Erdős Problem #1 — Maximum size of sets with distinct subset sums

About 95 years old · traced to

Let 1≤a1<a2<⋯<ak≤n1 \leq a_1 < a_2 < \cdots < a_k \leq n be a sequence of integers for which all the subset sums ∑i=1kεiai\sum_{i=1}^{k} \varepsilon_i a_i (εi=0\varepsilon_i = 0 or 11) are distinct. The powers of 22 have of course this property. Put f(n)=max⁡kf(n) = \max k. Is it true that

f(n)<log⁡nlog⁡2+c1f(n) < \frac{\log n}{\log 2} + c_1

for some absolute constant c1c_1? I offer 500 dollars for a proof or a disproof of (1).

References

Additional references

P. Erdős, Some of my favourite problems in number theory, combinatorics, and geometry, Resenhas IME-USP 2 (1995), 165-186.

Progress summary

Refreshed
Open

The conjecture remains open: newer work sharpens the lower bound, but no proof or counterexample has been found.

Erdős posed the question in 1931 or 1932: must every set with distinct subset sums and largest element aka_k have aka_k growing essentially like 2k2^k? Equivalently, is f(n)f(n) bounded above by log⁡nlog⁡2+O(1)\frac{\log n}{\log 2}+O(1)?

Known results

  • Erdős and Moser proved ak≥c2k/ka_k\geq c2^k/\sqrt{k}.
  • Elkies sharpened this to ak≥(2/π−o(1))k−1/22ka_k\geq(\sqrt{2/\pi}-o(1))k^{-1/2}2^k.
  • Bohman constructed examples with ak≤0.22002 2ka_k\leq0.22002\,2^k in 1998.
  • Exhaustive computation gives minimum largest element 309309 for k=10k=10, with no asymptotic conclusion.

February 5, 2026 note

A note dated February 5, 2026, records the conjecture as Conjecture 1.3 and gives the best preceding estimate ∣S(n)∣<log⁡2n+12log⁡2log⁡2n−log⁡2c∗+o(1)|S(n)|<\log_2 n+\frac12\log_2\log_2 n-\log_2 c^*+o(1), with c∗=2/πc^*=\sqrt{2/\pi} attributed to Dubroff, Fox, and Xu. It reports neither a proof nor a disproof.

Current status (as of September 2026): The logarithmic upper bound remains open; the best recorded general lower bound retains a factor k−1/2k^{-1/2}, and no counterexample is recorded.

Sources

Solutions 0

No solutions have been posted yet.