A functional central limit theorem for weighted occupancy processes of the Karlin model

arXiv:2412.02160 · math.PR · Submitted 2024-12-03 · Read on arXiv

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: "A functional central limit theorem for weighted occupancy processes of the Karlin model".

Emmy: Here's the gist of the paper. A functional central limit theorem is established for weighted occupancy processes of the Karlin model, which is important because it provides a functional central limit theorem for statistics in this model that can be used to investigate asymptotic behaviors of random permutation matrices induced by Chinese restaurant processes with (α, θ)-seating The main result is the establishment of a functional central limit theorem for the weighted sum of occupancy counts (Dn,j)j∈N with prescribed weights a = (aj)j∈N defined as Sn ≡ Sn, a:= Xn j=1 ajDn,j The statistics in this form include the (odd-)occupancy processes in (1.1) and hence we refer to Sn as weighted occupancy processes For the general weights (aj)j∈N we impose ai+j − ai ≤ Cjβ for some β ∈

0, α2/2) and all i, j ∈ N ∪ 0 Our main result is a functional central limit theorem for the counting process (S⌊nt⌋)t∈[0,1: appropriately normalized, and we identify a new family of Gaussian processes in the limit Limit Process Characterization The limit process that will arise in our limit theorem is denoted by Zα(t) ≡ Zα, a(t):= Z R+×omega′ aN′(rt) − E′ aN′(rt) Mα(dr, dω ′), t ≥ 0 The process Zα is in fact a linear combination of the above with coefficients (aj)j∈N We refer to

33: for background on stochastic integrals with respect to Gaussian random measures Properties of the Limit Process Proposition 2.1 states that under the assumption (2.3) ai+j − ai ≤ Cjβ for all i, j ∈ N with β ∈ h 0,α 2, then (Zα(t))t≥0 forms a stochastic process with a version that is γ-Hölder-continuous for every γ ∈ (0, α/2) Remark 1.3 notes that the process Zα is in fact a linear combination of the above with coefficients (aj)j∈N Poissonized Model Convergence Theorem 3.1 proves the convergence of the Poissonized model, stating that Wfn(t) σn!

Ada: t∈

0,1: ⇒ (Zα(t))t∈

0,1: , as n → ∞ This is achieved by proving finite-dimensional convergence in Section 3.1 and tightness in Section 3.2 De-Poissonization Step Section 4 completes the proof of Theorem 1.1 by showing that the difference of the two processes is negligible in an appropriate sense Proposition 4.3 suggests a natural candidate to replace An (instead of An,ǫ used in (4.13)) as An:= (sup t∈

0,1: n 1/2 λn(t) − t ≤ log n) Application in Random Matrix Theory The main motivation for this work came from random matrix theory: using the general functional central limit theorem established here we shall investigate asymptotic behaviors of random permutation matrices induced by the Chinese restaurant processes with (α, θ)-seating Theorem 1.5 shows that S⌊nt⌋(f) − E S⌊nt⌋(f) P nα/2!

Emmy: t∈

0,1: a.s.w.

Ada: → sα,θ Γ(1 − α) 1/2 (Zα(t))t∈

0,1: with respect to P in D(

0, 1: ) Local Central Limit Theorems Lemma 4.4 provides a refined control on the local central limit theorem for Poisson and binomial distributions This lemma shows that max k∈Jnp P(N(np) = k) − P(Bn,p = k) ≤ C√np p log2 n + log4 n np, for all n ∈ N, p ∈

n−ǫ, 1/2: The proof of (4.16) and (4.17) relies on Stirling’s formula and concentration inequalities Tightness Argument The tightness argument involves showing that both Wcn(t) σn!

Title and authors: Ada: So, we’re looking at this paper today, "A functional central limit theorem for weighted occupancy processes of the Karlin model." It deals with how we count things in an infinite urn scheme where the sampling frequencies can be prescribed by a specific sequence of numbers.

Emmy: Exactly. The authors are taking the standard infinite urn setup and generalizing it by introducing these weights, which they call the sequence a subscript j of N, and they want to see what happens when you look at these weighted sums as n gets very large.

Kolya: What’s the main thing they’re trying to nail down here? Is this just about making sure the counts converge nicely?

Ada: It's more than just convergence; they are establishing a functional central limit theorem for these weighted sums, which means we get a continuous process as n goes to infinity. This is important because it lets us study the asymptotic behavior of random permutations that come from Chinese restaurant processes with specific seating arrangements, like alpha theta seating.

