The Erdős distinct subset sums problem in a modular setting

∗†
Stijn Cambie Jun Gao∗ Younjin Kim∗ Hong Liu∗

Abstract
arXiv:2308.03748v1 [math.CO] 7 Aug 2023

We prove the following variant of the Erdős distinct subset sums problem. Given t ě 0 and
sufficiently large n, every n-element set A whose subset sums are distinct modulo N “ 2n ` t
satisfies ´1 ¯
max A ě ´ op1q N.
3
Furthermore, we provide examples showing that the constant 31 is best possible. For small values
of t, we characterise the structure of all sumset-distinct sets modulo N “ 2n ` t of cardinality n.

1 Introduction
The Erdős distinct subset sums problem, considered one of the oldest conjectures of Erdős [10], poses
the question whether there exists a fixed constant c ą 0 such that for every n-element set A “
ta1 , a2 , ¨ ¨ ¨ , an u of positive integers, all of whose subsets have distinct sums, the largest element, say
n 2n
an “ max A, has to be greater than ´ ¯c ¨ 2 . Erdős and Moser [11] first observed that an ě n and later
2 n
improved this lower bound to Ω ? n
by considering the sum of the squares of the elements [13]. Erdős
has posed this question multiple times and even offered monetary prize for its proof or disproof [9].
The lower bound on the largest element an has subsequently
´b been¯improved by a constant factor
2 2n
[2, 13, 8, 4, 1, 7]. The current best known lower bound π ´ op1q is due to Dubroff, Fox and
?
n
Xu [7]; an alternative proof is given by Steinerberger [16]. In the other direction, starting with a
construction by Conway and Guy [6], where they produced a construction with an ď 2n´2 , better
constructions have been obtained [14, 3, 15]. The best construction so far is the one by Bohman [5],
where he produced a construction with an ď 0.22002¨2n . Here he studied variations of the Conway-Guy
sequence. These sequences are derived from differences between an element and the smaller ones in an
increasing sequence pun qn which is exponential: S “ tun ´ u1 , un ´ u2 , . . . , un ´ un´1 u. As such, most
of the elements in the set have roughly the same magnitude. This implies that two subsets of different
sizes have a different subset-sum in most cases. For two subsets with the same size, it is sufficient to
have a different subset-sum modulo un .
Motivated by this, we consider the Erdős distinct subset sums problem in a modular setting. An
n-element set ta1 , a2 , ¨ ¨ ¨ , an u of positive integers is said to be sumset-distinct modulo N if all subsets
(including the empty set) have distinct sums modulo N. Note that we may assume N ě 2n for otherwise
not all subset sums of a size-n set can be distinct modulo N . We propose the following modular version
of the Erdős distinct subset sums conjecture.
Conjecture 1.1. For a fixed t ě 0 and sufficiently large n, let A be an n-element set of positive
integers, all of whose subsets have distinct sums modulo N “ 2n ` t, it holds that
max A ě N {3 ě 2n {3.
∗
Extremal Combinatorics and Probability Group (ECOPRO), Institute for Basic Science (IBS), Daejeon, South Korea,
supported by the Institute for Basic Science (IBS-R029-C4), E-mail: stijn.cambie@hotmail.com,{jungao, hongliu,
younjinkim}@ibs.re.kr.
†
Department of Computer Science, KU Leuven Campus Kulak-Kortrijk, 8500 Kortrijk, Belgium. Supported by
Internal Funds of KU Leuven (PDM fellowship PDMT1/22/005).

1
 The condition that n is sufficiently large is necessary, as shown by the following example. Let
n “ 9, t “ 104 and consider the set t77, 117, 137, 148, 154, 157, 159, 160, 161u. It is sumset-distinct
modulo 616 “ 29 ` 104, but 161 ă 13 29 ă N3 .
In this paper, we prove the following asymptotic version of Conjecture 1.1. Comparing the constant
1
3 below with the one (0.22002) obtained by Bohman [5], our result witnesses a slightly different
behaviour of sumset-distinct sets in the modular setting.

Theorem 1.2. Let 0 ă ε ă 1 and n, t be positive integers such that t ă p1 ` εqn . Assume that
A “ ta1 , a2 , ¨ ¨ ¨ , an u is an n-element set of positive integers in which the sums of all its subsets are
distinct modulo N “ 2n ` t. Then
max A ě p1 ´ εqN {3.
Furthermore, the constant 1{3 is optimal. That is, There exist sets that are sumset-distinct modulo
N “ p1 ` op1qq2n for which max A ď p 13 ` op1qqN.

A nice consequence of Conjecture 1.1 is that itřcan be used to prove the original Erdős distinct
subset sums problem when the sum of all elements ni“1 ai is close to 2n .

Proposition 1.3. If Conjecture

řn 1.1 is true, then for any A “ ta1 , a2 , . . . , an u, a1 ă a2 ă . . . ă an ,
with distinct subset sums and i“1 ai ă 2 ` t, we have 2i´1 ď ai ď 2i´1 ` t for every large i.
n

Lending support to Conjecture 1.1, we establish its validity for 0 ď t ď 3 in strong sense by
characterising the structure of all maximum-size sumset-distinct sets modulo N “ 2n ` t.
We show that a set of cardinality n is sumset-distinct modulo 2n if and only if each element has a
distinct 2-adic valuation. The p-adic valuation of an integer N is the integer k such that pk divides N ,
but pk`1 does not. This valuation is denoted as vp pN q.

