Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Definitions for Multiparty Quantum Mutual Information: An alternative definition

Definition
QAlgorithms_Papers_Independent_Support50_P43

by Goku · Oct 23, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

The objects the formalized statements of Multiparty Quantum Mutual Information: An alternative definition (arXiv:1504.07176) are stated in. Each was checked against the paper by a blind read-back and a moderator.

  • restrictLabel: Restrict a joint computational label to the selected registers (PDF p.3).
  • reducedDensity: Partial trace via the accepted register marginal, with a trivial external register attached and then removed (PDF pp.1–3). Both retained indices remain independent.
  • marginalEntropyBits: Entropy of the accepted marginal with its one-dimensional external factor (PDF pp.1,3,5). Attaching Unit does not change the marginal spectrum.
  • alternativeOn: Alternative information of the individual parties in s (PDF pp.2–3,(10),(13)). For a singleton this is S(empty), hence zero on density inputs.
  • conventionalOn: Conventional multipartite information, without a factor 1/2 (PDF p.1,(2)).
  • commonOn: Common information: alternating sum over every nonempty subset exactly once (PDF p.3, below (13)).
  • IsProductAcross: Density factors across a partition of the physical parties (PDF p.4, Theorem 5). This entrywise product is the tensor product after canonical regrouping.
  • cutProduct: Product of the two marginals at the cut s|complement(s), indexed on the original joint basis, so no replica-space identification is used (PDF p.3,(13), middle line).
  • singletonProduct: Tensor product of all single-party marginals on the original joint basis (PDF pp.1,3,(2),(13)).
  • HomBasis: Homogeneous n-party computational basis (PDF p.5, Table I).
  • dickeQubitVector: Reindex the accepted computational Dicke vector from Bool to Fin 2 (PDF p.5, Table I). The map sends 0 to false and 1 to true.
  • antisymmetric: Three-qutrit antisymmetric vector, labels 1,2,3 renamed 0,1,2 (PDF pp.4–5). The three even permutations are positive.
  • cluster: Four-qubit cluster vector with signs +,+,+,- (PDF p.5, Table I).
  • informationPair: The pair (common information, alternative information), in bits (PDF p.5).
  • noisyQubits: White-noise family from Figures 2(a,b) and 3, on three QUBITS (PDF pp.4–5); 0≤p≤1 and psi=GHZ3 or W3 in the statements.
  • partialTransposeA: Partial transpose on A in the computational basis, keeping B,C fixed (PDF p.4).
  • logarithmicNegativity: Logarithmic negativity E_N(A:BC), not (trace norm - 1)/2 (PDF p.4, Fig.2).
  • LocalBases: Independent choice of one local orthonormal qubit basis at each party (PDF p.5, (16)–(18)).
  • measurementProjector: Product rank-one outcome projector on measured parties s, identity elsewhere (PDF p.5). Outcomes have NO coordinates for unmeasured parties.
  • postMeasurement: Nonselective local projective measurement (PDF p.5, below (18)).
  • discordScore: Choice of information functional for the two panels of Figure 3 (PDF p.5).
  • discord: Optimized information loss, with one joint optimization over independent local bases on s (PDF p.5,(16)–(18)). For density inputs the value set is nonempty, bounded, and compact, so sSup equals the printed maximum.
Definition code
import Definitions.Def_QAlgorithms_Defs_Ensemble
import Definitions.Def_QAlgorithms_Defs_MatrixDistances
import Definitions.Def_QAlgorithms_Papers_Independent_Claimed50_P33
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P21
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P22
import Definitions.Def_QAlgorithms_Papers_Independent_Information100_P60

/-!
arXiv:1504.07176v2, finite-dimensional statements, entropies in bits.
The source's tensor-power relative-entropy identities and unrestricted CP
monotonicity are deferred, not assumptions. Compilation is pending batch handoff.
-/

noncomputable section

namespace QAlgorithms.Papers.Independent.Support50.P43

open scoped Matrix ComplexConjugate ComplexOrder

attribute [local instance] Classical.propDecidable Classical.decEq

-- These accepted register types are definitionally the required finite tensor bases.
open QAlgorithms.Papers.Independent.Claimed50.P33 (RegisterBasis SubBasis)

open QAlgorithms.Papers.Independent.Information100.P21 (ghzKet)

/-- Restrict a joint computational label to the selected registers (PDF p.3). -/
def restrictLabel {n : ℕ} {D : Fin n → Type} (s : Finset (Fin n))
    (x : RegisterBasis D) : SubBasis D s := fun i => x i.val

/-- Partial trace via the accepted register marginal, with a trivial external register
attached and then removed (PDF pp.1–3). Both retained indices remain independent. -/
def reducedDensity {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) (s : Finset (Fin n)) :
    Matrix (SubBasis D s) (SubBasis D s) ℂ :=
  fun a b => QAlgorithms.Papers.Independent.Claimed50.P33.registerMarginal
    (QAlgorithms.Papers.Independent.Claimed50.P33.trivialExtension rho) s (a, ()) (b, ())

/-- Entropy of the accepted marginal with its one-dimensional external factor
(PDF pp.1,3,5). Attaching Unit does not change the marginal spectrum. -/
def marginalEntropyBits {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) (s : Finset (Fin n)) : ℝ :=
  QAlgorithms.Papers.Independent.Claimed50.P33.entropyAt
    (QAlgorithms.Papers.Independent.Claimed50.P33.trivialExtension rho) s

/-- Alternative information of the individual parties in s (PDF pp.2–3,(10),(13)).
For a singleton this is S(empty), hence zero on density inputs. -/
def alternativeOn {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) (s : Finset (Fin n)) : ℝ :=
  (∑ i ∈ s, marginalEntropyBits rho (s.erase i)) - (s.card - 1 : ℕ) * marginalEntropyBits rho s

