LXXXVII.2 (1998)
On sums of distinct representatives
by
Hui-Qin Cao and Zhi-Wei Sun (Nanjing)
1. Introduction. In combinatorics, for a finite sequence
(1) {A
i}
ni=1of sets, a sequence
(2) {a
i}
ni=1of elements is called a system of distinct representatives (abbreviated to SDR) of (1) if a
1∈ A
1, . . . , a
n∈ A
nand a
i6= a
jfor all 1 ≤ i < j ≤ n. A celebrated theorem of P. Hall [H] says that (1) has an SDR if and only if (3)
[
i∈I
A
i≥ |I| for all I ⊆ {1, . . . , n}.
Clearly (1) has an SDR provided that |A
i| ≥ i for all i = 1, . . . , n, in particular an SDR of (1) exists if |A
1| = . . . = |A
n| ≥ n or 0 < |A
1| < . . .
< |A
n|.
Let G be an additive abelian group and A
1, . . . , A
nits subsets. We as- sociate any SDR (2) of (1) with the sum P
ni=1
a
iand set S({A
i}
ni=1) = S(A
1, . . . , A
n)
(4)
= {a
1+ . . . + a
n: {a
i}
ni=1forms an SDR of {A
i}
ni=1}.
Of course, S(A
1, . . . , A
n) 6= ∅ if and only if (3) holds. A fascinating and challenging problem is to give a sharp lower bound for |S({A
i}
ni=1)| and determine when the bound can be reached.
Let p be a prime. In 1964 P. Erd˝os and H. Heilbronn (cf. [EH] and [G]) conjectured that for each nonempty subset A of Z
p= Z/pZ there are at least min{p, 2|A| − 3} elements of Z
pthat can be written as the sum of two distinct elements of A. With the help of Grassmann spaces this was
1991 Mathematics Subject Classification: Primary 11B75; Secondary 05A05.
The research is supported by the Return-from-abroad Foundations of the Chinese Educational Committee and Nanjing City.
[159]
confirmed by J. A. Dias da Silva and Y. O. Hamidoune [DH] in 1994, in fact they proved the following generalization for n-fold sums: If A ⊆ Z
pthen (5) |n
∧A| ≥ min{p, n|A| − n
2+ 1}
where n
∧A denotes the set of sums of n distinct elements of A, i.e. n
∧A = S(A, . . . , A) with A repeated n times on the right hand side. In 1995 and 1996, N. Alon, M. B. Nathanson and I. Z. Ruzsa [ANR1, ANR2] introduced an ingenious polynomial method and obtained the following result by con- tradiction: Let F be any field of characteristic p and A
1, . . . , A
nits subsets with 0 < |A
1| < . . . < |A
n| < ∞. Then
(6) |S(A
1, . . . , A
n)| ≥ min
p,
X
n i=1|A
i| − n(n + 1)
2 + 1
providing N < p where N = P
ni=1
|A
i| − n(n + 1)/2. We mention that (6) also holds in the case N ≥ p. In fact, let s be the smallest positive integer with P
si=1
(|A
i| − i) > N − p and choose A
01⊆ A
1, . . . , A
0n⊆ A
nso that
|A
0i| = i for i < s, A
0i= A
ifor i > s, and
|A
0s| = s − 1 + X
s i=1(|A
i| − i) − (N − p) ≤ |A
s| − 1.
Then |A
01| < . . . < |A
0n|, N
0= P
ni=1
|A
0i| − n(n + 1)/2 = p − 1 and so
|S(A
1, . . . , A
n)| ≥ |S(A
01, . . . , A
0n)| ≥ N
0+ 1 = p = min{p, N }.
Alon, Nathanson and Ruzsa [ANR2] posed the question when the lower bound in (6) can be reached and considered it to be interesting.
In view of the fundamental theorem on finitely generated abelian groups (cf. [J]), if a finite addition theorem holds in Z then it holds in any torsion- free abelian groups. So, without any loss of generality, we may work within Z.
For a finite subset A of Z, in 1995 Nathanson [N] showed the inequality
|n
∧A| ≥ n|A| − n
2+ 1 and proved that if equality holds then A must be an arithmetic progression providing 2 ≤ n < |A| − 2. The same result was independently obtained by Y. Bilu [B].
