New method tracks hidden patterns in complex systems

Causal fingerprints reveal how hidden mechanisms explain individual events in partially-observed systems.

Researchers introduced causal fingerprints to study partially-observed dynamical systems. These fingerprints track how information about a hidden state is revealed through evolution.

The method compares a filtering posterior with a smoothing posterior. A filtering posterior is based on observations up to a given time. A smoothing posterior is obtained after adding an expanding window of future observations.

The researchers quantify retrospective information gain using Kullback-Leibler divergence. This tracks how the fingerprint evolves as the future window of data expands.

Causal fingerprints reveal the strength and temporal reach of hidden influence. This influence can vary across different events and their stages.

The fingerprints show how competing mechanisms explain the same observations. This works even when selected statistics agree.

In nonlinear examples, fingerprints reveal signatures for extreme bursts and regime switching. They can distinguish between multiplicative instability and additive forcing.

Key facts

  • Causal fingerprints track how information about a hidden state is revealed.
  • The method uses Kullback-Leibler divergence to quantify retrospective information gain.
  • Fingerprints can distinguish between different mechanisms like deterministic nonlinear feedback and state-dependent multiplicative noise.

All sections

Edition