When escape through a random environment forces a definite direction and speed

← All digests

Family 220 · Probability and statistical mechanics
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-11

1. The question

Imagine a walker moving through a landscape of local instructions. At each site, a randomly chosen probability vector tells it how likely it is to move to each neighboring site. The instructions remain in place: returning to a site means encountering the same biases again.

Two basic questions arise. Can the walker have a probability strictly between zero and one of eventually escaping in a prescribed direction? And if it does escape in that direction almost surely, must it make linear progress, or can increasingly long delays reduce its speed to zero?

This family contains three manuscripts claiming answers under carefully distinguished hypotheses. Two establish directional zero–one laws, including an extension to environments with dependence over a finite range. The third claims that almost-sure directional escape implies positive speed in independent uniformly elliptic environments in dimension at least two.

The distinction between escape and speed is central. A sequence can tend to infinity as slowly as $\log n$; reaching arbitrarily distant levels says little by itself about how much time the journey takes.

2. Background and essential definitions

The walker lives on $\mathbb Z^d$ and takes nearest-neighbor steps from

$$U=\{\pm e_1,\ldots,\pm e_d\},$$

where $e_i$ is the $i$-th coordinate vector. An environment assigns a row of transition probabilities $\omega(x,\cdot)$ to each site:

$$\omega(x,e)\geq0, \qquad \sum_{e\in U}\omega(x,e)=1.$$

For a fixed environment, the walk is a Markov chain with transition probability $\omega(x,e)$ from $x$ to $x+e$. Its law is called quenched. Averaging that law over the environment gives the annealed law $P_0$, when the walk starts at zero.

An iid environment has independent identically distributed rows at different sites. The entries within a row are not required to be independent; indeed they must sum to one.

There are two positivity assumptions:

• Strict ellipticity: every allowed transition probability is positive almost surely.
• Uniform ellipticity: one deterministic constant $\kappa>0$ bounds every transition probability below, at every site.

Strict ellipticity permits exceptionally small transition probabilities. Uniform ellipticity prohibits them, though larger regions can still delay a walker.

For a fixed nonzero real vector $\ell$, define the directional escape event

$$A_\ell=\{X_n\cdot\ell\longrightarrow+\infty\}.$$

A directional zero–one law says $P_0(A_\ell)\in\{0,1\}$. Ballisticity is stronger: the displacement divided by elapsed time approaches a vector $v$ with $v\cdot\ell>0$.

The dependent-environment manuscript uses finite-range dependence. There is a fixed range $R$ such that all rows in two deterministic sets are independent whenever the sets are more than $R$ apart in maximum-coordinate distance. This concerns the full collections of rows, including infinite sets. Merely having pairwise correlations that decay with distance is weaker.

3. Development history and previous partial results

Solomon’s one-dimensional theory showed that transience and positive speed can separate even in uniformly elliptic iid environments. A walker may escape to the right while spending enough time in unfavorable stretches to have zero limiting speed. Higher dimensions offer routes around obstacles, but making that geometric intuition quantitative has been difficult. Solomon’s paper

Kalikow’s work initiated much of the directional-transience theory. The classical zero–one statement concerns the union $A_\ell\cup A_{-\ell}$: escape in one of two opposite directions is either almost sure or has probability zero. That leaves open the possibility that the two signs each have positive probability.