Theorem 1.4. A set A of cardinality n is sumset-distinct modulo N “ 2n if and only if its elements
can be ordered as A “ ta0 , a1 , . . . , an´1 u such that v2 paj q “ j for every 0 ď j ď n ´ 1.

When N “ 2n ` 1 or N “ 2n ` 3, we prove that maximum-size sumset-distinct sets modulo N , up

to signs, are dilations of the powers of 2.

Theorem 1.5. A set A of cardinality n is sumset-distinct modulo N “ 2n ` 1 or 2n ` 3 for some

integer n ě 1 if and only if it can be expressed as A “ t˘λ20 , ˘λ21 , . . . , ˘λ2n´1 u pmod N q for some
λ satisfying gcdpλ, N q “ 1 and some choice of the signs.
2n φpN q
Corollary 1.6. For every integer n ě 1, there are exactly 2n sumset-distinct sets modulo N “
2n ` 1 of cardinality n.

Modulo N “ 2n ` 2 is closely related to modulo 2n´1 ` 1.

Theorem 1.7. A set A of cardinality n is sumset-distinct modulo N “ 2n ` 2 for some integer n ě 2

if and only if
(i) A is the union of an odd number and 2A1 , where A1 is sumset-distinct modulo N {2, or
(ii) A is the union of N {2 and A1 , where A1 is sumset-distinct modulo N {2. In particular, the number
of such sets is OpN 3 q.

Somewhat surprisingly, our results show that there are more solutions for N “ 2n (when every sum
appears exactly once) compared to N “ 2n ` t for any t ą 0 and n sufficiently large. In the former
2 2
case, the number of solutions is superpolynomial, at least 2Ωpn q “ ´ 2Ωplog N q¯, whereas in the latter
? 3`t
case, for t odd, the number of solutions is polynomial, bounded by O N .

2
2 Preliminaries
ř
For a given set A, we define spAq “ aPA a as the sum of its elements, and SpAq as the set of sums of
subsets of A, that is, SpAq “ tspBq : B Ď Au . A translate of a set is denoted by A` b “ ta` b : a P Au.
A primitive kth root of unity is a solution to the equation xk “ 1 that does not satisfy xm` “ ˘1 for
any m dividing k. We use ωk for the standard kth root of unity, which is equal to exp 2πi k , so
1 ` ωk ` ωk2 ` . . . ` ωkk´1 “
XN \ 0. For any x P Z, let respxq “ mintx pmod N q, N ´ x pmod N qu be the
element between 0 and 2 which is equivalent to x or ´x modulo N.
We start with two lemmas which indicate that there can be multiple sumset-distinct sets modulo
N which are essentially the same. This implies that one can partition the sumset-distinct sets modulo
N into equivalent classes of sets which are related with a composition of two operations: dilation or
changing signs.

Lemma 2.1. A set A “ ta1 , . . . , an u is sumset-distinct modulo N if and only if each of the 2n sets of
the form t˘a1 , . . . , ˘an u are sumset-distinct modulo N .

Proof. It suffices to observe that if ta1 , a2 . . . , an u is a sumset-distinct set modulo N , then t´a1 , a2 , . . . , an u
is also a sumset-distinct setřmodulo N ř . This can be easily seen: if there exist disjoint sets I1 and I2
from the set rns such that iPI1 ai ” iPI2 ai pmod N q, then the same sets work when 1 R I1 Y I2 .
Otherwise, if 1 is included in either I1 or I2 , we can simply switch the element 1 between the two
sets. In other words, let I11 “ I1 ∆t1u1 and I21 “ I2 ∆t1u. These are two distinct index subsets of rns,
corresponding to two subsets of elements in t´a1 , a2 , . . . , an u with the same sum modulo N.

Lemma 2.2. For a given λ where gcdpλ, N q “ 1, a set A “ ta1 , . . . , an u is sumset-distinct modulo N
if and only if tλa1 , . . . , λan u is also sumset-distinct modulo N .
ř ř ř ř
Proof. It suffices to note that iPI1 ai ” iPI2 ai pmod N q if and only if λ iPI1 ai ” λ iPI2 ai
pmod N q.

We will also need the following corollary of [12, Thr. 1.5]. Here we use that the group NZZ , ` is
` ˘

abelian and does not have elements of order 2 when N is odd.

Theorem 2.3 ([12]). Let A, B be multisets in G “ NZZ for an odd positive integer N. Then SpAq “
SpBq implies that A be obtained from B by applying one of the following moves

1. replace an element x by ´x;

2. for some α, β relative prime with N , replace the elements in the set αUk by βUk where Uk “
t1, 2, . . . , 2k´1 u and k ą 0 is the smallest integer for which 2k ” ˘1 pmod N q.

When the cardinality of A and B is n, we have that

• when N “ 2n ` 1, SpAq “ SpBq implies that the elements of B can be obtained up to sign from
multiplying all elements in A with some element λ “ βα´1 which is relative prime to N,

• when N ą 2n ` 1, SpAq “ SpBq implies that the elements in A and B are equal up to sign.

For N ě 2n ` 3, it is important to note that the smallest positive integer k ě 1 satisfying 2k ” ˘1

mod N must satisfy k ě n ` 1 and thus no multiple of Uk can be a subset of a set with cardinality n.
1
Here ∆ denotes the symmetric difference of the two sets.

