You are on page 1of 9

P -ADIC ANALYSIS COMPARED WITH REAL

LECTURE 2
NORMS ON RATIONAL NUMBERS

IVO ASCHWANDEN, ADRIAN CLOUGH,


AND ALEXANDROS GROSDOS KOUTSOUMPELIAS

1. Equivalence of Norms/Metrics (Ivo Aschwanden)


1.1. Remarks about this section. Whenever we write “Kelley” [1] we refere to
the splendid book “General Topology” written by John L. Kelley which is now
available at the Springer Verlag in the GTM series. A net is an object which gen-
eralizes the concept of a sequence; it is the same concept as a filter. A complete
discussion of this topic can be found in Kelley in Chapter “Moore-Smith conver-
gence”; it is sometimes convenient to define a topology by specifying which nets
converge to which points. Natural questions concerning this topic are answered in
the chapter on convergence in Kelley’s book. Any proofs given of stated properties
will be provided at the end of this chapter since we do not want to disturb the
reading flow. We will follow Halmos’ fashion and write “iff” for “if and only if”.

1.2. Discussion of metric toplogies. (pseudo) Metric spaces are first countable,
for given x in X one chooses for the neighborhood base the balls with rational
radii. Pseudo metric spaces are metric ones except that from d(x,y)=0 it must
not necessarily follow that x=y. Pseudo metric spaces are metric ones iff they are
Hausdorff. The first countability result in the fact that basic topological properties
like closure, accumulation point, compactness, and continuity (domain space first
countable, range space arbitrary topological) can be expressed in terms of sequences
instead of nets see [Kelley]:

• cl(A) = {x ∈ X | ∃ sequences (xn ) converging to x}, Lemma 1.1


• cl(A) = {x ∈ X | d(x, A) = 0}; d(x, A) := inf{d(x, a) | a ∈ A}, Lemma 1.1
• x is an accumulation point of A iff there is sequence in A − {x} which
converges to A, Lemma 1.1
• compactness ⇐⇒ sequential compactness
• continuity ⇐⇒ sequential continuity
• two first countable topologies are equal iff they have the same converging
sequences and the corresponding limit points coincide, Lemma 1.2

1.3. Equivalences of metrics. There are various different yet not equivalent con-
ditions on two metrics being equivalent. It depends on what one is interested in.
The best known are the following:
Two metrics are defined to be equivalent iff:

Date: October 18, 2011.


1
2IVO ASCHWANDEN, ADRIAN CLOUGH, AND ALEXANDROS GROSDOS KOUTSOUMPELIAS

• (X, d1 ) and (X, d2 ) are topologically isomorphic (they induce the same
topologies) or
• (X, d1 ) and (X, d2 ) are uniformly isomorphic (for example if the identity
map is d1 − d2 and d2 − d1 continuous) or
• x is an accumulation point of A iff there is sequence in A which converges
to A.
The most commonly known notion is the first one mentioned. We will also use
this definition and will write d1 ∼ d2 . If they are “Cauchy” equivalent, i.e sequences
are Cauchy in d1 iff it is Cauchy in d2 , we will write d1 ∼c d2 . Katok uses Cauchy
equivalence and Schikhof uses topological equivalence. For our purposes this is no
problem since soon enough we will deal only with complete normed fields and the
two notions will then coincide, Lemma 1.3. Two (pseudo) norms are equivalent iff
the induced (pseudo) metrics are. Two metrics are equivalent iff the balls nest i.e
each d1 -ball contains a d2 -ball and vice versa. Now we turn to normed fields.
1.4. Criteria for the equivalence of field norms. What a norm on a field is,
was defined earlier; our last purpose is to give various equivalent conditions of field
norms being equivalent. We have one norm which steps out of line: The trivial
one which is zero only on 0 and one everywhere else. This is not a very interesting
norm, even topologically. The trivial norm induces the discrete topology and a
norm which induces the discrete topology must be the discrete one. Therefore two
equivalent norms are either both discrete or none of them is, Lemma 1.4. This fact
justifies why most authors assume that their norms are not trivial. The non trivial-
ity assures that there are always numbers, say f and f 0 in F , for which 0 < |f | < 1
and 1 < |f 0 | (choose f 0 = f −1 ) . There is another important fact. The norm of a
number is less than 1 iff its powers converge to zero. This is the point where two
equivalent norms are connected over a non topological statement. This will then
also be the main idea of the proof of our target Theorem 1.1 that two norms are
equivalent iff one is a positive power of the other. We found some more equiva-
lent statements for two field norms being equivalent, these are more technical and
therefore in most cases easier to work with:

