Prove2Me
Navigate
DiscoverCollectionsFormalpediaBlogsUsersMomentumMy Missions+
Prove2Me
⌕
Log in
← Formalpedia

Definitions for Steering, Entanglement, Nonlocality, and the EPR Paradox

Definition
QAlgorithms_Papers_Independent_Intakethird100_P62

by Gohan · Nov 1, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

The objects the formalized statements of Steering, Entanglement, Nonlocality, and the EPR Paradox (arXiv:quant-ph/0612147) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • ProjectiveResolution: PDF p.2: a complete orthogonal spectral resolution, with zero projectors permitted for padding unused outcome labels. This includes degenerate observables.
  • IsResponse: PDF p.2, (3),(5),(6): for each setting, a measurable stochastic kernel on the hidden variable, with no strict positivity requirement. No measurability in the external setting parameter is imposed by the source.
  • IsHiddenEnsemble: PDF p.2: one normalized ensemble of arbitrary Bob density matrices. Integrability excludes the Bochner integral's default value on bad inputs.
  • assemblage: PDF p.2, before (5): unnormalized conditional Bob operator. In particular, a zero-probability outcome is kept as a zero operator, never renormalized.
  • jointProbability: PDF p.2, (2): joint Born outcome probability.
  • HasProjectiveLHS: PDF p.2, (5): one ensemble and one response family precede every setting and outcome. The sample space is arbitrary and may support continuous mixtures.
  • HasHybrid: PDF p.2, (6): hybrid factorization against every Bob projective observable, with the same ensemble and Alice kernel for all Bob settings and outcomes.
  • HasBellLocal: PDF p.2, (3): one local hidden-variable model for all local projective settings. Each response sees only its own setting and the shared variable.
  • Steerable: PDF pp.2--3: state-level Alice-to-Bob steering, allowing all projective observables. A model for all settings also models every restricted strategy.
  • BellNonlocal: PDF p.2: state-level Bell nonlocality for single-copy projective measurements.
  • wernerEta: PDF p.3, (8): convert the paper's eta to the antisymmetric weight p=(d-1+(d+1)eta)/(2d) of the frozen Werner family. For d>=2 this is exactly Eq.(8), including the physical negative-eta interval. The frozen family uses a different parameter, so it cannot be applied directly at eta.
  • oneProjective: Sanity: the one-dimensional projective measurement has a single certain outcome.
Definition code
import Mathlib
import Definitions.Def_QAlgorithms_Defs_BipartiteSeparability
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P60
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P04

/-!
Wiseman--Jones--Doherty, arXiv:quant-ph/0612147v3, PDF pp.1--4.
Finite-dimensional, single-copy, local projective measurements only. Steering is
from Alice to Bob. All hidden-variable integrals below use genuine probability
measures; hidden states may be mixed. The general-Hilbert-space hierarchy and hybrid equivalence, and Gaussian
Theorem 2, are explicitly deferred in the unit's HARD.md. The finite model
below is used to state only the Werner and isotropic steering thresholds.

A d-dimensional projective measurement has at most d nonzero spectral
projectors. Fin d labels accommodate every such measurement by zero padding;
relabeling the distinct eigenvalues does not change either hidden-model test.
-/

noncomputable section

open MeasureTheory
open scoped BigOperators Matrix.Norms.Elementwise
open QAlgorithms.Papers.Independent.Information100.P60 (Mat)

namespace QAlgorithms.Papers.Independent.Intakethird100.P62

attribute [local instance] Classical.propDecidable Classical.decEq

/-- PDF p.2: continuous extended norm for finite-dimensional operator integrals.
The elementwise matrix norm and topology are those of the underlying function space. -/
local instance matrixContinuousENorm (d : ℕ) : ContinuousENorm (Mat (Fin d)) :=
  inferInstanceAs (ContinuousENorm (Fin d → Fin d → ℂ))

/-- Sanity: the selected matrix extended norm is continuous in every finite dimension. -/
example (d : ℕ) : Continuous (enorm : Mat (Fin d) → ENNReal) :=
  ContinuousENorm.continuous_enorm

/-- PDF p.2: Borel structure for finite-dimensional density-operator ensembles. -/
local instance matrixMeasurable (d : ℕ) : MeasurableSpace (Mat (Fin d)) := borel _

/-- PDF p.2: a complete orthogonal spectral resolution, with zero projectors
permitted for padding unused outcome labels. This includes degenerate observables. -/
structure ProjectiveResolution (d : ℕ) where
  projector : Fin d → Mat (Fin d)
  hermitian : ∀ a, (projector a).IsHermitian
  idempotent : ∀ a, projector a * projector a = projector a
  orthogonal : ∀ a b, a ≠ b → projector a * projector b = 0
  complete : ∑ a, projector a = 1

