PROBLEMS IN ADDITIVE NUMBER THEORY, I

arXiv:math/0604340v2 [math.NT] 6 Aug 2006

MELVYN B. NATHANSON

Abstract. Talk at the Atelier en combinatoire additive (Workshop on Arith-

metic Combinatorics) at the Centre de recherches mathématiques at the Uni-
versité de Montréal on April 8, 2006.

Definition 1. A problem is a problem I cannot solve, not necessarily an unsolved

problem.

1. Sums and differences

For any set A of integers, we define the sumset
A + A = {a + a′ : a, a′ ∈ A}
and the difference set
A − A = {a − a′ : a, a′ ∈ A}.
In this section we consider finite sets of integers, and the relative sizes of their
sumsets and difference sets. If A is a finite set of integers and x, y ∈ Z, then the
translation of A by x is the set x + A = {x + a : a ∈ A} and the dilation of A by y
is y ∗ A = {ya : a ∈ A}. We have
(x + A) + (x + A) = 2x + 2A
and
(x + A) − (x + A) = A − A.
Similarly,
y ∗ A + y ∗ A = y ∗ (A + A)
and
y ∗ A − y ∗ A = y ∗ (A − A).
It follows that
|(x + y ∗ A) + (x + y ∗ A)| = |2A|
and
|(x + y ∗ A) − (x + y ∗ A)| = |A − A|
so the cardinalities of the sum and difference sets of a finite set of integers are
invariant under affine transformations of the set.

2000 Mathematics Subject Classification. 11B05, 11B13, 11B34.

Key words and phrases. Additive number theory, sumsets, difference sets, representation
functions.
Supported in part by grants from the NSA Mathematical Sciences Program and the PSC-
CUNY Research Award Program.
1
2 MELVYN B. NATHANSON

The set A is symmetric with respect to the integer z if A = z −A or, equivalently,

if a ∈ A if and only if z − a ∈ A. For example, the set {4, 6, 7, 9} is symmetric with
z = 13. If A is symmetric, then
A + A = A + (z − A) = z + (A − A)
and so |A + A| = |A − A|.
If A = {a, b, c} with a < b < c and a + c 6= 2b, then
|A + A| = 6 < 7 = |A − A|.
If
A = {0, 2, 3, 4, 7}
then
A + A = [0, 14] \ {1, 12, 13}
A − A = [−7, 7] \ {−6, 6}
and
|A + A| = 12 < 13 = |A + A|.
This is the typical situation. Since
2+7=7+2
but
2 − 7 6= 7 − 2
it is natural to expect that in any finite set of integers there are always at least as
many differences as sums. There had been a conjecture, often ascribed incorrectly
to John Conway1 that asserted that |A + A| ≤ |A − A| for every finite set A of
integers. This conjecture is false, and a counterexample is the set
(1) A = {0, 2, 3, 4, 7, 11, 12, 14},
for which
A + A = [0, 28] \ {1, 20, 27}
A − A = [−14, 14] \ {±6, ±13}
and
|A + A| = 26 > 25 = |A − A|.
On the other hand, the conjecture really should be true, and suggests the first
somewhat philosophical problem.
Problem 1. Why do there exist finite sets A of integers such that |A+A| > |A−A|?
Given the existence of such aberrant sets, we can ask for the smallest one. The
set A in (1) satisfies |A| = 8.
Problem 2. What is
min{|A| : A ⊆ Z and |A + A| > |A − A|}?
1The confusion may be due to the fact that the first published paper on the conjecture, by
John Marica [4], is entitled ”On a conjecture of Conway.” I asked Conway about this at the Logic
Conference in Memory of Stanley Tennenbaum at the CUNY Graduate Center on April 7, 2006.
He said that he had actually found a counterexample to the conjecture, and that this is recorded
in unpublished notes of Croft [2].
 PROBLEMS IN ADDITIVE NUMBER THEORY, I 3