3
3 The asymptotic modular Erdős distinct subset sums conjecture
Proof of Theorem 1.2. Let us define the sumset generating function
ź
gA pxq “ p1 ` xai q
i

and let ts1 , s2 , . . . , st u “ NZZ zSpAq be the set of t numbers that cannot be expressed as the sum of a
subset of A. Then we have
ÿt ´1
Nÿ
si i
gA pωN q ` ωN “ ωN “ 0.
i“1 i“0
Hence, by the triangle inequality, we have
t n n c ´a ¯
i
ÿ si
ź ai
ź
tě ωN “ |gA pωN q| “ 1` ωN “ 2 ` 2 cos ¨ 2π .
i“1 i“1 i“1
N
x2
Now, if max A ă p1 ´ εq N3 , then using cos π3 ` 2π 1
` ˘
3 x ă 2 ´ x ´ 2 for 0 ă x ă 1, we have for every
i P rns that
c d ˆ ˙ d ˆ ˙
´a
i
¯ 1´ε π 2επ
2 ` 2 cos ¨ 2π ą 2 ` 2 cos ¨ 2π “ 2 ´ 2 cos ` ą 1 ` ε.
N 3 3 3
This would imply t ą p1 ` εqn , contradicting the assumption.
We now construct the sets showing that the constant c “ 31 cannot be improved.
Let n " k " 1 be two integers, and consider N “ 2n ` 2k ` 1 such that 3 | pN ´ 1q, and so
3 | 2n´k ` 1. Let B “ ta1 , a2 , ¨ ¨ ¨ , an u, where
# i´1
2 , 1 ď i ď n ´ k,
ai “ i´1´pn´kq n´k
2 p2 ` 1q, n ´ k ă i ď n.
řj řn
Since i“1 ai ď aj`1 and i“1 ai ă N , B is sumset-distinct module N . Let λ “ N 3´1 , then gcdpλ, N q “
gcdpN ´ 1, N q “ 1. By Lemma 2.2, we know that λB is also sumset-distinct modulo N . Let A be
the set obtained from λB by restricting all elements to rN {2s, that is, replacing x by respxq “ mintx
pmod N q, N ´ x pmod N qu. Since λai “ a3i N ´ a3i , it is not hard to see that for i ď n ´ k, either
ai
λai ă N3 ` 2Nk pmod N q or λai ą 2N N
3 ´ 2k pmod N q. If i ą n ´ k, we have 3 | N ´ 1 and thus λai ” ´ 3
pmod N q. Thus, we conclude that max A ď N3 ` 2Nk “ p 13 ` op1qqN as n " k " 1.

We complement Theorem 1.2 with the following observation for the original setting.
` `˘
Proposition 3.1. Let A “ ta1 , a2 , . . . , an u P Zn be a set in which the sums of its subsets are distinct.
Let N “ 1 ` ni“1 ai . If ai ď 2n´1 ´ 1, then ai ă N3 .
ř

Proof. Let us consider an element ai that is at most 2n´1 ´ 1. Note that the interval r0, 2ai ´ 1s can
contain at most ai ă 2n´1 elements from the set SpAztai uq. Consequently, SpAztai uq contains an
element that is at least 2ai . This implies that SpAq contains an element that is at least 3ai , and
therefore 3ai ă N.

We can now prove Proposition 1.3 using Proposition 3.1.

ř us consider A “ ta1 , a2 , . . . , an u with a1 ă a2 ă . . . ă an having distinct

Proof of Proposition 1.3. Let
subset sums and satisfying ni“1 ai ă 2n `t for a fixed value of t. Then A has also distinct sums modulo
1` n
ř
n 2n `t i“1 ai
2 ` t. Assuming Conjecture 1.1, we have max A ě 3 ě . Applying Proposition 3.1, we
n´1
ř3 n´1
conclude that an ě 2 . Since A has distinct subset sums, i“1 ai ě 2n´1 ´ 1. Consequently,
an ď 2 n´1 ` t. Now one can repeat the argument for the set ta1 , a2 , . . . , an´1 u.

4
4 Sumset-distinct sets modulo N “ 2n
In this section, we prove Theorem 1.4. The “if” direction can be easily proven by induction. Since
the cardinality of SpAq is 2n , it suffices to observe that for every integer k, there is a solution for
ε0 a0 ` ε1 a1 ` . . . ` εn´1 an´1 ” k pmod 2n q, where εi P t0, 1u for every 0 ď i ď n ´ 1. Note that ε0 ” k
pmod 2q. By the induction hypothesis for n1 “ n ´ 1 and a1j “ aj`1 {2, 0 ď j ď n1 ´ 1, we find a solution
for ε1 a10 ` ε2 a11 ` . . . ` εn´1 a1n1 ´1 ” k´ε20 a0 pmod 2n´1 q, implying that ε1 a1 ` . . . ` εn´1 an´1 ” k ´ ε0 a0
pmod 2n q, as desired.
We provide two proofs of the “only if” direction. For each 0 ď j ď n´1, let Aj “ ta P A | v2 paq ď ju.

Proof 1. Consider the generating function Gj pxq “ aPAj p1 ` xa q , which, in its expanded form, equals
ś
a n j`1 appears equally often
ř
aPSpAj q x . Since A is sumset-distinct modulo 2 , every residue modulo 2
řN ´1 i
among the elements in SpAj q. By considering the expansion as a sum and using i“0 ω2j`1 “ 0, we
` πi ˘
can deduce that Gj pxq is zero when evaluated in ω2j`1 “ exp 2j . The factorization of Gj pxq implies
the existence of a factor 1 ` ω2aj`1 “ 0, indicating that there is at least one value a for which a ” 2j
pmod 2j`1 q, or equivalently, v2 paq “ j. Since this holds for every 0 ď j ď n ´ 1 and considering that
|A| “ n, we can conclude that the elements of A can be ordered such that v2 paj q “ j, establishing the
desired conclusion.

