The authors use a very sophisticated tool to solve a very old problem of model order reduction. They compare the performance of their result with several benchmark datasets of high academic interest. As referenced in the paper, researchers have developed techniques for the selection of interpolation points and directions, that is, (s_i, d_i), i = 1, ..., m, in a specific way [1,2].

In the present paper, the authors use an adaptive method based on the global Arnoldi method to obtain the parameters (s_i, d_i). In the course of doing this, the authors develop a new tangential Krylov-based method. In section 4.3, they propose an interesting result using proposition 3 and develop a suitable threshold to make the adaptive tangential interpolation approximation result exact. The adaptive global tangential Arnoldi algorithm (AGTAA) is very nicely summarized. Overall, the contribution is technically very sound and conceptually very enriched.

However, the authors do not mention the basic objective of reducing the model order. Do they want to achieve a low-cost design of the controller by reducing the model order? If so, there should be some cost-effective design study and comparison with the benchmark results. Time responses of the reduced order model and the corresponding steady-state error, transient overshoot and undershoot, and intermediate oscillations of the time response are not included in the paper. All of these results should be compared with the benchmark datasets.