Polynomial Decay in Absorbing Continuous-Time Markov Chains
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.
Ada: We're looking at the paper: "Polynomial Decay in Absorbing Continuous-Time Markov Chains".
Emmy: Here's the gist of the paper. Polynomial decay in absorbing continuous-time Markov chains studies asymptotic relations of the form pij (t) ∼ aijL(t), t → ∞, where L is independent of i and j, and investigates conditions under which this common asymptotic scale determines the global asymptotic structure of the transition function <ref:2609.27335#pg2> The analysis applies to both reducible and irreducible chains <ref:2609.27335#pg2> Asymptotic Structure and Rank-One Conditions The paper explores whether a common scale function L forces the coefficient matrix A = (aij) to have rank one, and describes the consequences for survival probabilities <ref:2609.27335#pg15> The analysis applies to both reducible and irreducible chains <ref:2609.27335#pg2> Conjecture 1.2 states that the coefficients factor as cij = Aximj, where m and x are a positive λ-invariant measure and vector, or more generally, positive λ-subinvariant quantities <ref:2609.27335#pg2> Several results below support this rank-one picture <ref:2609.27335#pg2> It does not, however, follow from irreducibility alone <ref:2609.27335#pg2>.
Ada: Renewal Framework and Fixed-State Asymptotics The paper develops a renewal approach to examine polynomial decay generated by a single exceptionally long excursion <ref:2609.27335#pg2> The diagonal renewal equation (6) is derived, showing that the tail of gaa is controlled by the singular behaviour of Kba at 0 <ref:2609.27335#pg2> For the killed stable M/M/1 queue, Seneta’s asymptotics yield a proper LCD <ref:2609.27335#pg20> Theorem 3.8 establishes that for every fixed i, j ∈ C, pij (t) ∼ Aximj t −κL(t), t → ∞ <ref:2609.27335#pg17> The right vector x and left vector m are 0-subinvariant <ref:2609.27335#pg17> Counterexamples and Spectral Representations The paper provides counterexamples to the rank-one conjecture, such as the hub-and-traps model, where fixed-state probabilities have a common polynomial exponent but their limiting coefficient kernel is not rank one <ref:2609.27335#pg2> The killed random walk on a homogeneous tree serves as an irreducible counterexample where the coefficient kernel is positive semidefinite but not rank one, and survival occurs on a different exponential scale <ref:2609.27335#pg2> Spectral representations show that in the scalar Karlin–McGregor setting, regular variation at the lower endpoint produces a common fixed-state exponent and rank-one coefficients <ref:2609.27335#pg2> In a multichannel representation, the symmetrized edge coefficient is instead a positive semidefinite kernel formed from inner products of edge vectors, and its rank records the number of asymptotically visible edge directions <ref:2609.27335#pg2> Survival Asymptotics and Tightness Survival asymptotics are governed by a different issue than fixed-state limits, requiring control of the normalized mass outside finite sets <ref:2609.27335#pg2> Proposition 5.1 identifies asymptotic tightness as the required condition for summing fixed-state limits to obtain a proper LCD <ref:2609.27335#pg2> The hub-and-traps model shows that summability of the limiting left factor does not imply asymptotic tightness <ref:2609.27335#pg2> Concluding Perspectives Two complementary pictures emerge from the results, where the renewal picture suggests a long-lived or late-returning path is often composed of an ordinary entrance segment, one exceptional excursion, and an ordinary exit segment <ref:2609.27335#pg2> The spectral picture suggests that polynomial decay is the Laplace-transform image of power-law spectral mass near the decay parameter <ref:2609.27335#pg2> The central structural question is when irreducibility reduces the asymptotically dominant long-time mechanism to one effective channel <ref:2609.27335#pg2> Survival is governed by a different issue, and fixed-state limits determine survival only when the normalized transition measures are asymptotically tight <ref:2609.27335#pg2> References
1: W.
Emmy: J.
Ada: Anderson and P.
Title and authors: Ada: So we're talking about this paper called "Polynomial Decay in Absorbing Continuous-Time Markov Chains". It looks at how probabilities change over a long time, specifically looking for relations like p i j t is proportional to some function L of t as t goes to infinity.
Emmy: Yeah, it’s about those asymptotic relations and whether that common scale L forces the matrix A of these transition probabilities to have rank one. It applies to both reducible and irreducible chains, which is a pretty broad scope for this kind of analysis.
Kolya: I'm curious if this rank one requirement actually holds up when you look at the whole picture, especially since we know that irreducibility by itself doesn't guarantee that common asymptotic scale.
Ada: Exactly. The paper explores whether having a common scale function L is strong enough to dictate the global structure of the transition function. It looks at how this affects survival probabilities and conditional distributions when we are dealing with these decaying probabilities.
Emmy: We see some conjectures, like conjecture number two, suggesting that the coefficients factor in a certain way involving measures and vectors. The paper supports that rank one picture in several places, but it also provides counterexamples to show it's not always true.
Kolya: Counterexamples are important because they show where the rank one assumption breaks down, like with a hub-and-traps model where fixed state probabilities have a common polynomial exponent but their limiting coefficient kernel isn't rank one.
Ada: That’s right. And it also shows that in some cases, you can have an irreducible chain where the coefficient kernel is positive semidefinite but not rank one, and survival occurs on a different scale than what you might expect from fixed-state limits.
Emmy: The renewal approach they develop uses a single exceptionally long excursion to examine polynomial decay. They derive a diagonal renewal equation, which shows that the tail of the probability of staying in state i is controlled by how something behaves at zero for some kernel K.
Title and authors: Kolya: And for specific models, like the killed stable M/M/one queue, Seneta's asymptotics give us a proper limiting conditional distribution <ref:2609.27335#pg2>. Theorem three point eight establishes that for every fixed i and j, p i j t is proportional to A x m L of t as t goes to infinity.