Proof 2. Assume, for the sake of contradiction, that there exists a set A of cardinality n that is sumset-
distinct modulo 2n but does not have the claimed form. Let us consider such a set A for which the
cardinality n is minimal. We observe that A must contain an odd number, as SpAq contains all 2n
residues modulo 2n and therefore includes the odd residues as well. If A0 “ ta0 u would be a singleton,
then pAzA0 q{2 “ ta{2 | a P AzA0 u would be sumset-distinct modulo 2n´1 and would not have the
claimed form either. This would contradict our choice of A being a minimal example. Thus, we
conclude that |A0 | ě 2.
Since A is sumset-distinct modulo 2n , SpAq “ t0, 1, 2, 3 ¨ ¨ ¨ , 2n ´ 1u pmod 2n q and so
n ´1
2ÿ
ÿ ÿ ÿ
n´1
2 a“ a” i “ 2n´1 p2n ´ 1q pmod 2n q,
aPA BĎA aPB i“1
ř
which implies thatř aPA a is odd.
Let Ne “ tB : aPB a is even, B Ď Au, and let t :“ |A0 | ě 2.
Claim 4.1. BPNe aPB a “ 2n´2 aPA a
ř ř ř

Proof of claim. When an odd element in A is fixed, the number of choices for the other elements (to
form B) such that the sum is even is given by
ˆ ˙
n´t
ÿ t´1
2 ¨ “ 2n´t`t´2 “ 2n´2 ,
0ďiďt´1,i is odd
i

where we have used the fact that t ě 2.

Similarly, when an even element in A is fixed, the number of choices for the other elements such
that the sum is even is
ˆ˙
n´t´1
ÿ t
2 ¨ “ 2n´t´1`t´1 “ 2n´2 .
0ďiďt,i is even
i

5
 Using Claim 4.1, we can now derive that:

ÿ ÿ ÿ 2n´1
ÿ´1
n´2
2 a“ a” 2i “ 2n´1 ¨ p2n´1 ´ 1q pmod 2n q.
aPA BPNe aPB i“1
ř
However, this implies that aPA a is even, a contradiction.

5 Sumset-distinct sets modulo N “ 2n ` 1 or N “ 2n ` 3

In this section, we prove Theorem 1.5. The “if” direction is trivial, following from applying Lemma 2.1
and Lemma 2.2 on A “ t1, 2, 22 , . . . , 2n´1 u.
For the “only if” direction, we begin by proving a claim that will be used for both cases.

Lemma 5.1. If N is 2n ` 1 or 2n ` 3, then every element in a sumset-distinct set (modulo N ) A must

be relatively prime to N.

Proof. Suppose there exists an element a P A and a common divisor d such that d | gcdpa, N q and
d ą 1. Note that if N “ 2n ` 3, then d ą 3. Since SpAq contains exactly the elements in SpAztauq along
with its translate a`SpAztauq, every residue modulo d appears an even number of times. Now consider
the fact that SpAq contains all residues modulo N , except for exactly one residue (when N “ 2n ` 1)
or at most three residues (when N “ 2n ` 3). Hence, at least one element (at least d ´ 1 in the case of
N “ 2n ` 1, or at least d ´ 3 in the case of N “ 2n ` 3 ) appears Nd times modulo d. However, since N
is odd, we know that Nd is also odd. This leads to a contradiction with the previous observation that
every residue appears an even number of times modulo d.

We first prove the “only if” direction for N “ 2n ` 1.

Proof for N “ 2n ` 1. Let us fix one element λ P A and consider A1 :“ λ´1 A “ tλ´1 a | a P Au, where
λ´1 is the inverse of λ in NZZ . Since gcdpλ, N q “ 1, we have that gcdpλ´1 , N q “ 1, and so A1 is also
sumset-distinct modulo N by Lemma 2.2. We note that by definition, 1 belongs to A1 and there is
exactly one residue x modulo N that is not part of SpA1 q. Furthermore, it is evident that for any y, if
y, y ` 1 ı x pmod N q, then exactly one of the two belongs to SpA1 zt1uq.
By writing the residues as the elements x`1´N up to x´1, we can deduce that SpA1 zt1uq pmod N q
contains exactly the even numbers within the interval from x` 1´ N up to x´ 1. Consequently, we can
express A1 “ t1, a2 , a3 , . . . , an u, where each ai is an even number within the interval rx ` 1 ´ N, x ´ 1s.
We now present three elementary observations:

• the sum of any set of negative numbers in A1 zt1u “ ta2 , a3 , . . . , an u is an even negative number,

• since N is odd, SpA1 zt1uq is not congruent modulo N to any even number between x ´ 2N and
x ´ N,

• for every 2 ď i ď n, we have |ai | ă N .

