SearcharxivSearch

arXiv subjects

Joscha Prochno

Publications and source records attributed to Joscha Prochno.

At least 19 recordsLinked to original sources

Intrinsic volumes of $\ell_p$-balls and a continuum of Maxwell--Poincar\'e--Borel laws for their curvature measures

For $p>1$, we derive explicit formulas for the intrinsic volumes $V_0(\mathbb B_p^n),\dots,V_{n-1}(\mathbb B_p^n)$ of the $n$-dimensional $\ell_p$-balls $$ \mathbb B_p^n = \{x\in\mathbb R^n:\ |x_1|^p+\ldots+|x_n|^p\le 1\} $$ and, more generally, of their coordinate-weighted analogues. The formula is given in terms of a one-dimensional integral involving the special function $$ \mathcal F_p(t;\nu) = \int_{\mathbb R}|u|^\nu e^{-|u|^p-t|u|^{2p-2}}\,du. $$ Previously known formulas for the intrinsic volumes of ellipsoids, weighted crosspolytopes, and rectangular boxes arise as special or limiting cases. We also obtain asymptotic formulas for $V_{j(n)}(\mathbb B_p^n)$ in the high-dimensional regime $n\to\infty$, where the index $j(n)$ is allowed to depend on $n$. We further investigate the curvature measures of $\mathbb B_p^n$. These are finite measures $$ \Phi_0(\mathbb B_p^n,\cdot),\dots,\Phi_{n-1}(\mathbb B_p^n,\cdot) $$ on $\partial\mathbb B_p^n$ that localize the intrinsic volumes. We prove a Maxwell--Poincar\'{e}--Borel type limit theorem: if $X_n$ is a random boundary point of $\mathbb B_p^n$ distributed according to the normalized curvature measure $\Phi_{j(n)}(\mathbb B_p^n,\cdot)/V_{j(n)}(\mathbb B_p^n)$, where $j(n)/n\to\alpha\in[0,1]$ as $n\to\infty$, then for every fixed $r\in\mathbb N$, the joint distribution of the first $r$ coordinates of $n^{1/p}X_n$ converges weakly to the product measure $\nu_{p,\alpha}^{\otimes r}$. Here $\nu_{p,\alpha}$ is an explicit probability measure on $\mathbb R$ depending on $p>1$ and $\alpha\in[0,1]$. The main tool underlying these results is an explicit characterization of the curvature measures of coordinate-weighted $\ell_p$-balls, and in particular an explicit formula for their mixed moments.

math.PR

Large deviations for sums of multivariate stretched-exponential random variables: the few-big-jumps principle

Large deviations for sums of i.i.d.\ random variables with stretched-exponential tails (also called Weibull or semi-exponential tails) have been well understood since the 60's, going back to Nagaev's seminal work. Many extensions in the $1$-dimensional setting have been developed since then, showing that such deviations are typically governed by a single big jump. In higher dimensions, a corresponding theory has remained largely undeveloped. This work provides such a multivariate extension and establishes large deviation results for sums of i.i.d.\ random vectors in $\mathbb{R}^k$ under fairly general assumptions. Roughly speaking, for some $\alpha\in(0,1)$, the log-probability of one random vector divided by $x$ exceeding a threshold $t$ in all components behaves asymptotically, for large $x$, as $x^\alpha$ times a negative infimum of a function $\mathcal{J}$. We prove large deviation results for sums of i.i.d.\ copies, where the rate function is given by a minimization of at most $k$ summands of $\mathcal{J}$. This establishes a few-big-jumps principle that generalizes the classical $1$-dimensional phenomenon: the deviation is typically realized by \emph{at most} $k$ independent vectors. The results are applied to absolute powers of multivariate Gaussian vectors as well as to various other examples. They also allow us to study random projections of high-dimensional $\ell_p^N$-balls, revealing interesting insights about the appearance of light- and heavy-tailed distributions in high-dimensional geometry.

math.PR

Limit theorems for the distance of random points in $l_p^n$-balls

In this paper, we prove that the Euclidean distance between two independent random vectors uniformly distributed on $l_p^n$-balls $(1 \leq p \leq \infty)$ or on its boundary satisfies a central limit theorem as $n$ tends to $\infty$. Also, we give a compact proof of the case of the sphere, which was proved by Hammersley. Furthermore, we complement our central limit theorem by providing large deviation principles for the cases $p \geq 2$.

math.PR

Arithmetic sensitivity of cumulant growth in lacunary sums: transcendental versus algebraic ratio limits

We study the asymptotic behavior of cumulants of lacunary trigonometric sums $S_n(\omega) := \sum_{k=1}^n \cos (2 \pi a_k \omega)$, $\omega\in[0,1]$, and show that cumulant growth is highly sensitive to the arithmetic structure of the sequence $(a_k)_{k \geq 1}$ of positive integers. In particular, if $\lim_{k \to \infty} a_{k+1}/a_k = \eta > 1$ for some transcendental number $\eta$, we prove that for every $m\in \mathbb N$ the $m$-th cumulant of $S_n$ is asymptotically equivalent to the $m$-th cumulant of the ``independent model'' $\widetilde{S}_n := \sum_{k=1}^n \cos (2 \pi a_k U_k)$, where $U_1, U_2, \dots$ are independent random variables having uniform distribution on $[0,1]$. In particular, the order of growth of the cumulants as $n \to \infty$ is linear in this case. We also show that the transcendence condition for $\lim_{k \to \infty} a_{k+1}/a_k$ is in general necessary: when the ratio limit $\eta$ is algebraic, the cumulants of $S_n$ may have a different asymptotic order from those of $\widetilde{S}_n$. For instance, for $a_k = 2^k+1$ (with $\eta = 2$), the sixth cumulant of $S_n$ grows quadratically in $n$. In contrast, for $a_k = 2^k$ (again $\eta = 2$) or when $(a_k)_{k \geq 1}$ is the Fibonacci sequence (with $\eta = (1+\sqrt 5)/2$), the $m$-th cumulant of $S_n$ grows linearly as $n\to\infty$, but with a growth rate that differs from the one of the independent model $\widetilde{S}_n$. Overall, our results show that the asymptotic behavior of the cumulants of lacunary trigonometric sums depends on arithmetic effects in a very delicate way. This is particularly remarkable since many other probabilistic limit theorems, such as the Central Limit Theorem, hold for lacunary trigonometric sums in a universal way without any such sensitivity towards arithmetic effects.

math.NT

A sharp threshold for arithmetic effects on the tail probabilities of lacunary sums

A classical observation in analysis asserts that lacunary systems of dilated functions show many properties which are also typical for systems of independent random variables. For example, if $(n_k)_{k \ge 1}$ is a sequence of integers satisfying the Hadamard gap condition $n_{k+1}/n_k\ge q > 1,~k \ge 1$, then the normalized sums $\sum_{n=1}^N \cos(2\pi n_k x)$, considered on the probability space $[0,1]$ with Borel $\sigma$-field and Lebesgue measure, satisfy the central limit theorem (CLT) and the law of the iterated logarithm (LIL). Remarkably, the situation becomes much more deliacate when the trigonometric function $\cos(2 \pi x)$ is replaced by a more general 1-periodic function $f$, and fine arithmetic properties of the sequence $(n_k)_{k \ge 1}$ come into play. The most relevant arithmetic property can be phrased in terms of the number of solutions of certain 2-variable Diophantine equations. Recently, the authors proved that the validity of the LIL requires a strictly stronger Diophantine criterion than the CLT. In the present paper we show that this is only a special case of a wide-ranging general principle: there is a sharp cutoff, which can be expressed in form of a Diophantine criterion on the sequence $(n_k)_{k \ge 1}$, at which the tail probabilities of $\sum_{k=1}^N f(n_k x)$ change from Gaussian to potentially erratic behavior. More precisely, let $L(N,a,b,c)$ be the number of solutions $(k,\ell)$ of the equation $a n_k - b n_\ell= c$, where $1\leq k,\ell \leq N$. Roughly speaking, we prove: if $L(N,a,b,c) \le N / g_N$ for some $g_N$, then $\mathbb{P} \left[\sum_{k=1}^N f(n_k x) > t \|f\|_2 \sqrt{N} \right]$ is asymptotically is accordance with standard normal behavior for all $t$ up to $\sqrt{2 \log g_N}$. We also show that this criterion is optimal in the sense that under the same premises, the conclusion can fail to be true for values of $t$ beyond this threshold.

math.NT

Limit Theorems for the Volume of Random Projections and Sections of $\ell_p^N$-balls

Let $\mathbb{B}_p^N$ be the $N$-dimensional unit ball corresponding to the $\ell_p$-norm. For each $N\in\mathbb N$ we sample a uniform random subspace $E_N$ of fixed dimension $m\in\mathbb{N}$ and consider the volume of $\mathbb{B}_p^N$ projected onto $E_N$ or intersected with $E_N$. We also consider geometric quantities other than the volume such as the intrinsic volumes or the dual volumes. In this setting we prove central limit theorems, moderate deviation principles, and large deviation principles as $N\to\infty$. Our results provide a complete asymptotic picture. In particular, they generalize and complement a result of Paouris, Pivovarov, and Zinn [A central limit theorem for projections of the cube, Probab. Theory Related Fields. 159 (2014), 701-719] and another result of Adamczak, Pivovarov, and Simanjuntak [Limit theorems for the volumes of small codimensional random sections of $\ell_p^n$-balls, Ann. Probab. 52 (2024), 93-126].

math.PR

The macroscopic shape of Gelfand-Tsetlin patterns and free probability

A compression is a function $F:\mathbb{R}\times[0,1]\to[0,1]$ such that each $F(\cdot,\tau)$ is the distribution function of a measure of mass $\tau$, while each $F(x,\cdot)$ is increasing and $1$-Lipschitz. Compressions are continuum analogues of Gelfand--Tsetlin patterns: if $(t_{k,j})_{ 0 \leq j \leq k \leq n}$ is a Gelfand--Tsetlin pattern, setting $F(t_{k,j},k/n) = j/n$ and interpolating creates a compression. For a differentiable compression $F$, we define the compression entropy \begin{align*} \mathcal{H}[F] :=\int_{-\infty}^\infty\int_0^1 F_x \left\{-\log F_x+\log\sin(\pi F_\tau)+1-\log\pi\right\} \mathrm{d}\tau\mathrm{d}x. \end{align*} If $\mu$ is absolutely continuous and compactly supported, we prove \begin{equation*} \sup\left\{\mathcal{H}[F]:F \text{ compression}, F(\cdot,1)\text{ is the distribution function of }\mu\right\} =\chi[\mu], \end{equation*} where $\chi[\mu]$ is Voiculescu's free entropy. By identifying the Euler--Lagrange equations for $\mathcal{H}[F]$ with a Burgers equation for Cauchy transforms, we show that the supremum is attained uniquely by the free compression of free probability theory. We also view Gelfand--Tsetlin patterns as Ginzburg--Landau $\nabla\phi$-interface models with a hard-core interaction, and compute the surface tension: \begin{equation*} \sigma(u_1,u_2) =-\log(u_1+u_2)-\log\sin\left(\pi\frac{u_1}{u_1+u_2}\right)-1+\log\pi. \end{equation*} Finally, we prove that uniform $n$-dimensional Gelfand--Tsetlin patterns with deterministic bottom rows converging to $\mu$ satisfy a large deviation principle with speed $n^2$ and rate function \begin{equation*} I_\mu[F]=-\mathcal{H}[F]+\chi[\mu]. \end{equation*} These results resolve a conjecture of Shlyakhtenko and Tao stating that the Euler--Lagrange equations for free compression arise from the statistical mechanics of interlacing point processes.