Sets A with the property that |A + A| > |A − A| should have structure. By

structure I do not mean that A contains arithmetic progressions or generalized
arithmetic progressions or even subsets of some polynomial type. There is a signif-
icant part of combinatorial and additive number theory, sometimes called additive
combinatorics, that consists of looking for arithmetic progressions inside sets of in-
tegers, or proving that certain sets can be approximated by generalized arithmetic
progressions. The results are beautiful, deep, and difficult, but it is hard to ignore
the fact that arithmetic progressions are fundamentally boring, and dense or even
relatively dense sets of integers must contain vastly more interesting structures that
we have not yet imagined.
The astronomers are trying to understand the large-scale structure of the uni-
verse. If they found an arithmetic progression or a generalized arithmetic pro-
gression of galaxies, they would be ecstatic, but it would also be obvious to them
that this fascinating and unexpected curiosity is only a small part of the universe,
and they would keep looking for other structures. Since the complexity of sets of
integers is comparable to that of the universe, we should also keep looking.
Problem 3. What is the structure of finite sets satisfying |A + A| > |A − A|?
If A is a finite set of integers and m is a sufficiently large positive integer (for
example, m > 2 max({|a| : a ∈ A}), then the set
( t−1 )
X
At = ai mi : ai ∈ A for i = 0, 1, . . . , t − 1
i=0
has the property that
|At + At | = |A + A|t
and
|At − At | = |A − A|t .
It follows that if |A + A| > |A − A|, then |At + At | > |At − At | and, moreover,
 t
|At + At | |A + A|
lim = lim = ∞.
t→∞ |At − At | t→∞ |A − A|

The sequence of sets {At }∞

t=1 is the standard parametrized family of sets with more
sums than differences.
Problem 4. Are there other parametrized families of sets satisfying |A + A| >
|A − A|?
Even though there exist sets A that have more sums than differences, such sets
should be rare, and it must be true with the right way of counting that the vast
majority of sets satisfies |A − A| > |A + A|.
Problem 5. Let f (n) denote the number sets A ⊆ [0, n − 1] such that |A − A| <
|A + A|, and let f (n, k) denote the number of such sets A ⊆ [0, n − 1] with |A| = k.
Compute
f (n)
lim
n→∞ 2n
and
f (n, k)
lim n

.
n→∞
k
4 MELVYN B. NATHANSON

The functions f (n) and f (n, k) are not necessarily the best functions to count
finite sets of nonnegative integers with respect to sums and differences.
Problem 6. Prove that |A − A| > |A + A| for almost all sets A with respect to
other appropriate counting functions.

2. Binary linear forms

The problem of sums and differences can be considered a special case of a more
general problem about binary linear forms
f (x, y) = ux + vy
where u and v are nonzero integers. For every finite set A of integers, let
f (A) = {f (a, a′ ) : a, a′ ∈ A}.
We are interested in the cardinality of the sets f (A). For example, the sets associ-
ated to the binary linear forms
s(x, y) = x + y
and
d(x, y) = x − y
are the sumset s(A) = A + A and the difference set d(A) = A − A.
We begin by putting our functions into a standard form. Let f0 (x, y) = u0 x+v0 y.
If d = (u0 , v0 ) > 1, let f1 (x, y) = (u0 /d)x + (v0 /d)y = u1 x + v1 y. If |u1 | < |v1 |, let
f2 (x, y) = v1 x + u1 y = u2 x + v2 y. If u2 < 0, let f3 (x, y) = −u2 x − v2 y = u3 x + v3 y.
Then
|f0 (A)| = |f1 (A)| = |f2 (A)| = |f3 (A)|
for every finite set A. To every binary linear form there is constructed in this way
a unique normalized binary linear form f (x, y) = ux + vy such that
u ≥ |v| ≥ 1 and (u, v) = 1.
The natural question is: If f (x, y) and g(x, y) are two distinct normalized binary
linear forms, do there exist finite sets A and B of integers such that |f (A)| > |g(A)|
and |f (B)| < |g(B)|, and, if so, is there an algorithm to construct A and B?
Brooke Orosz gave constructive solutions to this problem in some important
cases. For example, she proved the following: Let u > v ≥ 1 and (u, v) = 1, and
consider the normalized binary linear forms
f (x, y) = ux + vy
and
g(x, y) = ux − vy.
For u ≥ 3, the sets
A = {0, u2 − v 2 , u2 , u2 + uv}
and
B = {0, u2 − uv, u2 − v 2 , u2 }
satisfy the inequality
|f (A)| = 14 > 13 = |g(A)|
and
|f (B)| = 13 < 14 = |g(B)|.
 PROBLEMS IN ADDITIVE NUMBER THEORY, I 5