The following statements about two field norms | |1 and | |2 are equivalent:
• The induced topologies coincide
• One is a positive power of the other
• From |f |1 < 1 follows |f |2 < 1
• From |f |1 ≤ 1 follows |f |2 ≤ 1
• The following statements are true
◦ |f |1 < 1 iff |f |2 < 1
◦ |f |1 ≤ 1 iff |f |2 ≤ 1
◦ |f |1 = 1 iff |f |2 = 1
1.5. Proofs of the precised statements.
Lemma 1.1. For first countable spaces, x in X is an accumulation point of a
subset A iff there is a sequence in A − {x} converging to x, and cl(A) = {x ∈ X |
∃ sequences (xn ) converging to x}
Proof. If we have a sequence in A converging to x then by the definition of con-
vergence x must be an accumulation point. Let therefore x be an accumulation
P -ADIC ANALYSIS COMPARED WITH REALLECTURE 2NORMS ON RATIONAL NUMBERS3

point of A and choose


S a countable base for the neighbourhood system of x, say Un
then then Vn := k≤n Un is also a base for the neighbourhood system of x which
is furthermore decreasing. Choose now for each n ∈ N an xn in Vn ∩ {A − {x}}.
This is possible by definition of an accumulation point. But xn clearly converges
to x and we are done. The second statement follows from the fact that cl(A) is
the union of A with its accumulation points and that a constant sequence always
converges. 

Lemma 1.2. Two first countable topologies are equal iff they produce the same
convergence classes of sequences i.e the same set of pairs (xn , x) where xn is a
convergent sequence and x is a limit point of xn
Proof. By 1.1 we get via the converging sequences a closure operator which is indeed
one since they are derived from topologies. These two operators must coincide by
assumption hence we get that the closed sets, and hence the topologies coincide. 

Lemma 1.3. For complete normed fields (F, | |1 ), (F, | |2 ) it holds: | |1 ∼


| |2 ⇐⇒ | |1 ∼c | |2
Proof. “ =⇒ ” Since every convergent sequence is Cauchy and every Cauchy se-
quence is convergent by completeness this direction is clear.
“⇐=” Suppose that | |1 ∼c | |2 and let xn −→ x in | |1 , then it is Cauchy in | |1 ,
therefore Cauchy in | |2 and hence convergent in | |2 say to x2 . =⇒ xn − x1 and
xn − x2 are both Cauchy in | |2 =⇒ (xn − x1 ) − (xn − x2 ) is Cauchy in | |2 , but
being constant x2 − x1 and Cauchy it follows that x2 = x1 and 1.2 applies. 

Lemma 1.4. If | |1 ∼ | |2 then


(i) | |1 trivial ⇐⇒ | |2 trivial
(ii) the trivial norm induces the discrete topology
(iii) a norm which induces the discrete topology is discrete
Proof. (i) Let | |1 be trivial and suppose | |2 is not, that is ∃a ∈ F : 0 <
|a|2 < 1; =⇒ an −→ 0 in both topologies by equivalence of the norms, but
a 6= 0 and we have a contradiction.
(ii) Let a ∈ F then B 21 (a) = {a} and we have that every set is open, since each
set is the union of its singletons.
(iii) a norm which induces the discrete topology is equivalent to the trivial norm
by (ii) and hence (i) applies


Theorem 1.1. | |1 ∼ | |2 ⇐⇒ |s|1 = |s|c2 for some c > 0 ∀s ∈ F


Proof. “⇐=” This is clear since the balls nest
“ =⇒ ” |s|1 = |s|c2 is equivalent with ln |s|1 = c · ln |s|2 for s 6= 0. The case s = 0 is
trival, therefore let s 6= 0 and | |1 and | |2 be nontrivial (the trivial case is obvious)
a(s)
and choose k ∈ F : 0 < |k|1 < 1, then there is a a(s) ∈ R+ such that|s|1 = |k|1 .
a(s)
We will show later that |s|2 = |k|2 must also hold. Now we assume that both
equalities are true and conclude:
ln |s|1 = a(s) ln |k|1
ln |s|2 = a(s) ln |k|2
4IVO ASCHWANDEN, ADRIAN CLOUGH, AND ALEXANDROS GROSDOS KOUTSOUMPELIAS

