Physlib

QuantumInfo.Finite.Entropy.Relative

To do relative entropies, we start with the _sandwiched Renyi Relative Entropy_ which is a nice general form. Then instead of proving many theorems (like DPI, relabelling, additivity, etc.) several times, we just prove it for this one quantity, then it follows for other quantities (like the relative entropy) as a special case.

55 declarations

theorem

Weighted Jensen Inequality for Power Functions xqx^q with q≥1q \ge 1

Let wjw_j be a set of non-negative weights such that ∑jwj=1\sum_j w_j = 1, and let bjb_j be non-negative real numbers. For any real number q≥1q \geq 1, the following weighted Jensen inequality holds: (∑jwjbj)q≤∑jwjbjq\left( \sum_j w_j b_j \right)^q \le \sum_j w_j b_j^q

theorem

Doubly Stochastic Hölder Inequality

Let aia_i and bjb_j be non-negative real numbers for i,j∈{1,…,d}i, j \in \{1, \dots, d\}. Let wijw_{ij} be a doubly stochastic matrix, such that wij≥0w_{ij} \geq 0 for all i,ji, j, and its rows and columns sum to 1 (∑jwij=1\sum_{j} w_{ij} = 1 for all ii, and ∑iwij=1\sum_{i} w_{ij} = 1 for all jj). For any conjugate exponents p,q>1p, q > 1 satisfying 1p+1q=1\frac{1}{p} + \frac{1}{q} = 1, the following inequality holds: ∑i,jaibjwij≤(∑iaip)1/p(∑jbjq)1/q\sum_{i,j} a_i b_j w_{ij} \le \left( \sum_{i} a_i^p \right)^{1/p} \left( \sum_{j} b_j^q \right)^{1/q}

theorem

Hermitian Trace Hölder Inequality: ⟨A,B⟩≤Tr[Ap]1/pTr[Bq]1/q\langle A, B \rangle \le \text{Tr}[A^p]^{1/p} \text{Tr}[B^q]^{1/q}

Let AA and BB be complex Hermitian matrices of dimension dd that are positive semi-definite (A≥0,B≥0A \ge 0, B \ge 0). Let pp and qq be real conjugate exponents such that p>1p > 1 and 1p+1q=1\frac{1}{p} + \frac{1}{q} = 1. Then the real-valued Hilbert-Schmidt inner product of AA and BB, denoted ⟨A,B⟩R\langle A, B \rangle_{\mathbb{R}}, satisfies: ⟨A,B⟩R≤(Tr[Ap])1/p(Tr[Bq])1/q\langle A, B \rangle_{\mathbb{R}} \le (\text{Tr}[A^p])^{1/p} (\text{Tr}[B^q])^{1/q}

theorem

Mr≤1M^r \le 1 for Mixed State MM and r>0r > 0

Let σ\sigma be a quantum mixed state with associated matrix MM. For any real number r>0r > 0, the rr-th power of the matrix MM satisfies Mr≤IM^r \le \mathbb{I}, where I\mathbb{I} denotes the identity matrix.

theorem

If A≤IA \le I, then its eigenvalues satisfy λi≤1\lambda_i \le 1

Let AA be a Hermitian matrix in Cd×d\mathbb{C}^{d \times d}. If A≤IA \le I (where II is the identity matrix), then for every index ii in the dimension type dd, the ii-th eigenvalue of AA satisfies λi(A)≤1\lambda_i(A) \le 1. Combined with the condition A≥0A \ge 0, this implies that all eigenvalues of AA lie in the interval [0,1][0, 1].

theorem

Tr(Ap)≤Tr(A)\text{Tr}(A^p) \le \text{Tr}(A) for 0≤A≤I0 \le A \le I and p≥1p \ge 1

Let AA be a complex Hermitian matrix of dimension dd such that 0≤A≤I0 \le A \le I, where II is the identity matrix. For any real number p≥1p \ge 1, the trace of the pp-th power of AA is less than or equal to the trace of AA, i.e., Tr(Ap)≤Tr(A)\text{Tr}(A^p) \le \text{Tr}(A).

theorem

A−pAp=Psupp(A)A^{-p} A^p = P_{\text{supp}(A)} for PSD AA and p≠0p \neq 0

Let AA be a positive semi-definite (PSD) Hermitian matrix of dimension d×dd \times d over the complex numbers C\mathbb{C}. For any non-zero real number pp, the product of the matrix power A−pA^{-p} and ApA^p is equal to the orthogonal projection onto the support of AA, denoted as Psupp(A)P_{\text{supp}(A)}.

theorem

Left multiplication of a Hermitian matrix by its support projection PAA=AP_A A = A

For any complex Hermitian matrix A∈Cd×dA \in \mathbb{C}^{d \times d}, the product of its support projection PAP_A and the matrix itself is equal to AA, i.e., PAA=AP_A A = A.

theorem

⟨A,PA⟩=Tr(A)\langle A, P_A \rangle = \text{Tr}(A)

For any complex Hermitian matrix AA, the inner product of AA with its support projection PAP_A is equal to the trace of AA. Here, ⟨A,PA⟩\langle A, P_A \rangle denotes the Frobenius inner product (or Hilbert-Schmidt inner product), and A.traceA.trace denotes the sum of the diagonal elements of AA.

