Obstructions to a conjectured conditioning bound for random tensors

Vershynin conjectured that a lower bound for the smallest singular value of a matrix of independent simple random tensors extends to tensor orders $d=O(n)$ under independent subgaussian coordinate assumptions. We disprove this extension using two fixed, bounded, centered, variance-one coordinate distributions. A three-point distribution produces zero columns with probability tending to one at order $d\asymp n/\log n$. An absolutely continuous distribution gives full column rank almost surely at every finite size and an unnormalized smallest singular value at most $e^{-n^2/2}$ with probability tending to one along an explicit sequence $d=o(n)$. Both examples apply when the number of columns is a fixed positive fraction of the ambient dimension, as well as when this fraction tends to one. For the absolutely continuous example, all inverse moments of the smallest singular value diverge at each finite size. We also determine the exact zero-column probability, its Poisson transition, and the largest mass at zero under a prescribed subgaussian norm bound. The proofs are elementary and self-contained. The accompanying Lean formalization is available at \url{https://doi.org/10.5281/zenodo.24305675}.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24307560
Primary Topic
Random Matrices and Applications
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Obstructions to a conjectured conditioning bound for random tensors

Dingcheng Ji
2 links
Zenodo (CERN European Organization for Nuclear Research)
Random Matrices and Applications
preprint

Obstructions to a conjectured conditioning bound for random tensors

Dingcheng Ji
preprint en

Abstract

Vershynin conjectured that a lower bound for the smallest singular value of a matrix of independent simple random tensors extends to tensor orders $d=O(n)$ under independent subgaussian coordinate assumptions. We disprove this extension using two fixed, bounded, centered, variance-one coordinate distributions. A three-point distribution produces zero columns with probability tending to one at order $d\asymp n/\log n$. An absolutely continuous distribution gives full column rank almost surely at every finite size and an unnormalized smallest singular value at most $e^{-n^2/2}$ with probability tending to one along an explicit sequence $d=o(n)$. Both examples apply when the number of columns is a fixed positive fraction of the ambient dimension, as well as when this fraction tends to one. For the absolutely continuous example, all inverse moments of the smallest singular value diverge at each finite size. We also determine the exact zero-column probability, its Poisson transition, and the largest mass at zero under a prescribed subgaussian norm bound. The proofs are elementary and self-contained. The accompanying Lean formalization is available at \url{https://doi.org/10.5281/zenodo.24305675}.

Zenodo (CERN European Organization for Nuclear Research)
Jiangnan University (CN)
Random Matrices and Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon