Three proofs of the Cauchy-Davenport Inequality

Following Tao and Vu Additive Combinatorics

Trevor Karn

University of Minnesota Student Number Theory Seminar

April 27, 2020

Algebaic proof

Combinatorial proof

Fourier Analytic Proof

Additive sets

Definition
I An additive set is a pair (A, Z ) where Z is a group and A ⊆ Z
is a finite, nonempty subset of Z
I Given two additive sets A, B we can define the sum set
A + B := {a + b : a ∈ A, b ∈ B}

Example
A = {2, 3}, B = {4}, then A + B = {6, 7}

Nonexample
A = {2, 3}, B = 2Z, then A + B = Z, but B not finite.
Cauchy-Davenport theorem

Theorem (Cauchy, 1813 and Davenport, 1935)

Let p be a prime, and A, B are two additive sets in Z/pZ, then

|A + B| ≥ min(|A| + |B| − 1, p)
Cauchy-Davenport theorem

Example 1
Let A = {1, 3, 5, 9}, B = {2, 3, 4} be additive sets with ambient
group Z/11Z.
Then A + B = {3, 4, 5, 6, 7, 8, 9, 0, 1, 2} so

|A + B| = 10 ≥ min(6, 11) X
Cauchy-Davenport theorem

Example 2
Let A = {1, 3, 5, 7}, B = {2, 4, 6, 8, 10} in Z/11Z.

Then A + B = {3, 5, 7, 9, 0, 2, 4, 6} so

|A + B| = 8 ≥ min(8, 11) X

There are a few things to notice here:

I Compare to last example: |A| is the same, |B| is bigger, but
|A + B| is smaller
I “Consecutive” numbers in A, B have common difference of 2.
I Equality is achieved!
Arithmetic progression

I In Z, we call sequence with a common difference of

consecutive terms an arithmetic progression.
I In the language of additive sets, if we can write

A = a + [0, n) · r = {a, a + r , a + 2r , ..., a + nr }

then we call A an arithmetic progression.

I From last example: A = 1 + [0, 4) · 2
I A well known formula which makes me smile is
n
X n(n + 1)
i=
2
i=1

because of this proof:

Arithmetic progression

n Each line has a value

of n + 1.
n−1
When n is even, add
up for each of
n−2
1, 2, ..., n, but that has
overcounted by double,
··· n+1 n+1 n+1 so n(n+1)
2 .

3 When n is odd, same

idea but add up n + 1
2 (n − 1)-many times
(omitting n+1
2 ) so we
1 (n+1)(n−1)
get 2 + n+1
2 .
Arithmetic progression

I It turns out that there is a similar formula for arithmetic

progressions. If sn is the nth partial sum for an arithmetic
progression,
n(a1 + an )
sn =
2

I Same proof

I {1, 2, ..., n} = 1 + [0, n) · 1 is a special case

Vosper’s theorem

Theorem
If |A|, |B| ≥ 2, |A + B| ≤ p − 2, the Cauchy-Davenport theorem
achieves equality if and only if A, B are both arithmetic
progressions with a common difference.
Proof ∅ of Cauchy-Davenport

I Augstin Louis Cauchy, 1813

I As far as I can tell, Oevours complete d’Augustin Cauchy
starts in 1815.
Proof 1 of Cauchy-Davenport [Dav35, TV10]

Definition
Let A, B additive sets, e ∈ A − B define e-transform A(e) , B(e) as

A(e) := A ∪ (B + e) ⊇ A

B(e) := B ∩ (A − e) ⊆ B

Example
Let A = {1, 3, 5, 9}, B = {2, 3, 4}, 5 = 9 − 4 ∈ A − B

A(5) = {1, 3, 5, 9} ∪ {7, 8, 9} = {1, 3, 5, 7, 8, 9}

B(5) = {2, 3, 4} ∩ {7, 9, 0, 4} = {4}

Proof 1 of Cauchy-Davenport [Dav35, TV10]

I A(5) + B(5) = {5, 7, 9, 0, 1, 2} ⊆ A + B

I |A| + |B| = |A(5) | + |B(5) |.
I These are true in general
I Upshot: e-transformation keeps the total size of the two sets
the same, while making the sum set weakly smaller.
Proof 1 of Cauchy-Davenport [Dav35, TV10]

Lemma
If A, B ⊆ Z are additive sets, n, m ∈ Z then:

|A + nB − mB| = |A| if and only if there is a subgroup G ≤ Z so

that B ⊆ h + G and A is a union of cosets of G .

