1: \begin{abstract}                  

2: 	In this paper, we study distributed methods for solving a Sylvester equation in the form of  $ \A\X+\X\B=\CC $ for matrices $ \A, \B, \CC \in \R^{n\times n} $  with $ \X $ being the unknown variable.

3: 	The entries of $ \A, \B $ and $ \CC $ (called data) are partitioned into a number of pieces (or sometimes we permit these pieces to overlap). 

4: 	Then a network with a given structure is assigned, whose number of nodes is consistent with the partition. Each node has access to the corresponding set of data and holds a dynamic state. Nodes share their states among their neighbors defined from the network structure, and we aim to design flows that can asymptotically converge to a solution of this equation. 

5: 	The decentralized data partitions may be resulted directly from networks consisting of physically isolated subsystems, or indirectly from artificial and strategic design for processing large data sets.

6: 	 Natural partial row/column partitions, full row/column partitions and clustering block partitions of the data $ \A, \B $ and $ \CC $ are assisted by the use of the vectorized matrix equation. We show that the existing ``consensus $ + $ projection'' flow and the ``local conservation $ + $ global consensus'' flow for distributed linear algebraic equations can be used to drive distributed flows that solve this kind of equations. A ``consensus $ + $ projection $ + $ symmetrization'' flow is also developed for equations with symmetry constraints on the solution matrices. We  reveal some fundamental convergence rate limitations for such flows regardless of the choices of node interaction strengths and network structures. 

7: 	 For a special case with $ \B=\A^T $, where the equation  mentioned is reduced to a classical Lyapunov equation, we demonstrate that by exploiting the symmetry of data, we can obtain  flows with lower complexity for certain partitions.

8: \end{abstract}

9: