Linear Scaling Causal Discovery from High-Dimensional Time Series by Dynamical Community Detection.

Area of Science:

• Complex Systems
• Causal Inference
• Network Science

Background:

• Inferring causal links in dynamical systems from observational data is crucial but computationally challenging, especially for high-dimensional systems.
• Existing methods struggle with the computational complexity of analyzing large datasets without direct system manipulation.

Purpose of the Study:

• To develop a computationally efficient framework for constructing causal graphs from high-dimensional time series.
• To address the limitations of current methods in inferring causality in complex systems.

Main Methods:

• Introduced a novel framework based on automatic identification of "dynamical communities" within the system.
• Utilized "information imbalance" optimization to weight variables by their information content.
• Ordered communities based on their autonomy and dependence to build a community causal graph.

Main Results:

• The proposed framework achieves linear scaling with the number of variables, offering significant computational efficiency.
• Demonstrated accurate causal graph construction on both discrete-time and continuous-time dynamical systems.
• Successfully analyzed systems with up to 80 variables.

Conclusions:

• The developed framework provides an efficient and accurate method for causal discovery in high-dimensional time series.
• This approach facilitates a deeper understanding of interdependencies within complex dynamical systems.
• The method has broad applicability in fundamental and applied scientific research.