A NOTE ON THE ERDŐS DISTINCT SUBSET SUMS PROBLEM

QUENTIN DUBROFF, JACOB FOX, AND MAX WENQIANG XU

arXiv:2006.12988v2 [math.CO] 20 Jul 2020

Abstract. We present two short proofs giving the best known asymptotic lower bound for the maxi-
mum element in a set of n positive integers with distinct subset sums.

Let {a1 , . . . , an } be a set of positive integers with a1 < . . . < an and all subset sums distinct. Erdős
conjectured that an ≥ c · 2n for some constant c > 0 and offered $500 for a proof or disproof. See [8]
for more history on this. Using the second moment method, Erdős and Moser [6] (see also [2]) proved
1
an ≥ · n−1/2 · 2n .
4
There have been some improvements on the constant factor of the above lower bound, including the
work of Guy [7], Elkies [5], Bae [3], and Aliev [1]. In p particular, the previous best published lower
bound was due to Aliev, which gave the constant factor 3/2π. In the other direction, the best known
construction is due to Bohman [4], where he constructed such sets with an ≤ 0.22002 · 2n .
In this note, we give two proofs of the best known lower bound, matching an unpublished result of
Elkies and Gleason (see [1]).
Theorem 1. If a set {a1 , . . . , an } of integers with 0 < a1 < . . . < an has all subset sums distinct, then
r !
2
an ≥ − o(1) · n−1/2 · 2n .
π
n

The second proof uses an isoperimetric inequality to show that an ≥ ⌊n/2⌋ for all n, which asymp-
totically matches the lower bound above. The first proof uses the following result from probability.
Theorem 2 (Berry-Esseen). Let X1 , . . . , Xn be independent random variables with E[Xi ] = 0, E[Xi2 ] =
σi2 , and E[|Xi |3 ] = ρi < ∞. Let X = X1 + · · · + Xn and σ 2 = E[X 2 ]. Then
sup |F (x) − Ψ(x)| ≤ C · ψ,
x∈R
where F (x) and Ψ(x) are the cumulative distribution functions for X and the normal distribution
P with
mean zero and standard deviation σ, respectively, C is an absolute constant, and ψ = σ −3 · ni=1 ρi .

It is known [10] that one can take C = 0.56 in the Berry-Esseen theorem.
Using the Berry-Esseen theorem and an inequality of Moser, our proof verifies Erdős’ conjecture that
an = Ω(2n ) if the distribution of a random subset sum is not close to normal. If the distribution is
close to normal, we replace Chebyshev’s inequality in the second moment argument by the normal
distribution in bounding the probability that a random subset sum is in an interval.

Fox is supported by a Packard Fellowship and by NSF award DMS-1855635. Xu is supported by the Cuthbert C.
Hurd Graduate Fellowship in the Mathematical Sciences, Stanford.
1
2 QUENTIN DUBROFF, JACOB FOX, AND MAX WENQIANG XU

Proof of Theorem 1. Let ǫ1 , . . . , ǫn ∈ {−1, 1} be independently and uniformly distributed and let
X = ǫ1 a1 + . . . + ǫn an . The random variable X has mean 0 and is symmetric around 0. Moreover,
each of its possible values occurs with probability 2−n by the distinct subset sumsPproperty, and each
of its possible values are of the same parity. Note that the variance of X is σ 2 = ni=1 a2i .
Let δ = δn tend to 0 slowly as n tends to infinity, with δ > n−1/2 , e.g., δ = 1/ log n works. We first
notice that if σ ≤ an /δ, then Moser’s lower bound on the variance [7]:
X 4n − 1
σ2 = a2i ≥ 12 + 22 + 42 + · · · 22(n−1) =
3
1≤i≤n

gives the desired lower bound an ≥ δ · 2n−1 > n−1/2 · 2n .

Pn
It remains to consider the case σ P
> an /δ. We next apply Theorem 2. Note that X = i=1 Xi , where
3 n Pn 2 2
Xi = ǫi ai . We have ρi = ai and i=1 ρi ≤ an · i=1 ai = an σ , which implies ψ ≤ an /σ < δ by the
definition of ψ and the bound on σ. By Theorem 2, we have supx∈R |F (x) − Ψ(x)| ≤ δ, that is, X is
δ-close to normal. We bound Pr[|X| ≤ ℓ] in two different ways to get a lower bound for σ.
Since the 2n possible outcomes for X have the same parity and are distinct, we obtain
(1) Pr[|X| ≤ ℓ] ≤ (ℓ + 1) · 2−n .
On the
√ other hand, let ℓ = ασ with α = αn tending to 0 as n tends to infinity with δ = o(α), e.g.,
α = δ works. As X is δ-close to a normal distribution, we obtain that
Z ℓ r
1 x2
2 ℓ
(2) Pr[|X| ≤ ℓ] = √ exp − 2 dx + O(δ) ∼ .
σ 2π −ℓ 2σ π σ
p 
Comparing (1) and (2), we obtain that σ ≥ 2/π − o(1) ·2n . Comparing this with the upper bound
p 
σ 2 = a21 + · · · + a2n ≤ na2n , we have an ≥ 2/π − o(1) · n−1/2 · 2n as desired. ✷

