Random attractors and nonergodic attractors for diffusions with degeneracies

Lai-Sang Young; Renaud Raqu\'epas; Yuri Bakhtin

REVIEW 2 major objections 4 minor 27 references

Random attractors and nonergodic attractors for diffusions with degeneracies

T0 review · 2 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read For a diffusion whose noise dies at the boundary, this paper proves a complete trichotomy of long-run statistical behavior in dimensions one and two: random absorption, cycling without ergodic convergence, or a unique interior invariant mea

desk verdict A substantial classification of boundary-degenerate diffusions with a genuinely new nonergodic attractor, but the proof of case III has a real gap where the Lyapunov function is used on a set where it is not defined. read the letter →

arxiv 2508.20968 v1 pith:J3TRFGKE submitted 2025-08-28 math.PR math.DS

Yuri Bakhtin , Renaud Raquépas , Lai-Sang Young This is my paper

classification math.PRmath.DS MSC 37A2537H3060J60

keywords diffusionnoisedegeneracyboundarynonergodicbehaviorattractorheterocliniccyclingLyapunovfunctionempiricalmeasure

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper studies stochastic differential equations on a bounded domain whose noise vanishes on the boundary, so the domain's faces act as invariant sets of lower dimension. It claims that, in dimensions 1 and 2 and under boundary hyperbolicity assumptions, the long-run statistical behavior of every typical trajectory is completely classified into three exclusive scenarios: absorption to a randomly chosen attracting vertex or edge; convergence to a unique interior invariant measure; or, in the case of a stable stochastic cycle, perpetual cycling between corner measures with empirical measures that never converge but trace a fixed closed curve. The paper computes the set of weak limit points of the empirical measures in every case, and proves the needed recurrence and transience statements with a generalized Foster–Lyapunov hitting-time estimate that tolerates regions where the Lyapunov drift is positive. This classification is relevant because systems such as scalable reaction networks, where one or more species deplete, exhibit exactly this boundary degeneracy; the results spell out when an exponential growth rate is deterministic, random, or not defined.

What carries the argument

The arguments run on two mechanisms. Near each vertex, logarithmic coordinates turn the generator applied to a linear combination of log-distance terms into the vertex eigenvalues; near an edge, a corrected logarithmic distance Φ = θ(ln distance + ψ) with ψ solving a Poisson equation makes the generator approximately the transversal Lyapunov exponent Λ̄2. These local Lyapunov functions are stitched together with transition functions so that the generalized Foster–Lyapunov estimate (Theorem 6.2) applies, giving expected return times to a recurrent set without requiring uniform negative drift everywhere. For the cycling scenario, the stability index Π = ∏|λ₋|/λ₊ controls the iteration: each pa

What would settle it

Numerically integrate a 2D diffusion on [0,1]^2 satisfying Assumptions A–D3 whose drift makes all four vertices consistently oriented saddles with Π > 1, e.g. choosing the expanding eigenvalues (1,2,3,4) and contracting eigenvalues (−2,−2,−3,−5). The theorems predict that almost every trajectory is attracted to the boundary, visits the corners cyclically, and that the set of weak limit points of the empirical measures is exactly the four line segments connecting the measures µ_k defined by (5.8). If the empirical measures instead converge to a single measure, or accumulate on a set different f

Watch

Extended reading notes

Core claim

The central discovery is Theorem 2.9: under Assumptions A–D3 on a diffusion in the square (and its polygonal analogues), exactly one of three alternatives holds. (I) If at least one vertex or edge is attracting, then for every interior starting point the trajectory converges almost surely to one of these attractors, each with positive probability, and the empirical distribution converges to that attractor's unique ergodic measure. (II) If all vertices are consistently oriented saddles and the stability index Π, the product of the ratios of contracting to expanding eigenvalue magnitudes, exceeds 1, then almost every trajectory is attracted to the boundary and visits the four corners cyclicall

Load-bearing premise

The classification is complete only when the boundary is hyperbolic: the drift's eigenvalues at every vertex, the transversal Lyapunov exponent of every edge carrying an invariant measure, and the stochastic-cycle stability index must all be nonzero (and the index not equal to 1); the paper's own nonhyperbolic example shows the limit set then becomes the entire simplex of mixtures of vertex measures.