Ada: So they identify some zero subinvariant vectors and measures for this limit <ref:2609.27335#pg2>. It points toward the idea that fixed-state asymptotics can determine the behavior of the whole transition function under certain conditions.
Emmy: But there's a distinction between fixed-state limits and survival asymptotics, which requires controlling the mass outside finite sets. Proposition five point one identifies asymptotic tightness as being necessary for summing those fixed-state limits to get a proper limiting conditional distribution.
Kolya: The hub-and-traps model gives us another example where summability of the limiting left factor doesn't guarantee asymptotic tightness, which is a big caveat we have to keep in mind.
Ada: So the paper suggests that for survival asymptotics, you need an extra condition beyond just summing up the fixed state limits. It shows that this missing mass is carried by traps whose indices grow with t when asymptotic tightness fails.
Emmy: The conclusion is that two pictures emerge here: the renewal picture suggesting a path is made of an ordinary entrance segment, one exceptional excursion, and an ordinary exit segment. And the spectral picture suggests polynomial decay is related to the Laplace transform of power-law spectral mass near the decay parameter.
Kolya: So for someone just listening to this show, what does this mean? It suggests that a single dominant long-excursion mechanism corresponds naturally to one effective edge channel, whereas multiple distinct long-time mechanisms might look like higher edge multiplicity spectrally.
Ada: That's the central structural question they address: when irreducibility actually reduces the dominant long-time mechanism to one effective channel. And survival is tied to that tightness condition for fixed-state limits.
Emmy: It’s a nuanced picture where the renewal and spectral viewpoints complement each other, showing different ways polynomial decay can be understood in these chains.
Kolya: I just want to say that the paper makes it clear that we need to look at survival probabilities separately because limiting conditional behavior is governed by how those two things interact.
Title and authors: Ada: Right, so while fixed-state limits give us some information, they only determine survival when the normalized transition measures are asymptotically tight. That’s a key condition we need to satisfy before we can rely on those fixed-state results for survival questions.
Emmy: So we're left with two different perspectives on how these chains behave over time, one focusing on the path structure and the other on the underlying spectral mass near zero.
Kolya: That’s what I see—one picture is about a single long excursion, and the other is about spectral properties near that decay parameter. It gives us a framework to approach these complex decaying systems from different angles.
Ada: So we've covered the main points of "Polynomial Decay in Absorbing Continuous-Time Markov Chains" and seen how it connects fixed-state asymptotics with survival behavior through conditions like asymptotic tightness.
Emmy: We’ve also seen how the paper uses renewal theory and spectral analysis to build these pictures, even when irreducibility doesn't immediately force a rank one matrix.
Kolya: It really emphasizes that in complex systems, you often need to look at different aspects—the path structure versus the spectrum—to get a complete picture of the long-time behavior.
Ada: So we wrap up here with this paper, showing how polynomial decay can be understood through these complementary lenses. We’ve seen how it relates to renewal ideas and spectral mass near zero.
Emmy: It points us toward figuring out when irreducibility actually simplifies the dominant mechanism down to a single channel, and survival depends on that tightness condition we talked about.
Kolya: For anyone working on these kinds of systems, this paper is useful because it shows exactly where the assumptions break down and what extra conditions are needed for those limits to hold.
Ada: Exactly. We’ve looked at how fixed-state limits determine survival only when the normalized transition measures are asymptotically tight. That’s a crucial point we should keep in mind as we look at other models, like those with different decay rates or structures.
Emmy: So moving on, next time we'll be looking at some work on three-ends Furstenberg estimates and approximate rectangles to see how they connect to incidence estimates.
The paper's summary: Ada: So, to quickly recap, this paper is about studying how probabilities in absorbing continuous time Markov chains decay over time. They look at that relationship where the probability of going from state i to state j eventually looks like some function L of time multiplied by some constant a i j as time goes on.
Emmy: Right, and they’re asking if this common scale function L forces the matrix of these probabilities to have a rank one structure. They test this on both reducible and irreducible chains, which is important because it shows that having a common decay rate doesn't automatically mean the entire system has a very simple structure.
Ada: It seems like they’re exploring whether this common scale actually dictates the whole long-term behavior of the transition function. They look at how those fixed state probabilities behave and whether they get stuck in some kind of rank one pattern.
Emmy: And they bring in renewal theory to look at a single, very long path—an exceptionally long excursion—to see how that generates polynomial decay. This helps them derive some formulas for the diagonal entries, like the probability of staying in a state for a while.
Ada: What's interesting is how they use these ideas to analyze specific models, like that killed stable M/M/one queue. They get a proper limiting conditional distribution there which gives us some concrete numbers to work with.
Emmy: But then they have to switch gears because survival probabilities are governed by something different than just the fixed state limits. It requires controlling the mass outside any finite set of states, which is a whole other kind of analysis.
Ada: They show that you need a specific condition called asymptotic tightness to be able to sum up those fixed state limits and actually get a proper limiting conditional distribution for survival. Without that tightness, the results don't hold up for survival probabilities.
Emmy: And the paper suggests a couple of ways to understand this decay—a renewal picture where a long path is just an entrance, one big excursion, and an exit segment. Or they have a spectral picture where that polynomial decay is actually related to some power law spectrum near zero.
Ada: So the core question they tackle is whether having an irreducible chain simplifies the dominant mechanism down to just one effective channel. And survival depends entirely on whether those normalized transition measures are asymptotically tight for those fixed state limits to matter.
Emmy: It’s a lot of structure being pulled from these different angles, showing that polynomial decay can be understood both through the path it takes and through its underlying spectral mass near zero.
Ada: It really highlights how we need to look at survival probabilities separately because the behavior of fixed state limits and survival is governed by different things.
