Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Statement

Theorem 7.1 (p. 9, quoted). "There is no constant c>0c>0 such that every sum-distinct set A⊆{1,…,N}A\subseteq\{1,\ldots,N\} satisfies N>c 2∣A∣N>c\,2^{|A|}. Equivalently, the ratio 2∣A∣/N2^{|A|}/N is unbounded over such sets."

The print leaves the range of NN implicit. For N=0N=0 the only such set is empty and the bound fails trivially, so the content is the case N≥1N\geq1, which is the range Proposition 1.1 uses and the construction supplies. The proof below also deduces an equivalent form that the source does not state: for every ε>0\varepsilon>0 there are examples of arbitrarily large cardinality with

N≤ε2∣A∣.N\leq\varepsilon2^{|A|}.

Proof

Fix an integer multiplier k≥0k\geq0. [[additive_combinatorics/adamczewski_2026_erdos1/proposition_3_2|Proposition 3.2]] supplies an admissible real matrix with dyadic denominator, common column sum qq, and kdet⁡C<qk\det C<q. The [[additive_combinatorics/adamczewski_2026_erdos1/lattice_reduction|integer lattice reduction]] clears denominators and chooses an upper-triangular integer basis with

kD<2nrkD<2^{nr}

and the required balanced-cube separation. [[additive_combinatorics/adamczewski_2026_erdos1/lemma_5_1|Lemma 5.1]] and [[additive_combinatorics/adamczewski_2026_erdos1/proposition_5_2|Proposition 5.2]] preserve a quantitative separation after the unitriangular perturbation. The [[additive_combinatorics/adamczewski_2026_erdos1/normal_coefficients|normal coefficients]] are eventually positive and define exactly the perturbed lattice kernel. The [[additive_combinatorics/adamczewski_2026_erdos1/digit_injectivity|digit-box map]] is therefore injective. Finally, [[additive_combinatorics/adamczewski_2026_erdos1/binary_expansion|binary expansion]] gives N≥1N\geq1 and a sum-distinct A⊆{1,…,N}A\subseteq\{1,\ldots,N\} with

2∣A∣N=2nrD>k.\frac{2^{|A|}}{N}=\frac{2^{nr}}D>k.

Thus kN<2∣A∣kN<2^{|A|} for every kk, and [[additive_combinatorics/adamczewski_2026_erdos1/proposition_1_1|Proposition 1.1]] proves the first two formulations.

For the quantified ε\varepsilon statement, given also an integer LL, take

k≥max⁡(ε−1,2L).k\geq\max(\varepsilon^{-1},2^L).

Then N<2∣A∣/k≤ε2∣A∣N<2^{|A|}/k\leq\varepsilon2^{|A|}. Since N≥1N\geq1, the strict inequality 2∣A∣>k≥2L2^{|A|}>k\geq2^L also gives ∣A∣>L|A|>L. Hence the cardinalities can be made arbitrarily large.

Real variant

The same examples disprove the analogous assertion for finite A⊂(0,N]A\subset(0,N] whose distinct subset sums must be at least 11 apart. Indeed, the constructed elements are positive integers, and two unequal integer subset sums differ in absolute value by at least 11. This is a direct consequence of the integer theorem, not an additional construction.

Source and dependencies

An explanation of the proof of Erdős Problem 1, preliminary exposition with no named author (erdosproblems.com, 2026), §7, Theorem 7.1 and the summary on pp. 9–10. The edition read is named on the source card. The arbitrarily-large cardinality deduction and real-variant consequence are elementary deductions from the displayed theorem. No explicit or optimal dependence of the first cardinality on ε\varepsilon is claimed.

Bears on. #1: the problem asks whether every sum-distinct A⊆{1,…,N}A\subseteq\{1,\ldots,N\} with ∣A∣=n|A|=n has N≫2nN\gg2^n; Theorem 7.1 states that no constant c>0c>0 gives N>c2∣A∣N>c2^{|A|} for all such sets, the negation of that bound. The exposition is an account of a formal proof and is not itself refereed; the standing of the problem is recorded on its page.