ln |s|1 |k|1
=⇒ =
ln |s|2 |k|2
|k|1
=⇒ ln |s|1 = · ln |s|2
|k|2
and the proof will be complete when we verify the assumed equality. This we
will do by showing for every two rationals r, r0 ∈ Q : r0 < a(s) < r we get that
0 a(s)
|k|r2 < |s|2 < |k|r2 from which it clearly follows that |s|2 = |k|2 (prove this by
contradiction using the continuity of x 7→ ax ), therefore first let a(s) < p · q −1 and
−1
a(s)
then |s|1 = |k|1 < |k|p·q1 .
=⇒ |s|q1 < |k|p1 =⇒ |sq k −p |1 < 1 and here we can use the connection of | |1 and
−1
| |2 to conclude |sq k −p |1 < 1 =⇒ |s|2 < |k|p·q . The other inequality is proved
analogously. 
Corollary 1.1. Let F, | |1 , | |2 be a given field and nontrivial field norms. Then
the following statements are equivalent.
(i) | |1 ∼ | |2
(ii) | |1 = | |c2 for some c > 0
(iii) |x|1 < 1 =⇒ |x|2 < 1
(iv) |x|1 ≤ 1 =⇒ |x|2 ≤ 1
(v) • |x|1 < 1 ⇐⇒ |x|2 < 1
• |x|1 ≤ 1 ⇐⇒ |x|2 ≤ 1
• |x|1 = 1 ⇐⇒ |x|2 = 1
Proof. “(i) ⇐⇒ (ii)”: is the prooved theorem.
“(ii) =⇒ (iii)”: is clear
“(iii) =⇒ (iv)”: if |x|1 < 1 then by assumption |x|2 < 1. Let therefore |x|1 = 1
and choose by nontrivialility of the noms a ∈ F : |a|1 < 1 =⇒ |a|2 < 1 =⇒
−1
|k n a|1 < 1 ∀n ∈ N =⇒ |x|2 < |a|2 n ∀n ∈ N =⇒ |x|2 ≤ 1 (by taking the
limit). The same argument applied to x−1 yields |x|2 ≥ 1 “(iv) =⇒ (v)”: choose
a ∈ F : 0 < |a|2 < 1. We will just show that (iv) impies (iii) from which all points
in (v) easily follow. If therefore
xn |x|1 < 1 then there is a n ∈ N such that |x|n1 ≤ |a|1
which is not zero =⇒ a 1 ≤ 1 =⇒ |x|n2 ≤ |a|2 < 1 =⇒ |x|2 < 1.

“(v) =⇒ (iii)” this is really trivial
“(iii) =⇒ (ii)” in the proof of the theorem we just used the property (iii) to
construct such a c > 0, therefore this is essentially the same proof. 

2. All Norms on Q (Adrian Clough)


2.1. Introduction. In this section we will classify all norms on Q (up to equiva-
lence). First we will prove Ostrowski’s theorem 2.1 which states that all non-trivial
norms are equivalent either to the absolute value or some p-adic norm. By the
previous discussion we will quickly be able to conclude that these norms are dis-
tinct, that is non-equivalent, so every norm on Q is either trivial, Archimedean, or
equivalent to a specific p-adic norm. The trivial norm is systematically ignored as
it is clear from the preceding discussion that the trivial norm is called such for a
good reason and it is trivially equivalent only to itself. Before we turn to the proof
of Ostrowski’s theorem 2.1, which constitutes the main part of this section we will
briefly review some material relevant to the proof, which will provide some of the
ideas needed for the proof and hopefully make the proof more straight forward.
P -ADIC ANALYSIS COMPARED WITH REALLECTURE 2NORMS ON RATIONAL NUMBERS5

2.2. Preliminary Considerations. We recall that two field norms | |1 , | |2 are


equivalent when there exists some c > 0 such that | |1 = | |c2 . A ubiquitous strat-
egy encountered in numerous areas of mathematics when trying to prove a certain
property for all elements of a set is to first prove the property for some set of gener-
ators with which the correctness for all other elements may be shown. In our case
this phenomenon may be broken into several steps: ”some special element(s)” −→
N −→ Z −→ Q because for any n ∈ N we have | − n| = |n| andfor any c rational
c
|p|1 |p|c2 |p|2
number written as q p, q ∈ Z we have q = |q|1 = |q|c = |q|2 = pq . In the
p p
1 2 2
preceding discussion we showed that the information about a field norm is encoded
by which elements are sent to values greater or smaller than 1, so let us examine
this a bit more closely. Summarising the equivalent conditions for a norm being
Archimedean from the last lecture we see that the two main properties held by such
norms are unboundedness of the integers and not being an ultrametric. In fact the
boundedness of non-Archimedean norms may be expressed in increasingly precise
terms:

