POLYNOMIALS ASSUMING SQUARE VALUES

M. Ram Murty1
in honour of R.P. Bambah on his 80th birthday

Abstract. If f (x1 , ..., xn ) ∈ Z[x1 , ..., xn ] has the property that every integer specialization
gives an integral square value, then f is itself the square of a polynomial. We also give an
effective version of this result by using an effective version of a classical theorem of E. Noether
along with a theorem of Lang and Weil.

1. Introduction

Given a polynomial f (x1 , ..., xn ) ∈ Z[x1 , ..., xn ] with the property that every integer spe-
cialization of the xi ’s results in a square value, does it follow that f is itself the square of a
polynomial? Below, we will show that the answer is yes. In fact, it is not necessary to assume
that an infinite number of specializations give rise to a square value. A finite number, depending
on the size of the coefficients and the degree of the polynomial suffice and this is our main theo-
rem. The question raises other questions that belong to number theory and algebraic geometry.
We will discuss these questions at the end.
The case n = 1 of this problem is classical. For example, it appears as a problem in the book
by Pólya and Szegö (see p. 132 of [7]). Of course, an analogous result is true for k-th powers
also. As the referee points out, the multi-variable version of this problem was first investigated
by Kojima [4] in 1915. A modern treatment of it can be found in Theorem 52 of [9].
After giving an expository treatment of the single and several variable cases of the problem,
we will prove the following effective theorem:
Theorem 1. Let f (x1 , ..., xn ) ∈ Z[x1 , ..., xn ]. Then, there is an effectively computable constant
C = C(f ), depending only on f such that if every integer specialization of x1 , ..., xn with |xi | ≤ C
makes f (x1 , ..., xn ) a perfect integral square, then f (x1 , ..., xn ) is itself the square of a polynomial.
The effectively computable constant C(f ) seems to be humongous and depends on the size
of the coefficients of f , the degree d and n. It may be possible by using the work of Deligne,
to improve this estimate for C(f ), but at present, there are some technical difficulties in this
approach. How one may circumvent these difficulties will be addressed in a later paper. However,
a refinement of the argument used to prove Theorem 1 will enable us to show:
1
Research partially supported by a Natural Sciences and Engineering Research Council (NSERC) grant.
Mathematics Subject Classification(2000): Primary 11R09, Secondary 11C08.
Key words and phrases: polynomials, absolute irreducibility, squares, Lang-Weil theorem, resultants.

1
2 POLYNOMIALS ASSUMING SQUARE VALUES

Theorem 2. Let f (x1 , ..., xn ) ∈ Z[x1 , ..., xn ] have total degree d and let
 
n+d−1
k= .
n
t
Let ||f || denote the sum of the absolute values of all the coefficients of f . Put ψ = 2dt2 where
t = (d + 1)(d + 2)/2. If for all integer specializations with 0 ≤ ai ≤ H where
k
H = 2 max(6dψ, 9(d − 1)2 (d − 2)2 , 1319007) + k 2 log 4||f ||,

we have that f (a1 , ..., an ) is a perfect square, then f (x1 , ..., xn ) is itself the square of a polynomial.

2. Preliminaries

We begin by recalling several results we will need in the later discussion. These are of interest
in their own right. The first concerns an old result due to Schur [10]. Given a polynomial
f (x) ∈ Z[x], we say a prime p is a prime divisor of f if p|f (n) for some natural number n. An
excellent introduction to the topic of prime divisors of polynomials can be found in [2].

Lemma 3. (Schur) Let f (x) be a non-constant polynomial with integer coefficients. Then f has
infinitely many prime divisors.

Proof. The proof follows Euclid. We induct on the degree of f . For polynomials of degree one,
this is clear. Let
f (x) = an xn + an−1 xn−1 + · · · + a1 x + a0 .

If a0 = 0, we are done by induction, so we may suppose a0 6= 0. Since f (x) can assume the values
±1 only a finite number of times, we deduce that f has at least one prime divisor. Suppose
there are only finitely many such prime divisors, p1 , ..., pr (say). For each natural number m, let
Nm = (p1 · · · pr )m a0 and consider

f (Nm ) = a0 (an an−1

0 (p1 · · · pr )nm + · · · + a1 a0 (p1 · · · pr ) + 1).

For m sufficiently large, the term in the parentheses above is in absolute value greater than 1
and coprime to p1 · · · pr . This is a contradiction.

