In this paper, we explore the p-adic system, by defining it in
multiple ways: as an extension of the p-adic integers, as well as an
extension of the rationals. We then proceed to perform analysis in the
p-adics, by defining convergence, continuity and discs. We then
describe exponentiation and logarithmic functions over the p-adics
as functions derived from power series. We explore the radius of
convergence and other properties of these functions. We then explore
the Artin-Hasse exponential, which, though seemingly random, turns out
to be an integral power series.
Prove that R[[x]] is a ring under the natural operations of addition
and multiplication.
Addition : Let f,g∈R[[x]] such that
f(x)=∑i=0∞aixi and
g(x)=∑j=0∞bjxj. Then
f(x)+g(x)=∑i=0∞aixi+∑j=0∞bjxj=∑i,j=0∞(ai+bj)xi.
Since ai+bj is closed under addition as aiand bj∈R we
can say, R[[x]] is a ring under addition.
Multiplication : For proving the multiplication, we need to prove
the commutative property in a formal power series. Let f,g∈R[[x]]
such that f(x)=∑i=0∞aixi and
g(x)=∑j=0∞bjxj. Then
f(x)g(x)=i=0∑∞aixi×j=0∑∞bjxj=k=0∑∞(i=0∑kaibk−i)xk
Since the sum ∑i=0kaibk−i is a sum of products of elements
of R, the sum itself is in R since R is closed under addition and
multiplication. Therefore, the above sum is a power series in R, that
is the products fg∈R[[x]].
Now, since the ring is commutative over both + and ⋅, it follows
that R[[x]] is also commutative. Moreover, it is clearly associative
over + since the ring R is itself associative over +. Next, the
ring has a zero element, i.e. 0∈R[[x]] since for any
f∈R[[x]], we have f+0=f. Similarly, it also has an identity
element, i.e. 1. Also, the additive inverse of f is the power series
with the coefficients each being the additive inverse of the
coefficients of f. Thus, every element of R[[x]] has an inverse
element. Distributivity can also be easily seen from the fact that R
is itself distributive.
We will now prove associativity over multiplication. Let
a=i=0∑∞aixi,b=i=0∑∞bixi,c=i=0∑∞cixi
Now,
ab=k=0∑∞(i=0∑kaibk−i)xk=i=0∑∞pixi
So,
(ab)c=k=0∑∞(i=0∑kpick−i)xk=k=0∑∞(i=0∑k(j=0∑iajbi−j)ck−i)xk
The above can be expanded as
k=0∑∞(c0a0b0+c1(a0b1+a1b0)+⋯ck(a0bk+⋯akb0)xk
Now, a(bc)=(bc)a by commutativity. Therefore, we may use a similar
approach as above to show that
a(bc)=k=0∑∞(a0b0c0+a1(b0c1+b1c0)+⋯+ak(b0ck+⋯bkc0))xk
By rearranging the terms above, we get
a(bc)=k=0∑∞(c0a0b0+c0(a0b1+a1b0)+⋯+ck(a0bk+⋯akb0))xk=(ab)c
This proves associativity over multiplication.
Definition 2. Units in R[[x]] are all f and g in R[[x]]
such that [f⋅g](x)=1.
Examples of these units are:
f(x)=−(x−1)
f(x)=∑i=0∞xi
f(x)=1+∑i=1naixi for ai∈R
f(x)=(±(x−1)⋅∑i=0∞xi)n for
n∈N
Proposition 1. If f(x)∈R[[x]] and f(0)=1, then f(x) is
a unit in R[[x]].
Proof. Let f(x),g(x) be in R[[x]] such that f⋅g=1. We
need to prove that such a power series, g, exists if and only if we have
f(0)=1. Let f(x)=a0+a1x+⋯ and
g(x)=b0+b1x+⋯. Thus, proving the existence of g is
equivalent to proving that there exists a sequence {bi} such that
(a0+a1x+⋯)(b0+b1x+⋯)=1. Expanding this
product, and combining the terms of equal degree, we get:
a0b0+(a0b1+a1b0)x+⋯+(a0bn+⋯+anb0)xn+⋯=1
Since the RHS is simply 1, we need all terms
(a0bn+⋯anb0)=0 for all n≥1 and a0b0=1.
Now, if f(0)=a0 is not a unit, then we cannot find a b0 in R
such that a0b0=1. This shows that we cannot find the desired
sequence {bi}. Thus, if f(x) is a unit in R[x], it follows that
f(0) is a unit.
We now prove the converse. That is, if f(0) is a unit in R, then
there exists some sequence {bi} so that g(x)=i≥0∑bixi
and f(x)g(x)=1. We use induction in order to do so. We will first
prove that there exists some sequence {bi}i=0 (i.e. a single
term b0) so that a0b0=1. This finishes our base case. Now,
assume that there exists some sequence {bi}i=0n so that
a0b0=1 and for all 0<k≤n we have
(a0bk+⋯akb0)=0. We prove that there exists a sequence
{bi}i=0n+1 so that the same holds for k=n+1 as well. We
have an+1b0+⋯a0bn+1=0. Therefore, we have
bn+1=−a0an+1b0+⋯+a1bn Since we know
bi exist for all i less than or equal to n, we have that
bn+1 also exists, completing our inductive step. Thus, if f(0) is
a unit, then f(x) has an inverse. ◻
We will now generalize as to when f(g(x)) where g(x)∈R[[x]] is
an element of R[[x]] itself.
Proposition 2. Let f,g∈R[[x]]. Then, f(g(x))∈R[[x]] if
and only if g(0)=0.
Proof. Suppose g(0)=0. Then, we can write g(x)=xk+h(x)
where k is the smallest power of x in g(x), and k=0 since
g(0)=0. Moreover, the smallest power of x in h(x) is more than
k. Also, let f(x)=i=0∑∞aixi Therefore,
f(g(x))=a0+a1g(x)+a2g(x)2+⋯=a0+a1(xk+h(x))+a2(xk+h(x))2+⋯
Also let f(g(x))=c0+c1x+c2x2+⋯ In order to prove
that f(g(x))∈R[[x]] it suffices to prove that cn∈R for all
n. So, we will work on finding the coefficient of the xn term, i.e.
cn.
Consider the term an+1(xk+h(x))n+1 in the expansion for
f(g(x)). The term with the lowest power in this expansion is
x(n+1)k. Since k=0, it follows that (n+1)k>n. Therefore,
the coefficient cn of xn is independent of an+1. By a similar
argument, cn is independent of aj for all j>n. In other words,
it follows that cn=i=0∑naiti where ti∈R. Note
that there will be no powers of ai in the expansion since the
coefficients ai are not raised to a power in the expansion of
f(g(x)). Since ai,ti∈R, it follows that aiti∈R. Thus,
∑i=0naiti∈R. Therefore, for all n, we have cn∈R,
where f(g(x))=c0+c1x+⋯ Therefore, by definition of a
formal power series, f(g(x))∈R[[x]].
Now, suppose g(0)=c=0. Let g(x)=c+h(x). Then,
f(g(x))=a0+a1(c+h(x))+a2(c+h(x))2+⋯=c0+c1x+⋯
In the nth term of the above expansion, we have a constant cn.
Since there are infinitely many terms in the expansion, c0 is an
infinite sum, which is not defined in R. Therefore, c0∈/R. It
follows that f(g(x))∈/R[[x]]. ◻
We can define the system R[[x,y]] to (R[[x]])[[y]]. Inductively, we
may define R[[x1,…,xk]]=(R[[x1,…,xk−1]])[[xk]]
Since R is a ring implies R[[x]] is a ring (base case) Then, we
assume that R[[x1,x2,x3,…xk]] is a ring Now,
R[[x1,x2,X3,…xk+1]] is
(R[[x1,x2,x3,…xk]])[[xk+1]] which is a ring (inductive
case). Thus, by induction we can say that
R[[x1,x2,X3,…xn+1]] is a ring.
We will now generalize. For which power series g∈R[[x,y]] can we
define f(g(x,y)) for any f∈R[[t]]?
Proposition 3. For f,g∈R[[x,y]], we have that
f(g(x,y)) is defined in R[[x,y]] if and only if g(0,0)=0.
Proof. Suppose g(0,0)=0. Thus, the smallest power of x and y
can be represented as bmnxmyn where both m and n are not 0
at the same time. So, we let g(x,y)=bmnxmyn+h(x,y). Thus,
f(g(x,y))=f(bmnxmyn+h(x,y))=a0+a1(bmnxmyn+h(x,y))+a2(bmnxmyn+h(x,y))2+⋯=c00+c10x+c01y+⋯
Now, consider the coefficient, cpq. We know that
the term ap+q+1(bmnxm+yn+h(x,y)p+q+1) has smallest
power xm(p+q+1)yn(p+q+1) which is clearly greater than the power
of xpyq. Hence, cpq is independent of ak for all k>p+q.
