Article
Computer Science, Theory & Methods
Matthieu Dolbeault, Albert Cohen
Summary: This study demonstrates that near-best approximation of a function in a linear subspace of dimension n can be computed through a near-optimal budget of pointwise evaluations. Sampling points are drawn according to a random distribution, and the approximation is calculated using a weighted least-squares method, with error assessed in expected L-2 norm. This result improves upon previous findings and shows the dominance of the sampling number in the randomized setting by the Kolmogorov n-width.
JOURNAL OF COMPLEXITY
(2022)
Article
Computer Science, Theory & Methods
David Krieg, Mario Ullrich
Summary: In this paper, a similar result is proven for separable Banach spaces and other classes of functions in the worst-case setting for L-2 approximation. It shows that linear algorithms based on function values can achieve the same polynomial rate of convergence as arbitrary linear algorithms if the linear widths are square-summable.
JOURNAL OF COMPLEXITY
(2021)
Article
Computer Science, Theory & Methods
Nicolas Nagel, Martin Schaefer, Tino Ullrich
Summary: In this study, a new upper bound for sampling numbers associated with the compact embedding of a separable reproducing kernel Hilbert space into the space of square integrable functions is provided. The algorithm for realizing this bound uses a least squares algorithm based on a specific set of sampling nodes. The non-constructive result implies that for dimensions d>2, any sparse grid sampling recovery method does not perform asymptotically optimal.
FOUNDATIONS OF COMPUTATIONAL MATHEMATICS
(2022)
Article
Mathematics, Applied
Baasansuren Jadamba, Akhtar A. Khan, Fabio Raciti, Miguel Sama
Summary: This paper develops a stochastic approximation approach for estimating the flexural rigidity within the framework of variational inequalities. The nonlinear inverse problem is analyzed as a stochastic optimization problem using an energy least-squares formulation. A stochastic variational inequality is solved by a stochastic auxiliary problem principle-based iterative scheme, which satisfies the necessary and sufficient optimality condition for the optimization problem. The convergence analysis for the proposed iterative scheme is given under general conditions on the random noise. Detailed computational results demonstrate the feasibility and efficacy of the proposed methodology.
COMMUNICATIONS IN NONLINEAR SCIENCE AND NUMERICAL SIMULATION
(2022)
Article
Mathematics, Applied
Giovanni Migliorati
Summary: This paper introduces numerical algorithms based on weighted least squares for approximating bounded real-valued functions. Stable and optimally converging estimators can be constructed even when an orthonormal basis is not available, by using a suitable surrogate basis. The computational cost depends on specific functions, and numerical results validate the theoretical analysis.
IMA JOURNAL OF NUMERICAL ANALYSIS
(2021)
Article
Computer Science, Theory & Methods
V. Temlyakov
Summary: We have proven that the optimal error of recovery in the L-2 norm of functions from a class F can be bounded by the value of the Kolmogorov width of F. The obtained inequality is shown to be a powerful tool for estimating errors of optimal recovery in functions with mixed smoothness.
JOURNAL OF COMPLEXITY
(2021)
Article
Computer Science, Software Engineering
Sandra Merchel, Bert Juettler, Dominik Mokris, Maodong Pan
Summary: This paper proposes an acceleration method for least-squares fitting by tensor-product spline surfaces, reducing the time consumption of assembling the system of equations by using sum factorization and regular grids.
COMPUTER-AIDED DESIGN
(2022)
Article
Mathematics
Qianqian Hu, Zhifang Wang, Ruyi Liang
Summary: Geometric iterative methods, such as progressive iterative approximation and geometric interpolation methods, are efficient for fitting data sets. This paper presents an accelerated LSPIA method for tensor product surfaces by combining the Schulz iterative method with the traditional LSPIA method, showing that the corresponding iterative surface sequence converges to the least-squares fitting surface. The proposed method exhibits a faster convergence rate compared to other methods, as illustrated by numerical examples.
Article
Mathematics, Applied
David Krieg, Pawel Siedlecki, Mario Ullrich, Henryk Wozniakowski
Summary: In this paper, we study high-dimensional approximation and compare the power of function evaluations and arbitrary continuous linear measurements when different classes of information are available. We find that the number of linear measurements required for a given accuracy depends only on epsilon(-1) in a poly-logarithmic manner. It is shown that allowing only function evaluation instead of arbitrary linear information does not result in a significant loss and even allows for linear algorithms. Both types of available information satisfy several notions of tractability simultaneously.
