Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Violation of the causal inequality: a four-qubit process with psucc>3/4p_{succ}>3/4psucc​>3/4

Proved
OCB2012.causal_inequality_violation

by Alien60 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

indefinite-causal-orderprocess-matrixquantum-gravityquantum-information

There exist a four-qubit process matrix WWW on A1A2B1B2A_1A_2B_1B_2A1​A2​B1​B2​ (Eqs. (4)–(5)), a local instrument {MA(x∣a)}x\{M^A(x\mid a)\}_x{MA(x∣a)}x​ for Alice for each input bit aaa, and a local instrument {MB(y∣b,b′)}y\{M^B(y\mid b,b')\}_y{MB(y∣b,b′)}y​ for Bob for each pair of input bits (b,b′)(b,b')(b,b′) (positive semidefinite CJ matrices whose outcome sums are CJ matrices of CPTP maps) such that the success probability of the task of Eq. (1),

psucc=12[P(x=b∣b′=0)+P(y=a∣b′=1)],P(x,y∣a,b,b′)=Tr⁡[W(MA(x∣a)⊗MB(y∣b,b′))],p_{succ}=\tfrac12\big[P(x=b\mid b'=0)+P(y=a\mid b'=1)\big],\qquad P(x,y\mid a,b,b')=\operatorname{Tr}\big[W(M^A(x\mid a)\otimes M^B(y\mid b,b'))\big],psucc​=21​[P(x=b∣b′=0)+P(y=a∣b′=1)],P(x,y∣a,b,b′)=Tr[W(MA(x∣a)⊗MB(y∣b,b′))],

with a,b,b′a,b,b'a,b,b′ independent and uniform, strictly exceeds the causal bound: psucc>3/4p_{succ}>3/4psucc​>3/4.

Preamble
import Mathlib
import Definitions.Def_PeresTerno_kraus_basics
import Definitions.Def_OCB2012_defs

open Matrix
open scoped Kronecker ComplexOrder
Formal statement
namespace OCB2012

theorem causal_inequality_violation :
    ∃ (W : Matrix ((Qubit × Qubit) × (Qubit × Qubit)) ((Qubit × Qubit) × (Qubit × Qubit)) ℂ)
      (MA : Bool → Bool → Matrix (Qubit × Qubit) (Qubit × Qubit) ℂ)
      (MB : Bool → Bool → Bool → Matrix (Qubit × Qubit) (Qubit × Qubit) ℂ),
      IsProcessMatrix W ∧ IsAliceInstrument MA ∧ IsBobInstrument MB ∧
        3 / 4 < pSucc W MA MB := by sorry

end OCB2012
Source
O. Oreshkov, F. Costa, Č. Brukner, Quantum correlations with no causal order, Nat. Commun. 3, 1092 (2012), https://arxiv.org/abs/1105.4464v3, Abstract; p. 5, Eqs. (2), (7), (8)
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Read-back of OCB2012.causal_inequality_violation.

Setting. A "qubit" index set is Q={0,1}Q = \{0,1\}Q={0,1}, so every system A1,A2,B1,B2A_1, A_2, B_1, B_2A1​,A2​,B1​,B2​ is C2\mathbb{C}^2C2. Operators on A1A2A_1A_2A1​A2​ (and on B1B2B_1B_2B1​B2​) are complex 4×44\times 44×4 matrices indexed by pairs (i1,i2)∈Q×Q(i_1,i_2)\in Q\times Q(i1​,i2​)∈Q×Q, first component A1A_1A1​ (resp. B1B_1B1​), second A2A_2A2​ (resp. B2B_2B2​). Operators on all four systems are complex 16×1616\times 1616×16 matrices indexed by ((i1,i2),(k1,k2))((i_1,i_2),(k_1,k_2))((i1​,i2​),(k1​,k2​)), i.e. in the order A1A2B1B2A_1A_2B_1B_2A1​A2​B1​B2​. Bits are Booleans, with false↦0\mathrm{false}\mapsto 0false↦0 and true↦1\mathrm{true}\mapsto 1true↦1. Nothing in the statement is universally quantified, and it has no hypotheses or typeclass parameters beyond the fixed concrete types above. The whole statement is a single existential claim.

The claim. There exist

  • a complex 16×1616\times 1616×16 matrix WWW on A1A2B1B2A_1A_2B_1B_2A1​A2​B1​B2​,
  • a family Mx,aAM^A_{x,a}Mx,aA​ of complex 4×44\times 44×4 matrices on A1A2A_1A_2A1​A2​, indexed by two bits x,a∈{0,1}x, a\in\{0,1\}x,a∈{0,1} (first index xxx = Alice's output/guess, second aaa = her input),
  • a family My,b,b′BM^B_{y,b,b'}My,b,b′B​ of complex 4×44\times 44×4 matrices on B1B2B_1B_2B1​B2​, indexed by three bits y,b,b′∈{0,1}y,b,b'\in\{0,1\}y,b,b′∈{0,1} (first index yyy = Bob's guess, then inputs bbb and b′b'b′),

such that all four of the following hold simultaneously:

  1. WWW is a process matrix (defined below);
  2. MAM^AMA is an Alice instrument (defined below);
  3. MBM^BMB is a Bob instrument (defined below);
  4. the success probability satisfies the strict inequality
34<psucc(W,MA,MB).\tfrac34 < p_{\mathrm{succ}}(W, M^A, M^B).43​<psucc​(W,MA,MB).