Therefore, we have that cpq=i=0∑p+qaidi which is
clearly an element of R. Therefore, f(g(x,y)) makes sense in
R[[x,y]].
Now, if g(0,0)=c which is a constant, then let
g(x,y)=c+h(x,y). Therefore, we have
f(g(x,y))=a0+a1(c+h(x,y))+a2(c+h(x,y))2+⋯ The
constant term of the above expansion will be
a0+a1c+a2c2+⋯ which is an infinite sum. This does not
make sense in R since we can only compute finite sums in R.
Therefore, if c=0, then there is no way to make sense of
f(g(x,y)) as an element of R[[x,y]]. ◻
For some field k, we write k[x] to denote the set of polynomials
with coefficients in k. We write k(x) to denote the set
{h(x)g(x)g(x),h(x)∈R[[x]]}
We will now explore the relation between k(x),k[x] and k[[x]].
Firstly, note that k[x]∈k(x) by letting g(x)=1. Also, clearly
k[x]∈k[[x]]. This is because a polynomial is a formal power series
with the coefficients equal to 0 for all xn where n is greater
than the degree of the polynomial. We now check when an element of
k(x) can be written as a formal power series. Consider an example of
the power series expansion of a rational function g(x)/h(x), t(x).
Let
g(x)=2x+1 and h(x)=x1+1. Then,
h(x)g(x)=x2+12x+1=1+x+2x−x3−2x4+x5+2x6−⋯
Notice how the terms of the power series are recursive. The nth term
of t(x) can be found in terms of the previous terms of t(x). In
fact, this holds in general too!
Proposition 4. We claim that a rational function g(x)/h(x), i.e.
an element of k(x) with h(0)=0 can be written as a power series
t(x).
Proof. If h(0)=0, then we have already proved that h(x) is a
unit. Thus, 1/h(x)∈R[[x]]. So, since formal power series are
closed under multiplication, g(x).(1/h(x))∈R[[x]]. Thus,
g(x)/h(x) can be written as a power series t(x). If h(0)=0, then
h(x) is not a unit which implies 1/h(x)∈/R[[x]] and thus,
g(x)/h(x) cant be expressed as a formal power series. ◻
Proposition 5. Let t(x)=g(x)/h(x) be the power series expansion
of a rational function. Then, the terms an of t(x) satisfy a linear
recursion. Formally, this means that there exists a number m≥1 and
constants c1,…,cm∈k such that for all n that is
sufficiently large, an=i=1∑mcian−i
Proof. Firstly, let the degree of h(x) be m and the degree of
g(x) be k. Let h(x)=b0+b1x+⋯bmxm and
t(x)=a0+a1x+⋯. Now, since t(x)=g(x)/h(x), we may
write g(x)=t(x)h(x).
Since the degree of g(x) is k, it follows that the coefficients of
xn in the expansion of the product on the RHS is 0 for all n that
is greater than k.
Now, consider some n>max(m,k). Clearly, n>k. Therefore, the
coefficient of xn in the expansion will be 0. Let us find the
coefficient using the fact that t(x)=a0+a1x+⋯ and
h(x)=b0+b1x+⋯+bmxm. This will be
anb0+an−1b1+⋯+an−mbm Since n>k, the above
must be 0. Rearranging the terms, we get
anb0=−(an−1b1+⋯an−mbm) Since k is a field and
b0=h(0)=0, we have
an=−b0−1(an−1b1+⋯an−mbm)=i=1∑mcian−i
where ci=−b0−1bi. This proves that the coefficients of the
power series t(x) satisfy a linear recursion. ◻
Proposition 6. The converse of the previous lemma also holds true.
That is, if t(x)=∑n≥anxn∈k[[x]] is such that the
coefficients an satisfy a linear recursion, then there exists some
rational function g(x)/h(x) which has a power series expansion of
t(x).
Proof. We know that for t(x), we have
an=c1an−1+c2an−2+⋯cman−m for n that is
sufficiently large enough, since we are given that the coefficients of
t satisfy a linear recursion. Therefore, we have
an−c1an−1−⋯−cman−m=0 Let b0=1 and
bi=−ci when i≥1. Therefore, we may write
anb0+an−1b1+⋯an−mbm=0 Now, consider
h(x)=b0+b1x+⋯bmxm. Consider the product t(x)h(x).
The coefficient of xn in this product will be
b0an+b1an−1+⋯bman−m which is 0 by our
assumption. Therefore, all terms of t(x)h(x) have a coefficient of 0
for n sufficiently large. Suppose for all n>k we have that the
coefficient of xk is 0 (and not 0 for n not more than k).
Therefore, t(x)h(x)=g(x) is a polynomial of degree k. Therefore,
g(x)/h(x) is a rational function that is represented by t(x).
Therefore, we have found some rational function that is represented by
t(x) given that the coefficients of t(x) satisfy a linear recursion.
This completes our proof. ◻
Definition 3. The set of p-adic numbers Zp is defined as:
Zp={(a1,a2,…)∣ai∈Z/piZ and ai+1≡ai(modpi)
One can prove that the set of p-adic numbers, Zp forms a
ring under term-wise addition and multiplication.
Proposition 7. The ring Zp is an integral domain, i.e.,
if for some a,b∈Zp we have ab=0, then either a=0
or b=0.
Proof. In Zp, we have 0=(0,0,…). Thus, we need
to prove that either a=(0,0,…) or b=(0,0,…) given
that ab=(0,0,…). First, let a=(a1,a2,…) and
b=(b1,b2,…). Then,
ab=(a1b1,a2b2,…)=(0,0,…) This implies that we
must have aibi≡0(modpi) for all i. Clearly, we must
also have a1b1≡0(modp). Since Zp is an
integral domain, we must have a1≡0(modp) or
b1≡0(modp). Assume without loss of generality that
a1≡0(modp). We will prove that either ai=0 for all i
or bi=0 for all i.
We will prove this by contradiction. Suppose that ai=0 and
bi=0. Then, there exists some ak such that
ak≡0(modpk). Also assume without loss of generality that
bi=0 for some i≤k (if the smallest such i is greater
than k, then we just exchange a and b). If
ak≡0(modpk), then by the definition of a p-adic number,
we must have ak+1≡ak≡0(modpk). Thus,
ak+1=0. It follows that for any i≥k, we must have
ai=0 as well as bi=0.
Now, since ab=0, we must have aibi≡0(modpi) for all
i≥k. Now, we have ak−1≡0(modpk−1) but
ak≡0(modpk). Moreover, by definition, we have
ak≡ak−1(modpk−1). Therefore, it follows that
ak≡pk−1m1(modpk) where 0<m1<p. Next, we have
ak+1≡pkm0+pk−1m1(modpk+1) where
0<m0<p. In general, we therefore have
ak+i≡pk+i−1mi−1+⋯+pkm0(modpk+i)
Let us say the smallest value of i for which bi=0 is i=r.
We assumed earlier (WLOG) that r<k. Now, just as before, we may
write:
br+j≡pr+j−1nj−1+⋯+prn0(modpr+j)
Since k>r, let k=r+l where l is some natural number. Also,
let i=r−k+j. Therefore,
br+j=bk+i≡pk+i−1nj−1+⋯+prn0(modpk+i)
Now, vp(ak+ibk+i)=vp(ak+i)+vp(bk+i). Since the
smallest power of p in the expansion of ak+i is k. Therefore,
we have vp(ak+i)=k. Since the smallest power of p in the
expansion of bk+i is r, we have that vp(bk+i)=r<k.
Thus, vp(ak+ibk+i)=kr≤k2. Now, consider the case when
i=k2−k+1. We therefore have vp(ak+1bk2+1)≤k2.
This implies that ak2+1bk2+1≡0(modpk2+1)
However, since ab=0, we must have
ak2+1bk2+1≡0(modpk2+1). A contradiction. Thus,
we cannot have both a=0 as well as b=0. ◻
Let us explore what the units in Zp are. We first take an
example. Consider Z3. Consider
a=(1,4,13,40,…)∈Zp. Also, let
b=(1,7,25,79,…). Clearly, b also is an element of
Zp. Moreover,
ab=(1⋅1,4⋅7,13⋅25,40⋅79,…)=(1,1,1,1,…)
Thus, b is the inverse of a in Zp. Since a has an
inverse in Zp, it follows that a is a unit in
Zp. Note that however, a=(0,3,12,…) is not a unit
in Zp. It seems that in general, a∈Zp is a
unit if and only if the first term of a is non-zero. In other words,
we claim that a∈Zp is a unit if and only if
a≡0(modp).