Let A
1, . . . , A
nbe finite subsets of Z with 0 < |A
1| < . . . < |A
n|. Take a sufficiently large prime p greater than P
ni=1
|A
i| − n(n + 1)/2 and the largest element of S(A
1, . . . , A
n). Applying the Alon–Nathanson–Ruzsa re- sult stated above, we have the inequality
(7) |S(A
1, . . . , A
n)| ≥ X
n i=1|A
i| − n(n + 1)
2 + 1 =
X
n i=1(|A
i| − i) + 1.
In this paper we will make a new approach to sums of distinct repre-
sentatives. The method allows us to give a somewhat constructive proof of
(7) provided that A
1, . . . , A
nare finite nonempty subsets of Z with distinct
cardinalities. Furthermore we are able to make key progress in the equality case.
Let us first look at two examples.
Example 1. Let a ∈ Z and d ∈ Z \ {0}. Let k ≥ n ≥ 1, A = {a + jd : j = 0, 1, . . . , k − 1}, and A
1, . . . , A
nbe subsets of A with |A
i| = k − n + i for every i = 1, . . . , n. Obviously S(A
1, . . . , A
n) ⊆ n
∧A. If S ⊆ A and |S| = n, then for each i = 1, . . . , n at least i elements of S lie in A
isince |S \ A
i| ≤
|A \ A
i| = n − i, therefore we can write S in the form {a
1, . . . , a
n} where a
1∈ A
1, . . . , a
n∈ A
n. So n
∧A ⊆ S(A
1, . . . , A
n). Let X = {j
1+ . . . + j
n: 0 ≤ j
1< . . . < j
n< k}. For each j = 0, 1, . . . , n(k − n), there exist 0 ≤ u < n and 0 ≤ v ≤ k − n such that j = u(k − n) + v, hence
n(n − 1) 2 + j =
X
n i=1(i − 1) + u(k − n) + v
= X
0<i<n−u
(i − 1) + (n − u − 1 + v) + X
n−u<i≤n
(k − n + i − 1) belongs to X. Thus {n(n − 1)/2 + j : 0 ≤ j ≤ n(k − n)} ⊆ X. Apparently the least and the largest elements of X are 0 + 1 + . . . + (n − 1) = n(n − 1)/2 and (k − n) + . . . + (k − 1) = n(n − 1)/2 + n(k − n) respectively. So, by the above
S(A
1, . . . , A
n) = n
∧A = n X
ni=1
(a + j
id) : 0 ≤ j
1< . . . < j
n< k o
= {na + xd : x ∈ X}
=
na +
n(n − 1)
2 + j
d : 0 ≤ j ≤ n(k − n)
and hence
|S(A
1, . . . , A
n)| = |n
∧A| = n(|A| − n) + 1 = X
n i=1|A
i| − n(n + 1) 2 + 1.
Example 2 (cf. [N]). Let a
0, a
1, a
2, a
3∈ Z, a
0< a
1< a
2< a
3and a
3− a
2= a
1− a
0(but a
2− a
1may be different from a
1− a
0). Let A
1= {a
0, a
1, a
2} and A
2= {a
0, a
1, a
2, a
3}. Then
S(A
1, A
2) = {a
0+ a
1, a
0+ a
2, a
0+ a
3= a
1+ a
2, a
1+ a
3, a
2+ a
3}.
Note that in this example |A
1| = 3 < |A
2| = 4 < |S(A
1, A
2)| = 5 =
|A
1| + |A
2| − 2(2 + 1)/2 + 1.
Now we introduce some notations to be used throughout the paper. For a
subset A of Z, −A refers to {−a : a ∈ A}, min A and max A denote the least
and the largest elements of A respectively. If there exist a ∈ Z, d ∈ Z \ {0}
and a positive integer k such that
A = {a + jd : 0 ≤ j < k}, then we call A an arithmetic progression (for short, AP).
In this paper, by a novel method we obtain the following
Theorem. Let A
1, . . . , A
nbe subsets of Z with 0 < |A
1| < . . . <
|A
n| < ∞. Then inequality (7) holds. Moreover , in the equality case we have S
mi=1
A
i= A
mfor every m ∈ M = {1 ≤ j < n : |A
j+1| > |A
j| + 1} ∪ {n}, and A
nforms an AP unless n = 1 or |A
1| ≤ 3.
The result of Nathanson and Bilu stated above actually follows from the Theorem. For i = 1, . . . , n let A
i⊆ A and |A
i| = |A| − (n − i). Obviously 0 < |A
1| < . . . < |A
n| < ∞. It follows from Example 1 and the Theorem that
|n
∧A| = |S(A
1, . . . , A
n)| ≥ X
n i=1|A
i| − n(n + 1)
2 + 1 = n|A| − n
2+ 1.