Emmy: And the core idea they are using is defining this weighted sum Sn as X sub n j equals x sub one a j Dn,j, and they're imposing some structural constraint on these weights to make the theorem work properly <ref:2412.02160#pg1>.

Kolya: Can you tell us what that structural constraint actually looks like in simple terms? I’m trying to get a feel for what conditions are necessary for this kind of limit theorem to exist.

Ada: Well, they impose a condition on the differences between consecutive weights, saying that the difference between the weight at index i plus j and the weight at index i is bounded by C sub j beta. The crucial part is that this parameter beta has to be less than or equal to alpha squared over two, where alpha is related to how fast those sampling frequencies decay.

Emmy: That constraint on beta being less than or equal to alpha squared over two seems very specific, linking the weight structure directly back into the parameter defining the urn model itself. It suggests that if you want this nice Gaussian process limit, your weights can't be too erratic in a certain way.

Kolya: So, what is the actual result they get from setting up this functional central limit theorem for these weighted sums? What does that resulting limit process look like?

Ada: They establish a new family of Gaussian processes in the limit. They identify this limit process as Z sub alpha of t, which is defined as a linear combination involving something called the Gaussian random measure and some other terms related to the weights. It’s essentially a complicated process, but it’s characterized by these coefficients defined by those weights a sub j.

Title and authors: Emmy: That characterization is interesting because they show that this limit process Z sub alpha is actually a linear combination of several underlying components, which gives us more structure to analyze than just the final Gaussian object. They also proved that under certain conditions on those weights, this resulting process has a version that is Hölder continuous for every exponent less than alpha over two.

Kolya: So we have this continuous process Z sub alpha with some regularity properties, and the paper shows it converges from a Poissonized model, which is essentially looking at the counts in a slightly different way. What does that convergence actually prove about the system?

Ada: Theorem three point one proves that the Poissonized model converges to this limit process Z sub alpha as n goes to infinity. They do this by first showing convergence for finite dimensional cases and then proving tightness for the whole path of the process. This is a big step because it connects discrete counting processes, which are hard to handle directly, to a continuous Gaussian object we can analyze.

Emmy: And that convergence hinges on a de-Poissonization step in section four, where they show that the difference between the two processes—the Poissonized model and the actual weighted occupancy process—becomes negligible in an appropriate sense. They propose replacing some terms with something called An to simplify this analysis, suggesting it should be related to how fast those frequencies grow or decay.

Kolya: I'm a bit stuck on what this convergence means for the applications they mention, specifically with random permutation matrices from Chinese restaurant processes. What’s the payoff here?

Ada: The motivation for this work is really tied to random matrix theory. By having this functional central limit theorem established, they can now investigate the asymptotic behaviors of random permutation matrices that arise from Chinese restaurant processes with alpha theta seating. Theorem one point five shows a convergence result where the difference between the actual weighted count and its expected value falls with the square root of n times some factor involving alpha over two, all almost surely.

Emmy: That convergence is stated as S sub floor n t of f minus E S sub floor n t of f falling with that specific rate, which relates directly to the limit process Z sub alpha. It shows that for these permutation matrices, the deviation from the average behavior is controlled by this new continuous Gaussian object.

Kolya: That sounds like a very concrete statement about tail behavior. So, what are some of those local results they use to support this big functional convergence? Are there any specific lemmas that give us finer control over the underlying probabilities?

Title and authors: Ada: They have a lemma, four point four, which provides a refined control on the local central limit theorem for both Poisson and binomial distributions. It gives an upper bound on the difference between the actual probability of getting k balls in n times p samplings and the probability of getting k balls in n trials with success probability p.

Emmy: That bound is pretty sharp, showing that this difference falls with something like the square root of n times p logarithm to two plus a logarithmic term involving four over n times p, which helps them control the error terms they need for tightness and convergence.

Kolya: And what about the conditions on those weights again? I noticed they emphasize that for this whole structure to hold, you need that condition on beta being less than alpha squared over two. Is there anything else they point out as a major hurdle when applying this result to different kinds of weighted sums?

Ada: They stress that the condition ai plus j minus ai less than or equal to C sub j beta for some beta in zero to alpha squared over two is absolutely crucial because it’s what allows them to establish the moment inequality on the increments, which is a necessary step for proving the main functional central limit theorem.

Emmy: And they mention that when you set beta equal to zero, it just means the weights are bounded, which covers simpler cases like the standard occupancy process or even just counting odd-occupancy processes. But when you move beyond that, you need that strict bound on beta related to alpha squared over two.