Editorial extensions

If this is right

Where a stable stochastic cycle exists, time averages are not just slow to converge; they provably never converge, and the entire set of their limit points is a deterministic closed curve that can be computed from the local eigenvalue data.
The random-choice scenario shows that an interior starting point cannot be assigned an ergodic component: every boundary attractor has positive probability from every interior point, so the system has one future shared by all initial conditions even though it branches randomly.
For scalable reaction networks, the trichotomy translates into three growth regimes for total mass: a deterministic exponential rate, a random exponential rate chosen among finitely many measures, or no asymptotic rate at all, with the growth rate fluctuating over longer and longer epochs.
The generalized hitting-time estimate applies beyond this setup: it proves recurrence for Markov processes whose Lyapunov function is allowed to increase in regions the process rarely visits, so long as drift decay dominates elsewhere.
In higher dimensions, the same scenarios appear as building blocks, including combinations such as random choice between several nonergodic attractors, and the paper shows that a full higher-dimensional classification remains open.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The explicit formula for the curve Γ in Proposition 2.8 suggests a direct numerical test: simulate a case-II system, measure the fraction of time spent near each corner over successive cycles, and check whether the limiting weights match the ratios built from ρ and λ₊; agreement would confirm that the cycle's stability index controls the empirical measures.
The nonhyperbolic example of Section 8.2 hints that the trichotomy is sharp at the boundary of the hyperbolic regime: once an eigenvalue or Lyapunov exponent vanishes, null-recurrent fluctuations mix all boundary measures, so the set of empirical limits becomes the entire simplex rather than a curve or a point.
The proof strategy suggests that a complete higher-dimensional classification will need to treat attractors as cell complexes with possible overlaps, since product examples already produce unions of faces and multiple nonergodic attractors that can be chosen randomly.
The generalized Foster–Lyapunov estimate could be applied to metastability problems in which the rare region of positive drift is visited only during short excursions; the estimate's condition (d) is exactly a quantitative bound on how rarely those excursions can last.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

Here's my take on Bakhtin–Raquépas–Young. The paper is strong: it gives a complete classification of long-run statistical behavior for 1- and 2-dimensional diffusions with boundary degeneracies, under explicit hyperbolicity assumptions. The genuinely new piece is scenario C—a nonergodic attractor where empirical measures cycle on a curve—which cannot happen in the smooth-boundary setting of Freidlin–Koralov. The Foster–Lyapunov hitting-time estimate (Theorem 6.2) is a real, reusable contribution. The local analysis near saddles and edges is careful, and the paper is honest about what is conjectural in higher dimensions and nonhyperbolic cases.

But I have a concern that is more serious than the two minor issues the reader reported (omitted proof of Lemma A.16, asserted reduction of case-I configurations). In Section 7.2, the proof of case III of Theorem 2.9 applies Meyn–Tweedie and needs strong recurrence of the compact set R = X◦ \ F^{*}_{r′}. The displayed estimate is

sup_{x∈R} E_x inf{t≥s: X(t)∈R} ≤ s + 3(sup_R Φ + s) + 2KT.

The problem is that Φ is explicitly constructed only on So∪Q, the boundary strips, and R is the complementary compact set. On R, Φ is not defined, and no extension is given. Theorem 6.2 requires Φ∈C² with LΦ bounded above along the path; a naive constant extension would create a transition layer where LΦ could be large and positive. So the estimate as written is invalid, and the existence of an interior invariant measure in case III is not actually established as written.

I want to be careful: this is a proof-completeness gap, not a refutation. The same construction may be repairable—for instance, by extending Φ to R in a controlled way or by modifying the recurrence argument to avoid evaluating Φ on R. But the burden is on the authors, and the paper as submitted doesn't discharge it.

