Structural Controllability in Systems Theory

• Structural controllability is a concept in systems theory that determines if nearly all admissible parameterizations allow steering from any initial state to a desired state using suitable inputs.
• It relies on graph-based notions such as accessibility, no dilation, and cactus structures to assess control capabilities under uncertainty.
• The concept underpins robust analysis in networked, ensemble, and nonlinear systems, with scalable verification methods despite NP-hard optimal input placement challenges.

Structural controllability is a foundational concept in systems and control theory, referring to the property whereby almost all parameterizations of a system that respect a specified sparsity pattern are controllable—i.e., it is possible to steer the state from an arbitrary initial to a desired final state using admissible inputs. Unlike classic controllability, which depends on exact parameter values, structural controllability is determined solely by the system’s interconnection topology or zero–nonzero pattern. This abstraction yields robust, generic criteria applicable to large-scale, uncertain, or parameter-dependent systems, and underpins analysis and design in network control, multi-agent systems, and complex networks.

1. Fundamental Principles and Classical Characterization

The classical framework considers linear time-invariant (LTI) systems: $\dot x = A x + B u,\quad x\in\R^n,\ u\in\R^m$ with unknown but independent free parameters in nonzero entries of (A, B). Structural controllability asks whether, for almost all numerical choices of these parameters, the system is controllable in the sense of Kalman.

Lin's theorem provides necessary and sufficient graph-theoretic conditions:

• Accessibility (no inaccessible node): Every state node is reachable from at least one input node via a directed path.
• No dilation: For every subset of state-nodes $S$, the set of their in-neighbors (including inputs) $\Gamma(S)$ satisfies $|\Gamma(S)| \geq |S|$.

Alternatively, the existence of a so-called “cactus” (disjoint union of path and cycle subgraphs rooted at inputs) spanning the state-node set implies structural controllability (Pickard, 2023).

For parameterized (possibly dependent) matrices satisfying the binary assumption, structural controllability equivalently requires:

• Existence of a spanning forest rooted at input vertices.
• Existence of an “unbalanced” similarity class of multi-colored subgraphs (each color corresponding to one parameter) (Liu et al., 2017).

Structurally controllable systems are generic: the set of uncontrollable parameterizations forms an algebraic variety of measure zero.

2. Graph-Theoretic, Hypergraph, and Extended Criteria

Graph-theoretic frameworks extend to nonlinear and high-order systems:

• For odd-homogeneous polynomial systems (“polysystems”), the support is captured by a directed hypergraph, with nodes for states and inputs, and hyperedges for monomials. Structural controllability holds if and only if the hypergraph has no hyperedge dilation (i.e., each subset $S$ of states is targeted by at least $|S|$ hyperedges) and no inaccessible state node (every node is reachable via a hyper-walk from the inputs) (Pickard, 2023, Pickard et al., 20 Mar 2026).
• For ensemble systems (continuum families indexed by a parameter), structural controllability additionally requires the existence of a Hamiltonian decomposition (disjoint cycle cover) in the state-induced subgraph, beyond accessibility (Chen, 2020).
• For drifted bilinear systems on Lie groups, the structural controllability reduces to connectivity (plus colored-cycle conditions in SU(n)) in graphs determined by the drift and control zero patterns (Wang et al., 2021).

In switched, time-varying, or structured network settings, the theory extends with colored union graphs, matching, or auxiliary structures—e.g., controllability of switched linear systems is certified by the existence of $n$ S-disjoint edges in the colored union graph of subsystem patterns (Liu et al., 2011, Yin et al., 6 Dec 2025).

3. Minimal Patterns, Monotonicity, and Indices

The property of structural controllability is monotone: adding edges (relaxing constraints) preserves or improves controllability. Minimal structurally controllable patterns (i.e., those that lose the property upon deletion of any edge) are characterized by single-input arborescence skeletons with all strongly connected components as cycles (Chen, 2020).

The structural controllability index (SCOI) quantifies the minimal number of steps (or horizon length) to achieve generic controllability. For single-input systems, the index equals the size of the largest “cactus-structure family” in the system digraph, but classical heuristics or forest/cactus-based formulas can overestimate it, especially for systems with nontrivial symmetry or self-loops. Polynomial-time tight lower bounds are obtained via min-cost max-flow computations on associated dynamic graphs (Zhang et al., 19 Feb 2025).

