p-adic Analysis Compared with Real

Lecture 3
Claudio Hüni, Kathrin Naef, Daniel Schmitter

1 The ring of p-adic integers

Recall:

• Qp is the completion of Qp with respect to |·|p , where

(
p− ordp (x) , if x 6= 0,
|x|p :=
0, if x = 0
and
(
max{k ∈ N | pk divides x}, if x ∈ Z,
ordp (x) :=
ordp (a) − ordp (b), if x = ab , a, b ∈ Z, b 6= 0.

• For a ∈ Qp we have a canonical expansion

∞
X
a= ai pi ,
i=−m

where a−m 6= 0 and ai ∈ {0, 1, . . . , p − 1}. Therefore we get

( ∞
)
X
i
Qp = ai p .
i=−m m∈N

Definition (p-adic integers). a ∈ Qp is a p-adic integer if its canonical expansion

contains only non-negative powers of p. The set of p-adic integers is denoted by Zp .
Thus ( ) ∞
X
Zp = ai pi .
i=0

Remark. We have
Zp = B1 (0, Qp ) = {a ∈ Qp | |a|p ≤ 1}.
P
Indeed, let a ∞ i
i=−m ai p ∈ Qp . Then |a|p = p
− ordp (a) ≤ 1 ⇔ ord (a) ≥ 0. The above is
p
true if either ordp (a) = 0, i.e. a−m = a0 6= 0, or p|a, i.e. a−m = a0 = 0.
Theorem 1.1. Any infinite sequence of p-adic integers has a convergent subsequence.
Date: October 27, 2011

1
Proof. Let (xk )k∈N be a sequence in Zp . In the canonical expansion we have
∞
X
xk = . . . ak2 ak1 ak0 , i.e. xk = ai pi
i=0
for all k ∈ N. Note that ak0∈ {0, 1, . . . , p−1} =: I for all k ∈ N, i.e. there are only finitely
many possibilities for the value of ak0 . Thus one can find a b0 ∈ I and a subsequence of
(xk ), say (x0k ), such that for all elements in (x0k ) the first digit is b0 .
Analogously one can find b1 ∈ I and a subsequence (x1k ) of (x0k ) such that the first
two digits of each element in (x1k ) are b1 b0 . Inductively we obtain b0 , b1 , b2 , . . . and a
sequence of sequences as shown below.

x00 , x01 , x02 , ... ← (x0k )

x10 , x11 , x12 , ... ← (x1k )
x20 , x21 , x22 , ... ← (x2k )
.. .. .. ..
. . . .
By construction each sequence is a subsequence of the preceding one and each element
in the (j + 1)th row begins with bj . . . b1 b0 . Then the diagonal sequence x00 , x11 , . . . is
still a subsequence of the original one and it converges to . . . b3 b2 b1 b0 . 
Remark. The cardinality of Zp is |Zp | = 2|N| = p|N| . Indeed the map f : (Z/pZ)N → Z
given by
∞
X
f ((. . . , a2 , a1 , a0 )) := ai pi
i=0
is a bijection by the theorem which states that the canonical expansion of a p-adic
number is unique (cf. first report).
Examples. (i) The canonical expansion of (−1). As

0 = 1 + (p − 1) + p(p − 1) + p2 (p − 1) + . . .

we get
∞
X
−1 = (p − 1)pi
i=0
1
and the series converges since |p|p = p < 1. Also the p-adic norm of −1 is
|−1|p = 1.

(ii) The canonical expansion of (−3) in Zp for p 6= 2. We have

X
−3 = 2 − 5 = 2 + 5 · (−1) = 2 + (p − 1)pi
i≥1

and |−3|p = 1.
1
(iii) The canonical expansion of 2 for p = 5. As

1 1 α
=1− =1+
2 2 1−p

2
 we get
1 p−1
α = − (1 − p) = ∈ {0, . . . , p − 1},
2 2
since p − 1 < p is even. Thus, by the use of geometric series,
1 p − 1 X p + 1 X p − 1 i
=1+ · pi = + ·p ,
2 2 i≥0
2 i≥1
2
which is well-defined since p+1 p+1
2 ∈ {0, . . . , p − 1}, as p + 1 is even and 2 < p.
In this case one could also find the canonical expansion by the expansion of its
inverse 2: X X
2· (ai pi ) = 1 ⇔ (2 · ai pi ) = 1
i≥0 i≥0

and inductively construct the ai ’s.

