Alternating binomial sums weighted by logarithms — expressions of the form involving (-1)^k times binomial coefficients times logarithmic terms — arise in the analysis of collecting processes. An inequality bounding such sums was conjectured based on numerical evidence and the structure of minimum-order statistics.
Note (2026-11-12): checked against the paper's full text (Doumas). The proof matches: the combination of alternating binomial logarithmic sums equals the variance of the logarithm of the maximum of n independent exponential variables, which is positive, so the inequality follows. The paper says the inequality was introduced in the author's earlier work on the minimum of independent collecting processes (a variant of the coupon collector problem), where it appears in the asymptotic variance; I did not find a statement that it was conjectured from numerical evidence. What the paper does call a conjecture, and says remains open, is a different statement from the earlier work: that the sequence pi^2/6 + n w_n - n^2 c_n^2 is decreasing in n. The paper also notes the limit as n grows (the variance tends to zero), which the essay does not mention.
The proof exploits a probabilistic identity (arXiv:2603.11129). The alternating binomial logarithmic sum equals the variance of the logarithm of the maximum of n independent exponential random variables. Since variance is nonnegative, the inequality follows immediately.
The method is striking because the algebraic object — a complicated alternating sum — has no obvious sign. The terms oscillate, and direct estimation requires careful cancellation tracking. But the probabilistic representation converts the sum into a variance, which is nonnegative by definition. The inequality is not proved by bounding; it is proved by reinterpreting.
This is a specific instance of a broader technique: translating combinatorial identities into probabilistic statements where structural properties (nonnegativity, monotonicity, convexity) are automatic. The alternating sum looked like it needed estimation. It needed recognition.