/-- Conventional multipartite information, without a factor 1/2 (PDF p.1,(2)). -/
def conventionalOn {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) (s : Finset (Fin n)) : ℝ :=
  (∑ i ∈ s, marginalEntropyBits rho {i}) - marginalEntropyBits rho s

/-- Common information: alternating sum over every nonempty subset exactly once
(PDF p.3, below (13)). -/
def commonOn {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) (s : Finset (Fin n)) : ℝ :=
  ∑ t ∈ s.powerset.filter (fun t => t.Nonempty),
    (-1 : ℝ) ^ (t.card + 1) * marginalEntropyBits rho t

/-- Density factors across a partition of the physical parties (PDF p.4, Theorem 5).
This entrywise product is the tensor product after canonical regrouping. -/
def IsProductAcross {n m : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) (blocks : Fin m → Finset (Fin n)) : Prop :=
  ∀ a b, rho a b = ∏ j, reducedDensity rho (blocks j)
    (restrictLabel (blocks j) a) (restrictLabel (blocks j) b)

/-- Product of the two marginals at the cut s|complement(s), indexed on the original
joint basis, so no replica-space identification is used (PDF p.3,(13), middle line). -/
def cutProduct {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) (s : Finset (Fin n)) :
    Matrix (RegisterBasis D) (RegisterBasis D) ℂ :=
  fun a b => reducedDensity rho s (restrictLabel s a) (restrictLabel s b) *
    reducedDensity rho sᶜ (restrictLabel sᶜ a) (restrictLabel sᶜ b)

/-- Tensor product of all single-party marginals on the original joint basis
(PDF pp.1,3,(2),(13)). -/
def singletonProduct {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) : Matrix (RegisterBasis D) (RegisterBasis D) ℂ :=
  fun a b => ∏ i, reducedDensity rho {i} (restrictLabel {i} a) (restrictLabel {i} b)

/-- Homogeneous n-party computational basis (PDF p.5, Table I). -/
abbrev HomBasis (n d : ℕ) := Fin n → Fin d

/-- Reindex the accepted computational Dicke vector from Bool to Fin 2
(PDF p.5, Table I). The map sends 0 to false and 1 to true. -/
def dickeQubitVector (n r : ℕ) : EuclideanSpace ℂ (HomBasis n 2) :=
  WithLp.toLp 2 fun x =>
    QAlgorithms.Papers.Independent.Information100.P22.dicke (I := Fin n) r
      (fun i => x i == 1)

/-- Three-qutrit antisymmetric vector, labels 1,2,3 renamed 0,1,2
(PDF pp.4–5). The three even permutations are positive. -/
def antisymmetric : EuclideanSpace ℂ (HomBasis 3 3) :=
  WithLp.toLp 2 fun x =>
    if x = ![0,1,2] ∨ x = ![1,2,0] ∨ x = ![2,0,1] then
      ((Real.sqrt 6)⁻¹ : ℂ)
    else if x = ![0,2,1] ∨ x = ![1,0,2] ∨ x = ![2,1,0] then
      -((Real.sqrt 6)⁻¹ : ℂ) else 0

/-- Four-qubit cluster vector with signs +,+,+,- (PDF p.5, Table I). -/
def cluster : EuclideanSpace ℂ (HomBasis 4 2) :=
  WithLp.toLp 2 fun x =>
    if x = ![0,0,0,0] ∨ x = ![0,0,1,1] ∨ x = ![1,1,0,0] then 1/2
    else if x = ![1,1,1,1] then -(1/2) else 0

/-- The pair (common information, alternative information), in bits (PDF p.5). -/
def informationPair {n d : ℕ} (psi : EuclideanSpace ℂ (HomBasis n d)) : ℝ × ℝ :=
  (commonOn (pureDensity psi) Finset.univ, alternativeOn (pureDensity psi) Finset.univ)

/-- White-noise family from Figures 2(a,b) and 3, on three QUBITS
(PDF pp.4–5); 0≤p≤1 and psi=GHZ3 or W3 in the statements. -/
def noisyQubits (psi : EuclideanSpace ℂ (HomBasis 3 2)) (p : ℝ) :
    Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ :=
  (p : ℂ) • pureDensity psi + (((1-p)/8 : ℝ) : ℂ) • 1

/-- Partial transpose on A in the computational basis, keeping B,C fixed (PDF p.4). -/
def partialTransposeA (rho : Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ) :
    Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ :=
  fun a b => rho (Function.update a 0 (b 0)) (Function.update b 0 (a 0))

/-- Logarithmic negativity E_N(A:BC), not (trace norm - 1)/2 (PDF p.4, Fig.2). -/
def logarithmicNegativity (rho : Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ) : ℝ :=
  Real.logb 2 (traceNorm (partialTransposeA rho))

/-- Independent choice of one local orthonormal qubit basis at each party
(PDF p.5, (16)–(18)). -/
abbrev LocalBases := Fin 3 → Matrix.unitaryGroup (Fin 2) ℂ

/-- Product rank-one outcome projector on measured parties s, identity elsewhere
(PDF p.5). Outcomes have NO coordinates for unmeasured parties. -/
def measurementProjector (U : LocalBases) (s : Finset (Fin 3))
    (z : SubBasis (fun _ : Fin 3 => Fin 2) s) :
    Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ :=
  fun a b => ∏ i, if hi : i ∈ s then
    (U i : Matrix (Fin 2) (Fin 2) ℂ) (a i) (z ⟨i,hi⟩) *
      conj ((U i : Matrix (Fin 2) (Fin 2) ℂ) (b i) (z ⟨i,hi⟩))
    else if a i = b i then 1 else 0

