English

Fast and RIP-optimal transforms

Numerical Analysis 2013-02-19 v2 Information Theory math.IT

Abstract

We study constructions of k×nk \times n matrices AA that both (1) satisfy the restricted isometry property (RIP) at sparsity ss with optimal parameters, and (2) are efficient in the sense that only O(nlogn)O(n\log n) operations are required to compute AxAx given a vector xx. Our construction is based on repeated application of independent transformations of the form DHDH, where HH is a Hadamard or Fourier transform and DD is a diagonal matrix with random {+1,1}\{+1,-1\} elements on the diagonal, followed by any k×nk \times n matrix of orthonormal rows (e.g.\ selection of kk coordinates). We provide guarantees (1) and (2) for a larger regime of parameters for which such constructions were previously unknown. Additionally, our construction does not suffer from the extra poly-logarithmic factor multiplying the number of observations kk as a function of the sparsity ss, as present in the currently best known RIP estimates for partial random Fourier matrices and other classes of structured random matrices.

Keywords

Cite

@article{arxiv.1301.0878,
  title  = {Fast and RIP-optimal transforms},
  author = {Nir Ailon and Holger Rauhut},
  journal= {arXiv preprint arXiv:1301.0878},
  year   = {2013}
}
R2 v1 2026-06-21T23:04:18.284Z