Zerner and Merkl settled the one-sign question for planar iid elliptic environments. Their argument exploits the fact that suitably opposed paths must intersect in the plane. Zerner later simplified the proof. Higher-dimensional paths can pass around one another, so the same geometric mechanism does not extend directly. Zerner–Merkl, Zerner’s revised proof

Sznitman and Zerner developed regeneration times, which cut a transient walk into independent pieces after conditioning against backtracking. Finite mean duration of those pieces gives a law of large numbers with positive speed. The difficulty is that finite duration almost surely does not imply finite expected duration. The regeneration law of large numbers

Many sufficient conditions for ballisticity were subsequently proved. Kalikow-type drift conditions, Sznitman’s slab-exit conditions and effective criterion, and polynomial exit criteria all supply quantitative control strong enough to rule out excessive trapping. The conjecture addressed here starts from only qualitative escape in one direction. Fribergh and Kious explicitly formulate that gap and study the role of local trapping. Local trapping

Dependence changes the picture again. Bramson, Zeitouni, and Zerner constructed stationary, uniformly elliptic, polynomially mixing environments in higher dimensions with positive probabilities of escape in opposite directions. Thus a theorem for finite-range dependence cannot be replaced by a vague assertion that sufficiently weak correlations should suffice. The source manuscripts carefully preserve exact independence beyond the stated range.

4. What the new manuscripts claim

A zero–one law with only strict ellipticity

M2 states that for every dimension $d\geq3$, every iid nearest-neighbor environment with strictly positive transition probabilities, and every fixed nonzero real direction $\ell$,

$$P_0(A_\ell)\in\{0,1\}.$$

There is no uniform lower bound, inverse-probability moment condition, drift assumption, reversibility assumption, or positive-speed premise. The direction need not have rational coordinates. The theorem is asserted for each fixed direction; it does not require a single exceptional null set that works simultaneously for every real direction.

An extension to finite-range dependence

M1 gives the same directional conclusion for $d\geq3$ when the row environment is stationary, ergodic, uniformly elliptic, and finite-range dependent. Stationarity means that translating the whole environment preserves its law; ergodicity rules out nontrivial translation-invariant random alternatives.

The assumptions differ from M2 in two ways: independence of all sites is weakened, but uniform ellipticity is imposed. Combining the headlines into a theorem with both arbitrary finite-range dependence and mere strict ellipticity would go beyond what is stated.

Directional escape implies positive velocity

M3 states that for an iid uniformly elliptic nearest-neighbor environment in every dimension $d\geq2$,

$$P_0(A_\ell)=1 \quad\Longrightarrow\quad \frac{X_n}{n}\longrightarrow v \quad P_0\text{-almost surely}, \qquad v\cdot\ell>0,$$

where $v$ is deterministic and depends on the environment law. No regeneration-time moment is assumed; obtaining the required duration control is part of the claim.

For $d\geq3$, combining this with M2 upgrades positive-probability directional escape to the same positive-velocity conclusion. If one such direction exists, the manuscript identifies all deterministic escape directions:

$$\{u\in S^{d-1}:P_0(A_u)>0\} = \{u\in S^{d-1}:P_0(A_u)=1\} = \{u\in S^{d-1}:v\cdot u>0\}.$$

They form the open hemisphere pointing along the velocity. Each fixed equatorial direction, with $v\cdot u=0$, has zero escape probability. This last conclusion uses the combined theorems; convergence of $X_n/n$ alone would not rule out slower unbounded motion in an equatorial direction.

5. Why the advance matters

Importance: 5/5. Directional zero–one behavior and ballisticity are foundational questions about motion in a persistent random medium. The claims would settle both the higher-dimensional iid directional conjecture and the uniformly elliptic ballisticity conjecture, while adding an exact finite-range-dependent extension of the first.

The results separate two kinds of uncertainty. The zero–one theorem rules out a probabilistic choice between opposite eventual escape signs under its hypotheses. The speed theorem rules out indefinite slowdown once almost-sure escape is known under stronger ellipticity and independence assumptions.

The common methods also aim to extract quantitative information from very weak qualitative premises. That is a significant change from requiring a strong drift or a pre-established exit estimate at the outset.

6. The main idea

Cut the path where the past stops mattering

A regeneration point is a new height record below which the future path never returns. After such a point, the walk can use fresh environmental rows. Successive pieces then have a factorization that resembles independent increments.