/-- Nonselective local projective measurement (PDF p.5, below (18)). -/
def postMeasurement (U : LocalBases) (s : Finset (Fin 3))
    (rho : Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ) :
    Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ :=
  ∑ z : SubBasis (fun _ : Fin 3 => Fin 2) s,
    measurementProjector U s z * rho * measurementProjector U s z

/-- Choice of information functional for the two panels of Figure 3 (PDF p.5). -/
def discordScore (alternative : Bool)
    (rho : Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ) : ℝ :=
  if alternative then alternativeOn rho Finset.univ else conventionalOn rho Finset.univ

/-- Optimized information loss, with one joint optimization over independent local
bases on s (PDF p.5,(16)–(18)). For density inputs the value set is nonempty,
bounded, and compact, so sSup equals the printed maximum. -/
def discord (alternative : Bool) (s : Finset (Fin 3))
    (rho : Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ) : ℝ :=
  discordScore alternative rho - sSup {v : ℝ | ∃ U : LocalBases,
    v = discordScore alternative (postMeasurement U s rho)}

/-! Small sanity checks for the constructed adapters. -/

example {n : ℕ} {D : Fin n → Type} (s : Finset (Fin n)) (x : RegisterBasis D)
    (i : {i : Fin n // i ∈ s}) : restrictLabel s x i = x i.val := rfl

example {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (s : Finset (Fin n)) : reducedDensity (0 : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) s = 0 := by
  ext a b
  simp [reducedDensity, QAlgorithms.Papers.Independent.Claimed50.P33.registerMarginal,
    QAlgorithms.Papers.Independent.Claimed50.P33.trivialExtension]

example {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) (s : Finset (Fin n)) :
    marginalEntropyBits rho s = QAlgorithms.Papers.Independent.Claimed50.P33.entropyAt
      (QAlgorithms.Papers.Independent.Claimed50.P33.trivialExtension rho) s := rfl

example {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) : alternativeOn rho ∅ = 0 := by
  simp [alternativeOn]

example {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) : commonOn rho ∅ = 0 := by
  unfold commonOn
  apply Finset.sum_eq_zero
  intro t ht
  have hsubset : t ⊆ (∅ : Finset (Fin n)) :=
    Finset.mem_powerset.mp (Finset.mem_filter.mp ht).1
  obtain ⟨i, hi⟩ := (Finset.mem_filter.mp ht).2
  have hfalse : False := by simpa using hsubset hi
  exact hfalse.elim

example (x : HomBasis 3 2) : dickeQubitVector 3 1 x =
    QAlgorithms.Papers.Independent.Information100.P22.dicke (I := Fin 3) 1
      (fun i => x i == 1) := rfl

example : antisymmetric ![0,1,2] = ((Real.sqrt 6)⁻¹ : ℂ) := by
  simp [antisymmetric]

example : cluster ![1,1,1,1] = -(1/2 : ℂ) := by
  have h0 : (![1,1,1,1] : HomBasis 4 2) ≠ ![0,0,0,0] := by
    intro h
    have hh := congrFun h 0
    change (1 : Fin 2) = 0 at hh
    exact (by norm_num : (1 : Fin 2) ≠ 0) hh
  have h1 : (![1,1,1,1] : HomBasis 4 2) ≠ ![0,0,1,1] := by
    intro h
    have hh := congrFun h 0
    change (1 : Fin 2) = 0 at hh
    exact (by norm_num : (1 : Fin 2) ≠ 0) hh
  have h2 : (![1,1,1,1] : HomBasis 4 2) ≠ ![1,1,0,0] := by
    intro h
    have hh := congrFun h 2
    change (1 : Fin 2) = 0 at hh
    exact (by norm_num : (1 : Fin 2) ≠ 0) hh
  simp [cluster, h0, h1, h2]

example (psi : EuclideanSpace ℂ (HomBasis 3 2)) : noisyQubits psi 1 = pureDensity psi := by
  simp [noisyQubits]

example : partialTransposeA 0 = 0 := rfl

example (U : LocalBases) (s : Finset (Fin 3)) : postMeasurement U s 0 = 0 := by
  simp [postMeasurement]

example (U : LocalBases) (z : SubBasis (fun _ : Fin 3 => Fin 2) ∅)
    (a : HomBasis 3 2) : measurementProjector U ∅ z a a = 1 := by
  simp [measurementProjector]

example (rho : Matrix (RegisterBasis (fun _ : Fin 0 => Unit))
    (RegisterBasis (fun _ : Fin 0 => Unit)) ℂ) (a b : RegisterBasis (fun _ : Fin 0 => Unit)) :
    singletonProduct rho a b = 1 := by
  simp [singletonProduct]

example (rho : Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ) :
    discordScore true rho = alternativeOn rho Finset.univ := rfl

example {n : ℕ} {D : Fin n → Type} [∀ i, Fintype (D i)]
    (rho : Matrix (RegisterBasis D) (RegisterBasis D) ℂ) :
    conventionalOn rho ∅ = -marginalEntropyBits rho ∅ := by
  simp [conventionalOn]

example (alternative : Bool) (s : Finset (Fin 3))
    (rho : Matrix (HomBasis 3 2) (HomBasis 3 2) ℂ) :
    discord alternative s rho = discordScore alternative rho -
      sSup {v : ℝ | ∃ U : LocalBases,
        v = discordScore alternative (postMeasurement U s rho)} := rfl

end QAlgorithms.Papers.Independent.Support50.P43
Source
Multiparty Quantum Mutual Information: An alternative definition, arXiv:1504.07176v2, https://arxiv.org/abs/1504.07176v2
Read-back

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

QAlgorithms.Papers.Independent.Support50.P43.HomBasis

For every pair of natural numbers n,dn,dn,d, this abbreviates the type of functions from {0,…,n−1}\{0,\ldots,n-1\}{0,…,n−1} to {0,…,d−1}\{0,\ldots,d-1\}{0,…,d−1}, equivalently length-nnn tuples with entries in that ddd-element set. For n=0n=0n=0 it contains exactly the empty tuple, including when d=0d=0d=0; for n>0n>0n>0 and d=0d=0d=0 it is empty.