Proposition 8. In Zp, a p-adic number u is a unit
if and only if u1≡0(modp)
Proof. Let u=(u1,u2,…). Suppose u1=0. Then, there is
clearly no inverse for u since there is no
b1∈Z/pZ such that u1b1≡1(modp)
(0 is not a unit modulo p). Hence, if u is a unit, then it follows
that u1≡0 or equivalently, u≡0(modp).
We now prove the other direction. Suppose u≡0(modp). Then,
we must prove that u is a unit in Zp. Since p is prime,
if u=0(modp), it means that u1≡0(modp). Since
p is a prime, it follows that (u1,p)=1. Thus, u1 is a unit in
Z/pZ.
Now, suppose that (uk,p)=1. Now, by definition of a p-adic
number, we have uk+1≡uk(modpk). This in turn implies
that uk+1≡uk(modp). Thus, (uk+1,p)=1. Hence,
(uk+1,pk+1)=1. Thus, uk+1 is a unit in
Z/pk+1Z.
By induction, this implies that for all k, we have uk is a unit in
Z/pkZ. Since the multiplication operation is
termwise in Zp, it follows that u is a unit in
Zp. ◻
This leads to another important result:
Proposition 9. Let a=0 be a p-adic number. Then, there
exists a unique pair (u,k) such that a=pku and u is a unit.
Proof. We will first prove existence. Let a=(a1,a2,…). If
a1=0, then by the previous proposition, we have that a itself
is a unit. Thus, a=p0u where u=a.
Now, suppose that ak=0 for some k (this automatically implies
that a1,a2,…,ak=0 by the construction of p-adic numbers)
such that ak+1=0. Now, consider the p-adic number
b=(ak+1/pk,ak+2/pk,…) We must first prove that b
is indeed a p-adic integer.
For any i>k, by the definition of p-adics, we know that
ai≡ak≡0(modpk). Thus, ai/pk is an integer for
all ai.
Now, for some i, we have ak+i+1≡ak+i(modpk+i) by
the definition of p-adics. Thus, ak+i+1=pin+ak+i where
0<n<p. Dividing both sides by pk, we get
pkak+i+1=pin+pkak+i Hence, taking
mod pi on both sides, we get
pkak+i+1≡pkak+i(modpi) Hence,
we conclude that b∈Zp. Now, consider the p-adic integer
pkb. We get pkb=pk(ak+1/pk,ak+2/pk,…,ak+k/pk,ak+k+1/pk,…)=(ak+1,ak+2,…,ak+k,ak+k+1,…) Now, ai+1≡ak(modpi), or in general,
ak+i≡ai(modpi) by induction on k. Thus,
pkb=(a1,a2,…)=a Now, clearly,
ak+1/pk≡0(modp) since ak+1 is nonzero. Thus, b
is a unit. Thus, for arbitrary a, we have found a specific unit b
and power k such that a=pkb, which completes our proof for
existence.
Now, we prove uniqueness. To do so, suppose that
a=pk1u1=pk2u2. Assume WLOG that k1≥k2. Then,
pk1u1−pk2u2=pk2(pk1−k2u1−u2)=0 Since
Zp is an integral domain, it follows that either
pk2=0 or pk1−k2u1−u2=0. Since pk1 is clearly
not 0, it follows that pk1−k2u1=u2 If k1>k2, then the
first term of u2 will be 0, which is a contradiction since u2 is
a unit. Therefore, k1=k2. Thus, u2=p0u1=u1. Hence, the
pair (u,k) is unique. ◻
Now, we will explore some properties related to Zp[x], i.e.
polynomials with coefficients in Zp.
Theorem 1. Let f(x)∈Zp[x]. Consider some a1 (if
there exists one) such that f(a1)≡0(modp) such that
f’(a1)≡0(modp). Then, there exists a unique
a∈Zp such that f(a)=0 and a≡a1(modp).
Proof. If we prove that if there exists a root, ak to f(x) in
Z/pkZ with f’(ak)=0, then there exists a
unique ak+1 such that f(ak+1)=0 in
Z/pk+1Z, it will by induction imply that there
exists some unique p-adic number a=(a1,a2,…) such that
f(a)=0 in Zp. However, note the condition that
ak+1≡ak(modpk). Let us first prove this result for
k=1.
Suppose f(a1)≡0(modp) and f’(a1)≡0(modp). We
will try to find some a2 such that f(a2)≡0(modp2) and
a2≡a1(modp). f(x)=c0+c1x+⋯+cnxn in
Z/pZ. Then, we have
f(a1)=c0+c1a1+⋯+cna1n. Consider a2=a1+mp
for some integer 0<m<p. Then, we have f(a2)=f(a1+mp)=c0+c1(a1+mp)+⋯+cn(a1+mp)n=c0+c1(a1+mp)+⋯+cn(a1n+na1n−1mp+⋯+(mp)n) Now, if we take mod p2 on both sides of the above,
all terms with powers of p over 2 get cancelled to 0. Thus, we
get: f(a2)≡c0+c1(a1+mp)+c2(a12+2a1mp)+⋯+cn(a1n+na1n−1mp)≡(c0+c1a1+⋯cnan)+(c1(mp)+2c2(mp)a1+⋯ncn(mp)a1n−1)≡f(a1)+mpf’(a1)(modp2) Now, we know that a2≡a2(modp) since
a2=mp+a1. If a2 is a root modulo p2, it implies that
f(a2)≡0(modp2). Therefore,
f(a1)+mpf’(a1)≡0(modp2). Since
f(a1)≡0(modp), let f(a1)=kp. So, we must have
kp+mpf’(a1)≡0(modp). So,
m≡−k(f’(a1))−1(modp). We can do this step since
f’(a1)≡0(modp). Thus, we have found a unique m such that
f(a1+mp)≡0(modp2). In other words, given a value a1,
we have proved that there exists a unique a2 such that
f(a2)≡0(modp2).
Now, suppose we are given some ak such that
f(ak)≡0(modpk). We will prove that there exists a unique
ak+1 such that f(ak+1)≡0(modpk+1) and
ak+1≡ak(modpk).
Let f(x)=c0+⋯+cnxn. Then,
f(ak)≡c0+c1ak+⋯cnann≡0(modpk) Now,
since ak+1≡ak(modpk), we have
ak+1≡pkm+ak where 0<m<p. Therefore, f(ak+1)=c0+c1ak+1+⋯+cnak+1n=c0+c1(ak+pkm)+⋯+cn(ak+pkm)n=c0+c1(ak+pkm)+⋯+cn(akn+nakn−1mp+⋯(mp)n) If we take mod pk+1 on both sides, all terms apart
from the pk and constant terms cancel out since 2k>≥k+1 for
all positive k. Thus, f(ak+1)≡c0+c1(ak+pkm)+c2(ak2+2akpkm)+⋯+cn(akn+nakn−1pkm)≡(c0+c1ak+⋯cnakn)+pkm(c1+2c2ak+⋯ncnakn−1)≡f(ak)+pkmf’(ak)≡0(modpk+1) Since f(ak)≡0(modpk), we have
f(ak)=pkt. So, pkt+pkmf’(ak)≡0(modpk+1).
Hence, −t+mf’(ak)≡0(modp) Hence,
m≡t(f’(ak))−1(modp). Since
f’(ak)≡0(modpk), its inverse exists and is unique. Thus,
there exists a unique m, given as above, such that
ak+1=mpk+ak is a root of f(x) in
Z/pk+1Z and ak+1≡ak(modpk).
We are given the value a1 that is a root of f(x) in
Z/pZ. Thus, we can find the corresponding value of
a2 and hence a3 and so on. Thus, by induction, there exists a
unique sequence a=(a1,a2,…) that satisfies
f(ak)≡0(modpk) and ak+1≡ak(modpk). By
definition of a p-adic number, a∈Zp. Moreover, since
f(ak)≡0(modpk) it follows that f(a)=0 in
Zp. Thus, there exists a unique p-adic number
a∈Zp such that f(a)=0 given a1. ◻
For example,
1(1,4,13,…)+3(3,3,3,…)+9(2,5,14,…)∈Q3
Note that the above is NOT equal to
(1,4,13,…)+(1,1,1,…)+(92,95,913,…)
The divided by sybmol is merely used as a notation and does not
translate to the above. More generally, any element of Qp
is: a0+pa1+⋯+pkak By taking pk
as the common denominator, the above can be rewritten as:
pka0pk+a1pk−1+⋯+ak=pka where
ai,a∈Zp. Moreover, we have that two p-adic numbers
pka and pmb are equal if and only if
apm=bpk.
Proposition 10. Qp is a field. Moreover,
Q∈Qp.
Proof. Consider some p-adic number α=pka. We will
prove that there exists some p-adic number β=pmb
such that αβ=1 for every α∈Qp that is
non-zero, i.e. a=0.
