---
title: Virtual Quantum Markov Chains
url: https://www.emergentmind.com/topics/virtual-quantum-markov-chains-vqmcs
type: topic
---

# Virtual Quantum Markov Chains

Searching arXiv for the cited VQMC papers and closely related context.
I’ll look up the relevant arXiv entries to ground the article in the latest papers.
Virtual quantum Markov chains (VQMCs) are a generalization of quantum Markov chains in which the central recoverability requirement is shifted from exact physical state reconstruction to the recovery of measurement statistics of the global system from a subsystem by local operations and post-processing. Introduced for tripartite states in "Virtual Quantum Markov Chains" [2312.02031], the framework formalizes when arbitrary global shadow information can be recovered from local data, even when the recovery map is not completely positive. The extension to four-qubit systems in "Virtual Quantum Markov Chain of four-qubit systems" [2509.18803] shows that multipartite recoverability has a richer structural and geometric behavior: necessary kernel conditions cease to be sufficient, explicit entangled states separate recoverable from non-recoverable cases, and semidefinite programming provides computational tests and sampling-cost quantification.

## 1. Definition and operational scope

For a standard tripartite quantum Markov chain (QMC) in the order \(A \leftrightarrow B \leftrightarrow C\), a state \(\rho_{ABC}\) satisfies
\[
\rho_{ABC}=\mathcal{R}_{B\to BC}(\rho_{AB}),
\]
for some CPTP recovery channel \(\mathcal{R}_{B\to BC}\), and this is entropically equivalent to \(I(A:C|B)_\rho=0\) [2312.02031].

A tripartite VQMC relaxes the recovery requirement. Instead of demanding a physical quantum channel that reconstructs the full state, it requires a Hermitian, trace-preserving map
\[
\mathscr{R}_{B \to BC} = \sum_{i} \eta_i \mathcal{N}^{(i)}_{B \to BC}, \qquad \mathcal{N}^{(i)} \in (B, B \otimes C), \quad \eta_i \in \mathbb{R},
\]
such that
\[
\mathscr{R}_{B\rightarrow BC}(\rho_{AB}) = \rho_{ABC}.
\]
The map may be non-CP, but it remains operationally meaningful because it is statistically implementable through quasiprobability decomposition or measurement-controlled post-processing.

The operational content is explicit: a VQMC permits the retrieval of any global observable’s expectation value, \(\operatorname{Tr}[O\rho_{ABC}]\), by acting only on \(\rho_{AB}\). In this sense, the framework prioritizes observable statistics or “shadow information” over literal state preparation. This suggests that VQMCs are tailored to settings in which classical information extracted from measurements is the main resource.

## 2. Algebraic characterization in the tripartite setting

A central result for tripartite systems is an algebraic necessary-and-sufficient characterization in terms of block matrices [2312.02031]. For \(\rho_{AB}\), define
\[
Q_B^{(ij)} := \langle i|_A \rho_{AB} |j\rangle_A \in \mathbb{C}^{d_B\times d_B},
\]
and for \(\rho_{ABC}\),
\[
Q_{BC}^{(ij)} := \langle i|_A \rho_{ABC} |j\rangle_A \in \mathbb{C}^{d_B d_C\times d_B d_C}.
\]
The associated block operators are
\[
\mathrm{Rec}_B = [Q_B^{(ij)}]_{i,j}, \qquad \mathrm{Rec}_{BC} = [Q_{BC}^{(ij)}]_{i,j}.
\]

The theorem states that \(\rho_{ABC}\) is a VQMC in the order \(A\leftrightarrow B\leftrightarrow C\) if and only if
\[
\ker \mathrm{Rec}_B \subseteq \ker \mathrm{Rec}_{BC}.
\]
Equivalently, any linear dependence among the block matrices of \(\rho_{AB}\) must also be a dependence among the corresponding block matrices of \(\rho_{ABC}\). When the family \(\{Q_B^{(ij)}\}_{ij}\) is linearly independent, the inclusion is automatic and the state is a VQMC.

This criterion is significant because it replaces an a priori operational definition by a concrete linear-algebraic test. The paper also emphasizes that there is no intrinsic connection between vanishing quantum conditional mutual information and the VQMC property. That separation marks a conceptual divergence from standard QMC theory, where conditional mutual information plays the defining role.