When |n
∧A| = n|A| − n
2+ 1, we have
|S(A
1, . . . , A
n)| = X
ni=1
|A
i| − n(n + 1) 2 + 1,
hence by the Theorem if 2 ≤ n ≤ |A| − 3 (i.e., n ≥ 2 and |A
1| ≥ 4) then A = A
nis an AP.
In the next section we shall provide two lemmas. The proof of the The- orem will be given in Section 3.
2. Auxiliary results
Lemma 1. Let G be an additive abelian group, and A
1, . . . , A
nits fi- nite subsets. Let r ∈ {1, . . . , n} and suppose that {a
i}
i6=rforms an SDR of {A
i}
i6=r. Then
(i) There exists a J ⊆ {1, . . . , n} containing r such that if J ⊆ I ⊆ {1, . . . , n} then
S
r(I)[
j∈J
A
j⊆ S({A
i}
i∈I), n
i ∈ I \ {r} : a
i∈ [
j∈J
A
jo
= J \ {r}
and hence
S
r(I)[
j∈J
A
j=
[
j∈J
A
j− |J| + 1, where for any subset A of G we let
(8) S
r(I)(A) = n X
i∈I
a
i: a
r∈ A \ {a
i: i ∈ I \ {r}}
o
.
(ii) Let k
r= |A
r| < . . . < k
n= |A
n|. For any J described in (i) and I ⊆ {1, . . . , n} containing J, we have
(9)
S
(I)r[
j∈J
A
j≥ k
r− r + 1,
and equality holds if and only if there exists an l ∈ {r, . . . , n} for which J = {1, . . . , l}, S
lj=1
A
j= A
land k
l= k
r+ (l − r).
P r o o f. (i) Let J be the class of those J ⊆ {1, . . . , n} containing r such that if J ⊆ I ⊆ {1, . . . , n} then for each j ∈ J there exists a one- to-one mapping σ
I,j: I \ {r} → I \ {j} for which a
i∈ A
σI,j(i)for all i ∈ I \ {r}. Obviously J is nonempty (for, {r} belongs to J ) and finite.
Let J be any maximal set in J with respect to the semiorder ⊆, and let J ⊆ I ⊆ {1, . . . , n}.
Set A = S
j∈J
A
j. Apparently J
0= {r} ∪ {i ∈ I \ {r} : a
i∈ A} contains J. Let J
0⊆ I
0⊆ {1, . . . , n}. Since J ∈ J and J ⊆ I
0, for j ∈ J there is a one-to-one mapping σ
I0,j: I
0\ {r} → I
0\ {j} such that a
i∈ A
σI0,j(i)for all i ∈ I
0\ {r}. For j
0∈ J
0\ J, there is a j ∈ J with a
j0∈ A
j. Since J ∈ J and I
00= I
0\ {j
0} ⊇ J, there also exists a one-to-one mapping σ
I00,j: I
00\ {r} → I
00\ {j} such that a
i∈ A
σI00,j(i)for i ∈ I
00\ {r}. Obviously by letting j
0∈ I
0\ {r} correspond to j ∈ I
0\ {j
0} we can extend σ
I00,jto a one-to-one mapping σ
I0,j0: I
0\ {r} → I
0\ {j
0} for which a
i∈ A
σI0,j0(i)for all i ∈ I
0\ {r}. Thus J
0∈ J . As J ⊆ J
0and J is a maximal set in J , we must have J
0= J, i.e., {i ∈ I \ {r} : a
i∈ S
j∈J
A
j} = J \ {r}.
If j ∈ J and x
j∈ A
j\ {a
i: i ∈ I \ {r}}, then x
j+ P
i∈I\{r}
a
i∈ S({A
i}
i∈I) because a
i∈ A
σI,j(i)for i ∈ I \ {r}. So S
r(I)(A) ⊆ S ({A
i}
i∈I).
Note that
|S
r(I)(A)| = |A \ {a
i: i ∈ I \ {r}}| = |A| − |{i ∈ I \ {r} : a
i∈ A}|
= |A| − |J \ {r}| = |A| − |J| + 1.
This proves part (i).
(ii) Let J be as described in (i), A = S
j∈J
A
jand J ⊆ I ⊆ {1, . . . , n}.
If |J| < r, then
|A| − |J| ≥ |A
r| − |J| > k
r− r.
