Paul Pollack

Unreal Analysis

Visions of the p-adic Realm

October 8, 2024
 In memory of Kurt Hensel, Reinhold Strassmann, and the
countless others persecuted or murdered by the National Socialists.
Preface

A few months ago, on his last visit to In the long history of mathematics,
New Jersey, I was telling [Paul] Erdős ‘number’ has always meant real
something about p-adic analysis. number, and it is only relatively
Erdős was not interested. “You know,” recently that we have become aware
he said about the p-adic numbers, of the world of p-adic numbers. The
“they don’t really exist.” situation is akin to someone who has
only experienced daylight gazing in
Melvyn Nathanson astonishment at the night sky.

Kazuya Kato
Nobushige Kurokawa
Takeshi Saito

p-adic numbers were birthed into the world by Kurt Hensel in 1897, at a lecture
for the annual meeting of the German Mathematical Society. Initially, these
new objects were viewed with some skepticism. Helmut Hasse recollects that in
1920, no less a figure than Richard Courant advised him against studying with
Hensel, dismissing Hensel’s 1913 book on p-adic numbers as an unproductive
detour (“unfruchtbarer Seitenweg”). Yet today p-adic numbers have thoroughly
permeated number theory. Local-global principles, which relate solubility in
number fields to solubility in p-adic fields, are central to modern arithmetic
geometry. Tate’s thesis and the adelic viewpoint on class field theory are bread
and butter to algebraic number theorists. And those on the analytic side of
the fence can point to ‘local densities’ as p-adically motivated objects that
arise in nearly every study of arithmetic counting problems. We live in an age
when activists and arithmeticians universally agree on the need to think (and
act) both globally and locally.
These problem sets, written for a topics course at the 2024 Ross Indiana
summer mathematics camp, do not describe any of these significant modern
applications of p-adic numbers. Rather, they offer a hands-on — better, minds-
on! — approach to the foundational theory of Qp , working towards elementary
but attractive number-theoretic applications. Whenever there was a choice
to be made, I have elected to treat special cases rather than develop general
theory. This has allowed prerequisites to be kept to a minimum. Readers

I
are expected to have seen number theory and rigorous calculus, as well as
both abstract and matrix algebra, but most of what is needed belongs to the
standard undergraduate curriculum.
Notwithstanding the chosen title, these problem sets aim to offer a general in-
troduction to p-adic numbers, privileging neither analysis nor algebra. Analysts
will be disappointed by the absence of material on continuity, differentiability,
and p-adic interpolation. Algebraists will be disheartened that the discussion
never ventures beyond Qp to any of its finite or infinite extensions. Despite
the many omissions, I am optimistic that as individuals engage with these
problem sets, they will catch a glimpse of an alluring territory begging to
be charted. The suggested readings listed below have been selected as partic-
ularly approachable, and we commend their consideration to those with an
adventurous spirit.

How to use this book

This manuscript is intended as a resource for undergraduates and beginning

graduate students looking to dip their toes into non-Archimedean waters.
For the teacher using the text as a resource for a class on p-adic numbers:
First of all, congratulations! As you are probably aware, courses on this topic
at the specified level are rare.
If the book is used as the primary course text, problems should be distributed
to students independently of solutions. Class time can be spent discussing
ideas and approaches, with conversations led by students but facilitated by
the instructor. Experts will often see more in the problems than beginning
students and they are encouraged to use these discussions to share insights.
Solutions to Problem Set (p-Set) X can be passed out once the class, as a
whole (or at least, on the whole), is ready to move on to Set X + 1. For those
on the semester system, a reasonable aim is to cover roughly one set per week,
with the expectation that later material will take a bit more time. The book
can also be used in a supplementary fashion, with an instructor handpicking
interesting-seeming problems to spice up their own exercise sheets.
For students: Congratulations to you as well! p-adic numbers are a(n) (un)real
treat to think about. If you are using the book for a course, follow the
directions of your instructor. If you are using the book for self-study, I strongly
recommend working through the problems systematically, tackling one set
completely before moving on to the next. Please give yourself enough time
for the mental fermentation process to occur before giving up on a question!
Of course, you should not feel bad if after a long struggle certain problems
continue to elude you; solutions are included for precisely these moments.

