9.1 Equivalent Forms of Hensel's Lemma

Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

Chapter 9

Hensels Lemma
9.1 Equivalent forms of Hensels Lemma
For a valued eld K = (K, v), the property of being henselian is equivalent to a variety
of criteria for polynomials f O
K
[X] to admit a zero in O
K
. Recall that Lemma 5.7 has
shown that the residue map induces a bijection between the integral roots of f in

K and
the roots of f in

K (counted with multiplicity). In view of this, the question arises under
which conditions a root of f lying in K is the residue of some root of f which lies in K. If
b K is such that b is a root of f, then b could be thought of as an approximative root.
Hensels original lemma showed how an approximative root can be rened to a root of
f. This renement procedure works in every spherically complete eld (see Theorem 7.44
and Exercise ??). However, Hensels Lemma turned out to be one of a lot of properties of a
valued eld which are equivalent to the property of being henselian, the proof not anymore
involving a renement procedure. Note that the condition f O
K
[X] implies that also the
derivative f

and all higher derivatives of f are polynomials in O


K
[X] (cf. Lemma 24.59).
Since the residue map is a ring homomorphism, we have f

= f

(and the same holds for
the higher derivatives).
Theorem 9.1 For a valued eld K = (K, v), the property of being henselian is equivalent
to each of the following properties:
1) If f = 1+X+X
2
g(X) with g(X) M
K
[X], then f admits a root in O
K
whose residue
is 1.
2) Every monic polynomial f = X
n
+ c
n1
X
n1
+ . . . + c
1
X + c
0
O
K
[X] with c
n1
= 0
and c
n2
= . . . = c
0
= 0 admits a linear factor X + c in O
K
[X] such that c = c
n1
.
3) For every monic polynomial f O
K
[X] the following holds: if f has a simple root
b K, then f admits a root a O
K
such that a = b.
4) (Hensels Lemma)
For every monic polynomial f O
K
[X] the following holds: if b O
K
satises vf(b) > 0
and vf

(b) = 0, then f admits a root a O


K
such that a = b.
5) (Newtons Lemma)
For every monic polynomial f O
K
[X] the following holds: if b O
K
satises vf(b) >
2vf

(b), then f admits a root a O


K
such that v(a b) > vf

(b).
6) (strong Hensels Lemma)
215
216 CHAPTER 9. HENSELS LEMMA
For arbitrary polynomials f O
K
[X] the following holds: if there is a factorization f = gh
such that g is relatively prime to h in the polynomial ring K[X], then there exist polynomials
g, h O
K
[X] reducing to g, h respectively and such that f = gh and deg g = deg g.
7) (Krasners Lemma)
Assume v to be extended to the separable closure K
sep
of K. Then for every element
a K
sep
the following holds: if b K
sep
\ K satises
v(a b) > max{v(a a) | Gal K a = a} , (9.1)
then a K(b).
3

), 4

), 5

) The same as 3), 4), 5) respectively, but without the hypothesis that f be
monic.
2

), 3

), 4

), 5

) The same as 2), 3), 4), 5) respectively, but with the additional hypothesis
that f be separable over K.
Proof: During this proof, let f = c
n
X
n
+ c
n1
X
n1
+ . . . + c
0
always be a
polynomial in K[X] with roots a
1
, . . . , a
n


K, so that
f = c
n
n

i=1
(X a
i
) .
We show that 6) holds in every henselian eld (K, v). This eld admits a unique extension
of v to

K which we will again denote by v. Hence, v = v for all Gal K. Now let f, g
and h satisfy the hypothesis of 6). Observe that the condition that g be prime to h implies
that g = 0 = h, so f = gh = 0. We dene
g :=

iJ
(X a
i
)

K[X] with J = {i | 1 i n va
i
0 g(a
i
) = 0} .
Since (K, v) is assumed to be henselian, from (7.2) we know that every Gal K induces
an automorphism K (the reduction of ) via a = a. For all Gal K and
all i J we have that va
i
= va
i
0 and that a
i
is again a root of g, which yields
that a
i
is again a root of g. So the set of roots of g is closed under the application of
all Gal K. Since g is assumed to be prime to h, our construction yields that g and

h = c
n

i / J
(X a
i
)

K[X] have no common root in

K. Since f = g

h K[X], part b)
of Lemma 24.1 now shows that g,

h K[X].
We know from Lemma 5.7 that for every root a of g of multiplicity m, there are at
least m roots (counted with multiplicity) of f whose residue is equal to a. From our choice
of g it thus follows that g divides g =

iJ
(X a
i
). On the other hand, f = g

h yields
that g divides f = gh. By assumption, g and h are prime to each other, so by part a) of
Lemma 24.1 they have no common root in

K. Since by our choice of g, every root of g
is also a root of g, this shows that g must divide g. This proves that deg g = deg g and
that there is c O

K
such that g := c g O
K
[X] reduces to g. We nd that f = gh with
h = c
1

h O
K
[X] by virtue of Gau Lemma.
If 6) holds, then also 1) holds: If f = 1 + X + X
2
g(X) with g(X) M
K
[X], then
the reduction of f is just f = 1 +X =: g , so the factor g obtained by 6) is linear, and the
corresponding root of f has residue 1.
9.1. EQUIVALENT FORMS OF HENSELS LEMMA 217
If 1) holds, then also 5

) and thus also 5) and 5

) hold: Let f, b satisfy the assumptions


of 5). We may assume that f(b) = 0 because otherwise we are done. Then it follows from
the assumption vf(b) > 2vf

(b) that also f

(b) = 0. We employ a substitution used by


P. Ribenboim in [RIB17]. From the Taylor Expansion (24.13) we infer the equality
f(X + Y ) = f(X) + f

(X)Y + g(X, Y )Y
2
with g(X, Y ) O
K
[X, Y ]. We set X = b and Y = f(b)f

(b)
1
Z with Z a new variable.
We obtain
f(b + f(b)f

(b)
1
Z) = f(b) + f(b)Z +
f(b)
2
f

(b)
2
Z
2
g(b, f(b)f

(b)
1
Z) .
We have g(b, f(b)f

(b)
1
Z) O
K
[Z]. Since f(b)/f

(b)
2
is an element of M
K
by hypothesis,
g(Z) := f(b)f

(b)
2
g(b, f(b)f

(b)
1
Z) M
K
[Z] .
We obtain
h(Z) :=
f(b + f(b)f

(b)
1
Z)
f(b)
= 1 + Z + Z
2
g(Z) .
By 1), h admits a root a

O
K
such that a

= 1 and thus, va

= 0. Consequently,
a := b + f(b)f

(b)
1
a

is a root of f, and it satises v(a b) = v(f(b)f

(b)
1
a

) > vf

(b).
Furthermore, vf

(b) 0 since f

O
K
[X] and b O
K
.
Hensels Lemma in the versions 4), 4

), 4

) are special cases of Newtons Lemma 5),


5

), 5

) respectively, because a = b follows from v(a b) > 0.


Property 4) and property 3) are equivalent since their hypotheses are, and similarly
for 4

) and 3

) as well as for 4

) and 3

). Indeed, for b O
K
, b is a simple root of f if
and only if f(b) = 0 and f

(b) = 0. Since f(b) = f(b) and f

(b) = f

(b), this in turn is


equivalent to vf(b) > 0 and vf

(b) = 0.
If property 3) or 3

) holds, then also 2) holds, and if property 3

) holds, then also


2

) holds. For, if f = X
n
+ c
n1
X
n1
+ . . . + c
1
X + c
0
O
K
[X] with c
n1
= 0 and
c
n2
= . . . = c
0
= 0, then f = X
n
+ c
n1
X
n1
admits c
n1
as a simple root.
If property 2) or 2

) holds, then (K, v) is henselian: Assume that (K, v) admits more


than one extension of v to

K. Then there is already a nite Galois extension L|K admitting
more than one extension of v from K to L. Hence, Z := (L|K, v)
d
is a proper extension
of K. Let
1
= id,
2
, . . . ,
n
be the distinct embeddings of Z into L. From Theorem 7.9
we know that v
i
= v on Z for all i = 1. From Lemma 6.60 we infer the existence of some
a L such that va = 0 and v
i
a > 0 for all i = 1. Consequently, a / K is the only root
of value 0 of its minimal polynomial over K, and all other roots have value > 0. This
minimal polynomial is thus of the form as in the hypothesis of 2), but it is irreducible. We
have shown that 2) can not hold if (K, v) is not henselian.
If (K, v) is henselian, then Krasners Lemma holds: Assume that a, b K
sep
satisfy
the hypothesis of Krasners Lemma. We will show that a is xed by every automorphism
Gal K(b); since K(b) is the xed eld of Gal K(b) in K
sep
and a K
sep
, this will yield
that a K(b). Note that b = b because of Gal K(b). Since (K, v) is henselian, the
same holds for (K(b), v) by Lemma 7.33, hence we have
v(b a) = v(b a) = v(b a) = v(a b) > max{v(a a) | Gal (K) a = a} .
218 CHAPTER 9. HENSELS LEMMA
It follows that
v(a a) min{v(a b), v(b a)} > max{v(a a) | Gal K a = a}
which yields that a = a.
Krasners Lemma implies property 2): Assume that f satises the assumption of 2).
Then in view of Lemma 5.7, there is precisely one root of f in

K with residue c
n1
, say
a , and all other roots have residue zero. Observe that f is separable because c
n1
= 0
and not both n and n 1 can be divisible by the characteristic. Hence, a K
sep
. Since
all conjugates of a over K are among these roots, we nd that v(a a) = va = 0 for all
Gal K such that a = a. On the other hand, v(a + c
n1
) > 0, hence by Krasners
Lemma, a K(c
n1
) = K. Now the desired linear factor is X + c with c = a.
Remark 9.2 In Hensels Lemma, the root a satisfying a = b is uniquely determined. This follows from
Lemma 5.7 and the hypothesis that b be a simple root of f. The same holds for Newtons Lemma:
Let f O
K
[X] and b O
K
such that vf(b) > 2vf

(b). Then there is a unique root a



K of f such that
v(a b) > vf

(b), and if a = b, then v(a b) = vf(b) vf

(b) > vf

(b).
This is seen as follows. From the Taylor expansion (cf. Lemma 24.59), we know that 0 = f(a) = f(b) +
f

(b)(ab) +

h(b, a)(ab)
2
with

h(b, a) O
K
. Hence, f(b) = f

(b)(ab)

h(b, a)(ab)
2
. Now assume
that a = b, which means that v(a b) < . From our hypothesis that v(a b) > vf

