---
title: Multipartite Markov Gap in Quantum & Holographic Systems
url: https://www.emergentmind.com/topics/multipartite-markov-gap
type: topic
---

# Multipartite Markov Gap in Quantum & Holographic Systems

The multipartite Markov gap is a family of entropic diagnostics that quantify the failure of a multipartite quantum state to behave as a quantum Markov chain. In its basic form, it is defined as the difference between reflected entropy and mutual information; for pure tripartite states this difference reduces to a conditional mutual information, making the quantity a computable witness of irreducible tripartite structure. More recent work has extended the construction to higher-party settings through reflected multi-entropy and related multipartite entropic combinations, especially in holography, where the gap acquires a geometric interpretation in terms of entanglement-wedge cuts and recoverability obstructions. The same phrase also appears in a distinct, non-entanglement context in spectral compression of lumpable Markov chains, so the term is currently field-dependent rather than uniquely standardized [2408.12533][2507.15262][2604.10820].

## 1. Core definition and tripartite identity

For a bipartite mixed state $\rho_{AB}$, the basic Markov gap is
$$
h(A:B)=S_R(A:B)-I(A:B),
$$
where $I(A:B)=S(A)+S(B)-S(AB)$ is the mutual information and $S_R(A:B)$ is the reflected entropy obtained from the canonical purification. By strong subadditivity, $h(A:B)\ge 0$, and $h(A:B)=0$ if and only if $\rho_{AB}$ is a quantum Markov chain; equivalently, $\rho_{AB}$ saturates strong subadditivity. The same construction admits the canonical-purification identity
$$
h(A:B)=I(A':B'|AB),
$$
so the gap is literally a conditional mutual information in the purified four-party state, and it lower bounds the relative-entropy distance to the set of quantum Markov states [2506.02131].

In the tripartite setting used in nonequilibrium many-body physics, one considers three contiguous regions $A$, $B$, and $C$ and defines
$$
M(A,B,C)=S_R(A:B)-I(A:B).
$$
When $\rho_{ABC}$ is pure, this quantity coincides with
$$
I(A:C|B)=S_{AB}+S_{BC}-S_B,
$$
since $S_{ABC}=0$. Positivity of $M(A,B,C)$ signals that the tripartite correlations cannot be reduced to a classical-Markov chain or to purely bipartite entanglement. In the formulation given for the XX chain, there is no decomposition
$$
\rho_{ABC}=\sum_j p_j\,\rho_{AB}^{(j)}\otimes \rho_{BC}^{(j)}
$$
that is consistent with the marginals unless $M=0$. The same source also emphasizes a limitation: this witness does not detect GHZ-type three-party entanglement [2408.12533].

| Quantity | Definition | Special reduction |
|---|---|---|
| Bipartite Markov gap | $h(A:B)=S_R(A:B)-I(A:B)$ | $h=0$ iff quantum Markov chain |
| Tripartite Markov gap | $M(A,B,C)=S_R(A:B)-I(A:B)$ | For pure $ABC$, $M=I(A:C|B)$ |
| Multipartite Markov gap | $MG^{M(q-1)}=S_{R\,(q-1)}-\bigl[\sum_i S(A_i)-S(A_1\cdots A_{q-1})\bigr]$ | For $q-1=2$, reduces to $h(A:B)$ |

## 2. Multipartite generalizations

A higher-party construction was proposed by replacing reflected entropy with a reflected multi-entropy. For a pure state on $A_1\cdots A_q$, one first traces out one subsystem, canonically purifies the resulting $(q-1)$-party mixed state, and then computes the reflected multi-entropy $S_{R\,(q-1)}(A_1:\cdots:A_{q-1})$. The multipartite mutual-information analogue is
$$
I_{\rm multi}(A_1:\cdots:A_{q-1})=\sum_{i=1}^{q-1}S(A_i)-S(A_1\cdots A_{q-1}),
$$
which leads to
$$
MG^{M(q-1)}(A_1:\cdots:A_{q-1})
=
S_{R\,(q-1)}(A_1:\cdots:A_{q-1})
-\Bigl[\sum_i S(A_i)-S(A_1\cdots A_{q-1})\Bigr].
$$
In holographic states, this quantity is nonnegative. It vanishes when the entanglement wedge splits so that the multiway cut is trivial, or equivalently when the reflected multi-entropy saturates its lower bound. A coarse-graining inequality was also established,
$$
MG^{M(q-1)}(A_1:\cdots:A_{q-1})
\ge
MG^{M(q-2)}(A_1A_2:A_3:\cdots:A_{q-1}),
$$
so merging regions cannot increase reconstructability [2507.15262].

