Fractal Conditional Correlation Dimension Infers Complex Causal Networks.

Area of Science:

• Physical sciences
• Engineering applications
• Complex network analysis

Background:

• Causal inference is increasingly important in physical and engineering fields.
• Observing time series data is key to modeling complex networks.
• Existing methods face challenges in accurately distinguishing direct and indirect causal relationships.

Purpose of the Study:

• To introduce a novel principle, the optimal conditional correlation dimensional geometric information flow (oGeoC), for causal inference.
• To develop algorithms for discovering direct causal links and removing indirect ones in networks.
• To provide a geometric interpretation for understanding causal relations.

Main Methods:

• Development of the oGeoC principle based on geometric interpretations.
• Introduction of two algorithms to identify direct links and filter indirect links using oGeoC.
• Evaluation of algorithms on coupled logistic networks.

Main Results:

• The proposed algorithms accurately identify direct causal links in networks.
• A low false positive rate was observed when sufficient observations were available.
• The oGeoC principle effectively reveals both direct and indirect causal relations.

Conclusions:

• The oGeoC principle and associated algorithms offer a robust solution for causal inference in complex networks.
• The method demonstrates high accuracy and reliability for identifying direct causal links.
• This approach enhances the understanding of network dynamics through time series analysis.