(b), it then follows


that vf

(b)(ab) < v

h(b, a)(ab)
2
. Consequently, vf(b) = vf

(b)(ab), that is, v(ab) = vf(b) vf

(b).
Suppose that also a



K is a root of f with v(a

b) > vf

(b). Then v(a

a) min{v(a

b), v(ab)} >


vf

(b). By the Taylor expansion, 0 = f(a

) f(a) = f

(a)(a

a) +

h(a, a

)(a

a)
2
with

h(a, a

) O
K
.
From v(a b) > vf

(b), it also follows that vf

(a) = vf

(b). Hence if a

= a, then we would have


vf

(a)(a

a) = vf

(b)(a

a) < v

h(a, a

)(a

a)
2
, and it would follow that v0 = vf

(a)(a

a) < , a
contradiction. Hence a

= a, as asserted.
Let us give a rst application of Hensels Lemma.
Example 9.3 Consider the Artin-Schreier polynomial f = X
p
X c over the henselian
eld (K, v). If vc > 0, then f = X
p
X. This polynomial admits 0 and 1 as simple roots,
hence by Hensels Lemma, f admits roots in K with residue 0 and 1. If char K = p, then
X
p
X admits precisely the elements of the prime eld F
p
as its roots, that is, it splits
completely over F
p
. In this case, Hensels Lemma shows that f splits completely over K,
the residues of its roots being precisely the elements of F
p
.
If vc = 0 then f = X
p
Xc with c = 0. If this polynomial admits a root in K and
char K = p, then it splits completely over K, the roots being + i, i F
p
. In this case
again by Hensels Lemma, f splits completely over K, the residues of its roots being just
+ i, i F
p
. On the other hand, if f does not admit a root in K then f does not admit
a root in K.
By the preceding discussion, we see that an irreducible Artin-Schreier polynomial f =
X
p
X c over a henselian eld (K, v) with Artin-Schreier closed residue eld K must
satisfy vc < 0. The case of vc < 0 has already appeared in Lemma 6.39, Lemma 6.40 and
Lemma 6.41 and will reappear in Examples 11.44, 11.47, 11.56 and 11.59. It will play an
important role in several later chapters of this book.
Recall that (X) denotes the Artin-Schreier polynomial X
p
X. Further, we set
(S) := {(a) | a S} for every S K. From the above considerations, we conclude:
Lemma 9.4 Let (K, v) be as above. Then M
K
= (M
K
) (K). If in addition K is
Artin-Schreier closed, then O
K
= (O
K
) (K).
9.2. HENSELIAN FIELDS 219
Notes 9.5 The proof that 6) holds in a henselian eld is due to Rayner [RAY2]. The proof that 2) implies
henselian is due to Nagata [NAG1]. Ribenboims substitution which we have employed for the proof of
1)5) is a renement of a substitution that was used by Iversen in [IVER] to prove the equivalence of
Hensels Lemma with Newtons Lemma. This trick was apparently rediscovered by L. v. d. Dries in his
thesis [VDD1]. It works as follows: Suppose that f and b satisfy the hypothesis of Newtons Lemma. Set
g(X) := f(b + f

(b)X)/f

(b)
2
. Then vg(0) = vf(b) 2vf

(b) > 0. Further, g

(0) = 1 by virtue of the


chain rule, and thus, vg

(0) = 0. Then by Hensels Lemma, there is a M


K
such that g( a) = 0. Now
a := b +f

(b) a is a root of f which satises v(a b) > vf

(b).
We have found the name Hensels Lemma for property 4) as well as for property 6) or slightly dif-
ferent versions of them. We prefer to use it for property 4). Originally, Hensels Lemma is the lemma which
states that every complete valued eld of rank 1 satises property 4) resp. property 6) (cf. Theorem 9.6
below).
See P. Ribenboim [RIB17] for more equivalent forms of Hensels Lemma, alternative proofs and a
dierent, possibly historically more adequate naming of the various forms.
Exercise 9.1 Show the equivalence of 1) of Theorem 9.1 with a weaker version of 2) where c
n1
= 1
by means of a suitable substitution. What happens to the zeros under this substitution? Try to deduce
henselian from this weaker version of 2); what is the problem?
Exercise 9.2 Show that K is henselian if and only if it satises the following property: If f O
K
[X] is
irreducible over K and f is not constant, then f = c
m
where c O
K
and K[X] is an irreducible
polynomial such that deg f = mdeg .
9.2 Henselian elds
Although we know already from Theorem 7.44 as well as from Theorem 11.27 that every
spherically complete valued eld is henselian, we will now give alternative proofs on the
basis of Theorem 9.1. Let (K, v) be spherically complete.
First proof: We prove that Newtons Lemma (property 5) of Theorem 9.1) holds in
(K, v). Take a polynomial f O[X] and b O such that vf(b) > 2vf

(b). In particular,
we have that vf

(b) = , that is, s := f

(b) = 0. From part a) of Theorem 5.15 we


know that f induces a pseudo-linear isomorphism of ultrametric spaces from b +sM onto
f(b) + s
2
M. The hypothesis vf(b) > 2vf

(b) = vs
2
implies that f(b) s
2
M, whence
0 f(b) +s
2
M. Thus, there is a b +sM O such that f(a) = 0. This element satises
v(a b) > vs = vf

(b).
Second proof: Part b) of Theorem 5.15 can be used to prove a generalization of Newtons
Lemma: the multidimensional Newtons Lemma. We will do this in Theorem 9.11 below.
The multidimensional Newtons Lemma implies the one-dimensional Newtons Lemma.

Third proof: Following an idea of S. Prie-Crampe, we employ the Ultrametric Fixed


Point Theorem to prove that (K, v) satises property 1) of Theorem 9.1. So let f =
1 + X + X
2
g(X) with g(X) M[X]. For every b K, set
b := b f(b) = 1 b
2
g(b) . (9.2)
By assumption, all coecients of g have value > 0, hence vb
2
g(b) > 0 for all b O. It
follows that sends O into 1 +M. Moreover, we can show that is contractive on O.
For b, c O, we have
b c = c
2
g(c) b
2
g(b) .
220 CHAPTER 9. HENSELS LEMMA
Applying Lemma 24.59 with R = M to the polynomial X
2
g(X) M[X], we nd that
there exists G(X, Z) M[X, Z] such that
Z
2
g(Z) X
2
g(X) = (Z X)G(X, Z) .
Evaluating with Z = b and X = c, we nd G(c, b) M, whence v(b c) = v(c
2
g(c)
b
2
g(b)) > v(b c). This proves to be contractive on O. In view of Lemma ?? and our
assumption that (K, v) be spherically complete, we may infer from the Ultrametric Fixed
Point Theorem 1.12 that has a xed point a O. It satises a = a = a f(a) which
is the same as f(a) = 0. Hence a is a root of f and it satises a = 1 since it lies in
1 +M.
Fourth proof: The idea is to employ the results of Section 5.6 to prove that (K, v)
satises Newtons Lemma. Take f O[X] and b O such that vf(b) > 2vf

(b). For every


c b + f

(b)M, we set
c := c
f(c)
f

(b)
.
Since c b + f

(b)M, we have that v(c b) > vf

(b). Hence,
vf(c) = v(f(b) + f

(b)(c b) + (c b)
2
H
f
(b, c))
min{vf(b), vf

(b) + v(c b), 2v(c b) + vH


f
(b, c)}
> 2vf

(b)
This implies that
v(c c) = vf(c) vf

(b) > vf

(b) ,
and consequently,
c b + f

(b)M .
We show that : b + f

(b)M b + f

(b)M is contractive. Take any c, d b + f

(b)M.
Then (5.11), with c, d in place of y, z respectively, shows that
v (c d) = v
_
c d
f(c) f(d)
f

(b)
_
= v(f(c) f(d) f

(b)(c d)) vf

(b)
> vf

(b)(c d) vf

(b) = v(c d) .
Since the ball b + f

(b)M is spherically complete by Lemma 1.21, the Ultrametric Fixed


Point Theorem 1.12 shows that has a xed point a in b + f

(b)M. This means that


0 = a a =
f(a)
f

(b)
and consequently, f(a) = 0. Since a b + f

(b)M, we have that v(a b) > vf

(b). We
have thus proved that (K, v) satises Newtons Lemma.
Historically, the rst examples for henselian elds have been the complete elds of
rank 1. Recall that rank 1 just says that the value group is archimedean. If the value
group is not isomorphic to Z, then a complete eld may not be spherically complete (cf.
Example 11.50). So the above proofs do not work in this case. But Theorem 5.16 shows
that every complete eld of rank 1 satises Newtons Lemma. So we have:
9.3. THE KRASNER CONSTANT 221
Theorem 9.6 Every complete eld of rank 1 is henselian. If (K, v) is a valued eld of
rank 1, then its completion contains a henselization.
Here is an alternate proof for this theorem:
Let (K, v) be a complete eld of rank 1, and let the notation be as in the rst of the above
two proofs. Since G(X, Z) M[X, Z], we may write G(X, Z) = c

G(X, Z) with c M
and

G(X, Z) O[X, Z] (for instance, c may be taken to be the coecient of smallest value
in G). Then we have

G(b, a) O and v(a b) = v(ag(a) bg(b)) vc + v(a b).
By induction on n we nd that v(
n+1
a
n
a) nvc + v(a a) nvc. Since vK is
archimedean, for every vK there is some n N such that nvc > . Now we can use
Lemma 1.13 to infer the existence of the desired xed point.
The fact that a valued eld is henselian if and only if it satises some form of Hensels
Lemma as given in Theorem 9.1 can also be used for the construction of new classes of
henselian elds which may not even be complete.
Lemma 9.7 Take an ascending chain ((K

, v))
<
of henselian elds. Then their union
(K, v) is again a henselian eld.
Proof: Clearly, the union is a valued eld. In order to show that it is henselian, it suces
to prove that it has property 4) of Theorem 9.1, that is, it satises Hensels Lemma. Let
f K[X] and b K satisfy the assumptions of Hensels Lemma. Since K =

<
K

,
there must be some
f
< such that K

f
contains the nitely many coecients of f and
the element b. Since K

f
is henselian, we know from Theorem 9.1 that it satises Hensels
Lemma. Thus, the required root a of f can already be found in the subeld K

f
of K.
This proves that (K, v) is henselian.
Theorem 9.8 Every Puiseux series eld K over k is henselian with respect to its canonical
valuation v
t
.
Proof: K

f
is henselian with respect to its t
n
i
-adic valuation (cf. Corollary 11.28). Again
by Theorem 9.1, K
n
i
satises Hensels Lemma. Thus, the required root a of f can already
be found in the subeld K
n
i
of K. This proves that (K, v
t
) is henselian.
Another way of obtaining new henselian elds, again using the equivalence of being
henselian with satisfying Hensels Lemma, will be presented in Section 9.10.
9.3 The Krasner constant
Let (K, v) be any valued eld. If a