Emmy: That’s what I mean—one picture focuses on a single long excursion, and the other focuses on the spectral mass near that decay parameter. It gives us a framework to approach these decaying systems from different angles.
Ada: So we've seen how this paper connects renewal ideas and spectral analysis to polynomial decay, showing that fixed state limits only determine survival when those measures are tight.
Emmy: It points us toward figuring out exactly when irreducibility simplifies the dominant mechanism down to a single channel, and survival depends on that tightness condition we talked about.
The paper's improvements: Ada: So, we're looking at what the authors suggest they can do better with this work on polynomial decay in Markov chains. They’re trying to refine how we see that common asymptotic scale L and its connection to rank one matrices.
Emmy: Right, they are exploring how their renewal framework can be made more precise, specifically by developing a defective renewal equation for the diagonal entries. This helps them get a tighter bound on those probabilities.
Ada: That means they’re trying to show that even with the complexity of absorbing chains, you can use this single long path idea to control the tail behavior more strictly. It's about making sure the approximation isn't just an asymptotic guess but something we can pin down with better math.
Emmy: And they are looking at how they handle the multichannel representation. They’re trying to move beyond just saying rank one is governed by effective spectral multiplicity one at the dominant edge, and calculating that symmetrized limiting coefficient kernel more rigorously as a positive semidefinite kernel formed from inner products of edge vectors.
Ada: So, instead of just a general idea about how structure works, they’re proposing specific tools—like the defective renewal equation and tighter definitions for those kernels—to actually prove those structural claims.
Emmy: It moves the argument from just saying that rank one is what happens in certain scalar settings to showing exactly why that's true when you have multiple edges involved, linking it to the geometry of those edge vectors.
Ada: That’s a big step because it connects the algebraic structure—the rank one condition—directly to geometric properties of the underlying space, which I think is really powerful.
Emmy: It opens up avenues for other types of chains where that specific geometric insight might apply, maybe in situations where we have more complex interactions than just simple scalar settings.
Ada: So, they’re not just describing a phenomenon; they’re building a better machinery to prove the conditions under which that phenomenon occurs in more complicated environments.
Emmy: And this improved framework gives us a clearer map for when we can expect those common asymptotic scales to lead to the simple rank one structure we've been looking for.
Conclusion: Tom: So we're wrapping up on "Polynomial Decay in Absorbing Continuous-Time Markov Chains" by Ada and Emmy, looking at what this work means for us.
Ada: To recap, this paper tackles how probabilities in absorbing continuous time Markov chains decay over time, focusing on that common asymptotic scale L and whether it forces the matrix to have a rank one structure.
Emmy: Right, they found that polynomial decay can be understood through two different pictures: one based on a single long excursion and another based on spectral mass near zero.
Kolya: I think the most important part is seeing how those fixed state limits only determine survival when you have that tightness condition we talked about, which is a big caveat for applying these results broadly.
Ada: Exactly. So, while fixed state limits are useful for some things, they don't tell the whole survival story unless those transition measures are asymptotically tight.
Emmy: And the improvements they suggested—using that defective renewal equation and more rigorous kernels—give us better tools to prove those structural claims in more complex settings than just simple scalar cases.
Kolya: From a probabilistic view, it shows that the tail behavior is governed by how fast you can control that mass outside finite sets, which is where the tightness condition comes into play.
Ada: So, this paper gives us a solid framework to understand how these chains behave over long times using both renewal and spectral methods.
Emmy: It also points toward the central question of when irreducibility actually simplifies the dominant mechanism down to just one effective channel.
Kolya: For me, it’s useful because it shows exactly where the assumptions break down and what extra conditions you need for those limits to hold up in real applications.
Ada: Exactly. So, we've looked at how polynomial decay relates to fixed state limits and survival through these complementary lenses.
Emmy: We can use this paper as a guide for building better machinery when dealing with these kinds of decaying systems in the future.
Kolya: Next up, we’re looking at some work on three-ends Furstenberg estimates and approximate rectangles to see how those connect to incidence estimates.
Phil Pollett
math.PR
Submitted: 2026-09-23
Updated: 2026-11-13
Comments: 40 pages, 0 figures
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 80/100
The gist: Polynomial decay in absorbing continuous-time Markov chains studies asymptotic relations of the form pij (t) ∼ aijL(t), t → ∞, where L is independent of i and j…
Key concepts
- Polynomial Decay
- This refers to how quickly the probability of transitioning between states (or remaining in a state) decreases over time, often following a power law rather than simple exponential decay. The study focuses on finding common scaling functions $L(t)$ that govern this decay across different transition probabilities.
- Rank-One Condition
- This is a structural property of the coefficient matrix $A = (a_{ij})$ where all entries can be expressed as the product of two vectors, i.e., $a_{ij} = u_i v_j$. The paper investigates if the existence of a common asymptotic scale function $L(t)$ implies this rank-one structure, which would simplify the long-term dynamics.
- Renewal Framework
- This approach uses renewal theory to model polynomial decay by considering the path as a sequence of segments: an entrance, one exceptionally long excursion, and an exit. This helps analyze how a single very long event dictates the overall asymptotic behavior of probabilities in the chain.