2 Arithmetical operations in Qp respectively Zp

We will define operations in Qp . The operations in Zp are then given by the restriction
that the sum starts with 0 instead of −m.

2.1 Addition
P P
Let a = ∞ n
n=−m an p and b =
∞ n
n=−k bn p be elements in Qp such that a−m 6= 0 6= b−k .
W.l.o.g. assume m ≥ k. Then we define
∞
X ∞
X
a + b := (an + bn )pn = cn p n ,
n=−m n=−m
where cn = an + bn . Note that this is not the canonical form in general. For example
take an = p − 1 and bn = 1. Then it follows that an + bn = p which leads to a term of
the form ppn . Therefore a carry has to be taken for p + 1, i.e. cn+1 = an+1 + bn+1 + 1.
Example. Let a = 1 ∈ Qp . We are looking for b ∈ Qp such that a + b = 0. Since
0 = 0 + 0p + 0p2 + · · · and 1 = 1 + 0p + 0p2 + · · · we get by comparing coefficients
!
that 1 + b0 = 0. Because b0 ∈ {0, 1, . . . , p − 1} it follows that b0 = p − 1. Now it is
important that we take a carry of 1 to the left. Hence again by comparing coefficients
!
we get 1 + b1 = 0. Therefore b1 = p − 1. Comparing coefficients step by step leads to
P
bn = p − 1 for all n > 0. Hence b = ∞n=0 (p − 1)p .
n

2.2 Multiplication
P∞ n P∞ n
Let a = n=−m an p and b = n=−k bn p be elements in Qp . We define
∞
X
a · b := cn p n ,
n=−m−k

where c−m−k = a−m b−k , c−m−k+1 = a−m+1 b−k + a−m b−k+1 and in general c−m−k+l =
Pl
i=0 a−m+l−i b−k+i which is called the Cauchy product formula. Note that this does
not lead to the canonical form in general. The reason are again the carries which we
have to take to the left.

3
Note. We easily see that Zp is closed under addition and multiplication. Therefore
it follows that Zp is a commutative ring. Since Zp is embedded in Qp by definition it
contains no non-zero zero-divisors, i.e. it is an integral domain.
These definitions of addition and multiplication coincide with the addition and mul-
tiplication of the Cauchy sequences seen in the first presentation. Further the algorithms
for addition, subtraction, multiplication and division known for the reals can be used
for Qp if we proceed indefinitely from right to left. The example below shows the long
division algorithm for the division of . . . 421 by . . . 153 in Q7 .

. . . 615
. . . 153 | . . . 421
. . . 161
. . . 230
. . . 153
. . . 400
...4
......

Remark. p ∈ Zp has no multiplicative inverse in Zp .

Proof. Let a = p = 0 + 1p + 0p2 + · · · ∈ Zp . Suppose there exists a b ∈ Zp such that
ab = 1. We will show that this leads to a contradiction. Since we assume that b ∈ Zp it
P
has the form b = ∞ n
n=0 bn p . Hence we get
∞
X
ab = p bn p n
n=0
∞
X
= bn pn+1
n=0
= b0 p + b1 p 2 + · · ·
!
= 1 + 0p + 0p2 + · · · .

Now we get a contradiction comparing the constant coefficients, i.e. 0 = 1. Therefore p

has no multiplicative inverse in Zp . 
Theorem 2.1. a = . . . a2 a1 a0 ∈ Zp has a multiplicative inverse in Zp if and only if
a0 6= 0.
Proof. [⇒]: Suppose a0 = 0. Let b be the multiplicative inverse of a in Zp . Then
P !
ab = ∞ n
n=0 cn p with c0 = a0 b0 = 1. But since we assume that a0 = 0 it follows that
c0 = 0 which leads to a contradiction. Hence a0 must not be equal to 0.
P
[⇐]: Let b = ∞ n
n=−k bn p be the multiplicative inverse of a in Qp . Then ab =
P∞ ! !
n=−k cn p = 1. Comparing the coefficients of p−k we get c−k = a0 b−k = 0. Since
n

we assume a0 6= 0 it follows that b−k = 0. Comparing coefficients for p−k+1 (if k > 1)
!
leads to c−k+1 = a1 b−k + a0 b−k+1 = a0 b−k+1 = 0 such that it follows that b−k+1 = 0.
Comparing coefficients step by step we get that bn = 0 for all n < 0. Hence b ∈ Zp . 
This is completely different to the usual integers Z where only the inverses of 1 and
−1 are in Z. Nevertheless Zp is not a field.

