Lect 26

PROOF OF THUE’S THEOREM – PART I
In this lecture we’ll do the first half of the proof of Thue’s theorem.
Suppose that P is a polynomial with integer coefficients. Let r be a rational point.
We would like to understand the relationship between P vanishing to high order at
r and the “size” of P in terms of its degree and size of its coefficients. Define |P | to
be the maximum absolute value of the coefficients of P .
1. Polynomials vanishing to high order at a rational point

Let us start with the simple case where P is a one-variable polynomial. Let
r = p/q, written in lowest form. We suppose that P vanishes to ℓ orders at r, i.e.,
∂ j P (r) = 0, j = 0, 1, . . . , ℓ − 1.
Here is the first example of such a polynomial that comes to mind.
Example 1: P (x) = (qx − p)ℓ . Then |P | ∼ krkℓ , where krk := max(|p|, |q|).
Could we do better? In other words, can we find a polynomial P with smaller
coefficients, i.e., |P | smaller? As we shall see, the answer is no.
Proposition 1.1 (Gauss). If P ∈ Z[x] satisfies ∂ j P (r) = 0 for j = 0, 1, . . . , ℓ − 1,
then P (x) = (qx − p)ℓ P1 (x) for some P1 ∈ Z[x].
Proof. The vanishing condition tells us that P (x) = (qx − p)ℓ P2 (x) for some poly-
nomial P2 ∈ R[x]. It remains to show that the coefficients of P2 are integers. By
expanding and comparing coefficients, we see that we can solve for the coefficients
of P2 in terms of the coefficients of P , and we deduce that the coefficients of P2 are
at least rational.
Taking out the lowest common demoninator, we write P (x) = M1 (qx − p)ℓ P̃2 (x),
for some P̃2 ∈ Z[x] so that there is no prime dividing all the coefficients of P̃2 as well
as M. So MP (x) = (qx − p)ℓ P̃2 (x). If M 6= ±1, then let s be any prime divisor of
M. Then we get a contradiction modulo s, since qx − p is not 0 mod s as p/q was
already given in lowest terms, and P̃2 is also not 0 mod s. It follows that M = ±1
and hence P1 ∈ Z[x].
We’re not quite done yet, as it’s not always true that the norm of a polynomial is
always at least as large as its factors.
Example 2: The polynomial (x−1)2 = x2 −2x+1 has norm 2, while (x−1)2 (x+1) =
x − x2 − x + 1 has norm 1, which is smaller.
3
1
2 PROOF OF THUE’S THEOREM – PART I
Fortunately, this example does not really pose an issue. The follow corollary
answers our question for one variable polynomials.
Corollary 1.2. |P | ≥ krkℓ .
Proof. In P (x) = (qx − p)ℓ P1 (x), we see that q ℓ divides the top coefficient and pℓ
divides the bottom non-zero coefficient.
Now, what about polynomials in two variables? Let P ∈ Z[x1 , x2 ], and r =

(r1 , r2 ) = (p1 /q1 , p2 /q2 ) ∈ Q2 . We want to assume that P vanishes to high order at
r. Let’s say ∂j P (r) = 0 for all j ∈ J, where J is some list of pairs, e.g., all j = (j1 , j2 )
with |j| := j1 + j2 ≤ ℓ − 1.
Define krk := max(kr1 k, kr2 k). If krk is large, does P (r) = 0 imply something
about the norm of P (as in the single variable case)? The following examples show
that the answer is no.
Example 3: P (x1 , x2 ) = x1 − x2 and r = (r1 , r1 ). Then P (r) = 0 but |P | = 1.
What if we assume P vanishes at r to high order? Say ∂j P (r) = 0 for all j with
|j| ≤ ℓ − 1? Still the answer is no.
Example 4: P (x1 , x2 ) = (x1 − x2 )ℓ and r = (r1 , r1 ). Then ∂j P (r) = 0 for all j with
|j| ≤ ℓ − 1, but |P | ≤ 2ℓ , independent of krk.
These examples suggest that perhaps our notion of vanishing to high order at
a point isn’t very useful. It prompts us to modify the question. Let us consider
polynomials of the form
P (x1 , x2 ) = P1 (x1 )x2 + P0 (x1 ).
Suppose we have
∂1j P (r) = 0, j = 0, 1, . . . , ℓ − 1.
In this case, can we infer something about the size of P from the size of r?
Since we are only differentiating with respect to x1 , this condition is equivalent to
∂ j [p2 P1 + q2 P0 ](r) = 0, j = 0, 1, . . . , ℓ − 1.
It follows by Corollary 1.2 that kp2 P1 + q2 P0 k ≥ kr1 kℓ . We see that kp2 P1 + q2 P0 k ≤

