Random Additions in Urns of Integers

Mackenzie Simper

REVIEW 4 major objections 5 minor 22 references

Random Additions in Urns of Integers

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read An urn that adds the sum of two random draws converges, after rescaling, to an exponential whose random scale is the martingale limit of the mean.

desk verdict A novel infinite-type, multi-draw urn limit with a clean exponential law, but Theorem 1.1 overclaims for zero-sum initial configurations and the nontriviality proof has a gap. read the letter →

arxiv 1908.01370 v2 pith:XDU766YC submitted 2019-08-04 math.PR

Mackenzie Simper This is my paper · ORCID

classification math.PR MSC 60F0560G4260C05

keywords urnmodelsrandomadditionsintegerlabelsexponentiallimitlawcontractionmethodrecursivedistributionalequationsmartingaleconvergencemulti-drawurns

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 analyzes the $\mathbb{Z}$-urn: balls carry integer labels, and at each step two balls are drawn independently with replacement and a new ball labeled by their sum is added. The main result, Theorem 1.1, asserts that for any starting configuration, if $Z_n$ is a uniformly random draw from the urn after $n$ balls, then $A=\lim_{n\to\infty}\mathbb{E}[Z_n\mid \mu_n]/(n+1)$ exists almost surely, $P(A\neq 0)>0$, and the conditional law of $Z_n/n$ converges to that of $A$ times an $\mathrm{Exp}(1)$ variable. The proof splits the claim in two: a martingale argument shows the rescaled mean converges, and a contraction-method argument shows the rescaled added ball converges to a Gamma distribution, which a uniform index then turns into an exponential. This matters because the model combines drawing multiple balls and having infinitely many ball types, and the limit law is simple enough to use.

What carries the argument

The argument is carried by two mechanisms. First, the rescaled total $A_n=S_n/(n(n+1))$ is a martingale: adding a ball whose conditional mean is $2S_n/n$ leaves $A_n$ unchanged in expectation, and its second moment is bounded, so $A_n\to A$ almost surely. Second, the scaled added ball $\tilde{X}_n=X_n/n$ obeys the recursive distributional equation $\tilde{X}_n\stackrel{d}{=}(I_n^1/n)\tilde{X}_{I_n^1}+(I_n^2/n)\tilde{X}_{I_n^2}$, whose limit must satisfy $X\stackrel{d}{=}U_1X_1+U_2X_2$ with $U_i$ uniform. The unique solution with mean $\mu$ is $G\cdot\mu$ with $G\sim\Gamma(2,1)$ (Lemma 3.3); a uniformly chosen past index introduces an extra independent uniform factor $U$, and $U\cdot G$ is exponential (Lemma 3.2). The contraction method in the Wasserstein $L^2$ metric (a notion of distance between probability laws that also tracks second moments) turns the recursion into a contraction and forces the convergence.

What would settle it

The all-zero urn is a direct counterexample to the theorem as stated: with initial labels $\{0,0\}$, every added ball has label $0$, so $A=0$ almost surely and $P(A\neq 0)=0$. For the $\{-1,1\}$ configuration, the proof's lower bound for $\mathbb{E}[A_n^2]$ is proportional to the initial mean, which is zero, so the argument that $A$ is nontrivial does not apply; simulating many runs and checking whether the empirical distribution of $A$ has positive mass away from $0$ would settle whether the theorem's conclusion still holds there.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is the exponential limit law of Theorem 1.1. For the integer urn with any initial configuration, the martingale $A_n = S_n/(n(n+1))$ converges almost surely to a random variable $A$, and the conditional law of $Z_n/n$ given the urn history converges to $\mathcal{L}(A\cdot \mathrm{Exp}(1))$. In particular, for the starting configuration $\{-1,1\}$, the limiting urn is supported on one side of zero, positive or negative according to the sign of $A$, which explains the one-sided exponential histograms the simulations show. The vector version, Theorem 1.2, says that the $n$-th added ball $X_n$, rescaled by $n$, converges to $G\cdot A$ with $G\sim\Gamma(2,1)$; since a random draw is a uniform past ball, the factor $G$ is multiplied by an independent uniform variable and becomes exponential.

Load-bearing premise

The theorem's nontriviality assertion depends on the initial urn not being the all-zero urn—if every starting label is $0$, the limit $A$ is identically $0$—and the proof's second-moment lower bound does not cover configurations whose initial labels sum to $0$, such as $\{-1,1\}$.

