Documentation

MIPRE.Background.Orthonormalization.Orthogonalization.MvN.Perturb

Real scalars and the Loewner order #

theorem Orthogonalization.MvN.smul_nonneg_real {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {x : H →L[ℂ] H} {δ : ℝ} (hδ : 0 ≤ δ) (hx : 0 ≤ x) :
0 ≤ ↑δ • x

Multiplication by a nonnegative real scalar preserves positivity (it is the conjugation by √δ • 1).

theorem Orthogonalization.MvN.smul_le_smul_real {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {x y : H →L[ℂ] H} (h : x ≤ y) {δ : ℝ} (hδ : 0 ≤ δ) :
↑δ • x ≤ ↑δ • y

Multiplication by a nonnegative real scalar is monotone.

theorem Orthogonalization.MvN.smul_le_smul_scalar {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {x : H →L[ℂ] H} {α β : ℝ} (h : α ≤ β) (hx : 0 ≤ x) :
↑α • x ≤ ↑β • x

For 0 ≤ x and real α ≤ β, α • x ≤ β • x.

A real number is a self-adjoint complex number.

Corners #

theorem Orthogonalization.MvN.corner_mul_left {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {r y : H →L[ℂ] H} (hr : IsStarProjection r) (hy : r * y * r = y) :
r * y = y

An element of the corner r M r is fixed by left multiplication by r.

theorem Orthogonalization.MvN.corner_mul_right {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {r y : H →L[ℂ] H} (hr : IsStarProjection r) (hy : r * y * r = y) :
y * r = y

An element of the corner r M r is fixed by right multiplication by r.

theorem Orthogonalization.MvN.corner_conj {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {r : H →L[ℂ] H} (hr : IsStarProjection r) (x : H →L[ℂ] H) :
r * (r * x * r) * r = r * x * r

r y r lies in the corner of r.

theorem Orthogonalization.MvN.corner_of_corner {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] {c r y : H →L[ℂ] H} (hcr : c * r = r) (hrc : r * c = r) (hy : r * y * r = y) :
c * y * c = y

If r ≤ c (c r = r = r c) then the corner of r is contained in the corner of c.

theorem Orthogonalization.MvN.corner_star {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {r y : H →L[ℂ] H} (hr : IsStarProjection r) (hy : r * y * r = y) :
r * star y * r = star y

The adjoint of an element of the corner r M r lies in the corner.

The center of the corner p M p #

theorem Orthogonalization.MvN.commute_central_mul_proj {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {M : VonNeumannAlgebra H} {p r z y : H →L[ℂ] H} (hr : IsStarProjection r) (hpr : p * r = r) (hrp : r * p = r) (hz : IsCentralIn M p z) (hyM : y ∈ M) (hy : r * y * r = y) :
Commute (z * r) y

An element z r with z ∈ Z(p M p) and r ≤ p commutes with every element of r M r.

Step 1: the spectral projection χ_{[δ, 1-δ]}(e) #

noncomputable def Orthogonalization.MvN.clip (t : ℝ) :

The clip t ↦ max 0 (min t 1) of the real line onto [0, 1].

Equations
Instances For
    theorem Orthogonalization.MvN.clip_of_mem {t : ℝ} (h0 : 0 ≤ t) (h1 : t ≤ 1) :
    clip t = t
    theorem Orthogonalization.MvN.clip_le_of_le {t δ : ℝ} (hδ : 0 ≤ δ) (h : t ≤ δ) :
    clip t ≤ δ
    theorem Orthogonalization.MvN.le_clip_of_le {t δ : ℝ} (hδ : δ ≤ 1) (h : δ ≤ t) :
    δ ≤ clip t
    noncomputable def Orthogonalization.MvN.bump (t : ℝ) :

    t ↦ clip t (1 - clip t): bounded, continuous, equal to t (1 - t) on [0, 1].

    Equations
    Instances For
      theorem Orthogonalization.MvN.bump_le_of_notMem {t δ : ℝ} (hδ0 : 0 < δ) (ht : t ∉ Set.Icc δ (1 - δ)) :
      bump t ≤ δ

      Outside [δ, 1 - δ] the bump is at most δ.

