1: \begin{abstract}

2: Topology inference for network systems (NSs) plays a crucial role in many areas. 

3: This paper advocates a causality-based method based on noisy observations from a single trajectory of a NS, which is represented by the state-space model with general directed topology. 

4: Specifically, we first prove its close relationships with the ideal Granger estimator for multiple trajectories and the traditional ordinary least squares (OLS) estimator for a single trajectory. 

5: Along with this line, we analyze the non-asymptotic inference performance of the proposed method by taking the OLS estimator as a reference, covering both asymptotically and marginally stable systems. 

6: The derived convergence rates and accuracy results suggest the proposed method has better performance in addressing potentially correlated observations and achieves zero inference error asymptotically.  

7: Besides, an online/recursive version of our method is established for efficient computation or time-varying cases. 

8: Extensions on NSs with nonlinear dynamics are also discussed. 

9: Comprehensive tests corroborate the theoretical findings and comparisons with other algorithms highlight the superiority of the proposed method. 

10: \end{abstract}

11: