Lower Bounds for Sparse Oblivious Subspace Embeddings

12/21/2021

∙

by   Yi Li, et al.

∙

0

∙

share

An oblivious subspace embedding (OSE), characterized by parameters m,n,d,ϵ,δ, is a random matrix Π∈ℝ^m× n such that for any d-dimensional subspace T⊆ℝ^n, _Π[∀ x∈ T, (1-ϵ)x_2 ≤Π x_2≤ (1+ϵ)x_2] ≥ 1-δ. For ϵ and δ at most a small constant, we show that any OSE with one nonzero entry in each column must satisfy that m = Ω(d^2/(ϵ^2δ)), establishing the optimality of the classical Count-Sketch matrix. When an OSE has 1/(9ϵ) nonzero entries in each column, we show it must hold that m = Ω(ϵ^O(δ) d^2), improving on the previous Ω(ϵ^2 d^2) lower bound due to Nelson and Nguyen (ICALP 2014).