Everything else in the classification—case I via the local attractor analysis and case II via the heteroclinic cycling arguments—looks coherent to me. The paper's own caveats about higher dimensions and nonhyperbolic boundaries are appropriate.

Who is this for? Anyone working on stochastic persistence, boundary-degenerate diffusions, or heteroclinic cycling. It deserves a serious referee, and I would send it to review, but the referee should insist on fixing the case III gap before publication. I would not be comfortable citing the classification theorem in its current form, though the hitting-time estimate and the scenario-C construction are citable.

Referee Report

2 major / 4 minor

Summary. The paper studies diffusions on bounded polytopes that are irreducible in the interior and degenerate on the boundary. In dimension 1, Theorem 2.2 gives a dichotomy: either convergence to one of the attracting endpoints, or a unique ergodic interior measure. In dimension 2, Theorem 2.9 claims a complete classification under hyperbolicity assumptions D1–D3 into three mutually exclusive cases: (I) an attracting vertex or edge, with empirical measure convergence to its ergodic measure; (II) a stable stochastic cycle, with almost-sure convergence of the trajectory to the boundary and nonconvergence of empirical measures whose limit set is an explicit closed curve Γ (Proposition 2.8); and (III) a unique ergodic interior invariant measure to which empirical measures converge almost surely. The proofs combine local analyses near vertices and edges, corrector functions for transversal Lyapunov exponents, and a new generalized Foster–Lyapunov hitting-time estimate (Theorem 6.2). Higher-dimensional and nonhyperbolic extensions are discussed, and the paper explicitly notes where the classification is not complete.

Significance. If the proof gaps noted below are repaired, this is a substantial contribution. The paper gives a complete and fairly explicit classification in dimensions 1 and 2 under weak irreducibility and generic boundary hyperbolicity, and it identifies a new nonergodic phenomenon: empirical measures can fail to converge and instead cycle along a curve of mixtures of vertex measures. The generalized Foster–Lyapunov estimate in Section 6 is a useful standalone tool. The authors are appropriately careful about scope: the classification is explicitly conditional on hyperbolicity, and Section 8 clearly distinguishes conjectures and examples from proved results. The paper also connects to stochastic persistence and to reaction-network growth rates, which should make it of interest beyond probability.

major comments (2)

[§7.2, around Eq. (7.13) and the application of Theorem 7.6] The proof of strong recurrence for case III is not valid as written. The displayed estimate sup_{x∈R} E_x inf{t≥s : X(t)∈R} ≤ s + 3(sup_R Φ + s) + 2KT invokes (7.13) with Φ evaluated at X(s), but Φ is constructed only on So∪Q via (7.7) and (7.11), and R = X°∖F*_{r'} is disjoint from So∪Q. Thus sup_R Φ is undefined, and the Markov step cannot be applied when X(s)∈R because (7.13) is only stated for x∉R. A constant extension of Φ to R would create a transition layer where LΦ may be large and positive, so condition (b) of Theorem 6.2 would not automatically hold. This is a load-bearing gap: it is used to obtain the strong recurrence condition needed for the Meyn–Tweedie theorem, and therefore the existence of the interior invariant measure in case III is not established as written. The same formal issue affects Theorem 6.2 itself, whose proof evaluates Φ at X(ζ_R)∈R although Φ is only defin
[§7.2, first paragraph; §7.1] The proof of case I is carried out only for a particular configuration of attracting and nonattracting vertices and edges. The text asserts that the local definitions (7.7)–(7.9), the extensions (7.11), and Lemmas 7.2, 7.3, and Proposition 7.4 apply to all situations covered by case I, 'provided that one adapts the definition of R and the proof of Lemma 7.1'. Since Theorem 2.9 is a complete classification, this reduction is load-bearing. The adaptation of the geometric construction of Si, So, Q and of the transition functions for arbitrary arrangements of attracting vertices/edges and nonattracting saddles/sources is not fully written out. The authors should either provide the general construction or state precisely which configurations are proved and how the remaining ones reduce to the case study.

minor comments (4)

