Union College

Union | Digital Works

Honors Theses Student Work

6-2011

An Introduction to the p-adic Numbers

Charles I. Harrington
Union College - Schenectady, NY

Follow this and additional works at: https://digitalworks.union.edu/theses

Part of the Logic and Foundations of Mathematics Commons

Recommended Citation
Harrington, Charles I., "An Introduction to the p-adic Numbers" (2011). Honors Theses. 992.
https://digitalworks.union.edu/theses/992

This Open Access is brought to you for free and open access by the Student Work at Union | Digital Works. It has been accepted for inclusion in Honors
Theses by an authorized administrator of Union | Digital Works. For more information, please contact [email protected].
 AN INTRODUCTION TO

THE p-adic NUMBERS

By

Charles Irving Harrington

*********

Submitted in partial fulllment

of the requirements for

Honors in the Department of Mathematics

UNION COLLEGE

June, 2011

i
 Abstract
HARRINGTON, CHARLES An Introduction to the p-adic Numbers.
Department of Mathematics, June 2011.
ADVISOR: DR. KARL ZIMMERMANN

One way to construct the real numbers involves creating equivalence classes
of Cauchy sequences of rational numbers with respect to the usual absolute
value. But, with a dierent absolute value we construct a completely dierent
set of numbers called the p-adic numbers, and denoted Q . First, we take
p

an intuitive approach to discussing Q by building the p-adic version of √7.

p

Then, we take a more rigorous approach and introduce this unusual p-adic
absolute value, | | , on the rationals to the lay the foundations for rigor in Q .
p p

Before starting the construction of Q , we arrive at the surprising result that

p

all triangles are isosceles under | | . Then, we quickly construct Q and extend
p p

| | from the rationals. Next, we leave equivalence classes of Cauchy sequences

p

behind and introduce a more understandable view of numbers in Q . With this

p

view, we compute some p-adic numbers and observe that these computations are
similar to analogous computations in the real numbers. Then, we end our tour
of Q with a proof of Hensel's Lemmaa result describing a general approach
p

to building p-adic numbers. Lastly, we move to nite eld extensions of Q . We

p

extend | | to these eld extensions with the help of the norm function, and end
p

the paper with two important propositions that characterize most nite eld
extensions of Q .
p

ii
Contents

1 Introduction 1

2 Foundations 3
2.1 p -adic Expansions: An Intuitive Approach . . . . . . . . . . . . . . . 3

2.2 Absolute Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 The p -adic Numbers 16

3.1 Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.2 Interpreting Qp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.3 Calculations in Qp . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.4 Hensel's Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4 Finite Extensions 32
4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

4.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

iii
1 Introduction

At an early age we are only exposed to certain numbers, the integers and rationals,

restricting our grasp of mathematical concepts to only part of the number line. Later,

armed with our new favorite tool, the calculator, we take on the irrationals and

fully explore the real numbers while rumors of numbers existing illegally by taking

the square root of a negative number oat around the outskirts of our developing

mathematical minds. The boundaries of manipulating numbers with sophisticated

laws and axioms do not stop with the complex numbers and keep disturbing our

comfort with numbers. In this paper, we go one step further, and introduce another

such disturbance: the p-adic numbers.

√
Observe the familiar base-10 expansion of a real number, 7:

√
7 = 2.645 . . . = 2(10)0 + 6(10)−1 + 4(10)−2 + 5(10)−3 + · · · .

But, if we use a dierent perspective, viewing the 2-adic expansion of −1 leads to

a similar looking expansion which challenges our comfort with numbers and raises

interesting questions:

−1 = 1(2)0 + 1(2)1 + 1(2)2 + 1(2)3 + · · · .

This admittedly seems suspicious, however, if we add 1 to both sides of the equation,

1
we have

−1 + 1 = 1 + 1 + 1(2) + 1(22 ) + 1(23 ) + · · ·

0 = 0 + 1(2) + 1(2) + 1(22 ) + 1(23 ) + · · ·

= 0 + 2(2) + 1(22 ) + 1(23 ) + · · ·

= 0 + 0 + 1(22 ) + 1(22 ) + 1(23 ) + · · ·

= 0 + 0 + 2(22 ) + 1(23 ) + · · ·

= 0 + 0 + 0 + 2(23 ) + · · ·

= 0 + 0 + ··· + 0 + ··· .

Now, we are in fact simply shuing powers of 2 farther to the right of the equation,

but in the world of p-adic numbers, pn → 0 as n → ∞. So, our goal is to describe a

scenario in which we view 2n as a small number when n is large. In doing so, we are

really talking about  2-adic numbers.

√
A more interesting number, such as 7, has an analogous representation in the

3-adic numbers (see Example 2.2 for the derivation). For now, consider

√
7 = 1(3)0 + 1(3)1 + 1(3)2 + 0(3)3 + 2(3)4 + · · · .

Note again the positive exponents, a signature of p-adic integers. Also, just as our
√ √ √
familiar 7 ∈
/ Q but 7 ∈ R, this 7 is an element in a dierent eld extending

Q: the eld of 3-adic numbers Q3 . Recall from real analysis, one way to construct

the real numbers involves creating equivalence classes of Cauchy sequences of rational

numbers, relative to the usual absolute value. Analogously, the eld of p-adic num-

bers, Qp , may be constructed from the rationals with Cauchy sequences by the same

procedure with a dierent absolute value. Whereas the familiar || ignores the sign

of a number, the p-adic absolute value, | |p , unintuitively measures the divisibility

2
of a number by a xed prime p. We will make sense of our p-adic expansions using

the fact that pn is very small in terms of this absolute value. Now, we present other

concrete properties of the p-adics and show how we formed the 3-adic expansion of
√
7. Our rst approach is to intuitively explain the behavior of the p-adic numbers,

and then we will take a more mathematically rigorous approach.

2 Foundations

2.1 p -adic Expansions: An Intuitive Approach

Before diving into an example, we dene a useful property underlying our p-adic
expansions.

Denition 2.1. Let p be a prime. We say a sequence (αn ) of integers with 0 ≤ αn ≤

pn+1 − 1 is coherent if, for every n ≥ 0, we have

αn+1 ≡ αn (mod pn+1 ).

√
First, we examine a number less intimidating than the irrational 7. Take 58 ∈
Q5 , as 5 is a comfortable base to use. We will soon see that every p-adic number can

be represented in base-p, or more precisely a Laurent series [1]

Qp = {a−n p−n + · · · + a0 + a1 p + a2 p2 + · · · | 0 ≤ ai ≤ p − 1}. (2.1)

We see below that the 5-adic expansion of 58 works the same as expanding plain, old

58 ∈ Z into base-5.

3
Example 2.1. The 5-adic expansion of 58 is as expected, 58 = 3 + 1(5) + 2(5)2 . The

elements α0 , α1 , α2 of the sequence (αn ) are

α 0 = a0 = 3

α1 = a0 + a1 p = 3 + 1(5) = 8

α2 = a0 + a1 p + a2 p2 = 3 + 1(5) + 2(5)2 = 58.

Note that the rest of the elements in (αn ) are equal to 58, or 58 = α3 = α4 = · · · .
Since we constructed the expansion around powers of 5, the elements 3, 8, and 58 are
related by reducing modulo powers of 5. That is 58 ≡ 8 (mod 52 ) and 8 ≡ 3 (mod 5).

Introducing coherent sequences for integers does not seem particularly helpful, but

now we use additional information to unlock the sequence for the more complicated
√
7.

√
Example 2.2. To nd the 3-adic expansion of 7, we must build the expression

rather than dismantle it as in Example 2.1by using Denition 2.1 in reverse. We

want to start with the bottom term α0 and build to α = lim αn = a0 +a1 p+a2 p2 +· · · .
n→∞

So, keeping some number theory in mind, we begin with

α ≡ α0 (mod 3)

α0 ≡ α (mod 3)

(α0 )2 ≡ (α)2 (mod 3).

4
Then, note we may replace (α)2 with 7, and

(α0 )2 ≡ 7 (mod 3)

(α0 )2 ≡ 1 + 2(3) (mod 3)

(α0 )2 ≡ 1 (mod 3),

implying α0 equals 1 or 2. Semi-surprisingly, the 2 3-adic expansion

begins a second
√
that we represent with the other root to the equation x2 − 7 = 0, − 7. Let us
concentrate on α0 = a0 = 1. We continue building our expansion and arrive at an

expression for the next term in our sequence,

(α1 )2 ≡ (α)2 (mod 32 ).

Now, α1 = a0 + a1 (3) = 1 + a1 (3) since we must have α1 ≡ α0 (mod 3), and so

(1 + a1 (3))2 ≡ 7 (mod 32 )

1 + 2a1 (3) + (a1 )2 32 = 1 + 2(3) (mod 32 )

2a1 (3) = 2(3) (mod 32 )

a1 ≡ 1 (mod 3),

meaning a1 = 1 since 0 ≤ a1 ≤ 2. Next, since our coecients are to be taken from

{0, 1, . . . , p − 1} note that α2 = a0 + a1 p + a2 p2 = 1 + (3) + a2 (32 ) as α2 ≡ α1 (mod 32 ).

5
We show the steps for one more term;

(1 + (3) + a2 (32 ))2 ≡ 7 (mod 33 )

1 + 2(3) + (2a2 + 1)32 + 2a2 33 + (a2 )2 34 ≡ 1 + 2(3) (mod 33 )

(2a2 + 1)32 = 0 (mod 33 )

2a2 + 1 ≡ 0 (mod 3)

⇒ a2 = 1.

√
Continuing further, we obtain our result: the aforementioned 3-adic expansion + 7 =
1 + 1(3) + 1(32 ) + 0(33 ) + 2(34 ) + · · · . Had we continued the example with a0 = 2,
√ 2 3 4
we would get − 7 = 2 + 1(3) + 1(3 ) + 2(3 ) + 0(3 ) + · · · , the negative version of
√ √
our rst expansion. Just as + 7 and − 7 are related in the real numbers, we relate

these numbers in the 3-adics. Observe that

√
−(+ 7) = −(1 + 1(3) + 1(32 ) + 0(33 ) + 2(34 ) + · · · )

= −1 − 1(3) − 1(32 ) − 2(34 ) + · · ·

= (2 − 3) + (2 − 3)(3) + (2 − 3)(32 ) + (1 − 3)(34 ) + · · ·

= 2 + (2 − 1)(3) + (2 − 1)(32 ) − 1(33 ) + 1(34 ) + · · ·

= 2 + 1(3) + 1(32 ) + (2 − 3)(33 ) + 1(34 ) + · · ·

= 2 + 1(3) + 1(32 ) + 2(33 ) + 0(34 ) + · · ·

√
= − 7.