Parallel multipartite extensions were introduced through three symmetric quantities:
$$
\Delta^{(1)}(A:B:C),
\qquad
\Delta^{(o)}(A:B:C),
\qquad
\Delta^{(J)}(A:B:C).
$$
The first sums reflected-entropy-minus-mutual-information terms over the three one-vs-rest bipartitions, the second replaces reflected entropy with odd entanglement entropy, and the third is a purely von Neumann-entropy cyclic combination. All three are nonnegative, permutation-symmetric, and reduce to the bipartite gap in the appropriate limit. In the same framework, a genuine reflected multi-entropy $GS_R$ was proposed to vanish on states lacking full $q$-party entanglement; for $q=4$ and in the $(m=2,n=2)$ Rényi case it matches twice the genuine Rényi-2 multi-entropy. This refinement was introduced precisely because the multipartite Markov gap can remain nonzero on states that do not possess full $q$-party structure [2506.02131][2507.15262].

## 3. Holographic meaning and recoverability

In holography, the Markov gap is computed from minimal bulk surfaces. For two boundary regions $A$ and $B$, reflected entropy is related to the entanglement wedge cross section, while mutual information is constructed from Ryu–Takayanagi surfaces. In the multipartite proposal, the ordinary reflected entropy is replaced by reflected multi-entropy, and its bulk dual is a minimal entanglement-wedge multiway cut anchored on the RT surface of the union $A_1\cdots A_{q-1}$. The resulting gap is the leftover area after subtracting the single-region RT contribution, and it measures a geometric obstruction to collapsing the multiway cut to the union of individual RT surfaces [2507.15262].

This geometric formulation is also an information-theoretic statement about recoverability. In the bipartite or purified-tripartite limit one may write
$$
MG(A:B)=I(A:B^*|B),
$$
with $B^*$ the auxiliary purifying copy. Vanishing gap implies exact recovery of the state from $\rho_{AB}$ by a recovery channel acting on $B$ alone, via the Petz or rotated-Petz theorems. A nonzero gap lower-bounds the fidelity deficit in approximate recovery through
$$
I(A:B^*|B)\ge -\log\!\bigl[F\bigl(\rho_{ABB^*},\,\mathcal R_{B\to BB^*}(\rho_{AB})\bigr)\bigr].
$$
Numerical studies on four-qubit toy states in the $(m=2,n=2)$ Rényi approximation reported a threshold gap $\approx 0.35$ below which the recovery fidelity rises to unity. In the holographic nodal-line setup, a symmetrized tripartite version
$$
h(A:B:C)=\min\{h(A:B),h(B:C),h(C:A)\}
$$
was proposed, although for the strip/complement configuration used there it reduces to $h(A:B)$ [2507.15262][2602.01545].

## 4. Static and dynamical realizations in quantum many-body systems

A static realization appears in two-dimensional chiral topological liquids. Using the vertex-state and conformal-interface construction, the Markov gap
$$
h(A,B)=S_R(A,B)-I(A,B)
$$
was found to be universal in the three-vertex state:
$$
h(A,B)=\frac{c}{3}\ln 2.
$$
The same $\ln 2$ persists for arbitrary $p$-vertex states in rational conformal field theories and for adjacent pairs of regions. Area-law terms, corner contributions, and the topological pieces appear in both $S_R$ and $I(A,B)$ and cancel in the difference, leaving only the universal $\frac{c}{3}\ln 2$ term. For the free real-fermion CFT with $c=1/2$, Gaussian correlation-matrix numerics for $p=4$ gave $h\simeq 0.1155\ldots$, matching $(1/2\div 3)\ln 2\approx 0.1155$ [2301.07130].

A dynamical realization was developed for global quenches in the XX spin chain from the “triplet-generator” initial state
$$
|\psi_0\rangle=\prod_i c_{3i}^\dagger |0\rangle.
$$
In the free-fermion formulation, one computes $S(X)$ from correlation matrices and obtains
$$
M(t)=S_R(A:B)(t)-[S_A(t)+S_B(t)-S_{AB}(t)].
$$
In the hydrodynamic limit, the quasiparticle picture predicts an initial latency
$$
t_d=\min(\ell_A,\ell_B)/\sqrt 3,
$$
with $M(t)=0$ for $t<t_d$, followed by a sharp rise once triplets can be simultaneously shared by $A$, $B$, and $C$. For unequal subsystems there is a second rise at
$$
t_b=\max(\ell_A,\ell_B)/\sqrt 3.
$$
The gap then reaches a maximum—the “entanglement barrier”—and decays algebraically, roughly as $\sim 1/t$. For equal subsystems, finite-size curves of $M/\ell$ collapse onto a universal hydrodynamic curve that is zero for $\tau<1/\sqrt 3$, rises steeply near $\tau\approx 1/\sqrt 3$, peaks at $\tau\approx 0.6\ldots$, and decays as $\sim 1/(\tau\ell)$. Free-fermion numerics for subsystem sizes from $\ell\sim 90$ to $\ell\sim 2070$ reproduced the predicted delay, plateau, barrier, and tail [2408.12533].

