arXiv subjects

Gilles Pisier

Publications and source records attributed to Gilles Pisier.

At least 37 records · Page 2Linked to original sources

arXiv2016

Subgaussian sequences in probability and Fourier analysis

This is a review on subgaussian sequences of random variables, prepared for the Mediterranean Institute for the Mathematical Sciences (MIMS). We first describe the main examples of such sequences. Then we focus on examples coming from the harmonic analysis of Fourier series and we describe the connection of subgaussian sequences of characters on the unidimensional torus (or any compact Abelian group) with Sidon sets. We explain the main combinatorial open problem concerning such subgaussian sequences. We present the answer to the analogous question for subgaussian bounded mean oscillation (BMO) sequences on the unit circle. Lastly, we describe several very recent results that provide a generalization of the preceding ones when the trigonometric system (or its analogue on a compact Abelian group) is replaced by an arbitrary orthonormal system bounded in $L_\infty$.

math.PR↗

arXiv2016

On uniformly bounded orthonormal Sidon systems

In answer to a question raised recently by Bourgain and Lewko, we show, with their paper's terminology, that any uniformly bounded $ψ_2 (C)$-orthonormal system ($ψ_2 (C)$ is a variant of subGaussian)is 2-fold tensor Sidon. This sharpens their result that it is 5-fold tensor Sidon. The proof is somewhat reminiscent of the author's original one for (Abelian) group characters, based on ideas due to Drury and Rider. However, we use Talagrand's majorizing measure theorem in place of Fernique's metric entropy lower bound. We also show that a uniformly bounded orthonormal system is randomly Sidon iff it is 4-fold tensor Sidon, or equivalently $k$-fold tensor Sidon for some (or all) $k\ge 4$. Various generalizations are presented, including the case of random matrices, for systems analogous to the Peter-Weyl decomposition for compact non-Abelian groups. In the latter setting we also include a new proof of Rider's unpublished result that randomly Sidon sets are Sidon, which implies that the union of two Sidon sets is Sidon.

math.FA↗

arXiv2016

Strong convergence for reduced free products (Remarks on a result by Paul Skoufranis)

Using an inequality due to Ricard and Xu, we give a different proof of Paul Skoufranis's recent result showing that the strong convergence of possibly non-commutative random variables $X^{(k)}\to X$ is stable under reduced free product with a fixed non-commutative random variable $Y$. In fact we obtain a more general fact: assuming that the families $X^{(k)}=\{X_i^{(k)}\}$ and $Y^{(k)}=\{Y_j^{(k)}\}$ are $*$-free as well as their limits (in moments) $X =\{X_i \}$ and $Y =\{Y_j\}$, the strong convergences $X^{(k)}\to X$ and $Y^{(k)}\to Y$ imply that of $\{X^{(k)},Y^{(k)} \}$ to $\{X ,Y\}$. Phrased in more striking language: the reduced free product is "continuous" with respect to strong convergence. The analogue for weak convergence (i.e. convergence of all moments) is obvious. Our approach extends to the amalgamated free product, left open by Skoufranis.

math.OA↗

arXiv2016

The non-commutative Khintchine inequalities for p<1

We give a proof of the Khintchine inequalities in non-commutative $L_p$-spaces for all $0< p<1$. These new inequalities are valid for the Rademacher functions or Gaussian random variables, but also for more general sequences, e.g. for the analogues of such random variables in free probability. We also prove a factorization for operators from a Hilbert space to a non commutative $L_p$-space, which is new for $0<p<1$. We end by showing that Mazur maps are Hölder on semifinite von Neumann algebras. The main tool is a new form of Hölder inequality for non commutative Lp spaces with weights.

math.OA↗

arXiv2015

Quantum expanders and growth of group representations

Let $π$ be a finite dimensional unitary representation of a group $G$ with a generating symmetric $n$-element set $S\subset G$. Fix $\vp>0$. Assume that the spectrum of $|S|^{-1}\sum_{s\in S} π(s) \otimes \overline{π(s)}$ is included in $ [-1, 1-\vp]$ (so there is a spectral gap $\ge \vp$). Let $r'_N(π)$ be the number of distinct irreducible representations of dimension $\le N$ that appear in $π$. Then let $R_{n,\vp}'(N)=\sup r'_N(π)$ where the supremum runs over all $π$ with ${n,\vp}$ fixed. We prove that there are positive constants $δ_\vp$ and $c_\vp$ such that, for all sufficiently large integer $n$ (i.e. $n\ge n_0$ with $n_0$ depending on $\vp$) and for all $N\ge 1$, we have $\exp{δ_\vp nN^2} \le R'_{n,\vp}(N)\le \exp{c_\vp nN^2}$. The same bounds hold if, in $r'_N(π)$, we count only the number of distinct irreducible representations of dimension exactly $= N$.

