Using covariance of node states to design early warning signals for network dynamics

Area of Science:

• Complex systems science
• Network dynamics
• Statistical modeling

Background:

• Real-world systems can undergo abrupt changes called regime shifts.
• Early warning signals (EWS) aim to predict these critical transitions.
• Network structures are crucial for understanding regime shifts in many systems.

Purpose of the Study:

• To investigate the efficacy of sample covariance between node pairs as an EWS for networks.
• To compare the performance of covariance-based EWS with variance-based EWS.
• To identify optimal strategies for constructing EWS in complex networks.

Main Methods:

• Analytical calculations were performed on four-node networks.
• Numerical simulations were conducted on larger, complex networks.
• The performance of sample covariance was compared against sample variance for EWS construction.

Main Results:

• Sample covariance and its combinations over node pairs were found to be inferior to sample variance.
• Variance-based EWS, utilizing combinations over nodes, outperformed covariance-based approaches.
• The study supports using diagonal entries (variance) over off-diagonal entries (covariance) of the covariance matrix for EWS.

Conclusions:

• Sample variance is a more effective EWS component than sample covariance for predicting regime shifts in networks.
• The findings reinforce the utility of variance-based metrics in network EWS.
• Future EWS development should prioritize variance over covariance for improved early detection of critical transitions.