K \ K is not purely inseparable over K, we choose
some extension of v from K to

K and dene
kras(a, K) := max{v(a a) | , Gal K and a = a} v

K
and call it the Krasner constant of a over K. Since all extensions of v from K to

K
are conjugate, this does not depend on the choice of the particular extension of v. For the
222 CHAPTER 9. HENSELS LEMMA
same reason, over a henselian eld (K, v) our Krasner constant kras(a, K) coincides with
the Krasner constant
max{v(a a) | Gal K and a = a}
which already appeared in (9.1).
The following theorem states an alternate version of the original Krasners Lemma
(property 7) of Theorem 9.1).
Theorem 9.9 Assume that K(a)|K is a separable-algebraic extension and (K(a, b), v) is
any (possibly transcendental) extension of (K, v) such that
v(b a) > kras(a, K) . (9.3)
Then for every extension of v from K(a, b) to its algebraic closure

K(a, b) =

K(b), the
element a lies in the henselization of (K(b), v) in (

K(b), v).
Proof: Take any extension of v from K(a, b) to

K(b) and denote by K(b)
h
the
henselization of (K(b), v) in (

K(b), v). Since a is separable-algebraic over K, it is also


separable-algebraic over K(b)
h
. Since for every Gal K(b)
h
we have that a = |

K
a and
|

K
Gal K, we nd that
{v(a a) | Gal K(b)
h
and a = a}
{v(a a) | Gal K and a = a}
{v(a a) | , Gal K and a = a} .
This implies that
kras(a, K(b)
h
) kras(a, K) ,
and consequently, v(b a) > kras(a, K(b)
h
). Now a K(b)
h
follows from the original
Krasners Lemma.
9.4 The Hensel-Rychlik Property
We will now add two further properties to the list of equivalent forms of Hensels Lemma.
For an arbitrary polynomial f of degree n with roots a
1
, . . . , a
n
and leading coecient c
n
,
we dene the discriminant of f to be
discr f := (1)
n(n1)
2
c
2n2
n

1i<jn
(a
i
a
j
)
2
= c
2n2
n

i=j
(a
i
a
j
) . (9.4)
We will also need the following notion. Two valuations v and w of a eld K are called
independent if they do not admit a non-trivial common coarsening, that is, if there is no
non-trivial valuation ring O of K which contains both O
v
and O
w
.
9.4. THE HENSEL-RYCHLIK PROPERTY 223
Theorem 9.10 For a valued eld K = (K, v), the property of being henselian is equivalent
to each of the following properties:
1) (strong Hensel-Rychlik)
For every polynomial f O
K
[X] the following holds: if some b K satises vf(b) >
v discr f, then f admits a root in K.
2) (weak Hensel-Rychlik)
For every monic polynomial f O
K
[X] the following holds: if some b K satises
vf(b) > v discr f, then f admits a root in K.
Proof: Every henselian eld (K, v) satises property 1):
The following proof is due to E. Ehrhardt, a student of F. Lorenz (cf. [LOR2], pp. 112113).
Assume that f satises the assumption of 1), and let c denote its leading coecient. Let
n = deg f and a
1
, . . . , a
n
be the roots of f in

K, enumerated such that va
1
va
2
. . .
va
n
. Choose m n such that a
m+1
, . . . , a
n
are precisely the integral roots of f. Since f
has integral coecients, it follows from part b) of Lemma 5.6 that vc+va
1
+. . . +va
m
0.
Hence, vc + va
1
+ . . . + va
i
vc + va
1
+ . . . + va
m
0 for every i. We write
discr f =
n

i=2
d
i
with d
i
:= c
2
i1

j=1
(a
i
a
j
)
2
.
From our enumeration of the roots it follows that v(a
i
a
j
) va
j
for every j i. Hence
for every i,
vd
i
= 2vc +
i1

j=1
2v(a
i
a
j
) 2(vc +
i1

j=1
va
j
) 0 . (9.5)
Further, for every i and k such that 1 i, k, n,
vc +
i1

j=1
j=k
v(a
i
a
j
) vc +
i1

j=1
j=k
va
j
vc +
m

j=1
va
j
0 , (9.6)
because a
1
, . . . , a
m
are precisely the roots of negative value by assumption.
Now suppose that there are two indeces k < such that v(ba
k
) = v(ba

) v(ba
i
)
for every i. Note that v(a

a
i
) v(b a
i
) for every i. From (9.6), we obtain that
vd

= v(c
1

j=1
j=k
(a

a
j
)) + v(c(a

a
k
)
2
1

j=1
j=k
(a

a
j
)) v(c(a

a
k
)
2
1

j=1
j=k
(a

a
j
))
and for i > ,
vd
i
= v(c
i1

j=1
(a
i
a
j
)) + v(c
i1

j=1
j=
(a
i
a
j
)) + v(a
i
a

) v(a
i
a

) = v(a

a
i
) .
With these inequalities and with vd
i
0 for i < , which we know from (9.5), we deduce:
v discr f v

i
d
i
v(c(a

a
k
)
2
n

j=1
j=k,
(a

a
j
))
v(c(b a
k
)(b a

)
n

j=1
j=k,
(b a
j
)) = vf(b) .
224 CHAPTER 9. HENSELS LEMMA
But this contradicts our assumption that vf(b) > vdiscr f. Hence, there is an index k such
that v(a
k
b) > v(a
i
b) for all i = k. All conjugates of a
k
over K are among the a
i
.
Since v discr f < , we have that a
i
= a
j
for i = j. This shows that a
k
is separable over
K. Since Krasners Lemma holds in every henselian eld by Theorem 9.1, we nd that
a
k
K(b) = K, which shows that f admits a root in K.
The implication 1)2) is trivial. Now assume that (K, v) is not henselian. We wish to
show that 2) does not hold. Let L|K be a nite Galois extension which admits at least two
extensions of v from K to L. Let v
1
, . . . , v
n
be all extensions of v to L. If w
i
is a coarsening
of v
i
then let w
i
denote the valuation induced by v
i
on Lw
i
(such that v
i
= w
i
w
i
), let
w denote the restriction of w
i
to K and w the restriction of w
i
to Kw. We are going to
prove that there is some i and a coarsening w
i
of v
i
such that all extensions of w from Kw
to Lw
i
are independent.
If already v
1
, . . . , v
n
are independent, then there is nothing to show. Otherwise, we
have a nonempty set O of all non-trivial valuation rings of L which contain at least two
valuation rings O
v
i
and O
v
j
. Observe that the intersection over every descending chain of
rings O

, < , (O

for ) is again a member of O. Indeed, it is a valuation


ring since every intersection over a descending chain of valuation rings is a valuation ring,
and it satises the above condition since there are only nitely many valuation rings O
v
i
.
By Zorns Lemma, it follows that O has minimal elements. Let O be such a minimal
element in O. After a suitable renumbering, we can assume that precisely O
v
1
, . . . , O
v
m
are the valuation rings contained in O (1 < m n). Let w
1
be the common coarsening of
v
1
, . . . , v
m
such that O = O
w
1
, and let w
1
, . . . , w
m
be the valuations induced by v
1
, . . . , v
m
on Lw
1
. Further, let w be the restriction of w
1
to K and let w denote the valuation induced
by v on Kw. Then w
1
, . . . , w
m
are all extensions of w from Kw to Lw
1
. Indeed, if w

is
an extension of w from Kw to Lw
1
, then v

= w
1
w

is an extension of v from K to L
which satises O
v
O and thus, v

= v
i
for some i m. By the minimality of O, there
is no valuation ring properly contained in O which contains more than one of these rings.
Consequently, there is no proper valuation ring of Lw
1
which contains more than one O
w
i
.
This shows that the extensions w
1
, . . . , w
m
are independent, and our claim is proved.
Note that by Lemma 6.61, the extension Lw
1
|Kw is normal. Let Z denote the decom-
position eld of (Lw
1
|Kw, w
1
). Since m 2, Z is a proper extension of Kw. Since it is a
separable extension, we can choose a primitive element a such that Z = Kw(a). W.l.o.g. we
can choose a such that w
i
a 0 for all i. Let f O
(Kw,w)
[X] be the minimal polynomial of
a over Kw. Set := wdiscr f. Further, let
1
= id,
2
, . . . ,
m
be the distinct embeddings
of Z into L. Since the valuations w
1
, . . . , w
m
are independent on Lw
1
, it follows that also
the extensions of w from Kw to Z are independent. From Theorem 7.9 we can thus infer
that w
1

i
= w
1
on Z for all i = 1. From the Strong Approximation Theorem (cf. [ZA-SA2],
10, Theorem 18

) we infer the existence of some b


1
Z such that such that w
1
(ab
1
) >
and w
1

i
b
1
> for all i = 1. For b := Tr
Z|Kw
1
(b
1
) we nd:
w
1
(a b) = w
1
_
a b
1

i>1

i
b
1
_
min{w
1
(a b
1
), w
1

i
b
1
| i = 2, . . . , m} > .
Since w
1
b 0 and thus also w
1
(
i
a b) 0 for all i, we nd that
w
1
f(b) = w
1
(a b) +

i>1
w
1
(
i
a b) > = wdiscr f .
9.5. THE MULTIDIMENSIONAL NEWTONS LEMMA 225
Now choose a monic polynomial f O
K
[X] whose w-reduction fw is f. Further, choose
b K such that bw = b. It follows that vf(b) > v discr f. But f is irreducible over K
since f is irreducible over Kw. This proves that 2) does not hold.
Note that in contrast to Newtons Lemma, the Hensel-Rychlik property gives a lower bound
for the value of vf(b) which does not depend on the approximative root b. Property 2) is
the version of the Hensel-Rychlik property used by Ax and Kochen in their important
paper [AXKOC1].
9.5 The multidimensional Newtons Lemma
In this section, we will use to show that a multidimensional Newtons Lemma holds in
every spherically complete valued eld. From this we will deduce that it also holds in
every henselian eld.
Theorem 9.11 Let (K, v) be a spherically complete valued eld. Then (K, v) satises the
multi-dimensional Newtons Lemma:
Let f = (f
1
, . . . , f
n
) be a system of n polynomials in n variables with coecients in O.
Assume that b O
n
is such that vf(b) > 2v det J
f
(b). Then there exists a unique a O
n
such that f(a) = 0 and v(a b) = vJ

f
(b)f(b) v det J
f
(b) > v det J
f
(b).
Proof: The inequality vf(b) > 2v det J
f
(b) implies that s := det J
f
(b) = 0. Hence
by Theorem 5.15, J