Terminology
Summary
Polynomial decay in absorbing continuous-time Markov chains studies asymptotic relations of the form pij (t) ∼ aijL(t), t → ∞, where L is independent of i and j, and investigates conditions under which this common asymptotic scale determines the global asymptotic structure of the transition function <ref:2609. p. 0 Polynomial Decay in Absorbing Continuous-Time
Markov Chains
Phil Pollett
Abstract Let P(t) be the transition function of an absorbing continuoustime Markov chain on a countable state space. We study asymptotic relations
of the form pij (t) ∼ aijL(t), t → ∞, where L is independent of i and j. We obtain general conditions under which the coefficient matrix A = (aij ) has rank
one and describe the resulting consequences for survival probabilities and conditional distributions. The analysis applies to both reducible and irreducible
chains.Where the paper says it
Where the paper says it
Asymptotic Structure and Rank-One Conditions
The paper explores whether a common scale function L forces the coefficient matrix A = (aij) to have rank one, and describes the consequences for survival probabilities <ref:2609.27335#pg15> The analysis applies to both reducible and irreducible chains <ref:2609. p. 0 We obtain general conditions under which the coefficient matrix A = (aij ) has rank
one and describe the resulting consequences for survival probabilities and conditional distributions. The analysis applies to both reducible and irreducible
chains. We state them below
in a regular-variation form before describing the qualifications and counterexamples developed
here. Conjecture 1. 2 (Rank-one coefficients). The coefficients factor as
cij = Aximj
,
where m and x are a positive λ-invariant measure and vector, respectively, or, more generally,
positive λ-subinvariant quantities, with equality in (2) replaced by the corresponding inequalities (≤). The coefficients factor as
cij = Aximj
,
where m and x are a positive λ-invariant measure and vector, respectively, or, more generally,
positive λ-subinvariant quantities, with equality in (2) replaced by the corresponding inequalities (≤). Several results below support this rank-one picture: it holds for the killed stable M/M/1 queue,
the hub-and-traps model, and birth–death processes satisfying the spectral-edge hypothesis of
Theorem 4. Several results below support this rank-one picture: it holds for the killed stable M/M/1 queue,
the hub-and-traps model, and birth–death processes satisfying the spectral-edge hypothesis of
Theorem 4. It does not, however, follow from irreducibility alone.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Renewal Framework and Fixed-State Asymptotics
The paper develops a renewal approach to examine polynomial decay generated by a single exceptionally long excursion <ref:2609.27335#pg2> The diagonal renewal equation (6) is derived, showing that the tail of gaa is controlled by the singular behaviour of Kba at 0 <ref:2609. p. 4 Its Laplace transform is
gbaa(z) = 1
(z + qa − λ)(1 − Kba(z))
. (7)
Thus the tail of gaa is controlled by the singular behaviour of Kba at 0. A proper LCD exists for the killed
stable M/M/1 queue, but not for the hub-and-traps or homogeneous-tree examples, where
conditioned mass escapes to infinity. More generally, passage from fixed-state asymptotics to
an LCD requires tightness of the corresponding normalized transition measures. The coefficients factor as
cij = Aximj
,
where m and x are a positive λ-invariant measure and vector, respectively, or, more generally,
positive λ-subinvariant quantities, with equality in (2) replaced by the corresponding inequalities (≤). Several results below support this rank-one picture: it holds for the killed stable M/M/1 queue,
the hub-and-traps model, and birth–death processes satisfying the spectral-edge hypothesis of
Theorem 4. In particular, with T0 =
inf{t ≥ 0 : X(t) = 0}, we have Pi(T0 > t) = 1 − pi0(t) = P
j∈C
pij (t) → 0 for every i ∈ C. 1
arXiv:2609. 2 (Rank-one coefficients). The coefficients factor as
cij = Aximj
,
where m and x are a positive λ-invariant measure and vector, respectively, or, more generally,
positive λ-subinvariant quantities, with equality in (2) replaced by the corresponding inequalities (≤).Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Counterexamples and Spectral Representations
The paper provides counterexamples to the rank-one conjecture, such as the hub-and-traps model, where fixed-state probabilities have a common polynomial exponent but their limiting coefficient kernel is not rank one <ref:2609.27335#pg2> The killed random walk on a homogeneous tree serves as an irreducible counterexample where the coefficient kernel is positive semidefinite but not rank one, and survival occurs on a different exponential scale <ref:2609. p. 0 The analysis applies to both reducible and irreducible
chains. We also construct an irreducible counterexample, based on a killed
random walk on a homogeneous tree, for which a common asymptotic scale
exists but the coefficient matrix has rank greater than one. In the homogeneous-tree
example the fixed-state probabilities have a common polynomial exponent, but their limiting coefficient kernel is not rank one. The multichannel analysis suggests that the relevant
structural condition is effective spectral multiplicity 1 at the lower edge. In the homogeneous-tree
example the fixed-state probabilities have a common polynomial exponent, but their limiting coefficient kernel is not rank one. The multichannel analysis suggests that the relevant
structural condition is effective spectral multiplicity 1 at the lower edge.Where the paper says it
Where the paper says it
Where the paper says it
Survival Asymptotics and Tightness
Survival asymptotics are governed by a different issue than fixed-state limits, requiring control of the normalized mass outside finite sets <ref:2609.27335#pg2> Proposition 5. p. 5 Suppose that the tilted return measure Ka has a density ka, and write ρa =
R ∞
0
ka(s) ds ∈ (0, 1). The renewal-density asymptotic
needed below is a direct consequence of standard weighted-convolution and random-sum results for subexponential densities; see [3, Corollary 4, Proposition 8 and Theorem 3] and [9,
