Sorted rectangular SVD with orthonormal completion of zero singular directions
ProvedQuantumLinearAlgebra.sorted_svd_of_finrank_leLet and be finite-dimensional complex inner product spaces, with , and let be linear. There are orthonormal bases of and of , and real numbers , such that
Zero singular values are included; their left vectors are completed to an orthonormal basis of the full codomain. The conclusion also covers zero-dimensional spaces. This is the rectangular singular value decomposition in a form that can be applied directly to maps between projector image subspaces, as needed for Definition 11 of the QSVT paper.
Formalization Note. The index inclusion is Fin.castLE hnm; basis vectors and singular values use zero-based indexing. This supporting lemma states the standard linear-map SVD underlying the paper's projected-unitary instance, with the explicit dimension order shown above.
import Mathlib.Analysis.InnerProductSpace.SingularValues set_option autoImplicit false
namespace QuantumLinearAlgebra
theorem sorted_svd_of_finrank_le {E F : Type*} [NormedAddCommGroup E] [NormedAddCommGroup F]
[InnerProductSpace ℂ E] [InnerProductSpace ℂ F]
[FiniteDimensional ℂ E] [FiniteDimensional ℂ F] (T : E →ₗ[ℂ] F) (hnm : Module.finrank ℂ E ≤ Module.finrank ℂ F) :
∃ (b : OrthonormalBasis (Fin (Module.finrank ℂ E)) ℂ E)
(c : OrthonormalBasis (Fin (Module.finrank ℂ F)) ℂ F)
(s : Fin (Module.finrank ℂ E) → ℝ),
(∀ i, 0 ≤ s i) ∧ Antitone s ∧
∀ i, T (b i) = (s i : ℂ) • c (Fin.castLE hnm i) := by sorry
end QuantumLinearAlgebra