9 The Polynomial Method

Our main goal here is to introduce a very simple and useful tool for solving certain types of
combinatorial problems. The main idea is due to Alon and Tarsi, and uses only very basic
properties of polynomials to achieve some surprisingly powerful results. We will first use this
technique to get a new proof of the Cauchy-Davenport Theorem. Then, we will generalize
this proof to achieve a result on a type of restricted sumset problem. If G is an abelian group
and A, B ⊆ G, then we let A ⊕ B = {a + b : a ∈ A, b ∈ B, and a 6= b}. Our main result
from this section is the following theorem on restricted sumsets as conjectured by Erdös and
Heilbron.

Theorem 9.1 (Dias da Silva, Hamidoune) Let p be prime and let A ⊆ Zp be nonempty.
Then |A ⊕ A| ≥ min{p, 2|A| − 3}.

We begin with a quick review of polynomials. Throughout we will fix a finite field
F of order q, and we will use F[x1 , x2 , . . . , xn ] to denote the ring of polynomials over F
with variables x1 , x2 , . . . , xn (so members of F[x1 , . . . , xn ] are formal linear combinations of
monomials xd11 . . . xdnn with coefficients in F). Every polynomial in this ring gives rise to a
mapping from Fn to F, and we call two polynomials P, Q ∈ F[x1 , . . . , xn ] equivalent if they
give the same mapping.

Proposition 9.2 Let B = {xd11 xd22 . . . xdnn ∈ F[x1 , . . . , xn ] : di < q for 1 ≤ i ≤ n}. Then B
is a basis of the F-linear space of functions from Fn to F. In particular, two polynomials
P, Q ∈ F[x1 , . . . , xn ] are equivalent if and only if they reduce to the same polynomial by
repeatedly using the rewrite rule xqi = xi
Qn
Proof: If z1 , z2 , . . . , zn ∈ F, then the polynomial i=1 (1 − (xi − zi )q−1 ) has value 1 at
(z1 , . . . , zn ) and 0 elsewhere. Further, by expanding, this polynomial may be written as a
linear combination of elements from B. Since every function may be expressed as a linear
combination of such terms, it follows that every function from Fn to F may be written as a
linear combination of members of B. Since |B| = q n and the dimension of the F-linear space
of functions from Fn to F also has dimension q n , we find that B is a basis as required.
Since F has order q, the multiplicative group F \ {0} has order q − 1, and it follows
that every z ∈ F satisfies z q = z. By writing a polynomial P ∈ F[x1 , . . . , xn ] as a sum of
 2
monomials and then using the rewrite rule xqi = xi , we reduce P to a linear combination of
terms from B and the result follows. 
Pn
If d1 , d2 , . . . , dn < q, then we define the degree of the monomial xd11 . . . xdnn to be i=1 di .
More generally, by the above proposition, every polynomial function P may be expressed
uniquely as a linear combination of terms from B, and we define the degree of P to be the
largest degree of a term in the support of in this representation.

Theorem 9.3 (Alon, Tarsi) Let P ∈ F[x1 , . . . , xn ], and let xd11 . . . xdnn have degree equal
to the degree of P , and assume that xd11 . . . xdnn appears in the expansion of P with nonzero
coefficient. If A1 , A2 , . . . , An ⊆ F satisfy |Ai | ≥ di + 1, then there exists (z1 , z2 , . . . , zn ) ∈
A1 × A2 . . . × An so that P (z1 , z2 , . . . , zn ) 6= 0.
Pn
Proof: Let d be the degree of P which is also equal to i=1 di . We may assume without loss
that |Ai | = di + 1 for every 1 ≤ i ≤ n and we set Bi = F \ Ai . Using the rewrite rule in the
above proposition, we may assume that P is expressed as a linear combination of monomials
in B. Now, consider the following polynomial
n Y
Y
Q(x1 , . . . , xn ) = P (x1 , . . . , xn ) · (xi − z).
i=1 z∈Bi

It is immediate from this construction that Q is not identically 0 if and only if there exists
(z1 , . . . , zn ) ∈ A1 ×. . .×An with P (z1 , . . . , zn ) 6= 0. Thus, to complete the proof, it suffices to
show that Q is not identically 0. To see that Q is nonzero, consider the term xq−1
1 x2
q−1
. . . xq−1
n
in the expansion of Q. Since P is written as a linear combination of monomials with degree
< q in each variable and total degree ≤ d, every monomial xd11 xd22 . . . xdnn appearing in the
expansion of Q has ni=1 di ≤ (q − 1)n . It follows from this that after reducing, the only
P

contribution to the coefficient of x1q−1 . . . xq−1

n comes from the term xd11 . . . xdnn in P and is
nonzero. Thus, Q is not identically 0 and the proof is complete. 

The above result also holds without the assumption that F is finite. The proof of this
more general result is quite instructive and the interested reader is encouraged to see Alon’s
excellent survey article ”Combinatorial Nullstellensatz” for a proof of this, and many appli-
cations. Our first application will be a new proof of the Cauchy-Davenport Theorem.
 3
Theorem 9.4 (Cauchy-Davenport) If p is prime and A, B ⊆ Zp are nonempty, then
|A + B| ≥ min{p, |A| + |B| − 1}.