Kolya: So to summarize what we’ve heard about this paper, they've taken a weighted sum of counts in an urn scheme and shown it converges to a specific Gaussian process Z sub alpha under certain conditions on the weights. This convergence is then used to study how random permutations from Chinese restaurant processes behave asymptotically.

Ada: Exactly. The key finding is that we can analyze these complex counting statistics using this new continuous Gaussian object, which simplifies the problem significantly for researchers in random matrix theory and related fields.

Emmy: And the main takeaway is that the structure of the weights, specifically how much they vary across different indices j, dictates whether you can achieve this functional convergence to Z sub alpha.

Kolya: It seems like a solid piece of work because it connects discrete counting problems with continuous processes in a way that’s directly applicable to analyzing those random permutations we care about in the Chinese restaurant context.

Ada: It is definitely a solid piece of work, providing the necessary machinery to look at those permutation matrices in new ways.

Emmy: This functional central limit theorem for weighted occupancy processes of the Karlin model gives us a powerful tool for understanding these specific random structures.

The paper's summary: Ada: So, to summarize this paper, they've taken these weighted sums of counts in the Karlin model and shown that they converge to a specific continuous Gaussian process called Z alpha under certain conditions on those weights.

Emmy: That convergence is really neat because it gives us a way to analyze these discrete counting problems by looking at this smooth, continuous object.

Ada: Exactly, and they establish that this limit process Z alpha has properties like Hölder continuity for every exponent less than half of alpha, which means we have some regularity there.

Emmy: And the structure of that limit process is actually a linear combination of several underlying components defined by those weights themselves, so it’s not just some random blob.

Ada: That structure lets them prove convergence from a Poissonized model, meaning they can take discrete counting and turn it into something continuous as n gets huge.

Emmy: And the whole point of this is that you can use this continuous object to study the asymptotic behavior of random permutation matrices from Chinese restaurant processes with alpha theta seating.

Ada: So, for someone listening who isn't deep in math, this means we can now look at how those permutations behave in a very precise way as the size of the system grows.

Emmy: And this connection is made through Kingman’s representation theorem and studying their eigenvalue statistics to link it all together.

Ada: It really shows that these complicated counting scenarios aren't just random noise; they follow a predictable, continuous mathematical structure in the limit.

Emmy: And that structure is dictated by the specific way those weights are arranged across different indices j.

Ada: So, if you’re interested in how these weighted sums behave when you change those weights—say, if you make them grow faster—that’s exactly what this paper tells us about the resulting limit process.

Emmy: It sets up a framework where we can investigate the scaling laws for these permutations that arise from those restaurant seating arrangements.

Ada: We need to look at how they handle those constraints on beta, because that condition on the weight differences is really what makes this functional central limit theorem possible in the first place.

The paper's improvements: Kolya: So, after we saw how they set up the main theorem for weighted occupancy processes of the Karlin model, what are some of these improvements or tweaks they suggest for future work?

Ada: They really focus on making that functional central limit theorem even more robust by tightening those conditions on the weights.

Emmy: They suggest that when you move beyond just the standard odd-occupancy processes and introduce these general weights, you need to be very careful with the growth of those weight differences.

Ada: Specifically, they point out that for the theorem to hold in its strongest form, you need that difference between consecutive weights to be bounded by a constant C times j beta, where beta has to be strictly less than alpha squared over two.

Emmy: That's a refinement because it forces the weight distribution to be smoother in a certain way relative to the parameter alpha of the underlying process.

Kolya: From a probabilistic standpoint, that restriction on beta seems crucial for ensuring those moment inequalities they use earlier still hold up when you consider large n.

Ada: Exactly, because without that strict bound on beta less than alpha squared over two, you can't guarantee the necessary moment inequality on the increments of those sums.

Emmy: And in terms of structure, they suggest that replacing some of the standard terms in their limit process definition with something called An might make the de-Poissonization step cleaner.

Kolya: That makes sense because simplifying those error terms is always a good goal when you're trying to prove convergence to a limit process.

Ada: So, essentially, they are suggesting that if you want to apply this framework to more complex weighting schemes, like the ones from Chinese restaurant processes with alpha theta seating, you need to ensure your weights obey these tighter structural constraints.

Emmy: It’s about ensuring the algebraic structure of those weights is compatible with the analytic machinery of the functional limit theorem.

Kolya: And for someone who only listens to this show, it means that while the main result gives us a tool for those permutations, getting that tool to work in more complicated scenarios requires you to impose these specific structural rules on your input data.

