Example 3.3 — a locally controllable bilinear system for which x ↦ P̃(x)⁻¹x is not a gradient near 0, so (3.8) fails
ProvedBilinearGramian.Integrability.example_3_3Consider the single-input bilinear system of Example 3.3,
with a real parameter . For every let be the solution of the Lyapunov equation (3.9),
Then:
- the system is locally controllable, i.e. the pair is controllable;
- for every there is no function such that
In words: the vector field is not integrable on any punctured neighbourhood of the origin. In particular the formula of (3.8), claimed in the literature for the controllability energy functional of a locally controllable bilinear system "for with sufficiently small", cannot hold for this system.
Formalization Note The statement refutes every candidate potential , which is exactly "not integrable", and so does not need the definition of . is only required to be differentiable, with the given gradient, at the points of the punctured ball. is added: at the Gramian is a constant symmetric matrix and the field is a gradient, whereas the paper calls "irrelevant for the computation". is any function satisfying (3.9) at every point; since is stable there is exactly one. The plane is EuclideanSpace ℝ (Fin 2) so that the gradient is the Euclidean one. (Pt x)⁻¹ is Mathlib's matrix inverse, which is at singular matrices; is invertible near the origin, so this value does not enter. The analogous claim for the observability Gramian , which the paper says can be argued similarly, is not part of this statement.
import Mathlib import Definitions.Def_BilinearGramian_Integrability_LinGramian import Definitions.Def_BilinearGramian_Integrability_Example33 open Matrix
namespace BilinearGramian.Integrability
/-- Example 3.3 (Benner–Damm 2011, pp. 697–698). For the single-input bilinear system with
`A = diag(-1, -2)`, `N = ν [0 1; 1 0]` (`ν ≠ 0`), `b = (1, 1)ᵀ`: the system is locally
controllable (`(A, b)` controllable), and if `P̃(x)` is defined by (3.9) for every `x`, then on no
punctured ball `0 < ‖x‖ < ε` is the field `x ↦ P̃(x)⁻¹ x` the gradient of a real function `E`;
so (3.8), `∇E_c(x) = P̃(x)⁻¹ x` for `x ≠ 0` with `‖x‖` sufficiently small, cannot hold. -/
theorem example_3_3 (ν : ℝ) (hν : ν ≠ 0) (Pt : (Fin 2 → ℝ) → Matrix (Fin 2) (Fin 2) ℝ)
(hPt : ∀ x, IsLinGramian exA (fun _ : Fin 1 => exN ν) exB x (Pt x)) :
IsControllable exA exB ∧
∀ ε > 0, ¬ ∃ E : EuclideanSpace ℝ (Fin 2) → ℝ,
∀ x : EuclideanSpace ℝ (Fin 2), 0 < ‖x‖ → ‖x‖ < ε →
HasGradientAt E (WithLp.toLp 2 ((Pt x.ofLp)⁻¹ *ᵥ x.ofLp)) x := by sorry
end BilinearGramian.Integrability
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.