II
I should emphasize that this is not meant to be your only book on p-adic
numbers. The author, for instance, learned about p-adic numbers from entirely
different sources! After mastering the material on each problem set, it is
recommended that you look at how the corresponding content is treated in
the suggested references. One only truly understands a topic after examining
it from all angles, and different texts bring out different perspectives.

Saying what we mean and meaning what we say

Exercises are typically introduced as assertions, without the use of “show that”
or “prove.” (The Extra Explorations, embedded in the solutions, are exceptions.)
Problem solvers should recognize they are on the hook to validate every claim.
We take no position on the hotly contested issue of whether 0 is a natural
number. Integers greater than or equal to 0 are — naturally enough — referred
to as nonnegative integers. As is customary in English, positive means strictly
larger than zero. We write Z+ , Q+ , and R+ for the sets of positive integers,
rational numbers, and real numbers, respectively.
Our terminology around quotient rings is slightly unconventional. To start
with, the term ring always refers to a commutative ring with unity. If I is an
ideal of the ring R, and a ∈ R, the class of a with respect to congruence modulo
I is typically denoted “a mod I.” We break this rule when (and only when)
R = Z and I = mZ, instead writing “a mod m.” Additionally, we use “Z/m”
in place of “Z/mZ”; this is a mild concession to the notation Zm appearing
on the first-year Ross Program problem sets.

Acknowledgements

Assembling these problem sets involved much scouring and borrowing. De-
serving of particular mention: Exercise 4.53 was sourced from an article by
Catherine Crompton in the Rose Hulman Undergrad. Mathematics Journal.
Exercises 7.90 and 9.117 were inspired by discussions in Murty’s text [6].
Exercise 11.132 was taken from Koblitz’s book [5]. The proof of Ramanujan’s
conjecture on integer solutions to x2 + 7 = 2m follows closely the exposition in
Cassels’s monograph [1]. The proof “from-the-Book” of Skolem’s theorem on
integer linear recurrences, outlined on Set #11, is due to Tao (loosely based
on an argument of Georges Hansel):
https://terrytao.wordpress.com/2007/05/25/open-question-e
ffective-skolem-mahler-lech-theorem/
In this connection it was also helpful to consult a post of Seewoo Lee:
https://seewoo5.github.io/jekyll/update/2023/02/21/p-adi
c-numbers-application.html

III
 Exercises 9.112 and 10.129 are based on a paper of Mahler:
On some irrational decimal fractions. J. Number Theory 13 (1981),
268–269.
The proofs on Set #13 of the Adams and Kummer theorems on Bernoulli
numbers are adapted from a delightful article of Wells Johnson:
p-adic proofs of congruences for the Bernoulli numbers. J. Number
Theory 7 (1975), 251–265.
Various tech tools were employed to produce the book in front of you. The
cover, which features 19th-century illustrations from the Hirayama Fireworks
company∗ , was designed in Canva. Typesetting was done with LATEX, using
Springer’s SVMono document class. Several calculations were farmed out to
the fantastically capable PARI/GP. ChatGPT assisted by offering English
translations, hunting down typos, and suggesting alternative phrasings.
This manuscript would not exist without the backing and encouragement
of the Ross Mathematics Foundation. Site directors Timothy All and Jim
Fowler have my profound appreciation for their inspiring efforts, year after
year, to ensure that the Ross Program provides a supportive, welcoming, and
stimulating environment for all camp participants (students, counselors, and
even us lecturers!). Special thanks to Phoebe Watkins for her service as course
assistant, Paco Adajar for suggesting the term “p-set” as an alternative to the
more mundane “problem set,” and Jacob Bucciarelli for sharing his expertise
in orbital mechanics.
Work on this project was facilitated by a grant from the US National Science
Foundation, award DMS-2001581.

Suggestions for further reading

1. J. W. S. Cassels, Local fields, London Mathematical Society Student Texts, vol. 3,

Cambridge University Press, Cambridge, 1986.
2. K. Conrad, Expository papers, https://kconrad.math.uconn.edu/blurbs/.
3. F. Q. Gouvêa, p-adic numbers: An introduction, third ed., Universitext, Springer,
Cham, 2020.
4. S. Katok, p-adic analysis compared with real, Student Mathematical Library,
vol. 37, American Mathematical Society, Providence, RI; Mathematics Advanced
Study Semesters, University Park, PA, 2007.
5. N. Koblitz, p-adic numbers, p-adic analysis, and zeta-functions, second ed.,
Graduate Texts in Mathematics, vol. 58, Springer-Verlag, New York, 1984.
6. M. R. Murty, Introduction to p-adic analytic number theory, AMS/IP Studies in
Advanced Mathematics, vol. 27, Amrican Mathematical Society, Providence, RI;
International Press, Somerville, MA, 2002.
∗
see https://www.city.yokohama.lg.jp/kurashi/kyodo-manabi/library/shuroku/
hirayama.html