Conclusion: Ada: So we’ve covered how to establish a functional central limit theorem for weighted occupancy processes of the Karlin model, showing convergence to this Gaussian process Z alpha under specific weight conditions.

Emmy: It really shows that these discrete counting problems have a very structured, continuous mathematical description when you look at them in the large scale limit.

Kolya: I think what this means for us is that we gain a new way to analyze the statistics of random permutations arising from Chinese restaurant processes with alpha theta seating by linking them directly to this continuous object.

Ada: Exactly, and it gives us a precise tool to investigate those asymptotic behaviors in random matrix theory.

Emmy: The paper also lays out the conditions on those weights that must be met for this convergence to happen, which is important because it tells us exactly what kind of weight distributions you can use.

Kolya: From a probabilistic view, those constraints on beta being less than alpha squared over two are essential because they ensure the moment inequalities used in the proof still work even when n gets big.

Ada: That’s right, and if you want to apply this to more complex weighting schemes, you have to make sure they respect that bound.

Emmy: And for someone just listening, it means we can take these complicated counting scenarios and map them onto a predictable continuous process Z alpha instead of just dealing with the raw counts.

Kolya: It changes how we look at the tail behavior of those permutation matrices because now we have this Gaussian object that governs the deviation from the average.

Ada: So, to wrap up, this paper establishes that functional central limit theorem for weighted occupancy processes of the Karlin model, providing a powerful framework for studying random structures in probability.

Emmy: It sets up a clear connection between discrete counting and continuous Gaussian processes in a way that’s directly useful for applications.

Kolya: It’s solid work because it gives us the machinery to analyze those permutation matrices we care about in the Chinese restaurant context with high precision.

JAIME GARZA, YIZAO WANG

University of Cincinnati

math.PR

Submitted: 2024-12-03

Updated: 2026-11-10

Comments: 32 pages; fixed a mistake in the previous version in the estimate in (4.11) in Proposition 4.1. The statements of Proposition 4.1 and all other results remain unaffected

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 79/100

The gist: A functional central limit theorem is established for weighted occupancy processes of the Karlin model…

Key concepts

Weighted Occupancy Processes
These are statistics derived from the Karlin model that count how many times certain items are occupied, weighted by a sequence of prescribed weights $a_j$. The core idea is to analyze these counts when the weights have controlled growth conditions.
Functional Central Limit Theorem (FCLT)
An FCLT describes the limiting behavior of a sequence of stochastic processes. In this context, it shows that the normalized weighted occupancy process converges in distribution to a continuous Gaussian process, $Z_\alpha(t)$, which is useful for understanding long-term trends.
$Z_\alpha(t)$ Limit Process
This is the limiting Gaussian process that arises from the FCLT. It is defined as a linear combination of stochastic integrals with respect to Gaussian random measures. Its properties, such as $\gamma$-Hölder continuity, are key to characterizing the asymptotic behavior of the original counting process.
Chinese Restaurant Processes
These are probabilistic models used in random matrix theory. They describe how items are assigned to groups (or tables) based on specific seating rules, which is used here to generate random permutation matrices whose asymptotic properties are being investigated.

Terminology

Summary

A functional central limit theorem is established for weighted occupancy processes of the Karlin model, which is important because it provides a functional central limit theorem for statistics in this model that can be used to investigate asymptotic behaviors of random permutation matrices induced by Chinese restaurant processes with (α, θ)-seating

Main Result and Setup

The main result is the establishment of a functional central limit theorem for the weighted sum of occupancy counts (Dn,j)j∈N with prescribed weights a = (aj)j∈N defined as Sn ≡ Sn, a:= Xn j=1 ajDn,j The statistics in this form include the (odd-)occupancy processes in (1.1) and hence we refer to Sn as weighted occupancy processes For the general weights (aj)j∈N we impose ai+j − ai ≤ Cjβ for some β ∈ [0, α2/2) and all i, j ∈ N ∪ 0 Our main result is a functional central limit theorem for the counting process (S⌊nt⌋)t∈[0,1] appropriately normalized, and we identify a new family of Gaussian processes in the limit

Limit Process Characterization

The limit process that will arise in our limit theorem is denoted by Zα(t) ≡ Zα, a(t):= Z R+×omega′ aN′(rt) − E′ aN′(rt) Mα(dr, dω ′), t ≥ 0 The process Zα is in fact a linear combination of the above with coefficients (aj)j∈N We refer to [33] for background on stochastic integrals with respect to Gaussian random measures

