Schur multipliers, vanishing minors and Gaussian conditional independence

Conditional independence in a Gaussian vector is the vanishing of almost-principal minors of its covariance matrix $Σ$. A sensitivity analysis replaces $Σ$ by the Schur product $Δ\circΣ$, where the symmetric multiplier $Δ$ differs from one exactly at the changed positions. Such a change generally destroys some conditional independences, and we determine which minors stay zero. A vanishing minor is preserved for every matrix exactly when the corresponding block of $Δ$ has rank one. When variances or covariances change, each almost-principal minor of $Δ\circΣ$ is a polynomial in the amounts added, with minors of $Σ$ as coefficients, and a conditional independence survives every such change exactly when they vanish. In an undirected graphical model, changing variances joins the neighbours of the perturbed vertices, while changing covariances leaves a structure read from the graph by a separation rule, possibly not graphical. Both extend to ancestral graphs, directed acyclic graphs included, and identify the entries that can vary freely. For the other entries, blocks of rank one along a junction tree characterize the multipliers that preserve a decomposable model. A sensitivity analysis can therefore rest on the conditional independence structure after the change, as illustrated on examination marks of 88 students.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-16
DOI
https://doi.org/10.48550/arxiv.2611.20982
Primary Topic
Bayesian Modeling and Causal Inference
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Schur multipliers, vanishing minors and Gaussian conditional independence

Manuele Leonelli
arXiv (Cornell University)
Bayesian Modeling and Causal Inference
preprint

Schur multipliers, vanishing minors and Gaussian conditional independence

Manuele Leonelli
preprint en

Abstract

Conditional independence in a Gaussian vector is the vanishing of almost-principal minors of its covariance matrix $Σ$. A sensitivity analysis replaces $Σ$ by the Schur product $Δ\circΣ$, where the symmetric multiplier $Δ$ differs from one exactly at the changed positions. Such a change generally destroys some conditional independences, and we determine which minors stay zero. A vanishing minor is preserved for every matrix exactly when the corresponding block of $Δ$ has rank one. When variances or covariances change, each almost-principal minor of $Δ\circΣ$ is a polynomial in the amounts added, with minors of $Σ$ as coefficients, and a conditional independence survives every such change exactly when they vanish. In an undirected graphical model, changing variances joins the neighbours of the perturbed vertices, while changing covariances leaves a structure read from the graph by a separation rule, possibly not graphical. Both extend to ancestral graphs, directed acyclic graphs included, and identify the entries that can vary freely. For the other entries, blocks of rank one along a junction tree characterize the multipliers that preserve a decomposable model. A sensitivity analysis can therefore rest on the conditional independence structure after the change, as illustrated on examination marks of 88 students.

arXiv (Cornell University)
Bayesian Modeling and Causal Inference
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