math.PR

Talagrand's mathematical journey to the Abel Prize 2024

Michel Talagrand (Centre National de la Recherche Scientifique, France) has been awarded the prestigious Abel Prize for 2024 for his work in probability theory, functional analysis, and statistical physics. In this note, we introduce the Abel Prize Laureate and his main contributions.

math.HO

The largest fragment in self-similar fragmentation processes of positive index

We study a self-similar fragmentation process with dislocation measure $\nu$ and self-similarity index $\alpha > 0$. Let $e^{-m_t}$ denote the size of the largest fragment at time $t \geq 0$. For dislocation measures satisfying a regularity condition of the form $\nu(1 - s_1 > \delta) = \delta^{-\theta} \ell(1/\delta)$ with $\theta \in [0,1)$ and slowly varying $\ell$, we prove almost sure convergence \[ \lim_{t \to \infty} (m_t - g(t)) = 0, \] where $g(t) = (\log t - (1 - \theta) \log \log t + f(t))/\alpha$, and $f(t) = o(\log \log t)$ is a lower order correction that can be described explicitly in terms of $\ell$ and $\theta$. Our results sharpen substantially the best prior result on general self-similar fragmentation processes, due to Bertoin, which states that $m_t = (1+o(1)) \log (t)/\alpha$.

math.PR

Sharp concentration phenomena in high-dimensional Orlicz balls

In this article, we present a precise deviation formula for the intersection of two Orlicz balls generated by Orlicz functions $V$ and $W$. Additionally, we establish a (quantitative) central limit theorem in the critical case and a strong law of large numbers for the "$W$-norm" of the uniform distribution on $\mathbb{B}^{(n,V)}$. Our techniques also enable us to derive a precise formula for the thin-shell concentration of uniformly distributed random vectors in high-dimensional Orlicz balls. In our approach we establish an Edgeworth-expansion using methods from harmonic analysis together with an exponential change of measure argument.

math.PR

Asymptotic theory of Schatten classes

The study of Schatten classes has a long tradition in geometric functional analysis and related fields. In this paper we study a variety of geometric and probabilistic aspects of finite-dimensional Schatten classes of not necessarily square matrices. Among the main results are the exact and asymptotic volume of the Schatten-$\infty$ unit ball, the boundedness of its isotropy constant, a Poincar\'e-Maxwell-Borel lemma for the uniform distribution on the Schatten-$\infty$ ball, and Sanov-type large deviations principles for the singular values of matrices sampled uniformly from the Schatten-$p$ unit ball for any $0 < p \leq \infty$.

math.FA

Entropy numbers of finite-dimensional Lorentz space embeddings

The sequence of entropy numbers quantifies the degree of compactness of a linear operator acting between quasi-Banach spaces. We determine the asymptotic behavior of entropy numbers in the case of natural embeddings between finite-dimensional Lorentz spaces $\ell_{p,q}^n$ in all regimes; our results are sharp up to constants. This generalizes classical results obtained by Sch\"utt (in the case of Banach spaces) and Edmunds and Triebel, K\"uhn, as well as Gu\'edon and Litvak (in the case of quasi-Banach spaces) for entropy numbers of identities between finte-dimensional Lebesgue sequence spaces $\ell_p^n$. We employ techniques such as interpolation, volume comparison as well as techniques from sparse approximation and combinatorial arguments. Further, we characterize entropy numbers of embeddings between finite-dimensional symmetric quasi-Banach spaces in terms of best $s$-term approximation numbers.

