Quantum correlations, those exhibited between two or more quantum systems and that cannot be explained in classical terms, are at the core of quantum theory and its applications for information processing.
Previous research has shown that the causal modeling framework offers a powerful set of techniques to study and analyze nonclassicality, and to extend this notion to general networks with independent sources of correlations and communication between parties.
Interventions are an essential tool in causal inference, enabling us to determine the causal relationships between variables in a given process.
Unlike passive observations, interventions involve locally changing the underlying causal structure of an experiment, by erasing all external influences that a given variable might have and putting it under the exclusive control of an observer.
Interestingly, through interventions, we can demonstrate the quantum behavior of a system that may appear classical at the observational level.
One paradigmatic example is the instrumental causal structure, where violations of causal bounds that rely on a specific measure of causal influence, called the average causal effect (ACE), can be violated even when no violation is possible with observational data.
This suggests a new approach that takes into account all available interventional and observational data in a given experiment, and can be applied to arbitrary scenarios.
This approach extends the notion of classical correlations to include interventional data, defining a set that is now bounded by hybrid (observational and interventional) inequalities, which subsumes all Bell-like and causal bounds, and provides a general method to better detect and characterize nonclassical behavior.