bibkey: “ottolini2020oscillations” authors: “Andrea Ottolini” year: 2020 title: “Oscillations for order statistics of some discrete processes” doi: “10.1017/jpr.2020.25” url: “https://arxiv.org/abs/2010.09071v1” claim: “Discrete maxima can have oscillating asymptotic distributions, and ties affect their upper order statistics; the paper studies independent variables and specified allocation models.” strata_touched: [] license: “citation-only” triage: “anchor”

Oscillations for order statistics of some discrete processes

The inspected primary text is arXiv:2010.09071v1, dated 18 October 2020, 16 pages. The journal article appears in Journal of Applied Probability 57(3), 703–719 (2020), with the DOI above. Manuscript page numbers below refer to that pinned version.

Theorem 1.1, p. 2, studies the maximum of independent identically distributed discrete variables with an unbounded upper support. Its hypothesis is the existence of a limiting successive upper-tail ratio in the closed unit interval. The three displayed cases distinguish zero, strictly intermediate, and unit limiting ratios. Remark 1.2 interprets the first case through oscillating probabilities and subsequences. Theorem 1.3, pp. 2–3, addresses ties under the zero-tail-ratio hypothesis; it is not a general theorem for all discrete laws or all triangular arrays.

The allocation results in Theorems 1.5–1.7, p. 4, concern uniform multinomial allocations and symmetric Dirichlet mixtures under their stated fixed-ratio conditions. Equations (1.4)–(1.5), pp. 3–4, explain the conditional representation and probability ratio used to transfer extreme events from independent variables. This is a specific conditioning argument, not an assertion that dependence can always be removed by Poissonization.

Discrete extreme oscillations and the need to account for ties are literature-attested. In the arithmetic extension of PARITY_HIDDEN_ARROW_RECOVERY, there are two row classes with different compound-Poisson comparison laws, the Poisson time increases logarithmically, and the observed rows come from compensated stationary pairs or paths. The proof uses that model’s exact likelihood and fixed-number row comparison, followed by a posterior average over the observed boundary bin. No theorem from this paper is used to replace those steps. In particular, its zero-tail-ratio tie theorem does not give the reciprocal-binomial boundary factor or the two-class critical series.

The parity model’s analytic phase-dependent exponential rate, and its resulting nonconvergence of exact minimax risk along integer dimensions, are repository-derived consequences of that critical series. The bounded source search found no theorem here stating those conclusions. This is a source-scope comparison, not a global originality claim.

The manuscript cites C. W. Anderson’s 1970 paper, Extreme value theory for a class of discrete distributions with applications to some stochastic processes, DOI 10.2307/3212152, as prior work on discrete extremes. Its original full text was not inspected for this note; the primary source actually read is Ottolini’s pinned manuscript.