Corollary 4. Suppose that the tilted return measure Ka has a density ka, and write ρa =
R ∞
0
ka(s) ds ∈ (0, 1). The renewal-density asymptotic
needed below is a direct consequence of standard weighted-convolution and random-sum results for subexponential densities; see [3, Corollary 4, Proposition 8 and Theorem 3] and [9,
Corollary 4. A proper LCD exists for the killed
stable M/M/1 queue, but not for the hub-and-traps or homogeneous-tree examples, where
conditioned mass escapes to infinity. More generally, passage from fixed-state asymptotics to
an LCD requires tightness of the corresponding normalized transition measures. A proper LCD exists for the killed
stable M/M/1 queue, but not for the hub-and-traps or homogeneous-tree examples, where
conditioned mass escapes to infinity. More generally, passage from fixed-state asymptotics to
an LCD requires tightness of the corresponding normalized transition measures. The coefficients factor as
cij = Aximj
,
where m and x are a positive λ-invariant measure and vector, respectively, or, more generally,
positive λ-subinvariant quantities, with equality in (2) replaced by the corresponding inequalities (≤). Several results below support this rank-one picture: it holds for the killed stable M/M/1 queue,
the hub-and-traps model, and birth–death processes satisfying the spectral-edge hypothesis of
Theorem 4. The coefficients factor as
cij = Aximj
,
where m and x are a positive λ-invariant measure and vector, respectively, or, more generally,
positive λ-subinvariant quantities, with equality in (2) replaced by the corresponding inequalities (≤). Several results below support this rank-one picture: it holds for the killed stable M/M/1 queue,
the hub-and-traps model, and birth–death processes satisfying the spectral-edge hypothesis of
Theorem 4.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Concluding Perspectives
Two complementary pictures emerge from the results, where the renewal picture suggests a long-lived or late-returning path is often composed of an ordinary entrance segment, one exceptional excursion, and an ordinary exit segment <ref:2609.27335#pg2> The spectral picture suggests that polynomial decay is the Laplace-transform image of power-law spectral mass near the decay parameter <ref:2609. p. 8 Under the original law, long returning excursions from state 1 are exponentially
rare, with a polynomial correction. After tilting at the decay parameter, the return density has
the heavy tail
k(t) ∼
1
2
√
2πθ
t
−3/2
. Under the original law, long returning excursions from state 1 are exponentially
rare, with a polynomial correction. After tilting at the decay parameter, the return density has
the heavy tail
k(t) ∼
1
2
√
2πθ
t
−3/2
. A proper LCD exists for the killed
stable M/M/1 queue, but not for the hub-and-traps or homogeneous-tree examples, where
conditioned mass escapes to infinity. More generally, passage from fixed-state asymptotics to
an LCD requires tightness of the corresponding normalized transition measures. A proper LCD exists for the killed
stable M/M/1 queue, but not for the hub-and-traps or homogeneous-tree examples, where
conditioned mass escapes to infinity. More generally, passage from fixed-state asymptotics to
an LCD requires tightness of the corresponding normalized transition measures.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
References
[1] W. J. Anderson and P. M. McDunnough, “On the representation of symmetric transition functions”, Advances in Applied Probability 22 (1990), 548–563 <ref:2609.27335#pg2>
[2] W. J. Anderson, Continuous-Time Markov Chains: An Applications-Oriented Approach, Springer-Verlag, New York, 1991 <ref:2609. p. 0 Polynomial Decay in Absorbing Continuous-Time
Markov Chains
Phil Pollett
Abstract Let P(t) be the transition function of an absorbing continuoustime Markov chain on a countable state space. We study asymptotic relations
of the form pij (t) ∼ aijL(t), t → ∞, where L is independent of i and j.Where the paper says it
[3] S. Asmussen, S. Foss and D. Korshunov, “Asymptotics for sums of random variables with local subexponential behaviour”, Journal of Theoretical Probability 16 (2003), 489–518 <ref:2609.27335#pg2>
[4] S. Asmussen and H. Hering, Branching Processes, Birkh¨auser, Boston, 1983 <ref:2609. p. 0 In particular, with T0 =
inf{t ≥ 0 : X(t) = 0}, we have Pi(T0 > t) = 1 − pi0(t) = P
j∈C
pij (t) → 0 for every i ∈ C. 1
arXiv:2609.Where the paper says it
[5] N. H. Bingham, C. M. Goldie and J. L. Teugels, Regular Variation, Encyclopedia of Mathematics and its Applications, vol. p. 3 Regular variation naturally produces
asymptotic forms of the type t
−κL(t); it does not by itself imply convergence after multiplication
by a pure power of t. (iv) Polynomial rates also arise in subgeometric ergodicity, including applications to Markov
chain Monte Carlo, although there the object of study is convergence to equilibrium rather
than decay in an evanescent class; see, for example, [7, 17]. In particular, with T0 =
inf{t ≥ 0 : X(t) = 0}, we have Pi(T0 > t) = 1 − pi0(t) = P
j∈C
pij (t) → 0 for every i ∈ C. 1
arXiv:2609.Where the paper says it
Where the paper says it
[6] N. Champagnat and D. p. 3 The analyses below suggest that this
distinction extends beyond the exponential scale and persists in the polynomial corrections to
large-t asymptotics. (ii) Such polynomial corrections also arise in recent work of Champagnat and Villemonais [6]
on absorbed Markov chains with reducible state spaces. The analyses below suggest that this
distinction extends beyond the exponential scale and persists in the polynomial corrections to
large-t asymptotics. (ii) Such polynomial corrections also arise in recent work of Champagnat and Villemonais [6]
on absorbed Markov chains with reducible state spaces. 27335v1 [math. PR] 23 Sep 2026 In particular, with T0 =
inf{t ≥ 0 : X(t) = 0}, we have Pi(T0 > t) = 1 − pi0(t) = P
j∈C
pij (t) → 0 for every i ∈ C. 1
arXiv:2609.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