[Appendix A.2, Lemma A.16] Lemma A.16 is stated without proof, with only 'the same ingredients can be used' as justification. Since it is invoked in Lemma 7.3 to control exit times in Q, please provide the proof or a clear reference.
[§6.2] Several internal references are inaccurate: 'Proposition 6.2' should be 'Theorem 6.2' in the proof of Theorem 6.2, and in §7.2 'Theorem 7.4' should be 'Proposition 7.4'.
[§3, proof of Theorem 2.2] The phrase 'the process makes finally many transitions' should read 'finitely many transitions'.
[§7.2, display after Theorem 7.6] The displayed estimate for sup_{x∈R} E_x inf{t≥s : X(t)∈R} also uses '(sup_R Φ + s)' with the same undefined sup_R issue. Even after repairing the major issue, this display should be rewritten to avoid evaluating Φ on R, or the extension of Φ should be made explicit.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the classification is derived from explicit hyperbolicity assumptions and independent Lyapunov/hitting-time estimates.

full rationale

All three cases of Theorem 2.9 are obtained by proving the relevant asymptotic statements from the assumptions, rather than by assuming them. The case partition itself is generated by signs of quantities computed from the SDE coefficients: eigenvalues at vertices (D1), transversal Lyapunov exponents (2.6)-(2.7) averaged against the edge invariant measure (D2), and the product (2.11)-(2.12) of saddle ratios (D3). None of these quantities is chosen after seeing the target classification. The hitting-time theorem (Theorem 6.2) is a general Foster-Lyapunov estimate whose hypotheses (a)-(d) are verified in Section 7.1 using local Lyapunov functions (7.7), transition-function extensions (7.11), and exit-time estimates from Appendix A; the verification is independent of the conclusion that a particular attractor or invariant measure exists. In case I, convergence to πA on B^x_A follows from Lemmas 4.12, 4.13, and 4.15-4.16, not by definition of A. In case II, the closed curve Γ and the nonconvergence of empirical measures are derived quantitatively in Proposition 5.9 and Corollaries 5.10-5.12 from the stability index Π > 1; the limit curve is not an input. In case III, existence and uniqueness of π° is imported from Meyn-Tweedie (external, Theorem 7.6) after checking minorization from C1/C2 and strong recurrence from (7.13); the Lyapunov parameters are chosen to satisfy the linear inequalities (7.10), a feasibility condition that uses only the signs of the linearized eigenvalues and, in the cycle case, Π < 1. Self-citations in Sections 1 and 8 ([Bak10, Bak11, BCPG], [LKYJW20, NY22]) are contextual applications and connections, not load-bearing premises. The internal proof gap noted by the skeptical reader in §7.2 — the estimate sup_{x∈R} E_x inf_{t≥s} ... uses Φ on R where (7.7)/(7.11) may not define it — is a proof-completeness issue, not a circularity: the displayed inequality is not equal by construction to the desired conclusion, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper's central results rest on the explicit model assumptions (A-D3) plus a suite of standard theorems from Ito calculus, Markov processes, and ergodic theory. The proof also introduces ad hoc Lyapunov coefficients and a weight theta, but these are existence constructions with no free empirical parameters. No new entities are postulated. The hyperbolicity assumptions are the main domain assumptions that make the classification possible.

free parameters (2)

Lyapunov coefficients beta_k (and alpha_k, gamma_k derived from them) = chosen to satisfy inequalities (7.8)-(7.9); existence shown in Lemma 7.1
Introduced in the proof of Theorem 2.9 cases I and III to construct a Lyapunov function with negative drift on the non-attracting parts of the boundary. They are not empirical fits, but they are ad hoc construction choices.
Lyapunov weight theta in (4.6) = sign chosen so LPhi has the correct sign; magnitude arbitrary
Scaling parameter in the corrected logarithmic distance used to build a local Lyapunov function near an edge. Its exact value is irrelevant to the conclusions.