With an intuitive understanding of p-adic expansions, we now turn our attention

to the p-adic absolute value, | |p , to make our argument more rigorous.

6
2.2 Absolute Values

The absolute value lays the groundwork for the rigor required to construct the p-adics.
We dene it as follows. To set notation, let R+ = {r ∈ R | r ≥ 0}.

Denition 2.2. An absolute value on a eld F is a function

| | : F → R+

that satises the following conditions:

i) |x| = 0 if and only if x=0

ii) |xy| = |x||y| for all x, y ∈ F
iii) |x + y| ≤ |x| + |y| for all x, y ∈ F .
We will say an absolute value on F is non-archimedean if it satises the additional

condition:

iv) |x + y| ≤ max{|x|, |y|} for all x, y ∈ F ;

otherwise, we will say that the absolute value is archimedean.

Condition iv) of the denition is a bit unusual and will be explored after an

example.

Example 2.3. Let F =Q and let || be the usual absolute value dened by



x
 if x ≥ 0
|x| = .

−x if x < 0


This absolute value is archimedean, as taking x=y=1 for example, violates condi-

tion iv).

7
 For comparison, || dened by



1 if x 6= 0

|x| =

0 if x = 0


is non-archimedean, and is called the trivial absolute value.

The absolute value has the following properties.

Lemma 2.1. For any absolute value || on a eld F, we have for all x ∈ F:
i) |1| = 1
ii) |xn | = 1 ⇒ |x| = 1, for n∈Z
iii) | − 1| = 1
iv) | − x| = |x|

Def 2.2
Proof. This proof follows [2, p. 27]. For the rst statement, note that |1| = |1·1| =
|1| · |1| = |1|2 . Since |1| = r, a positive real number, and r = r2 ⇒ r = 1, and we

get our result. Next, |xn | = 1 ⇒ |x|n = 1, and once again since |x| ∈ R+ we get

i) ii)
|x| = 1. For iii), see that | − 1|2 = |(−1)2 | = |1| = 1 ⇒ | − 1| = 1. Finally,

iii)
| − x| = | − 1| · |x| = |x|.

Now, we reach a critical point in our groundwork, as the next denition is tied to

the p-adic absolute value.

Denition 2.3. Fix a prime number p ∈ Z. The p-adic valuation on Z is the function

vp : Z \ {0} → R

dened as follows: for each integer n ∈ Z, n 6= 0, let vp (n) be the unique positive

integer satisfying

n = pvp (n) n0 where p 6 | n0 .

8
We extend vp to the eld of rational numbers as follows: if x = a/b ∈ Q \ {0}, then

vp (x) = vp (a) − vp (b),

and if x = 0, then vp (0) = +∞, treating innity with the usual conventions.

Since vp is an exponent, it makes sense to extend it to the rationals as the dier-

ence between the p-adic valuation of the numerator and the p-adic valuation of the

denominator of a rational number. Let us compute an example for this seemingly

random number.

Example 2.4. Take v5 (3/35). First, 3 = 50 (3) and 35 = 51 (7). So, v5 (3) = 0 and

v5 (35) = 1, meaning v5 (3/35) = 0 − 1 = −1. Now, think back to Example 2.2. We

haven't dened what this p-adic valuation means in the eld extending Q, our desired
eld Qp (this denition will come much later). But, since we conveniently have an
√ √
element from Q3 , 7, we may as well ponder v3 ( 7). Recall from Example 2.2 that
√
7 = 1 + 1(3) + 1(32 ) + 0(33 ) + 2(34 ) + · · · . We will see that for elements of Qp ,
vp will be determined as it is for elements of Q. So, since a0 = 1 6= 0, we cannot
√ √
factor out any powers of 3 and 7 = 30 α 0 where α0 ∈ Q3 (clearly α0 = 7), meaning
√
v3 ( 7) = 0.
More generally, recall from Equation (2.1) that a p-adic number may be repre-

sented as a Laurent series. So, imagine some p-adic number α, given by the expan-

sion α = a2 p 2 + a3 p 3 + · · · . Then, α = p2 (a2 + a3 p + · · · ) = p2 α0 for α 0 ∈ Qp , and

so vp (α) = 2. Also, as we will see in Lemma 2.2, since vp (ab) = vp (a) + vp (b),
vp (α) = vp (p2 ) + vp (a2 + a3 p + · · · ) = 2 + 0 = 2. For expansions with neg-

ative powers, take λ ∈ Qp given by λ = a−2 p−2 + a−1 p−1 + a0 + a1 p + · · · =

1
(a−2 + a−1 p + a0 p2 + a1 p3 + · · · ). Then, λ is analogous to a rational number where
p2
λ = α/β with α, β ∈ Zp ⊆ Qp , the eld of p-adic integers. So, as with a rational num-
ber, vp (λ) = vp (α) − vp (β) = vp (a−2 + a−1 p + a0 p2 + a1 p3 + · · · ) − vp (p2 ) = 0 − 2 = −2.

9
We will conrm these p-adic musings later.

Example 2.4 suggests a nice interpretation of the p-adic valuation. If we expand

a number (taken from Q or Qp ), vp gives the power of the p multiplying the rst

non-zero number in the expansion. We will observe later that this is, in fact, the

case.

Now, back to rm footing, we examine some suggestive properties of the p-adic
valuation vp on Q in the form of a lemma.

Lemma 2.2. For all x, y ∈ Q, we have

i) vp (xy) = vp (x) + vp (y)

ii) vp (x + y) ≥ min{vp (x), vp (y)}.

Proof. i) First, let a, b ∈ Z and write a = pvp (a) a0 and b = pvp (b) b0 where p 6 | a0 and

p 6 | b0 , and note that since p is prime p 6 |a0 b0 .

ab = pvp (a)+vp (b) a0 b0 = pvp (ab) n
Then,

r
for some n ∈ Z, p 6 | n, and the result follows for vp on Z. So, if x, y ∈ Q, x =
s
t r
and y = with s, z 6= 0, and r, s, t, z ∈ Z, we see vp (x) = vp ( ) = vp (r) − vp (s) and
z s
t
vp (y) = vp ( ) = vp (t) − vp (z). Thus, vp (x) + vp (y) = (vp (r) + vp (t)) − (vp (s) + vp (z)) =
z
rt
vp (rt) − vp (sz) = vp ( ) = vp (xy), establishing the result for vp on Q.
sz
v (a) 0
ii) Assume vp (a) < vp (b). Then, a+b = p p a +pvp (b) b0 = pvp (a) (a0 +pvp (b)−vp (a) b0 ),
showing vp (a + b) = vp (a) ≥ vp (a). An analogous argument works if vp (b) < vp (a).
Now, let x, y ∈ Q be as above and assume vp (x) < vp (y). Then, vp (r) − vp (s) <

10
vp (t) − vp (z) ⇒ vp (r) + vp (z) < vp (t) + vp (s) ⇒ vp (rz) < vp (ts). So,

 
rz + ts
vp (x + y) = vp
sz
= vp (rz + ts) − vp (sz)

≥ min{vp (rz), vp (ts)} − (vp (s) + vp (z))

= vp (rz) − (vp (s) + vp (z))

= vp (r) − vp (s)

= vp (x),

and a similar argument works if we assume vp (y) < vp (x).

Sneakily, Lemma 2.2 suggests dening an absolute value based on the p-adic val-

uation. Looking back at the absolute value Denition 2.2, condition ii) and the

non-archimedean condition iv) slightly resemble the above observed properties of vp .

In fact, if we make vp an exponent and negate it, Lemma 2.2 clauses i) and ii) fulll

Denition 2.2 ii) and iv), respectively, bringing us to the p-adic absolute value.

Denition 2.4. For any x ∈ Q \ {0}, we dene the p-adic absolute value of x by

|x|p = p−vp (x)

and we set |0|p = 0.

Note that this matches our denition of vp when vp (0) = +∞, as |0|p = p−vp (0) =
p−∞ = 0, if we continue to treat ∞ as usual. As established in the preceding para-

graph, this new absolute value is non-archimedean, making | |p very unfamiliar.

Example 2.5. To become more familiar, we calculate some p-adic absolute values.

From Example 2.4, |3/35|5 = 5−v5 (3/35) = 5−(−1) = 5. Also, we will see that | |p
√
extends to the p-adic numbers and using v3 ( 7) = 0 from the example, we nd

11
 √
| 7|3 = 30 = 1. For the number crunchers, we try |4243686|29 . To start, 4243686 =
1
294 (6), meaning v29 (4243686) = 4, and |4243686|29 = 29−4 = . In the end, we
707281
are just calculating friendly rational numbers since | |p is an absolute value after all.

As revealed in the introduction, numbers divisible by a high power of p are small

under the p-adic absolute value. Looking to Example 2.1 for |58|5 , although 58 is

pretty close to 50 = 2(52 ), 58 is not divisible by 5, and |58|5 = 1. Then, while 50 is

1 1
divisible by 5, |50|5 = is not so small compared to , for example. But, take
√ 25 707281
pn (as in 7 = 1(3)0 + 1(3)1 + 1(3)2 + 0(3)3 + 2(3)4 + · · · + an (3n ) + · · · for p = 3).
n n 1 n
Then, as n → ∞, |p |p → 0 because |p |p = , meaning p becomes more divisible
pn
by p as n → ∞.

To end this section, we introduce some topology to fully ensure the oddities of

the p-adic absolute value (or any non-archimedean absolute value in general) sink in

before leaving behind such concrete mathematics for more abstract grounds.

Denition 2.5. Let F be a eld and | | an absolute value on F . The metric induced

by || is the distance d(x, y) between some x, y ∈ F and is given by

d(x, y) = |x − y|.

The set F on which the metric d(x, y) is dened is called a metric space.

A metric, for our lowly intent and purpose, is a fancy synonym for a distance

function, but distance-function-space fails to match the allure of metric space. A non-

archimedean metric spacea place where | |p could livemeasures distance strangely,

as seen below.

Proposition 2.1. Let F be a eld and let || be a non-archimedean absolute value on

F. If x, y ∈ F and |x| =
6 |y|, then

|x + y| = max{|x|, |y|}.

12
Proof. Assume x, y ∈ F and |x| > |y|, meaning max{|x|, |y|} = |x|. We will show |x+
y| ≤ |x| and |x| ≤ |x + y| to arrive at our desired equality. We get the rst inequality

directly from the non-archimedean condition, i.e., |x + y| ≤ max{|x|, |y|} = |x|. Next,

note again by the special condition that |x| = |(x+y)+(−y)| ≤ max{|x+y|, |y|} since
| − y| = |y| by Lemma 2.1. Then, either |x| ≤ |x + y| or |x| ≤ |y|. But, the second

relation contradicts our assumption that |x| > |y|, and so |x| ≤ |x + y|. Therefore,

|x + y| = |x| = max{|x|, |y|}. Assuming |y| > |x| gives an analogous result.

It is strange that the absolute value of a sum equals the absolute value of a

summand, but, applied to distances and our concept of geometry involving triangles

and circles, it is even stranger.

Corollary 2.1. Let F be as above. Then, all triangles in F are isosceles.

Proof. Let x, y, z ∈ F , a non-archimedean metric space, be the vertices of our tri-

angle. Then, the lengths of the sides of the triangle are: d(x, y) = |x − y|,
d(y, z) = |y − z|, and d(x, z) = |x − z|. Assume, for example, d(x, y) 6= d(y, z).
Then, |x − y| =
6 |y − z| and we use Proposition 2.1 to see that d(x, z) = |x − z| =
|(x − y) + (y − z)| = max{|x − y|, |y − z|} = max{d(x, y), d(y, z)}. Thus, two of the

sides are equal and we have an isosceles triangle.

We look to a concrete example to see that a non-isosceles triangle in the rationals

may be isosceles in the p-adics.

Example 2.6. First, let F =Q with | |5 and let x, y, z ∈ Q be vertices of a triangle.

Then, for example let x = 200, y = 185, and z = 85. By our p-adic metric, d(x, y) =
1 1
|15|5 = and d(y, z) = |100|5 = . Since d(x, y) 6= d(y, z) we know by Corollary
5 25
1
2.1, d(x, z) = |x − y + y − z|5 = |15 + 100|5 = max{d(x, y), d(y, z)} = . A simple
5
calculation of |115|5 conrms this result that d(x, z) = d(x, y) and we are in fact

dealing with some sort of isosceles triangle.

13
 Next, we think intuitively again and let F = Qp with | |p and examine the p-adic
metric on the eld of p-adic numbers, recalling Example 2.4 if necessary. Also, recall

from Equation 2.1 that we may use a Laurent series to represent elements in Qp . So,

let λ, µ, ν ∈ Qp be vertices of a triangle with λ − µ = a−1 p−1 + a0 + a1 p + a2 p2 + · · ·

1
and µ − ν = b1 p + b2 p 2 + · · · . Then, d(λ, µ) = p 6= d(µ, ν) = and the corollary says
p
1
d(λ, ν) = max{p, } = p, which is true, as d(λ, ν) = |λ − ν|p = |a−1 p−1 + a0 + (a1 +
p
2
b1 )p + (a2 + b2 )p + · · · |p = p.

Now, we move to circles, or balls, in this strange metric space. First, a quick

denition.

Denition 2.6. Let F be a eld with an absolute value | |. Let a∈F and r ∈ R+ .
Then, the open ball of radius r and center a is the set

B(a, r) = {x ∈ F | d(x, a) < r} = {x ∈ F | |x − a| < r}.

The closed ball of radius r and center a is the set

B(a, r) = {x ∈ F | d(x, a) ≤ r} = {x ∈ F | |x − a| ≤ r}.

Once again, we consider a non-archimedean metric space which includes our | |p .

Proposition 2.2. Let F be a eld with a non-archimedean absolute value | |. Then,

every point in an open or closed ball is a center of the ball.

Proof. This proof follows [2, p.34]. Consider the open ball B(a, r) with center a∈F
and radius r ∈ R+ , and let b ∈ F be an arbitrary point in B(a, r). We will show

B(b, r) = B(a, r). To that end, let x ∈ B(b, r). Then, by Denition 2.6, |x − b| < r.
But, also since b ∈ B(a, r), |b − a| < r. Thus, by the non-archimedean property,

|x − a| = |(x − b) + (b − a)| ≤ max{|x − b|, |b − a|} < r, meaning x ∈ B(a, r), and

14
so B(b, r) ⊆ B(a, r). Similarly, we nd B(a, r) ⊆ B(b, r), and therefore B(b, r) =
B(a, r). Replace < with ≤ to get the result for a closed ball.

As with the isosceles triangle, we discuss the open ball in Q and Qp with | |p in

one last example.

1
Example 2.7. We describe B(3, ) in Q with | |7 by rst describing the ball in the
7
integers and then informally describing it in the rationals. The ball contains elements

1
a∈Z such that |a − 3|7 < , meaning v7 (a − 3) > 1. So, if we expand a−3 base-7,
7
it must equal an expansion with all powers of 7 greater than 1. Therefore, since the

expansion of a−3 cannot contain . . . , 7−1 , 70 , 7 72 , 73 , 74 , . . . then

but may contain

1
72 |(a − 3) ⇒ a ≡ 3 (mod 72 ) ⇒ a ∈ {. . . − 95, −46, 3, 52, 101, . . .} ⊇ B(3, ) ∩ Z. Or,
7
1 2 3
from an expansion point of view, integers in B(3, ) include 3+2(7 ) = 101, 3−4(7 ) =
7
1
−1369, 3 + 6(74 ) + 1(717 ) + 3(788 ), etc. Then for determining x ∈ Q ∩ B(3, ), we take
7
a purely expansionist view. Since v7 (x − 3) > 1, for x − 3 the p-adic valuation of the

numerator must be at least two greater than that of the denominator so the dierence
3 4
5(7 ) + 5(7 ) 13720
is greater than 1. These elements include 3+ = 3 + ,
2 + 1(7) + 4(72 ) + 6(75 ) 101047
etc.

Now, we look at the same ball and absolute value in Q7 . Note that elements of

Q have nite expansions, but are nonetheless Laurent series, and so it is intuitively

clear that Q7 extends Q. Therefore, the ball in Q7 includes the above elements, yet

also contains elements following the same rules established above but with innite

expansions. That is, α ∈ B(3, 1), α = 3 + a2 (72 ) + a3 (73 ) + a4 (74 ) + · · · , where

0 ≤ ai ≤ 6. These elements cannot be pictured as easily as the rationals, and we will

work to interpret the p-adic numbers.

Now that we have dened and dissected the p-adic absolute value, we are ready to
construct the mysterious eld of p-adic numbers Qp that we have increasingly hinted

at in examples.

15
3 The p -adic Numbers

3.1 Construction

As the main goal of this paper is to introduce the p-adic numbers, we construct the

eld Qp without providing much proof so as not to distract the reader. We are more

concerned with providing a concrete understanding of the numbers than constructing

them. By the end of this section we will be able to forget parts of the construction,

but then can continue to extensions knowing that the p-adic numbers actually exist.

As stated in the introduction, the construction of Qp is similar to that of R. As the

real numbers complete the rationals with respect to the usual | |, the p-adics complete
the rational numbers with respect to the p-adic absolute value | |p . In fact, R and Qp
are the only elds which complete Q in this manner because no other absolute values
exist as asserted in the following theorem.

For convenience, | |∞ , called the prime at innity, represents the usual absolute

value.

Theorem 3.1. (Ostrowski's Theorem) Every non-trivial absolute value on Q is

equivalent to one of the absolute values | |p , where either p is a prime number or

p=∞ [2, p.43].

A nice application of Theorem 3.1 comes in the form of a product formula for

absolute values seen below.

Proposition 3.1. For any x ∈ Q \ {0}, we have

Y
|x|p = 1,
p≤∞

where p≤∞ means we take the product of all the primes of Q [2, p.46].

With this formula and all but one of the absolute values of a rational number,

16
we can determine the missing absolute value. We provide essentially a sketch of the

proof in an example.

Example 3.1. This argument follows [2, p.46]. Let x ∈ Z \ {0} with prime factor-

ization x = pa11 · pa22 · · · pakk (think back to factor trees). Then,






 |x|q = 1 if q 6= pi


−ai
 |x|pi = pi if i = 1, 2, . . . , k ,



|x|∞ = x = pa11 · pa22 · · · pakk



and so |x|q · |x|p1 · |x|p2 · · · |x|pk · |x|∞ = 1. Reinforcing our analysis, take 18928 ∈ Z.
This number breaks down to 18928 = 24 · 7 · 132 . So, for unused primes such as 3,
1 1
|18928|3 = 0
= 1. For primes included in the factorization like 2, |18928|2 = 4 and
3 2
Q 1 1 1
clearly |18928|∞ = 18928. Therefore, |18928|p = 4 · · 2 · 18928 = 1.
p≤∞ 2 7 13

Now, we recall important denitions from real analysis which will guide us to the

p-adic numbers.

Denition 3.1. Let F be a eld and let || be an absolute value on F.

i) A sequence of elements xn ∈ F is called a Cauchy sequence if for all ε > 0, there
exists a bound N ∈N such that |xm − xn | < ε whenever m, n ≥ N .
ii) The eld F is called complete with respect to || if every Cauchy sequence of

elements of F has a limit in F.

iii) A subset S⊂F is called dense in F if for every x∈F and for every ε>0 we

have B(x, ε) ∩ S 6= ∅.

Note, part iii) of the denition can also be worded as S is dense in F if every

open ball around every element of F contains an element of S. We remind ourselves

of these terms with an example.

