Synchronization preserving model reduction of multi-agent network systems by eigenvalue assignments

In this paper, structure preserving model reduction problem for multi-agent network systems consisting of diffusively coupled agents is investigated. A new model reduction method based on eigenvalue assignment is derived. Particularly, the spectrum of the reduced Laplacian matrix is selected as a subset of the spectrum of the original Laplacian matrix. The resulting reduced-order model retains the network protocol of diffusive couplings, and thus the synchronization property is preserved. Moreover, a concise expression for the upper-bound of the 2 approximation error is presented in the setting of a leader-follower network, and it provides a guideline to select the eigenvalues of the reduced Laplacian matrix. The effectiveness of the proposed method is finally illustrated via the application to a spacecraft network, with a comparison of performances with the graph clustering method in [1] and balanced truncation approach in [2].