math.FA

Random approximation of convex bodies in Hausdorff metric

While there is extensive literature on approximation, deterministic as well as random, of general convex bodies $K$ in the symmetric difference metric, or other metrics arising from intrinsic volumes, very little is known for corresponding random results in the Hausdorff distance when the approximant $K_n$ is given by the convex hull of $n$ independent random points chosen uniformly on the boundary or in the interior of $K$. When $K$ is a polygon and the points are chosen on its boundary, we determine the exact limiting behavior of the expected Hausdorff distance between a polygon as $n\to\infty$. From this we derive the behavior of the asymptotic constant for a regular polygon in the number of vertices.

math.MG

The large and moderate deviations approach in geometric functional analysis

The work of Gantert, Kim, and Ramanan [Large deviations for random projections of $\ell^p$ balls, Ann. Probab. 45 (6B), 2017] has initiated and inspired a new direction of research in the asymptotic theory of geometric functional analysis. The moderate deviations perspective, describing the asymptotic behavior between the scale of a central limit theorem and a large deviations principle, was later added by Kabluchko, Prochno, and Th\"ale in [High-dimensional limit theorems for random vectors in $\ell_p^n$ balls. II, Commun. Contemp. Math. 23(3), 2021]. These two approaches nicely complement the classical study of central limit phenomena or non-asymptotic concentration bounds for high-dimensional random geometric quantities. Beyond studying large and moderate deviations principles for random geometric quantities that appear in geometric functional analysis, other ideas emerged from the theory of large deviations and the closely related field of statistical mechanics, and have provided new insight and become the origin for new developments. Within less than a decade, a variety of results have appeared and formed this direction of research. Recently, a connection to the famous Kannan-Lov\'asz-Simonovits conjecture and the study of moderate and large deviations for isotropic log-concave random vectors was discovered. In this manuscript, we introduce the basic principles, survey the work that has been done, and aim to manifest this direction of research, at the same time making it more accessible to a wider community of researchers.

math.FA

The Maclaurin inequality through the probabilistic lens

In this paper we take a probabilistic look at Maclaurin's inequality, which is a refinement of the classical AM-GM inequality. In a natural randomized setting, we obtain limit theorems and show that a reverse inequality holds with high probability. The form of Maclaurin's inequality naturally relates it to U-statistics. More precisely, given $x_1, \ldots, x_n, p \in (0,\infty)$ and $k \in \mathbb{N}$ with $k \leq n$, let us define the quantity \[ S_{k, p}^{(n)} = \Big( \tbinom{n}{k}^{-1} \sum_{1 \leq i_1 < \ldots < i_k \leq n} x_{i_1}^p \cdots x_{i_k}^p \Big)^{1/(k p)}.\] Then as a consequence of the classical Maclaurin inequalities, we know that $S_{k_1}^{(n)} \geq S_{k_2}^{(n)}$ for $k_1 < k_2$. In the present article we consider the ratio \[ \mathcal{R}_{k_1, k_2, p}^{(n)} := \frac{S_{k_2, p}^{(n)}}{S_{k_1, p}^{(n)}}, \] evaluated at a random vector $(X_1, \ldots, X_n)$ sampled either from the normalized surface measure on the $\ell_p^n$-sphere or from a distribution generalizing both the uniform distribution on the $\ell_p^n$-ball and the cone measure on the $\ell_p^n$-sphere; by the Maclaurin inequality, we always have $\mathcal{R}_{k_1, k_2, p}^{(n)} \leq 1$. We derive central limit theorems for $\mathcal{R}_{k_1, k_2, p}^{(n)}$ and $\mathcal{R}_{k_1, n, p}^{(n)}$ as well as Berry--Esseen bounds and a moderate deviations principle for $\mathcal{R}_{k_1, n, p}^{(n)}$, keeping $k_1$, $k_2$ fixed, in order to quantify the set of points where $\mathcal{R}_{k_1, k_2, p}^{(n)} > c$ for $c \in (0, 1)$, i.e., where the Maclaurin inequality is reversed up to a factor. The present aricle partly generalizes results concerning the AM-GM inequality obtained by Kabluchko, Prochno, and Vysotsky (2020), Th\"ale (2021), and Kaufmann and Th\"ale (2023+).