Combining these observations, we can conclude that SpA1 zt1uq does not contain any element smaller
than x ` 1 ´ N. Similarly, no element in SpA1 zt1uq is larger than x ´ 1. Consequently, SpA1 q is precisely
equal to the interval rx ` 1 ´ N, x ´ 1s. A corollary of this is that the sum of all negative numbers
in A1 results in the smallest number, x ` 1 ´ N , and the sum of all positive numbers in A1 equals
x ´ 1. By replacing the negative numbers in A1 with their additive inverse, we obtain A2 , which is
still sumset-distinct modulo N due to Lemma 2.1. Additionally, the sum of all elements in A2 equals
N ´ px ` 1q ` px ´ 1q “ N ´ 2 “ 2n ´ 1 and so SpA2 q “ t0, 1, 2, . . . , 2n ´ 1u. It remains to show that
A2 “ t1, 2, 4, . . . , 2n´1 u. Suppose we know the i smallest elements of A2 are 1, 2, 4, . . . , 2i´1 , for some

6
i with 0 ď i ď n ´ 1, then since 2i is the smallest element in SpAq that cannot be expressed as a sum
of any subset of these i smallest elements, it must belong to A2 . By induction, we can establish the
desired conclusion.

X \ 1.6. Recall that respxq “ mintx pmod N q, N ´ x pmod N qu is the element be-
Proof of Corollary
tween 0 and N2 equivalent to x or ´x modulo N. For every λ P rN s for which gcdpλ, N q “ 1, we
consider Rpλq “ trespλq, resp2λq, . . . , resp2n´1 λqu as a representative set for all the sets of the form
t˘λ, ˘2λ, . . . ˘ 2n´1 λu. We note that Rp2λq “ Rpλq since resp2n λq “ resp´λq, as a consequence of
2n ” ´1 pmod N q. This implies that Rpλq “ Rp˘2i λq for every 0 ď i ď n ´ 1. On the other hand,
if λ1 R t˘λ, ˘2λ, . . . ˘ 2n´1 λu, as elements in NZZ , then Rpλq ­“ Rpλ1 q and furthermore these sets are
q
disjoint. There are φpN q choices for λ, and thus φpN 2n different representative sets Rpλq. Every set
Rpλq is associated with 2n different sets of the form t˘λ, ˘2λ, . . . ˘ 2n´1 λu. We conclude that there
n q
are exactly 2 φpN2n sumset-distinct sets modulo N “ 2n ` 1.

Finally, we prove the case N “ 2n ` 3 and begin by presenting a lemma.

Lemma 5.2. Suppose A is a sumset-distinct set modulo N “ 2n ` t, where t is an odd positive integer,
and A has cardinality n. In this case, the missing elements from SpAq can be divided into a singleton
x and t´1
2 pairs ty, zu such that y ` z “ 2x.

Proof. Let T :“ spAq be the sum of all elements in A. For every subset B Ă A, we have spAzBq`spBq “
T. Therefore, an element a belongs to SpAq if and only if T ´ a does. We note that no subset B can
satisfy spBq “ T2 , as otherwise AzB has the same sum. Since rN s can be partitioned into N 2´1 pairs
ta, bu with a` b “ T , along with the element T2 , it follows that the element T2 and t´1
2 pairs are missing,
thereby establishing the lemma.

Proof for N “ 2n ` 3. According to Lemma 5.2, the three missing numbers in the sumset-distinct set
modulo N form an arithmetic progression of length 3. We first argue that the common difference of
these three numbers is relatively prime to N.
Claim 5.3. Let A be a sumset-distinct set modulo N “ 2n ` 3, and let x, y, z be the three missing
elements in SpAq such that y ` z “ 2x. Then we have gcdpx ´ y, N q “ 1.

Proof of claim. Suppose to the contrary that there is a prime divisor p | gcdpx´y, N q. Since N “ 2n `3
is not divisible by 2 or 3, we have p ą 3. Furthermore, all three elements x, y, and z are congruent
modulo p. Thus, among the residue classes modulo p, SpAq contains Np ´ 3 elements from the one
containing x, and exactly Np elements from every other residue class. Let A “ ta1 , a2 , . . . , an u and
Ai “ ta1 , a2 , . . . , ai u. Then we have SpAi`1 q “ SpAi q Y pSpAi q ` ai`1 q . Let the number of elements in
SpAi q that are congruent to j modulo p be denoted as xi,j . By definition, we have xn,j ” Np pmod 3q
for every j P rps. Using induction, we shall prove that for a fixed i P rns, xi,j pmod 3q is the same for
every j P rps and not equal to zero. This suffices to conclude the proof as some of the x1,j values are
clearly equal to zero, a contradiction.
Assuming the statement is true for i ` 1 to n, we will prove that it is also true for i. Note first that
as SpAi`1 q “ SpAi q Y pSpAi q ` ai`1 q, by definition, we have

xi`1,j “ xi,j ` xi,j´ai`1 ,

where the index j ´ ai`1 is considered modulo p as well. It is important to note that, by Lemma 5.1,
p ∤ ai`1 . So we can consider these equations with variables xi,k¨ai`1 for k P rps and constants xi`1,j ,
and rewrite the system of linear equations as follows:

7
 » fi» x fi »
xi`1,2ai`1
fi
1 1 0 0 ... 0 0 i,ai`1
— 0 — xi,2ai`1 ffi — xi`1,3ai`1 ffi
1 1 0 ... 0 0ffi
ffi— ffi — ffi
.. xi,3ai`1 ffi ...
—
— .. .. ffi— —
“—
ffi
—. . .
— ... . . .
— ffi —
ffi — . . . ffi —xi`1,´ai`1 ffi
ffi— ffi
ffi
– 0 0 0 0 ... 1 1fl –x fl – xi`1,0 fl
ffi
i,´ai`1
1 0 0 0 ... 0 1 x x
i,0 i`1,ai`1

The coefficient matrix in this case is a p ˆ p-matrix. Since p is odd, the determinant of the matrix
Z
is 2. Consequently, in 3Z , there exists a unique solution. However, since all the constants xi`1,j , j P rps,
are the same by the induction hypothesis, this implies that all the variables xi,j are the same and equal
to 2xi`1,j modulo 3. 