17
 √
Example 3.2. Consider the rst real number introduced in this paper, 7 = 2.645 . . .,
and form the sequence (an ) = {2, 2.6, 2.64, 2.645, . . .}. This is certainly Cauchy. Let
1 1
ε > 0 and pick N ∈ N such that N < ε. Then if m, n ≥ N , |am −an | < min{m,n} ≤
10 10
1 √
< ε. Next, by construction we know lim (an ) = 7. Similarly, every limit of
10N n→∞

Cauchy sequences in R lives in R, and therefore the real numbers are complete with

respect to | |∞ . Unfortunately, since the elements of (an ) are rational numbers, but
√
7∈
/ Q, Q is not complete with respect to | |∞ . Finally, the rational numbers are
√
dense in the real numbers. For example, 7 ∈ R is surrounded by the two rationals
2.645 and 2.646, and we may use (an ) to nd even closer numbers if need be.
√
Analogously, and thinking intuitively again (for another page or so) with 7=
1 + 1(3) + 1(3)2 + 0(3)3 + 2(3)4 + · · · and the sequence (αn ) = {1, 1 + 1(3), 1 + 1(3) +
1(3)2 , 1+1(3)+1(3)2 +0(3)3 , 1+1(3)+1(3)2 +0(3)3 +2(3)4 . . .} = {1, 4, 13, 13, 175, . . .}
we nd Q is not complete with respect to | |3 .

From the above two examples and Theorem 3.1, we surmise the following lemma.

Lemma 3.1. The eld Q of rational numbers is not complete with respect to any of

its nontrivial absolute values [2, p.49].

So, let us construct a completion of Q with respect to | |p . This entails adding all

the limits of | |p -Cauchy sequences to Q. We constructed something that represents

√
a limit to 7 ∈ Q3 in Section 2. But, we clearly do not have any other limits and

cannot possibly conceive what they look like. Thus, our approach will be to replace

the limits we do not have with the equivalence classes of limits of Cauchy sequences

we do [2, p.52]. We start this process with some set notation.

Let C = {(xn ) | (xn ) is a Cauchy sequence with respect to | |p } be the set of all

Cauchy sequences of elements of Q with respect to | |p . This set has a ring structure

18
as shown,

(xn ) + (yn ) = (xn + yn )

(xn ) · (yn ) = (xn · yn ).

Clearly, C is a commutative ring with identity. Dene f :Q→C by the constant

sequence f (x) = {x, x, x, . . .} denoted by (x). Now, note that f is 1-1 and f (Q) ⊆ C .
So, we say there is an inclusion of Q into C denoted Q ,→ C .
Unfortunately, C is not a eld because not all non-zero elements are invertible.

Let (xn ) ∈ C where (xn ) 6= 0 = {0, 0, 0, . . .}. If any term in (xn ) equals zero (e.g.

{1, 0, 1, 0, . . .}), then (xn ) · (yn ) 6= 1 = {1, 1, 1, . . .} for any (yn ) ∈ C since 0 is not

invertible, and so (xn )−1 does not exist. Moreover C is not a eld, as it contains

zero divisors. The product of the non-zero two sequences in C , {0, 1, 1, 1, . . .} and

{1, 0, 0, 0, . . .}, is clearly {0, 0, 0, . . .}.

Naturally, some Cauchy sequences share the same limit. Since it is the limits we

are after for the construction, the limits dictate how we treat the elements in C, and

so we dene an equivalence relation that groups together sequences in C that have

the same limit. That is, we say two Cauchy sequences are equivalent when they share

the same limit and we dene a set N ⊂C of sequences that tend to zero with respect

to the absolute value | |p , or N = {(xn ) | lim |xn |p = 0}. Clearly, N is an ideal since
n→∞

for (xn ) ∈ C and (yn ) ∈ N , lim |xn yn |p = lim |xn |p · 0 = 0 and so (xn ) · (yn ) ∈ N ,
n→∞ n→∞

and for (zn ), (yn ) ∈ N (zn ) + (yn ) ∈ N . The next result is less obvious.

Lemma 3.2. N is a maximal ideal of C [2, p.52].

Thus, recall from abstract algebra that the quotient ring of a maximal ideal is a

eld, and we nally dene the eld of p-adic numbers.

Denition 3.2. We dene the eld of p-adic numbers to be the quotient of the ring

19
C by its maximal ideal N:
Qp = C/N .

Now, we must consider the p-adic absolute value in our new eld. By [2, p.54], a

sequence of real numbers (|xn |p ) is eventually stationary provided (xn ) is Cauchy and
so we dene | |p as expected.

Denition 3.3. If λ ∈ Qp and (αn ) is any Cauchy sequence representing λ, we dene

|λ|p = lim |αn |p .

n→∞

Without proof, we claim that the rationals are in fact dense in the eld of p-adic
numbers and this eld is complete. To summarize, we have the following.

Theorem 3.2. For each prime p∈Z there exists a eld Qp with a non-archimedean

absolute value | |p , such that:

i) there exists an inclusion Q ,→ Qp (dened via constant Cauchy sequences), and

the absolute value on Qp induced by | |p , as dened in Denition 3.3, is the p-adic

absolute value;

ii) f (Q) is dense in Qp with respect to | |p ; and

iii) Qp is complete with respect to | |p .

The eld Qp satisfying (i), (ii), and (iii) is unique up to unique isomorphism

preserving the absolute values [2, p.57].

As promised, the last part of Theorem 3.2 allows us to continue without paying

attention to the details of the completion. The eld of p-adic numbers is unique,

meaning no other eld shares its established properties, and so we concentrate on

these properties to give a concrete description of a p-adic number (as a Laurent

series) in the next section.

20
3.2 Interpreting Qp

First, we check that the p-adic valuation vp from Section 2 makes sense in Qp . Prop-

erty i) from Theorem 3.2 and Denition 3.3 implies that | |p returns the same values

1 1
for elements of Q and Qp , namely {. . . , , , 1, p, p2 , . . .}. So in that regard, vp
p2 p
behaves the same in Qp as it does in Q a la Denition 2.3.

Lemma 3.3. For each λ ∈ Qp \ {0}, there exists an integer vp (λ) such that |λ|p =
p−vp (λ) , meaning the p-adic valuation vp extends to Qp [2, p.58].

We rst met the valuation in Example 2.4 and interpreted vp as the power of the p
multiplying the rst non-zero number in an expansion of a rational or p-adic number.
Now, we return to this concrete interpretation of p-adic numbers as expansions that

has only been proposed in examples, and solidify it. Consider the following important

theorem.

Theorem 3.3. Every equivalence class [α] in Qp for which |α|p ≤ 1, has exactly one

representative Cauchy sequence of the form (αn ) for which:

i) 0 ≤ αn ≤ pn+1 − 1 for n = 0, 1, 2, . . .
ii) αn+1 ≡ αn (mod pn+1 ) for n = 0, 1, 2, . . . [3, p.11].

Using this theorem, we nally conrm the concrete interpretation of p-adic num-

bers we have only portrayed in examples. The properties of Theorem 3.3 mirror

Denition 2.1, the denition of coherent sequences dened back in Section 2. We

understand how these sequences work, remembering Example 2.1 and the famous
√
7 Example 2.2. So, we satisfy conditions i) and ii) with the familiar sequence

21
(αn ) = {α0 , α1 , α2 , . . .} given by

α0 = a0

α1 = a0 + a1 p

α2 = a0 + a1 p + a2 p2
.
.
.

αn = a0 + a1 p + a2 p2 + · · · + an pn

αn+1 = a0 + a1 p + a2 p2 + · · · + an pn + an+1 pn+1

.
.
..

Then, since Qp is complete by Theorem 3.2, we take the limit of this sequence to

produce an element α in Qp . That is, α = lim (αn ) = a0 + a1 p + a2 p2 + · · · where

n→∞

0 ≤ ai ≤ p − 1 and |α|p ≤ 1. We make these numbers a special subset of the p-adic

numbers, called the p-adic integers, dened by Zp = {λ ∈ Qp | |λ|p ≤ 1}. We briey
√
discussed p-adic integers in Example 2.4 and saw 7 ∈ Z3 , and we see in an example
that these numbers indeed have no negative powers of p and that the p-adic valuation
is the rst power of p multiplied by a non-zero coecient.

Example 3.3. Let β ∈ Q7 such that β = 6(72 ) + 4(73 ) + 1(74 ) + · · · . Then, (βn ) =
{β0 , β1 , β2 , . . .} is a Cauchy sequence representing β given by

β0 = 0

β1 = 0

β2 = 6(72 )

β3 = 6(72 ) + 4(73 )

β4 = 6(72 ) + 4(73 ) + 1(74 )

.
.
..

22
Clearly, after the rst couple zero terms in the sequence (βn ), the valuations of the pro-
ceeding terms β2 , β3 , β4 , . . . do not change. That is, since each new term βn+1 adds on

to the preceding term βn , v7 (β2 ) = v7 (β3 ) = v7 (β4 ) · · · = 2, or v7 (β) = lim v7 (βn ) = 2.

n→∞

Therefore, the 7-adic valuation is indeed the rst power of 7 multiplied by a non-zero
1 1
coecient. Then, by Denition 3.3, |β|7 = lim (βn ) = lim = ≤ 1, and
n→∞ n→∞ 7v7 (βn ) 72
β ∈ Z7 .

Now, Theorem 3.3 does not address λ ∈ Qp where |λ|p > 1. But, we rem-

edy this simply by multiplying by a power of p. Assume µ ∈ Qp with |µ|p = pm .

1 m
Then, multiply µ by pm to get |pm µ|p = |pm |p |µ|p = p = 1 and so pm µ ∈ Zp .
pm
Then, since pm µ is a p-adic integer, it is given by p m µ = a0 + a1 p + a2 p 2 + · · · +
am pm + · · · where 0 ≤ ai ≤ p − 1. Dividing by pm , we arrive at the expres-
a0 a1 a2
sion µ =
m
+ m−1 + m−2 + · · · + am + am+1 p + am+2 p2 + · · ·. Since we assumed
p p p
m 1
|µ|p = p = −m ⇒ vp (µ) = −m, agreeing with our established visualization of the
p
p-adic valuation.
Thus, every λ ∈ Qp can be written as a Laurent series

a−n a−n+1 a−1

λ= + + · · · + + a0 + a1 p + a2 p 2 + · · ·
pn pn−1 p

where 0 ≤ an ≤ p − 1 (we shued the subscripts to conveniently match the power

of p). Interestingly, note that the p-adic numbers extend innitely to the right, but

nitely to the left. We compare this attribute to the real numbers in an example.