assumptions (4)

domain assumption Assumptions A, B, C1, C2, D1-D3 from Section 2
Define the class of diffusions under study: smooth Stratonovich SDE on a polytope, invariance of all boundary components, irreducibility, one-point Hormander condition, and hyperbolicity at vertices, edges, and cycles.
standard math Standard Ito calculus, martingale inequalities, Feller semigroup and strong Markov property
Used throughout the proofs: Ito formula (4.2), exponential martingale inequality (Theorem A.1), optional stopping, and related tools.
standard math [Ben23, Theorem 2.7]: weak limit points of empirical distributions of a Feller Markov process are invariant
Invoked in Section 4.2 and 5.5 to restrict the set of possible limit points of empirical measures.
standard math Meyn-Tweedie criteria [MT93] as formulated in Theorem 7.6, and uniqueness/absolute continuity results [BH22, Theorems 6.34, 6.37]
Used in Section 7.2 to convert recurrence and minorization into convergence to a unique invariant measure in case III.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Random attractors and nonergodic attractors for diffusions with degeneracies." pith.science (2026). https://pith.science/paper/J3TRFGKE

@misc{pith2026250820968,
  author       = {Pith},
  title        = {Pith review of: Random attractors and nonergodic attractors for diffusions with degeneracies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J3TRFGKE}},
  note         = {Machine review of arXiv:2508.20968}
}

read the original abstract

We consider a diffusion on a bounded domain, assuming that the system is irreducible inside the domain and that the diffusion has varying degree of degeneracy on the domain's boundary. The long-term statistical properties of typical trajectories started inside the domain may be governed by one invariant measure or more than one invariant measure. We describe various possible scenarios. In dimensions 1 and 2 under boundary hyperbolicity assumptions, we give a complete classification of the limiting behavior and answer the question whether sequential averaging involving more than one invariant distribution occurs. In all cases, we compute the set of weak limit points of empirical measures. Our hitting-time estimates used to prove transience or recurrence are based on a new version of the Foster-Lyapunov technique. Extensions to nonhyperbolic boundaries and higher dimensions are discussed and an application to growth rates in scalable networks is given.

Figures

Figures reproduced from arXiv: 2508.20968 by the authors.

**Figure 2.1.** Sketch of the 2-dimensional setup in the coordinates ( [PITH_FULL_IMAGE:figures/full_fig_p006_2_1.png] view at source ↗

**Figure 5.2.** The setup for Propositions 5.2 and 5.3: neighborhoods Ur,2r and U2r,r of the origin O with segments introduced to define stopping times. Some subscripts have been omitted for readability. (0, r) × (0, 2r) using the segments γ0,r = (0, r) × {2r}, γ2,r = {r} × (0, r), γ1,r = (0, r) × {r}, γ′ 2,r = {r} × (0, 2r), see [PITH_FULL_IMAGE:figures/full_fig_p019_5_2.png] view at source ↗

**Figure 5.3.** The setup for Proposition 5.6. Some subscripts have been omitted for readability. alternative typically happens fast and leads only to a small change in the distance to the boundary. The proofs are postponed to Section 5.6. Next, we analyze exits from U2r,r = (0, 2r) × (0, r) using, in addition to γ1,r and γ2,r, the segments γ3,r = {2r} × (0, r) and γ ′ 1,r = (0, 2r) × {r}, see [PITH_FULL_IMAGE:figures/full_fig_p02… view at source ↗

Figures from the paper (4 more)

**Figure 5.4.** Figure 5.4: Visual reference for Proposition 5.7. Some subscripts have been omitted for readability. Lemma 5.4 and Proposition 5.6 say that the exit time from Gr,r has exponential tails and single out a typical scenario in which the process exits towards one of the two adjacent …

**Figure 5.5.** Figure 5.5: The setup for the proof of Proposition 5.7. 1. There is y0 > 0 such that for all y > y0 and for all n ∈ N, f n (y) ≤ y + n 2 y 2/3 . 2. There is y0 > 0 such that for all y > y0 and all n ∈ N satisfying n < y1/6 − 1, g n is well-defined and g n (y) ≥ y − n 2 y 2/3 . P…

**Figure 7.6.** Figure 7.6: Vertices and edges are separated into two categories: those that are attracting are associated with closed [PITH_FULL_IMAGE:figures/full_fig_p034_7_6.png]

