The Rayleigh quotient #
The Rayleigh quotient of a self-adjoint operator T on an inner product space E is the function
fun x ↦ ⟪T x, x⟫ / ‖x‖ ^ 2.
The main results of this file are IsSelfAdjoint.hasEigenvector_of_isMaxOn and
IsSelfAdjoint.hasEigenvector_of_isMinOn, which state that if E is complete, and if the
Rayleigh quotient attains its global maximum/minimum over some sphere at the point x₀, then x₀
is an eigenvector of T, and the iSup/iInf of fun x ↦ ⟪T x, x⟫ / ‖x‖ ^ 2 is the corresponding
eigenvalue.
The corollaries LinearMap.IsSymmetric.hasEigenvalue_iSup_of_finiteDimensional and
LinearMap.IsSymmetric.hasEigenvalue_iSup_of_finiteDimensional state that if E is
finite-dimensional and nontrivial, then T has some (nonzero) eigenvectors with eigenvalue the
iSup/iInf of fun x ↦ ⟪T x, x⟫ / ‖x‖ ^ 2.
TODO #
A slightly more elaborate corollary is that if E is complete and T is a compact operator, then
T has some (nonzero) eigenvector with eigenvalue either ⨆ x, ⟪T x, x⟫ / ‖x‖ ^ 2 or
⨅ x, ⟪T x, x⟫ / ‖x‖ ^ 2 (not necessarily both).
The Rayleigh quotient of a continuous linear map T (over ℝ or ℂ) at a vector x is
the quantity re ⟪T x, x⟫ / ‖x‖ ^ 2.
Equations
- T.rayleighQuotient x = T.reApplyInnerSelf x / ‖x‖ ^ 2
Instances For
For a self-adjoint operator T, a local extremum of the Rayleigh quotient of T on a sphere
centred at the origin is an eigenvector of T.
For a self-adjoint operator T, a maximum of the Rayleigh quotient of T on a sphere centred
at the origin is an eigenvector of T, with eigenvalue the global supremum of the Rayleigh
quotient.
For a self-adjoint operator T, a minimum of the Rayleigh quotient of T on a sphere centred
at the origin is an eigenvector of T, with eigenvalue the global infimum of the Rayleigh
quotient.
The supremum of the Rayleigh quotient of a symmetric operator T on a nontrivial
finite-dimensional vector space is an eigenvalue for that operator.
The infimum of the Rayleigh quotient of a symmetric operator T on a nontrivial
finite-dimensional vector space is an eigenvalue for that operator.