Let ∆ be the common difference of x, y, and z, which is coprime with N according to Claim 5.3.
Therefore, Sp∆´1 Aq contains all elements in rN s except for three consecutive integers q ` 1, q ` 2, and
q`3. All three integers are nonzero (since 0 P Sp∆´1ř Aq) and thus 2n ą q ě 0. Let us consider the binary
n´1
representation bn´1 bn´2 . . . b0 of q, denoted as q “ i“0 bi 2i . Now, we define B “ tp´1qbi `1 2i u, where
i ranges from 0 to n ´ 1. It follows that SpBq “ tq ` 4 ´ N, q ` 5 ´ N, . . . , qu “ Sp∆´1 Aq “ ∆´1 SpAq.
By Theorem 2.3, we can conclude that the set ∆´1 A is derived from a set B by possibly switching
the signs of some elements, and one can even conclude that B “ ∆´1 A, i.e. A “ ∆B.

6 Sumset-distinct sets modulo N “ 2n ` 2

Proof of Theorem 1.7. If A contains N {2, then AztN {2u must be sumset-distinct modulo N {2. There-
fore, it is sufficient to prove the characterization assuming that N {2 R A.
Let A be a sumset-distinct set modulo N which does not contain N {2. Let s1 and s2 be the two
missing elements in SpAq, which are the elements in rN s that cannot be expressed as a sum of a subset
of A. Suppose A “ ta1 , . . . , an u.
Claim 6.1. s1 ´ s2 ı N {2 pmod N q.
s1 s2 řN ´1 i
Proof of claim. Let gA pxq “ i p1 ` xai q. Then we have gA pωN q ` ωN
ś
` ωN “ i“0 ωN “ 0. Since
s1 s2 s2 ´s1
N {2 R A, gA pωN q is nonzero. It follows that ωN ` ωN is also nonzero, which implies that ωN ­“ ´1.
Therefore, s1 ´ s2 ı N {2 pmod N q, as claimed. 

Since 2n ą 2n´1 ` 1, SpAq contains odd elements. This implies that A contains at least one odd
element. By induction, we can conclude that SpAq contains an equal number of odd and even sums.
Consequently, s1 and s2 have different parities. Next, we will prove that A contains exactly one odd
element.
Claim 6.2. A contains precisely one odd element.

Proof of claim. Assume that A contains at least two odd numbers, with a being one of them. Let d “
gcdpa, N q. Since SpAq “ SpAztauqYpSpAztauq`aq, every element modulo d appears an even number of
times. In particular, s1 and s2 have the same residue modulo d. Let Si pAq “ ts P SpAq | s ” i pmod dqu
and let Si pAztauq be defined analogously. Since s P SpAztauq if and only if s ` a P a ` SpAztauq,
N
for every i ı s1 pmod dq, Si pAztauq contains 2d elements, all of which have the same parity. Since
d “ gcdpa, N q and the two elements s1 and s2 have the same residue modulo d but different parities,
we can assume that s2 “ s1 ´ p2r ´ 1qa pmod N q, for some r P rN {2ds. Then,

Ss1 pAztauq “ ts1 ´ 2ta | r ą t ě 1u Y ts1 ´ p2t ` 1qa | N {2d ą t ě ru pmod N q,

N
as s P SpAztauq if and only if s ` a P a ` SpAztauq. The set Ss1 pAztauq contains 2d ´ 1 elements, and
since s1 ´s2 ­“ N {2 pmod N q, we have r ‰ pN {2d`1q{2. This implies that the number of even and odd

8
numbers in this set is not equal. Nevertheless, since Aztau has at least one odd number by assumption,
SpAztauq contains the same number of odd as even numbers, which leads to a contradiction. 

Denoting the single odd element of A as a, we observe that Aztau consists entirely of even numbers.
As A is sumset-distinct modulo N , the same applies to Aztau. Hence, pAztauq{2 is sumset-distinct
modulo N {2, and the conclusion follows.

7 Concluding remarks
In this paper, we introduced a modular setting of the Erdős distinct subset sums conjecture, and
proved it asymptotically. One may also investigate if a similar result holds for sets that are almost
sumset-distinct modulo N , i.e. sets having at least p1 ´ op1qq2n different subset sums.
We also characterise the sumset-distinct sets modulo N “ 2n ` t of cardinality n, for 0 ď t ď 3.
For larger t, we suggest the following problems.

• For every t ě 1, is the number of sumset-distinct sets modulo N “ 2n `t of cardinality n bounded

by Ot pN 2`v2 ptq q?

• Given t ě 1, let n be sufficiently large. If t is odd, is every sumset-distinct set modulo N “ 2n ` t

of cardinality n a ‘perturbation’ of the powers of 2?

• If t is even and n sufficiently large, is it true that every sumset-distinct set modulo N “ 2n ` t
of cardinality n either contains N {2, or contains exactly one odd number, or is a ‘perturbation’
of the powers of 2?

