Operator-valued maximal f-divergences for completely positive maps
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.
Ada: I'm Ada.
Emmy: And I'm Emmy, here with Kolya.
Ada: We're looking at the paper: "Operator-valued maximal f-divergences for completely positive maps".
Emmy: Here's the gist of the paper. Operator-valued maximal f-divergences for completely positive maps introduces an operator-valued maximal f-divergence for completely positive (CP) maps between von Neumann algebras, associated with an operator convex function on (0, +∞) The construction takes values in the extended lower-semibounded self-adjoint part of the codomain von Neumann algebra For normal positive functionals, our construction recovers Hiai’s maximal fdivergence As a consequence, we establish its joint weak lower semicontinuity for arbitrary von Neumann algebras, answering a question left open by Hiai Definition and Construction of the Divergence The operator-valued maximal f-divergence is defined as Sbf (Φ∥Ψ):= V∗ϕf (A′, B′)V, where (π, V, K; A′, B′) is a minimal Stinespring–Radon–Nikodym (SRN) realization of the pair (Φ, Ψ), and ϕf is the perspective associated with f The divergence takes values in the extended lower-semibounded self-adjoint part Nb lb introduced in
HUW: This construction can be computed using any common CP upper bound of Φ and Ψ, up to positive scalar multiples Fundamental Properties of the Operator-Valued Divergence The paper establishes several fundamental properties for this operator-valued maximal f-divergence These include: 1.
Ada: Joint subadditivity: Sbf Xn i=1 Φi Xn i=1 Ψi!≤ Xn i=1 Sbf (Φi∥Ψi) 2.
Emmy: Transpose relation: Sb fe(Φ∥Ψ) = Sbf (Ψ∥Φ) 3.
Ada: Monotonicity under unital precomposition and normal postcomposition by CP maps 4.
Paper summary: Ada: So to wrap up this paper "Operator-valued maximal f-divergences for completely positive maps", what's the real implication for our field?
Emmy: The core idea is that we have built a tool that gives us an operator-valued measure of divergence between these complex maps, and it works across a wider class of algebras than before.
Kolya: It seems like this construction provides joint weak lower semicontinuity for arbitrary von Neumann algebras, which was something Hiai left open. That’s a significant piece of information because it helps us understand the limiting behavior of these divergences in general settings.
Ada: Exactly. And they showed that for normal positive functionals, we get back to Hiai's divergence, and then they extend that joint lower semicontinuity property to all von Neumann algebras. That answers a question about how these divergences behave under weak convergence when the maps are not even normal.
Emmy: Plus, they establish a direct link between their operator-valued divergence and the Hollands–Ranallo BS channel divergence for normal channels with sigma finite codomain, which ties together several different concepts we've been studying.
Kolya: It also connects this to the Jones index formula for finite index inclusions of factors, which is important for understanding how these divergences relate to the structure of sub-algebras.
Ada: So in simple terms, they've introduced a robust framework that gives us a way to measure divergence that respects the structure of operator algebras and behaves predictably under various map transformations. It’s about making sure our measurements stay sensible even when we step outside the realm of normal maps.
Conclusion: Ada: So, we're wrapping up this discussion on operator valued maximal f divergences for completely positive maps. Basically, these authors have put together a new way to measure how different completely positive maps are between von Neumann algebras using an operator perspective on some function called f.
Emmy: They take this construction and show that it works even when you're dealing with general von Neumann algebras, not just the normal ones we usually see. The main thing they prove is that this divergence has joint weak lower semicontinuity, which is a big deal for understanding how these divergences behave as you let maps get closer to each other in a certain sense.
Kolya: I mean, what they did is recover Hiai’s maximal f divergence when the maps are normal positive functionals, but they push that result out to arbitrary algebras. That's where it gets interesting—it means this framework has broader applicability than we thought for these kinds of measurements.
Ada: Exactly. It’s about extending a specific measurement technique to cover a much wider mathematical landscape, and proving its stability under weak limits is crucial because it gives us more confidence in using these divergences for anything.
Emmy: And they link this new divergence directly to the Hollands–Ranallo BS channel divergence for normal channels with sigma finite codomain, which connects the abstract operator stuff to something more concrete in quantum information theory.
Kolya: It also relates it back to that Jones index formula for inclusions of factors, which gives us a way to calculate these divergences based on the structure of the sub-algebra itself.
Ada: So what this paper really boils down to is a unified framework that connects abstract operator theory with concrete measures of divergence in map theory, and proving it’s well-behaved across different algebra types is the key achievement here.
Rui Okayasu
math.OA, math-ph, math.FA, math.MP
Submitted: 2026-09-01
Updated: 2026-11-16
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 89/100
The gist: Operator-valued maximal f-divergences for completely positive maps introduces an operator-valued maximal f-divergence for completely positive (CP) maps between von Neumann algebras…
Key concepts
- Operator-valued maximal f-divergence
- This is a new measure of distance between two CP maps, $\Phi$ and $\Psi$, defined using a specific construction involving their minimal Stinespring–Radon–Nikodym realization. It operates on operators and takes values in the extended lower-semibounded self-adjoint part of the target von Neumann algebra.
