Hankel Singular Value¶
A nonnegative state-importance measure for a stable linear system, equal to the square roots of eigenvalues of the controllability–observability Gramian product and used in balanced reduction.
Core Idea¶
A Hankel singular value ranks a stable linear system's state directions by joint controllability and observability. It is the square root of an eigenvalue of the Gramian product and becomes the common balanced-Gramian diagonal used to choose reduced-order states. Mathematically the values are the nonnegative square roots of eigenvalues of the product of controllability and observability Gramians, equivalently singular values of the Hankel operator under standard assumptions. Mathematically the values are the nonnegative square roots of eigenvalues of the product of controllability and observability Gramians, equivalently singular values of the Hankel operator under standard assumptions.
Scope of Application¶
The measure applies to stable linear-system analysis and balanced model reduction with defined input and output channels. Use it in stable linear input–output models and balanced reduction with channels, norms, numerical method, and truncation validation explicit.
- Balanced truncation. Ranks state directions for reduction.
- Control design. Preserves dominant input–output dynamics.
- Large-scale simulation. Constructs lower-order surrogate models.
- System identification. Compares effective dynamic order.
- Filter design. Characterizes realizable input–output importance.
Clarity¶
The measure separates large state amplitude from input–output importance. A direction matters for reduction only when it is both excitable and visible, and its rank belongs to the declared system realization, channels, and stability assumptions. The closest near miss sets the boundary: Ordinary singular values of the state or transfer matrix are the closest near miss: they measure a chosen matrix, not joint past-input to future-output influence through system dynamics.
Manages Complexity¶
High-order state models contain many coupled coordinates whose apparent importance changes under basis transformation. The Hankel spectrum compresses joint reachability and visibility into an ordered invariant sequence, revealing effective order and truncation candidates. The central coordinate invariance–physical interpretability tradeoff is this: Balanced directions are system-invariant in importance but can mix named physical states. A second aggressive reduction–error preservation tension matters because Discarding more small values improves efficiency while accumulating approximation error.
Abstract Reasoning¶
Use three linked moves: state the linear realization, stability assumptions, and input–output channels; compute or approximate controllability and observability Gramians consistently; derive the nonnegative joint spectrum and check numerical conditioning. As a collapse test, the case exits when the system assumptions fail, Gramians do not exist in the stated sense, or values come from an unmatched coordinate-dependent quantity. A fourth check is to balance the realization and inspect gaps in the Hankel singular values.
Knowledge Transfer¶
The measure transfers literally among stable linear systems after channels and norms are declared. ‘Low-energy state’ outside this formalism is analogy; neither physical-energy content nor nonlinear importance follows automatically from a small Hankel singular value. No canonical parent prime is currently asserted; broader structural comparisons remain related-prime analogies until separately adjudicated in the DAG. Controllability and observability contribute symmetrically to the ranking.
Neighborhood in Abstraction Space¶
Hankel Singular Value sits in a sparse region of the domain-specific corpus (63rd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Thermodynamics & Dissipative Systems (19 abstractions)
Nearest neighbors
- Circle criterion — 0.85
- Closed Linear Operator — 0.85
- Control-Lyapunov function — 0.85
- Floquet Theory — 0.84
- Mean-field theory — 0.84
Computed from structural-signature embeddings · 2026-10-08