When |J| ≥ r, clearly max J ≥ r and k
max J−k
r= P
r<i≤max J
(k
i−k
i−1) ≥ max J − r, therefore
|A| − |J| ≥ |A
max J| − |J| ≥ k
r+ max J − r − |J| ≥ k
r− r, and |A| − |J| = k
r− r if and only if
A = A
max J, k
max J= k
r+ max J − r, max J = |J|,
i.e.,
J = {1, . . . , |J|} , A = A
|J|, k
|J|= k
r+ |J| − r.
This together with the equality |S
r(I)(A)| = |A| − |J| + 1 yields the second part.
Lemma 2. Let A and B be finite subsets of Z with 4 ≤ k = |A| < l = |B|, A ⊆ B, min A = min B, max A 6= max B and |S(A, B)| = k + l − 2. Then B is an AP.
P r o o f. Let A = {a
1, . . . , a
k} and B = {b
1, . . . , b
l} where a
1< . . . < a
kand b
1< . . . < b
l. Put C = {a
1+ b
2, . . . , a
1+ b
l−1, a
1+ b
l, . . . , a
k+ b
l}.
Clearly C ⊆ S(A, B) and |C| = k + l − 2. As |S(A, B)| = k + l − 2, S(A, B) coincides with C. Since A ⊆ B and a
k6= b
l, for i = 2, . . . , k we may suppose that a
i= b
f (i)where i ≤ f (i) < l. Because
S(A, B) ⊇ {a
1+ b
j: 2 ≤ j < f (i)} ∪ {a
i+ b
j: j 6= f (i)}
∪ {a
j+ b
l: i < j ≤ k}, we have
k + l − 2 = |S(A, B)| ≥ (f (i) − 2) + (l − 1) + (k − i) = k + l + f (i) − i − 3, i.e. a
i∈ {b
i, b
i+1}. Observe that a
3< a
k≤ b
l−1. Since
a
1+ b
l−1< a
2+ b
l−1< a
3+ b
l−1< a
3+ b
l, a
2+ b
l−1= a
1+ b
land a
3+ b
l−1= a
2+ b
l, it follows that
a
2− a
1= b
l− b
l−1= a
3− a
2.
If a
26= b
2, then a
2= b
3, a
3= b
4, b
2− b
1< a
2− b
1= a
3− a
2, a
1+ b
3<
b
2+ a
2< a
3+ b
1= a
1+ b
4; this contradicts the fact a
2+ b
2∈ S(A, B) = C.
So a
2= b
2. As a
2+ b
l−1∈ C we must have a
2+ b
l−1= a
1+ b
l, similarly a
2+ b
l−2= a
1+ b
l−1, . . . , a
2+ b
3= a
1+ b
4. Thus
b
l− b
l−1= . . . = b
4− b
3= a
2− a
1= b
2− b
1= a
3− b
2.
If a
36= b
3, then a
3= b
4and hence b
2= b
2−a
3+b
4= b
3, which is impossible.
So a
3= b
3and B forms an AP.
3. Proof of Theorem. The case n = 1 is trivial. Below we let n ≥ 2 and assume the statement holds for smaller values of n.
Put k
i= |A
i| for i = 1, . . . , n. Set a = min S
ni=1
A
i, I = {1 ≤ i ≤ n : a ∈ A
i}, r = min I and t = max I. For i ∈ I let
A
0i=
A
i\ {a} if i 6= r,
{a} if i = r;
and for i ∈ ¯ I = {1, . . . , n} \ I put A
0i=
A
i\ {a
i} if r < i < t and i 6∈ M , A
iotherwise,
where a
iis an arbitrary element of A
i. Apparently all the A
0iare finite, nonempty and contained in Z, also |S(A
01, . . . , A
0n)| = |S({A
0i}
i6=r)|. Let k
0i= |A
0i| for i = 1, . . . , n. Observe that k
i0< k
0jif 1 ≤ i < j ≤ n and i, j 6= r.
By the induction hypothesis,
|S({A
0i}
i6=r)| ≥ X
i6=r
k
0i− (n − 1)(n − 1 + 1)
2 + 1 > 0.