Editorial extensions

If this is right

For the $\{-1,1\}$ start, the limiting draw is exponential on $(0,\infty)$ or $(-\infty,0)$ according to the sign of $A$; the urn's labels become almost all positive or almost all negative.
For vector-valued labels in $\mathbb{Z}^d$, every coordinate of the rescaled added ball converges to the same $\Gamma(2,1)$ factor multiplied by the coordinate's martingale limit, so the coordinates are asymptotically perfectly dependent.
The annealed limit of $Z_n/n$ is a mixture of exponentials with random scale $A$, and the mean of $A$ is determined by the initial configuration as $S_{\tau_0}/(\tau_0(\tau_0+1))$.
For the $k$-draw variant, the limit satisfies $X\stackrel{d}{=}\sum_{i=1}^k U_iX_i$; a Gamma law no longer solves this identity, so the exponential shape observed at $k=2$ should be replaced by a family of limit laws indexed by $k$.

Reading between the lines

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

The identity $X\stackrel{d}{=}U_1X_1+U_2X_2$ encodes the entire dependence on the initial state in the single random vector $A$; a reading the paper leaves implicit is that the two-draw addition rule has exactly one stable shape, a ray of scaled exponentials, and the initial labels only choose the ray and the scale.
For the signaling-game motivation discussed in the paper, the one-sided exponential limit means the long-run distribution of strategies is sign-selected: each realization ends up predominantly positive or predominantly negative, and the random scale $A$ acts as the strength of the convention, so independent runs produce different conventions.
A testable consequence of the vector theorem: with labels in $\mathbb{Z}^d$, rescaled draws should collapse onto the random ray $\{tA:t\ge 0\}$ (modulated by an exponential), so an empirical scatter plot of rescaled draws should look one-dimensional rather than filling the ambient space.
For $k>2$, the fixed-point equation loses its explicit Gamma solution, so the limiting law becomes a genuinely new family; simulating the $k=3$ and $k=4$ urns and comparing histograms with solutions of $\sum_{i=1}^k U_iX_i$ would show how quickly the exponential shape deforms.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

Here is my read. The paper gives the first limiting result for an urn that combines multiple draws per step with an infinite type space: if you draw two integers with replacement and add their sum, the empirical measure, normalized, converges to an exponential with a random scale. That is a real contribution. The martingale argument for the mean is straightforward and correct, the contraction method is standard but carefully applied, and the paper is honest that the distribution of the scale A is not identified. The related-work section is accurate; the model really does seem new, and the signal-game motivation is contextual rather than load-bearing.

The soft spot is exactly where the reader put it. Theorem 1.1 is stated for “any initial configuration.” The all-zero urn is a counterexample: every ball is 0, A=0 a.s., so P(A≠0)>0 fails. That is not an exotic corner case; it is the most degenerate possible start. Lemma 2.1, the only place nontriviality is proved, gives a lower bound for E[A_n^2] proportional to A_{tau0}, the initial value of the martingale. When S_{tau0}=0, including the paper's own {-1,1} configuration, that bound is zero and proves nothing. The bound should involve A_{tau0}^2, or the proof needs a different argument. For {-1,1} the claim is likely true, and the simulations suggest it, but the paper as written does not prove it. This is a fixable gap, not a collapse of the main idea.

I also find Lemma 2.1's statement dimensionally odd: it claims an infimum over expectations greater than zero using a product times A_{tau0}, which is a random variable, not a positive constant. Even when S_{tau0}>0, the displayed inequality is at least poorly phrased. A referee should ask for a clean statement and a repair for the zero-sum case.

Bottom line: the exponential limit mechanism is sound, the paper deserves serious refereeing, and with a small amendment to the theorem and a corrected nontriviality proof it becomes a solid contribution. I would send it out.

Referee Report

4 major / 5 minor

Summary. The paper studies an urn process in which, at each step, two balls are drawn independently and with replacement and a new ball labeled by their sum is added. The main result, Theorem 1.1, asserts that for any initial configuration the rescaled empirical mean A_n = S_n/(n(n+1)) converges almost surely to a limit A, that P(A ≠ 0) > 0, and that a random draw Z_n from the urn satisfies the conditional limit law Z_n/n → A·Exp(1). The proof uses an L2-bounded martingale argument for the mean and the contraction method for recursive distributional equations, with a vector-valued generalization in Theorem 1.2.