QAlgorithms.Papers.Independent.Support50.P43.IsProductAcross

For every pair of natural numbers n,mn,mn,m, every family of finite types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, every complex matrix ρ\rhoρ indexed by X=∏iDiX=\prod_iD_iX=∏i​Di​, and every indexed family of subsets (Bj)j=0m−1(B_j)_{j=0}^{m-1}(Bj​)j=0m−1​ of {0,…,n−1}\{0,\ldots,n-1\}{0,…,n−1}, this defines the proposition

∀a,b∈X,ρ(a,b)=∏j=0m−1RBj(a∣Bj,b∣Bj),\forall a,b\in X,\qquad \rho(a,b)=\prod_{j=0}^{m-1}R_{B_j}(a|_{B_j},b|_{B_j}),∀a,b∈X,ρ(a,b)=j=0∏m−1​RBj​​(a∣Bj​​,b∣Bj​​),

where restriction retains precisely the indicated coordinates and

Rt(u,v)=∑x,y∈Xx∣t=u,y∣t=vxi=yi for every i∉tρ(x,y).R_t(u,v)= \sum_{\substack{x,y\in X\\x|_t=u,\;y|_t=v\\ x_i=y_i\text{ for every }i\notin t}}\rho(x,y).Rt​(u,v)=x,y∈Xx∣t​=u,y∣t​=vxi​=yi​ for every i∈/t​∑​ρ(x,y).

The subsets BjB_jBj​ are not required to be disjoint, nonempty, distinct, or to cover all coordinates, and ρ\rhoρ is not assumed to be a density matrix. The finite types DiD_iDi​ may be empty; if XXX is empty, the proposition is vacuous. When m=0m=0m=0, the product is 111, so the proposition requires every entry of ρ\rhoρ to equal 111. Natural number n=0n=0n=0 is also allowed.

QAlgorithms.Papers.Independent.Support50.P43.LocalBases

This abbreviates the type of assignments i↦Uii\mapsto U_ii↦Ui​, for i∈{0,1,2}i\in\{0,1,2\}i∈{0,1,2}, where each UiU_iUi​ is a complex 2×22\times22×2 unitary matrix, so Ui†Ui=UiUi†=I2U_i^\dagger U_i=U_iU_i^\dagger=I_2Ui†​Ui​=Ui​Ui†​=I2​.

QAlgorithms.Papers.Independent.Support50.P43.alternativeOn

For every natural number nnn, every family of finite types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, every complex matrix ρ\rhoρ indexed by X=∏iDiX=\prod_iD_iX=∏i​Di​, and every subset sss of the coordinate indices, this defines

∑i∈sHρ(s∖{i})−max⁡(∣s∣−1,0) Hρ(s).\sum_{i\in s}H_\rho(s\setminus\{i\}) -\max(|s|-1,0)\,H_\rho(s).i∈s∑​Hρ​(s∖{i})−max(∣s∣−1,0)Hρ​(s).

Here Hρ(t)=−Re⁡tr⁡(f(Rt))H_\rho(t)=-\operatorname{Re}\operatorname{tr}(f(R_t))Hρ​(t)=−Retr(f(Rt​)), with f(u)=ulog⁡2uf(u)=u\log_2uf(u)=ulog2​u evaluated by real continuous functional calculus, and

Rt(a,b)=∑x,y∈Xx∣t=a,y∣t=bxi=yi for every i∉tρ(x,y)R_t(a,b)= \sum_{\substack{x,y\in X\\x|_t=a,\;y|_t=b\\ x_i=y_i\text{ for every }i\notin t}}\rho(x,y)Rt​(a,b)=x,y∈Xx∣t​=a,y∣t​=bxi​=yi​ for every i∈/t​∑​ρ(x,y)

for labels a,b∈∏i∈tDia,b\in\prod_{i\in t}D_ia,b∈∏i∈t​Di​. No positivity or normalization of ρ\rhoρ, or nonemptiness of the coordinate types, is assumed. The subtraction in ∣s∣−1|s|-1∣s∣−1 is natural-number subtraction: the value is 000 when s=∅s=\varnothings=∅, and is Hρ(∅)H_\rho(\varnothing)Hρ​(∅) when sss is a singleton. Natural number n=0n=0n=0 is allowed.

QAlgorithms.Papers.Independent.Support50.P43.antisymmetric

This defines a complex Euclidean vector indexed by triples x∈{0,1,2}3x\in\{0,1,2\}^3x∈{0,1,2}3. Its amplitude is 1/61/\sqrt61/6​ at (0,1,2)(0,1,2)(0,1,2), (1,2,0)(1,2,0)(1,2,0), and (2,0,1)(2,0,1)(2,0,1); it is −1/6-1/\sqrt6−1/6​ at (0,2,1)(0,2,1)(0,2,1), (1,0,2)(1,0,2)(1,0,2), and (2,1,0)(2,1,0)(2,1,0); and it is zero at every other triple, including every triple with a repeated coordinate value.

QAlgorithms.Papers.Independent.Support50.P43.cluster

This defines a complex vector indexed by four-tuples x∈{0,1}4x\in\{0,1\}^4x∈{0,1}4, with amplitudes 1/21/21/2 at 000000000000, 001100110011, and 110011001100, amplitude −1/2-1/2−1/2 at 111111111111, and amplitude 000 elsewhere; equivalently, it is 12(∣0000⟩+∣0011⟩+∣1100⟩−∣1111⟩)\frac12(|0000\rangle+|0011\rangle+|1100\rangle-|1111\rangle)21​(∣0000⟩+∣0011⟩+∣1100⟩−∣1111⟩), where ∣x⟩|x\rangle∣x⟩ denotes the coordinate vector supported at xxx.