Suppose that max S({A
0i}
i6=r) = P
i6=r
a
0iwhere {a
0i}
i6=ris an SDR of {A
0i}
i6=r. By Lemma 1 there exists a J ⊆ {1, . . . , n} containing r for which
J \ {r} = {i 6= r : a
0i∈ A}, S
r(A) ⊆ S({A
i}
ni=1), |S
r(A)| ≥ k
r− r + 1, where A = S
j∈J
A
jand S
r(A) = { P
ni=1
a
0i: a
0r∈ A \ {a
0i: i 6= r}}. As S(A
1, . . . , A
n) ⊇ S(A
01, . . . , A
0n) ∪ S
r(A) and
max S(A
01, . . . , A
0n) = a + X
i6=r
a
0i= min S
r(A), we have
|S(A
1, . . . , A
n)| ≥ |S(A
01, . . . , A
0n)| + |S
r(A)| − 1
≥ X
i6=r
k
0i− n(n − 1)
2 + 1 + (k
r− r + 1) − 1
≥ X
i6=r
k
i− (t − r) − n(n − 1)
2 + 1 + k
r− r
= X
n i=1k
i− n(n + 1)
2 + n − t + 1
≥ X
n i=1k
i− n(n + 1) 2 + 1.
From now on we assume that (10) |S(A
1, . . . , A
n)| =
X
n i=1k
i− n(n + 1) 2 + 1.
The above deduction yields (11) |S({A
0i}
i6=r)| = X
i6=r
k
i− (t − r) − n(n − 1)
2 + 1,
(12) r < i < t ⇒ i 6∈ ¯ I ∩ M,
|S
r(A)| = k
r− r + 1, (13)
t = n.
(14)
By (13) and Lemma 1 there is an l ∈ {r, . . . , n} for which J = {1, . . . , l}, S
lj=1
A
j= A
l, k
l= k
r+ (l − r) and
(15) {1, . . . , l} \ {r} = J \ {r} = {i 6= r : a
0i∈ A = A
l}.
(14) indicates that a ∈ A
n. Let b = max S
ni=1
A
i. Clearly a 6= b, for otherwise each A
icontains exactly one element, which contradicts the inequality k
1<
k
n. As −b = min S
ni=1
(−A
i) and |S(−A
1, . . . , −A
n)| = |S(A
1, . . . , A
n)| = P
ni=1
|−A
i| − n(n + 1)/2 + 1, similarly we have −b ∈ −A
n. So b ∈ A
0n= A
n\ {a}. Choose the smallest s ≤ n such that b ∈ A
s.
Let m ∈ M . We now show that S
mi=1
A
i= A
m, i.e. A
mcontains both S
mi=1,i6=r
A
iand A
r.
If m = r, then r ∈ M , hence l = r = m and A
m= S
mi=1
A
i⊇ S
mi=1,i6=r
A
i.
Since A
0i= A
ifor all i < r, by (11), (12) and the induction hypothesis, if m < r then S
mi=1
A
i= S
mi=1,i6=r
A
i= A
m.
Assume r < n. Clearly b ∈ {a
0i: i 6= r} (otherwise P
i6=r,n
a
0i+ b ∈ S({A
0i}
i6=r) would be greater than P
i6=r
a
0i= max S({A
0i}
i6=r)). Suppose that b = a
0jwhere j 6= r. In view of (15), b ∈ A
lif and only if j ≤ l. Since b = a
0j∈ A
0j⊆ A
j, we have j ≥ s. If l = s, then b ∈ A
l, s ≤ j ≤ l = s, s = j 6= r.
Now suppose that r < m ≤ n. If m < t = n, then m ∈ I by (12), and k
0m+1− k
0m≥ (k
m+1− 1) − (k
m− 1) > 1. By (11), (12) and the induction hypothesis S
mi=1,i6=r
A
0i= A
0m. If 1 ≤ i < r, then A
i= A
0i⊆ A
0m⊆ A
m; if r < i ≤ m and i ∈ I, then A
i= A
0i∪ {a} ⊆ A
0m∪ {a} = A
m; if r < i ≤ m but i 6∈ I, then |A
i| ≥ k
r+ 1 ≥ 2 and hence for any given x
i∈ A
iby taking a
i∈ A
idifferent from x
iat the beginning we find that x
i∈ A
i\ {a
i} = A
0i⊆ A
0m⊆ A
m. So S
mi=1,i6=r
A
i⊆ A
m.
Since s is the smallest index such that −A
scontains min S
ni=1
(−A
i) =
−b, by analogy S
mi=1,i6=s
(−A
i) ⊆ −A
m. Thus, if r 6= s then −A
r⊆ −A
m, i.e. A
r⊆ A
m. If r = s, then by the above l 6= s = r, also l ≤ m since k
l= k
r+ l − r, therefore A
r⊆ S
lj=1
A
j= A
l⊆ A
m. So S
mi=1
A
i= A
m. Now let us check that A
nis an AP except the case k
1≤ 3.
