Elementary inequalities for positive operators #
For a positive operator T on a possibly infinite-dimensional Hilbert space, 1 + T is
invertible (ContinuousLinearMap.IsPositive.isUnit_one_add), because
re ⟪(1 + T) x, x⟫ ≥ ‖x‖² bounds it below. This applies in particular to T = X†X, which makes
(1 + X†X)⁻¹ available in formulas such as the orthogonal projection onto a graph subspace.
The quadratic form of a positive endomorphism A controls its off-diagonal entries,
2 re ⟪x, A y⟫ ≤ re ⟪x, A x⟫ + re ⟪y, A y⟫
(LinearMap.IsPositive.two_mul_re_inner_le), and on a finite-dimensional space the diagonal
entries along any orthonormal family sum to at most the trace
(LinearMap.IsPositive.sum_re_inner_le_re_trace). These are the inputs to Ky Fan's maximum
principle for positive operators.
A contraction between real or complex Hilbert spaces satisfies D D† ≤ 1 in the Loewner
order (ContinuousLinearMap.self_comp_adjoint_le_one), so its defect operator 1 - D D†
is positive.
For a positive endomorphism A, 2 re ⟪x, A y⟫ ≤ re ⟪x, A x⟫ + re ⟪y, A y⟫.
For a positive endomorphism A of a finite-dimensional space, the diagonal entries of A along
an orthonormal family have real parts summing to at most the real part of the trace.
On a complete space, 1 + T is invertible for every positive operator T, since
re ⟪(1 + T) x, x⟫ = ‖x‖² + re ⟪T x, x⟫ ≥ ‖x‖². For instance 1 + X†X is invertible, by
ContinuousLinearMap.isPositive_adjoint_comp_self.
A contraction composed with its adjoint is bounded above by the identity in the Loewner order. Equivalently, its defect operator is positive.