Proof: Let A, B be a counterexample to the theorem with |A| + |B| minimum and set
k = |A| and ` = |B|. Note that by minimality, we must have k + ` ≤ p + 1. Now choose a
set C ⊇ A + B with |C| = k + ` − 2 and consider the following polynomial in two variables
Y
P (x, y) = (x + y − c).
c∈C

The degree of P is k + ` − 2 and the coefficient of the term xk−1 y `−1 in the expansion of P
is equal to k+`−2


k−1
6= 0. Applying the previous theorem to this polynomial for the sets A, B
gives us a pair a ∈ A and b ∈ B with P (a, b) 6= 0. But then a + b 6∈ C ⊇ A + B, giving us a
contradiction. 

This alternate proof of Cauchy-Davenport is quite useful since it leads to numerous

generalizations which do not obviously follow from the original proof. Perhaps the most
interesting of these is the following.

Theorem 9.5 (Alon, Nathanson, Ruzsa) Let p be prime, let A, B ⊆ Zp and assume
that |A| =
6 |B|. Then |A ⊕ B| ≥ min{p, |A| + |B| − 2}.

Proof: Let A, B be a counterexample to the above theorem with |A| + |B| minimum and
let k = |A| and ` = |B|. We may assume that 2 ≤ k ≤ `. By the minimality of our
counterexample, we may further assume that k + ` − 2 ≤ p (otherwise remove an element of
A). Now, choose a set C ⊇ A⊕B with |C| = k +`−3 and consider the following polynomial.
Y
P (x, y) = (x − y) (x + y − c).
c∈C

The coefficient of xk−1 y `−1 is equal to

   
k+`−3 k+`−3 (k − `)(k + ` − 3)!
− =
k−2 k−1 (k − 1)!(` − 1)!

which is nonzero modulo p. By applying Theorem 9.3 to this polynomial for the sets A
and B we find that there exist a ∈ A and b ∈ B with P (a, b) 6= 0. But then a 6= b and
a + b 6∈ C ⊇ A ⊕ B and we have a contradiction. 
 4
This gives us an easy proof of the Dias da Silva - Hamidoune Theorem as follows.

Proof of Theorem 9.1: If |A| = 1 there is nothing to prove. Otherwise choose a ∈ A and set
A0 = A \ {a}. Now by the previous theorem we have |A ⊕ A| ≥ |A ⊕ A0 | ≥ min{p, 2|A| − 3}
as required. 

As in the preceeding results in this section, the tool is very simple to prove, but is still
quite useful. Our main application of this tool is a lemma due to Alon called the Permanent
Lemma, which has found application in graph theory as well as additive number theory.

Lemma 9.6 (Alon’s Permanent Lemma) Let M be an n × n matrix over F and as-
sume that perm(M ) 6= 0. If A1 , A2 , . . . , An ⊆ F and |Ai | = 2 for every 1 ≤ i ≤ n and
z1 , z2 , . . . , zn ∈ F, then there exists a vector a = (a1 , a2 , . . . , an ) ∈ A1 × . . . × An so that the
ith coordinate of M a is not equal to zi for every 1 ≤ i ≤ n.

Proof: Let M = {mi,j } and consider the following polynomial

n
Y
P (x1 , . . . , xn ) = (mi,1 x1 + mi,2 x2 + . . . mi,n xn − zi ).
i=1

The coefficient of x1 x2 . . . xn in the expansion of P is equal to the permanent of M which is

nonzero. Thus, by Theorem 9.3 there exists a vector a = (a1 , . . . , an ) ∈ A1 × . . . × An with
P (a1 , . . . , an ) 6= 0, but then by construction, the ith coordinate of M a is not equal to zi for
every 1 ≤ i ≤ n. 

Next we will use the Permanent Lemma to get another proof of the Erdös Ginzburg Ziv
theorem for Zp . This proof is similar to that of Proposition 5.3, but uses the Permanent
Lemma instead of Cauchy-Davenport.

Theorem 9.7 (Erdös, Ginzburg, Ziv) If β is a sequence in Zp of length 2p − 1, then

there is a subsequence of β of length p with sum 0.

Proof: Let β be given by b1 , b2 , . . . , b2p−1 . Identify the elements of Zp with the representatives
0, 1, . . . , p−1 as usual. By possibly reordering our sequence, we may assume that b1 ≤ b2 . . . ≤
b2p−1 . If there exists 1 ≤ i ≤ p − 1 so that bi = bp−1+i then some element occurs p times and
this subsequence has zero sum. Otherwise, let M be the (p−1)×(p−1) matrix over Zp with all
 5
entries 1 and let z = (z1 , z2 , . . . , zp−1 ) be a list of all elements in Zp \ −b2p−1 . By the previous
lemma we may choose a vector a = (a1 , . . . , ap−1 ) ∈ {b1 , bp } × {b2 , bp+1 } . . . × {bp−1 , b2p−2 }
so that M a and z have no coordinates equal. But then by construction, a1 , a2 , . . . , ap−1 is a
subsequence of b1 , . . . , b2p−2 with sum equal to −b2p−1 so appending the term b2p−1 gives us
the desired subsequence. 

Conjecture 9.8 (Jaeger) If M is an invertible matrix over a finite field with order > 3,
then there exist a pair of vectors x, y with M x = y so that x and y have no coordinates equal
to zero.

Since permanents and determinants are the same over fields of characteristic two, the
Permanent Lemma implies the truth of Jaeger’s conjecture over all such fields. More gener-
ally, Alon and Tarsi have shown that Jaeger’s conjecture holds over all fields with order not
equal to a prime.