Physical Simulators as Do-Operators: Causal Discovery under Latent Confounders for AI-for-Science
For AI-for-Science applications like molecular design and materials science, CFM-SD addresses the critical problem of causal discovery with latent confounders and expensive real interventions.
CFM-SD uses physical simulators as do-operators to perform causal discovery under latent confounders, achieving F1=0.800 vs. 0.127–0.562 for baselines on synthetic data and 57–58% bias reduction in real scientific tasks.
Existing interventional causal discovery methods -- IGSP, DCDI, ENCO -- assume causal sufficiency (no latent confounders) and rely on virtual interventions in synthetic simulators. In AI-for-Science settings such as molecular design and materials science, latent confounders are ubiquitous and real interventions (e.g., physics-based simulations) require hours to days per data point. We propose CFM-SD (Causal Flow Matching with Simulation Data), which uses first-principles physical simulators as do-operators in Pearl's interventional calculus to simultaneously handle latent confounders and real interventional data. Theoretically, $d$-variable causal structure is identifiable with $O(d)$ single-variable interventions -- the minimum under physical realizability constraints. In Intrinsic Evaluation on synthetic data ($γ=0.2$--$0.8$), CFM-SD achieves average F1$=0.800$ vs. F1$=0.127$--$0.562$ for all baselines. In Extrinsic Evaluation on real scientific data, CFM-SD achieves 57--58\% bias reduction in molecular toxicity prediction and battery electrolyte optimization, demonstrating practical value beyond synthetic benchmarks.