## 5. Topological semimetals and anisotropic infrared scaling

In strongly coupled holographic nodal-line semimetals, the Markov gap was used as a nonlocal probe of topological phase transitions. For two equal-width strips of width $l$ separated by the same distance $l$, the large-$l$ behavior is
$$
\Delta_M(l)\equiv h(l)\sim l^{-\alpha},
$$
with anisotropic exponents determined by the IR Lifshitz scaling exponent $\mathbf z$:
$$
h_x(l_x)\sim l_x^{-(1+\mathbf z)},
\qquad
h_z(l_z)\sim l_z^{-2/\mathbf z}.
$$
Across the transition controlled by $M/b$, the model exhibits three IR geometries with $\mathbf z\approx 10.929$ in the topologically nontrivial phase, $\mathbf z\approx 6.3694$ at the critical point $M/b=0.8597$, and $\mathbf z=1$ in the trivial phase. Accordingly, $\alpha_x$ changes from about $11.929$ to $2$, while $\alpha_z$ changes from about $0.183$ to $2$. Since $h(l)\to 0$ in all phases as $l\to\infty$, the phases remain short-range entangled; the order parameter is instead the decay exponent, or equivalently the large-$l$ value of $l^\alpha h(l)$, which develops a sharp kink at the critical point [2602.01545].

In the zero-temperature holographic Weyl semimetal, the tripartite Markov gap was likewise computed for strip regions. In the nontrivial and trivial phases, both $h_x(l)$ and $h_z(l)$ decay as $l^{-2}$ at large $l$. At the critical Lifshitz solution, the exponents become direction-dependent:
$$
h_x(l)\sim l^{-(1+\beta)},
\qquad
h_z(l)\sim l^{-2/\beta},
$$
with $\beta\simeq 0.407$, giving numerical exponents $1+\beta\approx 1.407$ and $2/\beta\approx 4.91$. At fixed large $l$, $h_x$ develops a peak near the critical value $(M/b)_c\simeq 0.744$, whereas $h_z$ shows a dip. The angular dependence $h(l,\theta)\sim a(\theta)l^{-2}$ further resolves anisotropy: the residual coefficient for the Markov gap is $\alpha_N^{(0)}\approx 0.168$ in the nontrivial phase and $\alpha_T^{(0)}\approx 1.7\times 10^{-4}$ in the trivial phase, while the sharpness exponent is of order $\sim 3.7$ versus $\sim 20$. In this setting, $h>0$ was interpreted as a signal of genuine tripartite entanglement not reducible to a convex mixture of bipartite “triangle states” [2606.05757].

## 6. Limits of detection and terminological divergence

The multipartite Markov gap is not a universal detector of all forms of multipartite entanglement. In the XX-chain formulation, positive $M(A,B,C)$ witnesses irreducible tripartite entanglement but does not detect GHZ-type three-party entanglement [2408.12533]. In the higher-party reflected-multi-entropy framework, the analogous limitation is structural: $MG^{M(q-1)}$ can already be nonzero on a $(q-1)$-party entangled state, which is why the genuine reflected multi-entropy was introduced. The four-qubit examples in that work sharpen the distinction: GHZ states give vanishing multipartite gaps and vanishing genuine gaps, whereas W states give strictly positive values for both [2507.15262].

A separate terminological divergence occurs in finite-state reversible Markov chains. In a six-state lumpable model with $T=P^2$, the phrase “multipartite Markov gap” was used for the strict inequality between a relaxed determinant benchmark
$$
\mathfrak D^{\mathrm{rel}_3}(T)=\sup_{U^*U=I_3}\det(U^*TU)
$$
and a partition-constrained benchmark
$$
\sup_{\mathcal A}\det Q_{\mathcal A}(T),
\qquad
Q_{\mathcal A}(T)=H_{\mathcal A}^*TH_{\mathcal A}.
$$
For the explicit model analyzed there,
$$
\mathfrak D^{\mathrm{rel}_3}(T)\approx 0.0883986324,
\qquad
\sup_{\mathcal A}\det Q_{\mathcal A}(T)\approx 0.0702908835,
$$
so
$$
\sup_{\mathcal A}\det Q_{\mathcal A}(T)<\mathfrak D^{\mathrm{rel}_3}(T).
$$
This usage concerns spectral compression by partition-indicator frames rather than entanglement, and its “gap” refers to a strict mismatch between relaxed orthonormal frames and genuine partitions. The coexistence of this stochastic-process meaning with the entanglement-theoretic meaning indicates that “multipartite Markov gap” currently names a family of field-specific constructions rather than a single universally fixed object [2604.10820].

Source: https://www.emergentmind.com/topics/multipartite-markov-gap