Projection-valued measures #
A projection-valued measure on a measurable space X with values in the bounded
operators of a complex Hilbert space H is the object the spectral theorem
produces: a countably additive, multiplicative assignment of orthogonal
projections to measurable subsets of X.
Main definitions #
Physicslib4.Spectral.ProjectionValuedMeasure— the blueprint'sμ : Ω(X) → 𝓑(H).Physicslib4.Spectral.ProjectionValuedMeasure.assoc— the positive finite scalar measureμ_ψ : E ↦ ⟪ψ, μ E ψ⟫.
Main statements #
Physicslib4.Spectral.ProjectionValuedMeasure.existsUnique_assoc—μ_ψis a well-defined positive, finite, real-valued measure.Physicslib4.Spectral.ProjectionValuedMeasure.inner_eq_zero_of_disjoint,Physicslib4.Spectral.ProjectionValuedMeasure.norm_sq_eq_sum_of_partition— the vectorsμ(Eᵢ)ψof a finite measurable partition are pairwise orthogonal and their squared norms add up to‖ψ‖².Physicslib4.Spectral.exists_isStarProjection_tendsto_partialSum— a family of mutually annihilating projections sums to the projection onto the closed span of their ranges.
Implementation notes #
Following Mathlib's MeasureTheory.VectorMeasure, a ProjectionValuedMeasure is
a total function Set X → H →L[ℂ] H which is required to vanish on
non-measurable sets; this makes the object canonical and avoids carrying
measurability side conditions in the coercion.
The blueprint's countable additivity is stated as HasSum, Mathlib's unordered
summability. For a family of pairwise orthogonal projections this is equivalent to
convergence of the partial sums, which is the blueprint's formulation.
Two projection-valued measures are equivalent in the blueprint's sense exactly
when they are equal as functions on Set X, so no separate relation is introduced.
A projection-valued measure on X: an assignment E ↦ μ(E) of orthogonal projections on H to measurable sets, with μ(X) = 𝟙, countably additive and multiplicative on intersections.
It is the object the spectral theorem attaches to a self-adjoint operator A, with μ(E) the
spectral projection of A for the set E of possible measured values. The axioms:
each μ(E) is a self-adjoint idempotent; μ(X) = 𝟙; for pairwise disjoint measurable E_1, E_2, …
and every vector v, μ(⋃ E_j)v = Σ μ(E_j)v with convergence in the norm of H; and
μ(E ∩ F) = μ(E)μ(F), so projections of disjoint sets have orthogonal ranges. For each vector ψ,
E ↦ ⟨ψ, μ(E)ψ⟩ is then a finite positive measure (assoc), and bounded measurable functions
can be integrated against μ to give operators (integral).
Implementation note: as in Mathlib's MeasureTheory.VectorMeasure, μ is a total function
Set X → H →L[ℂ] H that is 0 on non-measurable sets, so the coercion carries no measurability
side conditions. The blueprint's μ(∅) = 0 is not a field: countable additivity for the constant
family E_j = ∅ implies it, and a field would add a derivable obligation to every constructor.
Countable additivity is stated with Mathlib's HasSum, unordered summation; since the vectors
μ(E_j)v are pairwise orthogonal, this is equivalent to convergence of the partial sums.
The blueprint's equivalence of two projection-valued measures is equality
(ProjectionValuedMeasure.ext).
Blueprint: def:projection-valued-measure
The underlying assignment E ↦ μ(E) of operators to all subsets of X.
- isStarProjection' (E : Set X) : IsStarProjection (self.toFun E)
Every value μ(E) is an orthogonal projection: self-adjoint and idempotent.
- notMeasurable' (E : Set X) : ¬MeasurableSet E → self.toFun E = 0
Non-measurable sets are sent to 0, so μ is determined by its values on measurable sets.
Normalisation: μ(X) is the identity operator.