## 3. Distinction from standard quantum Markov chains

The difference between QMCs and VQMCs is not merely terminological. In a QMC, the recovery map is a CPTP channel and the recovered object is the full quantum state. In a VQMC, the recovery map may be HPTP and possibly non-CP, while the operational target is the full set of observable statistics [2312.02031].

Several structural consequences follow. Every QMC is a VQMC, but the converse is not true. The VQMC condition therefore strictly enlarges the recoverability class beyond ordinary quantum Markovianity. The tripartite paper further frames this in terms of information flow: classical Markov chains describe statistical dependence among random variables, QMCs extend that notion to quantum channels and entropy, and VQMCs isolate statistical recoverability of observables, even when exact physical state recovery is unavailable.

A common misconception is that VQMCs are merely QMCs viewed through a different implementation model. The published results do not support that identification. Instead, the distinction is operational, algebraic, and geometric: the recovery map class is broader, the recoverable object is different, and the block-matrix criterion is disconnected from the usual entropic characterization.

## 4. Extension to four-qubit systems

The four-qubit extension studies states \(\rho_{ABCD}\) for which one can reconstruct the global state from the three-qubit marginal \(\rho_{ABC}\) by acting only on subsystem \(C\) [2509.18803]. In the formulation used there, a four-qubit state is a VQMC if there exists a CPTP map \(\mathcal{R}: C \to CD\) such that
\[
\rho_{ABCD} = (\mathrm{id}_{AB} \otimes \mathcal{R})(\rho_{ABC}).
\]

For a fixed orthonormal basis \(\{|j\rangle\}\) of subsystem \(C\), the paper defines the conditional block reductions
\[
\hat{\rho}_{AC|j} := \operatorname{Tr}_B \left[ (I_A \otimes \langle j|_C) \rho_{ABC} (I_A \otimes |j\rangle_C) \right],
\]
\[
\hat{\rho}_{BC|j} := \operatorname{Tr}_A \left[ (I_B \otimes \langle j|_C) \rho_{ABC} (I_B \otimes |j\rangle_C) \right].
\]
A necessary condition for the existence of a recovery map is then
\[
\operatorname{Ker}(\hat{\rho}_{AC|j}) \subseteq \operatorname{Ker}(\hat{\rho}_{BC|j}), \qquad \forall j.
\]

The key structural result is that this kernel-inclusion condition remains necessary but is no longer sufficient in the four-qubit setting. Thus, higher-partite recoverability is strictly subtler than the tripartite block-kernel criterion might suggest. The paper presents explicit states that satisfy the inclusion while still failing to admit a recovery map, showing that the state-space geometry becomes more intricate in the four-qubit regime.

## 5. Virtual recovery, semidefinite programs, and sampling overhead

The virtual-recovery formalism is expressed through quasiprobability decompositions. In the four-qubit paper, a virtual recovery map may be written as
\[
\mathcal{R} = c_1 \mathcal{N}_1 - c_2 \mathcal{N}_2, \qquad c_1, c_2 \geq 0,\quad \mathcal{N}_1, \mathcal{N}_2 \in \mathrm{CPTP},
\]
and the associated sampling overhead is
\[
\nu(\rho_{ABCD}) := \log\{ c_1 + c_2 \mid (\mathrm{id}_{AB}\otimes \mathcal{R})(\rho_{ABC}) = \rho_{ABCD} \}.
\]
Lower \(\nu\) indicates less virtuality, or less overhead, in the recovery [2509.18803].

The corresponding optimization is cast as a semidefinite program:
\[
\begin{aligned}
& \min_{c_1, c_2, J_1, J_2}\quad c_1+c_2 \\
& \text{s.t. }\ J_1 \succeq 0,\ J_2 \succeq 0, \\
& \qquad \operatorname{Tr}_{C'D}(J_1) = c_1 I_C,\ \operatorname{Tr}_{C'D}(J_2) = c_2 I_C, \\
& \qquad \operatorname{Tr}_{C'}\left[ (\rho_{ABC}^T \otimes I_{C'D})(I_{AB}\otimes (J_1-J_2)) \right] = \rho_{ABCD},
\end{aligned}
\]
where \(J_1\) and \(J_2\) are the Choi matrices for \(\mathcal{N}_1\) and \(\mathcal{N}_2\).

