New paper reveals fundamental limits of causal digital twins in feedback systems

Faraz Dadgostari and Neda Nazemi (arXiv, July 2026) tackle a critical gap in digital twin theory: how to validate and transport causal digital twins in systems with feedback loops (equilibrium causal games). Traditional digital twins assume a static environment, but when mechanisms change—such as in economic markets or power grids—a twin developed in one domain may fail. The authors formalize conditions under which a twin's predictions can be confirmed via experimental data and then reused in a new target domain. They introduce cyclic selection diagrams to represent equilibrium selection, and propose hybrid models that combine invariant source mechanisms with target-specific information.

The paper’s standout contribution is an impossibility result: they construct systems that agree with every experimental distribution in a finite design but disagree on the true counterfactual of interest. This proves that validation without structural assumptions is impossible. For linear models, they derive exactly which intervention requirements are needed based on which mechanisms changed, the observation model, and graph support. When point identification fails, they characterize the remaining range of query values. The work also provides practical statistical tests for verifying reconstructed means and covariances, making it directly applicable to real-world feedback systems.