7 7 7
Example 3.4. Look at ∈R and ∈ Q3 . In R, = 1(100 ) + 1(10−1 ) + 6(10−2 ) +
6 6 6
6(10−3 )+· · · = 1.166 . . .. So, according to our familiar notation, the coecients of the

negative powers of 10 in a decimal expansion are the digits to the right of the decimal
7 2
place. Then in Q3 , = +2+1(3)+1(32 )+· · · = · · ·+1(32 )+1(3)+2(30 )+2(3−1 ) =
6 3
. . . 112.2.
Also of interest is the uniqueness of a p-adic expansion asserted by Theorem

23
3.3. Whereas in R, 0.999 . . . and 1 are interchangeable, in Qp any two p-adic which

converge to the same p-adic number have the same digits.

So, in a sense the p-adic numbers are opposite the real numbers. While the reals

are nite to the left and innite to the right, the p-adics are nite to the right and

innite to the left. This makes arithmetic in Qp just slightly dierent from that in R.

3.3 Calculations in Qp

Calculations in Qp are similar to our usual mode of operating in R except here we

borrow and carry numbers from left to right, rather than right to left. First, we

show an example of p-adic multiplication and then we will divide two p-adic numbers.

Example 3.5. Multiply 2+3(5)+1(52 )+3(53 )+ · · · and 4+2(5)+2(52 )+1(53 )+ · · ·

in Q5 . We start at the left and move right.

2 + 31 (5) + 12 (52 ) + 31 (53 ) + · · ·

× 4 + 2(5) + 2(52 ) + 1(53 ) + · · ·
3 + 3(5) + 1(52 ) + 3(53 ) + · · ·

The italicized numbers in the top multiplier are the carried terms. We have

4(2 + 3(5) + 1(52 ) + 3(53 ) + · · · ) = 8 + 12(5) + 4(52 ) + 12(53 ) + · · ·

= (3 + 1 (5)) + 12(5) + 4(52 ) + 12(53 ) + · · ·

= 3 + (3 + 2 (5))(5) + 4(52 ) + 12(53 ) + · · ·

= 3 + 3(5) + (1 + 1 (5))(52 ) + 12(53 ) + · · ·

= 3 + 3(5) + 1(52 ) + 3(53 ) + · · · .

We continue and note that adding p-adic numbers is similar to adding polynomials.

24
 2 + 3(5) + 1(52 ) + 3(53 ) + · · ·
× 4 + 2(5) + 2(52 ) + 1(53 ) + · · ·
3 + 3(5) + 1(52 ) + 3(53 ) + · · ·
4(5) + 1(52 ) + 3(53 ) + · · ·
4(52 ) + 1(53 ) + · · ·
+ 2(53 ) + · · ·
3 + 2(5) + 2(52 ) + 0(53 ) + · · ·

Now, we provide an example of division in Q7 .

1 + 6(7) + 3(72 ) + 5(73 ) + · · ·

Example 3.6. Calculate in Q7 .
2 + 5(7) + 1(72 ) + 4(73 ) + · · ·

4
2 + 5(7) + 1(72 ) + 4(73 ) + · · · 1 + 6(7) + 3(72 ) + 5(73 ) + · · ·
− 8 + 20(7) + 4(72 ) + 16(73 ) + · · ·

Then,

8 + 20(7) + 4(72 ) + 16(73 ) + · · · = (1 + 7) + (6 + 2(7))(7) + 4(72 ) + (2 + 2(7))(73 ) + · · ·

= 1 + 1(7) + 6(7) + 2(72 ) + 4(72 ) + 2(73 ) + · · ·

= 1 + 7(7) + 6(72 ) + 2(73 ) + · · ·

= 1 + 3(73 ) + · · · .

So,

25
 4 + 3(7) + 1(72 ) + 3(73 ) + · · ·
2 + 5(7) + 1(72 ) + 4(73 ) + · · · 1 + 6(7) + 3(72 ) + 5(73 ) + · · ·
− 1 + 0(7) + 0(72 ) + 3(73 ) + · · ·
6(7) + 3(72 ) + 2(73 ) + · · ·
− 6(7) + 1(72 ) + 5(73 ) + · · · (= 6(7) + 15(72 ) + 3(73 ) + ·
2(72 ) − 3(73 ) + · · ·
− 2(72 ) + 5(73 ) + · · ·
6(73 ) + · · · (= −8(73 ) + · · · )

a1 p + a2 p 2 + a3 p 3 + · · ·
Consider a number in Qp such as λ= . If we solve for λ
b3 p 3 + b4 p 4 + b5 p 5 + · · ·
3 −2
as in our example, we need to multiply b3 p by some cp to subtract o a1 p. So, the

quotient λ starts its expansion at p−2 , meaning λ is not a p-adic integer. We easily

conrm this by recalling vp (λ) = 1 − 3 = −2, the power of the rst p in λ.

Now, we will nd the invertible elements of Zp , i.e., the p-adic units. So, we use

1
the above analysis and keep quotients in mind. Let α ∈ Zp and we require ∈ Zp .
α
Then, α ∈ Zp ⇒ |α|p ≤ 1 ⇒ vp (α) ≥ 0, and so vp (α−1 ) ≤ 0. But, vp (α−1 ) < 0 ⇒
|α−1 |p > 1 ⇒ α−1 ∈
/ Zp , meaning |α|p = 1 ⇒ α−1 ∈ Zp . Thus, the set of p-adic units

is dened by Z×
p = {α ∈ Zp | |α|p = 1}. Or equivalently, α = a0 + a1 p + a2 p 2 + · · ·
where a0 6= 0 which implies α 6≡ 0 (mod p), and Z×
p = {α ∈ Zp | α 6≡ 0 (mod p)}.

With a concrete representation of the p-adic numbers and a grasp on calculations,

we conclude this section with the proof of an important abstract algebra result.

3.4 Hensel's Lemma

With this theorem we test whether a polynomial has roots in Zp , which, recalling

some abstract algebra, is necessary to nd irreducible polynomials and create eld

extensions.

26
Theorem 3.4. (Hensel's Lemma) Let f (x) = c0 + c1 x + c2 x2 + · · · + cn xn be a

polynomial with coecients in Zp and suppose that there exists α0 ∈ Zp such that

f (α0 ) ≡ 0 (mod p) and

f 0 (α0 ) 6≡ 0 (mod p),

where f 0 (x) is the derivative of f (x). Then, there exists α ∈ Zp such that α ≡
α0 (mod p) and f (α) = 0.

Proof. We will construct a Cauchy sequence of rational integers which converges to a

root α to show it exists. We have built many of these sequences already so this proof

will not be too unusual. First, we claim there exists a sequence of non-p-adic integers

{α0 , α1 , α2 , . . .} such that for all n ≥ 0:

i) f (αn ) ≡ 0 (mod pn+1 ) and
ii) αn+1 ≡ αn (mod pn+1 ).
We proceed by induction to show for all n ≥ 0, we can construct αn satisfying

these properties.

Base Case: First, we show i) and ii) hold when n = 0. For i), let a0 ∈ {0, . . . , p −
1} satisfy α0 ≡ a0 (mod p). Then, f (α0 ) ≡ f (a0 ) (mod p) and f (α0 ) ≡ 0 (mod p).
Next, for any 0 ≤ a1 ≤ p − 1 we need to nd α 1 = a0 + a1 p , satisfying condition ii).

Then, we show a1 exists by solving for it using the given polynomial f (x) and the

fact that

f (α1 ) ≡ f (a0 + a1 p) (mod p2 ).

To continue we will use a Taylor series of f (x) centered around a0 . Recall f (x) =

27
 1
f (a0 ) + f 0 (a0 )(x − a0 ) + f 00 (a0 )(x − a0 )2 + · · · . Now, we plug in for α1 and simplify,
2

f (α1 ) = f (a0 + a1 p)
1
= f (a0 ) + f 0 (a0 )(a1 p) + f 00 (a0 )(a1 p)2 + · · ·
2
≡ f (a0 ) + f 0 (a0 )(a1 p) (mod p2 ).

If we are to have α1 , it should satisfy condition i). That is

f (a0 ) + f 0 (a0 )(a1 p) ≡ 0 (mod p2 ).

Next, f (a0 ) ≡ f (α0 ) ≡ 0 (mod p) ⇒ f (a0 ) = βp for some β ∈ Zp , and this implies

f (a0 ) ≡ b1 p (mod p2 ) where 0 ≤ b1 ≤ p − 1. So,

b1 p + f 0 (a0 )(a1 p) ≡ 0 (mod p2 )

b1 + f 0 (a0 )(a1 ) ≡ 0 (mod p).

But, by assumption f 0 (a0 ) ≡ f 0 (α0 ) 6≡ 0 (mod p), and thus we may divide by f 0 (a0 )
to get a1 ≡ −b1 [f 0 (a0 )]−1 (mod p), showing a1 exists. So, we have established the

base case and move to the inductive hypothesis.

Inductive Step: Assume we have the stated properties for a0 , α1 , . . . , αn . Similar

to our base step, we let αn+1 = αn + αn+1 pn+1 where 0 ≤ αn+1 ≤ p − 1 and try to

determine αn+1 using the fact that αn+1 must satisfy

f (αn+1 ) = f (αn + an+1 pn+1 )

1
= f (αn ) + f 0 (αn )(an+1 pn+1 ) + f 00 (αn )(an+1 pn+1 )2 + · · ·
2
≡ f (αn ) + f 0 (αn )(an+1 pn+1 ) (mod pn+2 ).

28
We want αn+1 to satisfy i), so f (αn+1 ) ≡ 0 (mod pn+2 ). Also by the inductive

assumption, f (αn ) ≡ 0 (mod pn+1 ) ⇒ f (αn ) ≡ bn+1 pn+1 (mod pn+2 ) for some 0≤
bn+1 ≤ p − 1 and

f (αn ) + f 0 (αn )(an+1 pn+1 ) ≡ 0 (mod pn+2 )

bn+1 pn+1 + f 0 (αn )(an+1 pn+1 ) ≡ 0 (mod pn+2 )

bn+1 + f 0 (αn )(an+1 ) ≡ 0 (mod p).