Firstly, we know that since a∈Zp, we have that there
exists some a’ that is a unit in Zp and n1 a
non-negative integer so that a=a’pn1. Let
β∈Qp be such that
β=1b=1(a’)−1pk−n1 The inverse of
a’ exists since a’ is a unit as defined. Then,
αβ=pka’pn1⋅1(a’)−1pk−n1=pkpk
Now since a/pk=b/pm⟺apm=bpk, it follows that the above
equals 1 if and only if pk=pk. Thus, αβ=1. Thus, we
have found a β∈Qp given
α∈Qp∖{0} such that αβ=1.
Hence, Qp is a field.
We will now prove that Q∈Qp. Consider an element
of Q, say a/b. Let a=pk1a’ and b=pk2b’
where a’ and b’ are co-prime with p. Thus,
ba=pk1pk2b’a’ where
α∈Zp. Thus, any rational number is also a p-adic
number. Hence Qp contains Q. ◻
Definition 5. We call a set X along with a function (metric)
d:X×X→R, (X,d), a metric space if the
following properties are satisfied for any x,y,z∈X:
d(x,x)=0
If x=y then d(x,y)>0
d(x,z)≤d(x,y)+d(y,z)
d(x,y)=d(y,x)
For example, (Q,∣⋅∣) is a metric space.
Definition 6. For a metric space (X,d), we define a Cauchy
sequence of X wrt d as follows: A sequence (xn)n≥0 where xn∈X
is called a Cauchy sequence if for all
ϵ∈R+, there exists someNϵ∈Nsuch that for all integersm,n≥N, we
haved(xn,xm)<ϵ.
In simple terms, any sequence that should converge under the given
metric is called a Cauchy sequence. We now define what a completion with
respect to a metric is.
Definition 7. Let S(X,d) be the set of Cauchy sequences in
X with respect to the metric d. Then, given two elements xn,yn
of S(X,d), we say that xn is equivalent to yn iff for every
real number ϵ, there exists a natural number Nϵ such
that for all integers n≥N, we have d(xn,yn)<ϵ. We
denote this equivalence by xn∼yn.
Definition 8. We define the completion of X with respect to d
as the set S(X,d)/∼.
In more intuitive terms, the completion of a metric space X with
respect to d is the set of limits of the Cauchy sequences in X with
respect to d.
As an example, R is a completion of Q with respect
to the usual metric, the absolute value.
Now, we will define Qp as a completion of Q. In
order to do so, we must first define the metric on Q that
yields Qp.
Definition 9. Let p be a prime and a be an integer. Then, we
define vp(a) to be the largest n such that pn∣a when a=0
and to be ∞ when a=0. Now, we extend this definition to
Q. Let q=a/b∈Q where a and b are integers where
b=0. Then, we define vp(q)=vp(a)−vp(b).
Lemma 1. The above definition of vp() is well-defined. That is,
if a/b=c/d=q, then we have vp(a/b)=vp(c/d).
Proof. Suppose a/b=c/d such that vp(a/b)=vp(c/d)+m. Then,
vp(a)−vp(b)=vp(c)−vp(d)+m. Let
a=pk1a’,b=pk2b’,c=pk3c’,d=pk4d’ with each
of a’,b’,c’,d’ relatively prime to p. Therefore,
k1−k2k3−k4+m. Moreover, we have a/b=pk1−k2a’/b’ and
c/d=pk3−k4c’/d’. Since a/b=c/d, we have
pk1−k2b’a’=pk3−k4d’c’ We can rewrite
this as pk1−k2a’d’=pk3−k4b’c’ Now, we assumed that
k1−k2=k3−k4+m. Let k3−k4=n. So, pn+ma’d’=pnb’c’
which implies that pma’d’=b’c’. Taking the vp of both sides, we
have m=0. Thus ◻
Definition 10. We define the p-adic absolute value to be as
follows. Let q∈Q. Then, we define
∣q∣p=p−vp(q) when q=0 and 0 when q=0.
We will first explore some properties of vp(a) which will help us
find a metric on Q that gives Qp.
Proposition 11. For any a,b∈Z, the following are
true:
vp(ab)=vp(a)+vp(b)
vp(a+b)≥min(vp(a),vp(b))
If vp(a)=vp(b), then vp(a+b)=min(vp(a),vp(b))
Proof. Let vp(a)=k1 and vp(b)=k2. WLOG, assume that
k1≥k2. Then, clearly a=pk1a’ and b=pk2b’ where
(a’,p)=1 and (b’,p)=1. Then:
We have ab=a’b’pk1+k2 where (a’b’,p)=1 since each of
a’ and b’ is co-prime to p. Hence,
vp(ab)=k1+k2=vp(a)+vp(b).
We have a+b=a’pk1+b’pk2. Since k1≥k2, we can
write a+b=pk2(a’pk1−k2+b’) Hence,
vp(a+b)=k2+vp(a’pk1−k2+b’) Clearly, the above is
at least k2 since vp(a’pk1−k2+b’)≥0. Hence,
vp(a+b)≥k2=min(vp(a),vp(b)) since we assumed that
vp(a)≤vp(b).
Since we have that vp(a)=vp(b), we know k1>k2. Thus,
a’pk1−k2+b’≡0+b’≡b’(modp). Since b’ is
co-prime with p, we have a’pk1−k2+b’≡0(modp).
Hence, p∤(pk1−k2a’+b’). Thus,
vp(a’pk1−k2+b’)=0. Hence,
vp(a+b)=k2=vp(b)=min(vp(a),vp(b)) since we assume
that vp(a)≤vp(b).
◻
Proposition 12. The above properties all hold for any
r,s∈Q.
Proof. Let r=a1/b1 and s=a2/b2 where ai,bi are
integers with bi=0. We know by definition that
vp(r)=vp(a1)−vp(b1) and similarly
vp(s)=vp(a2)−vp(b2). Hence:
We have
vp(rs)=vp(b1b2a1a2)=vp(a1a2)−vp(b1b2)
Since ai∈Z we have
vp(a1a2)=vp(a1)+vp(a2) and similarly
vp(b1b2)=vp(b1)+vp(b2). Thus,
vp(rs)=(vp(a1)−vp(b1))+(vp(a2)−vp(b2))=vp(r)+vp(s)
We can write vp(r+s) as
vp(b1a1+b2a2)=vp(b1b2a1b2+a2b1)=vp(a1b2+a2b1)−vp(b1b2)
Now, a1b2 and a2b1 are integers. So,
vp(a1b2+a2b1)≤min(vp(a1b2),vp(a2b1))
Thus,
vp(r+s)≥min(vp(a1)+vp(b2),vp(a2)+vp(b1))−(vp(b1)+vp(b2))=min(vp(a1)−vp(b1),vp(a2)−vp(b2))=min(vp(r),vp(s))
We can legally perform the above step since vp(b1b2) is a constant.
If vp(a1b2)=vp(a2b1), then we have
vp(a1)+vp(b2)=vp(a2)vp(b1). Thus, by rearranging the
terms, we have vp(r)=vp(s). Thus, if vp(r)=vp(s), it
follows that vp(a1b2)=vp(a2b1). Hence,
vp(a1b2+a2b1)=min(vp(a1b2),vp(a2b1)). It follows
directly that vp(r+s)=min(vp(r),vp(s)).
We have thus extended all the properties that hold for integers under
vp to rationals. ◻
We will now use these properties to prove some properties regarding
∣⋅∣p, which will in turn prove that
(Q,∣⋅∣p) is a metric space.
Proposition 13. The following properties are true regarding
∣⋅∣p:
∣q∣p≥0 and equality holds iff q=0
∣q+r∣p≤max(∣q∣p,∣r∣p)
∣qr∣p=∣q∣p∣r∣p
Proof.
We have ∣q∣p=p−vp(q). Clearly this
is at least 0 for all vp(q) since vp(q)∈Q. We
have p−vp(q)=0 iff vp(q)=∞ which happens only when
q=0.
We have
∣q+r∣p=p−vp(q+r)≤p−min(vp(q),vp(r))=pmax(−vp(q)),−vp(r)
since vp(q+r)≥min(vp(q),vp(r))). Thus,
∣q+r∣p≤max(∣q∣p,∣r∣p)
Clearly, equality holds when
∣q∣p=∣r∣p,
which follows from the property of vp.
We have
∣qr∣p=p−vp(qr)=p−vpq−vp(r)=p−vp(q)p−vp(r)=∣q∣p∣r∣p
◻
With these properties, we may now define a new metric on Q.
Consider the metric dp which is defined as follows:
Definition 11. We define the function
dp:Q×Q→R as follows:
dp(x,y)=∣x−y∣p.
Proposition 14. Q is a metric space over dp.
Proof. In order to prove this, we must prove all properties listed in
the definition of a metric space.
dp(x,x)=∣x−x∣p=∣0∣p=0
dp(x,y)=∣x−y∣p>0 when x=y by
the previous proposition.
dp(x,z)=∣x−z∣p=∣(x−y)+(y−z)∣p.