[7] R. Douc, E. Moulines and P. Soulier, “Subgeometric ergodicity of Markov chains”, in Dependence in Probability and Statistics, Lecture Notes in Statistics, Springer, 2006, 55–64 <ref:2609. p. 3 Regular variation naturally produces
asymptotic forms of the type t
−κL(t); it does not by itself imply convergence after multiplication
by a pure power of t. (iv) Polynomial rates also arise in subgeometric ergodicity, including applications to Markov
chain Monte Carlo, although there the object of study is convergence to equilibrium rather
than decay in an evanescent class; see, for example, [7, 17].Where the paper says it
[8] L. R. G. Fontes and P. p. 9 The construction is reminiscent of the complete-graph trap models studied by Fontes and
Mathieu [8]. Their scaling limits are described by K-processes with countably many stable
states and a distinguished instantaneous or fictitious state. The construction is reminiscent of the complete-graph trap models studied by Fontes and
Mathieu [8]. Their scaling limits are described by K-processes with countably many stable
states and a distinguished instantaneous or fictitious state.Where the paper says it
Where the paper says it
[9] S. Foss, D. Korshunov and S. Zachary, An Introduction to Heavy-Tailed and Subexponential Distributions, 2nd edn. p. 6 Under the original
law, exceptionally long returning excursions are exponentially rare. After exponential tilting
at the decay parameter, the return density is heavy-tailed. In particular, with T0 =
inf{t ≥ 0 : X(t) = 0}, we have Pi(T0 > t) = 1 − pi0(t) = P
j∈C
pij (t) → 0 for every i ∈ C. 1
arXiv:2609.Where the paper says it
Where the paper says it
[10] D. L. Iglehart, “Reversible competition processes”, Zeitschrift f¨ur Wahrscheinlichkeitstheorie und Verwandte Gebiete 2 (1964), 314–331 <ref:2609.27335#pg2>
[11] S. D. Jacka and G. p. 3 (i) The distinction between the decay parameter governing fixed-state transition
probabilities and that governing survival probabilities has appeared previously in the study of
conditioned processes. Jacka and Roberts [11] introduced separate decay parameters for these
quantities and observed that they need not coincide. (i) The distinction between the decay parameter governing fixed-state transition
probabilities and that governing survival probabilities has appeared previously in the study of
conditioned processes. Jacka and Roberts [11] introduced separate decay parameters for these
quantities and observed that they need not coincide.Where the paper says it
Where the paper says it
[12] W. B. Jurkat, “On the analytic structure of semigroups of positive matrices”, Mathematische Zeitschrift 73 (1960), 346–365 <ref:2609. p. 3 Then, although G
is not a transition function, because P
j
gij (t) = e
λtPi(T0 > t) may exceed 1, it satisfies the
semigroup property G(t+s) = G(t)G(s). Consideration of such semigroups dates back at least
to Jurkat [12].Where the paper says it
[13] D. G. Kendall, “Integral representations for Markov transition probabilities”, Bulletin of the American Mathematical Society 64 (1958), 358–362 <ref:2609. p. 0 MSC 2020: Primary 60J35; Secondary 60J27, 60K05, 47A10. Keywords: absorbing continuous-time Markov chains; polynomial asymptotics; regular variation; defective renewal theory; spectral edge; birth–death processes; quasi-stationarity
1 Introduction
For an irreducible continuous-time Markov chain with an absorbing state, the transition probabilities between transient states may satisfy pij (t) ∼ aijL(t) as t → ∞.Where the paper says it
[14] D. G. Kendall, “Unitary dilations of one-parameter semigroups of Markov transition operators, and the corresponding integral representations for Markov processes with a countable infinity of states”, Proceedings of the London Mathematical Society (3) 9 (1959), 417–431 <ref:2609. p. 0 Keywords: absorbing continuous-time Markov chains; polynomial asymptotics; regular variation; defective renewal theory; spectral edge; birth–death processes; quasi-stationarity
1 Introduction
For an irreducible continuous-time Markov chain with an absorbing state, the transition probabilities between transient states may satisfy pij (t) ∼ aijL(t) as t → ∞. Does the existence of a
common scale function L force the coefficient matrix A = (aij ) to have rank one?Where the paper says it
[15] M. Kijima, M. G. Nair, P. K. Pollett and E. p. 0 Polynomial Decay in Absorbing Continuous-Time
Markov Chains
Phil Pollett
Abstract Let P(t) be the transition function of an absorbing continuoustime Markov chain on a countable state space. We study asymptotic relations
of the form pij (t) ∼ aijL(t), t → ∞, where L is independent of i and j. MSC 2020: Primary 60J35; Secondary 60J27, 60K05, 47A10. Keywords: absorbing continuous-time Markov chains; polynomial asymptotics; regular variation; defective renewal theory; spectral edge; birth–death processes; quasi-stationarity
1 Introduction
For an irreducible continuous-time Markov chain with an absorbing state, the transition probabilities between transient states may satisfy pij (t) ∼ aijL(t) as t → ∞.Where the paper says it
Where the paper says it
[16] J. F. C. Kingman, “The exponential decay of Markov transition probabilities”, Proceedings of the London Mathematical Society (3) 13 (1963), 337–358 <ref:2609. p. 3 The examples below
show that two qualifications are needed. In the hub-and-traps model, survival and fixed-state
transition probabilities have the same exponential decay parameter, but κ0 = κ − 1 < κ.Where the paper says it
[17] S. P. Meyn and R. L. Tweedie, Markov Chains and Stochastic Stability, Springer, 1993 <ref:2609. p. 0 Polynomial Decay in Absorbing Continuous-Time
Markov Chains
Phil Pollett
Abstract Let P(t) be the transition function of an absorbing continuoustime Markov chain on a countable state space. We study asymptotic relations
of the form pij (t) ∼ aijL(t), t → ∞, where L is independent of i and j.Where the paper says it
[18] E. Papageorgiou, “Long-time asymptotics for the heat kernel and for heat equation solutions on homogeneous trees”, arXiv:2603. p. 9 Exponential-mixture asymptotics. The long-time behaviour of the model is governed
by mixtures of exponential holding-time tails over traps of increasing depth. In particular, with T0 =