theorem

ker⁡B⊆ker⁡A  ⟹  AΠB=A\ker B \subseteq \ker A \implies A \Pi_B = A

Let AA and BB be Hermitian matrices of dimension dd over the complex numbers C\mathbb{C}. If the kernel of the linear map associated with BB is a subspace of the kernel of the linear map associated with AA (i.e., ker⁡B⊆ker⁡A\ker B \subseteq \ker A), then the product of AA and the support projection of BB (denoted ΠB\Pi_B) satisfies AΠB=AA \Pi_B = A.

theorem

ker⁡B⊆ker⁡A  ⟹  ⟨A,ΠB⟩=Tr(A)\ker B \subseteq \ker A \implies \langle A, \Pi_B \rangle = \text{Tr}(A)

Let AA and BB be complex Hermitian matrices of dimension dd. If the kernel of BB is contained in the kernel of AA (i.e., ker⁡B⊆ker⁡A\ker B \subseteq \ker A), then the Frobenius inner product of AA and the support projection of BB (denoted ΠB\Pi_B) is equal to the trace of AA, such that ⟨A,ΠB⟩=Tr(A)\langle A, \Pi_B \rangle = \text{Tr}(A).

theorem

ker⁡σ⊆ker⁡ρ  ⟹  ⟨Πσ,ρ⟩R=1\ker \sigma \subseteq \ker \rho \implies \langle \Pi_{\sigma}, \rho \rangle_{\mathbb{R}} = 1

Let ρ\rho and σ\sigma be density matrices. If the kernel of the linear map associated with σ\sigma is contained within the kernel of the linear map associated with ρ\rho (denoted ker⁡σ⊆ker⁡ρ\ker \sigma \subseteq \ker \rho), then the real inner product of the support projection of σ\sigma and the density matrix ρ\rho is equal to 1, i.e., ⟨Πσ,ρ⟩R=1\langle \Pi_{\sigma}, \rho \rangle_{\mathbb{R}} = 1.

theorem

Non-negativity of Quantum Relative Entropy D(ρ∥σ)≥0D(\rho \| \sigma) \geq 0

Let ρ\rho and σ\sigma be density operators (represented by their matrices ρ.M\rho.M and σ.M\sigma.M). If the kernel of σ.M\sigma.M is contained in the kernel of ρ.M\rho.M (ker⁡(σ.M)⊆ker⁡(ρ.M)\ker(\sigma.M) \subseteq \ker(\rho.M)), then the quantity ⟨ρ.M,log⁡(ρ.M)−log⁡(σ.M)⟩\langle \rho.M, \log(\rho.M) - \log(\sigma.M) \rangle is non-negative, i.e., 0≤⟨ρ.M,log⁡ρ.M−log⁡σ.M⟩0 \leq \langle \rho.M, \log \rho.M - \log \sigma.M \rangle. Here ⟨A,B⟩\langle A, B \rangle denotes the Hilbert-Schmidt inner product Tr(A†B)\text{Tr}(A^\dagger B).

theorem

Non-negativity of Sandwiched Rényi Relative Entropy Dα(ρ∥σ)≥0D_{\alpha}(\rho \| \sigma) \ge 0

Let ρ\rho and σ\sigma be density states with associated matrices ρ.M\rho.M and σ.M\sigma.M. Suppose the kernel of σ.M\sigma.M is contained in the kernel of ρ.M\rho.M (ker⁡σ.M⊆ker⁡ρ.M\ker \sigma.M \subseteq \ker \rho.M). For any α∈R\alpha \in \mathbb{R} such that α>0\alpha > 0, the sandwiched Rényi relative entropy Dα(ρ∥σ)D_{\alpha}(\rho \| \sigma) is non-negative (0≤Dα(ρ∥σ)0 \le D_{\alpha}(\rho \| \sigma)), where Dα(ρ∥σ)D_{\alpha}(\rho \| \sigma) is defined as: - If α=1\alpha = 1, the relative entropy is given by the inner product ⟨ρ.M,log⁡ρ.M−log⁡σ.M⟩\langle \rho.M, \log \rho.M - \log \sigma.M \rangle. - If α≠1\alpha \neq 1, it is given by 1α−1log⁡Tr[(σ.M1−α2αρ.Mσ.M1−α2α)α]\frac{1}{\alpha - 1} \log \text{Tr}\left[ \left( \sigma.M^{\frac{1-\alpha}{2\alpha}} \rho.M \sigma.M^{\frac{1-\alpha}{2\alpha}} \right)^\alpha \right].

theorem

ker⁡A⊆ker⁡C\ker A \subseteq \ker C and ker⁡B⊆ker⁡D\ker B \subseteq \ker D implies ker⁡(A⊗B)⊆ker⁡(C⊗D)\ker(A \otimes B) \subseteq \ker(C \otimes D)

Let A,CA, C be square matrices of size d1×d1d_1 \times d_1 over C\mathbb{C}, and let B,DB, D be square matrices of size d2×d2d_2 \times d_2 over C\mathbb{C}. If the kernel of AA is contained in the kernel of CC (ker⁡A⊆ker⁡C\ker A \subseteq \ker C) and the kernel of BB is contained in the kernel of DD (ker⁡B⊆ker⁡D\ker B \subseteq \ker D), then the kernel of the Kronecker product A⊗BA \otimes B is contained in the kernel of the Kronecker product C⊗DC \otimes D (ker⁡(A⊗B)⊆ker⁡(C⊗D)\ker(A \otimes B) \subseteq \ker(C \otimes D)). Here, ker⁡M\ker M refers to the kernel of the linear map associated with a matrix MM.

theorem

ker⁡(σ1⊗σ2)⊆ker⁡(ρ1⊗ρ2)\ker(\sigma_1 \otimes \sigma_2) \subseteq \ker(\rho_1 \otimes \rho_2) implies ker⁡σ1⊆ker⁡ρ1\ker \sigma_1 \subseteq \ker \rho_1

Let ρ1,σ1\rho_1, \sigma_1 be quantum mixed states of dimension d1d_1, and ρ2,σ2\rho_2, \sigma_2 be quantum mixed states of dimension d2d_2. If the kernel of the Hermitian operator associated with the product state σ1⊗σ2\sigma_1 \otimes \sigma_2 is contained in the kernel of the Hermitian operator associated with the product state ρ1⊗ρ2\rho_1 \otimes \rho_2 (i.e., ker⁡(σ1⊗σ2).M⊆ker⁡(ρ1⊗ρ2).M\ker(\sigma_1 \otimes \sigma_2).M \subseteq \ker(\rho_1 \otimes \rho_2).M), then the kernel of σ1\sigma_1 is contained in the kernel of ρ1\rho_1 (ker⁡σ1.M⊆ker⁡ρ1.M\ker \sigma_1.M \subseteq \ker \rho_1.M).

theorem

ker⁡(Mσ1⊗σ2)⊆ker⁡(Mρ1⊗ρ2)  ⟹  ker⁡Mσ2⊆ker⁡Mρ2\ker(M_{\sigma_1 \otimes \sigma_2}) \subseteq \ker(M_{\rho_1 \otimes \rho_2}) \implies \ker M_{\sigma_2} \subseteq \ker M_{\rho_2}

Let ρ1,σ1\rho_1, \sigma_1 be quantum mixed states of dimension d1d_1 and ρ2,σ2\rho_2, \sigma_2 be quantum mixed states of dimension d2d_2. If the kernel of the Hermitian operator of the product state σ1⊗σ2\sigma_1 \otimes \sigma_2 is contained in the kernel of the Hermitian operator of the product state ρ1⊗ρ2\rho_1 \otimes \rho_2 (i.e., ker⁡(Mσ1⊗σ2)⊆ker⁡(Mρ1⊗ρ2)\ker(M_{\sigma_1 \otimes \sigma_2}) \subseteq \ker(M_{\rho_1 \otimes \rho_2})), then the kernel of the Hermitian operator of σ2\sigma_2 is contained in the kernel of the Hermitian operator of ρ2\rho_2 (i.e., ker⁡Mσ2⊆ker⁡Mρ2\ker M_{\sigma_2} \subseteq \ker M_{\rho_2}).

theorem

ker⁡(σ1⊗σ2)⊆ker⁡(ρ1⊗ρ2)  ⟺  ker⁡σ1⊆ker⁡ρ1∧ker⁡σ2⊆ker⁡ρ2\ker(\sigma_1 \otimes \sigma_2) \subseteq \ker(\rho_1 \otimes \rho_2) \iff \ker \sigma_1 \subseteq \ker \rho_1 \land \ker \sigma_2 \subseteq \ker \rho_2

Let ρ1,σ1\rho_1, \sigma_1 be mixed states on a Hilbert space of dimension d1d_1, and let ρ2,σ2\rho_2, \sigma_2 be mixed states on a Hilbert space of dimension d2d_2. The kernel of the density matrix of the product state σ1⊗σ2\sigma_1 \otimes \sigma_2 is contained in the kernel of the density matrix of the product state ρ1⊗ρ2\rho_1 \otimes \rho_2 (i.e., ker⁡(σ1⊗σ2)⊆ker⁡(ρ1⊗ρ2)\ker(\sigma_1 \otimes \sigma_2) \subseteq \ker(\rho_1 \otimes \rho_2)) if and only if ker⁡σ1⊆ker⁡ρ1\ker \sigma_1 \subseteq \ker \rho_1 and ker⁡σ2⊆ker⁡ρ2\ker \sigma_2 \subseteq \ker \rho_2.

theorem

⟨A⊗B,C⊗D⟩=⟨A,C⟩⟨B,D⟩\langle A \otimes B, C \otimes D \rangle = \langle A, C \rangle \langle B, D \rangle

Let AA and CC be complex Hermitian matrices of dimension d1d_1, and let BB and DD be complex Hermitian matrices of dimension d2d_2. Then the Frobenius inner product of the Kronecker products A⊗BA \otimes B and C⊗DC \otimes D is equal to the product of the individual inner products, namely: ⟨A⊗B,C⊗D⟩=⟨A,C⟩⟨B,D⟩\langle A \otimes B, C \otimes D \rangle = \langle A, C \rangle \langle B, D \rangle

theorem

Continuity of r↦trr \mapsto t^r in the topology of uniform convergence on compact sets for r>0r > 0

Let K⊂RK \subset \mathbb{R} be a compact set. The function that maps a real exponent rr to the power function t↦trt \mapsto t^r (viewed as an element of the space of functions on KK endowed with the topology of uniform convergence) is continuous on the interval (0,∞)(0, \infty).

theorem

Additivity of ⟨ρ,log⁡ρ−log⁡σ⟩\langle \rho, \log \rho - \log \sigma \rangle under tensor products of mixed states

Let ρ1,σ1\rho_1, \sigma_1 be mixed states of dimension d1d_1 and ρ2,σ2\rho_2, \sigma_2 be mixed states of dimension d2d_2. Let Mρ1,Mσ1,Mρ2,Mσ2M_{\rho_1}, M_{\sigma_1}, M_{\rho_2}, M_{\sigma_2} denote their respective density matrices, and let ⊗M\otimes^M denote the tensor product of mixed states. Suppose that the kernel of Mσ1M_{\sigma_1} is a subspace of the kernel of Mρ1M_{\rho_1} (ker⁡Mσ1⊆ker⁡Mρ1\ker M_{\sigma_1} \subseteq \ker M_{\rho_1}) and the kernel of Mσ2M_{\sigma_2} is a subspace of the kernel of Mρ2M_{\rho_2} (ker⁡Mσ2⊆ker⁡Mρ2\ker M_{\sigma_2} \subseteq \ker M_{\rho_2}). Then the following additivity property for the Frobenius inner product of the logarithm of the states holds: ⟨Mρ1⊗Mρ2,log⁡(Mρ1⊗Mρ2)−log⁡(Mσ1⊗Mσ2)⟩=⟨Mρ1,log⁡Mρ1−log⁡Mσ1⟩R+⟨Mρ2,log⁡Mρ2−log⁡Mσ2⟩\langle M_{\rho_1 \otimes^M \rho_2}, \log(M_{\rho_1 \otimes^M \rho_2}) - \log(M_{\sigma_1 \otimes^M \sigma_2}) \rangle = \langle M_{\rho_1}, \log M_{\rho_1} - \log M_{\sigma_1} \rangle_{\mathbb{R}} + \langle M_{\rho_2}, \log M_{\rho_2} - \log M_{\sigma_2} \rangle where log⁡\log denotes the functional calculus for Hermitian matrices.

definition

Sandwiched Rényi Relative Entropy Dα(ρ∥σ)D_\alpha(\rho \| \sigma)

Let ρ\rho and σ\sigma be quantum states (represented by the type `MState d`) with associated density matrices MρM_\rho and MσM_\sigma. Given a parameter α∈R\alpha \in \mathbb{R}, the Sandwiched Rényi Relative Entropy Dα(ρ∥σ)D_\alpha(\rho \| \sigma) is defined as an extended non-negative real number as follows: 1. If α≤0\alpha \le 0, the value is 00. 2. If α>0\alpha > 0 and the kernel condition ker⁡Mσ≤ker⁡Mρ\ker M_\sigma \le \ker M_\rho is not satisfied, the value is ∞\infty. 3. If α=1\alpha = 1 and ker⁡Mσ≤ker⁡Mρ\ker M_\sigma \le \ker M_\rho, it returns the standard relative entropy (Kullback-Leibler divergence) calculated using the natural logarithm: D1(ρ∥σ)=Tr(Mρ(log⁡Mρ−log⁡Mσ))D_1(\rho \| \sigma) = \text{Tr}(M_\rho (\log M_\rho - \log M_\sigma)) 4. If α∈(0,1)∪(1,∞)\alpha \in (0, 1) \cup (1, \infty) and ker⁡Mσ≤ker⁡Mρ\ker M_\sigma \le \ker M_\rho, it is given by: Dα(ρ∥σ)=1α−1log⁡Tr[(Mσ1−α2αMρMσ1−α2α)α]D_\alpha(\rho \| \sigma) = \frac{1}{\alpha - 1} \log \text{Tr} \left[ \left( M_\sigma^{\frac{1-\alpha}{2\alpha}} M_\rho M_\sigma^{\frac{1-\alpha}{2\alpha}} \right)^\alpha \right] where Tr\text{Tr} denotes the trace and the power of the matrices is defined via functional calculus.

definition

Sandwiched Rényi relative entropy D~α(ρ∥σ)\tilde{D}_\alpha(\rho \| \sigma)

This notation defines the sandwiched Rényi relative entropy of order α\alpha between two mixed states ρ\rho and σ\sigma, denoted as D~α(ρ∥σ)\tilde{D}_\alpha(\rho \| \sigma). It is defined for α∈R\alpha \in \mathbb{R} and maps a pair of mixed states ρ,σ\rho, \sigma to an extended non-negative real number [0,∞][0, \infty].

definition

Quantum Relative Entropy D(ρ∥σ)D(\rho \parallel \sigma)

The quantum relative entropy, also known as the Umegaki relative entropy, is defined for two quantum states ρ\rho and σ\sigma (elements of the space of mixed states MState d\text{MState } d) as: D(ρ∥σ)=Tr[ρ(log⁡ρ−log⁡σ)]D(\rho \parallel \sigma) = \text{Tr}[\rho (\log \rho - \log \sigma)] The value is represented as an extended non-negative real number (ENNReal\text{ENNReal}). In this library, it is implemented as a special case of the sandwiched Rényi relative entropy D~α(ρ∥σ)\tilde{D}_\alpha(\rho \parallel \sigma) evaluated at the limit α→1\alpha \to 1.

definition

Notation for Quantum Relative Entropy D(ρ∥σ)\mathbf{D}(\rho \parallel \sigma)

This represents the mathematical notation D(ρ∥σ)\mathbf{D}(\rho \parallel \sigma), which denotes the quantum relative entropy between two quantum states ρ\rho and σ\sigma. It is defined formally as the value of the function `qRelativeEnt ρ σ`, which calculates the Umegaki relative entropy D(ρ∥σ)=Tr[ρ(log⁡ρ−log⁡σ)]D(\rho \parallel \sigma) = \text{Tr}[\rho (\log \rho - \log \sigma)].