Archimedean non-Archimedean
Z is unbounded Z is bounded
∀x, y ∈ Q, x 6= 0 : ∃n ∈ N : |nx| > |y|
∀n ∈ Z : |n| ≤ 1
|2| > 1 |2| ≤ 1
| | is non-ultrametric | | is ultrametric

We have included a new property which we state as


Proposition 2.1. For any field norm on Q : |2| ≤ 1 iff for all n in N : |n| ≤ 1
Proof. The first statement is a special instance of the latter so we only have to
show that when the statement is true for 2 it is true for all n in N.
So let n ∈ N, n ≥ 2; writing it as a binary expansion we have:

n = a0 + a1 2 + · · · + as 2s
with a0 . . . as ∈ 0, 1 and as = 1. Then, since |2| ≤ 1,
s
X s
X
|n| ≤ |ai ||2i | ≤ |2|i ≤ s + 1
i=0 i=0
Now we may do the same for the k-th power of n with k ∈ N. We have n < 2s+1 ,
thus nk < 2k(s+1) so that for the binary representation of nk :

nk = b0 + b1 2 + · · · + bt 2t
with b0 · · · bt ∈ 0, 1, bt = 1 and t < (s + 1)k we have

|nk | ≤ t + 1 ≤ (s + 1)k
and thus
p
k
|n| ≤ lim (s + 1)k = 1
k→∞

6IVO ASCHWANDEN, ADRIAN CLOUGH, AND ALEXANDROS GROSDOS KOUTSOUMPELIAS

We just witnessed a first instance of the strategy we want to employ to prove


Ostrowski’s theorem 2.1 and also that the power behind the n0 -ary representation of
a given positive integer, where n0 is the smallest positive integer such that |n0 | > 1,
comes from how effectively we may relate the given positive integer to its order of
magnitude (with respect to n0 ).
For p-adic norms the situation could be described even more vividly however, it
will suffice to know that for a given p-adic norm | |p , p prime, for any n ∈ N we
have |n|p < 1 when p divides n and |n|p = 1 otherwise.

2.3. Ostrowski’s Theorem. Now we are ready to tackle the main theorem of this
section:
Theorem 2.1. Each non-trivial norm on the field of the rational numbers is equiv-
alent either to the absolute value function or to some p-adic norm.
In fact, we will prove a slightly more precise result:
Proposition 2.2. Let | | be a nontrivial norm on Q,
(i) if | | is Archimedean, it is equivalent to the absolute value function,
(ii) if | | is non-Archimedean it is equivalent to some p-adic norm.
Proof. (i): Let | | be an Archimedean norm on Q, we want to show that there
exists a c > 0 such that for all n in N, |n| = nc . For this it is sufficient to find
M, M 0 ∈ R, M ≥ 1, 0 < M 0 ≤ 1 :
(1) |n| ≤ nc M

(2) |n| ≥ nc M 0
As this is true for all natural numbers it will be true for nk , with k some positive
integer. So we get
|nk | ≤ nkc M
so that √
|n| ≤ lim nc M = nc
k

k→∞
The other inequality is completely analogous.
To determine c > 0, M and M 0 we first consider the smallest n0 ∈ N : |n0 | > 1,
which we know to be 2, we set c > 0 such that |2| = 2c and we write some n ∈ N
using its binary representation:
n = a0 + a1 2 + · · · + as 2s (a0 , a1 , · · · ∈ {0, 1}, as = 1)
Then the order of magnitude of n is 2s ≤ n < 2s+1 so that
(3) 2sc ≤ nc < 2c(s+1)
To find M we apply (3) and get
s
X s
X
|n| ≤ |ai ||2|i ≤ 2ic = 2sc (1 + 2−c + · · · + 2−sc ) ≤ nc M
i=0 i=1
P∞ −ic
setting M := i=0 2 .
To find M 0 :
|n| = |2s+1 − (2s+1 − n)| ≥ |2s+1 | − |(2s+1 − n)| = 2c(s+1) − |(2s+1 − n)|
P -ADIC ANALYSIS COMPARED WITH REALLECTURE 2NORMS ON RATIONAL NUMBERS7

