
A note on the antisymmetry in the speed of a random walk in reversible
dynamic random environment
Oriane Blondel
January 13, 2023
Abstract
In this short note, we prove that v(−ε) = −v(ε). Here, v(ε)is the speed of a one-dimensional random walk
in a dynamic reversible random environment, that jumps to the right (resp. to the left) with probability 1/2 + ε
(resp. 1/2−ε) if it stands on an occupied site, and vice-versa on an empty site. We work in any setting where
v(ε), v(−ε)are well-defined, i.e. a weak LLN holds. The proof relies on a simple coupling argument that holds only
in the discrete setting.
1 Introduction
We consider the so-called “ε–random walk”: a random walk in one-dimensional dynamic random environment with
two values that jumps to the right (resp. to the left) with probability 1/2 +ε(resp. 1/2−ε) if it stands on an occupied
site, and vice-versa on an empty site (Figure 1).
Figure 1: Jump rates for the ε–random walk.
A lot of energy has been devoted to describing the behavior of this ε–RW in various random environments,
mainly to find whether it satisfies the usual limit theorems (LLN, CLT...). This is generally a hard problem, since
the environment seen from the ε–RW is highly non-stationary (even when the environment is). Additionally, it
belongs to the class of nestling random walks, which do not have an a-priori preferred direction. It is possible to
find settings in which the LLN does not hold. [BHT20, Section 9] proposes an example in which space-time traps
lead the random walk into longer and longer stretches of motion with drifts to the right or the left. Cases in which
the LLN has been shown include perturbative regimes (small ε) [ABF18], well-known environments like the exclusion
process [HS15, HKT20] or the contact process [MV15], uniform mixing hypotheses [CZ04, AdHR11, RV13], or fast
enough decay of correlations [BHT20].
When the random walk does satisfy a law of large numbers, let us call v(ε)its asymptotic speed (we will simply
say that v(ε)is well-defined). We observed in [ABF16] that, if the environment is given by a reversible Markov process
with positive spectral gap, and |ε|is smaller than the spectral gap, v–in addition to being well-defined– satisfies the
antisymmetry property
v(−ε) = −v(ε).(1)
This property also holds in higher dimensions for random walks with a certain symmetry property (Assumption 3
in [ABF16]). Simulations moreover suggest that this property extends out of the perturbative regime of [ABF16]
(where the law of large numbers was not yet even established; see Figure 3 in [ABF16]). The proof relies on an
expansion of the speed in ε, in which the terms of even degrees are shown to cancel due to reversibility.
It is worth pointing out that this antisymmetry does not seem to follow from obvious symmetry properties of the
system. Indeed, it is tempting at first sight to think that the (−ε)–random walk reversed in time should have the same
distribution as the ε–random walk, from which (1) would immediately follow. But closer inspection reveals that this
is not true and those two processes have very different trajectories in general (see for instance Figure 4 in [ABF16]).
Rather, (1) holds iff the speed of the ε–random walk is the same in the reversible environment and in its image under
the mirror symmetry x7→ −x. While this property is obvious for reversible systems invariant under mirror symmetry
(e.g. simple symmetric exclusion process), there exist reversible, translation invariant processes which have no mirror-
invariance. One that has been extensively studied is the East model [JE91,FMRT13] , in which particles appear (resp.
disappear) at site xat rate p(resp. 1−p), but only if x+ 1 is empty. This dynamics is reversible w.r.t. the product
1
arXiv:2210.10427v2 [math.PR] 12 Jan 2023