Explicit matrices with the restricted isometry property
A matrix has the restricted isometry property of order k if it approximately preserves the length of every vector with at most k nonzero coordinates. Can a deterministic construction approach the sparsity achieved by random matrices while using far fewer rows than columns?
Why it matters and what counts as progress
Why it matters: Such matrices let compressed sensing preserve sparse signals after taking fewer measurements. Deterministic constructions give a way to choose the measurements without a random draw.
What counts as progress: Construct an explicit family with increasing sparsity order and distortion tending to zero while the ratio of rows to columns tends to zero. A lower bound can rule out a specified construction or class of methods.
How the problem developed
- 2007 · Problem statement [5]
Terence Tao Asked for deterministic constructions comparable to random restricted isometries.
- 2011 · Square-root barrier surpassed [6]
Jean Bourgain and colleagues Used cancellation to obtain an explicit family with sparsity exponent above one half.
Christian Kümmerle and Dustin G. Mixon with GPT-6 Astra Posted an explicit family with vanishing distortion and sparsity exponent approaching two thirds.
AI contribution
Explicit restricted isometries with
Kümmerle and Mixon give an explicit family of real matrices with rows and unit-norm columns. As grows, and the restricted isometry distortion go to zero at sparsity .
Sources
Publication history
Daily emails by default, only when there’s an update.