Christophe Letellier,Irene Sendiña-Nadal,Luis A. Aguirre
Abstract:A faithful description of the state of a complex dynamical network would require, in principle, the measurement of all its $d$ variables, an infeasible task for systems with practical limited access and composed of many nodes with high dimensional dynamics. However, even if the network dynamics is observable from a reduced set of measured variables, how to reliably identifying such a minimum set of variables providing full observability remains an unsolved problem. From the Jacobian matrix of the governing equations of nonlinear systems, we construct a {\it pruned fluence graph} in which the nodes are the state variables and the links represent {\it only the linear} dynamical interdependences encoded in the Jacobian matrix after ignoring nonlinear relationships. From this graph, we identify the largest connected sub-graphs where there is a path from every node to every other node and there are not outcoming links. In each one of those sub-graphs, at least one node must be measured to correctly monitor the state of the system in a $d$-dimensional reconstructed space. Our procedure is here validated by investigating large-dimensional reaction networks for which the determinant of the observability matrix can be rigorously computed.
What problem does this paper attempt to address?
This paper aims to solve the observability problem of complex dynamic networks. In particular, in high - dimensional systems, when it is impossible to measure all variables due to practical limitations, how to reliably identify the minimum set of variables that can provide complete observability. Specifically, the paper proposes a method based on non - linear graph theory to identify these key variables by constructing a "pruned fluence graph".
### Main contributions of the paper:
1. **Proposing a method**: The paper proposes a new method. By analyzing the Jacobian matrix of the system, a pruned fluence graph that only contains linear dynamic dependencies is constructed. This method can more accurately reflect the non - linear characteristics in the system, thereby improving the observability assessment of the system state.
2. **Verifying the method**: The author verifies the effectiveness of this method through multiple high - dimensional reaction network models (such as the Rössler system, the 5D rational system, the 9D Rayleigh - Bénard convection model and the 13D DNA replication model). The selection of these models covers typical complex systems in biology and physics.
3. **Comparing with existing methods**: The paper also compares the existing graph - theory - based methods (such as the inference graph method proposed by Liu et al.), points out the shortcomings of these methods in dealing with non - linear systems, and shows the advantages of the new method through specific examples.
### Specific problem - solving:
- **Definition of observability**: The paper first defines the observability of the system, that is, whether the initial state of the system can be uniquely determined by measuring a part of the variables of the system. This is usually judged by calculating the rank of the observability matrix.
- **Construction of the pruned fluence graph**: The paper proposes to construct a pruned fluence graph starting from the Jacobian matrix of the system. Nodes in the graph represent state variables, and edges represent linear dynamic dependencies between variables. By removing non - linear relationships and retaining linear relationships, the observability of the system can be more accurately reflected.
- **Root Strongly Connected Component (Root SCC)**: In the pruned fluence graph, the root strongly connected components (strongly connected sub - graphs without outgoing edges) are identified. The nodes in these components are the key variables that need to be measured. By measuring these variables, the complete observability of the system can be ensured.
### Conclusion:
Through theoretical analysis and experimental verification, the paper proves the effectiveness and superiority of the proposed method in dealing with high - dimensional non - linear systems. Compared with the existing graph - theory - based methods, the new method can more accurately identify the minimum set of variables that provide complete observability, which is of great significance for the monitoring and control of complex systems.