theorem

Factorization of the trace of the sandwiched term over tensor products

Let ρ1,σ1\rho_1, \sigma_1 be mixed states of dimension d1d_1 and ρ2,σ2\rho_2, \sigma_2 be mixed states of dimension d2d_2. For any real numbers α\alpha and β\beta, the trace of the sandwiched term of the combined states satisfies the following factorization property: Tr⁡[((ρ1⊗ρ2)⋅(σ1⊗σ2)β⋅(ρ1⊗ρ2)†)α]=Tr⁡[(ρ1⋅σ1β⋅ρ1†)α]⋅Tr⁡[(ρ2⋅σ2β⋅ρ2†)α] \operatorname{Tr}\left[ \left( (\rho_1 \otimes \rho_2) \cdot (\sigma_1 \otimes \sigma_2)^\beta \cdot (\rho_1 \otimes \rho_2)^\dagger \right)^\alpha \right] = \operatorname{Tr}\left[ \left( \rho_1 \cdot \sigma_1^\beta \cdot \rho_1^\dagger \right)^\alpha \right] \cdot \operatorname{Tr}\left[ \left( \rho_2 \cdot \sigma_2^\beta \cdot \rho_2^\dagger \right)^\alpha \right] where M.conjM.conj denotes the conjugation operation A⋅B⋅A†A \cdot B \cdot A^\dagger, ⊗\otimes is the tensor product of mixed states, and MM is the density matrix associated with a mixed state.

theorem

Additivity of Sandwiched Rényi Relative Entropy for α≠1\alpha \neq 1

For any real parameter α≠1\alpha \neq 1 and any quantum mixed states ρ1,σ1\rho_1, \sigma_1 on a Hilbert space of dimension d1d_1 and ρ2,σ2\rho_2, \sigma_2 on a Hilbert space of dimension d2d_2, the sandwiched Rényi relative entropy D~α\tilde{D}_\alpha is additive under the tensor product. That is, D~α(ρ1⊗ρ2∥σ1⊗σ2)=D~α(ρ1∥σ1)+D~α(ρ2∥σ2),\tilde{D}_\alpha(\rho_1 \otimes \rho_2 \parallel \sigma_1 \otimes \sigma_2) = \tilde{D}_\alpha(\rho_1 \parallel \sigma_1) + \tilde{D}_\alpha(\rho_2 \parallel \sigma_2), where ⊗\otimes denotes the tensor product of the mixed states.

theorem

Additivity of Sandwiched Rényi Relative Entropy for Product States

For any Renyi parameter α\alpha and mixed states ρ1,σ1\rho_1, \sigma_1 on a system of dimension d1d_1 and ρ2,σ2\rho_2, \sigma_2 on a system of dimension d2d_2, the sandwiched Rényi relative entropy D~α\tilde{D}_\alpha is additive under the tensor product of states: D~α(ρ1⊗ρ2 ∥ σ1⊗σ2)=D~α(ρ1 ∥ σ1)+D~α(ρ2 ∥ σ2)\tilde{D}_\alpha(\rho_1 \otimes \rho_2 \,\|\, \sigma_1 \otimes \sigma_2) = \tilde{D}_\alpha(\rho_1 \,\|\, \sigma_1) + \tilde{D}_\alpha(\rho_2 \,\|\, \sigma_2) where ⊗\otimes denotes the tensor product of mixed states.

theorem

Additivity of Quantum Relative Entropy for Product States D(ρ1⊗ρ2∥σ1⊗σ2)=D(ρ1∥σ1)+D(ρ2∥σ2)D(\rho_1 \otimes \rho_2 \parallel \sigma_1 \otimes \sigma_2) = D(\rho_1 \parallel \sigma_1) + D(\rho_2 \parallel \sigma_2)

For any quantum mixed states ρ1,σ1\rho_1, \sigma_1 on a Hilbert space of dimension d1d_1, and any quantum mixed states ρ2,σ2\rho_2, \sigma_2 on a Hilbert space of dimension d2d_2, the quantum relative entropy DD is additive under the tensor product of states. Specifically, D(ρ1⊗ρ2∥σ1⊗σ2)=D(ρ1∥σ1)+D(ρ2∥σ2)D(\rho_1 \otimes \rho_2 \parallel \sigma_1 \otimes \sigma_2) = D(\rho_1 \parallel \sigma_1) + D(\rho_2 \parallel \sigma_2) where ⊗\otimes denotes the tensor product of the mixed states.

theorem

Invariance of Sandwiched Rényi Relative Entropy under basis relabeling

For any quantum mixed states ρ,σ\rho, \sigma defined on a finite-dimensional Hilbert space with basis dd, and for any bijection e:d2≃de: d_2 \simeq d between basis sets, the sandwiched Rényi relative entropy of order α\alpha is invariant under relabeling, such that D~α(ρ′∥σ′)=D~α(ρ∥σ)\tilde{D}_\alpha(\rho' \| \sigma') = \tilde{D}_\alpha(\rho \| \sigma), where ρ′\rho' and σ′\sigma' are the states relabeled by ee.

theorem

D~α(ρ∥ρ)=0\tilde{D}_{\alpha}(\rho \parallel \rho) = 0 for α>0\alpha > 0

For any quantum state ρ\rho and any real parameter α>0\alpha > 0, the sandwiched Rényi relative entropy of the state with itself, denoted as D~α(ρ∥ρ)\tilde{D}_{\alpha}(\rho \parallel \rho), is equal to 0.

theorem

D~α(ρ∥σ)≠∞\tilde{D}_{\alpha}(\rho \| \sigma) \neq \infty for non-singular σ\sigma

Let ρ\rho and σ\sigma be quantum mixed states on a finite-dimensional Hilbert space. If the density matrix of σ\sigma (denoted σ.M\sigma.M) is non-singular (invertible), then for any real order α\alpha, the sandwiched Rényi relative entropy D~α(ρ∥σ)\tilde{D}_{\alpha}(\rho \| \sigma) is not equal to positive infinity (∞\infty).