Significance. If the main theorem is established for non-degenerate initial configurations, the paper provides an elegant rigorous analysis of a genuinely novel urn model that combines multiple draws with an infinite type space. The martingale identification of the random scale A and the reduction to a fixed-point equation are clean and convincing, and the use of known contraction-method results is appropriate. The vector-valued extension is a useful addition. The main obstruction is that Theorem 1.1 is false as stated for an all-zero initial urn, and the proof of the nontriviality assertion has a gap for zero-sum initial configurations, including the paper's featured {-1,1} example.

major comments (4)

[Section 1.1, Theorem 1.1] The theorem is stated for 'any initial configuration', but if U0 consists only of balls labeled 0, then every drawn ball and every added ball is 0, so μ_n = δ_0 and A = 0 almost surely. The assertion P(A ≠ 0) > 0 is then false. The statement must either exclude all-zero initial configurations or be modified to a form that remains true for them.
[Section 2, Lemma 2.1] The lower-bound display in the proof of Lemma 2.1 reads E[A_n^2] ≥ ∏_{k=τ0}^{n-1}(1 - 2/(k^2+4k+4)) · A_{τ0}. As written this is dimensionally inconsistent: the right-hand side is nonpositive when A_{τ0} < 0 and is zero whenever S_{τ0} = 0. Even with the likely intended factor A_{τ0}^2, the argument gives no positive lower bound for zero-sum initial configurations. In particular, for the paper's featured initial configuration {-1,1}, S_{τ0}=0, so Lemma 2.1 does not establish the nontriviality assertion P(A ≠ 0) > 0 that is needed for Theorem 1.1. A separate argument is required for zero-sum starts.
[Section 3, Lemma 3.3] Lemma 3.3 assumes X(i) ≥ 0 almost surely for all coordinates, but in the application the vector A can have negative coordinates (as in the {-1,1} initial configuration, where A can be negative). The characteristic-function computation actually extends to signed A, but the lemma as stated does not cover the case used in the proof of Theorems 1.1 and 1.2. The paper should state and prove the signed version, or explicitly condition on the sign of A.
[Section 3.3, proof of Theorems 1.1 and 1.2] The proof divides by A in the step 'Since \tilde{X}_{I_n}/A converges in L2 to X ∼ Gamma(2,1)'. This presupposes A ≠ 0, but P(A ≠ 0) > 0 is exactly one of the assertions being proved. The proof should first establish nontriviality of A, or should handle the event {A = 0} separately, before dividing by A.

minor comments (5)

[Section 1.1, Remark] There is a typo in the remark: 'the the nth labeled added' should read 'the nth label added'.
[Section 2, Lemma 2.1] The notation A_τ0 is used ambiguously: it should be made explicit that the lower bound concerns A_τ0(i)^2 for each coordinate i, and the displayed inequality should be corrected to include the square on A_τ0.
[Section 1.1 and Section 2] In Theorem 1.1, A is defined as lim E[Z_n|μ_n]/(n+1), while Section 2 defines A_n = S_n/(n(n+1)). Since E[Z_n|μ_n] = S_n/n, these agree, but the paper would be easier to follow if this identity were stated explicitly before the theorem.
[Section 3.3] The phrase 'converges in L2' for \tilde{X}_{I_n}/A should be 'converges in the Wasserstein L2 metric' or 'converges in distribution with convergence of second moments', since the contraction method gives distributional convergence, not almost-sure convergence.
[Section 3.1, proof of Lemma 3.3] The proof states that the integral equation for the characteristic function has a unique solution, but no justification or reference is given for this uniqueness. A citation or a short argument would be helpful.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the exponential limit law is derived from the process recursion and external characterizations, not from fitted inputs or self-citation.

full rationale

The derivation is self-contained with respect to the central claim. The limiting scale A is defined as the almost sure martingale limit of A_n = S_n/[n(n+1)] via Lemma 2.1, and it is not a fitted parameter. The recursive distributional equation (4) and its scaled form (5) are derived directly from the urn update rule, not assumed. Theorem 3.4, whose proof is included, shows convergence to the fixed point X = U_1 X_1 + U_2 X_2, and Lemmas 3.2 and 3.3 characterize the exponential and gamma limits using external results [KS88] and [Nei01] along with self-contained characteristic-function arguments. No prediction reduces to an input by construction, and there is no load-bearing self-citation chain. The all-zero initial configuration makes the statement P(A neq 0) > 0 false for that case, and the lower bound in Lemma 2.1 vanishes when S_{tau0}=0, but this is a correctness gap in the theorem as stated rather than a circularity in the derivation mechanism.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters are fit to data; the only inputs are the initial labels, which are part of the model. The argument relies on standard martingale and contraction-method theorems and on distributional characterizations from the cited literature. The main hidden assumptions are nondegeneracy of the starting configuration and asymptotic independence in the contraction limit.