f induces an isomorphism of ultrametric spaces from b + sM


n
into
J

f(b) + s
2
M
n
, where J

= J

f
(b). Since vf(b) > vs
2
, we have that f(b) s
2
M
n
and
hence also J

f(b) s
2
M
n
(since J

O
nn
). That is, J

f(b)+s
2
M
n
= s
2
M
n
. Therefore,
0 J

f(b) + s
2
M
n
. Since J

f induces a bijection from b + sM


n
onto J

s
2
f(b) + M
n
,
there is a unique a b + sM
n
such that J

f(a) = 0. Since J

is invertible, we have that


f(a) = 0 J

f(a) = 0. Hence, a is the unique element in b+sM


n
such that f(a) = 0. We
have that v(a b) = v
_
J

f
(b)f(a) J

f
(b)f(b)
_
v det J
f
(b) = vJ

f
(b)f(b) v det J
f
(b) >
v det J
f
(b).
Note that like in the one-dimensional case, also in the multi-dimensional case the proof
of Newtons Lemma can be reduced by transformation to a simpler case where we would in
fact obtain the identity as a pseudo-companion. But as we have already shown that even
in the general case we can derive suitable pseudo-linear maps from f, it is much easier to
employ them directly in the proof of the multidimensional Newtons Lemma.
We are now going to show that the multidimensional Newtons Lemma holds in every
henselian eld. Beforehand, we need the following lemma. See [LANG3], Chapter X, 7,
Proposition 8 for its proof, which uses the theory of derivations.
Lemma 9.12 Let x
1
, . . . , x
n
be elements in an arbitrary extension eld of the eld K.
Suppose that there are n polynomials f
1
, . . . , f
n
K[X
1
, . . . , X
n
] such that
1) f
i
(x
1
, . . . , x
n
) = 0 for i = 1, . . . , n,
2) J
f
(x
1
, . . . , x
n
) = 0 .
Then the elements x
1
, . . . , x
n
are separable algebraic over K.
226 CHAPTER 9. HENSELS LEMMA
Theorem 9.13 A valued eld (K, v) is henselian if and only if it satises the multidimen-
sional Newtons Lemma.
Proof: : Let (K, v) be henselian. Take (L, v) to be a maximal immediate extension
of (K, v). Then (L, v) is spherically complete. By the foregoing theorem, (L, v) satises
the multidimensional Newtons Lemma. Denote by O the valuation ring of K, and by O
L
that of L. Now assume that the hypothesis of the multidimensional Newtons Lemma is
satised by a system f of polynomials with coecients in O and by b O
n
. It follows
that there is a unique a = (a
1
, . . . , a
n
) O
n
L
such that f(a) = 0 and v(a b) > v det J
f
(b).
From the latter, it follows that v det J
f
(a) = v det J
f
(b) and in particular, det J
f
(a) = 0.
Now the previous lemma shows that the elements a
1
, . . . , a
n
are separable algebraic over
K. On the other hand, for every Aut (

K|K), the element a = (a
1
, . . . , a
n
) satises
f(a) = f(a) = 0 and v(a b) = min
i
v(a
i
b
i
) = min
i
v(a
i
b
i
) = min
i
v(a
i
b
i
) =
v(a b) > v det J
f
(b) (note that v = v because (K, v) is henselian). By the uniqueness
of a, it follows that a = a for every Aut (

K|K), that is, a K
n
, as required.
: If n = 1, then det J
f
(b) = f

1
(b
1
), and the assertion is precisely the assertion of the
one-dimensional Newtons Lemma. Hence the multidimensional Newtons Lemma implies
that (K, v) is henselian.
9.6 The Implicit Function Theorem
Using the multidimensional Newtons Lemma, one can prove the multidimensional
Implicit Function Theorem:
Theorem 9.14 Take a henselian eld (K, v) and polynomials
f
1
, . . . , f
n
O[X
1
, . . . , X
m
, Y
1
, . . . , Y
n
] with m > 0.
Set Z = (X
1
, . . . , X
m
, Y
1
, . . . , Y
n
) and
J(Z) :=
_
_
_
f
1
Y
1
(Z) . . .
f
1
Y
n
(Z)
.
.
.
.
.
.
f
n
Y
1
(Z) . . .
f
n
Y
n
(Z)
_
_
_
.
Assume that f
1
, . . . , f
n
admit a common zero
z = (x
1
, . . . , x
m
, y
1
, . . . , y
n
) O
m+n
and that the determinant of J(z) is nonzero. Then for all (x

1
, . . . , x

m
) O
m
with v(x
i

x

i
) > 2v det J(z), 1 i m, there exists a unique tuple (y

1
, . . . , y

n
) O
n
such that
(x

1
, . . . , x

m
, y

1
, . . . , y

n
) is a common zero of f
1
, . . . , f
n
, and
min
1in
v(y
i
y

i
) min
1im
v(x
i
x

i
) v det J(z) .
9.7. AN INFINITE-DIMENSIONAL IMPLICIT FUNCTION THEOREM 227
Proof: We observe that the entries of J(Z) and its adjoint matrix J

(Z) are polynomials


in X
1
, . . . , X
m
, Y
1
, . . . , Y
n
with coecients in O. We set b = (x

1
, . . . , x

m
, y
1
, . . . , y
n
). Then
J

(b) is the adjoint matrix for J(b), and the entries of both matrices lie in O. In particular,
this implies that vJ

(b)f(b) vf(b).
By assumption, f
i
(z) = 0 for 1 i m. Hence, the condition v(x
i
x

i
) > 2 det vJ(a),
1 i m, will imply that
vf
i
(b) = v (f
i
(x

1
, . . . , x

m
, y
1
, . . . , y
n
) f(x
1
. . . , x
m
, y
1
, . . . , y
n
))
min
1im
v(x
i
x

i
) > 2v det J(x
1
. . . , x
m
, y
1
, . . . , y
n
)
= 2v det J(x

1
, . . . , x

m
, y
1
, . . . , y
n
) = 2v det J(b)
for 1 i m. In particular, det J(b) = 0. Hence by the multidimensional Newtons
Lemma (which holds by Theorem 9.13), there is a unique common zero (y

1
, . . . , y

n
) O
n
of the polynomials f
i
(x

1
, . . . , x

m
, Y
1
, . . . , Y
n
), 1 i n, such that
min
1in
v(y
i
y

i
) vJ

(b)f(b) v det J(b) = vJ

(b)f(b) v det J(z)


min
1im
vf
i
(b) v det J(z)
min
1im
v(x
i
x

i
) v det J(z) .
This proves our assertion.
Exercise 9.3 Let (K, v) be a valued eld of rank 1.
a) Let 0 < vK and prove that (K, v) is henselian if and only if it satises the following condition
Every monic polynomial f = X
n
+ c
n1
X
n1
+ . . . + c
1
X + c
0
K[X] with vc
n1
= 0 and vc
i
> ,
0 i n 2, admits a linear factor X +c in O
K
[X] such that c = c
n1
.
(Hint: use the Approximation Theorem).
b) Prove that (K, v) is henselian if and only if it satises the Implicit Function Theorem.
c) Find a transformation that reduces the nultidimensional Newtons Lemma to a nultidimensional Hensels
Lemma.
9.7 An innite-dimensional Implicit Function Theo-
rem
From our result in Section 1.7 it follows that an innite power Y
I
of an ultrametric space Y
can be equipped with an ultrametric u
I
(analogous to the minimum valuation) if the value
set uY is well ordered. In this case, if (Y, u) is spherically complete, then so is (Y
I
, u
I
). So
we obtain the following corollary to our Main Theorem 1.26 and to Proposition 2.41:
Corollary 9.15 a) Take two ultrametric spaces (Y, u) and (Y

, u

), and an arbitrary index


set I. Assume that uY is well ordered, f : Y
I
Y

is immediate and that (Y, u) is
spherically complete. Then f is surjective and (Y

, u

) is spherically complete.
b) Take two valued abelian groups (G, v) and (G

, v

), and an arbitrary index set I. Assume


that vG is well ordered, b G
I
, B is a ball around 0 in G
I
, f : G
I
G

has a pseudo-
companion on b + B, and that (G, v) is spherically complete. Then f is surjective and
(G

, v

) is spherically complete.
228 CHAPTER 9. HENSELS LEMMA
In the case of a valued eld (K, v) we cannot do the same since if the valuation is
non-trivial, the value group will not be well ordered. If the valuation is not discrete (i.e.,
its value group is not isomorphic to Z), then not even the value set vO := v(O\{0}) of the
valuation ring is well ordered. But we may be interested in innite systems of polynomials
with coecients in a subring R of O with well ordered value set vR := v(R\ {0}). We set
M
R
:= {a R | va > 0}.
Note that (R, v) is not necessarily spherically complete, even if (K, v) is. So we will
assume that (R, v) is spherically complete.
We generalize the denitions of minimum valuation and of pseudo linear map in
the obvious way. If a = (a
i
)
iI
R
I
, then va := min
iI
va
i
. If Y R
I
, 0 = s R and
f a map from Y into R
I
, then f is pseudo-linear with pseudo-slope s if (??) holds for
all y, z Y such that y = z. We then have the following application of Proposition 2.41
together with Proposition 1.11:
Proposition 9.16 Take b R
I
and B a ball in (R
I
, v) around 0. Assume that f : b+B
R
I
is pseudo-linear with pseudo-slope s R and that (R, v) is spherically complete. Then
f is an isomorphism of ultrametric spaces from b + B onto fb + sB.
If the map is given by an innite system of polynomials f = (f
k
)
kI
in innitely many
variables X
i
, i I, and with coecients in R, then we may consider the innite matrix
J
f
(b) R
II
. Note that this matrix has only nitely many non-zero entries in every row.
We denote by R
(II)
all matrices in R
II
which have only nitely many non-zero entries
in every row and every column. If every variable appears only in nitely many f
k
, then
J
f
(b) R
(II)
.
If we assume that R is spherically complete, we can consider a larger class of matrices.
We denote by R
((II))
all matrices in R
II
which for each vR have only nitely many
entries of value in every row and every column. For every two matrices in R
((II))
,
their product can be computed and lies again in R
((II))
. It is possible that J
f
(b) R
((II))
even when there are variables that appear in innitely many f
k
.
We dene M
(II)
R
and M
((II))
R
analogously and note that R
(II)
, R
((II))
, M
(II)
R
and
M
((II))
R
are all closed under matrix addition and multiplication and under scalar multi-
plication. Further, R
(II)
M
(II)
R
M
(II)
R
, M
(II)
R
R
(II)
M
(II)
R
, R
((II))
M
((II))
R

M
((II))
R
and M
((II))
R
R
((II))
M
((II))
R
.
We are not able to use determinants here. Still, we can use our original approach if
J
f
(b) has an inverse. But we can even work with less than invertibility. Given matrices
M, M

in R
(II)
, or in R
((II))
if R is spherically complete, we will say that M

is a
pseudo-inverse of M if the matrices MM

E and M

M E are in M
II
R
, where E
denotes the I I-identity matrix.
Actually, we also do not need that the ring R is a subring of a valued eld. It suces to
assume that it is a valued abelian group with its multiplication satisfying (V3), and that
its value set is a well ordered subset of an ordered abelian group. It then follows that the
value set does not contain negative elements. In particular, all entries of M R
II
have
value 0. This implies that vMa va for all a R
I
. Since vR is well ordered, it contains
a minimal positive value
0
. If M is in M
(II)
R
or in M
((II))
R
, then all entries of M have
value
0
. It then follows that vMa va +
0
> va for all a R
I
.
9.7. AN INFINITE-DIMENSIONAL IMPLICIT FUNCTION THEOREM 229
Lemma 9.17 Take M, M

in R
(II)
, or in R
((II))
if R is spherically complete. Assume
that M

is a pseudo-inverse of M. Then the following holds:


1) For all a R
I
, vMa = va and vM

a = va; in particular, M, M

/ M
II
R
and the
value set vR must contain 0.
2) If M

is in R
(II)
, or in R
((II))
respectively, such that M

M M
II
R
, then M

is
also a pseudo-inverse of M

.
3) Both M and M

induce immediate embeddings of the ultrametric space R


I
in itself with
value map id, and the same holds on every ball around 0 in R
I
.
Proof: 1): For all a R
I
we have that v(MM

a a) = v((MM

E)a) > va and


hence va = vMM

a vM

a va. It follows that equality holds everywhere, which gives


vM

a = va. Interchanging M and M

, we obtain vMa = va.


