Lean, informalized

Nielsen & Chuang · Chapter 11, Entropy and information · p. 526

Problem 11.4 - Conditional strong subadditivity for every quantum state, and a classical inequality that can fail

Strong subadditivity conditioned on a fourth system.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Exercise

Four quantum systems AA, BB, CC and DD share a state ρ\rho. The problem asks whether strong subadditivity for AA, BB and CC still holds when every entropy in it is taken conditional on the fourth system DD.

The library states that the conditional inequality holds for every state of the four systems. For four random variables on a finite probability space it states a counterexample to a classical inequality of the same shape, one in which each entropy is that of DD given some of the other three variables.

Four Quantum Systems

A quantum system SS is a physical system whose state space is a finite dimensional complex Hilbert space.✦ The composite system S⊗TS \otimes T of two systems SS and TT is the system whose state space is the tensor product of their state spaces.✦ An identification ee of a system SS with a system TT is a unitary map from the state space of SS onto that of TT.✦ The regrouping αS,T,U\alpha_{S,T,U} of three systems SS, TT and UU is the identification of S⊗(T⊗U)S \otimes (T \otimes U) with (S⊗T)⊗U(S \otimes T) \otimes U that sends u⊗(v⊗w)u \otimes (v \otimes w) to (u⊗v)⊗w(u \otimes v) \otimes w.✦

A state ρ\rho of a system SS is a positive operator of unit trace on the state space of SS.✦ The relabelling of ρ\rho along an identification ee of SS with TT is the state eρe−1e \rho e^{-1} of TT.✦

For a state ρ\rho of S⊗TS \otimes T, the reduced state ρS=tr⁡Tρ\rho_S = \operatorname{tr}_T \rho of SS is the partial trace of ρ\rho over TT, computed in an orthonormal basis {|k⟩}\{|k\rangle\} of the state space of TT: tr⁡Tρ=∑k(I⊗⟨k|)ρ(I⊗|k⟩).\operatorname{tr}_T \rho = \sum_k \bigl(I \otimes \langle k|\bigr)\, \rho\, \bigl(I \otimes |k\rangle\bigr).(1)✦ Here II is the identity on the state space of SS, the map I⊗|k⟩I \otimes |k\rangle embeds |x⟩|x\rangle as |x⟩|k⟩|x\rangle|k\rangle✦, and I⊗⟨k|I \otimes \langle k| sends |x⟩|y⟩|x\rangle|y\rangle to ⟨k|y⟩|x⟩\langle k|y\rangle\,|x\rangle✦. The reduced state ρT=tr⁡Sρ\rho_T = \operatorname{tr}_S \rho of TT is the same sum with the two factors exchanged.✦

The four systems of the problem are composed as A⊗((B⊗D)⊗C)A \otimes ((B \otimes D) \otimes C), and ρ\rho is a state of this composite. Tracing out AA gives the reduced state ρBDC\rho_{BDC} of (B⊗D)⊗C(B \otimes D) \otimes C; tracing CC out of ρBDC\rho_{BDC} gives ρBD\rho_{BD}, and tracing BB out of ρBD\rho_{BD} gives ρD\rho_D. The reduced state ρABD\rho_{ABD} of A⊗(B⊗D)A \otimes (B \otimes D) is the partial trace over CC of the relabelling of ρ\rho along the regrouping αA,B⊗D,C\alpha_{A,\,B \otimes D,\,C}.

Entropies

The Shannon entropy of a probability distribution qq on a finite set is H(q)=−∑iqilog⁡qi,H(q) = -\sum_i q_i \log q_i ,(2)✦ with the natural logarithm, so that entropies are measured in nats. The von Neumann entropy of a state ρ\rho is the Shannon entropy of its eigenvalues λ1,…,λn\lambda_1, \dots, \lambda_n, counted with multiplicity, where nn is the dimension of the state space: S(ρ)=−tr⁡(ρlog⁡ρ)=H(λ1,…,λn).S(\rho) = -\operatorname{tr}(\rho \log \rho) = H(\lambda_1, \dots, \lambda_n).(3)✦ The conditional entropy of a state σ\sigma of a composite X⊗YX \otimes Y, with reduced state σY\sigma_Y on YY, is S(X|Y)=S(σ)−S(σY).S(X|Y) = S(\sigma) - S(\sigma_Y).(4)✦

Conditioned on DD, the entropy of AA, BB and CC together is S(A,B,C|D)=S(ρ)−S(ρD),S(A,B,C|D) = S(\rho) - S(\rho_D),(5)✦ and the entropy of BB is the conditional entropy of ρBD\rho_{BD}, S(B|D)=S(ρBD)−S(ρD).S(B|D) = S(\rho_{BD}) - S(\rho_D).(6)✦ The entropy of the pair AA, BB given DD is S(A,B|D)=S(ρABD)−S(ρD),S(A,B|D) = S(\rho_{ABD}) - S(\rho_D),(7)✦ and that of the pair BB, CC given DD is S(B,C|D)=S(ρBDC)−S(ρD).S(B,C|D) = S(\rho_{BDC}) - S(\rho_D).(8)✦

Random Variables

A finite probability space is a finite sample space Ω\Omega with probabilities p(ω)≥0p(\omega) \ge 0 that sum to one. A random variable ff on Ω\Omega is a function from Ω\Omega to a finite set, and its entropy H(f)H(f) is the Shannon entropy of its distribution: H(f)=H(Pf),Pf(b)=∑ω:f(ω)=bp(ω).H(f) = H(P_f), \qquad P_f(b) = \sum_{\omega :\, f(\omega) = b} p(\omega).(9)✦ The conditional entropy of a random variable ff given a random variable gg is H(f|g)=H(f,g)−H(g),H(f|g) = H(f,g) - H(g),(10)✦ where (f,g)(f,g) is the random variable ω↦(f(ω),g(ω))\omega \mapsto (f(\omega), g(\omega)). Several variables after the bar stand for their tuple, as in H(D|A,B,C)H(D|A,B,C).

The Claim

Theorem 1. Let AA, BB, CC and DD be quantum systems and ρ\rho a state of A⊗((B⊗D)⊗C)A \otimes ((B \otimes D) \otimes C), with the conditional entropies (5) to (8). Then S(A,B,C|D)+S(B|D)≤S(A,B|D)+S(B,C|D).S(A,B,C|D) + S(B|D) \le S(A,B|D) + S(B,C|D).(11)✦

Theorem 2. There are a finite probability space Ω\Omega and random variables AA, BB, CC and DD on Ω\Omega such that H(D|A,B,C)+H(D|B)>H(D|A,B)+H(D|B,C).H(D|A,B,C) + H(D|B) > H(D|A,B) + H(D|B,C).(12)✦

What Is Checked

The library proves that the partial trace of a state over either factor is a state of the other system, and that the relabelling of a state along an identification of systems is again a state.