ADVANCES IN COMPUTATIONAL MATHEMATICS
(2023)
Article
Computer Science, Theory & Methods
David Krieg, Mario Ullrich
Summary: The study examines the L-2-approximation of functions from a Hilbert space and compares the sampling numbers with the approximation numbers. It shows that the sampling numbers decay with the same polynomial rate as the approximation numbers, particularly for Sobolev spaces with dominating mixed smoothness. This result improves upon previous bounds and disproves the prevalent conjecture about the optimality of Smolyak's algorithm.
FOUNDATIONS OF COMPUTATIONAL MATHEMATICS
(2021)
Article
Mathematics, Applied
Z. El Majouti, R. El Jid, A. Hajjaj
Summary: The article expands the three-dimensional modified moving least-square method for solving high-dimensional linear and nonlinear integral equations, without the need for mesh connectivity, with support size having a significant effect on maximum errors. The MMLS method with a non-singular moment matrix achieves better results than MLS approximation, and numerical experiments demonstrate the differences between the two methods for multidimensional problems.
INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS
(2022)
Article
Mathematics, Applied
Fleurianne Bertrand, Daniele Boffi
Summary: The study focuses on the approximation of the spectrum of least-squares operators in linear elasticity problems. By considering two different formulations and conducting numerical experiments, the theoretical results are confirmed.
COMPUTERS & MATHEMATICS WITH APPLICATIONS
(2021)
Article
Mathematics, Applied
Kerstin Hesse, Quoc Thong Le Gia
Summary: This paper investigates discrete penalized least-squares approximation on the unit sphere, providing error estimates for the approximation of functions from the Sobolev Hilbert space and showcasing the effects of different choices of the regularization parameter.
JOURNAL OF COMPUTATIONAL AND APPLIED MATHEMATICS
(2022)
Article
Computer Science, Interdisciplinary Applications
Ji Hee Kim, Naeun Choi, Seongmin Heo
Summary: This work proposes a novel iterative least squares method to approximate nonlinear functions using constrained least squares to ensure continuity. The method improves upon the existing continuous piecewise linear (CPWL) method by modifying the main steps and employing partitioned least squares and constrained least squares to reduce computational complexity. An iterative procedure with gradient descent using momentum is used for breakpoint updates to improve convergence characteristics.
COMPUTERS & CHEMICAL ENGINEERING
(2022)
Article
Mathematics, Applied
Michael S. Floater
Summary: In this note, a solution is derived for the problem of finding a polynomial of degree at most $n$ that best approximates data at $n+2$ points in the $l_{p}$ norm. The solution can be expressed as a convex combination of Lagrange interpolants over subsets of $n+1$ points, and the error oscillates in sign.
IMA JOURNAL OF NUMERICAL ANALYSIS
(2023)
Article
Mathematics
David Krieg
CONSTRUCTIVE APPROXIMATION
(2019)
Article
Computer Science, Theory & Methods
David Krieg
JOURNAL OF COMPLEXITY
(2018)
Article
Mathematics
David Krieg, Daniel Rudolf
JOURNAL OF APPROXIMATION THEORY
(2019)
Article
Computer Science, Theory & Methods
David Krieg
JOURNAL OF COMPLEXITY
(2019)
Article
Computer Science, Theory & Methods
Aicke Hinrichs, David Krieg, Robert J. Kunsch, Daniel Rudolf
JOURNAL OF COMPLEXITY
(2020)
Article
Computer Science, Theory & Methods
David Krieg, Mario Ullrich
Summary: The study examines the L-2-approximation of functions from a Hilbert space and compares the sampling numbers with the approximation numbers. It shows that the sampling numbers decay with the same polynomial rate as the approximation numbers, particularly for Sobolev spaces with dominating mixed smoothness. This result improves upon previous bounds and disproves the prevalent conjecture about the optimality of Smolyak's algorithm.