2): We compute: M

E = (M

M)M

+ MM

E M
II
R
, and similarly for
M

E.
3): It suces to show that for every ball B around 0 in R
I
, M induces an immediate
embedding of B into itself with value map id. Since vMa = va for all a R
I
, we have
MB B and that M induces an injective map on B with value map id. As M induces a
group homomorphism, we only have to show now that for every a

B\{0} there is a B
such that (IH1) and (IH2) of Proposition 2.36 hold for M in the place of f. As vM

= va

,
we have that a := M

B. Further, v(a

Ma) = v(a

MM

) = v(EMM

)a

> va

.
Finally, if b B with va vb, then vMa = va vb = vMb.
Proposition 9.18 Assume that (R, v) is spherically complete. Take any index set I and
a system of polynomials f = (f
k
)
kI
in variables Y
i
, i I, with coecients in R. Take
b R
I
and suppose that J
f
(b) lies in R
((II))
and admits a pseudo-inverse in R
((II))
.
Then J
f
(b) is a pseudo-companion of f on b +M
I
R
, and f is an isomorphism from b +M
I
R
onto f(b) + M
I
R
with value map id. The system f has a zero on b + M
I
R
(which then is
unique) if and only if vf(b) > 0.
Proof: Since J = J
f
(b) has a pseudo-inverse, we know from the previous lemma that J
induces an immediate embedding of M
I
R
in itself with value map id.
Take
1
,
2
M
I
R
. An innite-dimensional version of the multidimensional Taylor
expansion gives the innite-dimensional analogue of (5.12) and (5.13), with s = 1. We
obtain that for y = b +
1
and z = b +
2
in b +M
I
R
with y = z,
v(f(y) f(z) J(y z)) > v(y z) = vJ(y z) .
This proves that J is a pseudo-companion of f on b +M
I
R
. From Proposition 2.41 we infer
that f induces an embedding of b + M
I
R
in f(b) + JM
I
R
f(b) + M
I
R
with value map
= id.
The remaining assertions now follow from Proposition 2.41 and Theorem 1.26.
Now we can prove an innite-dimensional Implicit Function Theorem:
Theorem 9.19 Take any index sets I and I

and a system of polynomials f = (f


k
)
kI
in
variables X
j
, j I

, and Y
i
, i I, with coecients in R, and such that each variable
230 CHAPTER 9. HENSELS LEMMA
Y
i
appears in only nitely many f
k
. Assume that (R, v) is spherically complete. Set Z =
(X
j
, Y
i
| j I

, i I) and
J(Z) :=
_
f
k
Y
i
(Z)
_
k,iI
.
Assume that the polynomials f
k
, k I, admit a common zero z = (x
j
, y
i
| j I

, i I) in
R
I

I
such that J(z) admits a pseudo-inverse in R
((II))
. Then for all (x

j
)
jI
R
I

with
v(x
j
x

j
) > 0 there exists a unique (y

i
)
iI
R
I
such that z

= (x

j
, y

i
| j I

, i I) is a
common zero of the polynomials f
k
, k I, and
min
iI
v(y
i
y

i
) min
jI

v(x
j
x

j
) .
Proof: We set z := (x

j
, y
i
| j I

, i I) and observe that our condition that


v(x
j
x

j
) > 0 implies that v
_
f
k
Y
i
( z)
f
k
Y
i
(z)
_
> 0. From part 2) of Lemma 9.17 it
thus follows that the pseudo-inverse of J(z) is also a pseudo inverse of J( z). (Note that
J(z), J( z) R
(II)
by our condition on the variables Y
i
.)
For each k I we set g
k
(Y
i
| j I) := f
k
(x

j
, Y
i
| j I

, i I). Further, we set


b := (y
i
| i I). We consider the system g = (g
k
)
kI
. From Proposition 9.18 we infer that
J
g
(b) = J( z) is a pseudo-companion of g on b +M
I
R
. By assumption, f
k
(z) = 0 for k I.
Hence, the condition v(x
j
x

j
) > 0 will imply that
vg
k
(b) = vf
k
( z) = v(f
k
( z) f
k
(z)) min
jI

v(x
j
x

j
) > 0 .
Hence vg(b) > 0 and by Proposition 9.18 the system g has a unique zero a = (y

i
| i I)
on b +M
I
R
. It satises
min
iI
v(y
i
y

i
) = v(b a) = v(g(b) g(a)) = vg(b) min
jI

v(x
j
x

j
) .

Remark 9.20 In our theorem we needed the assumption on the variables Y


i
in order to have only nitely
many non-zero polynomials in each row and each column of J(Z). Without this it is not automatic that
the conditions J(z) R
((II))
and v(x
j
x

j
) > 0 imply that J( z) R
((II))
. We can drop the condition
on the variables if we assume instead that J( z) R
((II))
and that it has a pseudo-inverse in R
((II))
.
9.8 Power series maps on valuation ideals
Take any eld k and any ordered abelian group G. As usual, we consider k((G)) with its
canonical valuation v = v
t
and denote the valuation ideal by M. Every power series
f(X) =

iN
c
i
X
i
k[[X]] (9.7)
denes in a canonical way a map f : M M (note: 0 / N in our notation). This can
be shown by use of Neumanns Lemma, cf. [DMM1]. We note that for every integer r > 1
and every y, z M,
v(y
r
z
r
) > v(y z) . (9.8)
9.9. POWER SERIES MAPS ANDINFINITE-DIMENSIONAL IMPLICIT FUNCTIONTHEOREMS231
Therefore, if c
1
= 0, we have that
v(f(y) f(z) c
1
(y z)) = v

i2
c
i
(y
i
z
i
) > v(y z) = vc
1
(y z) (9.9)
because vc
i
= 0 for all i. So we see that f is pseudo-linear with slope c
1
if c
1
= 0. By
Proposition ??, we obtain:
Theorem 9.21 If f : M M is dened by the power series (9.7), then f is an isomor-
phism of ultrametric spaces.
A similar result holds for power series with generalized exponents (which for instance
are discussed in [DS]). Take any subgroup G of R and a generalized power series of the
form
f(X) =

iN
c
i
X
r
i
k[[X
G
]] (9.10)
where r
i
, i N, is an increasing sequence of positive real numbers in G. Suppose that the
power functions y y
r
i
are dened on M for all i. Then again, the generalized power
series (9.10) denes a map f : M M. We note that (9.8) also holds for every real
number r > 1 for which y y
r
is dened on M. Hence if c
1
= 0 and r
1
= 1, then (9.9)
holds, with the exponent i replaced by r
i
. This shows again that f is pseudo-linear with
pseudo-slope c
1
. If, however, r
1
= 1, we may think of writing f(y) =

f(y
r
1
) with

f(X) =

iN
c
i
X
r
i
/r
1
.
If the power functions y y
r
i
/r
1
are dened on M for all i, then

f denes a pseudo-linear
map from M to M with pseudo-slope c
1
. So we obtain:
Theorem 9.22 Suppose that the power functions y y
r
i
and y y
r
i
/r
1
are dened on
M for all i, and that y y
r
1
is surjective. If f : M M is dened by the power series
(9.10) with c
1
= 0, then f is surjective.
9.9 Power series maps and innite-dimensional Im-
plicit Function Theorems
We use again the notations and assumptions from Section 9.7. We take R[[X
j
, Y
i
| j
I

, i I]] to be the set of all formal power series in the variables X


j
, Y
i
in which for every
n N only nitely many of the X
j
, Y
i
appear to a power less than n. In the previous
section, our power series had well dened values because we were operating in a power
series eld k((G)). Here, we will assume throughout that R is spherically complete. But
this alone does not a priori give us well dened values of the power series on M
I

I
R
. So we
will assume that we have some canonical way to determine the value of a given power series
at an element of M
I
R
. This holds for instance if vR is archimedean, i.e., is a subsemigroup
of an archimedean ordered abelian group.
To every power series g R[[Y
i
| i I]] we associate its 0-linear part L
0
g
, by which
we mean the sum of all of its monomials of total degree 1 and with a coecient in R of
value 0. This is a polynomial, i.e., contains only nitely many of the variables Y
i
. We set
Y = (Y
i
| i I).
232 CHAPTER 9. HENSELS LEMMA
Theorem 9.23 Assume that (R, v) is spherically complete. Take any index sets I and
I

and a system f = (f
k
)
kI
where f
k
R[[X
j
, Y
i
| j I

, i I]]. Assume that f


k
,
k I, admit a common zero z = (x, y), x M
I

R
, y M
I
R
, such that for the map
L(Y ) = L
0
f(x,Y )
(Y ) : M
I
R
M
I
R
the following holds: for every a

M
I
R
\ {0} there is
some a M
I
R
such that
v(a

La) > va

and va = va

.
Take x

= (x

j
)
jI
M
I

R
, set = v(x x

) and g(Y ) = f(x

, Y ) and suppose that for all


distinct w, w

(y),
v(gw gw

L(w w

)) > v(gw gw

) . (9.11)
Then there exists a unique (y

i
)
iI
M
I
R
such that z

= (x

j
, y

i
| j I

, i I) is a common
zero of f
k
, k I, and
min
iI
v(y
i
y

i
) .
Proof: Note that L
f(x

,Y )
(Y ) = L
f(x,Y )
(Y ) = L(Y ). We claim that L is a pseudo-
companion of f(x

, Y ) : M
I
R
M
I
R
on B

(y). Condition (PC2) holds by assumption.