Proof.
Construct such a G.
Proof 1 of Cauchy-Davenport [Dav35, TV10]

Idea: fix A, see how B acts on A. Induct.

Proof.
Base case: If |B| = 1, then |A + B| = |A| = |A| + |B| − 1 ≤ p. X
Induction step: Suppose we know the claim holds for |B 0 | < |B|.
Suppose ∃e ∈ A − B so |B(e) | < |B|. Then

|A + B| ≥ |A(e) + B(e) |
≥ min(|A(e) | + |B(e) | − 1, p)
= min(|A| + |B| − 1, p).
Proof 1 of Cauchy-Davenport [Dav35, TV10]

Now suppose |B(e) | = |B| ∀e.

I B ∩ (A − e) = B
I B ⊆ (A − e) ⇔ B + e ⊆ A
I Adding anything in A − B makes B + e ⊂ A
I So adding everything yields B + (A − B) ⊆ A
I By lemma, B in coset of subgroup of Z/pZ, A is union of
cosets.
I Only subgroups of Z/pZ are 0 and itself.

G = 0 case: G = Z/pZ case:

|B| = 1 so back in base case. |A| = p so A + B = Z/pZ, so
|A + B| = p.

A few comments

I This can be phrased nicely because of the e-transform.

Davenport’s proof required lots of keeping track of indices.
I This gives insight into why this is true for a cyclic group of
prime order.
Warmup question

I Let f be a polynomial over a field. Bound the size of the set

A := {α : f (α) = 0}.

|A| ≤ deg f

I Said another way, if |A| > deg f , then ∃a ∈ A with f (a) 6= 0

I Question: can we phrase a set (e.g. a sum set A + B) as the
zero locus of a polynomial? This is the polynomial method.

Theorem (The combinatorial Nullstellensatz)

Let F be a field, p ∈ F [t1 , t2 , ..., tn ] a degree-d polynomial which
has nonzero coefficient of t1d1 t2d2 ...tndn where d = d1 + · · · dn . If
Si ⊂ F with |Si | ≥ di ∀i, then there exists a tuple
x = (x1 , x2 , ..., xn ) ∈ S1 × S2 × · · · × Sn for which p(x) 6= 0.
Proof 2 of Cauchy-Davenport [TV10]

Lemma
Let h ∈ Fp [x, y ]. Let k ≥ 0, and let A, B be additive sets in Fp
with |A| + |B| = k + 2 + deg h. If (x + y )k h(x, y ) has a nonzero
coefficient of x |A|−1 y |B|−1 , then

|{α + β : α ∈ A, β ∈ B, h(α, β) 6= 0}| ≥ k + 1

Proof.
Contradict combinatorial Nullstellensatz.
Proof 2 of Cauchy-Davenport [TV10]
Proof.
If |A| + |B| > p, then |A + B| = Z/pZ. If |A| + |B| ≤ p,
Consider the polynomial
|A|+|B|−2  
|A|+|B|−2
X |A| + |B| − 2 n |A|+|B|−2−n
f = (x + y ) = x y .
k
n=0

The coefficient of x |A|−1 y |B|−1 is |A|+|B|−2

|A|−1
. Since |A| + |B| ≤ p,
no factors of p appear in the binomial coefficient, so it is nonzero.
Previous lemma with h = 1 tells us that
|{α + β : α ∈ A, β ∈ B, h(α, β) 6= 0}| ≥ |A| + |B| − 1

But since h 6= 0 always,

{α + β : α ∈ A, β ∈ B, h(α, β) 6= 0} = A + B.
So what?

Define a restricted sum set A+̂B = {a + b : a ∈ A, b ∈ B, a 6= b}

Conjecture (Erdos, Heilbronn 1964)

|A+̂A| ≥ min(2|A| − 3, p)

I Proved 1994 by da Silva, Hamidoune.