Here are two ways to construct perturbations of the powers of 2.

ři
• A “ t20 ` p0 , 21 ` p1 , . . . , 2n´1 ` pn´1 u, with pi ě 0, pn´1 ă t{2 and pi`1 ě j“0 pj for all i ě 0,

• A “ ta0 , a1 , . . . , ai´1 , 2i , 2i`1 , . . . , 2n´2 , 2n´1 u with v2 paj q “ j for every 0 ď j ď i ´ 1.

We remark that the condition that n is sufficiently large is necessary. For N “ 25 ` 4, the set
t1, 6, 11, 13, 15u is sumset-distinct modulo N , but it is not of the above form.
In Appendix A it is additionally proven that many of the constructions of the first kind belong to
different equivalence classes. It might be tempting to guess that for odd t and n large enough, every
sumset-distinct set modulo N “ 2n ` t is in the equivalence class of a set for which the sum of elements
is bounded by N . This turns out to be false, as explained in Appendix B.

Acknowledgement
The authors thank Rob Morris for helpful discussions at the early stage. The first author also thanks
Tijs Buggenhout, Lei Yu, Noah Kravitz and Federico Glaudo for discussions.

References
[1] I. Aliev. Siegel’s lemma and sum-distinct sets. Discrete Comput. Geom., 39(1-3):59–66, 2008.

[2] N. Alon and J. H. Spencer. The probabilistic method. Wiley Series in Discrete Mathematics and
Optimization. John Wiley & Sons, Inc., Hoboken, NJ, fourth edition, 2016.

[3] M. D. Atkinson, A. Negro, and N. Santoro. Sums of lexicographically ordered sets. Discrete Math.,
80(2):115–122, 1990.

9
 [4] J. Bae. On subset-sum-distinct sequences. In Analytic number theory, Vol. 1 (Allerton Park, IL,
1995), volume 138 of Progr. Math., pages 31–37. Birkhäuser Boston, Boston, MA, 1996.

[5] T. Bohman. A construction for sets of integers with distinct subset sums. Electron. J. Combin.,
5:Research Paper 3, 14, 1998.

[6] J. H. Conway and R. K. Guy. Sets of natural numbers with distinct subset sums. Notices Amer.
Math. Soc, 15:345, 1968.

[7] Q. Dubroff, J. Fox, and M. W. Xu. A note on the Erdős distinct subset sums problem. SIAM J.
Discrete Math., 35(1):322–324, 2021.

[8] 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(1):89–94, 1986.

[9] P. Erdős. Some problems in additive number theory. Amer. Math. Monthly, 77:619–621, 1970.

[10] P. Erdős. Problems and results on extremal problems in number theory, geometry, and combi-
natorics. In Proceedings of the 7th Fischland Colloquium, I (Wustrow, 1988), number 38, pages
6–14, 1989.

[11] P. Erdös. Problems and results in additive number theory. In Colloque sur la Théorie des Nombres,
Bruxelles, 1955, pages 127–137. Georges Thone, Liège; Masson & Cie, Paris, 1956.

[12] F. Glaudo and N. Kravitz. Reconstructing a set from its subset sums: 2-torsion-free groups. arXiv
e-prints, page arXiv:2305.11062, May 2023.

[13] R. K. Guy. Sets of integers whose subsets have distinct sums. In Theory and practice of com-
binatorics, volume 60 of North-Holland Math. Stud., pages 141–154. North-Holland, Amsterdam,
1982.

[14] W. F. Lunnon. Integer sets with distinct subset-sums. Math. Comp., 50(181):297–320, 1988.

[15] R. Maltby. Bigger and better subset-sum-distinct sets. Mathematika, 44(1):56–60, 1997.

[16] S. Steinerberger. Some Remarks on the Erdős Distinct Subset Sums Problem. arXiv e-prints,
page arXiv:2208.12182, Aug. 2022.

Appendix

A Perturbations of powers of 2 that are sumset-distinct

We start with two families of sets that are sumset-distinct modulo 2n ` t. Additionally, we can prove
that many of these sets belong to distinct equivalence classes.
Proposition A.1. Let r ě 1 be a fixed value, and set t “ 2r ` 3. Let us consider a set A “
t20 ` p0 , 21 ` p1 , . . . , 2n´1 ` pn´1 u, which is a sumset-distinct
ři sets modulo 2n ` t. This holds for any
choice of pi ě 0 that satisfies the condition pi`1 ě j“0 pj for all i ě 0, and r ě pn´1 . For sufficiently
large n, each set of this form belongs to a distinct equivalence class.
řn´1
Proof. Note that the given conditions imply that t ą 2r ě j“0 pj . It is now sufficient to observe that
n
the sum of all elements is bounded by 2 ´ 1ř` t, and no two different subsets can have equal sums.
This is a consequence of the fact that ai`1 ą ij“0 aj holds for every i ě 0, where aj “ 2j ` pj .
Next, we proceed to prove that two distinct sets of the given form, A “ ta1 , a2 . . . , an u “ t20 `
p0 , 21 ` p1 , . . . , 2n´1 ` pn´1 u and B “ t20 ` q0 , 21 ` q1 , . . . , 2n´1 ` qn´1 u, belong to different equivalence