Then, since αn ≡ α0 (mod p) ⇒ f 0 (αn ) ≡ f 0 (α0 ) 6≡ 0 (mod p), we get an+1 ≡

−bn+1 [f 0 (αn )]−1 (mod p). Therefore, αn+1 = αn + an+1 pn+1 satises condition ii), and

we have proved our claim with induction.

So, we have constructed the sequence (αn ) = {a0 , α1 , α2 , . . .} with our desired

properties. As expected, let α = a0 + a1 p + a2 p 2 + · · · . Clearly, α = a0 + a1 p +

a2 p2 + · · · ≡ a0 ≡ α0 (mod p) and since we have f (α) ≡ f (αn ) ≡ 0 (mod pn+1 ) for all

n ≥ 0, then f (α) = 0. Also, note that the choice at each step was unique, so the root

constructed is unique.

We end this section with a nal example using Hensel's Lemma.

Example 3.7. We determine whether three polynomials have roots in Zp using

Hensel's Lemma (Theorem 3.4). According to the theorem, we need f (α0 ) ≡ 0 (mod p)
and f 0 (α0 ) 6≡ 0 (mod p) where α0 ∈ Zp . Since we require these congruences mod p,
then only the rst coecient in the p-adic expansion of α0 matters, that is the digit

multiplying p0 . Clearly, this digit can be taken from the set {0, . . . , p − 1}.
√
First, we use our relentless example and conrm 7 ∈ Z3 . Then, let f (x) = x2 − 7

29
and f 0 (x) = 2x with α0 ∈ {0, 1, 2}. We test the possible α0

α0 f (α0 ) f 0 (α0 )

0 −7 ≡ 1 6≡ 0 (mod 3)

1 −6 ≡ 0 (mod 3) 2 6≡ 0 (mod 3)

2 −3 ≡ 0 (mod 3) 4 ≡ 1 6≡ 0 (mod 3)

Since the degree of f (x) is two, we expect at most two roots, and according to our

chart, there exists α ∈ Z3 such that α ≡ α0 ≡ 1 (mod p) and α ≡ 2 (mod p)

with f (α) = 0. We have seen these α0 already in Example 2.2. But, these simple

calculations using Hensel's Lemma quickly reveal 7 is a 3-adic square. To actually

√
build the 3-adic expansion of 7 refer back to the old example.

Next, we take a mildly more interesting polynomial, letting f (x) = x2 + x − 6

which clearly has roots −3 and 2 in Z. We will check for roots in Z3 and Z5 . So, we

need α0 ∈ Z3 and β0 ∈ Z5 satisfying the input of Theorem 3.4 where α0 ∈ {0, 1, 2}

and β0 ∈ {0, 1, 2, 3, 4}. Then, recalling f (x) = 2x + 1,

α0 ∈ Z3 f (α0 ) f 0 (α0 ) β0 ∈ Z5 f (β0 ) f 0 (β0 )

0 0 1 0 4
1 2 and 1 1
2 0 2 2 0 0
3 4
4 1

Since Z3 and Z5 extend Z, then 2 is certainly a root in both elds. So, note that

while all β0 ∈ Z5 fail Hensel's Lemma, this does not imply that f (x) has no roots in

Z5 . Rather, this implies that we cannot denitively use Hensel's Lemma to conclude

that f (x) has roots in Z5 . To compute 5-adic expansion of the second root, simply

30
compute the 5-adic expansion of −3.
Now, consider a more interesting polynomial in Z3 that does not have roots in
√
Z such as f (x) = 7x2 − 3x + 2. We are allowed to use Hensel's Lemma because
√ √
7 ∈ Z3 , meaning 7 is equivalent to an integer in the set {0, 1, 2} mod 3. That is,
√ √
we know from Example 2.2 (and above) that 7 ≡ 1 (mod 3) and 7 ≡ 2 (mod 3).
√
But we cannot have both, so we establish the convention that + 7 ≡ 1 (mod 3) and
√
− 7 ≡ 2 (mod 3). Then,

α0 f (α0 ) f 0 (α0 )

0 2

1 0 2

2 0 1

and this polynomial has roots in Z3 , and we would build these 3-adic integers using

Hensel or some 3-adic quadratic formula. For amusement, since we use the notation
√ 2
± 7 to represent the 3-adic expansions of the rootsp
of x −7 in Z3 , we could represent
√
√ 2 3± 9−8 7
the roots of 7x − 3x + 2 with the notation √ .
2 7

Now that we have lightly studied roots of polynomials, it makes sense to transition

to eld extensions. While, we saw that the construction and calculations in Qp are

quite similar to those in R, the two elds are not so similar. One glaring dierence is

each eld's algebraic closure. The eld of p-adic numbers is complete, but turns out

to not be algebraically closed. This is true for the real numbers as well, but there is

no easy analogous eld of complex numbers (both complete and algebraically closed)

for Qp . Whereas it takes one step to go from R to C, it takes two steps to go from

Qp to a complete and algebraically closed eld containing Qp . So, we now introduce

some important properties of nite eld extensions of Qp to assist in determining its

algebraic closure.

31
4 Finite Extensions

4.1 Preliminaries

This section heavily relies on concepts from abstract algebra, and the reader may

want to refer to [3, p.52] for a quick review on the subject or [4] for more detail.

Qp is not algebraically closed, and so in this section we categorize some of its eld

extensions to help nd the algebraic closure. Field extensions of Qp are simply elds

K that contain Qp . This also means K is a vector space over Qp , and we write

K/Qp to denote a eld extension K over Qp . In this section, we only consider nite

extensions K/Qp , that is, elds that are nite dimensional over Qp .
Just as we extended the p-adic absolute value | |p from Q to Qp , we now look to

extend | |p from Qp to K. Denote this new absolute value by | |. We require || to

satisfy the usual properties of a non-archimedean absolute value from Denition 2.2,

and additionally the property that |λ| = |λ|p for λ ∈ Qp . Without proof, we note the

following.

Proposition 4.1. There is at most one absolute value on K extending the p-adic
absolute value on Qp [2, p.129].

This proposition will help us later. Next, the following function will help us dene

this new absolute value.

Denition 4.1. The norm from K to F is a function NK/F : K → F and can be

dened in several (equivalent) ways. Here, are three denitions:

i) Take α ∈ K, a nite-dimensional F -vector space, and consider the F -linear

map from K to K given by multiplication by α. Since this is a linear transformation

of vector spaces, it corresponds to a matrix. Then we dene NK/F (α) to be the

determinant of this matrix.

ii) Let α ∈ K, and consider the subeld F (α). Then, let r = [K : F (α)] be the

32
degree of K as an extension of F (α). Let

f (x) = xn + an−1 xn−1 + · · · + a1 x + a0 ∈ F [x]

be the minimal polynomial of α over F such that f (α) = 0. Then we dene

NK/F (α) = (−1)nr ar0 .

Q
iii) Let K/F be a nite, normal extension. Then, NK/F (α) = σ(α) where σ is
σ
an F -automorphism of K.

We will use each denition depending on which one makes most sense in context.

The denitions can be proven equivalent (see [2, p.132]), but we settle with an

example instead.

√
Example 4.1. This argument follows [2, p.133]. Consider the eld K = Q5 ( 2)
√ √ √
over F = Q5 . Note that 2 6∈ Q5 . We calculate the norm of α = a + b 2 ∈ Q5 ( 2)
using all three interpretations:
√ √ √ √
i) A basis for Q5 ( 2) over Q5 is {1, 2}. LetTa+b√2 : Q5 ( 2) → Q5 ( 2) be
√ √ √ √
dened by Ta+b√2 (x) = (a + b 2)(x). Then, T (1) = a + b 2 and T ( 2) = 2b + a 2.
So, the corresponding matrix with respect to our basis is

 
 a 2b 
 ,
b a

√
which has determinant a2 − 2b2 . Therefore, NQ5 (√2)/Q5 (a + b 2) = a2 − 2b2 .
ii) We look at two cases. First, if b = 0, then α=a and we consider the subeld
√
Q5 (a) ⊆ Q5 ( 2). Since a ∈ Q5 , then Q5 (a) = Q5 . Thus, the minimal polynomial

of a f (x) = x − a, meaning n = [Q5 (a) : Q5 ] = deg(f ) = 1, and these elds

is just
√ √ 2
are equal. So, r = [Q5 ( 2) : Q5 (a)] = [Q5 ( 2) : Q5 ] and since g(x) = x − 2 is
√
the minimal polynomial of 2 over Q5 , r = deg(g) = 2. Therefore, NQ5 (√2)/Q5 (a) =
√
(−1)nr ar0 = (−1)(1)(2) (a)2 = a2 . Next, in the second case b 6= 0 and α = a + b 2,

33
 √ √
and we consider Q5 (a + b 2) ⊆ Q5 ( 2). Let h(x) be the minimal polynomial of
√
a+b 2 over Q5 . The degree of h(x) cannot be 1 because that would imply h(x) =
√ √
x − (a + b 2) ⇒ 2 ∈ Q5 , a contradiction. Then, h(x) = x2 − 2ax + (a2 − 2b2 )
√ 2 2
√ 2
because we note that (a + b 2) = a + 2ab 2 + 2b and working backwards nd that

√ √
0 = (a + b 2)2 − a2 − 2ab 2 − 2b2
√ √
= (a + b 2)2 − 2a(a + b 2) + (a2 − 2b2 )
√
= h(a + b 2).

√
Therefore n = [Q5 (a + b 2) : Q5 ] = deg(h) = 2. But, from above we know that
√ √ √
[Q5 ( 2) : Q5 ] = deg(x2 − 2) = 2, implying that Q5 (a + b 2) = Q5 ( 2). Thus
√ √ √ √ √
r = [Q5 ( 2) : Q5 (a + b 2)] = [Q5 ( 2) : Q5 ( 2)] = 1, and so NQ5 (√2)/Q5 (a + b 2) =
(−1)(2)(1) (a2 − 2b2 )1 = a2 − 2b2 , which agrees with our calculation from i). Also, if we

set b = 0, we get our rst result, namely a2 − 2(0)2 = a2 .