assumptions (3)

standard math The contraction method theorem (special case of Theorems 4.1 and 4.2 in [Nei01]) is valid and applicable in the quenched setting.
Used in Section 3.2 to prove convergence of tilde X_n to the fixed point; if this theorem's hypotheses fail, the central limit law does not follow.
standard math The exponential and gamma distributional characterizations in Lemma 3.2 and Lemma 3.3 are correct, including the uniqueness of the solution to the characteristic-function equation.
Used to identify the limit as exponential or gamma; uniqueness is asserted without a full proof in the vector case.
domain assumption The recursive representation X_n = X_{I^1_n} + X_{I^2_n} with I_n uniform on {1,...,n-1} exactly models the urn draws with replacement.
Assumed in Equation (4); it matches the model of drawing two balls with replacement from the current urn.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Random Additions in Urns of Integers." pith.science (2026). https://pith.science/paper/XDU766YC

@misc{pith2026190801370,
  author       = {Pith},
  title        = {Pith review of: Random Additions in Urns of Integers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XDU766YC}},
  note         = {Machine review of arXiv:1908.01370}
}

read the original abstract

Consider an urn containing balls labeled with integer values. Define a discrete-time random process by drawing two balls, one at a time and with replacement, and noting the labels. Add a new ball labeled with the sum of the two drawn labels. This model was introduced by Siegmund and Yakir ({\it Ann. Probab. 33(5), 2005}) for labels taking values in a finite group, in which case the distribution defined by the urn converges to the uniform distribution on the group. For the urn of integers, the main result of this paper is an exponential limit law. The mean of the exponential is a random variable with distribution depending on the starting configuration. This is a novel urn model which combines multi-drawing and an infinite type of balls. The proof of convergence uses the contraction method for recursive distributional equations.

Figures

Figures reproduced from arXiv: 1908.01370 by the authors.

**Figure 2.** Results for A from simulations of the Z-urn model from two different initial configurations. An urn realization was run for 5, 000 rounds and the value of An was recorded. A total of 5, 000 realizations of each urn process were run to display the empirical distribution of A. 4 Conclusion and future questions This paper introduced a new urn model, extending the product urn model from [SY05] for finite groups to the i… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 21 canonical work pages

[1]

Abrams, H

A. Abrams, H. Landau, Z. Landau, J. Pommersheim, and E. Zaslow. Random multiplication approaches uniform measure in finite groups. J. Theoret. Probab. , 20(1):107--118, 2007

work page 2007
[2]

Bickel and David A

Peter J. Bickel and David A. Freedman. Some asymptotic theory for the bootstrap. Ann. Statist. , 9(6):1196--1217, 1981

work page 1981
[3]

MacQueen

David Blackwell and James B. MacQueen. Ferguson distributions via P \' o lya urn schemes. Ann. Statist. , 1:353--355, 1973

work page 1973
[4]

P\' o lya urn schemes with infinitely many colors

Antar Bandyopadhyay and Debleena Thacker. P\' o lya urn schemes with infinitely many colors. Bernoulli , 23(4B):3243--3267, 2017

work page 2017
[5]

Svante Janson . A.s. convergence for infinite colour P\' o lya urns associated with random walks . arXiv e-prints , page arXiv:1803.04207, Mar 2018

work page Pith review arXiv 2018
[6]

Johnson and Samuel Kotz

Norman L. Johnson and Samuel Kotz. Urn models and their application . John Wiley & Sons, New York-London-Sydney, 1977. An approach to modern discrete probability theory, Wiley Series in Probability and Mathematical Statistics

work page 1977
[7]

Markus Kuba and Hosam M. Mahmoud. Two-color balanced affine urn models with multiple drawings. Adv. in Appl. Math. , 90:1--26, 2017

work page 2017
[8]

Analysis of a generalized F riedman's urn with multiple drawings

Markus Kuba, Hosam Mahmoud, and Alois Panholzer. Analysis of a generalized F riedman's urn with multiple drawings. Discrete Appl. Math. , 161(18):2968--2984, 2013

work page 2013

Show all 22 references

[9]

P\' o lya urns via the contraction method

Margarete Knape and Ralph Neininger. P\' o lya urns via the contraction method. Combin. Probab. Comput. , 23(6):1148--1186, 2014