If r = 1 then min{k
0i: i 6= r} = k
02≥ k
2− 1 ≥ k
1, and if r > 1 then min{k
0i: i 6= r} = k
10= k
1. So min{k
i0: i 6= r} ≥ k
1. Below we assume that k
1≥ 4.
Suppose n > 2. By (11), (12) and the induction hypothesis, if r < n then
A
0n= A
n\ {a} is an AP. Similarly, if s < n then −A
n\ {−b} is an AP and
hence so is A
n\ {b}. Thus, if r < n and s < n then A
n= {a} ∪ (A
n\ {a}) =
(A
n\ {b}) ∪ {b} forms an AP. (Note that |A
n| = k
n> k
1≥ 4.)
Now consider the case r = s = 1 < n = 2. By the above l = 2 (since l 6= s = 1) and k
2= k
1+ 1. Let a = b
1< . . . < b
k2= b be all the elements of A
2. If 1 ≤ i < j ≤ k
2, then either b
ior b
jbelongs to A
1because |A
2\A
1| = 1, therefore b
i+ b
j∈ S(A
1, A
2). So
S(A
1, A
2) ⊇ S(A
2\ {b}, A
2)
⊇ C = {b
1+ b
2, . . . , b
1+ b
k2, b
2+ b
k2, . . . , b
k2−1+ b
k2}.
As |S(A
1, A
2)| = k
1+ k
2− 2 = 2k
2− 3 = |C|, we have S(A
1, A
2) = C = S(A
2\{b}, A
2). Clearly |A
2\{b}| = k
2−1 = k
1≥ 4, min(A
2\{b}) = min A
2and max(A
2\ {b}) 6= max A
2. Applying Lemma 2 we find that A
n= A
2forms an AP.
With respect to the case r = n we make the following remarks:
(i) Since A
n= S
nj=1
A
j, we have A
n−1⊂ A
n. If n > 2, then by (11), (12) and the induction hypothesis A
0n−1= S
n−1i=1
A
0iforms an AP, i.e. S
n−1i=1
A
i= A
n−1is an AP. Set
A
−n= {x ∈ A
n: x ≤ max A
n−1} and A
+n= {x ∈ A
n: min A
n−1≤ x}.
Whether n = 2 or n > 2 we always have S
n−1i=1
A
i= A
n−1⊆ A
−n∩ A
+nand hence |A
−n∩ A
+n| ≥ k
n−1≥ k
1≥ 4. Among A
1, . . . , A
n−1, A
−n, the index r
−of the first one containing min( S
n−1i=1
A
i∪ A
−n) = a is identical with r while the index s
−of the first one containing max( S
n−1i=1
A
i∪ A
−n) = max A
n−1is less than n. Similarly, among A
1, . . . , A
n−1, A
+n, the index r
+of the first one containing min( S
n−1i=1
A
i∪ A
+n) = min A
n−1is less than n while the index s
+of the first one containing max( S
n−1i=1
A
i∪ A
+n) = b is equal to s.
(ii) Suppose that A
−n6= A
n−1. Then |A
−n| > k
n−1> . . . > k
1. According to the previous reasoning,
|S(A
1, . . . , A
n−1, A
−n)| ≥
n−1
X
i=1
k
i+ |A
−n| − n(n + 1)
2 + 1 > 0.
Observe that
max S(A
1, . . . , A
n−1, A
−n) ≤ max S(A
01, . . . , A
0n−1) + max A
n−1< max S({a
01}, . . . , {a
0n−1}, A
n\ A
−n), and that |S({a
01}, . . . , {a
0n−1}, A
n\ A
−n)| = |A
n\ A
−n|. So
|S(A
1, . . . , A
n)| ≥ |S(A
1, . . . , A
n−1, A
−n)| + |S({a
01}, . . . , {a
0n−1}, A
n\ A
−n)|
≥
n−1
X
i=1
k
i+ |A
−n| − n(n + 1)
2 + 1 + (k
n− |A
−n|)
= X
n i=1k
i− n(n + 1)
2 + 1.
Since (10) holds, we must have (16) |S(A
1, . . . , A
n−1, A
−n)| =
n−1
X
i=1
k
i+ |A
−n| − n(n + 1) 2 + 1.
(iii) By analogy, when −A
+n= {−x ∈ −A
n: −x ≤ max(−A
n−1)} 6=
−A
n−1(i.e. A
+n6= A
n−1), we have
|S(A
1, . . . , A
n−1, A
+n)| = |S(−A
1, . . . , −A
n−1, −A
+n)|
(17)
=
n−1
X
i=1