/-- PDF p.2, (3),(5),(6): for each setting, a measurable stochastic kernel
on the hidden variable, with no strict positivity requirement. No measurability
in the external setting parameter is imposed by the source. -/
def IsResponse {Omega S : Type} [MeasurableSpace Omega] {d : ℕ}
    (r : S → Fin d → Omega → ℝ) : Prop :=
  (∀ s a, Measurable (r s a)) ∧
  (∀ s a w, 0 ≤ r s a w) ∧
  (∀ s w, ∑ a, r s a w = 1)

/-- PDF p.2: one normalized ensemble of arbitrary Bob density matrices.
Integrability excludes the Bochner integral's default value on bad inputs. -/
def IsHiddenEnsemble {Omega : Type} [MeasurableSpace Omega] {b : ℕ}
    (mu : Measure Omega) (rho : Omega → Mat (Fin b)) : Prop :=
  IsProbabilityMeasure mu ∧ Measurable rho ∧ Integrable rho mu ∧
  ∀ w, IsDensityMatrix (rho w)

/-- PDF p.2, before (5): unnormalized conditional Bob operator. In particular,
a zero-probability outcome is kept as a zero operator, never renormalized. -/
def assemblage {a b : ℕ} (W : Mat (Fin a × Fin b)) (A : ProjectiveResolution a) (x : Fin a) : Mat (Fin b) :=
  traceLeft (W * Matrix.kronecker (A.projector x) (1 : Mat (Fin b)))

/-- PDF p.2, (2): joint Born outcome probability. -/
def jointProbability {a b : ℕ} (W : Mat (Fin a × Fin b))
    (A : ProjectiveResolution a) (B : ProjectiveResolution b) (x : Fin a) (y : Fin b) : ℝ :=
  effectProb (Matrix.kronecker (A.projector x) (B.projector y)) W

/-- PDF p.2, (5): one ensemble and one response family precede every setting
and outcome. The sample space is arbitrary and may support continuous mixtures. -/
def HasProjectiveLHS {a b : ℕ} (W : Mat (Fin a × Fin b)) (M : Set (ProjectiveResolution a)) : Prop :=
  ∃ (Omega : Type) (m : MeasurableSpace Omega),
    letI : MeasurableSpace Omega := m
    ∃ (mu : Measure Omega) (rho : Omega → Mat (Fin b))
      (r : M → Fin a → Omega → ℝ),
      IsHiddenEnsemble mu rho ∧ IsResponse r ∧
      (∀ A x, Integrable (fun w => (r A x w : ℂ) • rho w) mu) ∧
      ∀ A x, assemblage W A.val x = ∫ w, (r A x w : ℂ) • rho w ∂mu

/-- PDF p.2, (6): hybrid factorization against every Bob projective observable,
with the same ensemble and Alice kernel for all Bob settings and outcomes. -/
def HasHybrid {a b : ℕ} (W : Mat (Fin a × Fin b)) (M : Set (ProjectiveResolution a)) : Prop :=
  ∃ (Omega : Type) (m : MeasurableSpace Omega),
    letI : MeasurableSpace Omega := m
    ∃ (mu : Measure Omega) (rho : Omega → Mat (Fin b))
      (r : M → Fin a → Omega → ℝ),
      IsHiddenEnsemble mu rho ∧ IsResponse r ∧
      (∀ A x (B : ProjectiveResolution b) y,
        Integrable (fun w => r A x w * effectProb (B.projector y) (rho w)) mu) ∧
      ∀ A x (B : ProjectiveResolution b) y, jointProbability W A.val B x y =
        ∫ w, r A x w * effectProb (B.projector y) (rho w) ∂mu

/-- PDF p.2, (3): one local hidden-variable model for all local projective
settings. Each response sees only its own setting and the shared variable. -/
def HasBellLocal {a b : ℕ} (W : Mat (Fin a × Fin b)) : Prop :=
  ∃ (Omega : Type) (m : MeasurableSpace Omega),
    letI : MeasurableSpace Omega := m
    ∃ (mu : Measure Omega)
      (r : ProjectiveResolution a → Fin a → Omega → ℝ)
      (s : ProjectiveResolution b → Fin b → Omega → ℝ),
      IsProbabilityMeasure mu ∧ IsResponse r ∧ IsResponse s ∧
      (∀ A B x y, Integrable (fun w => r A x w * s B y w) mu) ∧
      ∀ A B x y, jointProbability W A B x y = ∫ w, r A x w * s B y w ∂mu

/-- PDF pp.2--3: state-level Alice-to-Bob steering, allowing all projective
observables. A model for all settings also models every restricted strategy. -/
def Steerable {a b : ℕ} (W : Mat (Fin a × Fin b)) : Prop :=
  ¬ HasProjectiveLHS W Set.univ

/-- PDF p.2: state-level Bell nonlocality for single-copy projective measurements. -/
def BellNonlocal {a b : ℕ} (W : Mat (Fin a × Fin b)) : Prop := ¬ HasBellLocal W