No particular WWW, MAM^AMA or MBM^BMB is required; in particular the statement does not refer to the specific matrices W7W_7W7​, ξ\xiξ, η\etaη defined in the same file, nor to causal separability or any no-signalling condition. Any witnesses meeting conditions 1–4 suffice.

Auxiliary definitions, unfolded.

Partial trace over the second factor. For a matrix MMM indexed by pairs (i,k)(i,k)(i,k), Tr⁡2M\operatorname{Tr}_2 MTr2​M is the matrix with entries

(Tr⁡2M)ij=∑kM(i,k),(j,k).(\operatorname{Tr}_2 M)_{ij} = \sum_{k} M_{(i,k),(j,k)}.(Tr2​M)ij​=k∑​M(i,k),(j,k)​.

For MMM on A1A2A_1A_2A1​A2​ this traces out A2A_2A2​ (the output system); for MMM on B1B2B_1B_2B1​B2​ it traces out B2B_2B2​.

CPTP Choi condition. A matrix MMM on X1X2X_1X_2X1​X2​ "is the CJ matrix of a CPTP map" means exactly:

M⪰0andTr⁡2M=1X1.M \succeq 0 \quad\text{and}\quad \operatorname{Tr}_{2} M = \mathbb{1}_{X_1}.M⪰0andTr2​M=1X1​​.

Here M⪰0M\succeq 0M⪰0 is Mathlib's positive semidefiniteness, which includes Hermiticity (M†=MM^\dagger = MM†=M) together with v†Mv≥0v^\dagger M v \ge 0v†Mv≥0 for all complex vectors vvv. No other condition (such as a particular Choi normalization convention or transpose) is imposed.

Probability of a pair of outcomes. For WWW on A1A2B1B2A_1A_2B_1B_2A1​A2​B1​B2​ and matrices MAM_AMA​ on A1A2A_1A_2A1​A2​, MBM_BMB​ on B1B2B_1B_2B1​B2​,

prob⁡(W,MA,MB)=Tr⁡[ W (MA⊗MB) ]∈C,\operatorname{prob}(W, M_A, M_B) = \operatorname{Tr}\big[\,W\,(M_A\otimes M_B)\,\big]\in\mathbb{C},prob(W,MA​,MB​)=Tr[W(MA​⊗MB​)]∈C,

where ⊗\otimes⊗ is the Kronecker product with (MA⊗MB)(i,k),(j,l)=(MA)ij(MB)kl(M_A\otimes M_B)_{(i,k),(j,l)} = (M_A)_{ij}(M_B)_{kl}(MA​⊗MB​)(i,k),(j,l)​=(MA​)ij​(MB​)kl​, i,ji,ji,j indexing A1A2A_1A_2A1​A2​ and k,lk,lk,l indexing B1B2B_1B_2B1​B2​.

Process matrix. WWW is a process matrix iff

W⪰0andTr⁡[W(MA⊗MB)]=1W\succeq 0 \quad\text{and}\quad \operatorname{Tr}\big[W(M_A\otimes M_B)\big] = 1W⪰0andTr[W(MA​⊗MB​)]=1

for every MAM_AMA​ on A1A2A_1A_2A1​A2​ and every MBM_BMB​ on B1B2B_1B_2B1​B2​ that both satisfy the CPTP Choi condition above. (This is an equality of complex numbers.)

Alice instrument. MAM^AMA is an Alice instrument iff

  • Mx,aA⪰0M^A_{x,a}\succeq 0Mx,aA​⪰0 for all x,a∈{0,1}x,a\in\{0,1\}x,a∈{0,1}, and
  • for each a∈{0,1}a\in\{0,1\}a∈{0,1}, the sum M0,aA+M1,aAM^A_{0,a}+M^A_{1,a}M0,aA​+M1,aA​ satisfies the CPTP Choi condition, i.e. it is PSD and its partial trace over A2A_2A2​ equals 1A1\mathbb{1}_{A_1}1A1​​ (the 2×22\times22×2 identity).

Bob instrument. MBM^BMB is a Bob instrument iff

  • My,b,b′B⪰0M^B_{y,b,b'}\succeq 0My,b,b′B​⪰0 for all y,b,b′∈{0,1}y,b,b'\in\{0,1\}y,b,b′∈{0,1}, and
  • for each pair (b,b′)(b,b')(b,b′), the sum M0,b,b′B+M1,b,b′BM^B_{0,b,b'}+M^B_{1,b,b'}M0,b,b′B​+M1,b,b′B​ satisfies the CPTP Choi condition (PSD, partial trace over B2B_2B2​ equals 1B1\mathbb{1}_{B_1}1B1​​).