      The spectrum of 0 ≤ e ≤ 1 lies in [0, 1].

      clip(e) = e when the spectrum of e lies in [0, 1].

      bump(e) = e (1 - e) when the spectrum of e lies in [0, 1].

      theorem Orthogonalization.MvN.exists_spectral_middle {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {e : H →L[ℂ] H} (he : IsSelfAdjoint e) (he0 : 0 ≤ e) (he1 : e ≤ 1) (hne : ¬IsStarProjection e) :
      ∃ (δ : ℝ), 0 < δ ∧ δ < 1 / 2 ∧ CommutingRepetition.BorelCalc.P e he (Set.Icc δ (1 - δ)) ≠ 0

      A self-adjoint contraction 0 ≤ e ≤ 1 which is not a projection has a nonzero spectral projection χ_{[δ, 1-δ]}(e) for some 0 < δ < 1/2: otherwise ‖e (1 - e)‖ ≤ δ for every such δ, so e² = e.

      δ r ≤ e r for r = χ_{[δ, 1-δ]}(e).

      e r ≤ (1 - δ) r for r = χ_{[δ, 1-δ]}(e).

      theorem Orthogonalization.MvN.P_le_smul_mul_P {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {e : H →L[ℂ] H} (he : IsSelfAdjoint e) (hs : spectrum ℝ e ⊆ Set.Icc 0 1) {δ : ℝ} (hδ0 : 0 < δ) :

      r ≤ δ⁻¹ e r for r = χ_{[δ, 1-δ]}(e).

      theorem Orthogonalization.MvN.proj_mul_eq_self_of_le_smul {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {r e c : H →L[ℂ] H} (hr : IsStarProjection r) (hc : IsStarProjection c) (hrc : Commute r c) (hce : c * e = e) {s : ℂ} (hle : r ≤ s • (e * r)) :
      r * c = r

      A projection r ≤ s • (e r) (as r = χ_{[δ, 1-δ]}(e) with s = δ⁻¹) lies below every projection c commuting with it such that c e = e.

      theorem Orthogonalization.MvN.exists_middle_projection {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] (M : VonNeumannAlgebra H) {e c a : H →L[ℂ] H} (heM : e ∈ M) (he0 : 0 ≤ e) (he1 : e ≤ 1) (hne : ¬IsStarProjection e) (hc : IsStarProjection c) (hce : c * e = e) (hec : e * c = e) (hea : Commute e a) :
      ∃ (δ : ℝ), 0 < δ ∧ δ < 1 / 2 ∧ ∃ (r : H →L[ℂ] H), r ≠ 0 ∧ IsStarProjection r ∧ r ∈ M ∧ Commute r a ∧ Commute r e ∧ r * c = r ∧ c * r = r ∧ ↑δ • r ≤ e * r ∧ e * r ≤ ↑(1 - δ) • r

      Step 1. For 0 ≤ e ≤ 1 in c M c (c a projection), commuting with a and not a projection, some spectral projection r = χ_{[δ, 1-δ]}(e) (0 < δ < 1/2) is nonzero; it lies in M, below c, commutes with a and e, and δ r ≤ e r ≤ (1 - δ) r.

      Step 2: a self-adjoint element of r M r commuting with a, not in Z(p M p) r #

      Real and imaginary parts: x = ½ ((x + x*) + (-i) (i (x - x*))).

      theorem Orthogonalization.MvN.commute_of_forall_selfAdjoint {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] (M : VonNeumannAlgebra H) {p r : H →L[ℂ] H} (hr : IsStarProjection r) (hpr : p * r = r) (hrp : r * p = r) (hall : ∀ b ∈ M, r * b * r = b → IsSelfAdjoint b → ∃ (z : H →L[ℂ] H), IsCentralIn M p z ∧ b = z * r) {x y : H →L[ℂ] H} (hxM : x ∈ M) (hrx : r * x * r = x) (hyM : y ∈ M) (hry : r * y * r = y) :

      If every self-adjoint element of r M r is of the form z r with z ∈ Z(p M p) (r ≤ p), then r M r is commutative.

      theorem Orthogonalization.MvN.exists_selfAdjoint_not_central {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] (M : VonNeumannAlgebra H) {p r a : H →L[ℂ] H} (hr : IsStarProjection r) (hrM : r ∈ M) (hpr : p * r = r) (hrp : r * p = r) (hnab : ¬IsAbelianProj M r) (haM : a ∈ M) (hasa : IsSelfAdjoint a) (hra : Commute r a) :
      ∃ b ∈ M, r * b * r = b ∧ IsSelfAdjoint b ∧ Commute b a ∧ ∀ (z : H →L[ℂ] H), IsCentralIn M p z → b ≠ z * r

      Step 2. If r M r is not abelian (r ≤ p a projection of M) and r commutes with the self-adjoint a ∈ M, there is a self-adjoint b ∈ r M r commuting with a which is not of the form z r with z ∈ Z(p M p): r a if r a is not of that form, and otherwise any self-adjoint element of r M r not of that form (all of r M r then commutes with a).