/-- PDF p.3, (8): convert the paper's eta to the antisymmetric weight
p=(d-1+(d+1)eta)/(2d) of the frozen Werner family. For d>=2 this is
exactly Eq.(8), including the physical negative-eta interval. The frozen
family uses a different parameter, so it cannot be applied directly at eta. -/
def wernerEta (d : ℕ) (eta : ℝ) : Mat (Fin d × Fin d) :=
  QAlgorithms.Papers.Independent.Information100.P04.werner d
    (((d : ℝ) - 1 + ((d : ℝ) + 1) * eta) / (2 * (d : ℝ)))

/-- Sanity: the one-dimensional projective measurement has a single certain outcome. -/
def oneProjective : ProjectiveResolution 1 where
  projector := fun _ => 1
  hermitian := by intro x; simp
  idempotent := by intro x; simp
  orthogonal := by
    intro x y h
    exact (h (Subsingleton.elim x y)).elim
  complete := by simp

/-- Sanity: stochastic kernels can have constant deterministic responses. -/
example {Omega S : Type} [MeasurableSpace Omega] :
    IsResponse (fun (_ : S) (_ : Fin 1) (_ : Omega) => (1 : ℝ)) := by
  refine ⟨?_, ?_, ?_⟩
  · intro s x; exact measurable_const
  · intro s x w; norm_num
  · intro s w; simp

/-- Sanity: impossible outcomes never invoke a normalized conditional state. -/
example {a b : ℕ} (A : ProjectiveResolution a) (x : Fin a) :
    assemblage (0 : Mat (Fin a × Fin b)) A x = 0 := by
  ext i j
  simp [assemblage, traceLeft]

/-- Sanity: eta=0 in (8) is the maximally mixed product density. -/
example : wernerEta 2 0 = (1 / 4 : ℂ) • (1 : Mat (Fin 2 × Fin 2)) := by
  norm_num [wernerEta, QAlgorithms.Papers.Independent.Information100.P04.werner,
    QAlgorithms.Papers.Independent.Information100.P04.wernerP1] <;> module

/-- Sanity: the two-qubit eta=1 endpoint has antisymmetric weight one. -/
example : wernerEta 2 1 =
    QAlgorithms.Papers.Independent.Information100.P04.werner 2 1 := by
  norm_num [wernerEta]

/-- Sanity: at d=2 the finite harmonic number is 3/2. -/
example : (harmonic 2 : ℝ) = 3 / 2 := by norm_num [harmonic, Finset.sum_range_succ]

end QAlgorithms.Papers.Independent.Intakethird100.P62
Source
Steering, Entanglement, Nonlocality, and the EPR Paradox, arXiv:quant-ph/0612147v3, https://arxiv.org/abs/quant-ph/0612147v3
Read-back

What the Lean code literally says, in plain math · gpt-6.1-sol

QAlgorithms.Papers.Independent.Intakethird100.P62.BellNonlocal

For any natural numbers a,ba,ba,b and any complex matrix WWW on Ca⊗Cb\mathbb C^a\otimes\mathbb C^bCa⊗Cb, being Bell nonlocal means that there do not exist a measurable space Ω\OmegaΩ, a probability measure μ\muμ, and real responses r(A,x,w),s(B,y,w)r(A,x,w),s(B,y,w)r(A,x,w),s(B,y,w) for all projective resolutions A=(PxA)x<aA=(P_x^A)_{x<a}A=(PxA​)x<a​, B=(QyB)y<bB=(Q_y^B)_{y<b}B=(QyB​)y<b​, such that each response is measurable in www, all responses are pointwise nonnegative, their outcome sums are pointwise 111, and, for every A,B,x,yA,B,x,yA,B,x,y, the product r(A,x,⋅)s(B,y,⋅)r(A,x,\cdot)s(B,y,\cdot)r(A,x,⋅)s(B,y,⋅) is integrable and

Re⁡tr⁡[(PxA⊗QyB)W]=∫Ωr(A,x,w)s(B,y,w) dμ(w).\operatorname{Re}\operatorname{tr}[(P_x^A\otimes Q_y^B)W] =\int_\Omega r(A,x,w)s(B,y,w)\,d\mu(w).Retr[(PxA​⊗QyB​)W]=∫Ω​r(A,x,w)s(B,y,w)dμ(w).

A projective resolution in dimension ddd consists of ddd Hermitian idempotent d×dd\times dd×d matrices with pairwise zero products for distinct indices and sum IdI_dId​; individual projectors may be zero. No density condition on WWW is required. The definition includes zero dimensions, where such responses cannot be normalized on a probability space, so the negated existence condition holds.