ℓ
kp2 P1 k + kq2 P0 k ≤ 2kr2 k|P |. It follows that |P | ≥ 12 kr 1k
kr2 k
.
Let us look at some examples of polynomials that satisfy the above vanishing
conditions.
Example 5: Let P = q2 x2 − p2 . Then |P | = kr2 k and ∂1j P (r) = 0 for all j.
Example 6: Let P = (q2 x2 − p1 )ℓ . Then |P | ≥ kr1 kℓ .
PROOF OF THUE’S THEOREM – PART I 3
2. Integer solutions to linear systems

So far we’ve been looking at explicit examples of polynomials that satisfy the
vanishing to high order condition. This is somewhat reminiscent of the first lecture
in the course where we wanted to know how big the degree of a polynomial P must
be if P (j, 2j ) = 0 for j = 1, 2, . . . , 106. There we also started by finding explicit
examples, but at the end we arrived at our bound by counting dimensions. In a
similar vein, we are going to find polynomials in Z[x1 , x2 ] by parameter counting.
Proposition 2.1. If L : ZM → ZN is a linear map, given by a matrix with integer
coefficients, with M > N, then there exists a nonzero x ∈ ZM such that Lx = 0.
For real vector spaces, this result follows from elementary results from linear alge-
bra. For integers, it’s actually even more elementary — it’s just pigeonhole principle.
Let’s quickly sketch a proof first. Afterwards we’ll be more careful quantitative
bounds.
Proof. (Sketch) Let QM S := {x ∈ Z

M
: |xi | ≤ S, i = 1, . . . , M}. Since the map L
M N
restricted to QS → QC·S where C = C(L) is some sufficiently large constant. We
have |QM S | ∼ S
M
and |QC·S | ∼ C N S N . So we can choose S so that |QM N
S | > |QC·S |.
Then by pigeonhole, there are x1 6= x2 ∈ QM S such that L(x1 ) = L(x2 ), so that
L(x1 − x2 ) = 0.
How big is the x produced by the proof? Let us look for some quantitative bounds.
We can take C = |L|op := max|x|∞=1,x∈RM |Lx|∞ by the operator norm of L. In
particular, |L|op ≤ M · max |coeff of L|. We need to take S so that (2S + 1)M >
(2|L|op S +1)N . It suffices to have (2S +1)M > |L|op (2S +1)N , or equivalently 2S +1 >
M/(M −N )
|L|op . It follows that we can always find a nonzero x ∈ ZM with Lx = 0 and
N/(M −N )
|x|∞ ≤ |L|op . So we can revise the proposition to a more quantitative version.
Proposition 2.2. If L : ZM → ZN is a linear map, given by a matrix with integer
N/(M −N )
coefficients, with M > N, then there exists a nonzero x ∈ ZM with |x|∞ ≤ |L|op
such that Lx = 0.
Note that if M = N + 1, then our bound is |L|N
op which is not too great, where as
100
if M = 1.01N then our bound is |L|op which is pretty good.
Let’s go back to the one-variable polynomial case for a moment. Recall that we
already know that (px − q)ℓ is the optimal polynomial vanishing to ℓ-th order at
r = p/q. Nevertheless, let us try this counting machinery here and see how well it
does in comparison.
4 PROOF OF THUE’S THEOREM – PART I
Suppose we are looking for a polynomial P of degree D such that ∂ j P (r) = 0 for
j = 0, 1, . . . , ℓ − 1. Let
XD
P = ai xi
i=0
We have
D D
j
X i! i−j
X i i−j
∂ P (x) = ai x = j! ai x .
i=0
(i − j)! i=0
j
(Extracting out the j! factor in the last step is a useful trick of the trade that makes
it easier to bound the coefficients.) Setting ∂ j P (r) = 0, we have
D
X i D−(i−j) i−j
ai q p = 0.
i=0
j
The coefficients of the ai ’s are all bounded in absolute value by 2D krkD . Viewing
(a0 , . . . , aD ) ∈ ZD+1 as our unknowns, it follows from Proposition 2.2 that we can
find a polynomial P of degree D with ∂ j P (r) = 0 for j = 0, 1, . . . , ℓ − 1 such that
|P | ≤ (2D krkD )ℓ/(D−ℓ) ∼ krkℓD/(D−ℓ).
So we could take, for example, D = 100ℓ to get |P | ∼ krk1.01ℓ . For comparison, the
optimal example (qx − p)ℓ has D = ℓ and |P | ∼ krkℓ .

Lect 26

Uploaded by

Copyright:

Available Formats

Lect 26

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lect 26

Uploaded by

Copyright:

Available Formats

PROOF OF THUE’S THEOREM – PART I

1. Polynomials vanishing to high order at a rational point

Corollary 1.2. |P | ≥ krkℓ .

Now, what about polynomials in two variables? Let P ∈ Z[x1 , x2 ], and r =

P (x1 , x2 ) = P1 (x1 )x2 + P0 (x1 ).

It follows by Corollary 1.2 that kp2 P1 + q2 P0 k ≥ kr1 kℓ . We see that kp2 P1 + q2 P0 k ≤

2. Integer solutions to linear systems

Proof. (Sketch) Let QM S := {x ∈ Z

You might also like