Extension Fields II: Derivatives and Multiple Roots

3 Derivatives and multiple roots

We begin by recalling the definition of a repeated root.

Definition 3.1. Let F be a field and let α ∈ F . Then there is a unique

integer m ≥ 0 such that (x − α)m divides f but (x − α)m+1 does not divide
f . We define this integer m to be the multiplicity of the root α in f . Note
that, by the correspondence between roots of a polynomial and its linear
factors, α has multiplicity 0 in f , i.e. m = 0 above, ⇐⇒ f (α) 6= 0. More
generally, if α has multiplicity m in f , then f = (x − α)m g with g(α) 6= 0,
and conversely.
If α has multiplicity 1 in f , we call α a simple root of f . If α has
multiplicity m ≥ 2 in f , then we call α a multiple root or repeated root of
f.

We would like to find conditions when a nonconstant polynomial does,

or does not have a multiple root in F or in some extension field E of F . To
do so, we introduce the formal derivative:

Definition 3.2. Let F be a field. Define the function D : F [x] → F [x] by

the formula
Xn n
X
D( ai xi ) = iai xi−1 .
i=0 i=1

Here the notation iai means the ring element i · ai = ai + · · · + ai , with the
| {z }
i times
convention that 0a0 = 0. We usually write D(f ) as Df . Note that either
Df = 0 or deg Df ≤ deg f − 1.

Clearly, the function D is compatible with field extension, in the sense

that, if F ≤ E, then we have D : F [x] → F [x] and D : E[x] → E[x], and
given f ∈ F [x], Df is the same whether we view f as an element of F [x] or
of E[x]. Also, an easy calculation shows that:

13
Proposition 3.3. D : F [x] → F [x] is F -linear.

This result is equivalent to the sum rule: for all f, g ∈ F [x], D(f + g) =
Df + Dg as well as the constant multiple rule: for all f ∈ F [x] and c ∈ F ,
D(cf ) = cDf . Once we know that D is F -linear, it is specified by the fact
D(1) = 0 and, that, for all i > 0, Dxi = ixi−1 . Also, viewing D as a
homomorphism of abelian groups, we can try to compute

Ker D = {f ∈ F [x] : Df = 0}.

Our expectation from calculus is that a function whose derivative is 0 is

a constant. But if char F = p > 0, something strange happens:

Proposition 3.4. If Ker D = {f ∈ F [x] : Df = 0}, then

(
F, if char F = 0;
Ker D = p
F [x ], if char F = p > 0.
Pn
Here F [xp ] = { i=0 ai x
ip : ai ∈ F } is the subring of all polynomials in xp .

Proof. Clearly, f = ni=0 ai xi is in Ker D ⇐⇒ for every i such that the

P
coefficient ai is nonzero, the monomial ixi−1 = 0. In case char F = 0, this
is only possible if i = 0, in other words f ∈ F is a constant polynomial. In
case char F = p > 0, this happens exactly when p|i for every i such that
ai 6= 0. This is equivalent to saying that f is a polynomial in xp .

As is well-known in calculus, D is not a ring homomorphism. In other

words, the derivative of a product of two polynomials is not in general the
product of the derivatives. Instead we have:

Proposition 3.5 (The product rule). For all f, g ∈ F [x],

D(f · g) = Df · g + f · Dg.

Proof. If f = xa and g = xb , then we can verify this directly:

D(xa xb ) = D(xa+b ) = (a + b)xa+b−1 ;

(Dxa )xb + xa (Dxb ) = axa−1 xb + bxa xb−1 = (a + b)xa+b−1 .

The general case follows from this by writing f and g as sums of monomials
and expanding (but is a little messy to write down). Another approach using
formal difference quotients is in the HW.

14
 If R is a ring, a function d : R → R which is an additive homomorphism
(i.e. d(r + s) = d(r) + d(s) for all r, s ∈ R) satisfying d(rs) = d(r)s + rd(s)
for all r, s ∈ R is called a derivation of R. Thus, D is a derivation of F [x].
As a corollary of the product rule, we obtain:
Corollary 3.6 (The power rule). For all f ∈ F [x] and n ∈ N,

D(f )n = n(f )n−1 Df.

Proof. This is an easy induction using the product rule and starting with
the case n = 1 (or 0).

The connection between derivatives and multiple roots is as follows:

Lemma 3.7. Let f ∈ F [x] be a nonconstant polynomial and let E be an
extension field of F . Then α ∈ E is a multiple root of f ⇐⇒ f (α) =
Df (α) = 0.
Proof. Write f = (x − α)m g with m equal to the multiplicity of α in f and
g ∈ F [x] a polynomial such that g(α) 6= 0. If m = 0, then f (α) = g(α) 6= 0.
Otherwise,
Df = m(x − α)m−1 g + (x − α)m Dg.
6 0. If m ≥ 2, then f (α) = Df (α) = 0.
If m = 1, then Df (α) = g(α) =
Thus we see that α ∈ F is a multiple root of f ⇐⇒ m ≥ 2 ⇐⇒
f (α) = Df (α) = 0.