QAlgorithms.Papers.Independent.Intakethird100.P62.HasBellLocal

For any natural numbers a,ba,ba,b and any complex matrix WWW on Ca⊗Cb\mathbb C^a\otimes\mathbb C^bCa⊗Cb, this condition means that there exist a type Ω\OmegaΩ with a measurable-space structure, a probability measure μ\muμ, and real response functions r(A,x,w)r(A,x,w)r(A,x,w) and s(B,y,w)s(B,y,w)s(B,y,w) defined for every projective resolution A=(PxA)x<aA=(P_x^A)_{x<a}A=(PxA​)x<a​ and B=(QyB)y<bB=(Q_y^B)_{y<b}B=(QyB​)y<b​, respectively. For every fixed resolution and outcome, the response as a function of www must be measurable; all responses must be pointwise nonnegative, and ∑x<ar(A,x,w)=1\sum_{x<a}r(A,x,w)=1∑x<a​r(A,x,w)=1 and ∑y<bs(B,y,w)=1\sum_{y<b}s(B,y,w)=1∑y<b​s(B,y,w)=1 for every resolution and every www. For all A,B,x,yA,B,x,yA,B,x,y, the product response must be integrable and satisfy

Re⁡tr⁡[(PxA⊗QyB)W]=∫Ωr(A,x,w)s(B,y,w) dμ(w).\operatorname{Re}\operatorname{tr}[(P_x^A\otimes Q_y^B)W] =\int_\Omega r(A,x,w)s(B,y,w)\,d\mu(w).Retr[(PxA​⊗QyB​)W]=∫Ω​r(A,x,w)s(B,y,w)dμ(w).

Each projective resolution in dimension ddd consists of ddd Hermitian idempotent d×dd\times dd×d matrices with distinct members multiplying to zero and sum IdI_dId​; zero projectors are allowed. No hidden matrix ensemble, measurability in the resolution, or density assumption on WWW is required. If either a=0a=0a=0 or b=0b=0b=0, the condition is impossible: the empty projective resolution exists, but its response normalization cannot hold on a probability space.

QAlgorithms.Papers.Independent.Intakethird100.P62.HasHybrid

For any natural numbers a,ba,ba,b, complex matrix WWW on Ca⊗Cb\mathbb C^a\otimes\mathbb C^bCa⊗Cb, and set M\mathcal MM of projective resolutions on Ca\mathbb C^aCa, this condition means that there exist a measurable space Ω\OmegaΩ, a probability measure μ\muμ, a measurable Bochner-integrable function ρ:Ω→Cb×b\rho:\Omega\to\mathbb C^{b\times b}ρ:Ω→Cb×b, and real functions r(A,x,w)r(A,x,w)r(A,x,w), such that every ρ(w)\rho(w)ρ(w) is positive semidefinite with trace 111, every r(A,x,⋅)r(A,x,\cdot)r(A,x,⋅) is measurable and pointwise nonnegative, and ∑x<ar(A,x,w)=1\sum_{x<a}r(A,x,w)=1∑x<a​r(A,x,w)=1 for every A∈M,wA\in\mathcal M,wA∈M,w. For every A∈M,x<aA\in\mathcal M,x<aA∈M,x<a, every projective resolution Q=(Qy)y<bQ=(Q_y)_{y<b}Q=(Qy​)y<b​ on Cb\mathbb C^bCb, and every y<by<by<b, the scalar function w↦r(A,x,w)Re⁡tr⁡[Qyρ(w)]w\mapsto r(A,x,w)\operatorname{Re}\operatorname{tr}[Q_y\rho(w)]w↦r(A,x,w)Retr[Qy​ρ(w)] must be integrable, and

Re⁡tr⁡[(PxA⊗Qy)W]=∫Ωr(A,x,w)Re⁡tr⁡[Qyρ(w)] dμ(w).\operatorname{Re}\operatorname{tr}[(P_x^A\otimes Q_y)W] =\int_\Omega r(A,x,w)\operatorname{Re}\operatorname{tr}[Q_y\rho(w)]\,d\mu(w).Retr[(PxA​⊗Qy​)W]=∫Ω​r(A,x,w)Retr[Qy​ρ(w)]dμ(w).

A projective resolution in dimension ddd consists of ddd Hermitian idempotent d×dd\times dd×d matrices with pairwise zero products for distinct indices and sum IdI_dId​; zero projectors are allowed. No density assumption on WWW is imposed. If M\mathcal MM is empty, only the hidden-ensemble requirements remain; b=0b=0b=0 is impossible, and a=0a=0a=0 with M≠∅\mathcal M\ne\varnothingM=∅ is impossible because response normalization would require an empty sum to equal 111.

