Causal discovery from nonstationary subsample time series.

Causal discovery from nonstationary subsample time series is a fundamental yet challenging problem in many scientific domains. In such data, observations are often recorded at a coarse time scale (subsampling). At the same time, the underlying data-generating mechanisms may change over time (nonstationarity), making standard methods that assume a single, fixed causal model unreliable. To address these challenges, we propose a local-to-global causal graph reconstruction framework for nonstationary subsample time series. Specifically, we first identify key latent variables via a pure-child-anchored strategy to recover boundary structures. We then infer contemporaneous causal relations through one-sided reduction and redundant-edge elimination. Finally, we reconstruct the full-time causal directed acyclic graph using recursive blocking and coefficient-invariant transfer, enabling consistent integration across nonstationary segments. We further provide theoretical guarantees for the correctness of the proposed method under mild conditions. Extensive experiments on synthetic datasets demonstrate the effectiveness of our method in recovering the underlying causal structure under both subsampling and nonstationarity. Moreover, an application to the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset reveals interpretable, stage-dependent causal pathways from mild cognitive impairment (MCI) to Alzheimer's disease (AD), highlighting the practical utility of our approach.