4
Definition. Let us denote the group of invertible elements in Zp by Z×
p . Hence
( ∞
)
X
Z×
p = an p n
a0 6= 0 .
n=0

This group is called the group of p-adic units.

Remark. We have Z×
p = {a ∈ Zp | |a|p = 1}.

Proof. [⊆]: Let a ∈ Z× p . Then a0 6= 0. Since we are in Zp we get ordp (a) = 0. Therefore
by definition of the p-adic norm |a|p = 1 for all p.
[⊇]: Let a ∈ Zp such that |a|p = 1 for all p. Then ordp (a) = 0. Hence a0 6= 0. 
Theorem 2.2. The canonical p-adic expansion represents a rational number if and only
if it is eventually periodic to the left.
Proof. [⇐]: Assume that we have an eventually periodic x ∈ Qp . After multiplying,
if necessary, by a power of p we can assume x ∈ Zp . After subtracting an integer, if
necessary, and again multiplying by a power of p we can assume that x is of the form

x = x0 + x1 p + x2 p2 + · · · + xk−1 pk−1 + x0 pk + x1 pk+1 + · · · ,

as each of the previous processes only changes a finite number of coefficients. Then the
number a := x0 + x1 p + · · · + xk−1 pk−1 is a rational number. Thus
1
x = a · (1 + pk + p2k + · · · ) = a · ,
| P {z } 1 − pk
= i≥0
(pk )i

by geometric series and therefore we get that x is a rational number.

[⇒]: This part is pretty technical. A proof can be found in [Kat07]. 
Remark. It is easy to see that we have

• Zp ∩ Q = { ab ∈ Q | p does not divide b} and

• Z× a
p ∩ Q = { b ∈ Q | p does not divide ab}.

3 Algebraic definitions of Zp and Qp

In the following we give alternative definitions of Zp and Qp and prove their equivalence
to the previous ones.

3.1 Zp as a projective limit

P
Let x = i≥0 ai pi ∈ Zp . We can define its reduction modulo pn for n ∈ N+ , denoted
εn : Zp → Z /pn Z, by
X
εn (x) := ai pi mod pn ∈ Z /pn Z .
i<n
P
For xn := i<n ai pi mod pn
∈ Z /pn Z we will denote its equivalence class by [xn ]. By
the definition of addition and multiplication of p-adic integers we see that εn is a ring

5
 P
homomorphism for all n ∈ N+ . As xn = i<n ai pi −→ x as n → ∞, we would like to
say that the rings Z /pn Z converge to Zp as n → ∞.
Note that we have canonical homomorphisms ϕn : Z /pn+1 Z → Z /pn Z for all n ∈ N+
defined by " #! " #
X X
ϕn ai pi := ai pi .
i≤n i<n

Therefore we get commutative diagrams

Z /pn+1 Z

εn+1 ϕn

εn
Zp Z /pn Z

for all n ∈ N+ . We can interpret this as “Zp being closer to Z /pn+1 than to Z /pn Z”.
Definition (Projective limit of sets). A sequence (En , ϕn )n≥0 of sets and maps ϕn :
En+1 → En is called a projective system. A set E given together with maps ψn : E → En
such that ψn = ϕn ◦ ψn+1 for all n ≥ 0 is called a projective limit of the sequence
(En , ϕn )n≥0 if the following condition is satisfied: For each set X and maps fn : X → En
satisfying fn = ϕn ◦ fn+1 for all n ≥ 0 there is a unique factorization f of fn through
the set E, i.e. fn = ψn ◦ f : X → E → En for all n ≥ 0.
The maps ϕn : En+1 → En are called transition maps of the projective system. A
projective system can be represented in the form
ϕ0 ϕ1 ϕn−1 ϕn ϕn+1
E0 E1 ··· En En+1 ···.

“The” projective limit E, denoted by lim En , is placed at the end of the projective system
←−
(note that this is an abuse of notation as a projective limit is a set E together with the
maps ψn : E → En ). With the notation from above we get the following commuting
diagram:
ψn
ψn+1
ϕn−1 ϕn ϕn+1
··· En En+1 ··· lim En
←−

fn+1
∃!f
fn

X.