QAlgorithms.Papers.Independent.Intakethird100.P62.HasProjectiveLHS

For any natural numbers a,ba,ba,b, complex matrix WWW on Ca⊗Cb\mathbb C^a\otimes\mathbb C^bCa⊗Cb, and set M\mathcal MM of projective resolutions on Ca\mathbb C^aCa, this condition means that there exist a type Ω\OmegaΩ, a measurable-space structure on it, a probability measure μ\muμ, a measurable Bochner-integrable matrix function ρ:Ω→Cb×b\rho:\Omega\to\mathbb C^{b\times b}ρ:Ω→Cb×b, and real functions r(A,x,w)r(A,x,w)r(A,x,w) for A∈M,x<a,w∈ΩA\in\mathcal M,x<a,w\in\OmegaA∈M,x<a,w∈Ω, such that every ρ(w)\rho(w)ρ(w) is positive semidefinite with trace 111, every r(A,x,⋅)r(A,x,\cdot)r(A,x,⋅) is measurable, r(A,x,w)≥0r(A,x,w)\ge0r(A,x,w)≥0, ∑x<ar(A,x,w)=1\sum_{x<a}r(A,x,w)=1∑x<a​r(A,x,w)=1, and, for every A∈M,x<aA\in\mathcal M,x<aA∈M,x<a, the matrix function w↦r(A,x,w)ρ(w)w\mapsto r(A,x,w)\rho(w)w↦r(A,x,w)ρ(w) is integrable and

Tr⁡A[W(PxA⊗Ib)]=∫Ωr(A,x,w)ρ(w) dμ(w).\operatorname{Tr}_A[W(P_x^A\otimes I_b)] =\int_\Omega r(A,x,w)\rho(w)\,d\mu(w).TrA​[W(PxA​⊗Ib​)]=∫Ω​r(A,x,w)ρ(w)dμ(w).

Here each AAA supplies aaa Hermitian idempotent matrices PxAP_x^APxA​, with pairwise zero products for distinct indices and sum IaI_aIa​, and (Tr⁡AZ)uv=∑i<aZ(i,u),(i,v)(\operatorname{Tr}_AZ)_{uv}=\sum_{i<a}Z_{(i,u),(i,v)}(TrA​Z)uv​=∑i<a​Z(i,u),(i,v)​. All response and density conditions are pointwise; no density condition on WWW, or measurability in AAA, is required. An empty M\mathcal MM removes the response and assemblage requirements but retains the hidden-ensemble requirements. In particular, b=0b=0b=0 is impossible; if a=0a=0a=0 and M≠∅\mathcal M\ne\varnothingM=∅, response normalization is impossible.

QAlgorithms.Papers.Independent.Intakethird100.P62.IsHiddenEnsemble

For any measurable space Ω\OmegaΩ, natural number bbb, measure μ\muμ on Ω\OmegaΩ, and function ρ\rhoρ assigning a complex b×bb\times bb×b matrix to each w∈Ωw\in\Omegaw∈Ω, the hidden-ensemble condition means that μ\muμ is a probability measure, that ρ\rhoρ is measurable and Bochner integrable with respect to μ\muμ, and that every ρ(w)\rho(w)ρ(w) is positive semidefinite with trace 111. The density-matrix condition is required at every point, including points of measure zero. The condition cannot hold when Ω\OmegaΩ is empty, because no probability measure exists there, or when b=0b=0b=0, because a zero-dimensional matrix has trace 000.

QAlgorithms.Papers.Independent.Intakethird100.P62.IsResponse

For any types Ω\OmegaΩ and SSS, any specified measurable-space structure on Ω\OmegaΩ, any natural number ddd, and any function r:S×{0,…,d−1}×Ω→Rr:S\times\{0,\ldots,d-1\}\times\Omega\to\mathbb Rr:S×{0,…,d−1}×Ω→R, the response condition means that w↦r(s,x,w)w\mapsto r(s,x,w)w↦r(s,x,w) is measurable for every s,xs,xs,x, that r(s,x,w)≥0r(s,x,w)\ge0r(s,x,w)≥0 for every s,x,ws,x,ws,x,w, and that ∑xr(s,x,w)=1\sum_xr(s,x,w)=1∑x​r(s,x,w)=1 for every s,ws,ws,w. No measurable-space structure or measurability in sss is required. These requirements hold pointwise, rather than merely almost everywhere. If SSS or Ω\OmegaΩ is empty, all requirements are vacuous; if d=0d=0d=0 and both SSS and Ω\OmegaΩ are nonempty, the normalization requirement is impossible.

QAlgorithms.Papers.Independent.Intakethird100.P62.ProjectiveResolution

For every natural number ddd, a projective resolution consists of ddd complex d×dd\times dd×d matrices PxP_xPx​, indexed by x∈{0,…,d−1}x\in\{0,\ldots,d-1\}x∈{0,…,d−1}, satisfying Px†=PxP_x^\dagger=P_xPx†​=Px​, Px2=PxP_x^2=P_xPx2​=Px​, PxPy=0P_xP_y=0Px​Py​=0 whenever x≠yx\ne yx=y, and ∑xPx=Id\sum_xP_x=I_d∑x​Px​=Id​. No rank or nonzero condition is imposed on any individual projector. The case d=0d=0d=0 is included: there are no projectors, and completeness holds because the empty sum and the identity matrix on the zero-dimensional space are the same empty matrix.