For u = 2, we have f (x, y) = 2x + y and g(x, y) = 2x − y. The sets

A = {0, 3, 4, 6}
and
B = {0, 4, 6, 7}
satisfy the inequality
|f (A)| = 13 > 12 = |g(A)|
and
|f (B)| = 13 < 14 = |g(B)|.
The problem of pairs of binary linear forms has been completely solved by
Nathanson, O’Bryant, Orosz, Ruzsa, and Silva [11].
Theorem 1. Let f (x, y) and g(x, y) be distinct normalized binary linear forms.
There exist finite sets A, B, C with |C| ≥ 2 such that
|f (A)| > |g(A)|
|f (B)| < |g(B)|
|f (C)| = |g(C)|.
Problem 7. Let f (x, y) and g(x, y) be distinct normalized binary linear forms.
Determine if |f (A)| > |g(A)| for most or for almost all finite sets of integers.
These results should be extended to linear forms in three or more variables.
Problem 8. Let f (x1 , . . . , xn ) = u1 x1 + · · · + un xn and g(x1 , . . . , xn ) = v1 x1 +
· · · + vn xn be linear forms with integer coefficients. Does there exist a finite set A
of integers such that |f (A)| > |g(A)|?

3. Polynomials over finite sets of integers and congruence classes

An integer-valued function is a function f (x1 , x2 , . . . , xn ) such that if x1 , x2 , . . . , xn ∈
Z, then f (x1 , x2 , . . . , xn ) ∈ Z. The binomial polynomial
 
x x(x − 1)(x − 2) · · · (x − k + 1)
=
k k!
is integer-valued, and every integer-valued
polynomial is a linear combination with
integer coefficients of the polynomials xk . For any set A ⊆ Z, we define
f (A) = {f (a1 , a2 , . . . , an ) : ai ∈ A for i = 1, 2, . . . , n} ⊆ Z.
Problem 9. Let f (x1 , . . . , xn ) and g(x1 , . . . , xn ) be integer-valued polynomials.
Determine if there exist finite sets A, B, C of positive integers with |C| ≥ 2 such
that
|f (A)| > |g(A)|
|f (B)| < |g(B)|
|f (C)| = |g(C)|.
There is a strong form of Problem 9.
Problem 10. Let f (x1 , . . . , xn ) and g(x1 , . . . , xn ) be integer-valued polynomials.
Does there exist a sequence {Ai }∞ i=1 of finite sets of integers such that
|f (Ai )|
lim = ∞?
i→∞ |g(Ai )|
6 MELVYN B. NATHANSON

There is also the analogous modular problem. For every polynomial f (x1 , x2 , . . . , xn )
with integer coefficients and for every set A ⊆ Z/mZ, we define
f (A) = {f (a1 , a2 , . . . , an ) : ai ∈ A for i = 1, 2, . . . , n} ⊆ Z/mZ.
Problem 11. Let f (x1 , . . . , xn ) and g(x1 , . . . , xn ) be polynomials with integer co-
efficients and let m ≥ 2. Do there exist sets A, B, C ⊆ Z/mZ with |C| > 1 such
that
|f (A)| > |g(A)|
|f (B)| < |g(B)|
|f (C)| = |g(C)|.
Problem 12. Let f (x1 , . . . , xn ) and g(x1 , . . . , xn ) be polynomials with integer co-
efficients. Let M (f, g) denote the set of all integers m ≥ 2 such that there exists
a finite set A of congruence classes modulo m such that |f (A)| > |g(A)|. Compute
M (f, g).
Note that if there exists a finite set A of integers with |f (A)| > |g(A)|, then
M (f, g) contains all sufficiently large integers.

4. Representation functions of asymptotic bases

A central topic in additive number theory is the study of bases for the integers
and for arbitrary abelian groups and semigroups, written additively. The set A
is called an additive basis of order h for the set X if every element of X can be
written as the sum of exactly h not necessarily distinct elements of A. The set A is
called an asymptotic basis of order h for the set X if all but at most finitely many
elements of X can be written as the sum of h not necessarily distinct elements of
A. The classical bases in additive number theory for the set N0 of nonnegative
integers are the squares (Lagrange’s theorem), the cubes (Wieferich’s theorem),
the k-th powers (Waring’s problem and Hilbert’s theorem), the polygonal numbers
(Cauchy’s theorem), and the primes (Shnirel’man’s theorem for sufficently large
integers). These classical results in additive number theory are in Nathanson [6, 7].
The representation function for a set A is the function rA,h (n) that counts the
number of representations of n as the sum of h elements of A. More precisely,
rA,h (n) is the number of h-tuples (a1 , a2 , . . . , ah ) ∈ Ah such that
a1 + a2 + · · · + ah = n
and
a1 ≤ a2 ≤ · · · ≤ ah .
The set A is an asymptotic basis of order h if rA,h (n) ≥ 1 for all but finitely many
elements of X. A fundamental unsolved problem in additive number theory is the
classification problem for representation functions.
Problem 13. Let f : N0 → N0 ∪ {∞} be a function. Find necessary and sufficient
conditions on f in order that there exists a set A in N0 such that rA,h (n) = f (n)
for all n ∈ N0 .
A special case is the classification problem for representation functions for as-
ymptotic bases for the nonnegative integers.
 PROBLEMS IN ADDITIVE NUMBER THEORY, I 7