      Step 3: normalization to 0 ≤ b ≤ r #

      A self-adjoint b = r b r satisfies b ≤ ‖b‖ r.

      A self-adjoint b = r b r satisfies -‖b‖ r ≤ b.

      theorem Orthogonalization.MvN.exists_normalized {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] (M : VonNeumannAlgebra H) {p r a : H →L[ℂ] H} (hp : IsStarProjection p) (hpM : p ∈ M) (hr : IsStarProjection r) (hrM : r ∈ M) (hpr : p * r = r) (hra : Commute r a) {b₁ : H →L[ℂ] H} (hb₁M : b₁ ∈ M) (hrb₁ : r * b₁ * r = b₁) (hb₁sa : IsSelfAdjoint b₁) (hb₁a : Commute b₁ a) (hb₁Q : ∀ (z : H →L[ℂ] H), IsCentralIn M p z → b₁ ≠ z * r) :
      ∃ b ∈ M, r * b * r = b ∧ 0 ≤ b ∧ b ≤ r ∧ Commute b a ∧ ∀ (z : H →L[ℂ] H), IsCentralIn M p z → b ≠ z * r

      Step 3. A self-adjoint b₁ ∈ r M r commuting with a and not of the form z r (z ∈ Z(p M p)) is renormalized to b₂ = (2 ‖b₁‖)⁻¹ (b₁ + ‖b₁‖ r), which satisfies 0 ≤ b₂ ≤ r and keeps the other properties.

      Step 5: the perturbation stays in [0, 1] #

      theorem Orthogonalization.MvN.perturb_mem_Icc {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] {e r b : H →L[ℂ] H} (hr : IsStarProjection r) (he0 : 0 ≤ e) (he1 : e ≤ 1) (hre : Commute r e) {δ : ℝ} (hlow : ↑δ • r ≤ e * r) (hupp : e * r ≤ ↑(1 - δ) • r) (hbsa : IsSelfAdjoint b) (hrb : r * b * r = b) (hbn : ‖b‖ ≤ 2) {t : ℝ} (ht : |t| ≤ δ / 2) :
      0 ≤ e + ↑t • b ∧ e + ↑t • b ≤ 1

      Step 5. If r is a projection commuting with 0 ≤ e ≤ 1 with δ r ≤ e r ≤ (1 - δ) r, and b = r b r is self-adjoint of norm ≤ 2, then e + t b ∈ [0, 1] for |t| ≤ δ / 2: decompose e = e (1 - r) + e r with 0 ≤ e (1 - r) ≤ 1 - r, and -δ r ≤ t b ≤ δ r.

      The perturbation step of Lemma 3.1 #

      theorem Orthogonalization.MvN.exists_perturbation {H : Type u} [NormedAddCommGroup H] [InnerProductSpace ℂ H] [CompleteSpace H] (M : VonNeumannAlgebra H) {p : H →L[ℂ] H} (hp : IsStarProjection p) (hpM : p ∈ M) {c : H →L[ℂ] H} (hc : IsStarProjection c) (hcc : IsCentralIn M p c) (hII : ∀ (r : H →L[ℂ] H), IsAbelianProj M r → r * c = r → r = 0) (E : (H →L[ℂ] H) →ₗ[ℂ] H →L[ℂ] H) (hE : IsCenterValuedTrace M p E) {a e : H →L[ℂ] H} (haM : a ∈ M) (hac : c * a * c = a) (ha0 : 0 ≤ a) (heM : e ∈ M) (hec : c * e * c = e) (he0 : 0 ≤ e) (he1 : e ≤ 1) (hea : Commute e a) (hne : ¬IsStarProjection e) :
      ∃ (b : H →L[ℂ] H), b ≠ 0 ∧ b ∈ M ∧ c * b * c = b ∧ IsSelfAdjoint b ∧ Commute b a ∧ E b = 0 ∧ ∃ (t₀ : ℝ), 0 < t₀ ∧ ∀ (t : ℝ), |t| ≤ t₀ → 0 ≤ e + ↑t • b ∧ e + ↑t • b ≤ 1

      The type II₁ perturbation of Lemma 3.1 (paper, Section 3, "The latter case is easier"): if e ∈ c M c, 0 ≤ e ≤ 1, commutes with a and is not a projection, then there is a nonzero self-adjoint b ∈ c M c commuting with a with E b = 0 such that e + t b ∈ [0, 1] for all small real t. Here c is a central projection of the finite corner p M p whose corner c M c has no nonzero abelian projection (type II₁), and E is the center-valued trace of p M p.