IV
 7. W. H. Schikhof, Ultrametric calculus: An introduction to p-adic analysis, Cam-
bridge Studies in Advanced Mathematics, vol. 4, Cambridge University Press,
Cambridge, 2006.
8. J.-P. Serre, A course in arithmetic, Graduate Texts in Mathematics, vol. 7,
Springer-Verlag, New York-Heidelberg, 1973.
9. J. Steuding, Die p-adischen Zahlen. Online survey article. URL: https://www.
uni-marburg.de/de/fb12/fachbereich/profil/geschichte-des-fachberei
chs/biographisches/hensel_p_adischen_zahlen.pdf

University of Georgia Paul Pollack

V
VI
 Part I

Problems

1
2
 Problem Set #1

Valuation Theory (the Theory of Absolute Values)

. . . true faith is belief in the reality of

absolute values. It is in this kingdom
of absolute values that we must look
for and find our immortality.

William R. Inge

Let K be a field. An absolute value on K is a function | · | : K → R satisfying

(i) |x| ≥ 0, with |x| = 0 if and only if x = 0,
(ii) |x + y| ≤ |x| + |y| (the triangle inequality),
(iii) |xy| = |x||y|.
for all x, y ∈ K. We refer to the pair (K, | · |) as a valued field.

1.1. Let (K, | · |) be a valued field. Then:

(a) | ± 1| = 1,
(b) |x − y| ≥ |x| − |y| for all x, y ∈ K,
(c) |xy −1 | = |x||y|−1 for all x, y ∈ K with y ̸= 0.

1.2. Let K be a field. Define |x| by letting |x| = 0 if x = 0 and |x| = 1 for all
x ̸= 0. Show that | · | is an absolute value on K (the trivial absolute value).

Let p be a prime number. For each x ∈ Q× , there is a unique integer v with

the property that x = pv ab for some integers a, b not divisible by p. We set
vp (x) = v. In order to have vp defined on all of Q, we take vp (0) = ∞. The
function vp : Q → Z ∪ {∞} is called the p-adic valuation.

1.3. Let p be a prime. For each x ∈ Q, define |x|p = p−vp (x) , where p−∞ is
taken to be 0. Then | · |p is an absolute value on Q (the p-adic absolute value).

3
The absolute value | · |p will play a starring role throughout the course. You
are strongly advised to compute several examples to develop a feel for this
notion of absolute value. Here are a few to get you started:

5 · 210 2
= 2−10 , |3−5 |2 = 1, = 3, |100!|7 = 7−16 .
210 + 1 2 21 3

If | · | is an absolute value on the field K, we call | · | non-Archimedean if it

satisfies the following strong triangle inequality:

|x + y| ≤ max{|x|, |y|} for all x, y ∈ K.

Sensibly enough, an absolute value that is not non-Archimedean is Archimedean.

For example, the trivial absolute value is non-Archimedean, while the usual
absolute value on Q is Archimedean.

1.4. For each prime p, the absolute value | · |p on Q defined in Exercise 1.3 is
non-Archimedean.

The following Darwinian property of non-Archimedean absolute values (“sur-

vival of the greatest”) is central to the theory.
1.5. If | · | is a non-Archimedean absolute value on a field K, then

|x + y| = max{|x|, |y|} whenever x, y ∈ K with |x| =

̸ |y|.

1.6. Let K be a field equipped with an absolute value | · | for which |2| ≤ 1.
Then |2e1 + 2e2
+ · · · + 2en | ≤ n for all nonnegative integers e1 , . . . , en . It
follows that | nk | ≤ n whenever n is a positive integer and 0 ≤ k ≤ n.

1.7 (Product Formula). Let |·|∞ denote the standard Archimedean absolute
value on Q. For every x ∈ Q× ,
Y
|x|∞ |x|p = 1.
p prime

Factorials and Factorization

1.8. For each prime p and positive integer n: |n!|p > p−n/(p−1) .

1.9. |n!|∞ grows faster than C n for any fixed C. Hence, the product formula
implies that there are infinitely many primes.

