Markov Gap: Extreme Values & Quantum Recovery

• Markov gap is a field-dependent concept that quantifies differences in key observables—such as the gap between record values in random walks, discrepancies in entropic measures in quantum information, or spectral gaps in Markov chains.
• In the random walk context, it is defined as the spatial and temporal difference between the first two maxima, computed via renewal decomposition and closed-form double generating functions.
• In quantum and holographic settings, the gap measures the excess of reflected entropy over mutual information, serving as an indicator of recovery fidelity and multipartite entanglement, while in Markov chains it relates to convergence rates.

“Markov gap” is not a single invariant across the arXiv literature. In one strand of probability theory it denotes the pair of observables attached to the two highest records of a one-dimensional Markovian random walk, namely the spatial gap between the first two maxima and the time interval separating their occurrence (Majumdar et al., 2013). In quantum information and holography it denotes the nonnegative difference between reflected entropy and mutual information, $\Delta_{A:B}=S_R(A:B)-I(A:B)$, with direct interpretations in recoverability, entanglement-wedge geometry, and multipartite entanglement (Hayden et al., 2021, Iizuka et al., 21 Jul 2025). In the theory of Markov chains and Markov operators, by contrast, the nearby standard terminology is “spectral gap” or “singular-value gap,” which measures convergence and variance decay rather than either extreme-value crowding or reflected-entropic excess (Chatterjee, 2023, Xu, 1 Jun 2026).

1. Terminological scope

The expression appears in at least two technically distinct senses, while a third, adjacent literature studies gap notions for Markov dynamics under different names.

Context	Quantity	Definition
Random-walk extremes	$(G_n,L_n)$	$G_n=M_{1,n}-M_{2,n}$, $L_n=n_1-n_2$
Quantum information / holography	$\Delta_{A:B}$ or $M$	$S_R(A:B)-I(A:B)$
Markov-chain convergence	spectral or singular-value gap	e.g. $1-\lambda_2$ or $\sigma_2(I-P)$

In the random-walk setting, the object of interest is explicitly the pair $(G_n,L_n)$ attached to the first and second maxima of a one-dimensional Markovian random walker (Majumdar et al., 2013). In the quantum-information setting, the Markov gap is the difference between reflected entropy and mutual information, and is also called the “Markov distance” in the Lifshitz-theory discussion (Berthiere et al., 2023). In finite-state, nonreversible, and continuous-state Markov-process theory, the corresponding gap quantities are instead defined through eigenvalues or singular values of the transition operator or generator (Chatterjee, 2023, Xu, 1 Jun 2026). This suggests that the term is field-dependent rather than a unique standardized invariant.

2. Extreme-value meaning for one-dimensional random walks

For a one-dimensional Markovian random walk started at $(G_n,L_n)$0 and evolving by

$(G_n,L_n)$1

with i.i.d. symmetric jumps whose characteristic small-$(G_n,L_n)$2 behavior is

$(G_n,L_n)$3

the first and second maxima after $(G_n,L_n)$4 steps are

$(G_n,L_n)$5

and, if $(G_n,L_n)$6 are their occurrence times, the “Markov Gap” is defined by

$(G_n,L_n)$7

The central object is the joint probability density

$(G_n,L_n)$8

together with its large-$(G_n,L_n)$9 limit $G_n=M_{1,n}-M_{2,n}$0 (Majumdar et al., 2013).

A three-interval renewal decomposition, combined with survival and conditioned propagators, yields a closed-form double generating function for $G_n=M_{1,n}-M_{2,n}$1. In the stationary limit,

$G_n=M_{1,n}-M_{2,n}$2

and the ordinary generating function in $G_n=M_{1,n}-M_{2,n}$3 takes the form

$G_n=M_{1,n}-M_{2,n}$4

with

$G_n=M_{1,n}-M_{2,n}$5

The gap marginal is

$G_n=M_{1,n}-M_{2,n}$6

while the time-interval marginal has generating function

$G_n=M_{1,n}-M_{2,n}$7

(Majumdar et al., 2013).

The significance of this construction is that it furnishes an exact description of the “crowding” near the global maximum in a correlated system. In classical extreme-value theory for independent variables, the emphasis is usually on the global maximum alone; here the geometry of the top two maxima becomes tractable despite Markovian correlations.

3. Asymptotic laws and universality in the random-walk setting

The marginal distribution of the spatial gap has a universal power-law tail for $G_n=M_{1,n}-M_{2,n}$8: $G_n=M_{1,n}-M_{2,n}$9 with amplitude

$L_n=n_1-n_2$0

For $L_n=n_1-n_2$1, the tail remains non-universal and depends on the details of $L_n=n_1-n_2$2 (Majumdar et al., 2013).

