Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Polynomial Method

Diljit Singh

1 Introduction
Noga Alon’s combinatorial nullstellensatz and the polynomial method will be the focus of
the first part of the presentation. We will attempt to build some intuition around the com-
binatorial nullstellensatz which, aided by induction and Lagrange Interpolation, will lead
to a proof of the theorem. Then we will go over a few problems, focusing on graph theory,
that utilize the polynomial method.

2 Combinatorial Nullstellensatz
2.1 Nullstellensatz is German for “zero-locus-theorem”
Before introducing the combinatorial nullstellenstaz we consider two well-known theorems.
Theorem 2.1 (Pigeonhole Principal) Given n pigeons and n − 1 boxes at least 2 pigeons must be
in the same box.

Theorem 2.2 A polynomial, p(x), of degree n has at most n distinct roots.

It naturally follows from the above theorems that given any n + 1 distinct values there
would be atleast one value s.t. any (non-zero) polynomial, p(x) of degree n is non-zero. This
is the combinatorial nullstellensatz in one variable. That is, we will generalize this idea to
polynomials of arbitrarily many variables. We don’t prove anything here, instead we try to
get a feel of the theorem now and prove it in the next part. The main idea here is that the
zero’s of a polynomial are limited by its degree, so if we look at the polynomial over a whole
bunch of points it should not always be zero.

Theorem 2.3 (Combinatorial Nullstellensatz) Let p be an element of the polynomial ring,


n n
d
xi i is
P Q
K[x1 , · · · , xn ], such that the degree of p is D = di with all di ≥ 0. If the coefficient of
i=1 i=1
non-zero then given S = S1 × S2 × · · · × Sn where S ⊂ K and |Si | > di there exists atleast one element
s ∈ S s.t. p(s) , 0.

That is a mouthful let’s try to break it up.


1. Let p be an element of the polynomial ring, K[x1 , x2 , x3 , · · · , xn ]: All this means is that
p is a polynomial in n variables (x1 , x2 , · · · , xn ) with coefficients from the field K. So
for instance if
q ∈ Zp [x, y, z]
then that means q is a polynomial in x,y,z and each term has coefficients from the field
Zp (integers modulo p).

1
n n
d
xi i is non-zero:
P Q
2. The degree of p is D = di with all di ≥ 0 and the coefficient of
i=1 i=1
Think about how we define the degree of a polynomial in n variables. What is the
degree of:
P (x, y, z) = xy 2 z5 + x8 + x2 y 2 z2 + x4 y 4 + 6?
Ok so the degree is 8, that wasn’t too hard.

What we did to find the degree of that was we broke the polynomial up into its sum-
mands (monomials) in this:

{xy 2 z5 , x8 , x2 y 2 z2 , x4 y 4 , 6}

and then for each monomial we added up the exponents of every variable, this gives,
respectively:
{8, 8, 6, 8, 0}
The greatest sum, in this case 8 is the degree of the polynomial. Easy enough, right?
Well this is all this part of the theorem is saying. We start off with a polynomial, p
and its degree is some number D. All that means is that there is some monomial in p
where the exponents add up to D, but there are a lot of ways this could happen*! We
are interested in the numbers used to sum to D in the polynomial. So for instance we
would care about a few sums for our example such as: 1 + 2 + 5 = 8 and 8 + 0 + 0 = 8 and
4+4+0 = 8 but I couldn’t care less about 3+3+2 = 8 no monomial with such exponents
exists in my polynomial. Seem’s easy enough but beware! Consider 5 + 2 + 1 = 8. Yes
those numbers appear in our polynomial and so it satisfies the first part of the theorem
n n
d
xi i ? Well that
P Q
(The degree of p is D = di ) but what is the coefficient of the of
i=1 i=1
depends if we let d1 = 1, d2 = 2, and d3 = 5 then it is 1 but if we jumble around the the
n
d
xi i
Q
di ’s for instance if we let d1 = 2, d2 = 1, and d3 = 5 then the coefficient of the of
i=1
is zero, which I don’t care about.
So we want to find the monomial in p where the sum of the degrees of its variables is
D, or atleast we want to find it upto its coefficient. Caring that the coefficient is non-
zero only means we want to confirm that the particular monomial actually appears in
p.

3. then given S = S1 × S2 × · · · × Sn where S ⊂ K and |Si | > di there exists atleast one
element s ∈ S s.t. p(s) , 0: This is the bulk of the combinatorial nullstellensatz. Given
S that is large in some sense (that is each of its dimensions, Si , is larger than the
respective exponent di ), then the polynomial should not be able to be zero everywhere
on it. If your thinking, ”that can’t be right what about the zero polynomial?” then refer
back to the previous point, since we have the coefficient to some monomial is non-zero,
p cannot be the zero polynomial.
*Given exponents should be non-negative is there an expression for the number of ways?
The proof is in N. Alon’s paper (first reference). It is both easy to follow and doesn’t
require any really high powered theorems.

