Strong Markov Property

• Strong Markov property is defined as the memoryless restart at stopping times, extending the standard Markov property.
• It plays a crucial role in diffusion theory, nonlinear expectations, and quantum dynamics by enabling conditional independence and restartability.
• The property underlies key proofs in hitting-time decompositions and the construction of singular SDE solutions in complex systems.

Searching arXiv for recent and foundational papers relevant to the strong Markov property and adjacent meanings of “Markov property.” The strong Markov property is the extension of the Markov property from deterministic times to stopping times. In its classical probabilistic form, it asserts that once a stopping time $\tau$ is reached, the post-$\tau$ evolution depends on the past only through the random present state $X_\tau$; a standard formulation is

$$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$

on $\{\tau<\infty\}$ for bounded measurable $f$ and $t\ge 0$ (Bruggeman et al., 2015). This restartability at random times is the defining content of the notion in diffusion theory and stochastic processes, but the term “Markov property” is also used in several nonclassical senses—quantum states, set-indexed fields, topological dynamics, and empirical scale cascades—where stopping times may play no role or are replaced by other structures (Ooi, 2022).

1. Classical probabilistic formulation

The ordinary Markov property conditions at deterministic times, whereas the strong Markov property replaces deterministic $t$ by a stopping time $\tau$. In the one-dimensional diffusion setting, the distinction is explicit: ordinary Markovianity takes the form

$E_x[f(X_{t+s})\mid \mathcal F_t]=E_{X_t}[f(X_s)],$

while strong Markovianity requires the same memoryless restart at $\tau$0 (Bruggeman et al., 2015). For diffusion models with stopping-time arguments, this is the property that allows one to decompose a trajectory into “first hit an intermediate point, then continue from there,” a mechanism that is indispensable in hitting-time proofs (Bruggeman et al., 2015).

The same structural extension appears in nonlinear expectation theory. For dynamic risk mappings on canonical path space, the paper on Markov risk mappings defines the Markov property by

$\tau$1

and proves the strong version

$\tau$2

for finite stopping times $\tau$3 (Kosmala et al., 2020). Here the restart principle is preserved, but conditional expectation is replaced by a nonlinear conditional risk operator.

2. Construction in diffusion theory

A precise existence theorem is available for degenerate Kimura diffusions with singular drift on

$\tau$4

Under Assumption 3.1, for every initial state $\tau$5 there exists a weak solution to the singular SDE

$\tau$6

and this constructed solution satisfies the strong Markov property (Pop, 2014). The stopping-time statement is written as

$\tau$7

for every stopping time $\tau$8, $\tau$9, and Borel set $X_\tau$0 (Pop, 2014).

The proof strategy is not a direct Stroock–Varadhan martingale-problem uniqueness argument for the singular equation. Instead, the standard Kimura diffusion is first constructed and shown to be strong Markov through uniqueness in law; then the singular drift is added by Girsanov after verifying Khas’minskii and Novikov estimates such as

$X_\tau$1

and the strong Markov property is transferred by an explicit stopping-time change-of-measure argument (Pop, 2014). The same paper is careful to distinguish this from full weak uniqueness: uniqueness in law is proved only within the class of weak solutions that satisfy the Markov property (Pop, 2014).

3. One-dimensional consequences and approximation

For one-dimensional continuous regular strong Markov processes, the strong Markov property has sharp pathwise consequences. If $X_\tau$2 is a one-dimensional, continuous, regular, and strong Markov process with state space $X_\tau$3, then for every $X_\tau$4 and every $X_\tau$5,

$X_\tau$6

so any interior point can be hit arbitrarily quickly with positive probability (Bruggeman et al., 2015). The proof uses strong Markovian restarting at the hitting time of an intermediate point to derive the triangle inequality

$X_\tau$7

for the minimal positive-probability hitting-time scale $X_\tau$8, and then combines this with local-martingale arguments to force $X_\tau$9 (Bruggeman et al., 2015).

The same class of processes admits simple discrete approximations. A functional limit theorem shows that every one-dimensional regular continuous strong Markov process in natural scale can be approximated by “coin-tossing” Markov chains with state-dependent step size,

$$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$0

provided the scale factors satisfy Condition (A),

$$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$1

on compact subsets $$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$2 (Ankirchner et al., 2019). The target class includes sticky Brownian motion and Brownian motion slowed down on the Cantor set, and the embedding of the chains into the limiting diffusion is explicitly based on stopping times of the limiting strong Markov process (Ankirchner et al., 2019).

4. Infinite-dimensional and generalized-index settings

In infinite-particle and infinite-dimensional systems, proving strong Markovianity is typically harder than proving ordinary Markovianity. For determinantal dynamics with the extended sine, extended Airy, and extended Bessel kernels, the paper on determinantal processes proves that the kernel-defined limits are reversible diffusion processes, and therefore strong Markov, and also identifies the associated quasi-regular Dirichlet forms and infinite-dimensional SDEs (Osada et al., 2014). The state space is the configuration space

$$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$3

and the key step is to strengthen vague continuity to continuity in carefully chosen spaces $$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$4, $$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$5, and $$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$6, then combine this with continuity in the initial condition and the previously known Markov property (Osada et al., 2014).