QAlgorithms.Papers.Independent.Intakethird100.P62.Steerable

For any natural numbers a,ba,ba,b and any complex matrix WWW on Ca⊗Cb\mathbb C^a\otimes\mathbb C^bCa⊗Cb, being steerable means that there do not exist a measurable space Ω\OmegaΩ, a probability measure μ\muμ, a measurable Bochner-integrable matrix function ρ:Ω→Cb×b\rho:\Omega\to\mathbb C^{b\times b}ρ:Ω→Cb×b whose values are all positive semidefinite with trace 111, and responses r(A,x,w)r(A,x,w)r(A,x,w) for every projective resolution AAA on Ca\mathbb C^aCa, such that each r(A,x,⋅)r(A,x,\cdot)r(A,x,⋅) is measurable, all responses are pointwise nonnegative, ∑x<ar(A,x,w)=1\sum_{x<a}r(A,x,w)=1∑x<a​r(A,x,w)=1, every w↦r(A,x,w)ρ(w)w\mapsto r(A,x,w)\rho(w)w↦r(A,x,w)ρ(w) is integrable, and

Tr⁡A[W(PxA⊗Ib)]=∫Ωr(A,x,w)ρ(w) dμ(w)\operatorname{Tr}_A[W(P_x^A\otimes I_b)] =\int_\Omega r(A,x,w)\rho(w)\,d\mu(w)TrA​[W(PxA​⊗Ib​)]=∫Ω​r(A,x,w)ρ(w)dμ(w)

for every A,xA,xA,x. Here a projective resolution is a family of aaa Hermitian idempotent a×aa\times aa×a matrices PxAP_x^APxA​ with pairwise zero products for distinct indices and sum IaI_aIa​, and (Tr⁡AZ)uv=∑i<aZ(i,u),(i,v)(\operatorname{Tr}_AZ)_{uv}=\sum_{i<a}Z_{(i,u),(i,v)}(TrA​Z)uv​=∑i<a​Z(i,u),(i,v)​. No density assumption on WWW is part of this definition. Zero dimensions are included: if a=0a=0a=0, response normalization is impossible, and if b=0b=0b=0, the hidden density matrices are impossible; hence the negated existence condition holds in either case.

QAlgorithms.Papers.Independent.Intakethird100.P62.assemblage

For any natural numbers a,ba,ba,b, any complex matrix WWW indexed by {0,…,a−1}×{0,…,b−1}\{0,\ldots,a-1\}\times\{0,\ldots,b-1\}{0,…,a−1}×{0,…,b−1}, any family PxP_xPx​ of aaa Hermitian idempotent a×aa\times aa×a matrices with PxPx′=0P_xP_{x'}=0Px​Px′​=0 for x≠x′x\ne x'x=x′ and ∑xPx=Ia\sum_xP_x=I_a∑x​Px​=Ia​, and any outcome x<ax<ax<a, the assemblage is the b×bb\times bb×b matrix Tr⁡A[W(Px⊗Ib)]\operatorname{Tr}_A[W(P_x\otimes I_b)]TrA​[W(Px​⊗Ib​)], where (Tr⁡AZ)uv=∑i<aZ(i,u),(i,v)(\operatorname{Tr}_AZ)_{uv}=\sum_{i<a}Z_{(i,u),(i,v)}(TrA​Z)uv​=∑i<a​Z(i,u),(i,v)​. Explicitly, its (u,v)(u,v)(u,v)-entry is ∑i,j<aW(i,u),(j,v)(Px)ji\sum_{i,j<a}W_{(i,u),(j,v)}(P_x)_{ji}∑i,j<a​W(i,u),(j,v)​(Px​)ji​. No positivity, Hermiticity, or trace normalization of WWW is assumed. When a=0a=0a=0, there is no possible outcome xxx; b=0b=0b=0 is allowed and gives an empty output matrix.

QAlgorithms.Papers.Independent.Intakethird100.P62.jointProbability

For any natural numbers a,ba,ba,b, any complex matrix WWW on the index set {0,…,a−1}×{0,…,b−1}\{0,\ldots,a-1\}\times\{0,\ldots,b-1\}{0,…,a−1}×{0,…,b−1}, projective resolutions P=(Px)x<aP=(P_x)_{x<a}P=(Px​)x<a​ and Q=(Qy)y<bQ=(Q_y)_{y<b}Q=(Qy​)y<b​, and outcomes x<a,y<bx<a,y<bx<a,y<b, the joint-probability expression is Re⁡tr⁡[(Px⊗Qy)W]\operatorname{Re}\operatorname{tr}[(P_x\otimes Q_y)W]Retr[(Px​⊗Qy​)W]. Here each resolution is a family of Hermitian idempotent matrices, with distinct members multiplying to zero and with sum equal to the identity; zero projectors are permitted. No density-matrix assumption is imposed on WWW, so this definition alone imposes no probability bounds on the resulting real number. If either dimension is zero, the corresponding outcome cannot be supplied.

