Searcharxiv⌕ Search

arXiv subjects

Mario Ullrich

Publications and source records attributed to Mario Ullrich.

At least 19 recordsLinked to original sources
arXiv2026

On bounds between all s-numbers and widths of convex sets

We prove $a_n(S) \le e\,(n+1)\, s_n(S)$ for every s-number sequence $(s_n)$, every bounded linear operator $S$ between normed spaces, and every $n \in \mathbb{N}_0$, where $a_n$ are the approximation numbers, which are the largest s-numbers. This is sharp up to the constant and settles conjectures of Mityagin, Henkin, Carl and Pietsch dating back to 1963. We also extend it to widths of convex sets and discuss optimality there. The proof is elementary.

math.FA↗
arXiv2026

Constructive discretization and approximation in reproducing kernel Hilbert spaces

We generalize the sparsification algorithm of Batson, Spielman and Srivastava, making one part of the result dimension-independent. In particular, we recover discretization inequalities in $L_2$- and sup-norms on general finite-dimensional subspaces, prove a suitable infinite-dimensional variant, and discuss the implications for the error of least-squares approximation based on samples. This gives a more constructive version of several recently established approximation bounds, some of which relied on the stronger and less constructive result of Marcus, Spielman and Srivastava. We also improve the constants and oversampling factors in these results.

math.NA↗
arXiv2026

Approximation of Functions: Optimal Sampling and Complexity

We consider approximation or recovery of functions based on a finite number of function evaluations. This is a well-studied problem in optimal recovery, machine learning, and numerical analysis in general, but many fundamental insights were obtained only recently. We discuss different aspects of the information-theoretic limit that appears because of the limited amount of data available, as well as algorithms and sampling strategies that come as close to it as possible. We also discuss (optimal) sampling in a broader sense, allowing other types of measurements that may be nonlinear, adaptive and random, and present several relations between the different settings in the spirit of information-based complexity. We hope that this article provides both, a basic introduction to the subject and a contemporary summary of the current state of research.

math.NA↗
arXiv2025

Noisy nonlinear information and entropy numbers

It is impossible to recover a vector from $\mathbb{R}^m$ with less than $m$ linear measurements, even if the measurements are chosen adaptively. Recently, it has been shown that one can recover vectors from $\mathbb{R}^m$ with arbitrary precision using only $O(\log m)$ continuous (even Lipschitz) adaptive measurements, resulting in an exponential speed-up of continuous information compared to linear information for various approximation problems. In this note, we characterize the quality of optimal (dis-)continuous information that is disturbed by deterministic noise in terms of entropy numbers. This shows that in the presence of noise the potential gain of continuous over linear measurements is limited, but significant in some cases.

math.NA↗
arXiv2025

Sparse grids vs. random points for high-dimensional polynomial approximation

We study polynomial approximation on a $d$-cube, where $d$ is large, and compare interpolation on sparse grids, aka Smolyak's algorithm (SA), with a simple least squares method based on randomly generated points (LS) using standard benchmark functions. Our main motivation is the influential paper [Barthelmann, Novak, Ritter: High dimensional polynomial interpolation on sparse grids, Adv. Comput. Math. 12, 2000]. We repeat and extend their theoretical analysis and numerical experiments for SA and compare to LS in dimensions up to 100. Our extensive experiments demonstrate that LS, even with only slight oversampling, consistently matches the accuracy of SA in low dimensions. In high dimensions, however, LS shows clear superiority.

math.NA↗
arXiv2025

Nonlocal techniques for the analysis of deep ReLU neural network approximations

Recently, Daubechies, DeVore, Foucart, Hanin, and Petrova introduced a system of piece-wise linear functions, which can be easily reproduced by artificial neural networks with the ReLU activation function and which form a Riesz basis of $L_2([0,1])$. This work was generalized by two of the authors to the multivariate setting. We show that this system serves as a Riesz basis also for Sobolev spaces $W^s([0,1]^d)$ and Barron classes ${\mathbb B}^s([0,1]^d)$ with smoothness $0<s<1$. We apply this fact to re-prove some recent results on the approximation of functions from these classes by deep neural networks. Our proof method avoids using local approximations and allows us to track also the implicit constants as well as to show that we can avoid the curse of dimension. Moreover, we also study how well one can approximate Sobolev and Barron functions by ANNs if only function values are known.

cs.LG↗
arXiv2024

How many continuous measurements are needed to learn a vector?

One can recover vectors from $\mathbb{R}^m$ with arbitrary precision, using only $\lceil \log_2(m+1)\rceil +1$ continuous measurements that are chosen adaptively. This surprising result is explained and discussed, and we present applications to infinite-dimensional approximation problems.

math.NA↗
arXiv2024

On the power of adaption and randomization

We present bounds on the maximal gain of adaptive and randomized algorithms over non-adaptive, deterministic ones for approximating linear operators on convex sets. If the sets are additionally symmetric, then our results are optimal. For non-symmetric sets, we unify some notions of $n$-widths and s-numbers, and show their connection to minimal errors. We also discuss extensions to non-linear widths and approximation based on function values, and conclude with a list of open problems.

math.NA↗
arXiv2024

Inequalities between s-numbers

Singular numbers of operators between Hilbert spaces were generalized to Banach spaces by s-numbers (in the sense of Pietsch). This allows for different choices, including approximation, Gelfand, Kolmogorov and Bernstein numbers. Here, we present an elementary proof of a bound between the smallest and the largest s-number.

