Intrinsic dimension and traces of spectral functions
DefinitionTroppMatrixConcentration_ch7_intrinsicFor a finite complex square matrix , define , using the spectral norm. On nonzero positive semidefinite matrices this is intrinsic dimension (Definition 7.1.1). The definition is total and gives . For a real scalar function , define through Mathlib's real continuous functional calculus. For Hermitian matrices it applies to each real eigenvalue. Non-Hermitian inputs retain the functional calculus's zero-default convention. No concentration conclusion is encoded in either definition.
import Definitions.Def_TroppMatrixConcentration_probability
import Definitions.Def_TroppMatrixConcentration_dilation
import Mathlib.Analysis.Convex.Function
open MeasureTheory ProbabilityTheory
open scoped Matrix.Norms.L2Operator ComplexOrder
noncomputable section
namespace TroppMatrixConcentration
def intrinsicDimension {d : Type*} [Fintype d] [DecidableEq d]
(A : Matrix d d ℂ) : ℝ := (Matrix.trace A).re / spectralNorm A
def traceFunction {d : Type*} [Fintype d] [DecidableEq d]
(φ : ℝ → ℝ) (A : Matrix d d ℂ) : ℝ := (Matrix.trace (cfc φ A)).re
end TroppMatrixConcentration
Read-back
What the Lean code literally says, in plain math · gpt-6
intrinsicDimension: For every finite index type equipped with decidable equality and every complex square matrix indexed by , this defines the real number , where the denominator is the operator norm for the Euclidean norm on complex coordinate vectors. No Hermitian or positivity condition is imposed on , and no nonzero condition is imposed on its norm. Real division is total with division by zero equal to zero, so the value is zero for the zero matrix. The index type may be empty, in which case the trace and operator norm are both zero and the defined value is zero.
traceFunction: For every finite index type equipped with decidable equality, every function , and every complex square matrix indexed by , this defines the real part of the trace of the matrix obtained by applying the real continuous functional calculus to and . Explicitly, when is Hermitian, choose a unitary diagonalization with real eigenvalues indexed by ; the matrix inside the trace is , so the defined real number is , with eigenvalues counted with multiplicity. The functional calculus returns the zero matrix when is not self-adjoint, hence the defined value is then zero. Its general definition also returns zero if is not continuous on the real spectrum of ; here that spectrum is finite, so every function is continuous on it and no continuity assumption on is required. No Hermitian, positivity, or nonemptiness assumption is included in this definition. If is empty, the trace is an empty sum and the defined value is zero for every .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.