The time-interval marginal exhibits three regimes: $L_n=n_1-n_2$3 The case $L_n=n_1-n_2$4 is characterized by a “freezing” of the tail exponent to $L_n=n_1-n_2$5. The mean interval $L_n=n_1-n_2$6 diverges for all $L_n=n_1-n_2$7, scaling as $L_n=n_1-n_2$8 for $L_n=n_1-n_2$9 and as $\Delta_{A:B}$0 for $\Delta_{A:B}$1 (Majumdar et al., 2013).

In the joint regime $\Delta_{A:B}$2, $\Delta_{A:B}$3, with $\Delta_{A:B}$4 fixed, the limiting law has the scaling form

$\Delta_{A:B}$5

where $\Delta_{A:B}$6 is integrable on $\Delta_{A:B}$7 and obeys

$\Delta_{A:B}$8

Monte Carlo simulations confirm the tail $\Delta_{A:B}$9, the change in the exponent of $M$0, the logarithmic correction at $M$1, and the collapse of the joint histogram in the scaling variables $M$2 and $M$3 (Majumdar et al., 2013).

4. Reflected entropy minus mutual information

In quantum information, one begins with a tripartite pure state $M$4, forms $M$5, and canonically purifies $M$6 to a pure state on $M$7. The reflected entropy is

$M$8

while the mutual information is

$M$9

The bipartite Markov gap is then

$S_R(A:B)-I(A:B)$0

(Iizuka et al., 21 Jul 2025).

This quantity admits an exact conditional-mutual-information representation: $S_R(A:B)-I(A:B)$1 Accordingly, $S_R(A:B)-I(A:B)$2 measures the obstruction to quantum Markovity. When $S_R(A:B)-I(A:B)$3, there exists a recovery map $S_R(A:B)-I(A:B)$4 — specifically the Petz or rotated-Petz map — that acting only on $S_R(A:B)-I(A:B)$5 recovers the full purification on $S_R(A:B)-I(A:B)$6 from $S_R(A:B)-I(A:B)$7 with perfect fidelity. More generally, one has the recoverability bound

$S_R(A:B)-I(A:B)$8

so the magnitude of the Markov gap controls the best achievable recovery fidelity from $S_R(A:B)-I(A:B)$9 alone (Iizuka et al., 21 Jul 2025).

An equivalent presentation emphasizes the same identity in the canonical purification and frames $1-\lambda_2$0 as the fidelity cost of a specific Markov recovery problem. In that formulation,

$1-\lambda_2$1

and the gap quantifies the failure of perfect recovery in a natural channel $1-\lambda_2$2 (Hayden et al., 2021).

5. Holographic, field-theoretic, and gravitational realizations

In AdS/CFT, reflected entropy is proposed to be dual to twice the area of the entanglement-wedge cross-section, while mutual information is expressed through RT areas. In the conventions summarized for the multipartite extension,

$1-\lambda_2$3

so that

$1-\lambda_2$4

In time-symmetric pure AdS$1-\lambda_2$5, the gap is universally lower bounded by a $1-\lambda_2$6 contribution per endpoint of the entanglement-wedge cross-section; in the island literature the parallel statement is that the lower bound is $1-\lambda_2$7 times the number of EWCS boundaries on minimal surfaces (Hayden et al., 2021, Lu et al., 2022).

With islands, explicit computations in the defect extremal surface model, JT gravity, and generic $1-\lambda_2$8 extremal black holes display the same counting rule. In vacuum states, a boundary formulation states that the lower bound is $1-\lambda_2$9 times the number of gaps between two boundary regions. For disjoint single intervals this gives $\sigma_2(I-P)$0, for adjacent intervals $\sigma_2(I-P)$1, and if the entanglement wedge is disconnected one has $\sigma_2(I-P)$2 (Lu et al., 2022).

In $\sigma_2(I-P)$3-dimensional Lifshitz theory, the reflected entropy for two disjoint intervals has the explicit form

$\sigma_2(I-P)$4

the mutual information is

$\sigma_2(I-P)$5

and therefore

$\sigma_2(I-P)$6

The gap is finite, cutoff-independent, and universal; for $\sigma_2(I-P)$7, $\sigma_2(I-P)$8, while for adjacent intervals the gap depends nontrivially on the ratio $\sigma_2(I-P)$9, unlike the constant adjacent-interval result in relativistic CFTs (Berthiere et al., 2023).

In non-inertial frames for free fermionic fields, the Markov gap

$(G_n,L_n)$0

shows monotonic behavior with respect to acceleration for Bell, Werner, and GHZ states. It is zero at $(G_n,L_n)$1 for Bell and GHZ, positive for the Werner state at $(G_n,L_n)$2, and strictly positive for all $(G_n,L_n)$3, signaling tripartite correlations involving the anti-Bob mode induced by the Unruh effect (Basak et al., 2023).