Applying (1) and (3)


|2s+1 − n| ≤ (2s+1 − n)c ≤ (2s+1 − 2s )c = 2sc
so that

|n| ≥ sc(s+1) − 2cs = 2c(s+1) (1 − 2−c )


Again by (3), 2c(s+1) > nc and by setting M 0 := 1 − 2−c we obtain
|n| ≥ nc M 0

(ii) Let | | be an Archimedean norm on Q, again we want to find a c > 0 such that
for all n in N, |n| = nc . This time it is sufficient to find some prime number p, such
that whenever a natural number is divisible by p it’s norm is strictly smaller than
1 and equal to 1 otherwise, because then for any n ∈ N written as n = pk m with
p - m we have |n| = |pk m| = |p|k and we set c = − log |p|(log p)−1 .
From our previous considerations it is clear that our p would be the smallest natural
number which norm is strictly smaller than one. To prove this set p = min{n ∈ N |
|n| < 1} and assume p is not a prime, then ∃a, b ∈ N, a, b 6= 1 : p = ab and due to
the compatibility of multiplication on N with its order we have a, b < p, but then
1 > |p| = |a||b| = 1 · 1 = 1. Now all that is left is to show that any natural number
not divisible by p has norm 1. Now let n be such a number and write n = ap + r
where a ∈ N and 1 ≤ r < p. Then |r| = 1 and |ap| = |a||p| ≤ |p| < 1. By the
strong triangle inequality 1 = |r| ≤ max(|ap + r|, | − ap|) = max(|n|, |ap|) = |n|. So
|n| ≥ 1, i.e. |n| = 1. 

2.4. Conclusion. Now all that is left to verify, is that the absolute value function
is not equivalent to any p-adic norm and that for different prime numbers p, q,
| |p is not equivalent to | |q . But this immediately follows from discussion in the
previous section as norms are only equivalent when they send the same numbers
in to the same subsets in the following partition of R = {0} ∪ (0, 1) ∪ {1} ∪ (1, ∞),
which is never the case.
Thus we have shown that that there is a one-to-one correspondence between the
prime numbers together with two additional elements and all the field norms on Q.

3. Applications and the Product Formula (Alexandros Grosdos


Koutsoumpelias)
Proposition 3.1 (Product Formula). For every x ∈ Q \ 0
Y
|x|∞ |x|p = 1
p prime

Proof. We can write any positive integer n in the form n = pa1 1 · . . . · pakk where the
pi are disntict prime numbers.
Q If for some prime number q, q 6= pi then |x|q = 1,
thus the infinite product p prime |x|p contains only finitely many terms different
than one. Multiplying |x|∞ and |x|pi = p−ai
i
we obtain the desired result. By the
multiplicative property of the norm, the product formula holds for any non-zero
rational number. 
1
Corollary 3.1. For any prime number p and any n ∈ N, |n|p ≥ n.
8IVO ASCHWANDEN, ADRIAN CLOUGH, AND ALEXANDROS GROSDOS KOUTSOUMPELIAS

Proof. Combining the product formula and the fact that for any prime number q,
|n|q ≤ 1 we get
Y Y
|n|∞ |n|q = 1 ⇒ n · |n|p |n|q = 1 ⇒ n · |n|p ≥ 1
q prime q prime
q6=p

and the result follows. 