**Figure 7.7.** Figure 7.7: A key feature of the construction is that hitting [PITH_FULL_IMAGE:figures/full_fig_p035_7_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 25 canonical work pages

[1]

Yuri Bakhtin, Small noise limit for diffusions near heteroclinic networks, Dyn. Syst. 25 (2010), no. 3, 413--431

work page 2010
[2]

Theory Relat

, Noisy heteroclinic networks, Probab. Theory Relat. Fields 150 (2011), no. 1, 1--42

work page 2011
[3]

Bass, Probabilistic techniques in analysis, Probability and its Applications, Springer-Verlag, New York, 1995

Richard F. Bass, Probabilistic techniques in analysis, Probability and its Applications, Springer-Verlag, New York, 1995

work page 1995
[4]

Yuri Bakhtin, Hong-Bin Chen, and Zsolt Pajor-Gyulai, Rare transitions in noisy heteroclinic networks, Mem. Amer. Math. Soc., in print

work page
[5]

Michel Bena \"i m, Stochastic persistence, arXiv preprint (2023), arXiv:1806.08450v3

arXiv 2023
[6]

Michel Bena\"im and Tobias Hurth, Markov chains on metric spaces: A short course, Universitext, Springer, 2022

work page 2022
[7]

Ethier and Thomas G

Stewart N. Ethier and Thomas G. Kurtz, Markov processes: characterization and convergence, John Wiley & Sons, Hoboken, 2009

work page 2009
[8]

Mark Freidlin and Leonid Koralov, Asymptotics in the D irichlet problem for second order elliptic equations with degeneration on the boundary , J. Differ. Equ. 332 (2022), 202--218

work page 2022

Show all 27 references

[9]

, Perturbations of parabolic equations and diffusion processes with degeneration: Boundary problems, metastability, and homogenization, Ann. Probab. 51 (2023), no. 5, 1752--1784

2023
[10]

Juraj F \"o ldes and Declan Stacy, Stochastic extinction, an average L yapunov function approach , arXiv preprint (2024), arXiv:2407.19606

2024 arXiv
[11]

Freidlin and Alexander D

Mark I. Freidlin and Alexander D. Wentzell, Random perturbations of dynamical systems, third ed., A Series of Comprehensive Studies in Mathematics, vol. 260, Springer-Verlag, Berlin, 2012

2012
[12]

Andrea Gaunersdorfer, Time averages for heteroclinic attractors, SIAM J. Appl. Math. 52 (1992), no. 5, 1476--1489

1992
[13]

Nguyen, Coexistence and extinction for stochastic K olmogorov systems , Ann

Alexandru Hening and Dang H. Nguyen, Coexistence and extinction for stochastic K olmogorov systems , Ann. Appl. Probab. 28 (2018), no. 3, 1893--1942

2018
[14]

Nguyen, and Peter Chesson, A general theory of coexistence and extinction for stochastic ecological communities, J

Alexandru Hening, Dang H. Nguyen, and Peter Chesson, A general theory of coexistence and extinction for stochastic ecological communities, J. Math. Biol. 82 (2021), no. 6, 56

2021
[15]

Nguyen, and Sebastian J

Alexandru Hening, Dang H. Nguyen, and Sebastian J. Schreiber, A classification of the dynamics of three-dimensional stochastic ecological systems, Ann. Appl. Probab. 32 (2022), no. 2, 893--931

2022
[16]

66, Springer-Verlag, Berlin, 2012

Rafail Khasminskii, Stochastic stability of differential equations, second ed., Stochastic Modelling and Applied Probability, vol. 66, Springer-Verlag, Berlin, 2012

2012
[17]

Shreve, Brownian motion and stochastic calculus, second ed., Graduate Texts in Mathematics, vol

Ioannis Karatzas and Steven E. Shreve, Brownian motion and stochastic calculus, second ed., Graduate Texts in Mathematics, vol. 113, Springer-Verlag, New York, 1991

1991
[18]