math.PR

Moderate deviation principles and Mod-Gaussian convergence for lacunary trigonometric sums

Classical works of Kac, Salem and Zygmund, and Erd\H{o}s and G\'{a}l have shown that lacunary trigonometric sums despite their dependency structure behave in various ways like sums of independent and identically distributed random variables. For instance, they satisfy a central limit theorem and a law of the iterated logarithm. Those results have only recently been complemented by large deviation principles by Aistleitner, Gantert, Kabluchko, Prochno, and Ramanan, showing that interesting phenomena occur on the large deviation scale that are not visible in the classical works. This raises the question on what scale such phenomena kick in. In this paper, we provide a first step towards a resolution of this question by studying moderate deviation principles for lacunary trigonometric sums. We show that no arithmetic affects are visible between the CLT scaling $\sqrt{n}$ and a scaling $n/\log(n)$ that is only a logarithmic gap away from the large deviations scale. To obtain our results, inspired by the notion of a dependency graph, we introduce correlation graphs and use the method of cumulants. In this work we also obtain results on the mod-Gaussian convergence using different tools.

math.PR

Diophantine conditions in the law of the iterated logarithm for lacunary systems

It is a classical observation that lacunary function systems exhibit many properties which are typical for systems of independent random variables. However, it had already been observed by Erd\H{o}s and Fortet in the 1950s that probability theory's limit theorems may fail for lacunary sums $\sum f(n_k x)$ if the sequence $(n_k)_{k \geq 1}$ has a strong arithmetic ''structure''. The presence of such structure can be assessed in terms of the number of solutions $k,\ell$ of two-term linear Diophantine equations $a n_k - b n_\ell = c$. As the first author proved with Berkes in 2010, saving an (arbitrarily small) unbounded factor for the number of solutions of such equations compared to the trivial upper bound, rules out pathological situations as in the Erd\H{o}s--Fortet example, and guarantees that $\sum f(n_k x)$ satisfies the central limit theorem (CLT) in a form which is in accordance with true independence. In contrast, as shown by the first author, for the law of the iterated logarithm (LIL) the Diophantine condition which suffices to ensure ''truly independent'' behavior requires saving this factor of logarithmic order. In the present paper we show that, rather surprisingly, saving such a logarithmic factor is actually the optimal condition in the LIL case. This result reveals the remarkable fact that the arithmetic condition required of $(n_k)_{k \geq 1}$ to ensure that $\sum f(n_k x)$ shows ''truly random'' behavior is a different one at the level of the CLT than it is at the level of the LIL: the LIL requires a stronger arithmetic condition than the CLT does.

math.NT

A probabilistic approach to Lorentz balls

We develop a probabilistic approach to study the volumetric and geometric properties of unit balls $\mathbb B_{q,1}^n$ of finite-dimensional Lorentz sequences spaces $\ell_{q,1}^n$. More precisely, we show that the empirical distribution of a random vector $X^{(n)}$ uniformly distributed on the volume normalized Lorentz ball in $\mathbb R^n$ converges weakly to a compactly supported symmetric probability distribution with explicitly given density; as a consequence we obtain a weak Poincar\'e-Maxwell-Borel principle for any fixed number $k\in\mathbb N$ of coordinates of $X^{(n)}$ as $n\to\infty$. Moreover, we prove a central limit theorem for the largest coordinate of $X^{(n)}$, demonstrating a quite different behavior than in the case of the $\ell_q^n$ balls, where a Gumbel distribution appears in the limit. Last but not least, we prove a Schechtman-Schmuckenschl\"ager type result for the asymptotic volume of intersections of volume normalized Lorentz and $\ell^n_p$ balls.

math.FA