QAlgorithms.Papers.Independent.Support50.P43.commonOn

For every natural number nnn, every family of finite types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, every complex matrix ρ\rhoρ indexed by X=∏iDiX=\prod_iD_iX=∏i​Di​, and every subset sss of the coordinate indices, this defines

∑∅≠t⊆s(−1)∣t∣+1Hρ(t).\sum_{\varnothing\ne t\subseteq s}(-1)^{|t|+1}H_\rho(t).∅=t⊆s∑​(−1)∣t∣+1Hρ​(t).

For each such ttt, Hρ(t)=−Re⁡tr⁡(f(Rt))H_\rho(t)=-\operatorname{Re}\operatorname{tr}(f(R_t))Hρ​(t)=−Retr(f(Rt​)), using real continuous functional calculus for f(u)=ulog⁡2uf(u)=u\log_2uf(u)=ulog2​u, and

Rt(a,b)=∑x,y∈Xx∣t=a,y∣t=bxi=yi for every i∉tρ(x,y),a,b∈∏i∈tDi.R_t(a,b)= \sum_{\substack{x,y\in X\\x|_t=a,\;y|_t=b\\ x_i=y_i\text{ for every }i\notin t}}\rho(x,y), \qquad a,b\in\prod_{i\in t}D_i.Rt​(a,b)=x,y∈Xx∣t​=a,y∣t​=bxi​=yi​ for every i∈/t​∑​ρ(x,y),a,b∈i∈t∏​Di​.

No positivity or trace normalization is required of ρ\rhoρ, and the finite coordinate types need not be nonempty. For s=∅s=\varnothings=∅, the sum is empty and the value is 000; this includes n=0n=0n=0.

QAlgorithms.Papers.Independent.Support50.P43.conventionalOn

For every natural number nnn, every family of finite types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, every complex matrix ρ\rhoρ indexed by X=∏iDiX=\prod_iD_iX=∏i​Di​, and every subset sss of the coordinate indices, this defines

∑i∈sHρ({i})−Hρ(s).\sum_{i\in s}H_\rho(\{i\})-H_\rho(s).i∈s∑​Hρ​({i})−Hρ​(s).

Here Hρ(t)=−Re⁡tr⁡(f(Rt))H_\rho(t)=-\operatorname{Re}\operatorname{tr}(f(R_t))Hρ​(t)=−Retr(f(Rt​)), where f(u)=ulog⁡2uf(u)=u\log_2uf(u)=ulog2​u is evaluated by real continuous functional calculus and the matrix indexed by ∏i∈tDi\prod_{i\in t}D_i∏i∈t​Di​ has entries

Rt(a,b)=∑x,y∈Xx∣t=a,y∣t=bxi=yi for every i∉tρ(x,y).R_t(a,b)= \sum_{\substack{x,y\in X\\x|_t=a,\;y|_t=b\\ x_i=y_i\text{ for every }i\notin t}}\rho(x,y).Rt​(a,b)=x,y∈Xx∣t​=a,y∣t​=bxi​=yi​ for every i∈/t​∑​ρ(x,y).

There is no density matrix assumption on ρ\rhoρ, and the finite coordinate types may be empty. Empty sss gives −Hρ(∅)-H_\rho(\varnothing)−Hρ​(∅), with R∅R_\varnothingR∅​ the one-by-one matrix having entry tr⁡ρ\operatorname{tr}\rhotrρ; singleton sss gives 000. The definition includes n=0n=0n=0.

QAlgorithms.Papers.Independent.Support50.P43.cutProduct

For every natural number nnn, every family of finite types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, every complex matrix ρ\rhoρ indexed by X=∏iDiX=\prod_iD_iX=∏i​Di​, and every subset sss of the coordinate indices, this defines the complex matrix CCC on XXX by

C(a,b)=Rs(a∣s,b∣s) Rsc(a∣sc,b∣sc),C(a,b)=R_s(a|_s,b|_s)\,R_{s^c}(a|_{s^c},b|_{s^c}),C(a,b)=Rs​(a∣s​,b∣s​)Rsc​(a∣sc​,b∣sc​),

where scs^csc is the complement among all nnn coordinates, restriction retains the designated coordinates, and

Rt(u,v)=∑x,y∈Xx∣t=u,y∣t=vxi=yi for every i∉tρ(x,y).R_t(u,v)= \sum_{\substack{x,y\in X\\x|_t=u,\;y|_t=v\\ x_i=y_i\text{ for every }i\notin t}}\rho(x,y).Rt​(u,v)=x,y∈Xx∣t​=u,y∣t​=vxi​=yi​ for every i∈/t​∑​ρ(x,y).

No positivity or normalization assumption is imposed on ρ\rhoρ, and the finite coordinate types may be empty. Empty sss and full sss are included; in either case C=(tr⁡ρ)ρC=(\operatorname{tr}\rho)\rhoC=(trρ)ρ. When n=0n=0n=0, ρ\rhoρ has a single entry and CCC has the square of that entry.

QAlgorithms.Papers.Independent.Support50.P43.dickeQubitVector

For every pair of natural numbers n,rn,rn,r, this defines a complex Euclidean vector indexed by tuples x∈{0,1}nx\in\{0,1\}^nx∈{0,1}n, with amplitude

ψn,r(x)={1(nr),∣{i∈{0,…,n−1}:xi=1}∣=r,0,otherwise.\psi_{n,r}(x)= \begin{cases} \displaystyle\frac{1}{\sqrt{\binom nr}},& \bigl|\{i\in\{0,\ldots,n-1\}:x_i=1\}\bigr|=r,\\[4pt] 0,&\text{otherwise}. \end{cases}ψn,r​(x)=⎩⎨⎧​(rn​)​1​,0,​​{i∈{0,…,n−1}:xi​=1}​=r,otherwise.​

The coefficient is a real number embedded in C\mathbb CC. There is no restriction r≤nr\le nr≤n: for r>nr>nr>n, every amplitude is zero. The inverse of zero is defined as zero. For n=0,r=0n=0,r=0n=0,r=0, the unique empty tuple has amplitude 111; for n=0,r>0n=0,r>0n=0,r>0, the vector is zero.