FOUNDATIONS OF COMPUTATIONAL MATHEMATICS
(2021)
Article
Computer Science, Theory & Methods
Aicke Hinrichs, David Krieg, Erich Novak, Jan Vybiral
Summary: We investigate the lower bounds for the worst case error of quadrature formulas using given sample points, focusing on optimal point sets and independently and uniformly distributed points. By utilizing recent results on the positive semi-definiteness of certain matrices related to the product theorem of Schur by Vybiral, our new technique extends to spaces of analytic functions where traditional methods are not applicable.
JOURNAL OF COMPLEXITY
(2021)
Article
Computer Science, Theory & Methods
David Krieg, Mario Ullrich
Summary: In this paper, a similar result is proven for separable Banach spaces and other classes of functions in the worst-case setting for L-2 approximation. It shows that linear algorithms based on function values can achieve the same polynomial rate of convergence as arbitrary linear algorithms if the linear widths are square-summable.
JOURNAL OF COMPLEXITY
(2021)
Article
Mathematics
Aicke Hinrichs, David Krieg, Erich Novak, Joscha Prochno, Mario Ullrich
Summary: The study focuses on the circumradius of the intersection of random subspaces with an rn-dimensional ellipsoid, showing that the random radius R-n can be of the same order as the minimal radius under certain conditions. The research delves into the worst-case error of algorithms based on random information for function approximation, with implications in different scenarios depending on the characteristics of the random variable. Various mathematical tools such as Gaussian processes are utilized in the proofs and analysis.
TRANSACTIONS OF THE AMERICAN MATHEMATICAL SOCIETY
(2021)
Article
Computer Science, Theory & Methods
Aicke Hinrichs, David Krieg, Erich Novak, Jan Vybiral
Summary: Function values are almost as informative as general linear information for L2-approximation of functions, and this paper mainly focuses on proving new lower bounds for this behavior. It is shown that sampling numbers can behave worse than approximation numbers, especially for Sobolev spaces with low smoothness. Additionally, new lower bounds for the integration problem are proven.
JOURNAL OF COMPLEXITY
(2022)
Article
Mathematics, Applied
David Krieg, Erich Novak, Mathias Sonnleitner
Summary: This paper studies Lq-approximation and integration for functions from the Sobolev space W-s (p) (Omega) and compares optimal randomized algorithms with algorithms that can only use identically distributed sample points. The main result is that the same optimal rate of convergence can be achieved if we restrict to identically distributed sampling, except when p = q = infinity.
MATHEMATICS OF COMPUTATION
(2022)
Article
Mathematics, Applied
David Krieg, Pawel Siedlecki, Mario Ullrich, Henryk Wozniakowski
Summary: In this paper, we study high-dimensional approximation and compare the power of function evaluations and arbitrary continuous linear measurements when different classes of information are available. We find that the number of linear measurements required for a given accuracy depends only on epsilon(-1) in a poly-logarithmic manner. It is shown that allowing only function evaluation instead of arbitrary linear information does not result in a significant loss and even allows for linear algorithms. Both types of available information satisfy several notions of tractability simultaneously.
ADVANCES IN COMPUTATIONAL MATHEMATICS
(2023)
Article
Mathematics, Applied
David Krieg, Mathias Sonnleitner
Summary: We demonstrate that independent and uniformly distributed sampling points are almost as effective as optimal sampling points for approximating functions from Sobolev spaces on bounded convex domains in the Lq-norm if q < p. More generally, we characterize the quality of arbitrary sampling point sets via the L-? (O)-norm of the distance function dist(., P), where ? = s(1/q - 1/p)(-1) if q < p and ? = 8 if q = p. This improvement surpasses previous characterizations based on the covering radius of P.
IMA JOURNAL OF NUMERICAL ANALYSIS
(2023)
Article
Mathematics, Applied
David Krieg, Jan Vybiral
Summary: This paper focuses on the integration problem of periodic functions in Hilbert spaces, presenting a new technique that utilizes the Hilbert space structure and a variant of the Schur product theorem. The results obtained using this technique have proven to be superior to the traditional bump-function technique.