4
‘p’s and Harmony
1.10. Let Hn = 1 + 12 + · · · + n1 , the nth harmonic number. From calculus,
|Hn |∞ tends to infinity. The same holds for |Hn |2 .

1
P
1.11 (Wolstenholme). When p is odd, Hp−1 = p 0<i<p/2 i(p−i) . Hence,
|Hp−1 |p ≤ p−2 for p > 3.

5
6
p-Set #2

Distance Learning

Today we are far from the standpoint

of viewing the measure or size of a
number or geometric figure as
something given by nature and
necessity. We view the size of a figure
or number rather as a function of its
essential components, whose
determination is entirely at our
discretion, and in whose choice we let
ourselves be guided by reasons of
expediency.

Kurt Hensel

In R it is often useful to view |x − y| as the distance between x and y. One

can adopt the same interpretation for any absolute value on any field K, but
only if one is prepared to welcome a few surprises!

2.12. If | · | is non-Archimedean, every triangle with vertices in K is isosceles.

Let (K, | · |) be a valued field. For x0 ∈ K and real r > 0, the open and closed
discs of radius r centered at x0 are defined, respectively, by
D<r (x0 ) = {x ∈ K : |x − x0 | < r} and D≤r (x0 ) = {x ∈ K : |x − x0 | ≤ r}.

2.13. Suppose | · | is non-Archimedean. Let D = D<r (x0 ) be an open disc in

K. Show that if x ∈ D, then D = D<r (x). That is: Every point of an open
disc is a center.

2.14. Suppose | · | is non-Archimedean. Then any two open discs are either
disjoint or one contains the other.

2.15. Let K = Q and | · | = | · |p with p prime. Every open disc is a closed

disc, and vice versa.

7
Valuation Theory
2.16. How many absolute values can you find on F2027 ? (Here and below, Fp
denotes the field with p elements, or equivalently the residue ring Z/p.)

2.17. Let (K, | · |) be a valued field. For all x, y ∈ K and all n ∈ Z+ ,

  1/n !
1/n n
|x + y| ≤ (n + 1) max max{|x|, |y|}.
0≤k≤n k

Let Fp (T ) denote the fraction field of the polynomial ring Fp [T ], so that

 
F (T )
Fp (T ) = : F (T ), G(T ) ∈ Fp [T ], G(T ) ̸= 0 .
G(T )

2.18. Every absolute value on Fp (T ) is non-Archimedean.

2.19. If π(T ), π̃(T ) are distinct monic irreducibles in Fp [T ], then there is an

absolute value | · | on Fp (T ) with |π(T )| < 1 and |π̃(T )| = 1. Is there an
absolute value on Fp (T ) with |T | > 1?

2.20. If K is a field equipped with a non-Archimedean absolute value | · |, then

D≤1 (0) = {x ∈ K : |x| ≤ 1} is a ring. When K = Q and | · | = | · |p , this ring
is denoted Z(p) and called the ring of p-integral rational numbers. Explain in
pedestrian terms what it means for a rational number to be p-integral.

T
2.21. Determine p prime Z(p) .

‘p’s and Harmony

k−1
2.22. For every prime p and each integer 0 < k < p: kp = kp k−1
p−1
≡ p· (−1)k

(mod p2 ). (Part of the problem is to make sense of the congruence, since

k−1
p · (−1)k ∈
/ Z if 1 < k < p.)

2.23 (Eisenstein). If p is an odd prime, then |1 − 12 + 13 − 14 + · · · − p−1

1
|p < 1
p 2
⇐⇒ 2 ≡ 2 (mod p ).
When 2p ≡ 2 (mod p2 ), we call p a Wieferich prime, alluding to the appearance of such primes in
Wieferich’s early 20th century investigations into Fermat’s last theorem. 1093 and 3511 are the
only known Wieferich primes. Other examples (if they exist) exceed 1019 .

8
Variations on a Theme of Euclid
2.24 (Schur). Let F (T ) ∈ Z[T ] be nonconstant. There are infinitely many
primes that divide F (n) for some n ∈ Z. In other words: F has a root in Z/p
for infinitely many primes p.

2.25. If p | n4 + 1 for some n ∈ Z, either p = 2 or p ≡ 1 (mod 8). Hence, there

are infinitely many primes p ≡ 1 (mod 8).