math.OA↗

arXiv2014

A continuum of $\mathrm{C}^*$-norms on $\IB(H)\otimes \IB(H)$ and related tensor products

For any pair $M,N$ of von Neumann algebras such that the algebraic tensor product $M\otimes N$ admits more than one $\mathrm{C}^*$-norm, the cardinal of the set of $\mathrm{C}^*$-norms is at least $ {2^{\aleph_0}}$. Moreover there is a family with cardinality $ {2^{\aleph_0}}$ of injective tensor product functors for $\mathrm{C}^*$-algebras in Kirchberg's sense. Let $\IB=\prod_n M_{n}$. We also show that, for any non-nuclear von Neumann algebra $M\subset \IB(\ell_2)$, the set of $\mathrm{C}^*$-norms on $\IB \otimes M$ has cardinality equal to $2^{2^{\aleph_0}}$.

math.OA↗

arXiv2014

Quantum Expanders and Geometry of Operator Spaces

We show that there are well separated families of quantum expanders with asymptotically the maximal cardinality allowed by a known upper bound. This has applications to the "growth" of certain operator spaces: It implies asymptotically sharp estimates for the growth of the multiplicity of $M_N$-spaces needed to represent (up to a constant $C>1$) the $M_N$-version of the $n$-dimensional operator Hilbert space $OH_n$ as a direct sum of copies of $M_N$. We show that, when $C$ is close to 1, this multiplicity grows as $\exp{βn N^2}$ for some constant $β>0$. The main idea is to relate quantum expanders with "smooth" points on the matricial analogue of the Euclidean unit sphere. This generalizes to operator spaces a classical geometric result on $n$-dimensional Hilbert space (corresponding to N=1). In an appendix, we give a quick proof of an inequality (related to Hastings's previous work) on random unitary matrices that is crucial for this paper.

math.OA↗

arXiv2013

On the metric entropy of the Banach-Mazur compactum

We study the metric entropy of the metric space $B_n$ of all n-dimensional Banach spaces (the so-called Banach-Mazur compactum) equipped with the Banach-Mazur (multiplicative) "distance" $d$. We are interested either in estimates independent of the dimension or in asymptotic estimates when the dimension tends to $\infty$. For instance, we prove that, if $N({B_n},d, 1+\varepsilon)$ is the smallest number of "balls" of "radius" $1+\varepsilon$ that cover $ B_n$, then for any $\varepsilon>0$ we have $$0<\liminf_{n\to \infty}n^{-1} \log\log N(B_n,d,1+\varepsilon)\le \limsup_{n\to \infty}n^{-1} \log\log N(B_n,d,1+\varepsilon)<\infty.$$ We also prove an analogous result for the metric entropy of the set of n-dimensional operator spaces equipped with the distance $d_N$ naturally associated to $N\times N$-matrices with operator entries. In that case $N$ is arbitrary but our estimates are valid independently of $N$. In the Banach space case (i.e.N=1) the above upper bound is part of the folklore, and the lower bound is at least partially known (but apparently has not appeared in print). While we follow the same approach in both cases, the matricial case requires more delicate ingredients, namely estimates (from our previous work) on certain $n$-tuples of $N\times N$ unitary matrices known as "quantum expanders".

math.FA↗

arXiv2013

Martingale inequalities and Operator space structures on $L_p$

We describe a new operator space structure on $L_p$ when $p$ is an even integer and compare it with the one introduced in our previous work using complex interpolation. For the new structure, the Khintchine inequalities and Burkholder's martingale inequalities have a very natural form:\ the span of the Rademacher functions is completely isomorphic to the operator Hilbert space $OH$, and the square function of a martingale difference sequence $d_n$ is $Σ\ d_n\otimes \bar d_n$. Various inequalities from harmonic analysis are also considered in the same operator valued framework. Moreover, the new operator space structure also makes sense for non commutative $L_p$-spaces with analogous results.