I Stronger result for A+̂B proved in 1996 by Alon, Nathanson,
Ruzsa

Proof.
Take h = (x − y ), and apply lemma.
Similar technique can be used to give results in particle physics.
Probabilistic method

Philosophy
If you can prove something has the right probability in an
appropriate space, that can interpreted as proof.

Example
Assign n balls to m bins at (uniform) random. Let P be the
probability that any of the bins contain two or more balls. If
P = 1, this is the pigeonhole principle.

Example
Existence proof: If I can draw something at random with nonzero
probability, it must exist.

Usage
Fourier analysis can be framed as statements about probability and
expectation.
Fourier analysis definitions

I Let p be a prime, f , g : Z/pZ → C

1 X
I fˆ(ξ) = f (x)e −2πixξ/p
p
x∈Z/pZ
1 X 1 X
I (f ∗ g )(ξ) = f (x − ξ)g (ξ) = f (ξ)g (x − ξ)
p p
x∈Z/pZ x∈Z/pZ
I supp f = {x ∈ Z/pZ : f (x) 6= 0}

Upshots
I supp(f ∗ g ) ⊆ supp(f ) + supp(g ) as an additive set.
I f[∗ g = fˆ · ĝ (among other standard identities)
Fourier analysis theorems

Theorem (Tao, ’05)

Let p be a prime, f : Z/pZ → C a random variable. Then
| supp f | + | supp fˆ| ≥ p + 1.

Theorem (Tao, ’05)

Let A, B, be nonempty subsets of Z/pZ with |A| + |B| ≥ p + 1.
Then ∃ a function f : Z/pZ → C such that supp f = A and
supp fˆ = B.
Proofs are not enlightening, so omitted.
Proof 3 of Cauchy-Davenport [TV10]
Let A, B be additive sets in Z/pZ. We can choose sets X , Y so
that
I |X | = p + 1 − |A|
I |Y | = p + 1 − |B|
I |X ∩ Y | = max(|X | + |Y | − p, 1).

Example
Let A = {1, 3, 5, 9}, B = {2, 3, 4} ⊆ Z/11Z so we want |X | = 8
and |Y | = 9, and |X ∩ Y | = 6

X = 0, 1, 2, 3, 4, 5 , 6, 7 , 6 8, 6 9, 6 10
Y = 0, 1, 2, 3, 4, 5 , 6 6, 6 7, 8, 9, 10

The point is that the only data from A and B which we retain is
their size.
Proof 3 of Cauchy-Davenport [TV10]
By previous theorem, since |A| + |X | = p + 1, there is:
I A function f with supp f = A and supp fˆ = X .
I g with supp g = B and supp ĝ = Y .
Now convolve f ∗ g . We know
I supp(f ∗ g ) ⊆ supp f + supp g = A + B
I supp(f[∗ g ) = supp(fˆ · ĝ ) = X ∩ Y
Then

| supp(f ∗ g )| + | supp(f[
∗ g )| ≥ p + 1
|A + B| + |X ∩ Y | ≥ p + 1
|A + B| ≥ p + 1 − max(|X | + |Y | − p, 1)
= p + 1 − max(p + 2 − |A| − |B|, 1)
= min(|A| + |B| − 1, p)


Poll

Which was your favorite?

1. Original proof exploiting (sub)group structure of Z/pZ?
2. Proof counting the zeros of a certain polynomial?
3. Exploiting the relationship between a Fourier transform
and its convolution?
Thank you!
References

H. Davenport, On the addition of residue classes, Journal of

the London Mathematical Society s1-10 (1935), no. 1, 30–32.
Ben Green, Additive combinatorics [book review of
mr2289012], Bull. Amer. Math. Soc. (N.S.) 46 (2009), no. 3,
489–497. MR 2507281
Terence Tao and Van H. Vu, Additive combinatorics,
Cambridge Studies in Advanced Mathematics, vol. 105,
Cambridge University Press, Cambridge, 2010, Paperback
edition [of MR2289012]. MR 2573797