2.2 Practice Problems


Let’s try to see this in action.
1.(Prof. Melvyn Nathanson Additive NT, this was a homework question)

2
For h ≥ 3 let {Ai }hi=1 with Ai = ki we will show that:

|Σhi=1 Ai | ≥ 1 − h + Σhi=1 ki

Take C = Σhi=1 Ai , assume that |C| ≤ −h + Σhi=1 ki . Define f for for ai ∈ Ai as follows:

def
Y
f (a1 , a2 , · · · , ah ) = (−c + Σhi=1 ai ).
c∈C

Qh ki −1
The term containing i=1 ai has the coefficient

−h + Σhi=1 ki
!

k1 − 1, k2 − 1, · · · , kh − 1

which is nonzero in R. By the nullstellensatz, if we take Si = Ai we should be able to find a


value where f is non-zero. This implies there ∃ (a1 , a2 , a3 , ..., ah ) ∈ A1 × A2 × · · · × Ah s.t. their
sum is not in C = Σhi=1 Ai , contradicting that |C| ≤ −h + Σhi=1 ki meaning that C = Σhi=1 Ai ≥
1 − h + Σhi=1 ki .

2.
Chevalley-Warning in 1 polynomial. That is given q ∈ Zp [x1 , · · · , xn ] such that q ≡ 0 (mod p)
iff xi = 0 ∀ i, what is a lower bound on the degree of q?

Lower Bound Proof Assume that there exists a polynomial q with such properties and
deg(q) < n. Define the polynomial,
n
Y Y
h = 1 − qp−1 + (−1)σ (xi − k)
i=1 k∈Zp k,0

where
σ ≡ n+1 (mod 2).
Taking into consideration both Fermat’s Little Theorem and Wilson’s Theorem we see that h
is zero everywhere on Zp .
Since q has degree less than n, we have that

deg(qp−1 ) < (p − 1)n

while
n
Y Y
(−1)σ (xi − k)
i=1 k∈Zp k,0

gives us deg((p − 1)n). So we have deg(h) = (p − 1)n. Consider the monomial


n
p−1
Y
xi
i=1

the coefficient is −1σ .


Taking Si = Zp for i = 1, 2, · · · , n we have h is zero everywhere on S = S1 ×S2 ×· · ·×Sn = Znp ,
however this contradicts the nullstellensatz. Thus the lower bound of the degree of any such
polynomial, q is then the number of inputs.

3
The motivation of h is as follows: we wanted to define a function that is zero everywhere,
to do this we needed some control, so we took qp−1 which is either 1 or 0 by Fermat’s little
theorem. We wanted the other part of the function to be 1 when qp−1 is zero and 0 oth-
erwise. The former case only occurs when all xi are zero so the latter part of the function
is just (−1)σ (p−1)!n = −1. In the case qp−1 is 1, the latter part of the h is zero and we are done.

Prof. Leonid Gurvits asked to construct, given p and n, the lowest such degree polyno-
mial, q. I found this to be a rather non-trivial exercise with a very nice solution.
3.
(Alon) For any prime, p, and any loopless graph, G = (V , E) with an average degree bigger
than 2p − 2 and max degree of 2p − 1, G contains a p-regular subgraph.
Comments The proof below shows an instance where we don’t infact use the nullstellensatz
to draw a contradiction, instead we use it to draw an existence proof. Also Whenever the
average is an integer the question is obvious, we just take one vertex away. This is not always
the case though. It was proven a few years back this holds for prime powers, but the result
for integers is still open.
Proof Let A denote the incidence matrix of G with entries av,e ∈ Z2 . With each edge associ-
ated with xe ∈ Z2 . Consider the polynomial
Y X Y
F= (1 − ( av,e xe ))p−1 − (1 − xe ).
i=v∈V e∈E i=e∈E

av,e xe ))p−1 is (p − 1)|V | and we know that the average degree is


Q P
The degree of (1 − (
i=v∈V e∈E
greater than 2p − 2, letting di be the degree of vertex i so we have that
X d
i
> 2p − 2
|V |

which means that |E| = d2i > (p − 1)|V |. The degree of


P Q
(1 − xe ) is |E| so the degree of F
i=e∈E
is |E|. The coefficient of e∈E xe is (−1)|E|+1 , that is the leading coefficient is not zero. Taking
Q
|E|
S = Z2 with all Si = Z2 we have that ∃ s ∈ S s.t. F(s) , 0. If s = 0̂, then F = 0 so we
have that the vector is non-zero, if instead the vector has any non-zero components then
Q
the second part of F is zero, that is: (1 − xe ) = 0. Now notice, since we are looking for
P i=e∈E
when F is nonzero, that if av,e xe is non-zero mod p then the F is zero and if the sum is
e∈E P
zero mod p then F = 1 , 0. So we have (from the nullztellensatz) for all v that av,e xe ≡ 0
e∈E
(mod p). This implies that subgraph of edges e ∈ E where all xe = 1, that is the i th component
of the vector s is 1 we have that the degree of every vertex of the subgraph is a multiple of
p. Since the maximum degree is 2p − 1 the degree of each vertex is p. Q.E.D.

3 Summary
A general way we used the polynomial method was that we inscribed an aspect of the ques-
tion into a polynomial. From here we assumed the polynomial was smaller (lower degree)
than we conjecture. Then we set up a new polynomial that was always zero in some field.
Next we showed the leading coefficient was not zero and used the nullstellensatz to produce
a contradiction.

4
References
• N. Alon, Combinatorial Nullstellensatz

• M. Nathanson 2015 Additive Number Theory, Lecture 1

You might also like