QAlgorithms.Papers.Independent.Support50.P43.discord

For any Boolean parameter bbb, subset S⊆{0,1,2}S\subseteq\{0,1,2\}S⊆{0,1,2}, and complex matrix ρ\rhoρ indexed by three-bit configurations, this defines

Fb(ρ)−sup⁡U0,U1,U2 unitaryFb ⁣(∑z:S→{0,1}PU,S,zρPU,S,z).F_b(\rho)-\sup_{U_0,U_1,U_2\ \mathrm{unitary}} F_b\!\left(\sum_{z:S\to\{0,1\}}P_{U,S,z}\rho P_{U,S,z}\right).Fb​(ρ)−U0​,U1​,U2​ unitarysup​Fb​​z:S→{0,1}∑​PU,S,z​ρPU,S,z​​.

The entries of PU,S,zP_{U,S,z}PU,S,z​ are ∏i∈S(Ui)ai,zi(Ui)bi,zi‾∏i∉Sδaibi\prod_{i\in S}(U_i)_{a_i,z_i}\overline{(U_i)_{b_i,z_i}}\prod_{i\notin S}\delta_{a_i b_i}∏i∈S​(Ui​)ai​,zi​​(Ui​)bi​,zi​​​∏i∈/S​δai​bi​​. Writing ρT\rho_TρT​ for the marginal obtained by summing over matching complementary labels and Hρ(T)=−Re⁡tr⁡f(ρT)H_\rho(T)=-\operatorname{Re}\operatorname{tr}f(\rho_T)Hρ​(T)=−Retrf(ρT​) with f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t applied through continuous functional calculus, the score is Ftrue(ρ)=∑i=02Hρ({0,1,2}∖{i})−2Hρ({0,1,2})F_{\mathrm{true}}(\rho)=\sum_{i=0}^2H_\rho(\{0,1,2\}\setminus\{i\})-2H_\rho(\{0,1,2\})Ftrue​(ρ)=∑i=02​Hρ​({0,1,2}∖{i})−2Hρ​({0,1,2}) and Ffalse(ρ)=∑i=02Hρ({i})−Hρ({0,1,2})F_{\mathrm{false}}(\rho)=\sum_{i=0}^2H_\rho(\{i\})-H_\rho(\{0,1,2\})Ffalse​(ρ)=∑i=02​Hρ​({i})−Hρ​({0,1,2}). The supremum is the real-valued supremum operation, and no maximizing basis is asserted. No density-matrix hypothesis is imposed; the empty subset is allowed and yields zero.

QAlgorithms.Papers.Independent.Support50.P43.discordScore

For any Boolean parameter bbb and any complex matrix ρ\rhoρ indexed by three-bit configurations, this defines

Fb(ρ)={Hρ({1,2})+Hρ({0,2})+Hρ({0,1})−2Hρ({0,1,2}),b=true,Hρ({0})+Hρ({1})+Hρ({2})−Hρ({0,1,2}),b=false.F_b(\rho)= \begin{cases} H_\rho(\{1,2\})+H_\rho(\{0,2\})+H_\rho(\{0,1\})-2H_\rho(\{0,1,2\}),&b=\mathrm{true},\\ H_\rho(\{0\})+H_\rho(\{1\})+H_\rho(\{2\})-H_\rho(\{0,1,2\}),&b=\mathrm{false}. \end{cases}Fb​(ρ)={Hρ​({1,2})+Hρ​({0,2})+Hρ​({0,1})−2Hρ​({0,1,2}),Hρ​({0})+Hρ​({1})+Hρ​({2})−Hρ​({0,1,2}),​b=true,b=false.​

Here ρS\rho_SρS​ is the matrix obtained by retaining coordinates in SSS and summing over equal row and column labels on the complement, and Hρ(S)=−Re⁡tr⁡f(ρS)H_\rho(S)=-\operatorname{Re}\operatorname{tr}f(\rho_S)Hρ​(S)=−Retrf(ρS​), where f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t is applied through continuous functional calculus. No density-matrix hypothesis is required.

QAlgorithms.Papers.Independent.Support50.P43.informationPair

For any natural numbers n,dn,dn,d, including zero, and any complex Euclidean vector ψ\psiψ indexed by configurations x:{0,…,n−1}→{0,…,d−1}x:\{0,\ldots,n-1\}\to\{0,\ldots,d-1\}x:{0,…,n−1}→{0,…,d−1}, this defines the ordered pair

(∑∅≠S⊆I(−1)∣S∣+1Hψ(S),∑i∈IHψ(I∖{i})−max⁡(n−1,0) Hψ(I)),I={0,…,n−1}.\left( \sum_{\varnothing\ne S\subseteq I}(-1)^{|S|+1}H_\psi(S),\quad \sum_{i\in I}H_\psi(I\setminus\{i\})-\max(n-1,0)\,H_\psi(I) \right), \qquad I=\{0,\ldots,n-1\}.​∅=S⊆I∑​(−1)∣S∣+1Hψ​(S),i∈I∑​Hψ​(I∖{i})−max(n−1,0)Hψ​(I)​,I={0,…,n−1}.