inf{t ≥ 0 : X(t) = 0}, we have Pi(T0 > t) = 1 − pi0(t) = P
j∈C
pij (t) → 0 for every i ∈ C. 1
arXiv:2609.Where the paper says it
Where the paper says it
[19] P. Pollett, “On the equivalence of µ-invariant measures for the minimal process and its q-matrix”, Stochastic Process. p. 9 Moreover, supi∈C qi = max{d + R, 1} < ∞. The q-matrix is therefore stable
and uniformly bounded, and thus regular, and so its minimal process is nonexplosive. In particular, with T0 =
inf{t ≥ 0 : X(t) = 0}, we have Pi(T0 > t) = 1 − pi0(t) = P
j∈C
pij (t) → 0 for every i ∈ C. 1
arXiv:2609.Where the paper says it
Where the paper says it
[20] P. Pollett, “Quasi-stationary distributions for queueing and other models”, Queueing Systems 100 (2022), 241–243 <ref:2609.27335#pg2>
[21] W. E. Pruitt, “Bilateral birth and death processes”, Transactions of the American Mathematical Society 107 (1963), 508–525 <ref:2609. p. 0 MSC 2020: Primary 60J35; Secondary 60J27, 60K05, 47A10. Keywords: absorbing continuous-time Markov chains; polynomial asymptotics; regular variation; defective renewal theory; spectral edge; birth–death processes; quasi-stationarity
1 Introduction
For an irreducible continuous-time Markov chain with an absorbing state, the transition probabilities between transient states may satisfy pij (t) ∼ aijL(t) as t → ∞.Where the paper says it
[22] E. Seneta, “Quasi-stationary behaviour in the random walk with continuous time”, Australian Journal of Statistics 8 (1966), 92–98 <ref:2609. p. 0 MSC 2020: Primary 60J35; Secondary 60J27, 60K05, 47A10. Keywords: absorbing continuous-time Markov chains; polynomial asymptotics; regular variation; defective renewal theory; spectral edge; birth–death processes; quasi-stationarity
1 Introduction
For an irreducible continuous-time Markov chain with an absorbing state, the transition probabilities between transient states may satisfy pij (t) ∼ aijL(t) as t → ∞.Where the paper says it
[23] E. Seneta and D. p. 7 Put
ν = p + q = q1, b =
p
p/q, θ = 2√
pq, λ = ν − θ. The transition probabilities are expressible using modified Bessel functions, and Seneta’s asymptotics [22] give
pij (t) ∼
2ijbj−i
θ
√
2πθ
e
−λtt
−3/2
. MSC 2020: Primary 60J35; Secondary 60J27, 60K05, 47A10. Keywords: absorbing continuous-time Markov chains; polynomial asymptotics; regular variation; defective renewal theory; spectral edge; birth–death processes; quasi-stationarity
1 Introduction
For an irreducible continuous-time Markov chain with an absorbing state, the transition probabilities between transient states may satisfy pij (t) ∼ aijL(t) as t → ∞. In particular, with T0 =
inf{t ≥ 0 : X(t) = 0}, we have Pi(T0 > t) = 1 − pi0(t) = P
j∈C
pij (t) → 0 for every i ∈ C. 1
arXiv:2609.Where the paper says it
Where the paper says it
Where the paper says it
[24] E. A. van Doorn, “Quasi-stationary distributions and convergence to quasi-stationarity of birth–death processes”, Advances in Applied Probability 23 (1991), 683–700 <ref:2609. p. 0 MSC 2020: Primary 60J35; Secondary 60J27, 60K05, 47A10. Keywords: absorbing continuous-time Markov chains; polynomial asymptotics; regular variation; defective renewal theory; spectral edge; birth–death processes; quasi-stationarity
1 Introduction
For an irreducible continuous-time Markov chain with an absorbing state, the transition probabilities between transient states may satisfy pij (t) ∼ aijL(t) as t → ∞.Where the paper says it
[25] E. A. van Doorn, “Spectral properties of birth–death polynomials”, Journal of Computational and Applied Mathematics 284 (2015), 251–258 <ref:2609. p. 0 MSC 2020: Primary 60J35; Secondary 60J27, 60K05, 47A10. Keywords: absorbing continuous-time Markov chains; polynomial asymptotics; regular variation; defective renewal theory; spectral edge; birth–death processes; quasi-stationarity
1 Introduction
For an irreducible continuous-time Markov chain with an absorbing state, the transition probabilities between transient states may satisfy pij (t) ∼ aijL(t) as t → ∞.Where the paper says it
[26] D. Vere-Jones, “Some limit theorems for evanescent processes”, Australian Journal of Statistics 11 (1969), 67–78 <ref:2609.27335#pg2>
[27] W. Woess, Random Walks on Infinite Graphs and Groups, Cambridge University Press, 2000 <ref:2609.27335#pg2>
How the training signal is built
The paper develops a defective-renewal equation for gaa to examine the mechanism of polynomial decay <ref:2609.27335#pg2> The tilted return measure Ka(ds) is introduced, leading to the defective renewal equation (6) <ref:2609. p. 4 Let qa = −qaa and define the defective return measure
Fa( ds) = Pa(τ
+
a ∈ ds, τ +
a < T0). A first-return decomposition gives the exact renewal equation
paa(t) = e
−qat +
Z
(0,t]
paa(t − s)Fa( ds). Suppose that the tilted return measure Ka has a density ka, and write ρa =
R ∞
0
ka(s) ds ∈ (0, 1). The renewal-density asymptotic
needed below is a direct consequence of standard weighted-convolution and random-sum results for subexponential densities; see [3, Corollary 4, Proposition 8 and Theorem 3] and [9,
Corollary 4. Proposition 2. 1 (A defective-renewal estimate). Sufficient conditions are provided by local subexponentiality of
the return density, suitable light-tail or convolution-equivalence properties of the entrance and
exit kernels, and negligibility of the taboo term. For the killed M/M/1 queue, the calculation below shows that the return kernel lies at
a critical exponential boundary. 3 The hub-and-traps model
We introduce an explicit hub-and-traps model in which polynomial decay arises from a mixture
of increasingly long holding times. The model makes the one-long-excursion mechanism transparent and separates fixed-state transition asymptotics from survival asymptotics.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the gains come from
The spectral edge theorem for birth–death processes establishes a common fixed-state exponent and rank-one coefficients <ref:2609.27335#pg2> The explicit hub transition asymptotic shows that for every fixed i, j ∈ C, pij (t) ∼ Aximj t −κL(t), t → ∞ <ref:2609. p. 0 This shows that
