Combinatorial Nullstellensatz

1 Combinatorial Nullstellensatz
See http://www.math.tau.ac.il/~nogaa/PDFS/null2.pdf for the original paper and https:
//artofproblemsolving.com/community/c7h19496p958907 for a discussion.

1.1 Background from algebraic geometry

Classical algebraic geometry is rooted in studying the solution sets of multivariate polynomial
equations over fields. Recall that a field is (loosely) a set with addition, multiplication, nonzero
inverses, and all the associativity/commutativity/distributivity properties you expect. Exam-
ples include Q, R, C, Fq , and K(t) for K a field. Sometimes we restrict to algebraically closed
fields like C, over which every single variable polynomial splits into linear factors.

Let K be a field. Given a set of polynomials {f1 , . . . , fk } ∈ K[x1 , . . . , xn ], we denote the set
of simultaneous solutions in K n by V ((f1 , . . . , fk )). Conversely, given a subset S ⊂ K n , we can
define I(S) to be the set of polynomials vanishing on S. Tautologically, we have

{f1 , . . . , fk } ⊂ I(V (f1 , . . . , fk )) and S ⊂ V (I(S)).

In general, these are far from being equalities of sets. Can you give examples for both being
proper and not proper subsets?

We are more interested in the first inclusion right now. There isn’t really a clean descrip-
tion of the second, except that in algebraic geometry one defines the Zariski topology in which
V (I(S)) is the closure of S.

We define the ideal (f1 , . . . , fk ) to be the set of all polynomials of the form ki=1 hi fi , where
P
hi ∈ K[x1 , . . . , xn ]. Note that this set is closed under addition, and also closed under multipli-
cation by any polynomial (these properties are essentially the definition of an ideal). Clearly,
we have
(f1 , . . . , fk ) ⊂ I(V (f1 , . . . , fk )).
Is this always an equality?

1.2 Combinatorial Nullstellensatz

Combinatorial Nullstellensatz, proven by Noga Alon (http://www.math.tau.ac.il/~nogaa/
PDFS/null2.pdf), says that this is an equality in a very special case. Namely, when the fk are
just products of linear polynomials in a single variable, so that the vanishing locus is just a
rectangular grid of points.
Theorem
Q 1.1. Take f ∈ K[x1 , . . . , xn ] and let S1 , . . . , Sn be nonempty subsets of K. Define gi =
s∈Si (xi − s). Then if f vanishes on S1 × · · · × Sn , then

f ∈ (g1 , . . . , gn ).
Pn
Furthermore, we can write f = i=1 hi gi with deg hi ≤ deg f − deg gi .

1
 Combinatorial Nullstellensatz

Proof. First, we make note of the following result. If f is 0 on T1 ×· · ·×Tn where |Ti | > degxi (f ),
then f = 0. Indeed, this follows immediately from induction on n. Now let us subtract from f
|S | |S |
suitable multiples of the gi by replacing each xi i with xi i − gi . By continuing to do this, we
reach a polynomial f such that degxi (f ) < |Si |. Then by our result f = 0, so f ∈ (g1 , . . . , gn ) as
desired. It is clear that this algorithm produces hi with deg hi ≤ deg f − deg gi .

Yes, it really is that easy! Here is a corollary.

Corollary 1.2. Say deg f = ni=1 ti with a nonzero coefficient of xt11 · · · xtnn . Then if |Si | > ti , there
P
is some tuple (s1 , . . . , sn ) ∈ S1 × · · · × Sn with f (s1 , . . . , sn ) ̸= 0.

Proof. Otherwise, by Combinatorial Nullstellensatz we know that f = ni=1 hi s∈Si (xi − s),
P Q
with deg hi ≤ deg f − |Si |. It is immediate that the right hand side has no term of the form
xt11 · · · xtnn , as desired.

1.3 Applications
Theorem 1.3 (Cauchy-Davenport theorem). Take subsets A, B ⊂ Z/pZ. Show that

|A + B| ≥ min{p, |A| + |B| − 1}.

Proof. If |A| + |B| > p, then for any x ∈ Z/pZ we have that x − A and B intersect. So let
|A| + |B| ≤ p and assume |A + B| ≤ |A| + |B| − 2. Define the polynomial f (x, y) ∈ Fp [x, y] by
Y
f (x, y) = (x + y − c).
c∈A+B

Then we  integers a, b with |A| > a, |B| > b, and a + b = |A + B|. Then the coefficient of
 can take
a + b
xa y b is which is nonzero, and so deg f = a + b. Then by Combinatorial Nullstellensatz,
a
we have that f cannot be 0 on A × B, which is a contradiction.

Example 1.4. For any prime p, any loopless graph G = (V, E) with average degree bigger than
2p − 2 and maximum degree at most 2p − 1 contains a p-regular subgraph.

Proof. We recall that 1 − xp−1 (mod p) is an indicator function to whether p|x. We want to
represent the degrees of the graph with a polynomial, so we fix a variable xe for each edge
and let av,e = 1 if v is a vertex of e and av,e = 0 otherwise. Then if we choose a P subgraph
by setting certain xe to be 1 and letting the rest be 0, for a fixed v we have that e av,e xe
is
Q the degreeP of v in p−1
that subgraph. Therefore we are motivated to consider the polynomial
v∈V [1 − (a x
e v,e e ) ]. Now as we will see, we also want to exclude the case when all xe = 0,
so we tack on another term and set
Y X Y
f= [1 − (av,e xe )p−1 ] − (1 − xe ).1
v∈V e e
Q
The condition tells us that (p − 1)|V | < |E|, so deg f = |E|. The highest degree term is e xe ,
so we know that f cannot vanish on all of {0, 1}n . It vanishes on (0, . . . , 0), so we see that there
is indeed some Pnonzero vector in {0, 1}n for the choices of e for which f does not vanish. This
implies that p| e|xe =1 av,e xe , so every vertex in this subgraph has degree p (by the problem
statement, it can’t be at least 2p).