theorem

The function α↦1−α2α\alpha \mapsto \frac{1-\alpha}{2\alpha} is continuous on (0,∞)(0, \infty)

The function f:R→Rf: \mathbb{R} \to \mathbb{R} defined by f(α)=1−α2αf(\alpha) = \frac{1 - \alpha}{2\alpha} is continuous on the interval (0,∞)(0, \infty).

theorem

Continuity of r↦zrr \mapsto z^r on (0,∞)(0, \infty)

For any complex number z∈Cz \in \mathbb{C}, the function f:R→Cf: \mathbb{R} \to \mathbb{C} defined by f(r)=zrf(r) = z^r is continuous on the interval (0,∞)(0, \infty). Here, zrz^r denotes the complex power function, and (0,∞)(0, \infty) is the set of positive real numbers.

theorem

α↦1−α2α\alpha \mapsto \frac{1-\alpha}{2\alpha} maps (1,∞)(1, \infty) to (−∞,0)(-\infty, 0)

The function f:R→Rf: \mathbb{R} \to \mathbb{R} defined by α↦1−α2α\alpha \mapsto \frac{1 - \alpha}{2\alpha} maps the interval (1,∞)(1, \infty) into the interval (−∞,0)(-\infty, 0).

theorem

The frontier of a singleton {p}\{p\} in a perfect T1T_1 space is {p}\{p\}

Let XX be a topological space that satisfies the T1T_1 separation axiom and is a perfect space (meaning it has no isolated points). For any point p∈Xp \in X, the frontier (boundary) of the singleton set {p}\{p\} is equal to the set {p}\{p\} itself.

theorem

Continuity of D~α(ρ∥σ)\tilde{D}_\alpha(\rho\|\sigma) for α>0\alpha > 0

Let ρ\rho and σ\sigma be mixed states of a dd-dimensional Hilbert space (MState dMState\, d). The sandwiched Rényi relative entropy D~α(ρ∥σ)\tilde{D}_\alpha(\rho\|\sigma), considered as a function of the order parameter α\alpha, is continuous on the interval (0,∞)(0, \infty).

theorem

Quantum Relative Entropy Formula for ker⁡σ⊆ker⁡ρ\ker \sigma \subseteq \ker \rho

For any two quantum states ρ\rho and σ\sigma in a dd-dimensional Hilbert space, if the kernel of the density operator σ.M\sigma.M is contained within the kernel of the density operator ρ.M\rho.M (denoted ker⁡σ⊆ker⁡ρ\ker \sigma \subseteq \ker \rho), then the quantum relative entropy D(ρ∥σ)D(\rho \parallel \sigma) is given by the inner product ⟨ρ,log⁡ρ−log⁡σ⟩\langle \rho, \log \rho - \log \sigma \rangle. In the context of operators, this corresponds to the trace formula Tr[ρ(log⁡ρ−log⁡σ)]\text{Tr}[\rho(\log \rho - \log \sigma)].

theorem

D(ρ∥σ)=−Svn(ρ)−⟨ρ,log⁡σ⟩D(\rho \parallel \sigma) = -S_{vn}(\rho) - \langle \rho, \log \sigma \rangle if ker⁡σ⊆ker⁡ρ\ker \sigma \subseteq \ker \rho else ∞\infty

For any two quantum mixed states ρ\rho and σ\sigma of dimension dd, the quantum relative entropy D(ρ∥σ)D(\rho \parallel \sigma) relates to the von Neumann entropy Svn(ρ)S_{vn}(\rho) as follows: D(ρ∥σ)=−Svn(ρ)+{−⟨ρ,log⁡σ⟩HSif ker⁡σ⊆ker⁡ρ∞otherwiseD(\rho \parallel \sigma) = -S_{vn}(\rho) + \begin{cases} -\langle \rho, \log \sigma \rangle_{HS} & \text{if } \ker \sigma \subseteq \ker \rho \\ \infty & \text{otherwise} \end{cases} where ⟨⋅,⋅⟩HS\langle \cdot, \cdot \rangle_{HS} denotes the Hilbert-Schmidt inner product and ker⁡σ⊆ker⁡ρ\ker \sigma \subseteq \ker \rho indicates that the kernel of the density operator σ\sigma is contained within the kernel of ρ\rho.

theorem

Quantum Relative Entropy is Invariant Under Relabeling

Let ρ\rho and σ\sigma be quantum mixed states on a system with basis dd. For any bijection (equivalence) e:d2≃de : d_2 \simeq d, let ρ.relabel e\rho.\text{relabel } e and σ.relabel e\sigma.\text{relabel } e be the mixed states on the system with basis d2d_2 obtained by reindexing the density matrices. Then the quantum relative entropy D(ρ∥σ)D(\rho \parallel \sigma) is invariant under such relabeling, i.e., D(ρ.relabel e∥σ.relabel e)=D(ρ∥σ)D(\rho.\text{relabel } e \parallel \sigma.\text{relabel } e) = D(\rho \parallel \sigma).

theorem

D~α(ρ‖σ)=0D̃_{\alpha}(\rho‖\sigma) = 0 for unique states on a 1D system

Let dd be a type with exactly one element (representing a 1-dimensional Hilbert space). For any two mixed states ρ\rho and σ\sigma in MState d\text{MState } d, the sandwiched Rényi relative entropy of order α\alpha, denoted D~α(ρ‖σ)D̃_{\alpha}(\rho‖\sigma), is equal to 0.