JOURNAL OF FOURIER ANALYSIS AND APPLICATIONS
(2023)
Article
Mathematics, Applied
Tian-Yi Zhou, Xiaoming Huo
Summary: This paper investigates the learning ability of deep convolutional neural networks (DCNNs) under both underparameterized and overparameterized settings. The study establishes the learning rates of underparameterized DCNNs without restrictions on parameter or function variable structure. Furthermore, it demonstrates that by adding well-defined layers to a non-interpolating DCNN, some interpolating DCNNs can maintain the good learning rates.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Amine Laghrib, Lekbir Afraites
Summary: This paper proposes a new PDE-based image denoising model that effectively deals with images contaminated by multiplicative noise. The model takes into account the gray level information by introducing a gray level indicator function in the diffusion coefficient, and has shown promising theoretical and numerical results.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Hartmut Fuehr, Irina Shafkulovska
Summary: In this article, we study the mapping properties of metaplectic operators on modulation spaces and provide a full characterization of the pairs for which the operator is well-defined and bounded. We also show that these two properties are equivalent and imply that the operator is a Banach space automorphism. Furthermore, we provide a simple criterion to determine the transferability of well-definedness and boundedness for polynomially bounded weight functions.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Jinjun Li, Zhiyi Wu
Summary: We prove that the Beurling dimensions of the spectra for a class of Moran spectral measures are between 0 and their upper entropy dimensions. Moreover, for such a Moran spectral measure, we show that the Beurling dimension for the spectra of the measure has the intermediate value property. Furthermore, we prove that the set of the spectra whose Beurling dimensions are equal to any fixed value in 0 and the upper entropy dimension has the cardinality of the continuum.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
M. M. Castro, F. A. Gruenbaum, I. Zurrian
Summary: This article introduces the existence of commuting differential operators for families of exceptional orthogonal polynomials, which can be found and exploited. The concept of Fourier Algebras is used, and the application of the result is illustrated through two examples.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Anton Kutsenko, Sergey Danilov, Stephan Juricke, Marcel Oliver
Summary: This paper discusses the relations between the expansion coefficients of a discrete random field analyzed with different hierarchical bases. The focus is on comparing Walsh-Rademacher basis and trigonometric Fourier basis, and it is proven that the rate of spectral decay computed in one basis can be translated to the other in a statistical sense. Explicit expressions for this translation on quadrilateral meshes are provided, and numerical examples are used to illustrate the results.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Albert Chua, Matthew Hirn, Anna Little
Summary: In this paper, we generalize and study finite depth wavelet scattering transforms. We provide norms for these operators and prove their continuity and invariance under specific conditions.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Hung-Hsu Chou, Carsten Gieshoff, Johannes Maly, Holger Rauhut
Summary: In deep learning, over-parameterization is commonly used and leads to implicit bias. This paper analyzes the dynamics of gradient descent and provides insights into implicit bias. The study also explores time intervals for early stopping and presents empirical evidence for implicit bias in various scenarios.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Elena Cordero, Gianluca Giacchi
Summary: This article introduces metaplectic Gabor frames as a natural extension of Gabor frames within the framework of metaplectic Wigner distributions. The authors develop the theory of metaplectic atoms and prove an inversion formula for metaplectic Wigner distributions on Rd. The discretization of this formula yields metaplectic Gabor frames. The study also reveals the relationship between shift-invertible metaplectic Wigner distributions and rescaled short-time Fourier transforms, providing a new characterization of modulation and Wiener amalgam spaces.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Ryan Vaughn, Tyrus Berry, Harbir Antil
Summary: In this paper, we propose a new method to solve elliptic and parabolic partial differential equations (PDEs) with boundary conditions using a finite collection of points sampled from a Riemannian manifold embedded in a Euclidean space. Unlike traditional methods that rely on triangulations, our approach defines quadrature formulas on the unknown manifold using only sample points. Our main result is the consistency of the variational diffusion maps graph Laplacian as an estimator of the Dirichlet energy on the manifold, which improves upon previous results and justifies the relationship between diffusion maps and the Neumann eigenvalue problem. Additionally, we derive the first uniform asymptotic expansion of the diffusion maps kernel integral operator for manifolds with boundary using semigeodesic coordinates. By combining various estimators, we demonstrate how to impose Dirichlet and Neumann conditions for common PDEs based on the Laplacian.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)
Article
Mathematics, Applied
Tal Shnitzer, Hau-Tieng Wu, Ronen Talmon
Summary: This paper proposes an operator-based approach for spatiotemporal analysis of multivariate time-series data. The approach combines manifold learning, Riemannian geometry, and spectral analysis techniques to extract different dynamic modes.
APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS
(2024)