Problem 14. Let F0 (N0 ) denote the set of all functions f : N0 → N0 ∪ {∞} such
that f (n) = 0 for only finitely many nonnegative integers n. For what functions
f ∈ F0 (N0 ) does there exist a set A ⊆ N0 such that rA,h (n) = f (n) for all n ∈ N0 ?
Nathanson [8] introduced these problems, and recently began to study the rep-
resentation functions of asymptotic bases for the set Z of integers. He proved [10]
that if f : Z → N0 ∪ {∞} is any function such that f (n) = 0 for only finitely
many integers n, then there exists a set A in Z such that rA,h (n) = f (n) for all
integers n. Moreover, arbitrarily sparse infinite sets A can be constructed with the
given representation function f . The important new problem is to determine the
maximum density of a set A of integers with given representation function f . For
any set A of integers, we define the counting function
X
A(x) = 1.
a∈A
|a|≤x

Problem 15. Let f : Z → N0 ∪ {∞} be any function such that f (n) = 0 for only
finitely many integers n. Let R(f ) denote the set of all sets A ⊆ Z such that
rA,2 (n) = f (n) for all n ∈ Z.
Compute
sup {α : A ∈ R(f ) and A(x) ≫ xα for all x ≥ x0 } .
Nathanson [9] proved that for any function f there exists a set A ∈ R(f ) with
A(x) ≫ x1/3 .
Cilleruelo and Nathanson [1] recently improved this to
√
2−1+ε
A(x) ≫ x
for any ε > 0.
A related problem is the inverse problem for representation functions. Associated
to a function f : Z → N0 ∪ {∞} can be infinitely many sets A of integers such that
A ∈ R(f ). On the other hand, the semigroup N0 of nonnegative integers is more
rigid than the group Z of integers. Given f : N0 → N0 , there may be a unique set
A ⊆ N0 such that rA,h (n) = f (n) for all n ∈ N0 .
Problem 16. Let f : N0 → N0 be the representation function of a set of integers.
Determine all sets A ⊆ N0 such that rA,h (n) = f (n) for all sufficiently large
integers n.
The problem was first studied by Nathanson [5], and subsequently by Lev [3]
and others. There is an excellent survey of additive representation functions by
Sárközy and Sós [12].

References
[1] J. Cilleruelo and M. B. Nathanson, Dense sets of integers with prescribed representation
fucntions, preprint, 2006.
[2] H. T. Croft, Research problems, Problem 7, Section 6, Mimeographed notes, University of
Cambridge, 1967.
[3] V. F. Lev, Reconstructing integer sets from their representation functions, Electron. J. Combin.
11 (2004), Research Paper 78, 6 pp. (electronic).
[4] J. Marica, On a conjecture of Conway, Canad. Math. Bull. 12 (1969), 233–234.
8 MELVYN B. NATHANSON

[5] M. B. Nathanson, Representation functions of sequences in additive number theory, Proc.

Amer. Math. Soc. 72 (1978), 16–20.
[6] M. B. Nathanson, Additive Number Theory: The Classical Bases, volume 164 of Graduate
Texts in Mathematics, Springer-Verlag, New York, 1996.
[7] M. B. Nathanson, Elementary Methods in Number Theory, volume 195 of Graduate Texts in
Mathematics, Springer-Verlag, New York, 1996.
[8] M. B. Nathanson, Unique representation bases, Acta Arith. 108 (2003), 1–8.
[9] M. B. Nathanson, The inverse problem for representation functions of additive bases, in:
Number Theory: New York Seminar 2003, Springer, New York, 2004, pages 253–262.
[10] M. B. Nathanson, Every function is the representation function of an additive basis for the
integers, Port. Math. (N. S.) 62 (2005), no. 1, 55–72.
[11] M. B. Nathanson, K. O’Bryant, B. Orosz, I. Z. Ruzsa, and M. Silva, Binary linear forms,
preprint, 2006.
[12] A. Sárközy and V. T. Sós, On additive representation functions, in: The mathematics of Paul
Erdős, I, Springer-Verlag, Berlin, 1997, pages 129–150.

Department of Mathematics, Lehman College (CUNY), Bronx, New York 10468

E-mail address: [email protected]