- Stinespring–Radon–Nikodym (SRN) realization
- This is a mathematical tool used to represent a pair of CP maps $(\Phi, \Psi)$ in terms of an operator $V$ and related structures. It provides a concrete way to compute the divergence by relating the maps to this specific realization.
- Extended norm (||m||ext)
- This is a specific way to measure the size or bound of an operator-valued quantity $m$. It is defined as the supremum over all positive normal states $\varphi$ on a subalgebra $N$, ensuring that the resulting value reflects how large the operator-valued divergence can be relative to certain states.
- Jones–Kosaki Index
- This index, denoted by $\lambda$, is associated with an inclusion of von Neumann algebras $N \subseteq M$. It is a finite number when the inclusion has finite index, and it appears in formulas relating the divergence to the ratio of algebra sizes.
Terminology
Summary
Operator-valued maximal f-divergences for completely positive maps introduces an operator-valued maximal f-divergence for completely positive (CP) maps between von Neumann algebras, associated with an operator convex function on (0, +∞) The construction takes values in the extended lower-semibounded self-adjoint part of the codomain von Neumann algebra For normal positive functionals, our construction recovers Hiai’s maximal fdivergence As a consequence, we establish its joint weak lower semicontinuity for arbitrary von Neumann algebras, answering a question left open by Hiai
Definition and Construction of the Divergence
The operator-valued maximal f-divergence is defined as Sbf (Φ∥Ψ):= V∗ϕf (A′, B′)V, where (π, V, K; A′, B′) is a minimal Stinespring–Radon–Nikodym (SRN) realization of the pair (Φ, Ψ), and ϕf is the perspective associated with f The divergence takes values in the extended lower-semibounded self-adjoint part Nb lb introduced in [HUW] This construction can be computed using any common CP upper bound of Φ and Ψ, up to positive scalar multiples
Fundamental Properties of the Operator-Valued Divergence
The paper establishes several fundamental properties for this operator-valued maximal f-divergence These include:
-
Joint subadditivity: Sbf Xn i=1 Φi Xn i=1 Ψi! p. 9≤ Xn i=1 Sbf (Φi∥Ψi)
-
Transpose relation: Sb fe(Φ∥Ψ) = Sbf (Ψ∥Φ)
-
Monotonicity under unital precomposition and normal postcomposition by CP maps
-
A martingale convergence theorem for normal CP maps
Recovery of Known Divergences and Specific Cases
The construction recovers Hiai’s maximal fdivergence for normal positive functionals For the function η(t) = tlog t, we obtain an operator-valued Belavkin–Staszewski (BS) relative entropy Moreover, for normal channels with σ-finite codomain, the BS channel divergence of Hollands and Ranallo coincides with the extended norm of our operator-valued divergence For finite-index conditional expectations E: M → N where N ⊆ M is an inclusion of σ-finite factors, Sbf (idM∥E) = f(λ) / λ + (1 − 1 / λ) f(+0) 1M in Mb lb
Characterization of Domain and Boundedness
The paper provides criteria for when Sbf (Φ∥Ψ) has a dense domain and is bounded These criteria are formulated in terms of the boundary values f(+0) and f′(+∞) and the absolute continuity relations between Φ and Ψ using Lebesgue-type decomposition of CP maps For instance, if f′(+∞) = +∞ > f(+0), Sbf (Φ∥Ψ) has a dense domain if and only if Φ ≪ Ψ and dom(Q′1/2 1 V) is dense in H
Agreement with Hollands–Ranallo Divergence
The operator-valued BS relative entropy D op BS(Φ∥Ψ) coincides with the extended norm of our operator-valued divergence for normal channels with σ-finite codomain This is shown by proving D HR BS (Φ∥Ψ) = D op BS(Φ∥Ψ)ext The finite-dimensional examples show that for Schur multipliers, Sbf (Φ∥Ψ)(Trn) = Trn ϕf (A, B), and for congruence maps where A and B are not proportional, the divergence is determined by the boundary values of the perspective: Sbf (Φ∥Ψ)(Trn) = f′(+∞)A + f(+0)B
Connection to Hiai’s Divergence
Under the identification of normal positive functionals with normal CP maps from M to C, Theorems 4.7 and 5.1 imply that (φ, ψ) → SbHiai f(φ∥ψ) is jointly lower semicontinuous with respect to the product σ(M∗, M)-topology Furthermore, Theorems 4. p. 27 and 5.1 imply that (φ, ψ) → SbHiai f(φ∥ψ) is jointly lower semicontinuous with respect to the product σ(M∗, M)-topology
Finite-Index Conditional Expectations
For an inclusion of σ-finite factors N ⊆ M with a faithful normal conditional expectation E: M → N and finite Jones–Kosaki index λ:= Ind E < +∞, Sbf (idM∥E) = f(λ) / λ + (1 − 1 / λ) f(+0) 1M in Mb lb If f(+0) = 0, then D op BS(idM∥E) = (log Ind E) 1M