From the condition fn = ϕn ◦ fn+1 for all n ≥ 0 we get

fn = ϕn ◦ fn+1 = ϕn ◦ ϕn+1 ◦ fn+2 = (ϕn ◦ ϕn+1 ◦ · · · ◦ ϕn+k ) ◦ fn+k+1

6
for all k ≥ 0. Therefore f behaves as a limit of the fn as n → ∞ and ψn behaves as a
limit of composition of transition mappings ϕn ◦ ϕn+1 ◦ · · · ◦ ϕn+k as k → ∞.
Theorem 3.1. For every projective system (En , ϕn )n≥0 of sets there exists a projective
Q
limit (E, ψn ) = lim En ⊆ n≥0 En , where the maps ψn : E → En are given by (restriction
←−
of) projections. Further, if (E ′ , ψn′ ) is another projective limit of the same sequence, there
is a unique bijection f : E ′ → E such that ψn′ = ψn ◦ f for all n ≥ 0.
Proof. We first prove existence. Let
Y
E := {(xn )n≥0 | ϕn (xn+1 ) = xn for all n ≥ 0} ⊆ En .
n≥0

Hence the elements of E are coherent sequences with respect to the transition maps ϕn
in the product of the En . For x ∈ E we have by definition

ϕn (πn+1 (x)) = πn (x)

Q
for all n ≥ 0, where πn : n≥0 En → En is the canonical projection for all n ≥ 0.
Therefore the restrictions ψn of the projections πn to E satisfy

ϕn ◦ ψn+1 = ψn .

We see that (E, ψn ) is an upper bound of the sequence (En , ϕn )n≥0 . It is left to show that
(E, ψn ) has the required universal property. Consider any other pair (E ′ , ψn′ ) satisfying
ψn′ = ϕn ◦ ψn+1
′ for all n ≥ 0. We need to prove that there is a unique factorization of
′
ψn by ψn . By the universal property of the product with its projection maps πn we get
Q
that there exists a unique g : E ′ → n≥0 En such that
Q
n≥0 En

∃!g
πn

ψn′
E′ En

commutes for all n ≥ 0. It is easy to see that

Y
(ψn′ ) : E ′ → En
n≥0
y 7→ (ψn′ (y))

satisfies this condition, i.e. g = (ψn′ ). As ψn′ = ϕn (ψn+1

′ (y)) for all n ≥ 0 and all y ∈ E ′
we see that Im(g) ⊆ E. Therefore we can define f : E ′ → E by f (y) := g(y), i.e. by
restricting the codomain of g (this is well-defined by the remark above), finishing the
proof of existence.
Let us now show uniqueness. Suppose that (E, ψn ) and (E ′ , ψn′ ) have the universal
property. Then there exist unique maps f : E ′ → E with ψn′ = ψn ◦ f and f ′ : E → E ′

7
with ψn = ψn′ ◦ f ′ . By substitution we get

ψn′ = ψn ◦ f = ψn′ ◦ f ′ ◦ f.

Therefore we get that the following diagram commutes:

E′

f′ ◦ f
ψn′
idE ′

ψn′
E′ En .

By the universal property of (E ′ , ψn′ ) we get f ′ ◦ f = idE ′ . Similarly we have

ψn = ψn′ ◦ f ′ = ψn ◦ f ◦ f ′

and therefore f ◦ f ′ = idE by the universal property of (E, ψn ). Hence f is a bijection

as needed. 
Remark. This construction can easily be generalized to other objects than sets. For
example the same result holds for groups and rings if we replace sets by groups or rings
respectively, replace map by homomorphism and define the operations on the product
componentwise. Similarly the statement also holds for topological spaces if we consider
continuous maps and put the subspace topology on E induced by the product topology.
One can also combine these two and consider projective limits of topological groups and
rings.
However projective limits do not need to be of the same kind as the En . For example
the projective limit of fields is only a ring in general and not a field. Another example
is given by the projective limit of finite abelian groups which need not be finite.
Remark. Let (En , ϕn )n≥0 be a projective system of topological spaces and continuous
maps. If the topological spaces En are Hausdorff spaces, the subspace (E, ψn ) := lim En
Q ←−
is closed in n≥0 En .
Q Q
Proof. We will prove that n≥0 En \ E is open. Therefore let x = (xn ) ∈ n≥0 En \ E.
Then there exists i ≥ 0 such that ϕi (xi+1 ) 6= xi . As Ei is Hausdorff we can find disjoint
open neighborhoods U of ϕi (xi+1 ) and V of xi . Then
( )
Y
−1
πi+1 (ϕ−1
i (U )) ∩ πi−1 (V )= x∈ En xi+1 ∈ ϕ−1
i (U ), xi ∈ V
n≥0
Q
is an open neighborhood of x not intersecting E. Hence n≥0 En \ E is open and
therefore we have that E is closed by definition. 
Theorem 3.2. The map ρ : Zp → lim Z /pn Z that associates to the p-adic integer
P ←− P
x = i≥0 ai pi the sequence (xn )n≥1 of its partial sums xn = i<n ai pi mod pn is a ring
isomorphism.