As L is a group homomorphism, our conditions together with Proposition 2.36 show that
L : M
I
R
M
I
R
is immediate; note that (IH2) holds because if va vb then vLa = va
vb vLb. Now the assertion of our theorem follows as in earlier proofs.
The following version of the above theorem has a similar proof:
Theorem 9.24 Assume that (R, v) is spherically complete. Take any index sets I and
I

and a system f = (f
k
)
kI
where f
k
R[X
j
| j I

][[Y
i
| i I]]. Assume that f
k
,
k I, admit a common zero z = (x, y), x R
I

, y M
I
R
, such that L(Y ) = L
0
f(x,Y )
(Y )
satises the same condition as in Theorem 9.23. Take x

= (x

j
)
jI
R
I

such that
= v(x x

) > 0. Suppose that (9.11) holds for g(Y ) = f(x

, Y ). Then there exists


a unique (y

i
)
iI
M
I
R
such that z

= (x

j
, y

i
| j I

, i I) is a common zero of the


polynomials f
k
, k I, and min
iI
v(y
i
y

i
) .
Alternatively, in order to obtain maps on all of R, one can consider convergent power
series. We let R{{X
j
, Y
i
| j I

, i I}} be the set of all formal power series in the variables


X
j
, Y
i
in which for every vR only nitely many monomials have coecients of value
less than . Again we assume that R is spherically complete. Then every convergent power
series denes a map from R into R. In a similar way as before, one can prove:
Theorem 9.25 Assume that (R, v) is spherically complete. Take any index sets I and I

and a system f = (f
k
)
kI
where f
k
R{{X
j
, Y
i
| j I

, i I}}. Assume that f


k
, k I,
admit a common zero z = (x, y), x R
I

, y R
I
, such that L(Y ) = L
0
f(x,Y )
(Y ) satises the
same condition as in Theorem 9.23. Take x

= (x

j
)
jI
R
I

such that = v(x x

) > 0.
Suppose that (9.11) holds for g(Y ) = f(x

, Y ). Then there exists a unique (y

i
)
iI
R
I
such that z

= (x

j
, y

i
| j I

, i I) is a common zero of the polynomials f


k
, k I, and
min
iI
v(y
i
y

i
) .
9.10. RELATIVELY ALGEBRAICALLY CLOSED SUBFIELDS 233
9.10 Relatively algebraically closed subelds
From the already given examples of henselian elds, we obtain an abundance of other exam-
ples by the fact that every relatively separable-algebraically closed subeld of a henselian
eld is again henselian. This was proved in Corollary 7.38. In the following lemma, we
rene this result and give a new proof. We will make use of the fact that a eld is henselian
if and only if it satises Hensels Lemma restricted to separable polynomials. The latter is
property 4

) of Theorem 9.1.
Lemma 9.26 Assume (L, v) to be henselian and K to be relatively separable-algebraically
closed in L. Then (K, v) is henselian too. Further, K is relatively separable-algebraically
closed in L. If in addition K = L, then the torsion subgroup of vL/vK is a p-group, with
p the characteristic exponent of K.
Proof: By Theorem 9.1, a valued eld is henselian if and only if it satises Hensels
Lemma in the version 4

). If (L, v) is henselian and f K[X] is separable over K and


satises the assumptions of 4

), then there exists a root a L according to 4

). The
minimal polynomial of a over K divides f and is thus separable over K. Since K is
assumed to be relatively separable-algebraically closed in L, it follows that a K. Hence
also (K, v) has property 4

), that is, (K, v) is henselian. (Note that a proof using the


uniqueness of extensions which serves as the denition for the property henselian is not
at all as obvious as this proof which uses Hensels Lemma. For the proof of the fact that
every algebraic extension of a henselian eld is henselian it is just the opposite case.)
Now let a L be separable-algebraic over K. We choose a monic polynomial g(X)
K[X] whose reduction g(X) K[X] modulo v is its irreducible polynomial over K. Then a
is a simple root of g. Hence by Hensels Lemma, there is a root a L of g whose residue is
a. As all roots of g(X) are distinct, Lemma 5.7 together with part a) of Lemma 5.6 shows
that also all roots of g(X) are distinct. Thus, a is separable-algebraic over K. By our
assumption, it follows that a K, showing that a K. This proves that K is relatively
separable-algebraically closed in L.
Assume that K = L and let vL and n N not divisible by p such that n vK.
Choose a L and b K such that va = and vb = n. Then v(a
n
/b) = 0. Since we
have K = L, there exists some c K satisfying vc = 0 and c = a
n
/b, hence a
n
/bc = 1. So
a
n
/bc is a 1-unit, and since n is not divisible by p, Corollary 9.33 (applied to (L, v)) shows
that it admits an n-th root d L. We have that (a/d)
n
= bc K. Since the polynomial
X
n
bc has no multiple roots, it is separable and thus, a/d is separable algebraic over
K. Since K was assumed to be relatively separable-algebraically closed in L, we nd that
a/d K. Since vc = 0, we have the equality nv(a/d) = vbc = vb = n, which shows that
= v(a/d) vK.
We will show in Example ?? below that the condition that vL = vK is necessary.
As an immediate consequence of Lemma 9.26, we obtain:
Corollary 9.27 Take a henselian eld (L, v) and a perfect relatively algebraically closed
subeld K of L. Then K is relatively algebraically closed in L. If in addition K = L, then
vK is pure in vL.
The result of Lemma 9.26 can be improved for extensions of nitely ramied elds.
234 CHAPTER 9. HENSELS LEMMA
Lemma 9.28 Let (L, v) be a henselian nitely ramied eld and K relatively algebraically
closed in L. If L|K is separable algebraic and if (L, v) and (K, v) have a common prime
element , then (K, v) is a henselian eld, L = K and vL/vK is torsion free.
Proof: As in the proof of Lemma 9.26 it is shown that (K, v) is henselian and that
L = K.
Let p be the residue characteristic of (L, v). It suces to show the following: If a L
such that pva vK, then va vK. Without loss of generality, we can assume that
va 0; otherwise, we replace va by va and a by a
1
. Let b = a
p
c, where c K is chosen
such that v c = va
p
, that is, vb = 0. We claim that mv
L
(b, K) for every positive
integer m. Assume the contrary, that is, that there exists a minimal m
0
0 such that
(m
0
+1)v /
L
(b, K). In view of m
0
v
L
(b, K), let c
0
K such that v(bc
0
) m
0
v.
Since vL = Zv, the inequality v(b c
0
) > m
0
v would imply that (m
0
+1)v
L
(b, K),
a contradiction. Hence v((b c
0
)
m
0
) = 0. Since L = K by assumption, there is some
c K such that v((b c
0
)
m
0
c) > 0. That is, v(b c
0

m
0
c) > m
0
v. As before,
it follows that (m
0
+ 1)v
L
(b, K), a contradiction. Hence, mv
L
(b, K) for every
positive integer m.
By what we have shown, we can choose c K such that v(b c) > 2ev = 2vp. Then
vc = vb = 0, and 1+
bc
c
is a 1-unit of level > 2vp. Hence, in the henselian eld (L, v) there
is some unit a such that a
p
= 1 +
bc
c
. Consequently, b = c a
p
, that is, (a/ a)
p
= c/ c K.
Since K was assumed to be relatively algebraically closed in L, we nd that a/ a K. This
shows that va = v(a/ a) vK.
Corollary 9.29 Let K be a relatively algebraically closed subeld of L and assume that
(L, v) is a -adically closed eld. If both (K, v) and (L, v) have a common prime element,
then also (K, v) is a -adically closed eld, and K = L.
Proof: Since L is a nite eld, also K is a nite eld and the extension L|K is separable
algebraic. Hence, Lemma 9.28 shows that L = K and that vL/vK is torsion free. From
the latter, it follows that also vK is a Z-group. Hence by the foregoing theorem, (K, v) is
-adically closed.
Concerning the embedding of a residue eld in a valued eld, we can derive the following
assertion:
Lemma 9.30 Let (K, v) be a henselian eld of equal characteristic. Let k be the prime
eld of K and K. If k

|k is a separably generated subextension of K|k, then there is a eld


embedding of k

in K such that = for all K. If (K, v) is of residue characteristic


0, then K itself admits such an embedding.
Proof: Let T be a transcendence basis of k

|k such that k

|k(T ) is a separable algebraic


extension. For every t T , choose an element a
t
K such that a
t
= t. Then by
Lemma 6.30, the elements a
t
, t T , are algebraically valuation independent and thus
also algebraically independent in K (over k). Hence, the assignment t a
t
induces a
eld embedding of k(T ) in K over k. In view of Lemma 6.35, it satises that =
for all k(T ). Now we take K
0
to be the relative algebraic closure of k(T ) in K.
9.11. N-TH POWERS IN HENSELIAN FIELDS 235
By Lemma 9.26, its residue eld contains the relative separable-algebraic closure of k(T )
in K. Thus, it also contains k

because k

|k(T ) is separable algebraic. But K


0
is an
algebraic extension of the trivially valued eld (k(T ), v) and is thus also trivially valued.
Consequently, the residue map restricted to K
0
is an isomorphism. Its inverse, restricted
to k

, is the required embedding of k

in K.
The second assertion follows from what we have just proved and the fact that every
extension of elds of characteristic 0 is separably generated.
As a further corollary to Theorem 9.6 and Lemma 9.26, we have something like a
converse to Theorem ??. Recall that the completion of a valued eld is an immediate
extension. So if a valued eld has rank 1, then the same is true for its completion.
Corollary 9.31 The relative separable-algebraic closure of a valued eld of rank 1 in its
completion is henselian.
9.11 n-th powers in henselian elds
Let K = (K, v) be a henselian eld, a K and n a natural number. We ask whether
a is an n-th power in K, that is, whether there exists b K such that a = b
n
. A rst
observation is the following. If a = b
n
with b K, then va = nvb nvK, and if a O

K
,
then a = b
n
(K

)
n
. A partial converse of the latter implication is given by the next
lemma.
Before we continue, let us note the following. If k is an arbitrary eld and the natural
number n is not divisible by the characteristic of k, then a polynomial X
n
a with 0 = a k
has no multiple roots. Indeed, a multiple root would also be a root of the derivative nX
n1
of X
n
a. But since n = 0 in k, the only root of nX
n1
is 0, which is not a root of X
n
a.
Lemma 9.32 Let (K, v) be a henselian eld, a O

K
and n a natural number not divisible
by the residue characteristic char K. If a (K