theorem

Quantum Relative Entropy of States on a Unique Type is 0

Let dd be a type with a unique element (i.e., a singleton type). For any two mixed states ρ\rho and σ\sigma defined over dd, the quantum relative entropy D(ρ∥σ)\mathbf{D}(\rho\|\sigma) is equal to 00.

theorem

D~α(ρ1∥σ1)=D~α(ρ2∥σ2)\tilde{D}_\alpha(\rho_1 \| \sigma_1) = \tilde{D}_\alpha(\rho_2 \| \sigma_2) for heterogeneously equal states and types

Let d1d_1 and d2d_2 be finite types with decidable equality, representing the index sets of two quantum systems. Let ρ1,σ1\rho_1, \sigma_1 be mixed states on the system indexed by d1d_1, and ρ2,σ2\rho_2, \sigma_2 be mixed states on the system indexed by d2d_2. If the index sets are equal (d1=d2d_1 = d_2) and the states are heterogeneously equal (denoted by ρ1≅ρ2\rho_1 \cong \rho_2 and σ1≅σ2\sigma_1 \cong \sigma_2), then their sandwiched Rényi relative entropies of order α\alpha are equal, i.e., D~α(ρ1∥σ1)=D~α(ρ2∥σ2)\tilde{D}_\alpha(\rho_1 \| \sigma_1) = \tilde{D}_\alpha(\rho_2 \| \sigma_2).

theorem

Invariance of Sandwiched Rényi Relative Entropy under Basis Relabeling (D~α(ρ1∥σ1)=D~α(ρ2∥σ2)\tilde{D}_\alpha(\rho_1 \| \sigma_1) = \tilde{D}_\alpha(\rho_2 \| \sigma_2))

Given two finite dimensional quantum systems with basis types d1d_1 and d2d_2 and states ρ1,σ1\rho_1, \sigma_1 on d1d_1 and ρ2,σ2\rho_2, \sigma_2 on d2d_2, if there is an equality between the basis types d1=d2d_1 = d_2 such that the states are related by the corresponding relabeling (ρ1=relabel(ρ2)\rho_1 = \text{relabel}(\rho_2) and σ1=relabel(σ2)\sigma_1 = \text{relabel}(\sigma_2)), then the sandwiched Rényi relative entropies are equal: D~α(ρ1∥σ1)=D~α(ρ2∥σ2)\tilde{D}_\alpha(\rho_1 \| \sigma_1) = \tilde{D}_\alpha(\rho_2 \| \sigma_2).

theorem

Quantum Relative Entropy is Invariant under Heterogeneous Equality of States

Let d1d_1 and d2d_2 be finite types. Let ρ1,σ1\rho_1, \sigma_1 be mixed states (of type `MState`) acting on a Hilbert space indexed by d1d_1, and ρ2,σ2\rho_2, \sigma_2 be mixed states acting on a Hilbert space indexed by d2d_2. If the underlying types are equal (d1=d2d_1 = d_2) and the states are heterogeneous equal (denoted by ρ1≅ρ2\rho_1 \cong \rho_2 and σ1≅σ2\sigma_1 \cong \sigma_2), then the quantum relative entropies are equal: D(ρ1∥σ1)=D(ρ2∥σ2)D(\rho_1 \Vert \sigma_1) = D(\rho_2 \Vert \sigma_2).

theorem

Quantum Relative Entropy for Full Rank σ\sigma acts as ⟨ρ,log⁡ρ−log⁡σ⟩\langle \rho, \log \rho - \log \sigma \rangle

Let ρ\rho and σ\sigma be quantum states (represented as density matrices in MState dMState \ d). If σ\sigma is non-singular (i.e., it has full rank), then the quantum relative entropy D(ρ∥σ)D(\rho \| \sigma) is given by the inner product ⟨ρ,log⁡ρ−log⁡σ⟩\langle \rho, \log \rho - \log \sigma \rangle where log⁡\log denotes the matrix logarithm and ⟨A,B⟩=Tr(A∗B)\langle A, B \rangle = \text{Tr}(A^* B) is the Hilbert-Schmidt inner product.

theorem

Upper Semicontinuity of ⟨ρ,log⁡σ⟩\langle \rho, \log \sigma \rangle under Kernel Constraint

Let ρ\rho and σ\sigma be mixed states in MState d\text{MState } d, where MM denotes the underlying density matrix. Suppose the kernel of σ\sigma is contained in the kernel of ρ\rho (ker⁡σ⊆ker⁡ρ\ker \sigma \subseteq \ker \rho). If the Hilbert-Schmidt inner product of ρ\rho and the matrix logarithm of σ\sigma is strictly less than some real value yy (i.e., ⟨ρ,log⁡σ⟩<y\langle \rho, \log \sigma \rangle < y), then for all mixed states xx in a sufficiently small neighborhood of σ\sigma, if ker⁡x⊆ker⁡ρ\ker x \subseteq \ker \rho, then ⟨ρ,log⁡x⟩<y\langle \rho, \log x \rangle < y.

theorem

Lower Semicontinuity of Quantum Relative Entropy at States with Mismatched Kernels