In the original tripartite framework, semidefinite programs are also derived to determine optimal sampling overhead and robustness, and the optimal sampling overhead is proved to be additive:
\[
\nu(\rho_{ABC} \otimes \sigma_{\hat{A}\hat{B}\hat{C}})=\nu(\rho_{ABC})+\nu(\sigma_{\hat{A}\hat{B}\hat{C}}).
\]
The paper interprets this as indicating no free lunch to further reduce the sampling cost of recovery from parallel calls of VQMC states [2312.02031].

The tripartite work additionally defines \(\varepsilon\)-approximate VQMCs by
\[
\min_{\mathscr{M} \in (B, B\otimes C)} \|\mathscr{M}_{B\to BC} \circ \operatorname{Tr}_C(\rho_{ABC}) - \rho_{ABC}\|_1 = \varepsilon,
\]
with the guarantee that the error in any observable’s expectation value is bounded by \(\varepsilon \|O\|_\infty\). This places virtual recoverability within a robustness framework rather than a purely exact one.

## 6. Canonical examples, geometry, and implications

The W and GHZ families supply the canonical separating examples in both the tripartite and four-qubit settings. For three qubits, the W state is a VQMC because the relevant block matrices are linearly independent, whereas the GHZ state is not, because there exists a nonzero coefficient vector \(\mathbf{c}\) with \(\mathrm{Rec}_B\mathbf{c}=0\) but \(\mathrm{Rec}_{BC}\mathbf{c}\neq 0\) [2312.02031]. For four qubits, the state
\[
|\mathrm{W}_4\rangle = \frac{1}{2}\left(|0001\rangle + |0010\rangle + |0100\rangle + |1000\rangle\right)
\]
admits a recovery channel and thus is a VQMC, with the map detailed via its Choi matrix (Eqn. 77), while the four-qubit GHZ state is not a VQMC because the kernel-inclusion condition fails [2509.18803].

The mixture
\[
\rho_p := p\, \rho_{\mathrm{W}_4} + (1-p)\, \rho_{\mathrm{GHZ}_4}
\]
is especially informative. For some values of \(p\), the kernel-inclusion condition holds but no recovery map exists. The four-qubit paper further states that a critical threshold \(p_c\) exists such that \(\rho_p\) is a VQMC if and only if \(p \geq p_c\), with \(p_c \geq 1/4\). This directly shows that the natural kernel test is incomplete in the multipartite setting.

The geometry of the VQMC set is non-convex. In the four-qubit analysis, if \(\rho_1\) is the W state and \(\rho_2\) is a classical-conditional VQMC state, then their mixture \(\rho_\lambda = \lambda \rho_1 + (1-\lambda)\rho_2\) is not a VQMC for \(\lambda \in (0,1)\); the mechanism is that the mixture can create new kernel directions in \(\hat{\rho}_{AC|j}\) that are absent in \(\hat{\rho}_{BC|j}\) [2509.18803]. The tripartite work likewise reports non-convexity through mixtures of W and GHZ states and through noise behavior: under depolarizing noise, the W state remains a VQMC for all nontrivial mixing with the maximally mixed state, whereas the GHZ state is not made into a VQMC by a tiny admixture of maximally mixed state, except in the fully mixed case [2312.02031].

| State or family | Criterion behavior | VQMC status |
|---|---|---|
| Three-qubit W state | Block matrices linearly independent | Yes |
| Three-qubit GHZ state | \(\ker \mathrm{Rec}_B \nsubseteq \ker \mathrm{Rec}_{BC}\) | No |
| Four-qubit W state | Kernel inclusion holds; recovery channel constructed | Yes |
| Four-qubit GHZ state | Kernel inclusion fails | No |
| Mixtures of VQMC states | Criterion may fail under mixing | Not always |

These results have several implications stated in the papers. VQMCs provide a tool to understand when and how multipartite quantum correlations can be reconstructed locally. The failure of sufficiency for kernel inclusion and the non-convexity of the VQMC set indicate a more intricate geometry and algebraic structure of state space than Markovian or separable sets. The four-qubit work also notes that recovery from smaller marginals is generally not possible; for example, two-qubit marginals are generally insufficient, and higher-order marginals are necessary. In application-oriented terms, the framework is positioned as relevant to shadow tomography, error mitigation, distributed quantum computing, entanglement detection, quantum simulation, and communication protocols, where measurement statistics rather than physical state synthesis are the primary objects of interest.

Source: https://www.emergentmind.com/topics/virtual-quantum-markov-chains-vqmcs