math.OA↗

arXiv2013

Random Matrices and Subexponential Operator Spaces

We introduce and study a generalization of the notion of exact operator space that we call subexponential. Using Random Matrices we show that the factorization results of Grothendieck type that are known in the exact case all extend to the subexponential case, but we exhibit (a continuum of distinct) examples of non-exact subexponential operator spaces, as well as a $C^*$-algebra that is subexponential with constant 1 but not exact. We also show that $OH$, $R+C$ and $\max(\ell_2)$ (or any other maximal operator space) are not subexponential.

math.OA↗

arXiv2012

Examples of non exact 1-subexponential $C^*$-algebras

This is a complement to our previous paper on the arxiv on quantum expanders and geometry of operator spaces. We show that there is a non-exact $C^*$-algebra that is 1-subexponential, and we give several other complements to the results of that paper. Our example can be described very simply using random matrices: Let ${X_j^{(m)}\mid j=1,2,...}$ be an i.i.d. sequence of random $m\times m$-matrices distributed according to the Gaussian Unitary Ensemble (GUE). For each $j$ let $u_j(ω)$ be the block direct sum defined by $$u_j(ω)= \oplus_{m\ge 1} X_j^{(m)}(ω)\in \oplus_{m\ge 1} M_m.$$ Then for almost every $ω$ the $C^*$-algebra generated by ${u_j(ω) \mid j=1,2,...}$ is 1-subexponential but is not exact. The GUE is a matrix model for the semi-circular distribution. We can also use instead the analogous circular model.

math.OA↗

arXiv2012

Quantum Expanders and Geometry of Operator Spaces II

In this appendix to our paper with the same title posted on arxiv we give a quick proof of an inequality that can be substituted to Hastings's result, quoted as Lemma 1.9 in our previous paper. Our inequality is less sharp but also appears to apply with more general (and even matricial) coefficients. It shows that up to a universal constant all moments of the norm of a linear combination of the form $$S=\sum\nolimits_j a_j U_j \otimes \bar U_j (1-P)$$ are dominated by those of the corresponding Gaussian sum $$S'=\sum\nolimits_j a_j Y_j \otimes \bar Y'_j .$$ The advantage is that $S'$ is now simply separately a Gaussian random variable with respect to the independent Gaussian random matrices $(Y_j)$ and $(Y'_j)$, and hence is much easier to majorize. Note we plan to incorporate this appendix into our future publication.

math.OA↗

arXiv2012

Tripartite Bell inequality, random matrices and trilinear forms

In this seminar report, we present in detail the proof of a recent result due to J. Briët and T. Vidick, improving an estimate in a 2008 paper by D. Pérez-Garc\'ıa, M. Wolf, C. Palazuelos, I. Villanueva, and M. Junge, estimating the growth of the deviation in the tripartite Bell inequality. The proof requires a delicate estimate of the norms of certain trilinear (or $d$-linear) forms on Hilbert space with coefficients in the second Gaussian Wiener chaos. Let $E^n_{\vee}$ (resp. $E^n_{\min}$) denote $ \ell_1^n \otimes \ell_1^n\otimes \ell_1^n$ equipped with the injective (resp. minimal) tensor norm. Here $ \ell_1^n$ is equipped with its maximal operator space structure. The Briët-Vidick method yields that the identity map $I_n$ satisfies (for some $c>0$) $\|I_n:\ E^n_{\vee}\to E^n_{\min}\|\ge c n^{1/4} (\log n)^{-3/2}.$ Let $S^n_2$ denote the (Hilbert) space of $n\times n$-matrices equipped with the Hilbert-Schmidt norm. While a lower bound closer to $n^{1/2} $ is still open, their method produces an interesting, asymptotically almost sharp, related estimate for the map $J_n:\ S^n_2\stackrel{\vee}{\otimes} S^n_2\stackrel{\vee}{\otimes}S^n_2 \to \ell_2^{n^3} \stackrel{\vee}{\otimes} \ell_2^{n^3} $ taking $e_{i,j}\otimes e_{k,l}\otimes e_{m,n}$ to $e_{[i,k,m],[j,l,n]}$.