Here ρxy=ψxψy‾\rho_{xy}=\psi_x\overline{\psi_y}ρxy​=ψx​ψy​​, ρS\rho_SρS​ is obtained from ρ\rhoρ by retaining the coordinates in SSS and summing over matching row and column configurations on its complement, and Hψ(S)=−Re⁡tr⁡f(ρS)H_\psi(S)=-\operatorname{Re}\operatorname{tr}f(\rho_S)Hψ​(S)=−Retrf(ρS​), with f(t)=tlog⁡2tf(t)=t\log_2tf(t)=tlog2​t applied through continuous functional calculus. No normalization of ψ\psiψ or positivity of n,dn,dn,d is assumed. The empty-register marginal is the scalar matrix [∥ψ∥2][\|\psi\|^2][∥ψ∥2]; when n=0n=0n=0, both components of the defined pair are zero.

QAlgorithms.Papers.Independent.Support50.P43.logarithmicNegativity

For any complex matrix ρ\rhoρ indexed by three-bit configurations, this defines

log⁡2 ⁣(Re⁡tr⁡B†B),B(a0,a1,a2),(b0,b1,b2)=ρ(b0,a1,a2),(a0,b1,b2),\log_2\!\left(\operatorname{Re}\operatorname{tr}\sqrt{B^\dagger B}\right), \qquad B_{(a_0,a_1,a_2),(b_0,b_1,b_2)} =\rho_{(b_0,a_1,a_2),(a_0,b_1,b_2)},log2​(RetrB†B​),B(a0​,a1​,a2​),(b0​,b1​,b2​)​=ρ(b0​,a1​,a2​),(a0​,b1​,b2​)​,

where the matrix square root is defined through continuous functional calculus. No normalization, positivity, or Hermiticity hypothesis is required. The real logarithm used here is total, with log⁡20=0\log_2 0=0log2​0=0, so a zero trace norm produces the value zero.

QAlgorithms.Papers.Independent.Support50.P43.marginalEntropyBits

For every natural number nnn, every family of finite types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, every complex matrix ρ\rhoρ indexed by X=∏i=0n−1DiX=\prod_{i=0}^{n-1}D_iX=∏i=0n−1​Di​, and every subset s⊆{0,…,n−1}s\subseteq\{0,\ldots,n-1\}s⊆{0,…,n−1}, define the matrix

Rs(a,b)=∑x,y∈Xx∣s=a,y∣s=bxi=yi for every i∉sρ(x,y),a,b∈∏i∈sDi.R_s(a,b)= \sum_{\substack{x,y\in X\\ x|_s=a,\;y|_s=b\\ x_i=y_i\text{ for every }i\notin s}}\rho(x,y), \qquad a,b\in\prod_{i\in s}D_i.Rs​(a,b)=x,y∈Xx∣s​=a,y∣s​=bxi​=yi​ for every i∈/s​∑​ρ(x,y),a,b∈i∈s∏​Di​.

The declared real number is

Hρ(s)=−Re⁡tr⁡(f(Rs)),f(t)=tlog⁡2t,H_\rho(s)=-\operatorname{Re}\operatorname{tr}\bigl(f(R_s)\bigr), \qquad f(t)=t\log_2 t,Hρ​(s)=−Retr(f(Rs​)),f(t)=tlog2​t,

where f(Rs)f(R_s)f(Rs​) is evaluated using the real continuous functional calculus. No density matrix assumption is imposed on ρ\rhoρ, and the finite coordinate types may be empty. Empty sss and n=0n=0n=0 are included: for empty sss, the matrix used is the one-by-one matrix with entry tr⁡ρ\operatorname{tr}\rhotrρ, so its entropy is not stipulated to be zero.

QAlgorithms.Papers.Independent.Support50.P43.measurementProjector

For any three complex 2×22\times22×2 unitary matrices U0,U1,U2U_0,U_1,U_2U0​,U1​,U2​, any subset S⊆{0,1,2}S\subseteq\{0,1,2\}S⊆{0,1,2}, and any assignment z:S→{0,1}z:S\to\{0,1\}z:S→{0,1}, this defines the matrix PU,S,zP_{U,S,z}PU,S,z​ indexed by a,b∈{0,1}3a,b\in\{0,1\}^3a,b∈{0,1}3 through

(PU,S,z)ab=∏i=02{(Ui)ai,zi (Ui)bi,zi‾,i∈S,δaibi,i∉S.(P_{U,S,z})_{ab} =\prod_{i=0}^{2} \begin{cases} (U_i)_{a_i,z_i}\,\overline{(U_i)_{b_i,z_i}},&i\in S,\\ \delta_{a_i b_i},&i\notin S. \end{cases}(PU,S,z​)ab​=i=0∏2​{(Ui​)ai​,zi​​(Ui​)bi​,zi​​​,δai​bi​​,​i∈S,i∈/S.​

Equivalently, its factor at a selected coordinate is the outer product of column ziz_izi​ of UiU_iUi​ with itself, and its factor at an unselected coordinate is the identity. The empty subset is allowed and gives the identity matrix.

QAlgorithms.Papers.Independent.Support50.P43.noisyQubits

For any complex Euclidean vector ψ\psiψ indexed by a∈{0,1}3a\in\{0,1\}^3a∈{0,1}3 and any real number ppp, this defines the 8×88\times88×8 complex matrix

ρab=p ψaψb‾+1−p8 δab.\rho_{ab}=p\,\psi_a\overline{\psi_b}+\frac{1-p}{8}\,\delta_{ab}.ρab​=pψa​ψb​​+81−p​δab​.

Here δab\delta_{ab}δab​ is 111 when a=ba=ba=b and 000 otherwise. Neither ∥ψ∥=1\|\psi\|=1∥ψ∥=1 nor 0≤p≤10\le p\le10≤p≤1 is required.