We also need to recall some basic facts about resultants (see p. 200ff of [5]). Let R be a
commutative ring. Given two polynomials

f (x) = an xn + an−1 xn−1 + · · · + a1 x + a0

and
g(x) = bm xm + bm−1 xm−1 + · · · + b1 x + b0
 POLYNOMIALS ASSUMING SQUARE VALUES 3

in R[x], we define the resultant R(f, g) to be the determinant of the (m + n) × (m + n) matrix

 
an an−1 · · · a0 0 · · · 0
 

 an an−1 · · · a0 · · · 0  

 ··· ··· 

 

 an · · · a0 
 m bm−1 · · · 0 ···
 b b0 0  
 

 bm bm−1 · · · b0 · · · 0  

 ··· ··· 

bm · · · b0

It is not hard to see that there are polynomials A(x), B(x) ∈ R[x] so that

A(x)f (x) + B(x)g(x) = R(f, g).

It is well-known that if the coefficients of f and g lie in a field K such that an bm 6= 0 and f, g
split into factors of degree 1 in K[x], then R(f, g) = 0 if and only if f and g have a common
root (see p. 203 of [5]). In particular, if f (x) ∈ Z[x] has n distinct roots in C, the natural
number R(f, f 0 ) is non-zero. Thus, if f (x) is squarefree, that is, a product of distinct irreducible
polynomials, then R(f, f 0 ) is non-zero.

3. The case n = 1

We can now prove:

Theorem 4. Suppose that f (x) ∈ Z[x] is a polynomial such that f (n) is a perfect square for
every integer n. Then f (x) = g(x)2 for some g(x) ∈ Z[x].

Proof. Since Z[x] is a UFD, we may factor f (x) as a product of irreducible polynomials.
By grouping the even powers of the irreducibles occurring in the factorization, we may write
f (x) = g(x)2 h(x) where h(x) is squarefree, that is, a product of distinct irreducible polynomials.
Let us suppose that the degree of h is ≥ 1. By Lemma 3, h has infinitely many prime divisors
and so we choose one p which is coprime to R(h, h0 ). Thus, there is a natural number n so that
p|h(n). Hence p|f (n). As f (n) is a perfect square, we see that the power of p dividing f (n) is an
even power. The same must be true for h(n). Thus, p2 |h(n). Now, h(n + tp) ≡ h(n)(mod p) so
that by the same reasoning, we deduce p2 |h(n+tp). However, h(n+tp) = h(n)+pth0 (n)(mod p2 )
so that p|h0 (n) if we choose t coprime to p (such as t = 1 say). In particular, p|R(h, h0 ) contrary
to our choice of p. Thus, the degree of h is zero so that h must be a constant. This constant
must be a square.
4 POLYNOMIALS ASSUMING SQUARE VALUES

4. The higher dimensional case

Before we begin the discussion of the multi-variable case, it is useful to recall the classical
theorem that if R is a UFD then so is R[x]. In particular, Z[x1 , ..., xn ] = Z[x1 , ..., xn−1 ][xn ] is a
UFD. Given two polynomials f, g ∈ Z[x1 , ..., xn ] we may consider them as polynomials in xn with
coefficients in Z[x1 , ..., xn−1 ]. Thus, we may consider the resultant Rxn (f, g) as a polynomial in
x1 , ..., xn−1 . By taking an algebraic closure of Q(x1 , ..., xn−1 ), we see that any polynomial factors
as a product of linear factors, and we may deduce, as before that Rxn (f, g) = 0 if and only if f
and g have a common root. In particular, if fxn = ∂f /∂xn , then R(f, fxn ) = 0 if and only if f
and ∂f /∂xn have a common root. Thus, if f is squarefree, then the resultant R(f, fxn ) 6= 0.

Theorem 5. (Kojima, 1915) Now suppose that we have f (x1 , ..., xn ) ∈ Z[x1 , ..., xn ] having
the property that every integer specialization of x1 , ..., xn makes f (x1 , ..., xn ) a square. Then,
f (x1 , ..., xn ) is the square of a polynomial.