Properties of the Limit Process

Proposition 2.1 states that under the assumption (2. p. 23) ai+j − ai ≤ Cjβ for all i, j ∈ N with β ∈ h 0,α 2, then (Zα(t))t≥0 forms a stochastic process with a version that is γ-Hölder-continuous for every γ ∈ (0, α/2) Remark 1. p. 63 notes that the process Zα is in fact a linear combination of the above with coefficients (aj)j∈N

Poissonized Model Convergence

Theorem 3.1 proves the convergence of the Poissonized model, stating that Wfn(t) σn! p. 3 t∈[0,1] ⇒ (Zα(t))t∈[0,1], as n → ∞ This is achieved by proving finite-dimensional convergence in Section 3. p. 31 and tightness in Section 3. p. 92

De-Poissonization Step

Section 4 completes the proof of Theorem 1.1 by showing that the difference of the two processes is negligible in an appropriate sense Proposition 4. p. 43 suggests a natural candidate to replace An (instead of An,ǫ used in (4. p. 413)) as An:= (sup t∈[0,1] n 1/2 λn(t) − t ≤ log n)

Application in Random Matrix Theory

The main motivation for this work came from random matrix theory: using the general functional central limit theorem established here we shall investigate asymptotic behaviors of random permutation matrices induced by the Chinese restaurant processes with (α, θ)-seating Theorem 1.5 shows that S⌊nt⌋(f) − E S⌊nt⌋(f) P nα/2! p. 5 t∈[0,1] a.s.w. → sα,θ Γ(1 − α) 1/2 (Zα(t))t∈[0,1] with respect to P in D([0, 1])

Local Central Limit Theorems

Lemma 4.4 provides a refined control on the local central limit theorem for Poisson and binomial distributions This lemma shows that max k∈Jnp P(N(np) = k) − P(Bn,p = k) ≤ C√np p log2 n + log4 n np, for all n ∈ N, p ∈ [n−ǫ, 1/2] The proof of (4. p. 1 p. 116) and (4.17) relies on Stirling’s formula and concentration inequalities

Tightness Argument

The tightness argument involves showing that both Wcn(t) σn! t∈[0,1] and Wfn(t) − Wcn(t) σn! p. 9 t∈[0,1] are tight in D([0, 1]) Proposition 3. p. 36 establishes conditions under which the process σ−1 n Wcn is tight in D([0, 1])

De-Poissonization Convergence

The final step involves showing that the difference of the two processes is negligible in an appropriate sense Proposition 4.1 shows that sup t∈[n−(1−ǫ), 1] 1σn (qel(nλn(t)) − ql(⌊nt⌋)) → 0 in probability as n → This is achieved by analyzing In,1 and In,2 terms

Local Central Limit Theorems for Poisson and Binomial Distributions

Lemma 4.4 provides a refined control on the local central limit theorem for Poisson and binomial distributions This lemma shows that max k∈Jnp P(N(np) = k) − P(Bn,p = k) ≤ C√np p log2 n + log4 n np, for all n ∈ N, p ∈ [n−ǫ, 1/2] The proof of (4.16) and (4.17) relies on Stirling’s formula and concentration inequalities

The Role of Conditions

The constraints on the weights are crucial, as the condition ai+j − ai ≤ Cjβ for some β ∈ [0, α/2) is needed to establish the moment inequality on the increments in (3.9) The constraint β < 1 − α/2 is equivalent to the condition β < 1 - α/2 < alpha/2, which is a stronger condition than initially imposed

The constraints on the weights are crucial, as the condition ai+j − ai ≤ Cjβ for some β ∈ [0, α/2) is needed to establish the moment inequality on the increments in (3.9) The constraint β < 1 - α/2 is equivalent to the condition β < 1 - α/2 < alpha/2, which is a stronger condition than initially imposed

The constraints on the weights are crucial, as the condition ai+j − ai ≤ Cjβ for some β ∈ [0, α/2) is needed to establish the moment inequality on the increments in (3.

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

  1. Improved AI can perform functional central limit theorems for weighted occupancy processes of Karlin models by establishing a functional central limit theorem for any sequence of real numbers (aj) with specific growth conditions on its increments, namely ai+j - ai <= Cj beta.

  2. The improved system can analyze asymptotic behaviors of random permutation matrices induced by Chinese restaurant processes with alpha, theta seating by linking them to the weighted occupancy counts via Kingman’s representation theorem and studying their eigenvalue statistics.

  3. The AI can determine if a sequence of weights (aj) satisfies the conditions for establishing a functional central limit theorem, specifically checking if ai+j - ai <= Cj beta for some beta in [0, alpha 2/2).

Sources

Related papers