Comparison with Channel Divergence
The Hollands–Ranallo BS channel divergence is bounded by the extended norm of the operator-valued divergence, D HR BS (Φ∥Ψ) ≤ D op BS(Φ∥Ψ)ext This inequality holds for every von Neumann algebra A, every binormal unital ∗-representation θ: N ⊙ Aop → B(Hθ), and every unit vector ξ ∈ Hθ
The Extended Norm Definition
For m ∈ Nb +, its extended norm is defined as mext:= sup s m(φ) φ ∈ N+∗, φ(1N) = 1 This is equivalent to mext = inf c ≥ 0 m ≤ c1N
For a faithful normal state ψ on N, DHR BS (Φ∥Ψ) ≥ m(ψ) for every faithful normal state ψ on N If m(ψ0) = +∞, then DHR BS (Φ∥Ψ) = +∞
The Jones–Kosaki Index Formula
For an inclusion of σfinite factors with finite index, the divergence is DHR BS (idM∥E0) = log[M: N] This result is derived from the fact that D op BS(idM∥E0) = log[M: N] 1M
The Hollands–Ranallo Variational Formula
The Hollands–Ranallo variational formula (8.1) provides an upper bound for the divergence: D B BS(φ∥ψ) ≤ ϕη(C, D)(ωζ) The equality holds when ζ is cyclic for ρ(B), and the supremum over all admissible pairs yields the exact value of the divergence
The Hollands–Ranallo BS Channel Divergence
The Hollands–Ranallo BS channel divergence is defined as D HR BS (Φ∥Ψ):= sup (A,θ,ξ) D B BS (ωΦ,θ,ξ∥ωΨ,θ,ξ), where the supremum is taken over all von Neumann algebras A all binormal unital ∗-representations θ as above and all unit vectors ξ ∈ Hθ The result is that D HR BS (Φ∥Ψ) ≤ c if D op BS(Φ∥Ψ) ≤ c1N
The Extended Norm Equivalence
The operator-valued maximal f-divergence is equal to the extended norm of the operator perspective, D HR BS (Φ∥Ψ) = D op BS(Φ∥Ψ)ext This equivalence is proven by showing that m is positive and then proving the reverse inequality using a sequence of states
For an inclusion of σfinite factors with finite index, DHR BS (idM∥E0) = log[M: N] This is derived from the fact that D op BS(idM∥E0) = log[M: N] 1M
The Operator-Valued Belavkin–Staszewski Relative Entropy
The operator-valued Belavkin–Staszewski relative entropy is defined by D op BS(Φ∥Ψ):= Sbη(Φ∥Ψ) ∈ Nb lb For Schur multipliers, D op BS(Φ∥Ψ)(Trn) = Trn ϕf (A, B), which recovers the usual matrix maximal f-divergence [Mat, Hia]
The Congruence Map Divergence
For congruence maps in Example 6.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Enhanced ability to compute operator-valued maximal f-divergences for completely positive maps between von Neumann algebras by leveraging independence results such as
independence of the common CP upper bound used in its computation.
-
Capability to recover scalar divergences from operator-valued ones, specifically demonstrating that
the BS channel divergence of Hollands and Ranallo coincides with the extended norm of our operator-valued divergence
for normal channels with sigma-finite codomain. -
Precision in handling conditional expectations by providing an explicit formula for a finite-index conditional expectation, which
reduces to the logarithm of the Jones–Kosaki index
in the BS case. -
Extension to arbitrary von Neumann algebras by establishing
joint weak lower semicontinuity
for operator-valued maximal f-divergences, answering questions left open by Hiai. -
Ability to calculate operator-valued Belavkin–Staszewski relative entropy, denoted as D op BS(Phi Psi), which is defined via the perspective function phi f and can be computed using
the weighted geometric means of CP maps developed in [Oka].
-
Robustness in analyzing map monotonicity by proving
precomposition monotonicity
andpostcomposition monotonicity,
ensuring that Sbf (Phi o Theta Psi o Theta) <= Sbf (Phi Psi). -
Functionality to handle complex operator structures through the use of the extended Pusz–Woronowicz functional calculus, which allows for the definition of perspective functions phi f for arbitrary operators A and B in B(K)+.
-
Integration of index theory with quantum divergences by relating Sbf (idM E) to
the logarithm of the Jones–Kosaki index
when considering a faithful normal conditional expectation E with finite Jones–Kosaki index lambda.
Sources
Related papers
- Cocycle perturbations and ergodicity for actions on type III factors
- The Dynamical Radius of Comparison for C*-Dynamical Systems
- C* -selflessness for free Burnside groups
- Kac lemma on noncommutative probability spaces
- The generator problem for C*-algebras with comparison
- Operator space theory from the maximal viewpoint