonlinear spectral calculus inequalities, as well as a unified framework for Lipschitz extension, including new Lipschitz extension results for CAT(0) targets. An example that elucidates the relation between metric Markov cotype and Rademacher cotype is analyzed, showing that a classical Lipschitz extension theorem of Johnson, Lindenstrauss and Benyamini is asymptotically sharp.

math.MG↗
arXiv2013

Pisier's inequality revisited

Given a Banach space $X$, for $n\in \mathbb N$ and $p\in (1,\infty)$ we investigate the smallest constant $\mathfrak P\in (0,\infty)$ for which every $f_1,...,f_n:{-1,1}^n\to X$ satisfy \int_{{-1,1}^n}\Bigg|\sum_{j=1}^n \partial_jf_j(\varepsilon)\Bigg|^pdμ(\varepsilon) \leq \mathfrak{P}^p\int_{{-1,1}^n}\int_{{-1,1}^n}\Bigg\|\sum_{j=1}^n \d_jΔf_j(\varepsilon)\Bigg\|^pdμ(\varepsilon) dμ(δ), where $μ$ is the uniform probability measure on the discrete hypercube ${-1,1}^n$ and ${\partial_j}_{j=1}^n$ and $Δ=\sum_{j=1}^n\partial_j$ are the hypercube partial derivatives and the hypercube Laplacian, respectively. Denoting this constant by $\mathfrak{P}_p^n(X)$, we show that $\mathfrak{P}_p^n(X)\le \sum_{k=1}^{n}\frac{1}{k}$ for every Banach space $(X,|\cdot|)$. This extends the classical Pisier inequality, which corresponds to the special case $f_j=Δ^{-1}\partial_j f$ for some $f:{-1,1}^n\to X$. We show that $\sup_{n\in \N}\mathfrak{P}_p^n(X)<\infty$ if either the dual $X^*$ is a $\mathrm{UMD}^+$ Banach space, or for some $θ\in (0,1)$ we have $X=[H,Y]_θ$, where $H$ is a Hilbert space and $Y$ is an arbitrary Banach space. It follows that $\sup_{n\in \N}\mathfrak{P}_p^n(X)<\infty$ if $X$ is a Banach lattice of finite cotype.

math.FA↗
arXiv2013

Nonlinear spectral calculus and super-expanders

Nonlinear spectral gaps with respect to uniformly convex normed spaces are shown to satisfy a spectral calculus inequality that establishes their decay along Cesaro averages. Nonlinear spectral gaps of graphs are also shown to behave sub-multiplicatively under zigzag products. These results yield a combinatorial construction of super-expanders, i.e., a sequence of 3-regular graphs that does not admit a coarse embedding into any uniformly convex normed space.

math.MG↗
arXiv2012

Vertical versus horizontal Poincaré inequalities on the Heisenberg group

Let $\H= < a,b | a[a,b]=[a,b]a \wedge b[a,b]=[a,b]b>$ be the discrete Heisenberg group, equipped with the left-invariant word metric $d_W(\cdot,\cdot)$ associated to the generating set ${a,b,a^{-1},b^{-1}}$. Letting $B_n= {x\in \H: d_W(x,e_\H)\le n}$ denote the corresponding closed ball of radius $n\in \N$, and writing $c=[a,b]=aba^{-1}b^{-1}$, we prove that if $(X,|\cdot|_X)$ is a Banach space whose modulus of uniform convexity has power type $q\in [2,\infty)$ then there exists $K\in (0,\infty)$ such that every $f:\H\to X$ satisfies {multline*} \sum_{k=1}^{n^2}\sum_{x\in B_n}\frac{|f(xc^k)-f(x)|_X^q}{k^{1+q/2}}\le K\sum_{x\in B_{21n}} \Big(|f(xa)-f(x)|^q_X+\|f(xb)-f(x)\|^q_X\Big). {multline*} It follows that for every $n\in \N$ the bi-Lipschitz distortion of every $f:B_n\to X$ is at least a constant multiple of $(\log n)^{1/q}$, an asymptotically optimal estimate as $n\to\infty$.

math.MG↗
arXiv2012

Locally decodable codes and the failure of cotype for projective tensor products

It is shown that for every $p\in (1,\infty)$ there exists a Banach space $X$ of finite cotype such that the projective tensor product $\ell_p\tp X$ fails to have finite cotype. More generally, if $p_1,p_2,p_3\in (1,\infty)$ satisfy $\frac{1}{p_1}+\frac{1}{p_2}+\frac{1}{p_3}\le 1$ then $\ell_{p_1}\tp\ell_{p_2}\tp\ell_{p_3}$ does not have finite cotype. This is a proved via a connection to the theory of locally decodable codes.

math.FA↗
arXiv2012

Krivine schemes are optimal

It is shown that for every $k\in \N$ there exists a Borel probability measure $μ$ on $\{-1,1\}^{\R^{k}}\times \{-1,1\}^{\R^{k}}$ such that for every $m,n\in \N$ and $x_1,..., x_m,y_1,...,y_n\in S^{m+n-1}$ there exist $x_1',...,x_m',y_1',...,y_n'\in S^{m+n-1}$ such that if $G:\R^{m+n}\to \R^k$ is a random $k\times (m+n)$ matrix whose entries are i.i.d. standard Gaussian random variables then for all $(i,j)\in {1,...,m}\times {1,...,n}$ we have \E_G[\int_{{-1,1}^{\R^{k}}\times {-1,1}^{\R^{k}}}f(Gx_i')g(Gy_j')dμ(f,g)]=\frac{ }{(1+C/k)K_G}, where $K_G$ is the real Grothendieck constant and $C\in (0,\infty)$ is a universal constant. This establishes that Krivine's rounding method yields an arbitrarily good approximation of $K_G$.

math.FA↗
arXiv2012

An introduction to the Ribe program

This article accompanies the 10th Takagi Lectures, delivered by the author at RIMS, Kyoto, on May 26 2012. It contains an exposition of results, applications, and challenges of the Ribe program.

math.FA↗
arXiv2012

Ultrametric subsets with large Hausdorff dimension

It is shown that for every $\e\in (0,1)$, every compact metric space $(X,d)$ has a compact subset $S\subseteq X$ that embeds into an ultrametric space with distortion $O(1/\e)$, and $$\dim_H(S)\ge (1-\e)\dim_H(X),$$ where $\dim_H(\cdot)$ denotes Hausdorff dimension. The above $O(1/\e)$ distortion estimate is shown to be sharp via a construction based on sequences of expander graphs.

math.MG↗
arXiv2012

Discretization and affine approximation in high dimensions

Lower estimates are obtained for the macroscopic scale of affine approximability of vector-valued Lipschitz functions on finite dimensional normed spaces, completing the work of Bates, Johnson, Lindenstrass, Preiss and Schechtman. This yields a new approach to Bourgain's discretization theorem for superreflexive targets.

math.FA↗
arXiv2011

Ultrametric skeletons

We prove that for every $ε\in (0,1)$ there exists $C_ε\in (0,\infty)$ with the following property. If $(X,d)$ is a compact metric space and $μ$ is a Borel probability measure on $X$ then there exists a compact subset $S\subseteq X$ that embeds into an ultrametric space with distortion $O(1/ε)$, and a probability measure $ν$ supported on $S$ satisfying $ν(B_d(x,r))\le (μ(B_d(x,C_εr))^{1-ε}$ for all $x\in X$ and $r\in (0,\infty)$. The dependence of the distortion on $ε$ is sharp. We discuss an extension of this statement to multiple measures, as well as how it implies Talagrand's majorizing measures theorem.

math.MG↗