8
 Q
Proof. The coherent sequences in the product n≥1 Z /pn Z are the sequences (xn ) of
P
partial sums of a formal series i≥0 ai pi with 0 ≤ ai ≤ p − 1 for all i ≥ 0. These are
precisely the p-adic integers as every x ∈ Zp has a unique representation in canonical
Q
form (cf. first report). As the operations in n≥1 Z /pn Z are defined componentwise,
we see that ρ(x) = (εn (x)) for all x ∈ Zp . Therefore ρ is a homomorphism. It is left to
show that ρ is a bijection. ρ is injective as we have
( )
X
−1 −1 i
ker(ρ) = ρ (0) = ρ (([0], [0], . . .)) = ai p ∈ Zp ai = 0 for all i ≥ 0 = {0}
i≥0

by the uniqueness of the canonical representation. Further it is easy to see that ρ is

Q
surjective, as if ([x1 ], [x2 ], . . .) ∈ n≥1 Z /pn Z is a coherent sequence, we can choose
yi ∈ [xi ] such that 0 ≤ yi < pi for all i ≥ 1. We can then define b0 := y1 and
P
bi := yi+1pi−yi for all i ≥ 1. Therefore ρ( i≥0 bi pi ) = ([y1 ], [y2 ], . . .) = ([x1 ], [x2 ], . . .), i.e.
ρ is surjective. 
Remark. We can consider Z /pn Z together with the discrete topology as a topological
ring for all n ≥ 1. As Z /pn Z is finite for all n ≥ 1 we see that these spaces are all
Q
compact. Hence by Tychonoff’s theorem n≥1 Z /pn Z is compact. Therefore Zp =
Q
lim Z /pn Z ⊆ n≥1 Z /pn Z is also compact as a closed subspace of a compact space.
←−
We also see that Zp is totally disconnected.
Further as the map ρ in the theorem above is a continuous map between two compact
spaces it is a homeomorphism, i.e. it is an isomorphism of topological rings.

3.2 Qp as a fraction field

As seen in the previous section, we have that Zp is an integral domain. Therefore the
following definition is well-defined.
e := Frac(Zp ) be the fraction field of Zp .
Definition. Let Q p
e is isomorphic to Q .
Theorem 3.3. Q p p

Proof. It is known that the field of fractions of an integral domain R satisfies the
following universal property: For every field F and any injective ring homomorphism
f : R → F there exists a unique ring homomorphism g : Frac(R) → F that extends f ,
i.e. such that the diagram

Frac(R)

i ∃!g

f
R F

commutes, where i denotes the embedding of R into Frac(R), i.e. the map sending x ∈ R
to the equivalence class of (x, 1) for all x ∈ R. We will now show that Qp satisfies this
universal property for Zp , i.e. for all fields F and injective maps f : Zp → F there exists
a unique ring homomorphism g : Qp → F such that the diagram

9
 Qp

i ∃!g

f
Zp F

commutes, where i : Zp → Qp is defined by i(x) := x for all x ∈ Zp . By the uniqueness

e ∼
(up to isomorphism) of the field of fractions it will then follow that Q p = Qp holds.
If g : Qp → F extends f , we need f (x) = g(i(x)) = g(x) for all x ∈ Zp . Further let
yk := f (pk ) = g(i(pk )) ∈ F for all k ∈ N+ . As g is a ring homomorphism we get

yk−1 = g(i(pk ))−1 = g(i(pk )−1 ) = g(p−k )

for all k ∈ N+ . As Qp is generated by Zp ∪{p−k }k>0 (cf. first report) and g(z) is
uniquely determined by f for all z ∈ Zp ∪{p−k }k>0 , the ring homomorphism g : Qp → F
is uniquely determined by f . 

References
[Kat07] Svetlana Katok: p-adic Analysis Compared with Real. American Mathe-
matical Society, 2007.

[Rob00] Alain Robert: A course in p-adic analysis. Springer, 2000.

10