Let ρ\rho and xx be mixed states in MState d\text{MState } d such that the kernel of the matrix xx is not contained in the kernel of the matrix ρ\rho (ker⁡x≰ker⁡ρ\ker x \not\le \ker \rho). For any value y∈[0,∞]y \in [0, \infty], if y<∞y < \infty, then there exists a neighborhood of xx such that for all mixed states x′x' in that neighborhood, the value yy is strictly less than the quantity defined by: {⟨ρ,log⁡ρ−log⁡x′⟩if ker⁡x′≤ker⁡ρ∞otherwise \begin{cases} \langle \rho, \log \rho - \log x' \rangle & \text{if } \ker x' \le \ker \rho \\ \infty & \text{otherwise} \end{cases} where the expression is treated as an element of the extended real numbers R‾\overline{\mathbb{R}}.

theorem

Lower Semicontinuity of Quantum Relative Entropy D(ρ∥σ)\mathbf{D}(\rho \| \sigma) in its Second Argument

For any fixed mixed state ρ\rho of dimension dd, the quantum relative entropy D(ρ∥σ)\mathbf{D}(\rho \| \sigma) is lower semicontinuous as a function of the mixed state σ\sigma. That is, for every σ∈MState d\sigma \in \text{MState } d, and for any y<D(ρ∥σ)y < \mathbf{D}(\rho \| \sigma), there exists a neighborhood of σ\sigma such that for all σ′\sigma' in that neighborhood, y<D(ρ∥σ′)y < \mathbf{D}(\rho \| \sigma').

theorem

Joint Convexity of Quantum Relative Entropy D(ρ∥σ)D(\rho \| \sigma)

For any quantum states ρ1,ρ2,σ1,σ2\rho_1, \rho_2, \sigma_1, \sigma_2 in the state space MStated\text{MState}_d and any probability p∈[0,1]p \in [0, 1], the quantum relative entropy D(⋅∥⋅)D(\cdot \| \cdot) satisfies joint convexity: D(pρ1+(1−p)ρ2∥pσ1+(1−p)σ2)≤pD(ρ1∥σ1)+(1−p)D(ρ2∥σ2)D(p\rho_1 + (1-p)\rho_2 \| p\sigma_1 + (1-p)\sigma_2) \le p D(\rho_1 \| \sigma_1) + (1-p) D(\rho_2 \| \sigma_2) where p[ρ1↔ρ2]p [\rho_1 \leftrightarrow \rho_2] denotes the convex mixture (or "mix") of the states ρ1\rho_1 and ρ2\rho_2 with weights pp and 1−p1-p respectively.

theorem

D(ρ∥ρ)=0D(\rho \Vert \rho) = 0

For any mixed state ρ\rho in a dd-dimensional Hilbert space, the quantum relative entropy of ρ\rho with respect to itself is zero, denoted as D(ρ∥ρ)=0D(\rho \Vert \rho) = 0.

theorem

Quantum Relative Entropy with Non-Singular σ\sigma is Finite

Let ρ\rho and σ\sigma be mixed states of dimension dd. If σ\sigma is non-singular (i.e., its density matrix is invertible), then the quantum relative entropy D(ρ∥σ)D(\rho\|\sigma) is not equal to ∞\infty (it is finite in the extended non-negative real numbers R‾≥0\overline{\mathbb{R}}_{\ge 0}).

theorem

I(A:B)=D(ρ∥ρA⊗ρB)I(A:B) = D(\rho \parallel \rho^A \otimes \rho^B)

For a bipartite quantum mixed state ρ\rho on a composite system with dimensions dA×dBd_A \times d_B, the quantum mutual information I(A:B)I(A:B) is equal to the quantum relative entropy between ρ\rho and the tensor product of its reduced density matrices, ρA⊗ρB\rho^A \otimes \rho^B. That is, I(A:B)=D(ρ∥ρA⊗ρB)I(A:B) = D(\rho \parallel \rho^A \otimes \rho^B) where ρA=TrB(ρ)\rho^A = \text{Tr}_B(\rho) and ρB=TrA(ρ)\rho^B = \text{Tr}_A(\rho) are the partial traces over the right and left subsystems, respectively.

theorem

D(ρ∥σ1)≤D(ρ∥σ2)+log⁡αD(\rho \parallel \sigma_1) \le D(\rho \parallel \sigma_2) + \log \alpha if σ2≤ασ1\sigma_2 \le \alpha \sigma_1

Let ρ\rho, σ1\sigma_1, and σ2\sigma_2 be quantum mixed states in a dd-dimensional Hilbert space. Let Mσ1M_{\sigma_1} and Mσ2M_{\sigma_2} denote the density operators of σ1\sigma_1 and σ2\sigma_2, respectively. If Mσ2M_{\sigma_2} is bounded by a scaling of Mσ1M_{\sigma_1} such that Mσ2≤α⋅Mσ1M_{\sigma_2} \le \alpha \cdot M_{\sigma_1} for some scalar α∈R\alpha \in \mathbb{R}, then the quantum relative entropy satisfies the inequality D(ρ∥σ1)≤D(ρ∥σ2)+log⁡αD(\rho \parallel \sigma_1) \le D(\rho \parallel \sigma_2) + \log \alpha.

theorem

Quantum Relative Entropy is bounded by the Logarithm of the Dominance Ratio α\alpha

Let ρ\rho and σ\sigma be mixed states of dimension dd, and let MρM_\rho and MσM_\sigma denote their respective density matrices. If Mρ≤α⋅MσM_\rho \le \alpha \cdot M_\sigma for some scalar α∈R\alpha \in \mathbb{R}, then the quantum relative entropy D(ρ∥σ)D(\rho \| \sigma) is bounded by log⁡α\log \alpha, specifically D(ρ∥σ)≤ENNReal.ofReal(log⁡α)D(\rho \| \sigma) \le \text{ENNReal.ofReal}(\log \alpha).