√
iii) The Q5 -automorphisms of Q5 ( 2) map roots of the minimal polynomial of
√
2 over Q5 to each other and leave elements of Q5 unchanged. There are two roots

of x2 − 2, ι and τ. That is, ι and

meaning there are two automorphisms, the identity
√√ √√ √
τ map a 7→ a and b 7→ b, while and τ ( 2) = − 2. So, τ (a + b 2) =
ι( 2) = 2
√ √ √ √
τ (a) + τ (b)τ ( 2) = a − b 2 and as expected ι(a + b 2) = a + b 2. Therefore,
√ Q √ √ √ √ √
NK/F (a + b 2) = σ(a + b 2) = ι(a + b 2) · τ (a + b 2) = (a + b 2)(a − b 2) =
σ
a2 − 2b2 .

From part i) of Denition 4.1, it is clear that the norm is multiplicative since

determinants are multiplicative. Next, to see what the extended absolute value must

be, assume K is normal. Then, let | |0 : K → R+ be dened by |x|0 = |σ(x)| where

σ is a K -automorphism. This is clearly an absolute value, but by Proposition 4.1

there is only one | | extending | |p , and so |x| = |x|0 = |σ(x)|. Now, recall that

if K/Qp is normal, there are n = [K : Qp ] K -automorphisms, namely σ1 , . . . , σn .

34
 |x|n = | σ(x)|
Q
So, and using part iii) of the denition for a norm, we see that
σ
|x|n = |x| · · · |x| = |σ1 (x)||σ2 (x)| · · · |σn (x)| = |σ1 (x) · · · σn (x)| = |NK/Qp (x)|. But,
p
the norm lies in Qp , and therefore |x| = |NK/Qp (x)|p . We formally put this into a
n

theorem.

Theorem 4.1. Let K/Qp be a nite extension of degree n. The function | | : K → R+

dened by
q
|x| = n |NK/Qp (x)|p

is a non-archimedean absolute value on K which extends the p-adic absolute value on

Qp [2, p.135].

Note that the extension K does not need to be normal. Also, this new | | is clearly
an absolute value, as it satises Denition 2.2 (to get multiplication, recall the norm

is multiplicative). Also, || satises our requirement that |λ| = |λ|p for λ ∈ Qp . Use

part ii) of Denition 4.1 and remember that the minimal polynomial over Qp for any

λ ∈ Qp is simply x − λ. Now, we calculate two easy examples.

√ √
Example 4.2. First, we calculate the absolute value of 10+5 2 in Q5 ( 2). From Ex-
√ √ 2
ample 4.1 ii) we know that n = [Q5 ( 2) : Q5 ] = 2 and NQ (√2)/Q (10+5 2) = (10) −
5 5
1 √ q √
2(5)2 = 50. Then, |50|5 = |2(52 )|5 = 2 . Therefore, |10+5 2| = n |NQ5 (√2)/Q5 (10 + 5 2)|5 =
r 5
p 1 1
|50|5 = = .
52 5 √ √ √ √
Next, we calculate |2 7 + 3| in Q3 ( 3) (recall that 7 ∈ Q3 ). Similar to the
√ 2 2
calculation in Example 4.1 iii), we see that NQ (√3)/Q (a + b 3) = a − 3b . Also,
3 3
√ √ √
r
p 1
again n = [Q3 ( 3) : Q3 ] = 2, and so |2 7 + 3| = |25|3 = = 1.
30
Before we begin examining eld extensions of Qp , we introduce one last familiar

proposition.

Proposition 4.2. (Eisenstein Irreducibility Criterion) Let

f (x) = an xn + an−1 xn−1 + · · · + a1 x + a0 ∈ Zp [x]

35
be a polynomial satisfying the conditions

i) |an |p = 1,
ii) |ai |p < 1 for 0 ≤ i < n, and
1
iii) |a0 |p = .
p
Then f (x) is irreducible over Qp [2, p.140].

This is an extremely useful result and we provide an example.

Example 4.3. Here we show that x2 − 3 is irreducible over Q3 . For the rst two

1
conditions, |1|3 = 1 and |0|3 = 0 < 1. Then, | − 3|3 = , and this polynomial is
3
irreducible by Eisenstein. Furthermore, we can adjoin a root of x2 − 3 to Q3 to gain
√
the extension Q3 ( 3) that we dealt with in Example 4.2.

4.2 Properties

Now, we are ready to present some information about nite extensions of Qp . First, we

note that the p-adic valuation extends to nite extensions and is dened, as expected,
in the usual way.

Denition 4.2. Let K be a nite extension of Qp , and let || be the p-adic absolute

value on K. For any x ∈ K \ {0}, we dene the p-adic valuation vp (x) to be the

unique rational number satisfying

|x| = p−vp (x)

and we set vp (0) = ∞.

Next, combining this denition and Theorem 4.1, we can solve for the p-adic
1
valuation, and get vp (x) = vp (NK/Qp (x)) for any x ∈ K \ {0}. Now, an example.
n
√ √
Example 4.4. Let 10 + 5 2 ∈ Q5 ( 2); we will calculate its 5-adic valuation. From
√ √
Example 4.2, NQ (√2)/Q (10 + 5 2) = 50 and n = [Q5 ( 2) : Q5 ] = 2. So, v5 (10 +
5 5

36
 √ 1 √ 1
5 2) = v5 (NQ5 (√2)/Q5 (10 + 5 2)) = v5 (50) = 1 which agrees with our previous
n 2
√ 1
result that |10 + 5 2| = .
√ √ 5 √ √
Now, let 2 7 + 3 ∈ Q3 ( 3). Then, using Example 4.2 again, we nd v3 (2 7 +
√ 1
3) = v3 (25) = 0.
2
1
Clearly, the image of vp is contained in Z = {x ∈ Q | nx ∈ Z}. More precisely,
n
1
this image is a non-trivial additive subgroup of Z that contains Z, and therefore,
n
1
it must equal Z for some positive integer e dividing n [3, p.66]. We give this e a
e
special name, as it will help us sort out some properties of nite eld extensions of

Qp .

Denition 4.3. Let K/Qp be a nite extension, and let e be the unique positive

integer dividing n = [K : Qp ] dened by

1
vp (K \ {0}) = Z = {x ∈ Q | ex ∈ Z}.
e

We call e the ramication index of K over Qp . We say the extension K/Qp is unram-

ied if e = 1. We say the extension is ramied if e > 1, and totally ramied if e = n.

n
Finally, we write f= .
e
Soon, we will see that f corresponds to something much more important than

1
simply the quotient of two integers. Intuitively, we want to play the same role in
e
nite eld extensions of Qp as vp (p) plays in Qp . That is, since vp (p) = 1, any λ ∈ Zp
can be written as λ = pvp (λ) u where u is a p-adic unit, or vp (u) = 0. So, we dene

1
such a number analogous to p with valuation equal to .
e
Denition 4.4. Let K/Qp be a nite extension, and let e be the ramication index.

1
We say an element π∈K is a uniformizer if vp (π) = .
e
Note that there are many α ∈ Zp with vp (α) = 1, and similarly there are many

uniformizers for a eld K. Also, in the unramied case where e = 1, we can set

37
 1
π = p. Finally, notice that if vp (K \ {0}) = Z, then there exists π ∈ K \ {0} such
e
1
that vp (π) = , meaning vp (π e ) = 1. So rather than expansions of p as in Zp , we
e
have expansions of π in K. Now, we make the above denitions concrete through

calculation.

Example 4.5. In this example we nd the ramication index e and uniformizer π of
√
Q3 ( 3). From Example 4.2, we know n = 2, e must divide
and from Denition 4.3,
√
2. Therefore, e = 1 or e = 2. Now since the valuation extends Q3 to Q3 ( 3), v3 (Q3 \
√ 1 1 1 1
{0}) = Z ⊆ v3 (Q3 ( 3) \ {0}) = Z (and remember Z ⊆ Z = Z). That is, if
e e n 2
√ 1 1
e = 1, v3 (Q3 ( 3) \ {0}) = Z = Z = Z, and 3 (for example) would be a uniformizer
√ e 1
of Q3 ( 3). So to show e = 2, we just need to nd an element whose valuation is not

1 1 3 1 1 3
simply contained in Z, but contained in Z = Z = {. . . , − , −1, − , 0, , 1, , . . .}.
e √2 √ 2 2 2 2√
√
Well, recall from Example 4.2 that for a + b 3 ∈ Q3 ( 3), NQ ( 3)/Q (a + b 3) =
3 3
√ 1
a2 −3b2 , meaning v3 (a+b 3) = v3 (a2 −3b2 ). Thus, we do not want v3 (a2 −3b2 ) to be a
2
√ √ √ 1 1 1
multiple of 2. Take 3 ∈ Q3 ( 3). Then, v3 ( 3) = v3 ((0)2 − 3(1)2 ) = v3 (−3) = .
√ 2 √ 2 2
So, the ramication index of Q3 ( 3) is e = 2 and π = 3 is a uniformizer (another
√
easy uniformizer to nd is 3 + 3).

Now, we set up the algebraic structure of K.

Proposition 4.3. Let K be a nite extension of Qp of degree n, and let

A = {x ∈ K | |x| ≤ 1} = {x ∈ K | vp (x) ≥ 0},

M = {x ∈ K | |x| < 1} = {x ∈ K | vp (x) > 0}.

Then, A is a ring, M is its unique maximal ideal, and A/M is a nite extension of

Fp of degree at most n [3, p.64].

The eld A/M is called the residue eld of K, and [A/M : Fp ] = f . The ring A
is called the valuation ring of | |p in K. This is the f that we revealed in Denition

38
 n
4.3, and it indeed has the property f= .
e

Proposition 4.4. Let K/Qp be a nite extension with [K : Qp ] = n and ramication

index e. Then, the degree of the nite eld with pf elements over the nite eld of p
n
elements is [A/M : Fp ] = . In other words, A/M = Fpf [2, p.146].
e

Now, we are ready to give our rst description of a eld extension of Qp . Recall

that we may nd an extension of a eld F by adjoining the root of an irreducible

polynomial. With totally ramied extensions, we can specify this polynomial.