QAlgorithms.Papers.Independent.Intakethird100.P62.lhs_correlation_equivalence

For every pair of natural numbers a,ba,ba,b with a>0a>0a>0 and b>0b>0b>0, every positive semidefinite matrix WWW of trace 111 on Ca⊗Cb\mathbb C^a\otimes\mathbb C^bCa⊗Cb, and every set MMM of projective measurements on Ca\mathbb C^aCa, the following two existence statements are equivalent. A projective measurement on Cd\mathbb C^dCd here consists of exactly ddd Hermitian idempotents (Px)x=0d−1(P_x)_{x=0}^{d-1}(Px​)x=0d−1​, with PxPy=0P_xP_y=0Px​Py​=0 for x≠yx\ne yx=y and ∑xPx=Id\sum_xP_x=I_d∑x​Px​=Id​; zero projectors are allowed. The first existence statement asks for a measurable space Ω\OmegaΩ, a probability measure μ\muμ, a measurable and μ\muμ-integrable family σω\sigma_\omegaσω​ of density matrices on Cb\mathbb C^bCb, and measurable real functions rA,x(ω)r_{A,x}(\omega)rA,x​(ω) for A∈MA\in MA∈M, with rA,x(ω)≥0r_{A,x}(\omega)\ge0rA,x​(ω)≥0 and ∑xrA,x(ω)=1\sum_xr_{A,x}(\omega)=1∑x​rA,x​(ω)=1 for every A,ωA,\omegaA,ω, such that every matrix-valued function rA,xσr_{A,x}\sigmarA,x​σ is integrable and

Tr⁡1 ⁣(W(Ax⊗Ib))=∫ΩrA,x(ω)σω dμ(ω)(A∈M, 0≤x<a),\operatorname{Tr}_1\!\bigl(W(A_x\otimes I_b)\bigr) =\int_\Omega r_{A,x}(\omega)\sigma_\omega\,d\mu(\omega) \quad(A\in M,\ 0\le x<a),Tr1​(W(Ax​⊗Ib​))=∫Ω​rA,x​(ω)σω​dμ(ω)(A∈M, 0≤x<a),

where (Tr⁡1Z)u,v=∑i=0a−1Z(i,u),(i,v)(\operatorname{Tr}_1 Z)_{u,v}=\sum_{i=0}^{a-1}Z_{(i,u),(i,v)}(Tr1​Z)u,v​=∑i=0a−1​Z(i,u),(i,v)​. The second existence statement asks for a measurable space, probability measure, density-matrix family, and response functions with those same measurability, integrability of the density-matrix family, nonnegativity, and normalization requirements, such that, for every A∈MA\in MA∈M, every x<ax<ax<a, every projective measurement BBB on Cb\mathbb C^bCb, and every y<by<by<b, the scalar function rA,x(ω)Re⁡tr⁡(Byσω)r_{A,x}(\omega)\operatorname{Re}\operatorname{tr}(B_y\sigma_\omega)rA,x​(ω)Retr(By​σω​) is integrable and

Re⁡tr⁡ ⁣((Ax⊗By)W)=∫ΩrA,x(ω)Re⁡tr⁡(Byσω) dμ(ω).\operatorname{Re}\operatorname{tr}\!\bigl((A_x\otimes B_y)W\bigr) =\int_\Omega r_{A,x}(\omega)\operatorname{Re}\operatorname{tr}(B_y\sigma_\omega)\,d\mu(\omega).Retr((Ax​⊗By​)W)=∫Ω​rA,x​(ω)Retr(By​σω​)dμ(ω).

The witnesses in the two existence statements need not be identical. The set MMM may be empty, in which case all requirements indexed by A∈MA\in MA∈M are vacuous.

QAlgorithms.Papers.Independent.Intakethird100.P62.oneProjective

This defines the projective resolution in dimension 111 whose sole projector is the 1×11\times11×1 identity matrix [1][1][1]. It is Hermitian and idempotent, and its sum is the identity. The requirement that projectors with distinct indices multiply to zero is vacuous because the outcome index set has only one member.

QAlgorithms.Papers.Independent.Intakethird100.P62.strict_hierarchy

For positive subsystem dimensions s,ts,ts,t, call a matrix a density matrix when it is positive semidefinite and has trace 111. A projective measurement on Cs\mathbb C^sCs means a family (Ax)x=0s−1(A_x)_{x=0}^{s-1}(Ax​)x=0s−1​ of Hermitian idempotent matrices satisfying AxAx′=0A_xA_{x'}=0Ax​Ax′​=0 for x≠x′x\ne x'x=x′ and ∑xAx=Is\sum_xA_x=I_s∑x​Ax​=Is​; zero projectors are allowed. Say that WWW has a projective hidden-state model if there exist a measurable space Ω\OmegaΩ, a probability measure μ\muμ, a measurable, μ\muμ-integrable family of density matrices σω\sigma_\omegaσω​ on Ct\mathbb C^tCt, and response functions rA,x(ω)r_{A,x}(\omega)rA,x​(ω), indexed by every such projective measurement AAA, which are measurable, nonnegative, and satisfy ∑xrA,x(ω)=1\sum_xr_{A,x}(\omega)=1∑x​rA,x​(ω)=1 for every A,ωA,\omegaA,ω, such that each rA,xσr_{A,x}\sigmarA,x​σ is integrable and

Tr⁡1 ⁣(W(Ax⊗It))=∫ΩrA,x(ω)σω dμ(ω).\operatorname{Tr}_1\!\bigl(W(A_x\otimes I_t)\bigr) =\int_\Omega r_{A,x}(\omega)\sigma_\omega\,d\mu(\omega).Tr1​(W(Ax​⊗It​))=∫Ω​rA,x​(ω)σω​dμ(ω).

Here (Tr⁡1Z)u,v=∑iZ(i,u),(i,v)(\operatorname{Tr}_1 Z)_{u,v}=\sum_iZ_{(i,u),(i,v)}(Tr1​Z)u,v​=∑i​Z(i,u),(i,v)​. “Steerable” means that no such model exists. Say that WWW is Bell local if there exist a measurable probability space and measurable, nonnegative response functions rA,x,sB,yr_{A,x},s_{B,y}rA,x​,sB,y​, normalized pointwise over their respective outcomes, for all projective measurements A,BA,BA,B on the two subsystems, such that every product rA,xsB,yr_{A,x}s_{B,y}rA,x​sB,y​ is integrable and

Re⁡tr⁡ ⁣((Ax⊗By)W)=∫ΩrA,x(ω)sB,y(ω) dμ(ω);\operatorname{Re}\operatorname{tr}\!\bigl((A_x\otimes B_y)W\bigr) =\int_\Omega r_{A,x}(\omega)s_{B,y}(\omega)\,d\mu(\omega);Retr((Ax​⊗By​)W)=∫Ω​rA,x​(ω)sB,y​(ω)dμ(ω);

“Bell nonlocal” means the absence of this model. Say that WWW is separable positive if W=∑j=0q−1wjρj⊗σjW=\sum_{j=0}^{q-1}w_j\rho_j\otimes\sigma_jW=∑j=0q−1​wj​ρj​⊗σj​ for some natural number qqq, nonnegative real weights wjw_jwj​, and density matrices ρj,σj\rho_j,\sigma_jρj​,σj​; this definition imposes no separate normalization on the weights and allows an empty sum. The declaration asserts the conjunction of three statements: for every a,b∈Na,b\in\mathbb Na,b∈N and every density matrix WWW on Ca+1⊗Cb+1\mathbb C^{a+1}\otimes\mathbb C^{b+1}Ca+1⊗Cb+1, Bell nonlocality implies steerability, and steerability implies that WWW is not separable positive; there exists a real η\etaη with 1/3<η≤1/21/3<\eta\le1/21/3<η≤1/2 for which Wη=1+η4I4−η2FW_\eta=\frac{1+\eta}{4}I_4-\frac{\eta}{2}FWη​=41+η​I4​−2η​F, where F(u⊗v)=v⊗uF(u\otimes v)=v\otimes uF(u⊗v)=v⊗u, is a density matrix that is not separable positive and is not steerable; and there exists a real η\etaη with 1/2<η≤11/2<\eta\le11/2<η≤1 for which this same formula gives a density matrix that is steerable and Bell local. The universal statement includes a=0a=0a=0 or b=0b=0b=0, corresponding to a one-dimensional subsystem.

QAlgorithms.Papers.Independent.Intakethird100.P62.wernerEta

For every natural number ddd and every real number η\etaη, set p=[d−1+(d+1)η]/(2d)p=[d-1+(d+1)\eta]/(2d)p=[d−1+(d+1)η]/(2d), with the arithmetic in this expression taking place in R\mathbb RR. Let FFF be the complex matrix indexed by pairs from {0,…,d−1}\{0,\ldots,d-1\}{0,…,d−1}, with F(i,j),(k,l)=1F_{(i,j),(k,l)}=1F(i,j),(k,l)​=1 when i=li=li=l and j=kj=kj=k, and 000 otherwise, and set P0=(I+F)/2P_0=(I+F)/2P0​=(I+F)/2, P1=I−P0P_1=I-P_0P1​=I−P0​. The defined matrix is

2(1−p)d(d+1)P0+2pd(d−1)P1.\frac{2(1-p)}{d(d+1)}P_0+\frac{2p}{d(d-1)}P_1.d(d+1)2(1−p)​P0​+d(d−1)2p​P1​.

The real coefficients are regarded as complex scalars, and d−1d-1d−1 in the displayed denominator is real subtraction. There is no restriction on η\etaη and no density-matrix requirement. All divisions use the total convention x/0=0x/0=0x/0=0; thus d=0d=0d=0 gives the empty matrix, and d=1d=1d=1 gives the 1×11\times11×1 matrix with entry 1−η1-\eta1−η.

Human review
  • Endorsed by Shuze Chen · Nov 1, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Gohan · Nov 1, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paper
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me