Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Mangerel 2022 additive functions short intervals gaps conjecture
Mangerel, Alexander P., Additive functions in short intervals, gaps and a conjecture of Erdős. Ramanujan J. 59 (2022), no. 4, 1023--1090, DOI 10.1007/s11139-022-00623-y. The held PDF is the arXiv preprint, version 1 of 27 August 2021, and the labels cited here are that version's. The arXiv record (https://arxiv.org/abs/2108.12351, read 2026-10-02) names the Creative Commons Attribution 4.0 license.
Mangerel develops analogs of the Matomäki-Radziwill theorem for additive functions, approximating the average of an additive function g over a typical short interval (n - h, n] by the corresponding long average. Theorem 1.1 gives, for any additive function g and any integer 10 <= h <= X/100, a bound on the first absolute discrepancy between short and long averages that improves on the trivial Turán-Kubilius bound and decays as h tends to infinity; Theorem 1.4 gives an l^2 (mean-square) analog for the restricted class A_s of additive functions, whose definition rules out the pathology of g(p)/B_g(X) being large on many primes, and rests on a Matomäki-Radziwill variant for divisor-bounded multiplicative functions (Theorem 4.3, from the author's earlier paper). Two families of applications follow. Theorem 1.11 shows that the average gap (1/X) sum |g(n) - g(n-1)| is o(B_g(X)) if and only if the first centered moment (1/X) sum |g(n) - A_g(X)| is o(B_g(X)), with a second-moment version for g in A_s, complementing results of Elliott and Hildebrand. On Erdős's 1946 conjecture (Conjecture 1.6) that an additive function non-decreasing outside a density-zero set B must equal c log n, Corollary 1.7 proves the conjecture for completely additive g satisfying lim F_g(epsilon) = 0 and the stronger sparseness |B(X)| << X/(log X)^(2+delta). Theorem 1.8 shows that any real additive g in A_s with |B(X)| = o(X) is close to lambda(X) log at prime powers, in a weighted mean square, and Theorem 1.9 shows that any real additive g with |B(X)| = o(X) satisfies g(n) = lambda(X) log n - eta(X) + o(B_g(X)) for all but o(X) integers n <= X, with slowly varying parameters. The paper bears on problem 1122 as partial progress on that Erdős conjecture characterizing constant multiples of log n among almost everywhere monotone additive functions.
Source: https://arxiv.org/abs/2108.12351.
Bears on. #1122
Results to transcribe.
- theorem_1_1: For any additive function g and integers 10 <= h <= X/100, the average over n in (X/2, X] of the absolute difference between the short-interval average of g on (n - h, n] and the long average is bounded in terms of B_g(X), improving on the trivial Turán-Kubilius bound by a factor tending to 0 as h tends to infinity.
- theorem_1_4: For additive g in the class A_s and an integer h = h(X) tending to infinity with 10 <= h <= X/100, the mean square of the difference between the short-interval average of g and the long average over n in (X/2, X] is o(B_g(X)^2).
- corollary_1_7: If g is completely additive with lim_{epsilon to 0} F_g(epsilon) = 0 and its set of decrease satisfies |B(X)| << X/(log X)^(2+delta) for some delta > 0, then g(n) = c log n for all n and some constant c, a partial case of Erdős's Conjecture 1.6.
- theorem_1_9: If g is additive with |B(X)| = o(X) then there are slowly varying parameters lambda(X), eta(X) with g(n) = lambda log n - eta + o(B_g(X)) for all but o(X) integers n <= X.
- theorem_1_8: For additive g in A_s with |B(X)| = o(X) there is lambda(X) with |lambda(X)| << B_g(X)/log X such that sum over prime powers p^k <= X of |g(p^k) - lambda log p^k|^2/p^k is o(sum g(p^k)^2/p^k).
- theorem_1_11: For additive g, (1/X) sum_{n<=X} |g(n) - g(n-1)| = o(B_g(X)) if and only if (1/X) sum_{n<=X} |g(n) - A_g(X)| = o(B_g(X)); for g in A_s the same equivalence holds for the corresponding second moments.