1
We note that in the course of solving this problem, this would probably only be done after you play around with
the original polynomial and see what is missing.

2
 Combinatorial Nullstellensatz

2 Algebraic geometry: Hilbert’s Nullstellensatz

Let J = (f1 , . . . , fk ) be some ideal of polynomials in K[x1 , . . . , xn ] where K is a field. Let us
return to the question: what is the relationship between

(f1 , . . . , fk ) and I(V (f1 , . . . , fk ))?

√
Define the radical of an ideal J to consist of all g such that g n ∈ J for some n. We see that
√
J ⊂ J ⊂ I(V (J)).

Theorem 2.1 (Hilbert’s Nullstellensatz). If K is algebraically closed (e.g. C), then

√
I(V (J)) = J.

The complete proof of this is nontrivial. We begin with an important step, which is some-
times known as a weak version of the Nullstellensatz. To reiterate, K is an algebraically closed
field.

Proposition 2.2. The maximal ideals of K[x1 , . . . , xn ] are the ideals of the form (x1 −a1 , . . . , xn −
an ).

Proof. To show every such ideal is a maximal ideal, assume we can add another element to
obtain strict equalities I = (x1 − a1 , . . . , xn − an ) ⊂ J = (x1 − a1 , . . . , xn − an , f ) ⊂ (1). By
taking f modulo the original ideal, we see that it is either 0, in which case f ∈ I, or it is some
nonzero constant, in which case J = (1). Contradiction.
We note that this argument shows that an ideal I ⊂ A is maximal if and only if A/I is a
field.
The other direction is more difficult, and uses Zariski’s lemma. Zariski’s lemma states that
if a finitely generated (as a module) algebra over a field is a field, then it is a finite field extension
of the base field. This follows, for example, from the Noether normalization lemma. This
lemma essentially says that every variety is finite over some affine space. In any case, Zariski’s
lemma requires at least a little bit of commutative algebra.
Now with Zariski’s lemma, we have that if J is a maximal ideal, then K[x1 , . . . , xn ]/J is a
finitely generated (as a module) field extension of K, so it is a finite field extension of K, so it
is equal to K. Thus we see it contains some (x1 − a1 , . . . , xn − an ), so by the previous direction
we are done.

Note: (x2 + 1) is a maximal ideal in R[x].

∼
Proof of Hilbert’s Nullstellensatz. Consider the localization K[x √ 1 , . . . , xn ]f = K[x1 , . . . , xn , y]/(yf −
1). Let I = (g1 , . . . , gk ) ⊂ K[x1 , . . . , xn ] and assume that f ̸∈ I. We claim that the ideal gener-
ated by I forms a proper ideal in K[x1 , . . . , xn , y]/(yf −1). Indeed, if not we have h0 +h1 y +· · ·+
hm y m = 1 for some hi ∈ I. Multiplying by f m , we see that f m = g ′ in K[x1 , . . . , xn , y]/(yf − 1)
for some g ′ ∈ I. This implies that f m = g ′ in K[x1 , . . . , xn ], contradiction.
The maximal ideals of K[x1 , . . . , xn , y]/(yf −1) are given by the maximal ideals of K[x1 , . . . , xn ]
which contain yf − 1. Thus by the previous result, the ideal generated by I is contained
in some ideal (x1 − a1 , . . . , xn − an , y − yn ) ∈ K[x1 , . . . , xn , y] that contains yf − 1. Then
(a1 , . . . , an ) ∈ V (I) and y1 f (a1 , . . . ,√
an ) − 1 = 0. But since f vanishes on V (I), this is impossi-
ble. Thus the assumption that f ̸∈ I is false, and we are done.

Note: Although Hilbert’s Nullstellensatz was one of the founding results of algebraic ge-
ometry in the early 20th century, it is no longer that important because modern algebraic ge-
ometry goes much further using schemes, which look at all prime ideals rather than just the
maximal ones.

3
 Combinatorial Nullstellensatz

3 Problems
Some of the problems were taken from https://web.evanchen.cc/handouts/BMC_Combo_Null/
BMC_Combo_Null.pdf.

1. (Russia MO 2007/5) Two distinct numbers are written on each vertex of a convex 100-
gon. Prove one can remove a number from each vertex so that the remaining numbers
on any two adjacent vertices differ.

2. (IMO 2007 #6) Let n be a positive integer. Consider

S = {(x, y, z) : x, y, z ∈ {0, 1, . . . , n}, x + y + z > 0}

as a set of (n+1)3 −1 points in three-dimensional space. Determine the smallest possible

number of planes, the union of which contain S but does not include (0, 0, 0).

3. Let H1 , H2 , . . . , Hm be a family of hyperplanes in Rn that cover all vertices of the unit

cube {0, 1}n but one. Then m ≥ n.

4. Let n ≥ 2 be even and let v1 , . . . , vk ∈ {0, 1}n be vectors of length n such that any v ∈ {±1}
is orthogonal to at least one of the vi . Prove that k ≥ n and that this estimate is sharp.

5. (Erdös-Ginzburg-Ziv) From any 2n − 1 integers one may choose n with sum divisible by
n.

6. (St. Petersburg 2003) Let p be a prime number, and n be an integer such that n ≥ p.
Assume that a1 , a2 , ..., an are arbitrary integer numbers. For every k, we denote by fk the
number of k-subsets {s1 , s2 , ..., sk } of the
P set {1, 2, ..., n} such that the sum as1 + as2 +
... + ask is divisible by p. Prove that p | nk=0 (−1)k fk .