2014
[10]

Kotz and F

S. Kotz and F. W. Steutel. Note on a characterization of exponential distributions. Statist. Probab. Lett. , 6(3):201--203, 1988

1988
[11]

Multiple drawing multi-colour urns by stochastic approximation

Nabil Lasmar, C\' e cile Mailler, and Olfa Selmi. Multiple drawing multi-colour urns by stochastic approximation. J. Appl. Probab. , 55(1):254--281, 2018

2018
[12]

Hosam M. Mahmoud. P\' o lya urn models . Texts in Statistical Science Series. CRC Press, Boca Raton, FL, 2009

2009
[13]

Inventing new signals

Jason McKenzie Alexander, Brian Skyrms, and Sandy Zabell. Inventing new signals. Dynamic Games and Applications , 2:129--145, 03 2012

2012
[14]

On a multivariate contraction method for random recursive structures with applications to Q uicksort

Ralph Neininger. On a multivariate contraction method for random recursive structures with applications to Q uicksort. Random Structures Algorithms , 19(3-4):498--524, 2001. Analysis of algorithms (Krynica Morska, 2000)

2001
[15]

A general limit theorem for recursive algorithms and combinatorial structures

Ralph Neininger and Ludger R\" u schendorf. A general limit theorem for recursive algorithms and combinatorial structures. Ann. Appl. Probab. , 14(1):378--418, 2004

2004
[16]

A survey of random processes with reinforcement

Robin Pemantle. A survey of random processes with reinforcement. Probab. Surv. , 4:1--79, 2007

2007
[17]

A limit theorem for `` Q uicksort''

Uwe R\" o sler. A limit theorem for `` Q uicksort''. RAIRO Inform. Th\' e or. Appl. , 25(1):85--100, 1991

1991
[18]

S. T. Rachev and L. R\" u schendorf. Probability metrics and recursive algorithms. Adv. in Appl. Probab. , 27(3):770--799, 1995

1995
[19]

S ignals: E volution, L earning, and I nformation

Brian Skyrms. S ignals: E volution, L earning, and I nformation . OUP Oxford, 2010

2010
[20]

Siegmund and B

D. Siegmund and B. Yakir. An urn model of D iaconis. Ann. Probab. , 33(5):2036--2042, 2005

2005
[21]

Infinite Color Urn Models

Debleena Thacker. Infinite Color Urn Models . PhD thesis, Indian Statistical Institute, Dehli Centre, 2015

2015
[22]

Probability with martingales

David Williams. Probability with martingales . Cambridge Mathematical Textbooks. Cambridge University Press, Cambridge, 1991

1991

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Abrams, H

A. Abrams, H. Landau, Z. Landau, J. Pommersheim, and E. Zaslow. Random multiplication approaches uniform measure in finite groups. J. Theoret. Probab. , 20(1):107--118, 2007

work page 2007

[2] [2]

Bickel and David A

Peter J. Bickel and David A. Freedman. Some asymptotic theory for the bootstrap. Ann. Statist. , 9(6):1196--1217, 1981

work page 1981

[3] [3]

MacQueen

David Blackwell and James B. MacQueen. Ferguson distributions via P \' o lya urn schemes. Ann. Statist. , 1:353--355, 1973

work page 1973

[4] [4]

P\' o lya urn schemes with infinitely many colors

Antar Bandyopadhyay and Debleena Thacker. P\' o lya urn schemes with infinitely many colors. Bernoulli , 23(4B):3243--3267, 2017

work page 2017

[5] [5]

Svante Janson . A.s. convergence for infinite colour P\' o lya urns associated with random walks . arXiv e-prints , page arXiv:1803.04207, Mar 2018

work page Pith review arXiv 2018

[6] [6]

Johnson and Samuel Kotz

Norman L. Johnson and Samuel Kotz. Urn models and their application . John Wiley & Sons, New York-London-Sydney, 1977. An approach to modern discrete probability theory, Wiley Series in Probability and Mathematical Statistics

work page 1977

[7] [7]

Markus Kuba and Hosam M. Mahmoud. Two-color balanced affine urn models with multiple drawings. Adv. in Appl. Math. , 90:1--26, 2017

work page 2017

[8] [8]

Analysis of a generalized F riedman's urn with multiple drawings

Markus Kuba, Hosam Mahmoud, and Alois Panholzer. Analysis of a generalized F riedman's urn with multiple drawings. Discrete Appl. Math. , 161(18):2968--2984, 2013

work page 2013