A different extension appears for Gaussian fields associated with local Dirichlet forms. There the ordinary Markov property is a spatial conditional-independence statement,

$$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$7

and it is proved equivalent to locality of the underlying irreducible regular Dirichlet form (Ooi, 2022). The paper then introduces a strong Markov property for random closed sets: if $$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$8 are random sets with $$E_x[f(X_{\tau+t})\mid \mathcal F_\tau]=E_{X_\tau}[f(X_t)],$$9, the strong property is

$\{\tau<\infty\}$0

under a sufficient measurability condition on the random sets (Ooi, 2022). This is modeled on classical strong Markov ideas, but the stopping objects are random closed sets rather than stopping times.

Set-indexed processes admit an analogous development. For a $\{\tau<\infty\}$1-Markov process, the paper on set-indexed Markov properties proves a strong Markov theorem at bounded simple stopping sets $\{\tau<\infty\}$2: $\{\tau<\infty\}$3 provided the process is homogeneous, $\{\tau<\infty\}$4-Feller, and has outer-continuous sample paths (Balança, 2012). Here stopping times are replaced by random sets, and the role of the semigroup is played by a $\{\tau<\infty\}$5-transition system satisfying a Chapman–Kolmogorov-type identity.

By contrast, not every infinite-dimensional Markov analysis reaches a strong Markov theorem. For the 3D stochastic primitive equations, an abstract selection theorem yields an almost sure Markov family satisfying

$\{\tau<\infty\}$6

for all deterministic times outside a Lebesgue-null exceptional set, but the paper does not formulate a full stopping-time strong Markov property (Dong et al., 2016).

5. Quantum, entropic, and topological analogues

Outside classical stochastic-process theory, the phrase “Markov property” often refers to structural factorization or recoverability rather than restart at stopping times. In relativistic quantum field theory, the vacuum state of a conformal field theory for regions with boundary on the lightcone is a quantum Markov state in the sense that it saturates strong subadditivity: $\{\tau<\infty\}$7 equivalently,

$\{\tau<\infty\}$8

This null-boundary Markov property is the ingredient used to derive the entropic $\{\tau<\infty\}$9-theorem in $f$0; it is not a stopping-time statement (Casini et al., 2017).

A distinct “strong quantum Markov property” appears in quantum many-body dynamics. There the strong property is an outcome-by-outcome recovery requirement: for every operator $f$1 supported on $f$2 with $f$3,

$f$4

The paper characterizes this strengthened local recoverability by combining approximate detailed balance on $f$5 with clustering between $f$6 and $f$7, and derives consequences such as repeatable measurement–recovery protocols from a single copy (Chen, 4 May 2026). Again, the terminology is explicitly not the classical probabilistic strong Markov property.

Topological dynamics supplies another nonprobabilistic usage. For an expansive action $f$8, the strong topological Markov property means that if two configurations agree approximately on a finite collar $f$9, then one can glue the inside of one and the outside of the other into a global point. In subshifts, this is equivalent to Gromov’s splicability (Ceccherini-Silberstein et al., 2021). The resulting property is a finite-boundary gluing rule, not restartability at random times.

6. Adjacent notions and recurring misconceptions

Several papers illustrate why “Markov property” should not be conflated with the strong Markov property.

The turbulence paper studies one-component Lagrangian velocity increments

$t\ge 0$0

as a stochastic process in the lag variable $t\ge 0$1, and tests the finite-step relation

$t\ge 0$2

It finds empirical Markovianity only for lag spacings above an Einstein–Markov coherence time and is explicit that no stopping-time or strong Markov theorem is involved (Fuchs et al., 2021).

The driving-behavior study tests whether sampled trajectories satisfy the ordinary discrete-time condition

$t\ge 0$3

or a higher-order finite-memory analogue, using conditional-characteristic-function tests. It does not introduce filtrations or stopping times, so its conclusions concern only ordinary finite-order Markov structure, not strong Markovianity (Li et al., 18 Jan 2025).

The SPDE note on the stochastic heat equation proves that the associated semigroup is strong Feller at every time $t\ge 0$4 under the Fourier lower bound

$t\ge 0$5

but it explicitly does not prove a strong Markov theorem; its subject is semigroup regularization, not restartability at stopping times (Liu et al., 26 May 2026). Likewise, the stable-limit paper for additive functionals of Markov chains relies on ordinary deterministic-time conditioning through the transition operator $t\ge 0$6, not on stopping-time arguments (Machkouri et al., 2018).

The classical strong Markov property is therefore best reserved for the restart principle at stopping times. Ordinary one-step conditional independence, strong Feller regularization, quantum recovery identities, null-boundary entropy equalities, and empirical finite-step closures are all important Markovian notions, but they are different notions. The modern literature preserves the phrase because each setting isolates a comparable “memoryless” structure; the substantive distinction is the object with respect to which memorylessness is asserted—stopping times, random sets, boundary algebras, local operations, or scale cascades.