Wiki
Wiki

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

Updated

Problem 963

../


Statement. Let f(n)f(n) be the maximal kk such that in any set $A\subset \mathbb{R}$ of size nn there is a subset B⊆AB\subseteq A of size $\lvert B\rvert\geq k$ which is dissociated that is, the sums ∑b∈Sb\sum_{b\in S}b are distinct for all S⊆BS\subseteq B. Estimate f(n)f(n) - in particular, is it true that

f(n)≥⌊log⁡2n⌋?f(n)\geq \lfloor \log_2 n\rfloor?

Status. Open. The site labels the problem OPEN, with its note that no finite computation can settle it (page last edited 23 January 2026).

Source. erdosproblems.com/963, accessed 2026-09-04. Cite as: T. F. Bloom, Erdős Problem #963, https://www.erdosproblems.com/963.

References.

  • [Er65] Erdős, P., Extremal problems in number theory. Proc. Sympos. Pure Math. VIII, Amer. Math. Soc. (1965), 181--189. Printed p. 188: the bound k≥⌊log⁡n/log⁡3⌋k\ge\lfloor\log n/\log3\rfloor, called not difficult and given without proof, the question whether ⌊log⁡n/log⁡2⌋\lfloor\log n/\log2\rfloor is attainable, and the example ai=ia_i=i. Library home: erdos_1965_extremal_problems_number_theory.
  • [Va99] Various, Some of Paul's favorite problems. Booklet produced for the conference "Paul Erdős and his mathematics", Budapest, July 1999 (1999); the site cites item 1.22.

Formalization. No external statement recorded.

Current assessment

The question (site formulation, page last edited 23 January 2026). The statement above; OPEN. The site's one remark records that Erdős noted the greedy bound f(n)≥⌊log⁡3n⌋f(n)\ge\lfloor\log_3 n\rfloor. In [Er65] (p. 188) he calls the bound not difficult and prints no proof; the greedy argument, worked out on the card for that paper, takes a maximal dissociated B⊆AB\subseteq A, so that every element of AA is a combination of elements of BB with coefficients in {−1,0,1}\{-1,0,1\} and n≤3∣B∣n\le3^{|B|}. The same paragraph of [Er65] asks whether ⌊log⁡2n⌋\lfloor\log_2 n\rfloor is attainable and says that the example ai=ia_i=i, 1≤i≤n1\le i\le n, shows that this bound, if true, is nearly best possible: a dissociated kk-subset of {1,…,n}\{1,\ldots,n\} has 2k2^k distinct subset sums in [0,kn][0,kn], so 2k≤kn+12^k\le kn+1 and f(n)≤log⁡2n+O(log⁡log⁡n)f(n)\le\log_2 n+O(\log\log n). The known bounds are ⌊log⁡3n⌋≤f(n)≤log⁡2n+O(log⁡log⁡n)\lfloor\log_3 n\rfloor\le f(n)\le\log_2 n+O(\log\log n); whether f(n)≥⌊log⁡2n⌋f(n)\ge\lfloor\log_2 n\rfloor for every nn is the open question, and the problem has no claim page.

The finite comparison below rules out the initial interval as a universal minimizer of the largest dissociated-subset size. It does not resolve the logarithmic lower-bound question.

The site's discussion thread carries two arguments, neither a dated manuscript, so neither gets a claim page; the discussion card records the thread. In post 2027 (5 December 2025) KoishiChan claims that every nn-element set of reals contains a dissociated subset of size (1−o(1))log⁡2n(1-o(1))\log_2 n, by a recursion that dilates the set modulo a prime and finds well-populated progression cells through a character-sum second moment. The replies report an off-by-one defect, which the author says is fixed by lowering a parameter by one, and a review carried out with ChatGPT Pro, as the post names it, which found minor issues; on 23 January 2026 the site's curator, Thomas Bloom, wrote in post 3664 that the argument looked good to him and asked for a formal write-up; the site labels the problem OPEN and lists no proof claim (page last edited 23 January 2026). The bound is recorded here as an unverified community claim: an asymptotic lower bound, which would not by itself decide the ⌊log⁡2n⌋\lfloor\log_2 n\rfloor question. In post 3658 (23 January 2026) a commenter claims the exact bound f(n)≥⌊log⁡2n⌋f(n)\ge\lfloor\log_2 n\rfloor, labeling the proof AI-assisted; it rests on the unproved assertion that {1,…,n}\{1,\ldots,n\} minimizes the largest dissociated subset among nn-element real sets, which post 3662 rejects and the 13-element example below refutes, so the argument gives neither a proof nor a disproof of the problem.

Search scope, 2026-10-06: the site's problem page (OPEN, last edited 23 January 2026, source keys [Er65] and [Va99, 1.22], no proof claims) and its discussion thread of 20 posts through 3 September 2026; the community database lists the problem as open and unformalized as of its last update, and formal-conjectures has no statement file for it.

Known Results

Write d(A)d(A) for the largest size of a dissociated subset of a finite real set AA. BAKKAOUI's posts 8701 and 8709, 3 September 2026, give the fixed set

A∗={1,2,3,4,5,6,7,8,9,10,12,13,15}A^*=\{1,2,3,4,5,6,7,8,9,10,12,13,15\}

with d(A∗)=4d(A^*)=4, whereas d({1,…,13})=5d(\{1,\ldots,13\})=5. The source-owned reconstruction gives the exact witnesses, all 1287 required five-subset checks, and the heredity and interval upper-bound arguments. Post 8709 corrects post 8701; the author says that AI agents assisted the searches and the literature check and that the displayed example was checked by hand in exact integer arithmetic.

Because A∗A^* is itself an allowed real set, this yields f(13)≤4f(13)\le4. It does not yield f(13)=4f(13)=4; ⌊log⁡213⌋=3\lfloor\log_2 13\rfloor=3, so this example does not refute the catalog's proposed lower bound. It settles no instance of the question, so it has no claim page.

The larger searches through n=16 and window 34, the claimed OEIS identity and the negative literature search remain source-reported computations. Their bounded domains cannot establish the minimum over all real sets, the first possible positive-integer failure, or exceptional behavior at n=13.

Linked library material

These entries are derived from explicit links on library pages. They are navigation only and do not by themselves record mathematical progress.