)
n
, then a is an n-th power in K.
Proof: Consider the polynomial X
n
a O
K
[X]. Its reduction is X
n
a. By
hypothesis, this polynomial has a root in K. Since n is not divisible by the characteristic
of K and a = 0, this root is a simple root. Hence by Hensels Lemma in the version 3) of
Theorem 9.1, X
n
a admits a root in K, and a is thus an n-th power in K.
Since a 1-unit has residue 1 and the polynomial X
n
1 always admits 1 as a root, we
obtain the following corollary:
Corollary 9.33 Let (K, v) be a henselian eld and n a natural number not divisible by
the residue characteristic char K. Then every 1-unit in (K, v) is an n-th power.
In the case of a O

K
, what can be said if n is divisible by the residue characteristic?
In this case one can employ Newtons Lemma. The derivative of X
n
a is nX
n1
, hence
an approximative root b will satisfy the hypothesis of Newtons Lemma if v(b
n
a) >
2vn + 2(n 1)vb. But then, vb
n
= va = 0 and thus, vb = 0. So in fact our condition is
v(b
n
a) > 2vn. (Note that the expression vn denotes the value of the element n 1,
where 1 is the multiplicative unit of K.) Now Newtons Lemma shows:
236 CHAPTER 9. HENSELS LEMMA
If a is an n-th power up to a summand of value > 2vn, then a is an n-th power.
Let us again consider the important special case of 1-units. If a is a 1-unit, then the
value v(a 1) is called the level of the 1-unit a. We have proved:
Lemma 9.34 Let (K, v) be a henselian eld and n a natural number. If a is a 1-unit of
level > 2vn, then a is an n-th power in K.
This lemma will be applied in the next section to improve the assertion of Lemma 9.26 for
an important class of valued elds of mixed characteristic (cf. Lemma 9.28). Let us now
discuss the special case of p-th roots of 1-units over henselian elds of mixed characteristic.
Throughout this section, we take C to be an element in the algebraic closure of Q such
that C
p1
= p, where p > 0 is a prime. Take a henselian eld (K, v) of characteristic 0
and residue characteristic p. Then C

Q

K. Extend the valuation v to

K. Note that
C
p
= pC and vC =
1
p 1
vp > 0 .
Consider the polynomial
X
p
(1 + b) (9.12)
with b K. Performing the transformation
X = CY + 1 , (9.13)
dividing by C
p
and using that C
p
= pC, we obtain the polynomial
f(Y ) = Y
p
+ g(Y ) Y
b
C
p
(9.14)
with
g(Y ) =
p1

i=2
_
p
i
_
C
ip
Y
i
(9.15)
a polynomial with coecients in K(C) of value > 0.
Lemma 9.35 Take (K, v) and C as above. Then K contains C if and only if it contains
all p-th roots of unity.
Proof: Since char K = p, the restriction of v to Q K is the p-adic valuation. Since
(K, v) is henselian, it contains (Q
h
, v
p
) (Lemma ??). Let = 1 be a p-th root of unity. It
suces to show that Q
h
() = Q
h
(C). Applying the transformation (9.13) to the polynomial
(9.12) with b = 0, we obtain the polynomial f(Y ) = Y
p
+g(Y ) Y Q(C)[Y ] which splits
over Q
h
by Hensels Lemma because f(Y ) = Y
p
Y splits over F
p
. Since the non-zero
roots of f have nonzero residue and thus value zero, v( 1) = vC = vp/(p 1). We nd
that (vQ
h
() : vQ
h
) p 1. Consequently,
[Q
h
() : Q
h
] p 1 [Q
h
(C) : Q
h
] [Q
h
() : Q
h
] ,
showing that equality holds everywhere and that Q
h
() = Q
h
(C).
The following lemma supplements Lemma 9.33 in the case of p-th roots of 1-units:
9.11. N-TH POWERS IN HENSELIAN FIELDS 237
Lemma 9.36 Let (K, v) be a henselian eld containing all p-th roots of unity. Then
vb >
p
p 1
vp 1 + b (K

)
p
for all b K. In other words, every 1-unit of level >
p
p1
vp has a p-th root in K.
Proof: Consider the polynomial (9.14), derived from (9.12). If vb >
p
p1
vp = vC
p
, then
f(Y ) = Y
p
Y , which splits over K. By Hensels Lemma, this implies that f(Y ) splits
over K. Via the transformation (9.13), it follows that 1 + b has a p-th root in K.
Corollary 9.37 Let (K, v) be a henselian eld containing all p-th roots of unity. Take any
1-units 1 + b and 1 + c in K. Then:
a) 1 + b (1 + b + c) (K

)
p
if vc >
p
p1
vp .
b) 1 + b (1 + b + c) (K

)
p
if 1 + c (K

)
p
and vbc >
p
p1
vp .
c) 1 + c
p
+ pc (K

)
p
if vc
p
> vp .
d) 1 + b pc (1 + b + c
p
) (K

)
p
if vb
1
p1
vp and vc
p
> vp .
Proof: a): 1 + b (1 + b + c)(K

)
p
is true if the quotient
1 + b + c
1 + b
= 1 +
c
1 + b
is an element of (K

)
p
. By hypothesis we have vb > 0 and thus v
c
1+b
= vc. Now our
assertion follows from Lemma 9.36.
b): An application of part a) shows that
(1 + b + c) (1 + b)(1 + c) (K

)
p
if vbc >
p
p1
vp.
The assertion of b) is an immediate consequence of this.
c): If vc
p
> vp then for every i = 2, . . . , p 1 we have
v
_
p
i
_
c
i
vp + 2vc >
p + 2
p
vp
p
p 1
vp ;
note that the last inequality holds for every p 2. This together with assertion a) yields
1 + c
p
+ pc
_
1 + c
p
+ pc +
p1

i=2
_
p
i
_
c
i
_
(K

)
p
= (1 + c)
p
(K

)
p
= (K

)
p
.
d): In view of part c), the assertion follows from part b) where b is replaced by bpc and c is
replaced by c
p
+pc. Note that b) can be applied since v(bpc)(c
p
+pc) >
1
p1
vp+vp =
p
p1
vp.

What can we say about elements of value = 0? We have seen that va nvK if a is
an n-th power in K. So we can answer: If there is some c K such that vac
n
= 0 and if
ac
n
satises one of the above criteria for being an n-th power, then a is an n-th power. In
particular, we have:
238 CHAPTER 9. HENSELS LEMMA
Corollary 9.38 Let (K, v) be a valued eld and n a natural number. If K is closed under
n-th roots, then vK is n-divisible, and K is closed under n-th roots. The converse holds if
(K, v) is henselian and n is not divisible by the residue characteristic char K.
(For the rst assertion, cf. Lemma 6.40 and Lemma 6.41).
But this answer is not satisfactory if we want to decide for every single element whether
it is an n-th power. There is an easy answer to this problem if (K, v) is equipped with
a multiplicative coecient map co : K K. In this case, we can prove the following
generalization of Lemma 9.32:
Lemma 9.39 Let (K, v) be a henselian eld and n a natural number not divisible by the
residue characteristic char K. Assume that (K, v) admits a multiplicative coecient map.
Then a K is an n-th power in K if and only if va nvK and co a is an n-th power in
K.
Proof: Since co is assumed to be multiplicative, co a is an n-th power in K if a K is
an n-th power in K. The latter also implies that va nvK. Now assume that va nvK
and co a is an n-th power in K. We choose c K such that va = nvc. Then co ac
n
=
(co a)(co c)
n
is again an n-th power in K. Since vac
n
= 0, we have that ac
n
= co ac
n
.
By Lemma 9.32 it follows that ac
n
is an n-th power in K and thus, the same holds for
a.
9.12 When polynomials are close to each other
The following is an application of Theorem 9.1:
Theorem 9.40 Let (K, v) be a henselian eld and f = f
1
. . . f
r
where f
1
, . . . , f
r
are
distinct monic separable irreducible polynomials over K. Then for every vK there is
some vK such that the following holds: If h is any monic polynomial over K satisfying
v(f h) > , then h = h
1
. . . h
r
where h
1
, . . . , h
r
are distinct monic separable irreducible
polynomials over K and for each k {1, . . . , r}, deg f
k
= deg h
k
, v(f
k
h
k
) > , for all
roots a of f
k
and b of h
k
, K(a) and K(b) are isomorphic over K, and f
k
and h
k
have the
same splitting eld.
Proof: Let n = deg f. By our hypothesis on f, it has n distinct roots a
1
, . . . , a
n


K.
Extend v to

K and choose some vK such that
> max{v(a
i
a
j
) | 1 i < j n}
(recall that vK is conal in

vK = v

K). By the Continuity of Roots (Theorem 5.11) there
exists some vK such that for every monic polynomial h of degree n over K such that
v(f h) > , there is an enumeration b
1
, . . . , b
n
of its roots in

K such that v(a
i
b
i
) >
for 1 i n. By our choice of it follows that b
i
is the unique root of h satisfying
v(a
i
b
i
) > . Consequently, h has no multiple roots.
For every k {1, . . . , r}, we dene h
k
=

(X b
i
) where the product is taken over all
i such that a
i
is a root of f
k
. We have to show that h
k
is an irreducible polynomial over
K; since it has no multiple roots, it must be separable. Note that for every i {1, . . . , n}
9.12. WHEN POLYNOMIALS ARE CLOSE TO EACH OTHER 239
and every Gal K, b
i
is a root of h. Assume that a
i
= a
j
. Since (K, v) is henselian,
we have that v(a
j
b
i
) = v(a
i
b
i
) > . By our choice of , it follows that b
i
= b
j
.
Hence, b
i
and b
j
are conjugate over K if and only if a
i
and a
j
are, which in turn holds if
and only if they are roots of the same f
k
. By denition, this is the case if and only if b
i
and b
j
are roots of the same h
k
. Consequently, h
k
is a polynomial over K, and since all
its roots are conjugate over K, it is irreducible.
Since h is separable over K, every b
i
lies in K
sep
. By our choice of and by Krasners
Lemma (cf. Theorem 9.1), we obtain that K(a
i
) K(b
i
). But if k is such that a
i
is a root
of f
k
, then [K(a
i
) : K] = deg f
k
= deg h
k
= [K(b
i
) : K], showing that K(a
i
) = K(b
i
).
Therefore, f
k
and h
k
have the same splitting eld. Moreover, if a = a
i
and b is an arbitrary
root of h
k
, then K(b) is isomorphic to K(b
i
) = K(a) over K.
From Exercise 5.5 we infer that there is some vK such that if v(a
i
b
i
) > for all
i, then v(f
k
h
k
) > . So if we choose , then will have all required properties.
Corollary 9.41 Let (K, v) be an arbitrary eld and f an irreducible polynomial over K,
and let the notation be as in Lemma 7.46, where we take K