math.FA↗
arXiv2024

Sampling projections in the uniform norm

We show that there are sampling projections on arbitrary $n$-dimensional subspaces of $B(D)$ with at most $2n$ samples and norm of order $\sqrt{n}$, where $B(D)$ is the space of complex-valued bounded functions on a set $D$. This gives a more explicit form of the Kadets-Snobar theorem for the uniform norm and improves upon Auerbach's lemma. We discuss consequences for optimal recovery in $L_p$.

math.FA↗
arXiv2023

On the power of iid information for linear approximation

This survey is concerned with the power of random information for approximation in the (deterministic) worst-case setting, with special emphasis on information consisting of functionals selected independently and identically distributed (iid) at random on a class of admissible information functionals. We present a general result based on a weighted least squares method and derive consequences for special cases. Improvements are available if the information is ``Gaussian'' or if we consider iid function values for Sobolev spaces. We include open questions to guide future research on the power of random information in the context of information-based complexity.

math.NA↗
arXiv2023

Sampling recovery in $L_2$ and other norms

We study the recovery of functions in various norms, including $L_p$ with $1\le p\le\infty$, based on function evaluations. We obtain worst case error bounds for general classes of functions in terms of the best $L_2$-approximation from a given nested sequence of subspaces and the Christoffel function of these subspaces. In the case $p=\infty$, our results imply that linear sampling algorithms are optimal up to a constant factor for many reproducing kernel Hilbert spaces.

math.NA↗
arXiv2022

Exponential tractability of $L_2$-approximation with function values

We study the complexity of high-dimensional approximation in the $L_2$-norm when different classes of information are available; we compare the power of function evaluations with the power of arbitrary continuous linear measurements. Here, we discuss the situation when the number of linear measurements required to achieve an error $\varepsilon \in (0,1)$ in dimension $d\in\mathbb{N}$ depends only poly-logarithmically on $\varepsilon^{-1}$. This corresponds to an exponential order of convergence of the approximation error, which often happens in applications. However, it does not mean that the high-dimensional approximation problem is easy, the main difficulty usually lies within the dependence on the dimension $d$. We determine to which extent the required amount of information changes, if we allow only function evaluation instead of arbitrary linear information. It turns out that in this case we only lose very little, and we can even restrict to linear algorithms. In particular, several notions of tractability hold simultaneously for both types of available information.

math.NA↗
arXiv2022

A sharp upper bound for sampling numbers in $L_{2}$

For a class $F$ of complex-valued functions on a set $D$, we denote by $g_n(F)$ its sampling numbers, i.e., the minimal worst-case error on $F$, measured in $L_2$, that can be achieved with a recovery algorithm based on $n$ function evaluations. We prove that there is a universal constant $c\in\mathbb{N}$ such that, if $F$ is the unit ball of a separable reproducing kernel Hilbert space, then \[ g_{cn}(F)^2 \,\le\, \frac{1}{n}\sum_{k\geq n} d_k(F)^2, \] where $d_k(F)$ are the Kolmogorov widths (or approximation numbers) of $F$ in $L_2$. We also obtain similar upper bounds for more general classes $F$, including all compact subsets of the space of continuous functions on a bounded domain $D\subset \mathbb{R}^d$, and show that these bounds are sharp by providing examples where the converse inequality holds up to a constant. The results rely on the solution to the Kadison-Singer problem, which we extend to the subsampling of a sum of infinite rank-one matrices.

math.NA↗
arXiv2020

Function values are enough for $L_2$-approximation: Part II

In the first part we have shown that, for $L_2$-approximation of functions from a separable Hilbert space in the worst-case setting, linear algorithms based on function values are almost as powerful as arbitrary linear algorithms if the approximation numbers are square-summable. That is, they achieve the same polynomial rate of convergence. In this sequel, we prove a similar result for separable Banach spaces and other classes of functions.

math.NA↗
arXiv2020

On the worst-case error of least squares algorithms for $L_2$-approximation with high probability

It was recently shown in [4] that, for $L_2$-approximation of functions from a Hilbert space, function values are almost as powerful as arbitrary linear information, if the approximation numbers are square-summable. That is, we showed that \[ e_n \,\lesssim\, \sqrt{\frac{1}{k_n} \sum_{j\geq k_n} a_j^2} \qquad \text{ with }\quad k_n \asymp \frac{n}{\ln(n)}, \] where $e_n$ are the sampling numbers and $a_k$ are the approximation numbers. In particular, if $(a_k)\in\ell_2$, then $e_n$ and $a_n$ are of the same polynomial order. For this, we presented an explicit (weighted least squares) algorithm based on i.i.d. random points and proved that this works with positive probability. This implies the existence of a good deterministic sampling algorithm. Here, we present a modification of the proof in [4] that shows that the same algorithm works with probability at least $1-{n^{-c}}$ for all $c>0$.

math.NA↗
arXiv2019

On the fixed volume discrepancy of the Fibonacci sets in the integral norms

This paper is devoted to the study of a discrepancy-type characteristic -- the fixed volume discrepancy -- of the Fibonacci point set in the unit square. It was observed recently that this new characteristic allows us to obtain optimal rate of dispersion from numerical integration results. This observation motivates us to thoroughly study this new version of discrepancy, which seems to be interesting by itself. The new ingredient of this paper is the use of the average over the shifts of hat functions instead of taking the supremum over the shifts. We show that this change in the setting results in an improvement of the upper bound for the smooth fixed volume discrepancy, similarly to the well-known results for the usual $L_p$-discrepancy. Interestingly, this shows that ``bad boxes'' for the usual discrepancy cannot be ``too small''. The known results on smooth discrepancy show that the obtained bounds cannot be improved in a certain sense.

math.NA↗