Proof. We proceed by induction on the number of variables. As before, we may factor

f = g 2 h, with h a squarefree polynomial. If h has fewer than n variables, we are done by
induction. Suppose then that h is a polynomial of n variables. As h is squarefree, the resultant
R(h, hxn ) (which is a polynomial in x1 , ..., xn−1 ) is not identically zero. Thus, we may choose
(x1 , ..., xn−1 ) = (a1 , ..., an−1 ) so that the resultant R := R(h, hxn )(a1 , ..., an−1 ) is not zero. Thus,
h(a1 , ..., an−1 , xn ) is a non-zero polynomial in the single variable xn . For otherwise, its derivative
would also be zero and consequently R would be zero which is not the case. By Lemma 3, this
polynomial has infinitely many prime divisors. Choose a prime p not dividing R. As before,
there is an an so that p|h(a1 , ..., an ). As f (a1 , ..., an ) is a square, we deduce p2 |h(a1 , ..., an ). The
same is true for h(a1 , ..., an + p). But then,

h(a1 , ..., an−1 , an + p) ≡ h(a1 , ..., an ) + phxn (a1 , ..., an ) (mod p2 )

from which we deduce that

p|hxn (a1 , ..., an )

so that p|R, a contradiction.

5. Absolutely irreducible polynomials mod p

We now turn to the question of making Theorem 5 effective. The results we invoke form a
chapter in classical elimination theory and we refer the reader to (pages 177-215) of [11]. We
summarise these results below.
Recall that a polynomial f with coefficients in a field K is said to be absolutely irreducible
if it is irreducible over the algebraic closure of K.
 POLYNOMIALS ASSUMING SQUARE VALUES 5

Let f (x1 , .., xn ) ∈ Z[x1 , .., xn ] be a polynomial of the form g 2 h with h non-constant. Let us
note that a polynomial of the form

y 2 − f (x1 , ..., xn ) (1)

with f (x1 , .., xn ) ∈ Z[x1 , ..., xn ] is irreducible over Q. For if it is reducible, it must be a product of
two factors which are linear in y and this can only happen only if the discriminant 4f (x1 , ..., xn )
is a perfect square, which is not the case, by hypothesis. Thus, (1) is absolutely irreducible over
Q.
Given a polynomial f (x1 , ..., xn ), suppose we can find a prime p such that the number of
solutions (x1 , ..., xn , y) mod p of the congruence

y 2 ≡ f (x1 , ..., xn )(mod p)

is strictly less than pn . Then, there is an n-tuple (a1 , ..., an ) so that

f (a1 , ..., an )

is not a square mod p. Under which conditions can we do this? If the polynomial (1) is absolutely
irreducible, that is irreducible over Fp , then a famous theorem of Lang and Weil[6] allows us
to do this. The question that arises now is if we can find a prime p for which (1) is absolutely
irreducible mod p.
Given a polynomial f with integer coefficients, we denote by ||f || the sum of the absolute
values of its coefficients. Clearly, ||f g|| ≤ ||f ||||g||. However, a more natural height function
for polynomials is given by taking the maximum of the absolute values of the coefficients of f
and denoting this by H(f ). It is evident that H(f ) ≤ ||f || ≤ C(d, n)H(f ), where C(d, n) is a
constant depending on the number of variables and the total degree d. In fact, we may take
 
d+n
C(d, n) = ,
n
by a simple calculation. Thus, the two heights are comparable.

Proposition 6. (Gelfond’s inequality)

H(f1 ) · · · H(fr ) ≤ ed1 +···+dn H(f1 · · · fr ),

where di is the degree in xi of the product f1 · · · fr .

Proof. See page 229 of [3].

Corollary 7. If f |g, then ||f || ≤ c1 (d, n)||g|| for some effectively computable constant c1 (d, n)
depending only on n and the total degree d of g.
6 POLYNOMIALS ASSUMING SQUARE VALUES

How is all this relevant to our situation? As before, let us write f = g 2 h, with h a squarefree
polynomial. Then, by Corollary 7, we have that ||h|| ≤ c1 (d, n)||f ||. We will need to use this
inequality in applying the following theorem due to E. Noether, in our context.