- hasSum' (E : ℕ → Set X) : (∀ (j : ℕ), MeasurableSet (E j)) → (Pairwise fun (i j : ℕ) => Disjoint (E i) (E j)) → ∀ (v : H), HasSum (fun (j : ℕ) => (self.toFun (E j)) v) ((self.toFun (⋃ (j : ℕ), E j)) v)
Countable additivity: for pairwise disjoint measurable sets E_j and every vector v, the series Σ μ(E_j)v converges in the norm of H to μ(⋃ E_j)v.
- inter' (E F : Set X) : MeasurableSet E → MeasurableSet F → self.toFun (E ∩ F) = self.toFun E * self.toFun F
Multiplicativity: μ(E ∩ F) = μ(E)μ(F) for measurable E and F.
Instances For
Equations
- Physicslib4.Spectral.ProjectionValuedMeasure.instFunLike = { coe := Physicslib4.Spectral.ProjectionValuedMeasure.toFun, coe_injective := ⋯ }
Two projection-valued measures agreeing as functions on Set X are equal.
The five structure fields, restated through the FunLike coercion.
For every ψ ∈ H there is a unique positive finite measure μ_ψ on X whose
value on a measurable set E is ⟪ψ, μ(E) ψ⟫.
Blueprint reference: thrm:projection-valued-measures-associated-measure.
The finite positive measure μ_ψ(E) = ⟨ψ, μ(E)ψ⟩ of a projection-valued measure μ and a vector ψ.
Since μ(E) is an orthogonal projection, ⟨ψ, μ(E)ψ⟩ = ‖μ(E)ψ‖² ≥ 0, and countable additivity of μ
makes E ↦ ‖μ(E)ψ‖² a measure; its total mass is ⟨ψ, ψ⟩ = ‖ψ‖², by assoc_apply with E = X.
For a unit vector ψ and the spectral measure of an observable A, μ_ψ(E) is the probability that a
measurement of A in the state ψ gives a value in E. The values on measurable sets are
assoc_apply; the operator-valued integral satisfies ⟨ψ, (∫ f dμ)ψ⟩ = ∫ f dμ_ψ
(inner_integral).
Implementation note: the measure is chosen from existsUnique_assoc, which also gives its
uniqueness and finiteness (the MeasureTheory.IsFiniteMeasure instance).
Blueprint: the measure μ_ψ of thrm:projection-valued-measures-associated-measure
Equations
- μ.assoc ψ = Exists.choose ⋯
Instances For
For a finite pairwise disjoint measurable family E₁, …, Eₙ, the vectors μ(Eᵢ)ψ
are pairwise orthogonal.
The blueprint additionally assumes that the Eᵢ cover X; that hypothesis is used
only for the norm identity norm_sq_eq_sum_of_partition, not here, so it is omitted.
Blueprint reference: lmm:lemma2-of-operator-valued-integration (part 1).
For a finite measurable partition E₁, …, Eₙ of X, ‖ψ‖² = ∑ᵢ ‖μ(Eᵢ)ψ‖².
Blueprint reference: lmm:lemma2-of-operator-valued-integration (part 2).
Sums of mutually annihilating projections #
The partial sums ∑_{i<n} Pᵢ ψ of a family of mutually annihilating orthogonal
projections converge in H.
Blueprint reference: prpstn:orthogonal-sum-converges.
The pointwise limit of the partial sums of a family of mutually annihilating orthogonal projections is again a bounded orthogonal projection.
Blueprint reference: prpstn:orthogonal-sum-is-projection.
The range of the limit projection is the smallest closed subspace containing the
ranges of the Pᵢ.
Blueprint reference: prpstn:orthogonal-sum-range.
A family of mutually annihilating orthogonal projections sums pointwise to the orthogonal projection onto the smallest closed subspace containing all their ranges.
This blueprint node is exactly the conjunction of
exists_isStarProjection_tendsto_partialSum' and
range_eq_topologicalClosure_iSup_of_tendsto_partialSum, so it is proved from them
rather than restated.
Blueprint reference: lmm:lemma-4.
A family of mutually annihilating orthogonal projections is unconditionally
summable, and the (HasSum) value is the pointwise limit of the partial sums
over Finset.range.
Blueprint reference: prpstn:mua-countably-additive.