2.26. Let n ∈ Z+ . If p | (2n + 1)2 − 2, then p ≡ ±1 (mod 8). Not every prime
dividing (2n + 1)2 − 2 can be congruent to 1 (mod 8). Hence, (2n + 1)2 − 2 is
always divisible by some prime congruent to −1 (mod 8). By varying n, we
can find infinitely many primes p ≡ −1 (mod 8).

2.27. Every coprime residue class mod 8 contains infinitely many primes.
In this last statement, “mod 8” can be replaced with “mod m,” for any m ∈ Z+ . This is a
celebrated (and difficult!) theorem of Dirichlet that you will meet in courses on analytic number
theory.

9
10
p-Set #3

Knowing Your Limits

One cannot blame a respectable

mathematician for looking twice at
the equation

−1 = 4 + 4 · 5 + 4 · 52 + 4 · 53 + . . . .

. . . It is obvious that [this] is absurd if

ordinary convergence is intended. The
whole point to Hensel’s theory is that
this is not ordinary convergence, but
a new type of convergence which,
from the point of view of abstract
algebra, is equally worthy of the
name.

Cyrus Colton MacDuffee

Let (K, | · |) be a valued field, and let {xn } be a sequence of elements of

K. Suppose x ∈ K. We say {xn } converges to x, and write xn → x or
limn→∞ xn = x or simply lim xn = x, if the following holds.

For every real number ϵ > 0, there is an N ∈ Z+ with the property that

|xn − x| < ϵ whenever n ≥ N.

(The inequality “|xn − x| < ϵ” could have been written as “xn ∈ D<ϵ (x).”)
We say {xn } converges if it converges to some x in K. Those who already
grok convergence in the context of calculus will notice that xn → x in (K, | · |)
precisely when |xn − x| → 0 in the familiar sense.

3.28. Every sequence in a valued field has at most one limit.

3.29 (calculus classics).

11
 (i) If xn = x for all n, then xn → x.
(ii) If xn → x and yn → y, then xn + yn → x + y.
(iii) If xn → x and yn → y, then xn yn → xy.

3.30. xn = 1 + 3 + 32 + · · · + 3n converges to − 12 in
P∞ (Q, | · |3 ). So defining the
value of a series as the limit of its partial sums, k=0 3k = − 12 . Does {xn }
converge in (Q, | · |p ) for any other values of p?
P∞ P∞
3.31. Evaluate n=0 n · n! and n=0 n2 · 2n in (Q, | · |2 ).

3.32. Explain how the calculations

11 −1
=2+3·
7 7 −6 −2
=0+3·
−1 −5 7 7
=2+3·
7 7 −2 −3
=1+3·
−5 −4 7 7
=1+3·
7 7 −3 −1
=0+3·
−4 −6 7 7
=2+3·
7 7

imply that in (Q, | · |3 ),

11
= 2 + 2 · 3 + 1 · 32 + 2 · 33 + 0 · 34 + 1 · 35 + 0 · 36 + 2 · 37 + 1 · 38 + . . . ,
7
where the “digits” in the right-hand expansion follow the eventually periodic
pattern 2, 2, 1, 2, 0, 1, 0.
P∞ k 2
3.33. Find c0 , c1 , c2 , . . . ∈ {0, 1, 2, 3, 4} with k=0 ck 5 = 7 in (Q, | · |5 ).

Valuation Theory
3.34. Let (K, | · |) be a valued field. Then | · | is non-Archimedean ⇐⇒ |2| ≤ 1.
(Here 2 means 1 + 1.) Is this equivalence true with 3 in place of 2?

3.35. If | · | is a nontrivial non-Archimedean absolute value on Q, then |p| < 1

for some prime p.

3.36. Let | · | be a non-Archimedean absolute value on Q. If m and n are

relatively prime integers, either |m| = 1 or |n| = 1. (Use Bézout!)

3.37. Let | · | be a non-Archimedean absolute value on K and O = D≤1 (0).

We have seen in Problem 2.20 that O is a subring of K.
Show that x ∈ O is a unit in O ⇐⇒ |x| = 1. Furthermore, if M is the collection
of all nonunits of O — that is, M = D<1 (0) — then M is a proper ideal of
O containing every proper ideal of O. Hence, M is the unique maximal ideal
of O.

12
 m 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
um 0 1 1 −1 −3 −1 5 7 −3 −17 −11 23 45 −1 −91 −89 93

Sequence {um }m≥0 appearing in Exercise 3.40.