For Haar-random tripartite states, every bound-entangled marginal $(G_n,L_n)$4 must have strictly positive Markov gap. Conversely, states with weakly non-zero Markov gap almost surely have an undistillable marginal state — bound entangled or separable — for sufficiently large systems. The average gap exhibits distinct large-system regimes depending on $(G_n,L_n)$5, including a strongly nonzero regime, a weakly nonzero regime, and a regime where the gap is zero with probability $(G_n,L_n)$6 (Jin et al., 7 Apr 2025).

6. Multipartite extensions

A multipartite generalization is built from a reflected multi-entropy. Given a $(G_n,L_n)$7-partite pure state $(G_n,L_n)$8, one traces out $(G_n,L_n)$9, canonically purifies the remaining mixed state on doubled systems $(G_n,L_n)$00, and defines the von Neumann reflected multi-entropy

$(G_n,L_n)$01

as the $(G_n,L_n)$02 limit of the corresponding Rényi construction. The $(G_n,L_n)$03-partite Markov gap is then

$(G_n,L_n)$04

In holographic states, $(G_n,L_n)$05 by construction (Iizuka et al., 21 Jul 2025).

The proposed interpretation is geometric as well as operational. The gap vanishes if and only if the canonical-purification cross-section can be reassembled from the ordinary RT surfaces, so that no “multi-way” geometric obstruction remains. Under holographic assumptions one also has a hierarchy

$(G_n,L_n)$06

meaning that the $(G_n,L_n)$07-partite gap controls all coarser recoveries. Operationally, a vanishing $(G_n,L_n)$08 implies that the full $(G_n,L_n)$09-purification can be recovered by successive local rotated-Petz maps on one region at a time (Iizuka et al., 21 Jul 2025).

The same work introduces the genuine reflected multi-entropy because $(G_n,L_n)$10 can fail to vanish on states lacking genuine $(G_n,L_n)$11-way entanglement. The prescriptions require the genuine reflected quantity to vanish on any pure state whose entanglement involves fewer than $(G_n,L_n)$12 parties, to be a linear combination of reflected multi-entropies and ordinary multi-entropies of lower arity, and, at $(G_n,L_n)$13, to reproduce twice the Rényi-2 genuine multi-entropy. For $(G_n,L_n)$14, an explicit formula for $(G_n,L_n)$15 is given and shown to vanish on any state factorizing a single party and on any bipartite-entangled pure state (Iizuka et al., 21 Jul 2025).

7. Relation to spectral-gap terminology in Markov-process theory

In the broader theory of Markov chains and Markov operators, “gap” usually refers to convergence-rate quantities rather than to either of the two Markov-gap notions above. For a reversible finite-state chain with eigenvalues

$(G_n,L_n)$16

the spectral gap is

$(G_n,L_n)$17

and the mixing time satisfies

$(G_n,L_n)$18

In optimal prediction, a fixed positive spectral gap eliminates the logarithmic penalty that appears when the chain may have arbitrarily small gap, leading to the parametric-rate bound $(G_n,L_n)$19 for large alphabets (Han et al., 2021).

For nonreversible finite-state chains, the generator is $(G_n,L_n)$20, the spectral gap is defined as the second-smallest singular value

$(G_n,L_n)$21

and the relaxation time is $(G_n,L_n)$22. This relaxation time characterizes the convergence of empirical averages and can be substantially smaller than the total-variation mixing time. Cheeger-type bounds and path-argument lower bounds generalize to this setting (Chatterjee, 2023).

For general Markov semigroups on $(G_n,L_n)$23, the corresponding quantity is the singular-value gap

$(G_n,L_n)$24

which yields uniform finite-time variance bounds, bounded invertibility of $(G_n,L_n)$25, solvability of the Poisson equation $(G_n,L_n)$26, and asymptotic-variance formulas. In discrete time one takes $(G_n,L_n)$27 (Xu, 1 Jun 2026).

Related operator-theoretic work studies $(G_n,L_n)$28-spectral gaps of Markov maps and criteria for spectral gaps of Markov operators. For Markov maps on noncommutative probability spaces, an $(G_n,L_n)$29-spectral gap implies an $(G_n,L_n)$30-spectral gap for $(G_n,L_n)$31, and the converse holds under factorizability (Conde-Alonso et al., 2017). For general Markov operators, the symmetrization $(G_n,L_n)$32 has a spectral gap if and only if the $(G_n,L_n)$33-norm satisfies $(G_n,L_n)$34 (Wang, 2013). Decomposition formulas of the form

$(G_n,L_n)$35

unify finite-cover decompositions, hybrid Gibbs samplers, localization schemes, and several new hybrid constructions (Qin, 1 Apr 2025).

These usages are conceptually adjacent but formally distinct. They concern the contraction, relaxation, or variance decay of Markov dynamics, whereas the random-walk Markov Gap measures the local geometry of extreme events and the quantum-information Markov gap measures the excess $(G_n,L_n)$36, equivalently a recoverability obstruction.