AI summaryⓘ
The authors introduce SVAR-FM, a method for discovering cause-and-effect in time series data by using a physics simulator like a real experiment where a variable is fixed to see its effect. This approach cuts off hidden influences and produces data that shows direct causal impacts, which they learn using Conditional Flow Matching. They prove the method can fully identify causal relationships under certain conditions and show that if the simulator is too inaccurate, it can even reverse the estimated causal direction. Their experiments across different fields confirm SVAR-FM finds the right causal effects when other methods fail, and a laser physics case validates their theory by showing how changing simulator accuracy flips the causal sign.
Structural VARFlow MatchingCausal DiscoveryPearl's do operatorInterventional DataSimulator FidelityMonte Carlo MethodsConfoundingError BoundUltrafast Laser Physics
Abstract
We propose SVAR-FM (Structural VAR with Flow Matching), a framework for time series causal discovery that treats a physics-based simulator as a mechanical realization of Pearl's do operator. Clamping a variable inside the simulator physically severs confounding paths, producing interventional data by construction. Conditional Flow Matching then learns the nonlinear interventional conditionals. Theoretically, we prove that the full structural VAR becomes identifiable under a coverage condition on the simulator-clampable variables, and derive an end-to-end error bound that decomposes into Monte Carlo, simulator fidelity, and Flow Matching terms. A sign-flip corollary predicts that when simulator accuracy falls below a threshold, the estimated causal effect reverses sign. Empirically, a benchmark across four scientific domains confirms that SVAR-FM recovers the correct causal sign where observational methods produce sign-reversed estimates due to confounding. A case study in ultrafast laser physics verifies the sign-flip prediction by physically varying the accuracy level of a first-principles quantum solver: the low-accuracy setting reverses the causal sign, while the high-accuracy setting recovers the correct direction (R-squared = 0.983, zero bias).