10
classes when n ě t ą 2pr ` 1q. Assume, without loss of generality, that the number of pi values equal
to zero is at least the number of qi values equal to zero. Let j0 be the smallest index for which qj0 ą 0.
řn´2
Since r ě pn´1 ě j“0 pj , there are at most r ` 1 occurrences of zero among the pi values. Since
n ą 2pr ` 1q, we can conclude that more than half of the pi values(and similarly for qi ) are zeros.
Therefore, we have j0 ą n2 ą r ` 1. If A and B belong to the same equivalence class, there must
exist a value λ P B such that tλa1 , ˘λa2 , . . . , ˘λan u is equal to B. When λ “ 2s for some value of
1 ď s ď j0 ´ 1, it would imply that 2j0 P B, leading to a contradiction. Therefore, we must have
λ “ 2`j ` qj for some j ě j0 . Now, considering 2n´j´1 λ “ 2n´1 ` 2n´j´1 qj , we need it to be equal
to ˘ 2n´1 ` qn´1 pmod N q. The latter implies that 2n´j λ “ ´ p2n ` 2qn´1 q pmod N q, specifically
˘

equal to t ´ 2qn´1 . However, this value is an odd number that lies strictly between 1 and 2r`2 , and
therefore, it does not belong to B.

For t P t1, 3, 5, 7u, it appears that the examples in Proposition A.1 generate all sumset-distinct
sets modulo 2n ` t. The number of examples, and therefore the number of equivalence classes in
Proposition A.1, is exponential in t (related with https://oeis.org/A131205). For larger odd values
of t, there exist alternative constructions as well.

Example A.2. A set A “ ta0 , a1 , . . . , ař i i`1 , . . . , 2n´2 , 2n´1 u is sumset-distinct if v pa q “ j for

i´1 , 2 , 2 2 j
n i i´1
every 0 ď j ď i ´ 1 and N ą 2 ´ 2 ` j“0 aj .

Note that these simple examples are all sumset-distinct in the original scenario and have a sum
smaller than N.

B Sumset-disctinct sets could have large total sum

Our exact results for N “ 2n ` t, 0 ď t ď 3, and the plausible characterization for t P t5, 7u, suggest
that for every odd t ą 0, every sumset-distinct set modulo N “ 2n ` t could be equivalent to a set
where the sum of elements is bounded by N . This turns out to be false for larger values of t.

Proposition B.1. For n ě 8 and N “ 2n ` 9, there exists an equivalence class of sumset-distinct sets
modulo N “ 2n ` 9 that does not have a representative where the sum of the elements is smaller than
N.

Proof. We will prove the validity of the statement for the sets t3, 6, 12, 24, 24 , 25 , . . . , 2n´2 , 2n´1 u „
t1, 2, 4, 8, . . . , 2n´5 , 5¨2n´4 `3, 3¨2n´3 `3, 2n´2 `3, 2n´1 `3u „ t3, 4, 6, 8, 16, . . . , 2n´3 , 2n´2 `3, 2n´1 `3u.
First, we will prove that the sets in the class are sumset-distinct modulo N “ 2n ` 9.
Claim B.2. The set A “ t3, 6, 12, 24, 24 , 25 , . . . , 2n´2 , 2n´1 u is sumset-distinct modulo N “ 2n ` 9.

Proof of claim. Note that spAq “ 2n ` 29. The numbers have different 2-adic valuations, thus the two
subsets with equal sums should have sum x and 2n ` 9 ` x. Since x ď 10, it must be a multiple of
3. The elements not involved, summing to 20 ´ 2x, should be multiples of 3, or equal to 16. However,
since 20 ı 0, 16 pmod 3q, this is impossible. 

Next, we prove that no representative of this equivalence class has a sumset(sum of elements)
smaller than N.
Claim B.3. The set A “ t1, 2, 4, 8, . . . , 2n´5 , 5 ¨ 2n´4 ` 3, 3 ¨ 2n´3 ` 3, 2n´2 ` 3, 2n´1 ` 3u has no
representative whose elements sum to less than N.

Proof of claim. Recall respxq “ mintx pmod N q,řN ´ px pmod Nřqqu, where ` x˘ pmod N q. We need
n´5
to prove that for every 0 ă λ ă N2 , we have i“0 resp2i λq ` 4i“1 res 23i λ ě N. According to
Proposition 3.1, we only need to consider values of λ for which none of the elements belong to the

11
interval N3 , 2n´1 ´ 1 . For λ ě 17, we know that respλ2i q ě N3 for some 0 ď i ď n ´ 5. Therefore, we
“ ‰

can focus on the cases where either 2n´1 ď λ2i ď 2n´1 ` 9 or λ ď 17.
Case 1: 1 ď λ ď 16 These cases can be verified using a computer.
See https://github.com/StijnCambie/ErdosSumSet/blob/main/lambda_till16.sagews.
case 2: λ ě 17, and λ ă N3 is even, or λ ă N6 Note that there exists some i ď n ´ 6 such that
2i λ ă N2 ă 2i`1 λ. Now consider i`1
` j ˘
2 λ “ p2i`1 ´ 1qλ ` pN ´ 2i`1 λq “ N ´ λ. If λ is even,
ř
j“0 res
then res 2 λ “ 2 λ and we are done. If λ ă N6 is odd, then res 23 λ “ N2 ´ 32 λ ą 23 λ as well.
`3 ˘ 3 ` ˘

Case 3: λ is among a few remaining possibilities

These remaining cases can be also verified using a computer.
See https://github.com/StijnCambie/ErdosSumSet/blob/main/largelambda.sagews. 

12