Proposition 8. (E. Noether, 1922) Let K be a field. Given a polynomial

X
f (x1 , ..., xn ) = ai1 ,...,in xi11 · · · xinn ∈ K[x1 , ..., xn ],
i1 +···+in ≤d

there exist forms g1 , ..., gs in variables Ai1 ,··· ,in with i1 + · · · + in ≤ d, such that the polynomial
f (x1 , ..., xn ) is reducible over K or of degree < d if and only if all the s polynomials gj vanish
when we specialize Ai1 ,...,in with ai1 ,...,in . Moreover, if
 
n+d−1
k= ,
n
k
then the degree of gj is bounded by k 2 . These forms depend only on n and d and are independent
of the field K in the sense that if the characteristic of K is zero, they are fixed forms with rational
integer coefficients, while if the characteristic of K is a prime p, they are obtained by reducing
the integral coefficients modulo p. In the case characteristic of K is zero, then
2k
||gj || ≤ 4k , 1 ≤ j ≤ s.

Proof. See page 190 of [11].

Corollary 9. Let k be as in the previous proposition. Given a polynomial f (x1 , ..., xn ) ∈

Z[x1 , ..., xn ], which is absolutely irreducible, there exists a natural number Nf such that
2k
Nf ≤ (4||f ||)k ,

and f is absolutely irreducible mod p for every prime p coprime to Nf .

Proof. By Proposition 8, there is a polynomial gj such that when we specialize to the coefficients
of f , the value of gj is a non-zero integer. We let Nf be the smallest of the absolute values of
the non-zero values among the gj ’s thus obtained. Since the coefficients of f are bounded by
k
H(f ), and the degree of gj is bounded by k 2 , we get
2k 2k
0 < Nf ≤ ||gj ||||f ||k ≤ (4||f ||)k .

Moreover, for every prime p coprime to Nf , we deduce by Noether’s theorem that f is absolutely
irreducible mod p. This completes the proof.

We will need the following effective version of the theorem of Lang and Weil, proved by
Schmidt [11].
 POLYNOMIALS ASSUMING SQUARE VALUES 7

Proposition 10. Let F (x1 , ..., xn ) be a polynomial over Fp which is of total degree d > 0 and
absolutely irreducible. Let N be the number of zeros of F in Fnp . Then,

|N − pn−1 | ≤ pn−2 (ω(p, d) + 2dψ),

where
ω(p, d) = (d − 1)(d − 2)p1/2 + d2 ,
t
and ψ = 2dt2 with t = (d + 1)(d + 2)/2.

6. Proof of Theorem 1

We apply Corollary 9 to the polynomial

F (x1 , ..., xn , y) := y 2 − f (x1 , ..., xn ).

Thus, if p is coprime to NF , F is absolutely irreducible mod p. By Proposition 10, the number

of zeros N mod p satisfies
|N − pn | ≤ pn−1 (ω(p, d) + 2dψ).

Of these zeros, the number with y = 0 can be at most dpn−1 since the number of solutions of

f (x1 , ..., xn ) ≡ 0(mod p),

is bounded by this quantity. So if, we let N ∗ be the number of zeros of F with y 6= 0, we get

|N ∗ − pn | ≤ pn−1 (ω(p, d) + 2dψ + d).

In particular, the number of specializations mod p for which f (x1 , ..., xn ) is a perfect square mod
p is
1 1
≤ pn + pn−1 (ω(p, d) + 2dψ + d)
2 2
n
and this is strictly less than p if

ω(p, d) + 2dψ + d < p.

If we choose
√
p > 6dψ, and p > 3(d − 1)(d − 2),

we see that
ω(p, d) + 2dψ + d < p.

So we need p coprime to NF and p > max(6dψ, 9(d − 1)2 (d − 2)2 ). This completes the proof of
Theorem 1.
8 POLYNOMIALS ASSUMING SQUARE VALUES

7. Proof of Theorem 2

In the proof of Theorem 1, let us consider all specializations of f (a1 , ..., an ) with 0 ≤ ai ≤ H
with H given as in the statement of Theorem 2. We claim that there is a prime p < H which is
coprime to NF and satisfying

p > C(d) := max(6dψ, 9(d − 1)2 (d − 2)2 ).

Indeed, by [12]
X
θ(H) := log p > .998684H
p<H

provided H ≥ 1, 319, 007. Since H = 2(C(d) + log NF ), we see that

X 9
log p > (C(d) + log NF ) − log NF > 0
5
p<H,(p,NF )=1

and so there is a prime of the desired type. This completes the proof.

