Dynamics-preserving graph sparsification

Many dynamical systems unfold on networks of nodes coupled by edge interactions. This work is motivated by the following question: can we remove most edges from a networked dynamical system without significantly altering its behavior? For a given network, our goal is to find a sparser network that approximately preserves the dynamics on the nodes. We first study linear diffusion, where node values evolve according to the graph Laplacian. The dominant network sparsification method, effective resistance sampling, preserves all Laplacian eigenvalues within a uniform multiplicative bound. We show that with a limited edge budget, this is dynamically inefficient: diffusion is dominated by the slowest-decaying modes, so enforcing a uniform guarantee spends edges on fast modes that barely affect long-term dynamics. Based on this observation, we derive Diffusion Spectral Error (DSE), which evaluates in closed form the typical error in trajectories between the original and sparsified graphs over random initial conditions. Minimizing this error, which naturally prioritizes slow modes, yields sparse networks explicitly designed to preserve dynamics. At the same sparsification levels, DSE sparsification often preserves diffusion dynamics with several-fold lower error than effective resistance sampling, and up to two orders of magnitude lower on strongly modular networks. Although derived in a linear context, we empirically find that DSE sparsification gives similar improvements in the Kuramoto model (a nonlinear system of coupled oscillators), where DSE sparsification approximately preserves the synchronization threshold and, at strong coupling, order-parameter trajectories.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-17
DOI
https://doi.org/10.48550/arxiv.2611.23176
Primary Topic
Complex Network Analysis Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Dynamics-preserving graph sparsification

Cristopher Moore, Yuanzhao Zhang, Maximilian Jerdee, Rosemary Braun et al.
arXiv (Cornell University)
Complex Network Analysis Techniques
preprint

Dynamics-preserving graph sparsification

Cristopher Moore, Yuanzhao Zhang, Maximilian Jerdee, Rosemary Braun, Emmie Fitz-Gibbon
preprint en

Abstract

Many dynamical systems unfold on networks of nodes coupled by edge interactions. This work is motivated by the following question: can we remove most edges from a networked dynamical system without significantly altering its behavior? For a given network, our goal is to find a sparser network that approximately preserves the dynamics on the nodes. We first study linear diffusion, where node values evolve according to the graph Laplacian. The dominant network sparsification method, effective resistance sampling, preserves all Laplacian eigenvalues within a uniform multiplicative bound. We show that with a limited edge budget, this is dynamically inefficient: diffusion is dominated by the slowest-decaying modes, so enforcing a uniform guarantee spends edges on fast modes that barely affect long-term dynamics. Based on this observation, we derive Diffusion Spectral Error (DSE), which evaluates in closed form the typical error in trajectories between the original and sparsified graphs over random initial conditions. Minimizing this error, which naturally prioritizes slow modes, yields sparse networks explicitly designed to preserve dynamics. At the same sparsification levels, DSE sparsification often preserves diffusion dynamics with several-fold lower error than effective resistance sampling, and up to two orders of magnitude lower on strongly modular networks. Although derived in a linear context, we empirically find that DSE sparsification gives similar improvements in the Kuramoto model (a nonlinear system of coupled oscillators), where DSE sparsification approximately preserves the synchronization threshold and, at strong coupling, order-parameter trajectories.

arXiv (Cornell University)
Complex Network Analysis Techniques
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