By property 2 above,
∣(x−y)+(y−z)∣p≤max(∣x−y∣p,∣y−z∣p)≤∣x−y∣p+∣y−z∣p
since ∣q∣p≥0. Hence,
dp(x,z)≤dp(x,y)+dp(y,z).
dp(x,y)=∣x−y∣p=∣y−x∣p=dp(y,x)
Thus, Q is a metric space over dp. ◻
We can thus finally define Qp in terms of Q as
follows:
Definition 12. We define Qp to be the completion of
Q with respect to dp. Thus,
Qp={[an]∣an∈Sp} where Sp denotes the set
of Cauchy sequences in Q with respect to dp.
Now, we know that Q is a field with operations (+,⋅).
We will now try to find a pair of binary operations (+,⋅) on
Qp such that (Qp,+,⋅) is a field.
Let α∈Qp. Therefore, we can write α=[a]
for some a∈Sp by definition of Qp. Similarly, let
β=[b]∈Qp. We define the addition of two elements
in the p-adics as α+β=[a]+[b]=[a+b] and their
product as α⋅β=[a]⋅[b]=[a⋅b] However, we
have not yet defined what addition and multiplication are in Sp (a
and b are elements of Sp). We define the sum of two sequences
a=(ai) and b=(bi) to be a+b=(ai+bi), i.e. the termwise sum
of the terms of the sequence. Similarly, we define their product to be
a⋅b=(aibi). Since both a and b converge, it follows that
both a+b as well as ab converge. Thus, Sp is closed under
multiplication.
Definition 13. Let S be the set of all Cauchy sequences of
Q, we define a set of equivalence classes in S such that
S/∼={[a]∣a∈S}[a]={(ai)∼y∣∃(yi)∈Si→∞lim∣ai−yi∣p=0}
Lemma 2. The set S/∼ has a well-defined addition and
multiplication[a]+[b]∼[a+b][a][b]∼[ab]
Proof. We need to show that,
∣(ai+bi)−(ai’+bi’)∣p=0. To prove this, we can
observe the following,
∣(ai+bi)−(ai’+bi’)∣p=∣ai−ai’+bi−bi’∣p
It follows that,
∣ai−ai’+bi−bi’∣p≤∣ai−ai’∣p+∣bi−bi’∣p
the RHS converges to 0 as i→∞. For multiplication,
we have, ∣aibi−ai’bi’∣=∣aibi−aibi’+aibi’+ai’bi’∣∣ai(bi−bi’)+bi’(ai−ai’)∣p≤∣ai(bi−bi’)∣p+∣bi’(ai−ai’)∣p∣ai(bi−bi’)∣p+∣bi’(ai−ai’)∣p=∣ai∣p∣(bi−bi’)∣p+∣bi’∣p∣(ai−ai’)∣p Since (ai) and (bi’) are bounded, then we can proceed:
∣ai∣pi→∞lim∣(bi−bi’)∣p+∣bi’∣p∣i→∞lim(ai−ai’)∣p=ϵ⋅0+ϵ⋅0=0 ◻
Now, we will prove that Qp is a field.
Definition 14. If λ is an element of Qp and
(xn)∈S/∼ is any Cauchy sequence representing λ, we
define ∣λ∣p=n→∞lim∣xn∣p
Proposition 15. Qp=S/∼ is a field
Proof. Let (1/xi) denote the multiplicative inverse of (xi) in
Qp. To know that 1/xn is Cauchy, observe the following
xi1−xj1p=xjxixj−xip
Since (xn) is Cauchy, we have that (xj−xi) is bounded by
ϵ. It follows that for a fix large N and for every
i,j>N, we have that for every ϵ>0, the following holds
xi1−xj1p<ϵ ◻
Till now, we only talked about sequences in Q, and used them
to define the system Qp. We will now talk about sequences in
Qp itself. They can be thought of as sequences of sequences,
since each element of Qp is itself a sequence of elements in
Q. We will now extend the p-adic absolute value to even
p-adic numbers. There are two things we need to take care of here.
Firstly, we only know what ∣q∣p where q is a
rational number is, not a sequence of rational numbers. So, we need to
define the notion of p-adic absolute value for Cauchy sequences. Next,
[a] is not a single sequence. It is a set of sequences. Therefore, we
need to show that for any sequence in [a], the result of
∣[a]∣p is the same. In other words, if
x∼y are two sequences, then we must show that
∣x∣p=∣y∣p.
Definition 15. Let (xn) be a Cauchy sequence. Then, the sequence
(∣xn∣p) converges to some real number, say
y∈R. Then, we define ∣(xn)∣p
to be y. In other words,
∣(xn)∣p=limn→∞∣xn∣p.
Proof. We will prove that if (xn) is a Cauchy sequence, then
(∣xn∣p) converges in R. So, we
must prove that for all ϵ∈R+, we have that there
exists some Nϵ∈N such that for all
m,n≥Nϵ we have
∣∣xn∣p−∣xm∣p∣<ϵ
By the definition of a Cauchy sequence with respect to
∣⋅∣p, we have that for every
ϵ∈R+, there exists some natural number
Nϵ such that for all m,n≥Nϵ, we have
∣xn−xm∣p<ϵ. Let us now keep
ϵ and Nϵ fixed. Now, consider some integers
m,n≥Nϵ such that
∣xn∣p=∣xm∣p.
Then, we have
∣xn−xm∣p=max(∣xn∣p,∣xm∣p)<ϵ
Therefore, we have
0<∣xm∣p<∣xn∣p<ϵ
(WLOG). Thus,
∣xn∣p−∣xm∣p<ϵ.
In general,
∣∣xn∣p−∣xm∣p∣<ϵ.
Now, if
∣xn∣p=∣xm∣p, this
implies that
∣∣xn∣p−∣xm∣p∣=0<ϵ
since ϵ∈R+. Thus, for all m,n≥Nϵ
we have
∣∣xm∣p−∣xn∣p∣<ϵ.
This holds true for every value of ϵ that is a positive real
and its corresponding Nϵ. Therefore, the sequence
(∣xn∣p) converges. ◻
We will now show that ∣[a]∣p is well defined
where [a] is the equivalence class of a Cauchy sequence a of
Q. Recall that this is equivalent to proving that if
xn∼yn then
∣xn∣p=∣yn∣p.
Proposition 16. If (xn),(yn)∈Sp are equivalent (where
equivalence is as defined before), then
∣(xn)∣p=∣(yn)∣p.
In other words,
n→∞lim∣xn∣p=n→∞lim∣yn∣p
Proof. Given that (xn)∼(yn), by definition, we have that for
all ϵ∈R+, there exists some
Nϵ∈N such that for all n≥Nϵ, we
have ∣xn−yn∣p<ϵ. In other words, if
(xn)∼(yn), then
n→∞lim∣xn−yn∣p=0 Let us
assume for the sake of contradiction that
limn→∞∣xn∣p=limn→∞∣yn∣p.
Also, let us assume WLOG that
limn→∞∣xn∣p>limn→∞∣yn∣p.
Since
∣xn−yn∣p=max(∣xn∣p,∣yn∣p)
when
∣xn∣p=∣yn∣p, it
follows that
n→∞lim∣xn−yn∣p=n→∞lim∣xn∣p=0
However, we assumed that
limn→∞∣yn∣p<limn→∞∣xn∣p=0.
A contradiction since the absolute value is always positive. It follows
that
limn→∞∣xn∣p=limn→∞∣yn∣p.
Hence,
∣(xn)∣p=∣(yn)∣p. ◻
From this, one may conclude that ∣α∣p is
well defined for all α∈Qp. Moreover, the same
properties hold for ∣α∣p for
α∈Qp as in Q, which can be easily seen
from the fact that
∣(xn)∣p=limn→∞∣xn∣p.
With the new extension of the p-adic absolute value in the p-adic
numbers, we may now extend the notion of convergence to Qp
as well.
Definition 16. Consider a sequence (xn) where
xn∈Qp. We say that (xn) converges to a limit
L∈Qp, iff for all ϵ∈R+, there is
some natural number Nϵ such that for all n≥Nϵ
we have ∣xn−L∣p<ϵ. Note that here,
each xk is itself a p-adic number. So, (xn) is essentially a
sequence of sequences of rationals.
We now give the condition of convergence for a sequence in
Qp.
Definition 17. A sequence (an) converges iff there exists an
element of Qp that it converges to.
Proposition 17. In Qp, a sequence (xn) converges
then (∣xn∣p) converges in R.