The input index is allowed to change the whole instrument (the CPTP sum may differ for different inputs), and Bob's instrument may depend on both bbb and b′b'b′.

Success probability. With P(x,y∣a,b,b′):=Re⁡Tr⁡[W(Mx,aA⊗My,b,b′B)]P(x,y\mid a,b,b') := \operatorname{Re}\operatorname{Tr}\big[W(M^A_{x,a}\otimes M^B_{y,b,b'})\big]P(x,y∣a,b,b′):=ReTr[W(Mx,aA​⊗My,b,b′B​)] (the real part of the complex trace),

psucc=12⋅14∑a∈{0,1}∑b∈{0,1}∑y∈{0,1}P(x=b, y∣a,b,b′=0)  +  12⋅14∑a∈{0,1}∑b∈{0,1}∑x∈{0,1}P(x, y=a∣a,b,b′=1).p_{\mathrm{succ}} = \frac12\cdot\frac14\sum_{a\in\{0,1\}}\sum_{b\in\{0,1\}}\sum_{y\in\{0,1\}} P(x{=}b,\,y\mid a,b,b'{=}0) \;+\; \frac12\cdot\frac14\sum_{a\in\{0,1\}}\sum_{b\in\{0,1\}}\sum_{x\in\{0,1\}} P(x,\,y{=}a\mid a,b,b'{=}1).psucc​=21​⋅41​a∈{0,1}∑​b∈{0,1}∑​y∈{0,1}∑​P(x=b,y∣a,b,b′=0)+21​⋅41​a∈{0,1}∑​b∈{0,1}∑​x∈{0,1}∑​P(x,y=a∣a,b,b′=1).

Concretely: the first term uses Alice's matrix Mb,aAM^A_{b,a}Mb,aA​ (her guess index set equal to Bob's input bbb) and Bob's My,b,0BM^B_{y,b,0}My,b,0B​, summed over Bob's guess yyy; the second uses Alice's Mx,aAM^A_{x,a}Mx,aA​, summed over her guess xxx, and Bob's Ma,b,1BM^B_{a,b,1}Ma,b,1B​ (his guess index set equal to Alice's input aaa). Each of the eight input triples (a,b,b′)(a,b,b')(a,b,b′) thus carries weight 18\tfrac1881​, i.e. a,b,b′a,b,b'a,b,b′ are treated as independent uniform bits; when b′=0b'=0b′=0 success means Alice's guess equals bbb, and when b′=1b'=1b′=1 success means Bob's guess equals aaa.

Edge cases and logical strength.

  • The conclusion is a strict inequality 34<psucc\tfrac34 < p_{\mathrm{succ}}43​<psucc​ of real numbers; equality psucc=34p_{\mathrm{succ}} = \tfrac34psucc​=43​ would not suffice.
  • Only real parts of the traces enter psuccp_{\mathrm{succ}}psucc​. Since WWW and all Mx,aAM^A_{x,a}Mx,aA​, My,b,b′BM^B_{y,b,b'}My,b,b′B​ are required to be PSD (hence Hermitian, and Mx,aA⊗My,b,b′BM^A_{x,a}\otimes M^B_{y,b,b'}Mx,aA​⊗My,b,b′B​ is PSD), each trace Tr⁡[W(⋅⊗⋅)]\operatorname{Tr}[W(\cdot\otimes\cdot)]Tr[W(⋅⊗⋅)] is automatically a nonnegative real, so the real part loses nothing for witnesses satisfying conditions 1–3.
  • The process-matrix condition is satisfiable (e.g. W=14116W = \tfrac14\mathbb{1}_{16}W=41​116​ satisfies it, since then the trace equals 14Tr⁡MATr⁡MB=14⋅2⋅2\tfrac14\operatorname{Tr}M_A\operatorname{Tr}M_B = \tfrac14\cdot 2\cdot 241​TrMA​TrMB​=41​⋅2⋅2), so the hypotheses on the witnesses are not contradictory; the statement is not vacuous in that sense, and the content lies in exhibiting witnesses that additionally exceed 34\tfrac3443​.
  • Given conditions 1–3, the normalization makes ∑x,yP(x,y∣a,b,b′)=1\sum_{x,y}P(x,y\mid a,b,b')=1∑x,y​P(x,y∣a,b,b′)=1 for each input triple, so psucc≤1p_{\mathrm{succ}}\le 1psucc​≤1 for any valid witnesses; the claim asks for a value in (34,1](\tfrac34, 1](43​,1].
  • All dimensions are fixed at 2 for every system; the statement says nothing about other dimensions.
  • The proof is currently sorry (unproved).
Human review
  • Endorsed by Shuze Chen · Oct 8, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Alien60 · Oct 8, 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