math.OA↗

arXiv2011

On the Dixmier problem (Seminar report after Monod-Ozawa, JFA 2010)

This seminar report contains a detailed account of the proof of the main results in Monod and Ozawa's recent JFA paper on the Dixmier unitarizability problem. The proof is exactly identical to their proof, but our more pedestrian presentation is hopefully more accessible to nonexperts. This text is not intended for publication (but it might end up as part of an updated version of our Springer Lecture Notes 1618).

math.OA↗

arXiv2011

Real interpolation between row and column spaces

We give an equivalent expression for the $K$-functional associated to the pair of operator spaces $(R,C)$ formed by the rows and columns respectively. This yields a description of the real interpolation spaces for the pair $(M_n(R), M_n(C))$ (uniformly over $n$). More generally, the same result is valid when $M_n$ (or $B(\ell_2)$) is replaced by any semi-finite von Neumann algebra. We prove a version of the non-commutative Khintchine inequalities (originally due to Lust--Piquard) that is valid for the Lorentz spaces $L_{p,q}(τ)$ associated to a non-commutative measure $τ$, simultaneously for the whole range $1\le p,q< \infty$, regardless whether $p<2 $ or $p>2$. Actually, the main novelty is the case $p=2,q\not=2$. We also prove a certain simultaneous decomposition property for the operator norm and the Hilbert-Schmidt one.

math.OA↗

arXiv2011

Real interpolation and transposition of certain function spaces

Our starting point is a lemma due to Varopoulos. We give a different proof of a generalized form this lemma, that yields an equivalent description of the $K$-functional for the interpolation couple $(X_0,X_1)$ where $X_0=L_{p_0,\infty}(μ_1; L_q(μ_2))$ and $X_1=L_{p_1,\infty}(μ_2; L_q(μ_1))$ where $0<q<p_0,p_1\le \infty$ and $(Ω_1,μ_1), (Ω_2,μ_2)$ are arbitrary measure spaces. When $q=1$, this implies that the space $(X_0,X_1)_{θ,\infty}$ ($0<θ<1$) can be identified with a certain space of operators. We also give an extension of the Varopoulos Lemma to pairs (or finite families) of conditional expectations that seems of independent interest. The present paper is motivated by non-commutative applications that we choose to publish separately.

math.FA↗

arXiv2011

Grothendieck's Theorem, past and present

Probably the most famous of Grothendieck's contributions to Banach space theory is the result that he himself described as "the fundamental theorem in the metric theory of tensor products". That is now commonly referred to as "Grothendieck's theorem" (GT in short), or sometimes as "Grothendieck's inequality". This had a major impact first in Banach space theory (roughly after 1968), then, later on, in $C^*$-algebra theory, (roughly after 1978). More recently, in this millennium, a new version of GT has been successfully developed in the framework of "operator spaces" or non-commutative Banach spaces. In addition, GT independently surfaced in several quite unrelated fields:\ in connection with Bell's inequality in quantum mechanics, in graph theory where the Grothendieck constant of a graph has been introduced and in computer science where the Grothendieck inequality is invoked to replace certain NP hard problems by others that can be treated by "semidefinite programming" and hence solved in polynomial time. In this expository paper, we present a review of all these topics, starting from the original GT. We concentrate on the more recent developments and merely outline those of the first Banach space period since detailed accounts of that are already available, for instance the author's 1986 CBMS notes.

math.FA↗

arXiv2011

Random Series of Trace Class Operators

In this lecture, we present some results on Gaussian (or Rademacher) random series of trace class operators, mainly due jointly with F. Lust-Piquard. We will emphasize the probabilistic reformulation of these results, as well as the open problems suggested by them. We start by a brief survey of what is known about the problem of characterizing a.s. convergent (Gaussian or Rademacher) series of random vectors in a Banach space. The main result presented here is that for certain pairs of Banach spaces $E,F$ that include Hilbert spaces (and type 2 spaces with the analytic UMD property), we have $$ R(E\hat\otimes F) =R(E)\hat\otimes F + E\hat\otimes R(F) $$ where $R(E)$ denotes the space of convergent Rademacher series with coefficients in $E$ and $E\hat\otimes F$ denotes the projective tensor product.

math.FA↗