Moreover,
limn→∞∣xn∣p=∣limn→∞xn∣p
Proof. Suppose (xn) converges to L∈Qp. Then, by
definition, we have that
∀ϵ∈R+∃Nϵ∈N st ∀n≥Nϵ,∣xn−L∣p<ϵ
Now,
∣xn−L∣p=max(∣xn∣p,∣L∣p)<ϵ
if ∣xn∣p=∣L∣p.
Therefore, we have 0≤∣xn∣p<ϵ and
0≤∣L∣p<ϵ for all real numbers
ϵ>0. Thus, we have
∣∣xn∣p−∣L∣p∣<ϵ.
If ∣xn∣p=∣L∣p the
previous statement still holds. Therefore,
n→∞lim∣xn∣p=∣L∣p=n→∞limxnp
which proves the desired result. ◻
Definition 18. A series ∑i≥0ai is said to converge if
and only if the sequence (Sn) given by Sn=∑i=0nai
converges.
We will now give the condition of convergence for a series in
Qp. The condition turns out to be much simpler than that in
R, in which there are multiple convergence tests for series.
Proposition 18. The series ∑n≥0an converges in
Qp⟺ the sequence (an)n≥0 converges to 0 in
Qp.
Proof. Consider the sequence (Sn) given by Sn=∑i=0nai.
We need to show that (Sn) converges if and only if (an)
converges.
Suppose (Sn) converges. We know that, by definition, (Sn)
converges if and only if for all ϵ∈R+ we have
that there exists some natural number Nϵ such that for all
m,n≥Nϵ we have
∣Sm−Sn∣p<ϵ. Consider some
n≥Nϵ. Then, we clearly have that
∣Sn+1−Sn∣p<ϵ Now, by
definition of Sn, we have Sn+1−Sn=an+1. Thus,
∣an+1∣p<ϵ for all
n≥Nϵ. Therefore, in general, we have that for every
ϵ∈R+, there is some Mϵ=Nϵ+1
such that for all n≥Mϵ we have
∣an∣p<ϵ. It follows that (an)
converges to 0, by definition. Note that this direction of the proof
holds in R as well. The opposite direction is what makes
convergence in R more difficult.
Now, we will prove that if (an) converges to 0, then (Sn)
converges. By definition, we have that for every
ϵ∈R+, there exists some natural number
Nϵ such that for all n≥Nϵ we have
∣an∣p<ϵ. Now, consider some integers
m,n such that m>n≥Nϵ. We have
Sm−Sn=an+1+⋯+am Thus, ∣Sm−Sn∣p=∣an+1+⋯+am∣p≤max(an+1,…,am) Since for all n≥Nϵ, we have that
an<ϵ, it follows that max(an+1,…,am)<ϵ.
Hence ∣Sm−Sn∣p<ϵ for all
m,n>Nϵ.
Thus, in general, for all ϵ∈R+, we have that
there exists some Nϵ∈N such that for all
m,n≥Nϵ, ∣Sm−Sn∣p. Hence,
(Sn) converges. ◻
Note that the above proof holds entirely because of the fact that
∣a+b∣p≤max(∣a∣p,∣b∣p).
We define the radius of convergence of a power series
∑n≥0anxn to be the value r so that the sequence
∣an∣pcn converges to 0 for all c<r and
does not converge for c>r. The following result is fundamental:
Proposition 19. The radius of convergence of ∑n≥0anxn
is given by
r=(limsup∣an∣p1/n)−1
Definition 19. For any a∈Qp and r∈R+,
we define a closed disc of radius r, centered at a to be the set
D(a;r):={z∈Qp:∣z−a∣p≤r}
and an open disc of radius r, centered at a to be the set
D(a;r−):={z∈Qp:∣z−a∣p<r}.
Now, consider f(x)=∑n≥0anxn∈Qp[[x]], a
power series, and suppose that its radius of convergence is r.
Therefore, we can define a function f:D(0;r−)→Qp
so that for any t∈D(0;r−), we have
f(t)=n→∞lim(k=0∑naktk) Since
t∈D(0;r−), the above sum indeed converges, and therefore, f is
well defined.
We now defined continuity in Qp.
Definition 20. We say that a function f:S→Qp
is continuous at a point x∈S if for all
ϵ∈R+, there exists some positive real δ
such that ∣x−y∣p<δ implies
∣f(x)−f(y)∣p<ϵ.
Definition 21. We say that a function f:S→Qp is
continuous, if it is continuous at every point in S.
Proposition 20. Every function f:D(0;r−)→Qp
such that f(x)=n→∞limk=0∑nakxk for all
x∈D(0;r−) is continuous in Qp.
Proof. We first prove a lemma:
Lemma 3. Let Sn=∑k=0nak and let
S=limn→∞Sn. Then,
∣S∣p≤limn→∞max(∣a1∣p,…,∣an∣p).
This can also be written as
max(∣a1∣p,∣a2∣p,…).
Notationally, we shall express this as
max(∣ak∣p)n≥0.
Proof. We know that S=n→∞limSn Thus,
∣S∣p=n→∞limSnp
As we saw earlier, the above is equal to
limn→∞∣Sn∣p, which is at most
limn→∞max(∣a1∣p,…,∣an∣p). ◻
Let x∈D(0;r−) be any point in D(0;r−). Let y be another
point in D(0;r−). Therefore, we have
∣x∣p<r and ∣y∣p<r.
Consider some positive real ϵ. Now, suppose that there is some
δ∈R+ such that
∣x−y∣p<δ. Now, we have ∣f(x)−f(y)∣p=n→∞lim(k=0∑nakxk)−n→∞lim(k=0∑nakyk)p=n→∞lim(k=0∑nak(xk−yk))p=n→∞lim((x−y)k=0∑nak(xk−1+xk−2y+⋯+xyk−2+yk−1))p=∣x−y∣pn→∞lim(k=0∑nak(xk−1+xk−2y+⋯+xyk−2+yk−1))p By the Lemma, we may simplify the above to get:
∣f(x)−f(y)∣p≤∣x−y∣pmax(an(xn−1+xn−2y+⋯+xyn−2+yn−1)p)n≥0
Now,
xn−1+xn−2y+⋯xyn−2+yn−1p≤max(xn−kyk−1p)k=1n−1
Since ∣x∣p<r and
∣y∣p<r, it follows that
xn−kyk−1p<rn−krk−1=rn−1.
Therefore,
xn−1+xn−2y+⋯xyn−2+yn−1p<rn−1
Thus,
∣f(x)−f(y)∣p<∣x−y∣pmax(∣an∣prn−1)n≥0
Now, by the definition of continuity, we must prove that for all
ϵ∈R+, there exists some δ∈R+
such that ∣x−y∣p<δ implies
∣f(x)−f(y)∣p<ϵ. We can let
δ=max(∣an∣prn−1)ϵn≥0
for any given positive real ϵ. Then, clearly
∣x−y∣p<δ implies
∣f(x)−f(y)∣p. Therefore, there always exists
such a real number δ and hence f(x) is continuous. ◻
For an alternative proof, we will use the notion of continuity between
the mapping of two topological spaces.
Definition 22. Let X and Y be topological spaces. The map
f:X→Y is continuous ⟺ the preimage of the open set is
open.
In other words, if you have a function f mapping from one topological
space X to another topological space Y, and for any open set
U⊂Y, the set of all points in X that map to points in U
(i.e., the preimage of U) is open in X, then f is a continuous
map.
Proof. To begin, let’s prove first the following lemma:
Lemma 4. Let a,b∈Qp and r,s∈R+,
we have the following properties of D(a;r−):
If b∈D(a;r−), then D(a;r−)=D(b;r−).
The open disc D(a;r−) is also a closed set.
D(a;r−)∩D(b;s−)=∅⟺D(a;r−)⊂D(b;s−)
or D(a;r−)⊃D(b;s−)
Proof.
Observe the following, we can rewrite x∈D(a;r−) as,
∣x−a∣p<r∣x−a∣p=(∣x−b+b−a∣p)≤max(∣x−b∣p,∣b−a∣p)<r We have that max(∣x−b∣p,∣b−a∣p) is contained
D(b;r−), but then ∣x−a∣p≤max(∣x−b∣p,∣b−a∣p), thus
D(a;r−)⊂D(b;r−), and since it is given that
b∈D(a;r−) we also have D(b;r−)⊃D(a;r−).
Hence, D(a;r−)=D(b;r−) as claimed.
By definition, D(a;r−) is an open set. We will show that it is
also a closed set. Pick a boundary point in D(a;r), and call it
x, and also choose s≤r. Since x is a boundary point, we
have D(a;r)∩D(x;s)=∅, then
∃y∈D(a;r)∩D(x,s), this means that ∣y−a∣<r
and ∣y−x∣<s≤r. Using the non-archimedean inequality, we
have: ∣x−a∣<max(∣x−y∣,∣y−a∣)<max(s,r)≤r thus
x∈D(a;r) such that D(a;r) contains each of its boundary
points, making D(a;r) a closed set by definition.
Assume W.L.O.G that r≤s. If the intersection is non-empty then
there exists a c in D(a;r)∩D(b;s). Then we know, from
(i), that D(a;r)=D(c;r) and D(b;s)=D(c;s). Hence
D(a;r)=D(c;r)⊂D(c;s)=D(b;s)
◻
Now to begin the proof, we define the preimage of D(y;s−) under
f as,
f−1(D(y;s−))={a∈D(0;r−)∣f(a)∈D(y;s−)}
For a sketch-proof, when these set of points in D(0;r−) that is in
the preimage of f are open then f is continuous. Now, fix an element
of the preimage of D(y;s−) under f, and call it t such that
∣t∣p<r. By definition, f(t) converges in D(y;s−). By
Proposition 16, it follows that ∣antn∣ converges to 0 in
D(y;s−). We have that ∣antn∣p is a Cauchy sequence. Since
an∈Qp, it is Cauchy, then we have
∣antn∣p=∣an∣p∣tn∣p<ϵ⋅∣tn∣p. For which it
follows that (tn) is Cauchy since we have ∣amtm∣p<ϵ
for every m>M (for a fixed large M). Then it follows that t converges
in D(y;s−). Next, we have that t∈D(y;s−), and also
t∈D(0;r−) then D(y;s−)∩D(0;r−)={t}. By
Lemma 4, we know that
D(0;r−)=D(t;r−)⊂D(t;s−)=D(y;s−). Then we
have a union of open disks which are the preimage of D(y;s−) under
f, f−1(D(y;s−))=∣t∣p<r⋃D(t;r−) Hence it
follows that f is continuous as claimed. ◻
Remarks 1. The characterization all power series
f(x)∈Qp[[x]] such that f(x) converges at every point
of the closed unit disk D(0;1) are as follows:
The function f:D(0;r−)→D(0;1−) must be continuous.
Given t∈D(0;r−), the sequence of coefficient of f,
given by (an) satisfy that ∣antn∣p converges to 0 in
D(0;1).
For a power series to converge to 1, it needs f(0)=1. Since
there exists g∈Qp[[x]] such that f⋅g=1 for
which multiplication is closed in Qp[[x]]. In other
words, f(x)∈1+xQp[[x]].
We define the exponentiation of x in the p-adics to be the following
element of Qp[[x]]:
exp(x)=n≥0∑n!xn We see that the exp
function has the following properties:
Proposition 21. We have exp(x+y)=exp(x)⋅exp(y) in
Qp[[x,y]].
Proof. We have
exp(x)=1+x+2!x2+3!x3+⋯ and
exp(y)=1+x+2!y2+3!y3+⋯ Thus,
exp(x)⋅exp(y)=(1+x+2!x2+⋯)(1+y+2!y2+⋯)
Expanding the above product and combining all terms with the same
degree, we get:
exp(x)⋅exp(y)=1+(x+y)+(2!x2+xy+2!y2)+⋯+k=0∑n(k!(n−k)!xkyn−k)+⋯
Writing the above as a summation with respect to n, we get:
exp(x)⋅exp(y)=n≥0∑(k=0∑nk!(n−k)!xkyn−k)=n≥0∑(n!∑k=0n(kn)xkyn−k)=n≥0∑n!(x+y)n=exp(x+y)
by the binomial theorem, and we are done. Hence, we have a homomorphism
ϕ:Qp2→Qp such that exp(x) satisfies
ϕ(x∘y)=ϕ(x)∘ϕ(y) ◻
Proposition 22. The radius of convergence of exp(x) is
p−p−11
Proof. We know that the radius, r, of convergence of any power
series in the p-adics is given by
r=(limsup∣an∣p1/n)−1. Therefore, we
know that the radius of convergence, r of exp(x) is given by
(limsupn!1p)−1. We need to
therefore prove that
(limsupn!1p)−1=p−1/(p−1)
In order to do so, we need the following lemma
Lemma 5. vp(n!)=p−1n−Sp(n) (Sp(n) is the sum of
all digits of n over base p)
Proof. Notice that
vp(n!)=vp(n)+vp(n−1)+⋯=∑l≤nvp(l). We take
vp(l) (l≤n), and expand l over base p. Take
l=lmpm+⋯+lrpr (m≤r, lm=0), where we have
vp(l)=m. Using telescoping techniques, observe that:
−1=(p−1)+(p−1)p+(p−1)p2+⋯+(p−1)pm−pml−1=(p−1)+(p−1)p+(p−1)p2+⋯+(p−1)pm−1+(lm−1)pm+⋯+lrpr We have that the sum of the digits (over base p) of
l−1 is, Sp(l−1)=m(p−1)+Sp(l)−1 The reason why there
is −1 on the RHS since we have lm−1 from the previous equation,
thus it follows we have Sp(l)−1. We know that vp(l)=m, then
solving for m, we have
m=p−11[Sp(l−1)−Sp(l)+1] Then we
have,
vp(n!)=l≤n∑vp(l)=p−11l≤n∑[Sp(l−1)−Sp(l)+1]
Since this is a telescoping series, we have that:
vp(n!)=p−11(−Sp(n)+n)=p−1n−Sp(n) ◻
Lemma 6. limsup((vp(n!)/n) converges to 1/(p−1).
Proof. Consider some non-negative integer k. Consider the sequence
of all reals vp(n!)/n where n is so that pk≤n<pk+1.
The maximum value of this sequence occurs when n=pk. Thus, the
supremum of the sequence vp(n!)/n when pk≤n<pk+1 is
vp((pk)!)/pk. By Legendre’s formula
vp(n!)=⌊pn⌋+⌊p2n⌋+⋯
When n=pk, this simplifies to:
vp(n!)=pk−1+⋯+1=p−1pk−1 Thus,
nvp(n!)=pk(p−1)pk−1=p−11⋅pkpk−1
The above gives the maximum value of vp(n!) for n between pk and
pk+1. Thus, one may conclude that
limsup(vp(n!)/n)=k→∞lim(p−11⋅pkpk−1)
The 1/(p−1) term is constant. Moreover, the sequence
(p−1)pxpx−1 where x is a real number converges to the
same real number as (pk−1)/(pk(p−1)) where k is an integer since
Z∈R. Therefore
k→∞limp−11⋅pkpk−1=x→∞limp−11⋅pxpx−1=p−11
Hence limsup(vp(n!)/n)=1/(p−1). ◻
Now, our goal is to find limsup of the sequence
∣1/n!∣p1/n. The sequence
(∣1/n!∣p1/n) can be rewritten as
(p−vp(1/n!)/n)=(pvp(n!)/n) Now,
limsup(pvp(n!)/n)=plimsup(vp(n!)/n) since
pk is a strictly increasing function with respect to k. By our
lemma, we therefore have
limsup(∣1/n!∣p1/n)=p1/(p−1) Thus,
r=p−1/(p−1) and we are done. ◻
Proposition 23. For all
a,b∈D(0;(p−1/(p−1))−), we have
a+b∈D(0;(p−1/(p−1))−). Therefore, we have
exp(a+b)=exp(a)⋅exp(b).
Proof. By definition, from
a,b∈D(0;(p−1/(p−1))−) we have
∣a∣p,∣b∣p<p−1/(p−1).
Thus, we have
max(∣a∣p,∣b∣p)<p−1/(p−1).
Hence,
∣a+b∣p≤max∣a∣p,∣b∣p<p−1/(p−1).
Thus, we have a+b∈D(0;(p−1/(p−1))−).
Now, we know
exp(a)=n→∞lim(k=0∑nk!ak)
and
exp(b)=n→∞lim(k=0∑nk!bk)
Thus,
exp(a)⋅exp(b)=(n→∞lim(k=0∑nk!ak))⋅(n→∞lim(k=0∑nk!bk))=n→∞lim(k=0∑nk!ak⋅k=0∑nk!bk)
We can expand the product of the summations and combine the terms of the
same degree to get:
k=0∑nk!ak⋅k=0∑nk!bk=k=0∑nk!(a+b)k+fn(a,b)
where fn(x,y) is some polynomial in Qp[x,y] such that
the smallest degree of the terms is n+1. Therefore,
n→∞lim(k=0∑nk!ak⋅k=0∑nk!bk)=n→=∞lim(k=0∑nk!(a+b)k)+n→∞limfn(a,b)=exp(a+b)+n→∞limfn(a,b)
by definition of exp(a+b). We know that the degree of the term with
lowest degree in f(x,y) is n+1. Moreover, we have
∣x∣p,∣y∣p<p−1/(p−1).
Assume WLOG that
∣x∣p≤∣y∣p<p−1/(p−1).