In practice, an (unknown) root of f will only exist in some (unknown)

extension field E of F . We would like to have a criterion for when a polyno-
mial f has some multiple root α in some extension field E of F , without
having to know what E and α are explicitly. In order to find such a criterion,
we begin with the following lemma, which says essentially that divisibility,
greatest common divisors, and relative primality are unchanged after passing
to extension fields.
Lemma 3.8. Let E be an extension field of a field F , and let f, g ∈ F [x],
not both 0.
(i) f g in F [x] ⇐⇒ f g in E[x].

(ii) The polynomial d ∈ F [x] is a gcd of f, g in F [x] ⇐⇒ d is a gcd of

f, g in E[x].

(iii) The polynomials f, g are relatively prime in F [x] ⇐⇒ f, g are rela-

tively prime in E[x].

15
Proof. (i): =⇒ : obvious. ⇐= : We can assume that f 6= 0, since
otherwise f g (in either F [x] or E[x]) ⇐⇒ g = 0. Suppose that f g in
E[x], i.e. that g = f h for some h ∈ E[x]. We must show that h ∈ F [x].
By long division with remainder in F [x], there exist q, r ∈ F [x] with either
r = 0 or deg r < deg f , such that g = f q + r. Now, in E[x], we have both
g = f h and g = f q + r. By uniqueness of long division with remainder in
E[x], we must have h = q (and r = 0). In particular, h = q ∈ F [x], as
claimed.
(ii): =⇒ : Let d ∈ F [x] be a gcd of f, g in F [x]. Then, by (i), since d f ,
d g in F [x], d f , d g in E[x]. Moreover, there exist a, b ∈ F [x] such that
d = af + bg. Now suppose that e ∈ E[x] and that e f , e g in E[x]. Then
e af + bg = d. It follows that d satisfies the properties of being a gcd in
E[x]. ⇐= : Let d ∈ F [x] be a gcd of f, g in E[x]. Then d f , d g in E[x],
hence by (i) d f , d g in F [x]. Suppose that e ∈ F [x] and that e f , e g in
F [x]. Then e f , e g in E[x]. Hence e d in E[x]. Since both e, d ∈ F [x], it
again follows by (i) that e d in F [x]. Thus d is a gcd of f, g in F [x].
(iii): The polynomials f, g are relatively prime in F [x] ⇐⇒ 1 ∈ F [x] is
a gcd of f and g in F [x] ⇐⇒ 1 ∈ F [x] is a gcd of f and g in F [x], by (ii),
⇐⇒ f, g are relatively prime in E[x].

Corollary 3.9. Let f ∈ F [x] be a nonconstant polynomial. Then there

exists an extension field E of F and a multiple root of f in E ⇐⇒ f and
Df are not relatively prime in F [x].
Proof. =⇒ : If E and α exist, then, by Lemma 3.7, f and Df have a
common factor x − α in E[x] and hence are not relatively prime. Thus by
Lemma 3.8 f and Df are not relatively prime in F [x].
⇐= : Suppose that f and Df are not relatively prime in F [x], and let
g be a common nonconstant factor of f and Df . There exists an extension
field E of F and an α ∈ E which is a root of g. Then α is a common root
of f and Df , and hence a multiple root of f .

We now apply the above to an irreducible polynomial f ∈ F [x].

Corollary 3.10. Let f ∈ F [x] be an irreducible polynomial. Then there
exists an extension field E of F and a multiple root of f in E ⇐⇒ Df = 0.
Proof. =⇒ : By the previous corollary, if there exists an extension field
E of F and a multiple root of f in E, then f and Df are not relatively
prime in F [x]. In this case, since f is irreducible, it must be that f divides
Df . Hence, if Df 6= 0, then deg Df ≥ deg f . But we have seen that either
deg Df < deg f or Df = 0. Thus, we must have Df = 0.

16
 ⇐= : Clearly, if Df = 0, then f is a gcd of f and Df , hence f and Df
are not relatively prime in F [x].

Corollary 3.11. Let F be a field of characteristic 0 and let f ∈ F [x] be an

irreducible polynomial. Then there does not exist an extension field E of F
and a multiple root of f in E. In particular, if E is an extension field of F
such that f factors into linear factors in E, say

f = c(x − α1 ) · · · (x − αn ),

then the αi are distinct, i.e. for i 6= j, αi 6= αj .

If char F = p > 0, then it is possible for an irreducible polynomial

f ∈ F [x] to have a multiple root in some extension field, but it takes a little
effort to produce such examples. For example, it is not possible to find such
an example for a finite field. The basic example arises as follows: consider
the field Fp (t), where t is an indeterminate (here we could replace Fp by
any field of characteristic p). Then t is not a pth power in Fp (t), and in fact
one can show that the polynomial xp − t is irreducible in Fp (t)[x]. Let E
be an extension field of Fp (t) which contains a root α of xp − t, so that by
definition αp = t. Then

xp − t = xp − αp = (x − α)p ,

because we are in characteristic p. Thus α is a multiple root of xp − t, of

multiplicity p.
The key property of the field Fp (t) which made the above example work
was that t was not a pth power in Fp (t). More generally, define a field F
of characteristic p to be perfect if every element of F is a pth power, or
equivalently if the Frobenius homomorphism σp : F → F is surjective. For
example, we shall show below that a finite field is perfect. An algebraically
closed field is also perfect. We also declare every field of characteristic zero
to be perfect. By a problem on HW, if F is a perfect field and f ∈ F [x] is
an irreducible polynomial, then there does not exist an extension field E of
F and a multiple root of f in E.

17