irreducibility alone does not imply rank-one asymptotics. The results identify conditions under which fixed-state transition asymptotics determine the
asymptotic behaviour of the entire transition function and clarify the limitations of such conclusions in the absence of additional structure. We also construct an irreducible counterexample, based on a killed
random walk on a homogeneous tree, for which a common asymptotic scale
exists but the coefficient matrix has rank greater than one. This shows that
irreducibility alone does not imply rank-one asymptotics. In the homogeneous-tree
example the fixed-state probabilities have a common polynomial exponent, but their limiting coefficient kernel is not rank one. The multichannel analysis suggests that the relevant
structural condition is effective spectral multiplicity 1 at the lower edge. In the homogeneous-tree
example the fixed-state probabilities have a common polynomial exponent, but their limiting coefficient kernel is not rank one. The multichannel analysis suggests that the relevant
structural condition is effective spectral multiplicity 1 at the lower edge.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Survival Asymptotics and Tightness Criterion
Survival asymptotics are governed by a different issue than fixed-state limits, requiring control of the normalized mass outside finite sets <ref:2609.27335#pg2> Proposition 5.1 identifies asymptotic tightness as the required condition for summing fixed-state limits to obtain a proper LCD <ref:2609.27335#pg2> The hub-and-traps model shows that summability of the limiting left factor does not imply asymptotic tightness <ref:2609.27335#pg2> The homogeneous-tree example exhibits a stronger failure at the fixed-state normalization <ref:2609. p. 2 It does not, however, follow from irreducibility alone. In the homogeneous-tree
example the fixed-state probabilities have a common polynomial exponent, but their limiting coefficient kernel is not rank one.Where the paper says it
Concluding Perspective
The paper concludes that two complementary pictures emerge, where the renewal picture suggests a long-lived or late-returning path is often composed of an ordinary entrance segment, one exceptional excursion, and an ordinary exit segment <ref:2609.27335#pg2> The spectral picture suggests that polynomial decay is the Laplace-transform image of power-law spectral mass near the decay parameter <ref:2609.27335#pg2> A single dominant long-excursion mechanism corresponds naturally to one effective edge channel, whereas several geometrically distinct long-time mechanisms may appear spectrally as higher edge multiplicity <ref:2609.27335#pg2> Survival is governed by a different issue, and fixed-state limits determine survival only when the normalized transition measures are asymptotically tight <ref:2609.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved system can analyze asymptotic relations of fixed-state transition probabilities by identifying conditions under which a common asymptotic scale exists, such as
pij (t) ∼ aijL(t), t → ∞,
and determining if this forces the coefficient matrix A to have rank one. -
Improved system can distinguish between the behavior of survival probabilities and fixed-state transition asymptotics by analyzing their respective decay parameters, noting that
the distinction between the decay parameter governing fixed-state transition probabilities and that governing survival probabilities has appeared previously in the study of conditioned processes.
-
Improved system can utilize renewal theory to derive asymptotic formulas for diagonal entries, such as
gaa(t) ∼ k(t) (qa − λ)(1 − ρa)2,
and apply these methods to analyze specific models like the killed stable M/M/1 queue. -
Improved system can implement spectral edge theorem analysis for birth-death processes, using the Karlin-McGregor formula to derive rank-one factorizations based on the behavior of a single scalar spectral measure near its lower edge.
-
Improved system can handle multichannel representations by determining that rank one is governed by
effective spectral multiplicity 1 at the dominant edge,
and calculating the symmetrized limiting coefficient kernel as a positive semidefinite kernel formed from inner products of edge vectors. -
Improved system can perform a spatial analysis of conditioned laws, showing that for survival up to time t,
the process is typically in a trap with index comparable to t,
leading to the scaled conditional limit law Li(Zt T0 > t) ⇒ L(Y), where Y has density fY (y) = y−κe−1/yΓ(κ − 1), y > 0. -
Improved system can identify the structural conditions for achieving a proper limiting conditional distribution, stating that
the missing mass is carried by traps whose indices grow with t
when asymptotic tightness fails.
Abstract
Let P(t) be the transition function of an absorbing continuous-time Markov chain on a countable state space. We study asymptotic relations of the form p ij(t) about a ij L(t), t to infinity, where L is independent of i and j. We obtain general conditions under which the coefficient matrix A=(a ij) has rank one and describe the resulting consequences for survival probabilities and conditional distributions. The analysis applies to both reducible and irreducible chains. We also construct an irreducible counterexample, based on a killed random walk on a homogeneous tree, for which a common asymptotic scale exists but the coefficient matrix has rank greater than one. This shows that irreducibility alone does not imply rank-one asymptotics. The results identify conditions under which fixed-state transition asymptotics determine the asymptotic behaviour of the entire transition function and clarify the limitations of such conclusions in the absence of additional structure.
Sources
Related papers
- Rank-One HCIZ Integrals and Multiplicative Convolutions
- Repeated averaging on random d-regular graphs
- Precise cover times for branching random walks on Hamming graphs: (iterated) logarithmic corrections
- Markov Chain CLTs: Resolving Open Problems
- Marked stable L'evy processes in Brownian local time
- A Sharp Refinement of Samuels' Inequality via Cap Moves