approximation-theory

first commit

Code for the paper "Rational neural networks", NeurIPS 2020

Codes for the paper 'Rational Neural Networks for Approximating Jump Discontinuities of Graph Convolution Operator' ICDM 18

This directory contains lecture notes for an undergraduate numerical analysis course offered in Spring 2018, at UT Austin.

Neural network architecture with learnable fractional power bases

Official code repo for STAI-X 2026 paper "Transformer as Provable Approximators of Sparse Principal Component Analysis".

From the Weierstrass M-test to the RBF kernel: uniform convergence with explicit error bounds, applied to kernel methods and Gaussian blur

Numerical study of convergence rates for approximating transport maps

A numerical methods project exploring different approaches to computing trigonometric functions. Combining modern C++, Python, and LaTeX to investigate approximation algorithms through reproducible numerical experiments.

Reproducible code, data and figures for the paper "Optimal selection of the preserved exponential in Szász–Mirakyan operators: a leading-term criterion validated at finite n" (submitted to DRNA).

Preliminary numerical experiments for some greedy algorithms.

Description: A straightedge-and-compass constructible cubic error cascade for approximate angle trisection.

Volume VII of Learning Real Analysis: Numerical analysis and approximation theory.

Rigorous analysis of the Amundson sequence G(n) = n^(n+1)/(n+1)^n and the constant A_G: Hausdorff moment representation, compound-Poisson/Levy form, fibre lattices, and endpoint Christoffel asymptotics (Theorem C0).

Certified minimax approximation of the logistic-normal integral

Code for the technical note "Chebyshev-Controlled Chernoff Approximations of the Heat Semigroup" (doi:10.5281/zenodo.21708475).

Function code in python of The fourth order Runge Kutta method to compute approximated solutions of ordinary differential equations and its documentation.

A port of Chebfun from MATLAB to GNU Octave.

Proof 7: Why the Cantor function (Devil's Staircase) is structurally hard for ReLU networks. All theorems verified.