4. Strong Structural Controllability and Robustness

Strong structural controllability (SSC) demands controllability for all nonzero parameter assignments (not just almost all). For LTI networks, this coincides with the input set being a zero-forcing set: every node is colored black by the zero-forcing rule, guaranteeing coverage for any parameter realization (Mousavi et al., 2019, Jia et al., 2020). In signed, colored, or symmetry-constrained networks, refined (e.g., signed zero-forcing or colored zero-forcing) combinatorial conditions apply (Mousavi et al., 2019, Jia et al., 2018). Critical edge-set, network-of-networks construction, and composition rules are developed to analyze SSC under perturbations and modular synthesis (Mousavi et al., 2019).

SSC is strictly stronger than standard structural controllability, and specialized algorithms exist for verification in large networks via dimension-reduction and pattern-transformation techniques (Jia et al., 2020). Notably, the property remains stable (“robust”) under addition of up to a computable critical number of edges, and one can explicitly characterize permissible structural perturbations before loss of SSC (Mousavi et al., 2019).

5. Applications Across Networked, Ensemble, and Nonlinear Systems

Networked Systems

In networks of diffusive, relative-coupling, or consensus agents, structural controllability reduces to global input-reachability of the network or leader-follower connectivity (i.e., each state is reachable from at least one input/leader), subject to subsystem controllability/observability and absence of fixed modes (Zhang et al., 2020, Zhang et al., 2019, Kazemi et al., 2018).

Ensemble and Switched Systems

For ensemble networks (continuum or multicopy systems), structural controllability entails Hamiltonian decomposition, accessibility, and—when switching is allowed—matching conditions in time-unfolded bipartite graphs or flow networks. Switching relaxes constraints and can strictly enlarge the class of achievable patterns compared to the LTI ensemble case (Yin et al., 6 Dec 2025, Chen, 2020).

Nonlinear, Hypergraph, and Influence Diagram Contexts

For nonlinear and hypergraph-modeled systems (e.g., biochemical networks, ecological systems), combinatorial accessibility and hyperedge dilation criteria provide efficient, generic certificates—even for large systems, via matching-augmented greedy algorithms (Pickard et al., 20 Mar 2026, Pickard, 2023). Influence diagrams generalize the framework beyond linear algebra to arbitrary generically “nimble” deterministic maps in causal graphs, using flow and path-cover principles (Chan et al., 2013).

Robustness and Perturbation-Tolerance

A further generalization, perturbation-tolerant structural controllability (PTSC), ensures that for almost all parameterizations, any admissible numerical perturbation (with specified sparsity) preserves controllability. Polynomial-time verification is achieved via rank and matching-based tests (Zhang et al., 2021).

6. Algorithmic and Complexity Aspects

Efficient algorithms for structural controllability typically involve:

• Construction and traversal of support graphs or hypergraphs to test accessibility.
• Maximum matching or flow computations in bipartite/expanded graphs to detect (hyper)dilations.
• Symbolic determinant expansion and randomized evaluation for detection of unbalanced similarity classes (Liu et al., 2017).
• Specialized color-change and edge manipulation rules for SSC in colored/signed graphs (Jia et al., 2018). Recent advances enable dimension-independent analysis for large structured networks (reducing high-order nodes to 1–2D patterns) and scalable selection of minimal driver sets in hypergraphs, with tight complexity guarantees (Jia et al., 2020, Pickard et al., 20 Mar 2026).

Table: Structural Controllability—Criteria and Algorithms

Setting/Model	Structural Controllability Criterion	Key Algorithmic Test
LTI systems (Lin, classic)	Accessibility & no dilation	BFS + matching (polynomial)
Hypergraph/polynomial systems	No hyperedge dilation & accessibility	Hyperwalk + bipartite matching
Signed/colored/SSC networks	Zero-forcing (signed/colored) set covers all	Forcing rule + color-change
Ensemble systems	Accessibility & Hamiltonian decomposition	Cycle-cover (matching)
Switched/sparse ensemble	Matching in time-unfolded graph (k-switches)	Max-flow (polynomial)

7. Limitations, Open Problems, and Future Directions

Complete, purely graph-theoretic characterizations of indices (SCOI) for arbitrary structured systems remain open, with only upper and lower bounds currently established (Zhang et al., 19 Feb 2025). Strong and robust variants, particularly for nonlinear or bilinear systems over Lie groups, demand further investigation into minimality and necessity of connectivity-plus-cycle criteria (Wang et al., 2021, Chen, 2020). The computational complexity of optimal input placement under structural constraints (minimal controllability) is generally NP-hard for large-scale instances, but scalable heuristics with provable performance are available in many cases (Zhang et al., 2018, Pickard et al., 20 Mar 2026).

Rapid growth in networked, multi-agent, and nonlinear systems continues to drive extensions of structural controllability theory, including integration with perturbation analysis, higher-order and hybrid settings, and robust design tools for large, uncertain, or evolving system topologies.