QAlgorithms.Papers.Independent.Support50.P43.partialTransposeA

For any complex matrix ρ\rhoρ indexed by three-bit configurations, this defines the matrix whose entries are

(ρTA)(a0,a1,a2),(b0,b1,b2)=ρ(b0,a1,a2),(a0,b1,b2).(\rho^{T_A})_{(a_0,a_1,a_2),(b_0,b_1,b_2)} =\rho_{(b_0,a_1,a_2),(a_0,b_1,b_2)}.(ρTA​)(a0​,a1​,a2​),(b0​,b1​,b2​)​=ρ(b0​,a1​,a2​),(a0​,b1​,b2​)​.

Thus the first row and column coordinates are exchanged while the other coordinates remain in place. No density-matrix or Hermiticity assumption is imposed.

QAlgorithms.Papers.Independent.Support50.P43.postMeasurement

For any three complex 2×22\times22×2 unitary matrices UiU_iUi​, any subset S⊆{0,1,2}S\subseteq\{0,1,2\}S⊆{0,1,2}, and any complex matrix ρ\rhoρ indexed by three-bit configurations, this defines

∑z:S→{0,1}PU,S,z ρ PU,S,z,(PU,S,z)ab=∏i∈S(Ui)ai,zi(Ui)bi,zi‾∏i∉Sδaibi.\sum_{z:S\to\{0,1\}}P_{U,S,z}\,\rho\,P_{U,S,z}, \qquad (P_{U,S,z})_{ab} =\prod_{i\in S}(U_i)_{a_i,z_i}\overline{(U_i)_{b_i,z_i}} \prod_{i\notin S}\delta_{a_i b_i}.z:S→{0,1}∑​PU,S,z​ρPU,S,z​,(PU,S,z​)ab​=i∈S∏​(Ui​)ai​,zi​​(Ui​)bi​,zi​​​i∈/S∏​δai​bi​​.

No density-matrix assumption is made on ρ\rhoρ. When SSS is empty, there is one outcome assignment, its matrix is the identity, and the result is ρ\rhoρ.

QAlgorithms.Papers.Independent.Support50.P43.reducedDensity

For every natural number nnn, every family of finite types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, every complex matrix ρ\rhoρ indexed by X=∏i=0n−1DiX=\prod_{i=0}^{n-1}D_iX=∏i=0n−1​Di​, and every subset s⊆{0,…,n−1}s\subseteq\{0,\ldots,n-1\}s⊆{0,…,n−1}, this defines the complex matrix RsR_sRs​ indexed by ∏i∈sDi\prod_{i\in s}D_i∏i∈s​Di​ whose entries are

Rs(a,b)=∑x,y∈Xx∣s=a,y∣s=bxi=yi for every i∉sρ(x,y).R_s(a,b)= \sum_{\substack{x,y\in X\\ x|_s=a,\;y|_s=b\\ x_i=y_i\text{ for every }i\notin s}} \rho(x,y).Rs​(a,b)=x,y∈Xx∣s​=a,y∣s​=bxi​=yi​ for every i∈/s​∑​ρ(x,y).

There is no positivity, Hermiticity, or trace normalization assumption on ρ\rhoρ, and the finite types DiD_iDi​ may be empty. For s=∅s=\varnothings=∅, RsR_sRs​ is the one-by-one matrix with entry tr⁡ρ\operatorname{tr}\rhotrρ; for sss equal to all coordinates, it reproduces ρ\rhoρ under restriction of labels. The definition also includes n=0n=0n=0.

QAlgorithms.Papers.Independent.Support50.P43.restrictLabel

For every natural number nnn, every family of types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, every subset s⊆{0,…,n−1}s\subseteq\{0,\ldots,n-1\}s⊆{0,…,n−1}, and every full label x∈∏i=0n−1Dix\in\prod_{i=0}^{n-1}D_ix∈∏i=0n−1​Di​, this defines its restriction to the coordinates in sss: the resulting label belongs to ∏i∈sDi\prod_{i\in s}D_i∏i∈s​Di​ and has value xix_ixi​ at each i∈si\in si∈s. No finiteness or nonemptiness assumption is imposed on the types DiD_iDi​. When s=∅s=\varnothings=∅, the result is the unique empty label; this includes n=0n=0n=0.

QAlgorithms.Papers.Independent.Support50.P43.singletonProduct

For every natural number nnn, every family of finite types (Di)i=0n−1(D_i)_{i=0}^{n-1}(Di​)i=0n−1​, and every complex matrix ρ\rhoρ indexed by X=∏iDiX=\prod_iD_iX=∏i​Di​, this defines the complex matrix SSS on XXX whose entries are

S(a,b)=∏i=0n−1R{i}(a∣{i},b∣{i}),Rt(u,v)=∑x,y∈Xx∣t=u,y∣t=vxℓ=yℓ for every ℓ∉tρ(x,y).S(a,b)=\prod_{i=0}^{n-1}R_{\{i\}}(a|_{\{i\}},b|_{\{i\}}), \qquad R_t(u,v)= \sum_{\substack{x,y\in X\\x|_t=u,\;y|_t=v\\ x_\ell=y_\ell\text{ for every }\ell\notin t}}\rho(x,y).S(a,b)=i=0∏n−1​R{i}​(a∣{i}​,b∣{i}​),Rt​(u,v)=x,y∈Xx∣t​=u,y∣t​=vxℓ​=yℓ​ for every ℓ∈/t​∑​ρ(x,y).

Here restriction retains precisely the coordinates in its subscript. No density matrix assumption is imposed on ρ\rhoρ, and the finite types DiD_iDi​ may be empty. For n=0n=0n=0, the label space has one element and the empty product gives the one-by-one matrix with entry 111, independently of ρ\rhoρ.

Human review
  • Endorsed by Shuze Chen · Oct 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 24, 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 paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me