These times depend on the future, so they are not ordinary stopping times. Independence must be established from the weights of finite path pieces. A particularly useful detail is that a path uses the row at a site when it departs, not merely when it arrives. Stopping on first arrival leaves that row available to a fresh continuation.

Under strict ellipticity, a short prescribed path through previously unused rows has positive annealed probability given by a product of positive row averages. There is no need for one uniform lower bound across all environments.

Rule out coexistence by counting contacts two ways

The zero–one arguments suppose that escape in both signs has positive probability. They first derive finite mean spatial radius for regeneration pieces. Very large sideways excursions could otherwise be spliced to an opposite continuation, forcing incompatible probabilities for arbitrarily large finite height maxima.

Next, two independent sequences of regeneration pieces are placed in opposed arrangements. The manuscripts estimate how many geometric scales contain contacts between them. An entropy comparison with independent bridges makes contacts relatively scarce. A second argument starts one path near the other’s random endpoint and forces more contacts.

Selecting that endpoint naively would be too costly. The proof tracks posterior probabilities of possible endpoints as a vector of martingales, controlling their accumulated maxima. The two resulting contact counts are incompatible, excluding coexistence.

For finite-range dependence, disjoint paths are not necessarily independent. M1 inserts short sequences of environment-independent forced steps to cross a gap wider than the dependence range. Contacts are then defined by proximity within that range, and all transfers retain the actual row information already used.

Control time only after controlling space

Ballisticity requires more than spatially finite regeneration pieces. A piece can stay in a modest region for a very long time. M3 studies the quenched probability of reaching a high level before a backward level and seeks to show that extremely small success probabilities are exceptionally rare.

The argument compares two walks in the same environment with two walks in independent environments, using spatial moments and truncated variances rather than assuming a central limit theorem based on positive speed. It then follows the distribution of successful endpoints across scales.

If survival repeatedly lost too much probability, normalized endpoint profiles would have to sustain that loss in a stationary construction. Entropy bounds restrict how their mass can diffuse or split among separated groups. An order comparison in the distant tails contradicts persistent loss. The resulting crossing estimate supplies finite mean regeneration duration, after which the renewal law of large numbers gives the positive velocity.

7. Scope, limitations, and open questions

The nearest-neighbor assumption is present throughout. The stated results do not cover arbitrary long jumps or environments that evolve in time.

The three hypotheses should remain separate. M2’s zero–one theorem needs only strict ellipticity but iid rows. M1 allows finite-range dependence but requires uniform ellipticity. M3’s speed theorem requires both iid rows and uniform ellipticity. It does not claim ballisticity for every strictly elliptic environment or for the dependent class of M1.

Dimension also matters: the speed theorem begins at two, and the one-dimensional counterexamples remain valid. The zero–one manuscripts address dimensions at least three, building on the earlier planar result.

A positive limiting velocity does not by itself specify Gaussian fluctuations, quantitative slowdown probabilities, or an explicit formula for the speed. Nor does the theorem decide which environment laws are directionally transient in the first place. It determines the consequences once that escape premise is met.

8. Sources and further reading

M2 is the iid strict-ellipticity directional theorem, M1 its uniformly elliptic finite-range-dependent extension, and M3 the ballisticity theorem. Their introductions explain the differences between spatial radius, projected width, and elapsed time in regeneration arguments.

For context, read Solomon on one-dimensional environments, Zerner’s planar zero–one proof, and Fribergh–Kious on trapping and the ballisticity conjecture. These predecessors make clear why positivity of transition probabilities, independence, and dimension play distinct roles.

Source manuscripts

• M1. OpenAI, A directional zero-one law for finite-range-dependent random environments (October 5, 2026).
• M2. OpenAI, A directional zero–one law under strict ellipticity (September 23, 2026).
• M3. OpenAI, Directional transience implies ballisticity (September 23, 2026).

Sources consulted on 11 October 2026 at repository revision fd4aeeb. Download the manuscript citations.