Hence,
xkyn+1−kp<yn+1p<p−n/(p−1)
for any k. Moreover, for any term in f(x,y) with the power of x
and y being s,t respectively, we must have
∣xsyt∣p<p−n/(p−1) since s+t≥n+1 for
f(x,y). But, each term of f(x,y) also has coefficients. We know
that the coefficients of f(x) are at least (1/n!)2. Thus, the
p-adic absolute value of a term of f(x,y) is at most
p2vp(n!)p−n/(p−1) Now, we know that vp(n!)≤1/(p−1) as
n goes to infinity as we saw earlier. Thus, the maximum value of the
p-adic absolute value of the individual terms of f(x,y) is
pp−12−n as n approaches ∞. As n approaches
infinity, the above approaches 0. Thus,
limn→∞∣fn(a,b)∣p=0. Hence,
fn(a,b) itself approaches 0 in Qp as we have seen
earlier. Hence, exp(a+b)=exp(a)exp(b). ◻
Corollary 1. exp(na)=exp(an) for all integers n and
a∈D(0;(p−1/(p−1))−).
Proposition 24. We have
∣exp(x)−exp(y)∣p=∣x−y∣p.
Proof. We first prove that the statement is true when y=0. When
y=0, we have exp(y)=0. Therefore, we need to prove that
∣exp(x)−1∣p=∣x∣p.
We have exp(x)−1=n≥1∑n!xn As we saw earlier,
we have
∣exp(x)−1∣p≤max(n!xnp)n≥1
We claim that the maximum absolute value is
∣x∣p. In order to do so, let us suppose for
the sake of contradiction that there is some term xn/n! with an
absolute value that is more than or equal to the absolute value of x.
So, we have
∣xn/n!∣p≥∣x∣p,
which happens if and only if
∣x∣pn−1∣1/n!∣p≥1
or
∣x∣pn−1≥∣n!∣p.
By the definition of the p-adic absolute value, this happens if and only if
p−(n−1)vp(x)≥p−vp(n!)⟺−(n−1)vp(x)≥−vp(n!)⟺(n−1)vp(x)≤vp(n!)⟺vp(n!)/(n−1)≥vp(x)
Now, ∣x∣p<p−1/(p−1).
Therefore, vp(x)>1/(p−1). Thus, we finally get
vp(n!)/(n−1)>1/(p−1) which is a contradiction. Therefore, we must
have that ∣x∣p is the unique maximum value in
the sequence ∣xn/n!∣p. Since the maximum
value is unique, we have
n≥1∑n!xnp=max(n!xnp)n≥1=∣x∣p
This proves that
∣exp(x)−1∣p=∣x∣p.
Now, consider any y in D(0;(p−1/(p−1))−).
Therefore, we have
∣exp(x)−exp(y)∣p=∣(exp(x)−1)−(exp(y)−1)∣p≤max(∣exp(x)−1∣p,∣exp(y)−1∣p)
Since
∣exp(x)−1∣p=∣x∣p,
we have the above may be written as
max(∣x∣p,∣y∣p). If
∣x∣p=∣y∣p, we
have
∣exp(x)−1∣p=∣exp(y)−1∣p.
Therefore, we have
∣exp(x)−exp(y)∣p=max(∣x∣p,∣y∣p)=∣x−y∣p. ◻
To prove this theorem, we need to prove the following Lemmas first:
Lemma 7. Let f(x)∈1+xQp[[x]] be a power series
with p-adic rational coefficients. Then
f(x)∈1+xZp[[x]]⟺f(x)pf(xp)∈1+pxZp[[x]]
Proof. We will begin with the assumption that
f(x)∈1+xZp[[x]]. We can see that the constant term is
given by,
F(0)=f(0)p−f(0p)=(1+i≥1∑ai0i)p−(1+i≥1∑ai0pi)=1p−1=0
Thus we have that f(0)p=f(0p)=1, then it follows that f(x)p
and f(xp) are both invertible formal power series. Then there exists
t(x)∈1+pxZp[[x]] such that
f(x)pf(xp)=t(x). The reason why
t(x)∈1+pxZp[[x]] is because that the coefficients of
t(x) satisfy a linear recursion that is derived from the product
f(x)p⋅t(x)=f(xp). By construction, the coefficients of
t(x) are an=∑i=1mcian−1 (∃m≥0,
n>0), and (ci) are the coefficients of f(x)p. By the
multinomial theorem, the coefficients (ci) is given by
p!/(r1!r2!r3!(⋯)) such that ∑i≥0ri=p. From
this, we know that the coefficients (except the constant term) of
f(x)p is a multiple of p. Hence it follows that
t(x)∈1+pxZp[[x]] exists. Now, supposed that
f(xp)=f(x)p⋅g(x) with g(x)∈1+pxZp[[x]]. Let
f(x)=∑n≥0anxn, g(x)=∑n≥0bnxn. By
assumption, a0=1, and suppose that we have the required integrality
for an with 0≤n<N−1. We will show that the Nth coefficient
of f(x)p⋅g(x) is equal to the Nth coefficient of
(∑n≤Nanxn)p+∑n≤Nbnxn for which this sum
is just a redefinition of f(xp) if we take N→∞. The proof
will proceed by induction. To begin, let’s expand these power series to
get a sense of the terms,
f(x)p=(i=0∑N−1aixi+aNxN+k≥N+1∑akxk)p=t=0∑p(tp)(i=0∑N−1aixi+k≥N+1∑akxk)p−t(aNxN)t(n≤N∑anxn)p=i=0∑p(ip)(n≤N−1∑anxn)p−i(aNxN)i
We have two cases, p∣N and p∣N. Suppose
p∣N, then there is no m∈Z such that N=pm
(that would give us another Nth coefficient which is
apNp). Thus we have that the Nth coefficient of
(∑n≤Nanxn)p+∑n≤Nbnxn is
(1p)aNa0+bN=p(1)(aN)+bN=paN+bN, while we have
that the Nth coefficient of f(x)p⋅g(x) is
g(0)(1p)aN+a0bN. Since we have assumed that a0=1,
then by the linear recursion definition of the coefficient of g(x), we
have that a0b0=1 then b0=1. It follows that the Nth term of
f(x)p⋅g(x) is (1)paN+(1)bN=paN+bN. The intuition
behind the multiplication of f(x)p⋅g(x) is just termwise
multiplication, thus we can find the pairs such that they’re the Nth
term. Since by definition, g(x)∈1+pZp[[x]], it follows
that bN∈pZp. Now, it follows that we can always
construct the Nth coefficient using the linear-recursion definition of
the coefficient of g(x) then the construction is followed by
induction. Hence both of them have the same Nth coefficient such that
p∣N. Also, we can conclude that aN∈Zp, since
the Nth coefficient is not divisible by p2 but by p only, thus it
follows that the Nth coefficient is in pZp and
aN∈Zp. Now for our second case, suppose that
p∣N, then there exists m∈Z such that N=pm. We
have that the Nth coefficient of
(∑n≤Nanxn)p+∑n≤Nbnxn is
apNp+(1p)aN+bN=apNp+(1)aN+bN=apNp+aN+bN.
To find apNp in (∑n≤Nanxn)p, consider the
pth term and set i=0, observe the following:
Then by the binomial theorem, it follows that we have
apNp as the additional term for the Nth coefficient of
(∑n≤Nanxn)p+∑n≤Nbnxn. While for the
Nth coefficient of f(x)p⋅g(x), we have
apNp+g(0)(1p)aN+a0bN. To find the value of
apNp, consider the pth term and set t=0, observe
the following:
(i=0∑N−1aixi+k≥N+1∑akxk)p=i=0∑(N/p)−1aixi+apNxpN+k≥pN+1∑akxkpi=0∑(N/p)−1aixi+apNxpN+k≥pN+1∑akxkp=i=0∑(N/p)−1aixi+k≥pN+1∑akxk+apNxpNp Then by the binomial theorem, it follows that we have
apNp as the additional term for the Nth coefficient of
f(x)p⋅g(x). Now, it follows that we can always construct the
Nth coefficient using the linear-recursion definition of the
coefficient of g(x) then the construction is followed by induction.
Hence both of them have the same Nth coefficient such that p∣N.
Also in our second case, we can conclude that aN∈Zp,
since the Nth coefficient is not divisible by pp but by p only,
thus it follows that the Nth coefficient is in pZp and
aN∈Zp. Since we have successfully concluded that
aN∈Zp, it follows that
f(x)∈1+xZp[[x]]. ◻
Lemma 8. exp(−px)∈1+pZp[[x]]
Lemma 9. E(x)pE(xp)=exp(−px)
Now we are ready to prove Theorem 2.
Proof. It follows that E(x)∈1+xZp[[x]] by Lemma 7
due to Lemma 9, thus it follows that the coefficients of E(x) is in
Zp by Lemma 7. ◻