Proposition 4.5. Let K/Qp be a totally ramied nite extension of Qp of degree n.

Then, K = Qp (π), where π is a uniformizer. Furthermore, π is a root of a polynomial

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

which satises the conditions of the Eisenstein criterion.

1
Proof. Let π be a uniformizer (Denition 4.4). Then, vp (π) = , and since K/Qp
e
1
is totally ramied, e = n = [K : Qp ] so that vp (π) = . Or equivalently, |π| =
n
p−vp (π) = p−1/n . But also, using the denition (Theorem 4.1) of absolute value in a
p
nite extension of Qp , |π| = n
|NK/Qp (π)|p .
Now, let f (x) be the minimal polynomial of π over Qp with degree s where s|n
such that n = [K : Qp ] = [K : Qp (π)] · [Qp (π) : Qp ] = r · s. Then, f (x) =
xs + as−1 xs−1 + · · · + a1 x + a0 where ai ∈ Q p and f (π) = 0. Using denition ii)

of the norm (Denition 4.1), we have NK/Qp (π) = (−1)nr ar0 .

Then, using our absolute values dened above, p−1/n = |π| = (|(−1)nr ar0 |p )1/n =
|ar0 |p 1/n = |a0 |p r/n = |a0 |p 1/s . Now, note that |a0 |p ∈ {. . . , p−2 , p−1 , 1, p, p2 , . . .} since

a0 ∈ Qp . Let |a0 |p = pz for some z ∈ Z. Then, p−1/n = |a0 |p 1/s = (pz )1/s = pz/s . So,
z −1
= ⇒ zn = −s, meaning s ≥ n. But, we know n = rs ⇒ s ≤ n, and so s = n.
s n
−1
Thus, z = −1 and |a0 |p = p .

39
 Thus since f (x), the minimal polynomial of π over Qp , is of degree s = n =
[K : Qp ], K = Qp (π) as desired. Next, we have |a0 |p = p−1 , condition iii) of the

Eisenstein criterion (Proposition 4.2), and we also clearly have |an |p = 1, condition

i). For condition ii), let π1 , π2 , . . . , πn be roots of f (x). Since the roots have the

same minimal polynomial f (x), they have the same norm by denition ii), and thus

|π1 | = |π2 | = · · · = |πn | = p−1/n < 1. The coecients ai of f (x) are combinations of

the roots, that is f (x) = (x − π1 )(x − π2 ) · · · (x − πn ), and it follows that |ai |p < 1 for

0≤i<n by the non-archimedean principle. There, conditions i), ii), and iii) hold

and f (x) is an Eisenstein polynomial.

In the next proof, we describe the unramied extensions of Qp . Here, we use a bar

represent an element in a nite eld. Also in a eld F, F× denotes the multiplicative

group of non-zero elements of F.

Proposition 4.6. There is exactly one unramied extension Kfunram of Qp of degree

f, and it can be obtained by adjoining a primitive (pf − 1)th root of unity. If K is an

extension of Qp of degree n, index of ramication e, and residue eld degree f (so that

n = ef as in Proposition 4.4), then K = Kfunram (π), where π satises an Eisenstein

polynomial with coecients in Kfunram .

Proof. This proof follows [3, p.67]. Let α be a generator of the multiplicative group

f −1
F×
pf
so that F×
pf
= {α, α2 , . . . , αp }, and let P (x) = xf + af −1 xf −1 + · · · + a0 , ai ∈ Fp
be its minimal polynomial over Fp . That is, deg(P ) = [Fp (α) : Fp ] = [Fpf : Fp ] = f .
For each i, let ai ∈ Zp be any element such that ai ≡ ai (mod p), and let P (x) =
xf + af −1 xf −1 + · · · + a0 . Clearly, P (x) is monic. Now, if P (x) were reducible it would
be the product of two polynomials, P (x) = (xm + bm−1 xm−1 + · · · + b1 x + b0 )(xl +
cl−1 xl−1 + · · · + c1 x + c0 ) where m + l = f , m, l ≥ 1, and coecients bi , ci ∈ Zp could

be reduced (mod p) to coecients in Fp . But, this new product would equal P (x),
contradicting that P (x) is irreducible. Therefore, P (x) is irreducible. So, take a root

40
α of P (x) and adjoin this to Qp to get Qp (α) = K , and let A = {x ∈ K | |x| ≤ 1}
and M = {x ∈ K | |x| < 1} as in Proposition 4.3. Then, n = [Qp (α) : Qp ] = f ,
while the coset α+M has the degree f minimal polynomial P (x) over Fp . Thus,

n
[A/M : Fp ] = f and using the relation f = from Proposition 4.4 shows e = 1,
e
meaning Qp (α) is an unramied extension of Qp of degree f.
For uniqueness, let K
e be another nite extension of Qp of degree n, index of

ramication e, and residue eld degree f. Also, let e = {x ∈ K

A e | |x| ≤ 1} and

f = {x ∈ K
M e | |x| < 1}, so that A/
e Mf = Fp f . Then, let β ∈ Fp f be a generator of

the multiplicative group F×

pf
. Let β0 ∈ A
e be any element that reduces to β mod M
f.
1
Finally, let π∈K
e be any element with vp (π) = .
e
f −1
We claim that there exists β ≡ β0 (mod π) where β∈K
e such that βp − 1 = 0,
i.e., we want β to be a (pf − 1)th root of unity. The technique used to prove this

claim is very similar to that used to prove Hensel's Lemma (Theorem 3.4). That

f −1
is, we write β ≡ β0 + β1 π (mod π 2 ). Then, we want βp − 1 = 0 ⇒ (β0 +
f
β1 π)p −1 − 1 ≡ 0 (mod π 2 ). Next, recall the binomial theorem which says, (x + y)n =
         
n n 0 n n−1 1 n n−2 2 n 1 n−1 n 0 n
x y + x y + x y +· · ·+ xy + x y . So, we expand
0 1 2 n
 f− 1  n
pf −1 pf −1 f pf −2 p − 1 pf −3
(β0 +β1 π) = β0 +(p −1)β0 β1 π+ β0 (β1 π)2 +· · · , but π 2 ≡ π 3 ≡
2
f
· · · ≡ π p −1 ≡ 0 (mod π 2 ), and we may ignore these higher terms. Also, since pf is the
f −1 f −2
order of A/
e Mf, pf ≡ 0 (mod π), and so we reduce our equation to β0p − β0p β1 π −
f −1 f −2
1 ≡ 0 (mod π 2 ). Solving, we get β1 ≡ (β0p − 1)/(πβ0p ) (mod π). Continuing in

f −1
this way, we nd a solution β = β0 + β1 π + β2 π 2 + · · · to the equation βp = 1.
2 pf −1 f −1
Finally, since the elements β, β , . . . , β of Fp f are all distinct, β, β 2 , . . . , β p
are distinct, meaning β is indeed a primitive (pf − 1)th root of unity. Note that

[Qp (β) : Qp ] ≥ f = [A/

e Mf : Fp ] = [Fpf : Fp ] by Proposition 4.3.

Now, the above construction of β also applies to our α from the rst paragraph.

Therefore, β ∈ Qp (α) and Qp (β) ⊂ Qp (α), where we just showed β is a primitive

(pf − 1)th root of unity. But, we saw f = [Qp (α) : Qp ] ≥ [Qp (β) : Qp ] ≥ f , which

41
implies Qp (α) = Qp (β). Thus, the unramied extension of degree f is unique, and

we denote it Kfunram .
For the last part of the proof, we must show K = Kfunram (π), where π is a uni-

formizer of K over Qp and satises an Eisenstein polynomial. This is similar to our ar-

gument in Proposition 4.5. So, let E(x) be the minimal polynomial with degree d of π
over Kfunram such that E(x) = xd +cd−1 xd−1 +· · ·+c1 x+c0 = (x−π1 )(x−π2 ) · · · (x−πd ).
This polynomial is clearly monic (|cd |p = |1|p = 1). Then, by Denition 4.1 each πi
has the same norm, and so |π1 | = |π2 | = · · · = |πd | = p−1/e < 1. Now, since the coe-

cients ci are combinations of the roots πi , it follows by the non-archimedean property

that |ci |p < 1. Finally, observe that the constant term is c0 = (−1)d π1 · π2 · · · · · πd .
1
Then, vp (c0 ) = vp (π1 ) + vp (π2 ) + · · · + vp (πd ) = d( ). But since ef = n = [K :
e
unram unram unram
Qp ] = [K : Kf ] · [Kf : Qp ] = [K : Kf ] · f ⇒ e = [K : Kfunram ], it fol-
lows that d ≤ e. c0 ∈ Kfunram , vp (c0 ) is an integer, and we conclude that
Then since

1
d = e with vp (c0 ) = 1 ⇒ |c0 |p = . Therefore, E(x) is an Eisenstein polynomial and
p
unram
K = Kf (π).

By classifying the totally ramied and unramied extensions, these last two propo-

sitions do a good job of categorizing arbitrary nite extensions of Qp because it turns

out any extension is obtained by adjoining a combination of the two [3]. So, with a

basic description of nite extensions of Qp in hand, the reader can move on to describe
the algebraic closure of the eld of p-adic numbers. This eld is not complete, so an

additional step is necessary to nd a eld containing Qp that is both complete and

algebraically closed. Then, with such a eld, many options, including analysis of the

p-adic numbers, are open to the reader.

42
References

[1] C.C. MacDuee. The p-Adic Numbers of Hensel. The American Mathematical

Monthly, Vol. 45, No. 8 (Oct., 1938), pp. 500-508.

[2] F.Q. Gouvêa. p-adic Numbers: An Introduction. Springer-Verlag, New York,

Berlin, Heidelberg, rst edition, 1993.

[3] N. Koblitz. p-adic Numbers, p-adic Analysis, and Zeta-functions. Springer-Verlag,

New York, Berlin, Heidelberg, second edition, 1984.

[4] J. Howie. Fields and Galois Theory. Springer-Verlag, London, New York, rst

edition, 2005.

43