The importance of the product formula lies in the fact that it establishes a close
relationship between the norms on Q. If we know the values of all norms but one
we can easily recover the missing value. This property can often be very useful in
algebraic geometry and number theory.
Consider a polynomial P (x) with rational coefficients. If P (x) has a rational
solution then it obviously has a solution in Qp for all prime numbers p and it
also has a real solution. One naturally wonders wether the inverse statements
is also true. The answer, in general, turns out to be no. However, there are
collections of polynomial equations which satisfy the Hasse Principle (or Local-to-
Global Principle), that is the equation has a solution in mathbbQ if and only if it
has a solution in Qp for all prime numbers p and a real solution. A very important
result is the Hasse-Minkowski theorem.
Theorem 3.1 (Hasse-Minkowski). A quadratic form with rational coefficients rep-
resents zero in the field of rational numbers if and only if it represents zero in the
field of real numbers and in all fields of p-adic numbers.
Later during this seminar we will prove the easier statement that a number x ∈ Q
is a square if and only if it is a square in R and in Qp for all primes p.
We will now prove two remarkable as well as counter-intuitive results concerning
norms on Q using the equivalence of norms and Ostrowski’s theorem.
Proposition 3.2. If | | and | |0 are inequivalent norms on Q, then for each pair of
rational numbers s and t there exists a sequence a1 , a2 , . . . of rational numbers such
that limn→∞ |s − an | = 0 and limn→∞ |t − an |0 = 0
pn
Proof. We will first consider the sequence pn +q n where p and q are two prime
numbers with p < q. Then:
pn
• limn→∞ pn +q n = 0 in (Q, | |∞ )
n
pn
• | pnp+qn |p ≤ p−n ||qn |p −|p
1
n| | = p
p
−n 1
1−p−n = 1
pn −1 , thus limn→∞ pn +q n =0
in (Q, | |p )
n n
pn
• | pnp+qn − 1|q = | pnq+qn |q ≤ q −n ||qn |q −|p
1
n| | =
q
1
q n −1 , thus limn→∞ pn +q n =1
in (Q, | |q )
Without loss of generality we assume that | | is equivalent to | |∞ or | |p for some
prime number p and | |0 is equivalent to | |q for some prime number q. By theorem
1.1 , limn→∞ |s − an | = 0 if and only if limn→∞ |s − an |p = 0 (or limn→∞ |s − an | = 0
if and only if limn→∞ |s − an |∞ = 0) and similarly limn→∞ |t − an |0 = 0 if and only
if limn→∞ |t − an |q = 0, therefore it suffices to find a sequence of rational numbers
an , such that limn→∞ |s − an |p = 0 = limn→∞ |t − an |q .
We are now ready to define our sequence an . We will consider three different
cases:
P -ADIC ANALYSIS COMPARED WITH REALLECTURE 2NORMS ON RATIONAL NUMBERS9

• Case 1: p and q are primes with p < q


n n
Define an = t pnp+qn + s pnq+qn . Then
qn pn pn
s − an = s(1 − pn +q n ) + t pn +q n = (s − t) pn +q n −
−−−→ 0 in (Q, | |p ) and
n→∞
qn
t − an = (t − s) pn +qn −−−−→ 0 in (Q, | |q ).
n→∞
• Case 2: p = ∞ and q > 2
n n
Define an = t 2n2+qn + s 2nq+qn . In a similar way,
n
s − an = (s − t) 2n2+qn −−−−→ 0 in (Q, | |∞ ) and
n→∞
n
t − an = (t − s) 2nq+qn −−−−→ 0 in (Q, | |q ).
n→∞
• Case 3: p = ∞ and q = 2
1 qn
Define an = t 1+q n + s 1+q n . The result is similar to that of the previous

case.

Proposition 3.3. For every prime number p, there exists a sequence a1 , a2 , . . . of
rational numbers such that limn→∞ |an |p = 1 and limn→∞ an = 0 in (Q, | |∞ ) and
in (Q, | |q ) for all prime numbers q different than p.
Proof. We define Lk to be the product of the first k prime numbers, for example
L1 = 2, L2 = 2 · 3, L5 = 2 · 3 · 5 · 7 · 11. For any prime number p and and for
sufficiently big k (specifically take k ≥ i, where i is the unique natural number such
that p is the i-th prime number):
|Lk
k |p p−k
|Lkk L−1
k2 |p = |Lk2 |p = p−1 −−−−→ 0
k→∞
One can also easily show that Lkk L−1
k2 <
1
2k−1
−−−−→ 0. Thus limn→∞ Lkk L−1
k2 = 0
k→∞
with respect to every norm on Q.
Fix a prime number p. From the multiplicative property of the norm one can
easily conclude that the sequence ak = p1−k Lkk L−1
k2 has the desired property. 

References
[1] John L. Kelley, General Topology, D. Van Nostrand, Pinceton, 1955.
[2] Svetlana Katok, p-adic Analysis Compared with the Real, American Mathematical Society,
2007.
[3] W.H.Schikhof, Ultrametric calculus, Cambridge University Press, Cambridge, 1984.
[4] Z. I. Borevich, I. R. Shafarevich, Number Theory, Academic Press, New York, 1966

You might also like