8. Concluding remarks

It is clear that the bounds obtained in Theorems 1 and 2 are not optimal. However, Propo-
sition 8 is optimal in the following sense. Ruppert [8] has shown that if f ∈ Z[x, y] is absolutely
irreducible and has deg x f = m, deg y f = n and height H(f ) = H, then for any prime p with

p > [m(n + 1)n2 + (m + 1)(n − 1)m2 ]mn+(n−1)/2 H 2mn+n−1 ,

the reduction of f mod p is absolutely irreducible. Moreover, if we assume the Bouniakowsky

conjecture [1] that predicts that for any irreducible polynomial g(t) ∈ Z[t], g(n)/δ is prime for
infinitely many values of n (here, δ = gcd{g(r) : r ∈ Z}), then there are infinitely many abso-
lutely irreducible polynomials f ∈ Z[x, y] which are reducible mod p where p is a prime with
p > H 2m . Thus, in the case n = 2, the power of H in Proposition 8 cannot be substantially
improved, modulo the Buniakowsky conjecture, which is generally believed. A closer examina-
tion of the proof in [8] shows that what is actually proved is that given f ∈ Z[x, y], there is a
non-zero natural number Nf satisfying

Nf < [m(n + 1)n2 + (m + 1)(n − 1)m2 ]mn+(n−1)/2 H 2mn+n−1 ,

such that f is absolutely irreducible mod p whenever p is coprime to Nf . Thus, refining Propo-
sition 8 will not lead to any substantial improvement of our bounds. It may be possible to
improve these bounds by other techniques, partly geometric and partly analytic in nature. This
will be investigated in a future paper.
Concerning the case of higher powers, the methods extend, in principle. However, getting
definitive bounds is not all that straightforward by the methods of this paper. Part of the
 POLYNOMIALS ASSUMING SQUARE VALUES 9

difficulty is first to establish the absolute irreducibility of the polynomial

y k − f (x1 , ..., xn ),

when f (x1 , .., xn ) is not a perfect k-th power. This is easily done if k is prime and then the
argument of this paper easily extends. Thus, one can proceed inductively in this fashion. It is
clear that this will again lead to humongous bounds. Consequently, it is thus desirable to think
of alternate ways in which better bounds can be obtained.
Acknowledgements. This paper was first presented at the Chandigarh conference in honour of
Professor R.P. Bambah organized by Professors Madhu Raka and R.J. Hans-Gill. I thank them
both for their kind hospitality. I would also like to thank Michael Roth, Dinesh Thakur and the
referee for their comments on an earlier version of this paper.

References

[1] V. Buniakowsky, Nouveaux théorèmes relatifs à la distinction des nombres premiers et à la

décomposition des entiers en facteurs, Mém. Acad. Sci. St. Pétersbourg Sci. Math. Phys.,
6 (1857), 305-329.
[2] I. Gerst and J. Brillhart, Prime divisors of polynomials, American Math. Monthly, 78
(1971), 250-266.
[3] M. Hindry and J. Silverman, Diophantine geometry, an introduction, Graduate Texts in
Math., 201 Springer-Verlag, New York, 2000.
[4] T. Kojima, Note on number-theoretical properties of algebraic functions, Tohoku Mathe-
matical Journal, 8 (1915), Series 1, 24-37.
[5] S. Lang, Algebra, Second edition, Addison-Wesley, 1984.
[6] S. Lang and A. Weil, Number of points on varieties in finite fields, American Journal of
Math., 76 (1954), 819-827.
[7] G. Pólya and G. Szegö, Problems and Theorems in Analysis II, translated from the German
by C.E. Billigheimer, Classics in Mathematics, Springer-Verlag, 1998.
[8] W. M. Ruppert, Reducibility of polynomials f (x, y) modulo p, Journal of Number Theory,
77 (1999), 62-70.
[9] A. Schinzel, Polynomials with special regard to reducibility, Cambridge University Press,
2000.
[10] I. Schur, Über die existenz unendlich vieler primzahlen in einiger speziellen arithmetischen
progressionen, S-B Berlin Math. Ges., 11 (1912), 40-50.
[11] W. Schmidt, Equations over finite fields, An elementary approach, Springer Lecture Notes,
536, 1976.
10 POLYNOMIALS ASSUMING SQUARE VALUES

[12] J. B. Rosser and L. Schoenfeld, Sharper bounds for the Chebyshev functions θ(x) and
ψ(x), Math. Comp., 29 (1975), 243-269.
Address of author:
Department of Mathematics,
Queen’s University,
Kingston, Ontario,
K7L 3N6, Canada
e-mail: [email protected]