Conditional Mutual Information and Quantum Markov Structure

Conditional mutual information measures the correlation between $A$ and $C$ that remains when $B$ is available. Its vanishing characterizes an exact quantum Markov structure for a compatible tripartite state; smallness supports approximate recovery, not a classical time-local Markov process.

Required background. Use Strong Subadditivity and Entropic Inequalities. Helpful background. Markov generators and semigroups clarify the different dynamical usage of “Markov,” while Mutual Information and Regulator-Independent Correlations fixes the algebraic correlation language.

Conditional correlation

For a compatible state $\rho_{ABC}$,

$$I(A{:}C\mid B)_\rho =S(AB)+S(BC)-S(B)-S(ABC)\geq0.$$

The ordering matters: $B$ is the retained mediator, and recovery will reconstruct $BC$ from $B$. In a regulated QFT, all four entropies must use one state, cutoff, and subsystem prescription. Algebraic formulations use relative modular objects and conditional expectations rather than assuming a sharp tensor product.

The structural diagram locates conditional mutual information at the junction between correlations and recovery.

$I(A{:}C\mid B)$ quantifies residual $A$–$C$ correlation given $B$. Zero gives exact Markov structure in the theorem’s domain; small values motivate a separate recovery task with specified fidelity and channel domain. Schematic.

Exact and approximate Markov structure

For a faithful finite-dimensional state, $I(A{:}C\mid B)=0$ is equivalent to the existence of a channel $\mathcal R_{B\to BC}$ such that

$$\rho_{ABC}=({\rm id}_A\otimes\mathcal R_{B\to BC})(\rho_{AB}).$$

It also yields a direct-sum tensor decomposition of $B$ into sectors in which $A$ and $C$ are conditionally independent, as characterized by Hayden, Jozsa, Petz, and Winter 2004, pp. 359–374. This is a structural statement about a state and its algebras, not about memoryless evolution in time.

In QFT, nearly vanishing conditional mutual information can arise for specially ordered regions, including controlled null-surface limits. The scale matters: a dimensionless value must be compared with the desired error tolerance, copy number, and observable class. A small number alone does not imply a local, low-energy, or experimentally implementable map.

For three contiguous intervals, label the middle interval $B$ before forming unions. Reordering regions changes which correlations are conditioned upon. Taking continuum limits term by term with different endpoint regulators can spoil ultraviolet cancellation.

Domain checks

The failure map makes the exact tripartite matching requirements visible.

Exact or approximate Markov claims concern one ordered tripartite state. Changing the middle algebra, using nonmatching regulators, or silently postselecting changes the conditional information and the recovery problem. Schematic.

Before calling a QFT state Markovian, specify the ordered algebras, whether the statement is exact or approximate, the limiting prescription, and the metric in which recovery is assessed.

References

• Hayden, Patrick, Richard Jozsa, Dénes Petz, and Andreas Winter. “Structure of States Which Satisfy Strong Subadditivity of Quantum Entropy with Equality.” Communications in Mathematical Physics 246 (2004): 359–374. DOI. Open preprint.

Further reading

• Lieb, Elliott H., and Mary Beth Ruskai. “Proof of the Strong Subadditivity of Quantum-Mechanical Entropy.” Journal of Mathematical Physics 14 (1973): 1938–1941. DOI.