194, Cambridge University Press, Cambridge, 2012

Sergei Kuksin and Armen Shirikyan, Mathematics of two-dimensional turbulence, Cambridge Tracts in Mathematics, vol. 194, Cambridge University Press, Cambridge, 2012

2012
[19]

Wei-Hsiang Lin, Edo Kussell, Lai-Sang Young, and Christine Jacobs-Wagner, Origin of exponential growth in nonlinear reaction networks, Proc. Natl. Acad. Sci. U.S.A. 117 (2020), no. 45, 27795--27804

2020
[20]

Meyn and R.L

Sean P. Meyn and R.L. Tweedie, Stability of M arkovian processes II : Continuous-time processes and sampled chains , Adv. Appl. Probab. 25 (1993), 487--517

1993
[21]

Peter Nandori and Lai-Sang Young, Growth and depletion in linear stochastic reaction networks, Proc. Natl. Acad. Sci. U.S.A. 119 (2022), no. 51, e2214282119

2022
[22]

Bernt ksendal, Stochastic differential equations: an introduction with applications, sixth ed., Universitext, Springer-Verlag, Berlin, 2003

2003
[23]

Veretennikov, On the P oisson equation and diffusion approximation

\'E tienne Pardoux and Alexander Yu. Veretennikov, On the P oisson equation and diffusion approximation. I , Ann. Probab. 29 (2001), no. 3, 1061--1085

2001
[24]

Schreiber, Michel Bena \" m, and Kolawol \'e A.S

Sebastian J. Schreiber, Michel Bena \" m, and Kolawol \'e A.S. Atchad \'e , Persistence in fluctuating environments, J. Math. Biol. 62 (2011), 655--683

2011
[25]

Stroock, Markov processes from K

Daniel W. Stroock, Markov processes from K . I t\^o's perspective , Annals of Mathematics Studies, vol. 155, Princeton University Press, Princeton, NJ, 2003

2003
[26]

Floris Takens, Heteroclinic attractors: time averages and moduli of topological conjugacy, Bol. Soc. Bras. Mat. 25 (1994), no. 1, 107--120

1994
[27]

Meyn, Random-time, state-dependent stochastic drift for M arkov chains and application to stochastic stabilization over erasure channels , IEEE Trans

Serdar Y \"u ksel and Sean P. Meyn, Random-time, state-dependent stochastic drift for M arkov chains and application to stochastic stabilization over erasure channels , IEEE Trans. Autom. Control 58 (2012), no. 1, 47--59

2012

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Yuri Bakhtin, Small noise limit for diffusions near heteroclinic networks, Dyn. Syst. 25 (2010), no. 3, 413--431

work page 2010

[2] [2]

Theory Relat

, Noisy heteroclinic networks, Probab. Theory Relat. Fields 150 (2011), no. 1, 1--42

work page 2011

[3] [3]

Bass, Probabilistic techniques in analysis, Probability and its Applications, Springer-Verlag, New York, 1995

Richard F. Bass, Probabilistic techniques in analysis, Probability and its Applications, Springer-Verlag, New York, 1995

work page 1995

[4] [4]

Yuri Bakhtin, Hong-Bin Chen, and Zsolt Pajor-Gyulai, Rare transitions in noisy heteroclinic networks, Mem. Amer. Math. Soc., in print

work page

[5] [5]

Michel Bena \"i m, Stochastic persistence, arXiv preprint (2023), arXiv:1806.08450v3

arXiv 2023

[6] [6]

Michel Bena\"im and Tobias Hurth, Markov chains on metric spaces: A short course, Universitext, Springer, 2022

work page 2022

[7] [7]

Ethier and Thomas G

Stewart N. Ethier and Thomas G. Kurtz, Markov processes: characterization and convergence, John Wiley & Sons, Hoboken, 2009

work page 2009

[8] [8]

Mark Freidlin and Leonid Koralov, Asymptotics in the D irichlet problem for second order elliptic equations with degeneration on the boundary , J. Differ. Equ. 332 (2022), 202--218

work page 2022