English

Lower Bounds for Sparse Oblivious Subspace Embeddings

Data Structures and Algorithms 2021-12-22 v1 Computational Geometry Discrete Mathematics

Abstract

An oblivious subspace embedding (OSE), characterized by parameters m,n,d,ϵ,δm,n,d,\epsilon,\delta, is a random matrix ΠRm×n\Pi\in \mathbb{R}^{m\times n} such that for any dd-dimensional subspace TRnT\subseteq \mathbb{R}^n, PrΠ[xT,(1ϵ)x2Πx2(1+ϵ)x2]1δ\Pr_\Pi[\forall x\in T, (1-\epsilon)\|x\|_2 \leq \|\Pi x\|_2\leq (1+\epsilon)\|x\|_2] \geq 1-\delta. For ϵ\epsilon and δ\delta at most a small constant, we show that any OSE with one nonzero entry in each column must satisfy that m=Ω(d2/(ϵ2δ))m = \Omega(d^2/(\epsilon^2\delta)), establishing the optimality of the classical Count-Sketch matrix. When an OSE has 1/(9ϵ)1/(9\epsilon) nonzero entries in each column, we show it must hold that m=Ω(ϵO(δ)d2)m = \Omega(\epsilon^{O(\delta)} d^2), improving on the previous Ω(ϵ2d2)\Omega(\epsilon^2 d^2) lower bound due to Nelson and Nguyen (ICALP 2014).

Keywords

Cite

@article{arxiv.2112.10987,
  title  = {Lower Bounds for Sparse Oblivious Subspace Embeddings},
  author = {Yi Li and Mingmou Liu},
  journal= {arXiv preprint arXiv:2112.10987},
  year   = {2021}
}
R2 v1 2026-06-24T08:25:40.687Z