Lean, informalized

Nielsen & Chuang · Chapter 11, Entropy and information · p. 520

Theorem 11.11 - Joint concavity in a pair of positive operators, for every exponent between zero and one

A trace of two operator powers is jointly concave.

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 Theorem

Theorem 11.11 is Lieb’s theorem, a statement about positive operators and their powers. Take positive operators AA and BB, an operator KK that carries the space on which BB acts into the space on which AA acts, with adjoint K†K^\dagger carrying the space of AA back to that of BB, and an exponent tt between 00 and 11, and form the trace of K†AtKB1−tK^\dagger A^t K B^{1-t}, where AtA^t and B1−tB^{1-t} are the powers of AA and BB to the exponents tt and 1−t1-t. Mix two such pairs as one mixes two density operators into an ensemble: with probabilities aa and bb, the pairs (A1,B1)(A_1, B_1) and (A2,B2)(A_2, B_2) give the pair (aA1+bA2,aB1+bB2)(aA_1 + bA_2,\ aB_1 + bB_2). The theorem compares the trace for the mixed pair with the average, under the same probabilities, of the traces for the two pairs.

The library takes AA acting on a complex Hilbert space EE and BB on ℂn\mathbb{C}^n.

Positive Operators and Their Powers

An operator AA on EE is positive, written A≥0A \ge 0, if ⟨ψ|A|ψ⟩\langle\psi|A|\psi\rangle is real and nonnegative for every vector |ψ⟩|\psi\rangle in EE. The positive operators on EE form the set 𝒫(E)\mathcal{P}(E), and those on ℂn\mathbb{C}^n the set 𝒫(ℂn)\mathcal{P}(\mathbb{C}^n).

The power AsA^s of a positive operator AA, for a real exponent s≥0s \ge 0, is the operator with the same eigenvectors as AA and each eigenvalue λ\lambda replaced by λs\lambda^s: where A=∑iλi|i⟩⟨i|A = \sum_i \lambda_i |i\rangle\langle i|, As=∑iλis|i⟩⟨i|.A^s = \sum_i \lambda_i^s\, |i\rangle\langle i| .(1)

The function ftf_t the theorem is about takes a pair (A,B)(A, B) in 𝒫(E)×𝒫(ℂn)\mathcal{P}(E) \times \mathcal{P}(\mathbb{C}^n) to the real part of the trace of the operator K†AtKB1−tK^\dagger A^t K B^{1-t} on ℂn\mathbb{C}^n: ft(A,B)=Re⁡tr⁡(K†AtKB1−t).f_t(A, B) = \operatorname{Re} \operatorname{tr}\bigl(K^\dagger A^t K B^{1-t}\bigr).(2)

The Claim

Theorem 1. Let KK be an operator from ℂn\mathbb{C}^n to a complex Hilbert space EE, and let 0≤t≤10 \le t \le 1. Then ftf_t is jointly concave on the convex set of pairs of positive operators, so that for A1,A2A_1, A_2 in 𝒫(E)\mathcal{P}(E), B1,B2B_1, B_2 in 𝒫(ℂn)\mathcal{P}(\mathbb{C}^n) and weights a,b≥0a, b \ge 0 with a+b=1a + b = 1, aft(A1,B1)+bft(A2,B2)≤ft(aA1+bA2,aB1+bB2).a\, f_t(A_1, B_1) + b\, f_t(A_2, B_2) \le f_t(aA_1 + bA_2,\ aB_1 + bB_2).(3)✦ ft(a(A1,B1)+b(A2,B2))≥aft(A1,B1)+bft(A2,B2).f_t\bigl(a(A_1, B_1) + b(A_2, B_2)\bigr) \ge a\, f_t(A_1, B_1) + b\, f_t(A_2, B_2).(4)✦

What Is Checked

The claim is stated over the positive operators on EE and on ℂn\mathbb{C}^n, their powers (1), and the trace on ℂn\mathbb{C}^n through which ftf_t is defined in (2).

Besides the inequalities (3) and (4), the library proves that the pairs of positive operators form a convex set, the domain on which ftf_t is concave.