Terminal Defects, Growing Multiplicity, and Variance Extremality in the Double Dixie Cup Problem

Abstract: We develop a terminal-defect method for the double Dixie cup problem and use it to prove the finite-variance extremality conjecture of Doumas and Papanicolaou. For every (m\ge1) and (N\ge2), among all positive coupon probability vectors (p=(p_1,\ldots,p_N)), the variance of the time (T_m(N)) to collect (m) complete sets is uniquely minimized at the uniform vector. We prove the stronger radial statement that the variance is strictly increasing along every ray from the uniform vector. The proof is finite-(N) and exact: after Poissonization, the completion time is a maximum of independent Erlang variables, and the radial derivative of its distribution is compared to a size-biased law using a monotone-likelihood-ratio argument based on a log-scale monotonicity property of the Gamma reverse hazard. The same framework gives a growing-multiplicity Gumbel theorem in the equal-probability case, with expectation and variance asymptotics on the inverse gamma-tail scale. This recovers the fixed-(m) equal-probability variance asymptotic stated as Conjecture 1 by Doumas and Papanicolaou, classically known for (m=1), and extends the mechanism to (m=m_N). We also illustrate the unequal-probability theory with endpoint-Laplace limits for power-law probabilities.

• The paper establishes that the variance of Tₘ(N) is uniquely minimized at the uniform distribution over all strictly positive probabilities.
• It introduces a terminal-defect approach with Poissonization that decouples the coupon collection process into independent Erlang clocks, enabling exact finite-N analysis.
• The results extend classical coupon collector findings by providing detailed moment formulas, Gumbel limit laws, and practical insights for randomized sampling systems.

Terminal Defects, Variance Extremality, and Distributional Asymptotics in the Double Dixie Cup Problem

Problem Formulation and Background

The paper addresses the double Dixie cup (or $m$-set coupon collector) problem: for $N$ coupon types and $m\ge1$, what is the distribution and moments of the smallest number $T_m(N)$ of iid draws needed to collect at least $m$ of each type, where each type $j$ occurs with probability $p_j>0$? The classical uniform case ($p_j=1/N$) connects to results by Newman-Shepp and Erdős-Rényi, who established expectation and limit laws for $T_m(N)$, notably the Gumbel limit for the normalized completion time.

The present work addresses exact finite-$N$ properties, especially for non-uniform probability vectors $N$0, extending beyond previous asymptotic and expectation-centric approaches. The central open question resolved is the finite variance extremality conjecture (Doumas–Papanicolaou, 2016): Does the variance of $N$1 attain its minimum uniquely at the uniform probability vector among all strictly positive $N$2?

Terminal-Defect Framework and Main Techniques

A terminal-defect approach is developed, establishing that, upon Poissonization, the coupon collection process decouples into independent Erlang clocks for each type. The time to complete $N$3 copies for all types is the maximum of these clocks. For equal probabilities, the defect scheme gives an explicit binomial structure for the number of types still missing $N$4 copies after time $N$5, enabling uniform asymptotic and tail analysis using incomplete gamma function expansions.

Terminal defects refer to coupon types that have not yet reached the $N$6 threshold. In the Poissonized model, at time $N$7, the probability that type $N$8 is still defective is $N$9. The zero-defect probability (all sets completed) admits sharp asymptotic inversion, and explicit, exact integral representations for expectations, variances, and higher moments of $m\ge1$0 are obtained.

Key reusable analytic modules include:

• Terminal-defect theorems: If the average number of defects at a scaling window converges and no single defect dominates, the limiting law is Gumbel (or more generally, exponential of the limiting terminal profile).
• Gamma-tail asymptotics: The correct scaling and normalization follow by inverting the mean terminal defect function.
• Poisson clock de-Poissonization: Ensures discrete and continuous (Poissonized) times share limit laws when clock entropy is dominated by the fluctuation scale.
• Endpoint-Laplace methods: Handle regimes where the rarest coupons dominate, encompassing power-law (Zipf-type) probabilities.