The second proof will use the following special case of Harper’s vertex-isoperimetric inequality. We
write 2[n] for the set of subsets of {1, 2, . . . , n}, and for F ⊆ 2[n] we define the vertex boundary
∂F = {A ∈ F c : |A△B| = 1 for some B ∈ F}.
n

Theorem 3 (Harper). If F ⊂ 2[n] such that |F| = 2n−1 , then |∂F| ≥ ⌊n/2⌋ .

See [9] for a full statement of Harper’s theorem as well as [11] and [12] for a particularly nice proof.

A second proof of Theorem 1. Set a = (a1 , . . . , an ) and note that a · ǫ 6= 0 for all ǫ ∈ {− 21 , 12 }n . Let F
be the 2n−1 points ǫ in {− 12 , 12 }n such that a · ǫ < 0. Identifying {− 12 , 21 }n with 2[n] in the natural way,
n

we have η ∈ ∂F if η ± ei ∈ F for some standard basis vector ei . Theorem 3 then gives |∂F| ≥ ⌊n/2⌋ ,
where each η ∈ ∂F must satisfy 0 < a · η < an . Therefore there are distinct ǫ, η ∈ ∂F such that
n
. However, ǫ − η ∈ {−1, 0, 1}n , so |a · (ǫ − η)| = | i∈I ai − j∈J aj | for some

P P
|a · (ǫ − η)| ≤ an / ⌊n/2⌋
I, J ⊆ [n], and we get |a · (ǫ − η)| ≥ 1 by the distinct subset sum property. Comparing the lower and
n

p
upper bound on |a · (ǫ − η)| gives an ≥ ⌊n/2⌋ = 2/π − o(1) · n−1/2 · 2n . ✷

Acknowledgements. We would like to thank Noga Alon, Matthew Kwan, Bhargav Narayanan, Joel
Spencer and Yufei Zhao for helpful comments.
 A NOTE ON THE ERDŐS DISTINCT SUBSET SUMS PROBLEM 3

References
[1] I. Aliev, Siegel’s lemma and sum-distinct sets, Discrete Comput. Geom. 39 (2008), 59–66.
[2] N. Alon and J. H. Spencer, The probabilistic method, 4th ed. Wiley, Hoboken, NJ, 2016.
[3] J. Bae, On subset-sum-distinct sequences. Analytic number theory, Vol. 1, Progr. Math., 138, Birkhäuser, Boston,
1996, 31–37.
[4] T. Bohman, A construction for sets of integers with distinct subset sums, Electron. J. Combin. 5 (1998), Research
Paper 3, 14 pages.
[5] N. D. Elkies, An improved lower bound on the greatest element of a sum-distinct set of fixed order, J. Combin.
Theory Ser. A 41 (1986), 89–94.
[6] P. Erdős, Problems and results in additive number theory, Colloque sur la Théorie des Nombres, Bruxelles, 1955, pp.
127–137, George Thone, Liège; Masson and Cie, Paris, 1956.
[7] R. K. Guy, Sets of integers whose subsets have distinct sums, Theory and practice of combinatorics, 141–154, North-
Holland Math. Stud., 60, Ann. Discrete Math., 12, North-Holland, Amsterdam, (1982).
[8] R. K. Guy, Unsolved Problems in Intuitive Mathematics, Vol. I, Number Theory, Problem C8, Springer-Verlag (1981).
[9] L. H. Harper, Optimal numberings and isoperimetric problems on graphs, J. Combin. Theory 1 (1966), 385-393.
[10] I. G. Shevtsova, An improvement of convergence rate estimates in the Lyapunov theorem, Dokl. Math. 82 (2010),
862–864.
[11] D. J. Kleitman, Extremal hypergraph problems. Surveys in combinatorics, Proc. Seventh British Combinatorial
Conf., Cambridge, (1979), pp. 4465, London Math. Soc. Lecture Note Ser., 38, Cambridge Univ. Press, Cambridge-New
York, (1979).
[12] I. Leader, Discrete isoperimetric inequalities. Probabilistic combinatorics and its applications (San Francisco, CA,
1991), 57–80, Proc. Sympos. Appl. Math., 44, AMS Short Course Lecture Notes, Amer. Math. Soc., Providence, RI,
1991.

Department of Mathematics, Rutgers University, Piscataway, NJ, USA

E-mail address: quentin.dubroff@rutgers.edu

Department of Mathematics, Stanford University, Stanford, CA, USA

E-mail address: jacobfox@stanford.edu

Department of Mathematics, Stanford University, Stanford, CA, USA

E-mail address: maxxu@stanford.edu