= K
h
. Then there is some
vK such that the following holds: If h is any monic polynomial over K satisfying
v(f h) > , then h = h
1
. . . h
r
where h
1
, . . . , h
g
are irreducible polynomials over K
h
and
for each k {1, . . . , g} and all roots a of f
k
and b of h
k
, K(a) and K(b) are isomorphic
over K
h
.
Proof: We choose the separable polynomial

f with f(X) =

f(X
p

) as in the proof of
part d) of Lemma 7.46. We observe that v(

f

h) > if and only if v(



f(X
p

h(X
p

)) >
for every polynomial

h K[X]. Now we apply the foregoing theorem to

f and

h in the
place of f and h. Setting h(X) =

h(X
p

) and h
k
(X) =

h
k
(X
p

), we argue further as in
the proof of part d) of Lemma 7.46.
Corollary 9.42 Let (K, v) be an arbitrary valued eld, f K[X] an irreducible polynomial
over K and a

K a root of f. Further, let v
1
, . . . , v
g
be all extensions of v from K to K(a).
Then there is some vK such that the following holds: If h is any monic polynomial over
K satisfying v(f h) > , and if b

K is a root of h and w
1
, . . . , w
g
are all extensions of
v from K to K(b), then g = g

and after a suitable renumeration of the w


i
, we have that
d(K(a)|K, v
i
) = d(K(b)|K, w
i
)
e(K(a)|K, v
i
) = e(K(b)|K, w
i
)
f(K(a)|K, v
i
) = f(K(b)|K, w
i
) .
Choosing large enough, we also get that h is separable if f is.
Proof: Fix an extension of the valuation v to

K and let (K
h
, v) denote the corresponding
henselization of (K, v). We choose as in the foregoing lemma and assume that v(f h) >
. Then f = f
1
. . . f
g
and h = h
1
. . . h
g
such that f
k
and h
k
are irreducible polynomials
over K
h
satisfying the further assertions of that corollary. From Lemma 7.47 we infer that
g = g

. We choose automorphisms

1
, . . . ,

g
according to that lemma for h in the place
of f. Now let b be a root of h. Then after a suitable renumeration, we can assume that
240 CHAPTER 9. HENSELS LEMMA

i
b is a root of h
i
if and only if
i
a is a root of f
i
. Hence,
i
K(a).K
h
= K(
i
a).K
h
and

i
K(b).K
h
= K(

i
b).K
h
are isomorphic over K
h
. Consequently, in view of Lemma 11.2,
d(K(a)|K, v
i
) = d(
i
K(a).K
h
|K
h
, v) = d(

i
K(b).K
h
|K
h
, v) = d(K(b)|K, w
i
), and similarly
for the ramication index and the inertia degree.
The last assertion follows directly from Theorem 9.40.
With this corollary, we can also prove:
Theorem 9.43 Let (K, v) be a valued eld of characteristic p > 0. Then (K, v) is a
separably defectless eld if and only if (K, v)
c
is a defectless eld.
Proof: By virtue of Theorem 11.77, (K, v)
c
is separably defectless if and only if it is
defectless. Thus it suces to prove that (K, v)
c
is separably defectless if and only if (K, v)
is separably defectless. Suppose rst that (K, v)
c
is separably defectless. We wish to show
that (K, v) is separably defectless. Let (L|K, v) be a nite separable extension. Using
Lemma 11.99 and Lemma ??, we nd that d(L|K, v) = d(L
c
|K
c
, v) = 1. Hence again by
Lemma ??, (K, v) is separably defectless.
Now suppose that (K, v) is separably defectless, and let (L

|K
c
, v) be a nite separable
extension. Let L

= K
c
(a) and f the minimal polynomial of a over K
c
. We choose
according to Corollary 9.42. Since (K, v) is dense in (K, v)
c
, there exists a polynomial
h K[X] such that v(fh) > . Let b be a root of h. Then by Corollary 9.42, b is separable
over K and d(L

|K
c
, v) = d(K
c
(b)|K
c
, v). By Lemma 11.99 and our assumption that
(K, v) be separably defectless, d(K
c
(b)|K
c
, v) = d(K(a)|K, v) = 1. Hence, d(L

|K
c
, v) = 1,
showing that (K, v)
c
is separably defectless.
9.13 Fields with two henselian valuations
Theorem 9.44 Let v
1
and v
2
be two non-trivial independent henselian valuations on the
eld K. Then K is separable-algebraically closed.
Proof: Let f = X
n
+c
1
X
n1
+. . . +c
0
K[X] be any separable irreducible polynomial;
we wish to show that it is linear. Let n = deg f and choose distinct elements a
1
, . . . , a
n
K
(note that K is innite since it admits non-trivial valuations). Set h :=

n
i=1
(X a
i
) =
X
n
+ d
1
X
n1
+ . . . + d
0
. Choose any
i
v
i
K , i = 1, 2. Let
i
v
i
K be the values
given by the foregoing theorem, depending on
i
(for i = 2, we replace f by h). By the
Strong Approximation Theorem, we can nd elements c

j
K such that v
1
(c
j
c

j
) >
1
and v
2
(d
j
c

j
) >
2
for 0 j < n. Then by the foregoing theorem, the polynomial
g := X
n
+ c

1
X
n1
+ . . . + c

0
is irreducible over K since f is. On the other hand, its
irreducible factors are linear since the same holds for h. This proves that f is linear.
Corollary 9.45 Let v
1
and v
2
be incomparable henselian valuations on the eld K. Let w
denote their nest common coarsening (its valuation ring is the smallest ring containing
O
v
1
and O
v
2
). Then Kw is separable-algebraically closed.
9.14. THE CORE AND THE HENSELIAN PART OF A VALUATION 241
Proof: By Theorem ??, the induced valuations v
1
/w and v
2
/w on Kw are henselian.
Since v
1
and v
2
are incomparable by assumption, w = v
1
and w = v
2
so that both v
1
/w
and v
2
/w are non-trivial on Kw. They are independent because otherwise, they would
admit a non-trivial common coarsening w

, so ww

would be a common coarsening of v


1
and v
2
, ner than w. Now our assertion follows from the foregoing theorem.
9.14 The core and the henselian part of a valuation
In [POP4], F. Pop has given the denition of the core of a valuation (it is a revised version
of an earlier denition given in [POP2]). In this section, let v be a xed valuation on the
algebraic closure

K of the eld K. By this, also the henselization (K
h
, v) of (K, v) is xed.
Let
V := {v} {w coarsening of v | K
h
w is separable-algebraically closed} .
Then we dene the core of v on K to be the valuation inf V. We will denote it by v
K
c
.
Its valuation ring is the union of all valuation rings O
w
for w V, and its valuation ideal
is the intersection of their valuation ideals M
w
.
Lemma 9.46 Let (K, v) be henselian. If v = v
K
c
, then Kv
K
c
and Kv are separable-
algebraically closed. In any case, v
K
c
V.
Proof: Assume that v = v
K
c
, then there is a coarsening w of v such that Kw is
separable-algebraically closed. Hence by ??, also the residue eld Kv of (Kw, v/w) is
separable-algebraically closed.
By virtue of Theorem ??, every coarsening w of v is henselian. Hence in view of
??, Kw is separable-algebraically closed if and only if Gal K is equal to the absolute
inertia group of K, i.e., aw = (a)w for all Gal K and every a O
w
. Here, O
w
can be replaced by O
v
because Kw is the quotient eld of the valuation ring O
v
/M
w
of
(Kw, v/w). On the other hand, aw = (a)w is equivalent to a a M
w
. So we see
that a a

wV
M
w
= M
v
K
c
for every a O
v
and every Gal K. This proves that
Kv
K
c
is separable-algebraically closed.
Suppose that v
1
and v
2
are henselian valuations with cores w
1
resp. w
2
on the eld K.
Since the latter are coarsenings of the former, they are henselian too. So if w
1
and w
2
were incomparable, then we could apply Corollary 9.45 to nd that their nest common
coarsening w has a separable-algebraically closed residue eld. On the other hand, w
would be coarser than w
1
, which contradicts the denition of the core valuation. So we
have proved:
Lemma 9.47 Let v
1
and v
2
be two henselian valuations on a eld K. Then their cores
are comparable.
Lemma 9.48 Assume that (L|K, v) is a normal extension and that (L, v) is henselian.
Then v
K
c
is henselian already on K.
Proof: By ??, all extensions of v from K to L are conjugate. Since L|K is normal,
(L, v) and (L, v) are isomorphic for every Gal K. Consequently, also v is henselian
242 CHAPTER 9. HENSELS LEMMA
on L. So by the foregoing lemma, v and v have comparable cores on L. The reader
may verify that (v)
K
c
= v
K
c
. Hence, comparability implies equality (cf. ??). This shows
that v
K
c
is the only extension of its restriction to K. Since v
K
c
is henselian on L (being a
coarsening of v), this implies that v
K
c
is henselian also on K.
Let us introduce a second signicant coarsening of a given valuation. Beforehand, we
observe:
Lemma 9.49 Let (L|K, v) be an algebraic extension. Dene
Uniq(L|K, v) := {w | w v and w admits a unique extension to L} .
Then the valuation
v
L|K
:= sup Uniq(L|K, v)
also admits a unique extension from K to L. If L|K is nite, then v/v
L|K
admits an
extension from Kv
L|K
to Lv
L|K
which is independent from v/v
L|K
on Lv
L|K
.
Proof: The valuation ring of v
L|K
is

wUniq(L|K,v)
O
(K,w)
, and its valuation ideal is

wUniq(L|K,v)
M
(K,w)
. For every w Uniq(L|K, v), there is a unique valuation ring O
(L,w)
lying above O
(K,w)
. Hence if v

is an extension of v
L|K
to L, then O
(L,w)
O
v
and
M
(L,w)
M
v
. This shows that the valuation ring

wUniq(L|K,v)
O
(L,w)
of L lies above
the valuation ring O
v
of L, which proves that they are equal.
Now assume that L|K is nite, and let v
1
= v, v
2
, . . . , v
n
denote all extensions of v
to L. Let w = min{v
1
, . . . , v
n
}. Then w is the only extension of its restriction to K.
Hence, w v
L|K
. On the other hand, ?? shows that there is i {1, . . . , n} such that
w = min{v, v
i
}. That is, v/w and v
i
/w are independent extensions of v/w from Kw to
Lw. This implies that w v
L|K
, so w = v
L|K
and our assertions are proved.
Now we dene the henselian part of v on K to be the valuation v
K
h
|K
. By the
foregoing lemma, the henselian part has a unique extension to K
h
. Since this extension
is a coarsening of the henselian valuation v on K
h
, it is also henselian. This shows that
the henselian part of v on K is itself henselian. On the other hand, it follows from the
denition that it is the nest coarsening of v which is henselian on K.

You might also like