Strongly Supported Results

Variance Extremality

For every $m\ge1$1 and $m\ge1$2, the variance of $m\ge1$3 is uniquely minimized (over all strictly positive $m\ge1$4) at the uniform distribution; the variance strictly increases along every simplex ray outward from uniform.

This is achieved via:

• An exact, finite-$m\ge1$5 analysis of the variance, combining explicit Poissonized rising moment identities (using inclusion-exclusion over coupon subsets) with differentiation under the simplex.
• A radial derivative argument: Moving from uniform in any direction, the derivative of the Poissonized completion distribution is shown (via MLR and monotonicity of the reverse Gamma hazard) to be “later” in a size-bias sense than the current law, forcing the variance to increase.
• Alternate local analysis via Hessian calculations confirms strict quadratic positivity at uniform, independent of pairwise smoothing counter-examples.

The result validates and generalizes the conjecture of Doumas and Papanicolaou, previously known only for $m\ge1$6 (Yu, 2017), to all $m\ge1$7, including growing $m\ge1$8.

Precise Asymptotics: Expectation, Variance, and Gumbel Law

The approach yields the following for fixed $m\ge1$9 and $T_m(N)$0 with $T_m(N)$1:

• $T_m(N)$2 (expectation),
• $T_m(N)$3 (variance).
• $T_m(N)$4, where $T_m(N)$5 is standard Gumbel and the centering/scaling parameters $T_m(N)$6 are determined via the inversion of the mean defect function.

For the general non-uniform case, two broad regimes (Cases I and II, in the sense of Doumas–Papanicolaou) are treated:

• Case I: When the coupon probability weights decrease slowly enough, the normalized completion time converges to the maximum of an infinite sequence of independent Gamma (Erlang) variables.
• Case II (Endpoint-dominated/Power-law): When rarest coupons dominate, terminal-defect/inverse-Laplace methods yield Gumbel limits under appropriate scaling for families such as $T_m(N)$7, with detailed power-law dependence on $T_m(N)$8.

Analytical Modules for Coupon Collector-Type Problems

Multiple technical tools for analyzing maxima of gamma and Erlang clocks under non-uniform rates are given, with modular proofs grounded in monotonicity of key hazard and reverse-hazard functions. The analysis covers both asymptotics (large $T_m(N)$9 and $m$0) and finite-$m$1 exactness, enabling extension to generalized coupon-collector scenarios.

Theoretical and Practical Implications

The strict variance minimality at uniform probabilities sharpens the understanding of randomness efficiency in repeated sampling systems—relevant to randomized algorithms, data streaming, hashing, and sequential discovery protocols. The result implies that, regardless of the desired coupon multiplicity $m$2, maximizing per-type sampling entropy (i.e., uniformity) achieves not only optimal expectation but also optimal concentration about the mean, strictly in both cases. The characterization of variance growth away from uniform may inform robust system design against source inhomogeneity.

On the theoretical side, the use of Poissonization, reverse-hazard monotonicity, and measure-derivative ordering establishes techniques applicable to other order-statistics and extreme-value tasks, especially where independence structures facilitate modular analysis. Open directions include extending endpoint-Laplace approaches to broader non-smooth classes, multidimensional or grouped coupon settings, and finer quantitative bounds on the variance gap as a function of probability vector deviation from uniformity.

Conclusion

This work comprehensively resolves the variance extremality conjecture for the double Dixie cup problem, employing a terminal-defect approach that is both modular and powerful. The joint analysis of moments, distributional limits, and the fine structure of the Poissonized process provides a unified perspective, illuminating both classical and novel regimes, symmetric and asymmetric, fixed and variable $m$3. The foundational analytic modules and positivity principles presented offer a transferable toolkit for broader classes of terminal-defect and extreme-value problems, with substantive and verifiable implications for discrete random processes and their applications.

Reference: “Terminal Defects, Growing Multiplicity, and Variance Extremality in the Double Dixie Cup Problem” (2604.25108).