3.38. Every nonzero element of Z(p) is uniquely expressible in the form pv u

where v is a nonnegative integer and u is a unit of Z(p) .

3.39. Z(p) is a principal ideal domain (PID) with pZ(p) its only maximal ideal.

A Question of Ramanujan
When is a power of 2 equal to 7 more than a square? This happens for
23 , 24 , 25 , 27 , and 215 (in this last case, 215 = 32 768 = 7 + 1812 ). In a 1913
issue of the Journal of the Indian Mathematical Society, Ramanujan listed all
of these examples and posed the problem of finding others.
√ √ √
Let R = Z[ 1+ 2 −7 ] (viewed as a subring of C). Since ( 1+ 2 −7 )2 = 1+ −7
2 − 2,
√
√
 
1+ −7 1
R=Z+Z = (a + b −7) : a, b ∈ Z and a ≡ b (mod 2) .
2 2

3.40 (Nagell). Assume as known that R is a unique factorization domain

whose only units are ±1. Show that if x, m are integers with x2 + 7 = 2m , then
um−2 = ±1, where um is the sequence defined by the Binet-type formula
√ √
αm − β m 1 + −7 1 − −7
um = for α = , β= , m = 0, 1, 2, 3, . . . .
α−β 2 2
Does the converse hold?

On our final problem set (Set #13) you will determine all solutions to um−2 = ±1 by p-adic
methods.

13
14
 p-Set #4

Berning Questions

. . . It took me less than half a quarter of an

hour to find that the tenth powers of the first
1000 numbers, starting from 1, being added
together make

91 409 924 241 424 243 424 241 924 242 500.

This renders apparent the futility of the work

Ismaël Boulliau spent on the compilation of
his voluminous Arithmetica Infinitorum, in
which he did nothing more than laboriously
compute the sums of the first six powers. . .

Jacob Bernoulli

The Bernoulli numbers Bk (k = 0, 1, 2, 3, . . . ) are defined as the coefficients in

the formal power series∗ expansion
∞
T X Tk
T
= Bk .
e −1 k!
k=0

Equivalently, B0 , B1 , B2 , . . . are determined by the identity

T2 T3 T4 T2 T3
  
T = T+ + + + ... B0 + B1 T + B2 + B3 + ... .
2! 3! 4! 2! 3!

T 2
Since e T−1 = 1 + 2!
T
+ T3! + . . . is a power series with rational coefficients and
nonzero constant term, its reciprocal eTT−1 is also a power series with rational
coefficients. Therefore, every Bk ∈ Q.
For positive integers n and k, define Sk (n) = 1k + 2k + · · · + (n − 1)k .
∗
We assume acquaintance with basic facts about formal power series, as found for
instance in §1.1 of Stanley’s Enumerative Combinatorics.

15
 k 0 1 2 3 4 5 6 7 8 9 10 11 12
Bk 1 − 12 1
6
0 1
− 30 0 1
42
0 1
− 30 0 5
66
0 691
− 2730

First several Bernoulli numbers.

P∞ k
4.41. 1 + eT + e2T + · · · + e(n−1)T = k=0 Sk (n) Tk! (as formal power series).

k  
X k nj+1
4.42 (Faulhaber’s Formula). Sk (n) = Bk−j .
j=0
j j+1

In view of the widespread applications of Faulhaber’s formula, the Bernoulli

numbers are natural objects of study. The following two exercises ask you to
pluck some low-hanging fruit.

2T P∞ k
4.43. If coth T := ee2T −1
+1
, then T coth T = T + e2T2T−1 = T + k=0 Bk (2Tk!) (as
formal series). T coth(T ) is invariant under the substitution T 7→ −T . Hence,
Bk = 0 for every odd k > 1.

22k
4.44. coth T = T1 + k≥1 B2k (2k)! d
T 2k−1 and dT coth T = 1 − coth2 T .
P

As a consequence: (−1)k+1 B2k > 0 for each k ∈ Z+ .

Other properties of Bernoulli numbers lie a bit further below the surface.
For instance, based on the above table one might conjecture that Bernoulli
numbers always have squarefree denominators. We will prove this — and much
more — in due time!

Take It to the Limit, One More Time

P∞
4.45. Compute the first several partial sums of k=1 2k /k, noting their 2-adic
absolute values. Any conjectures?

4.46. For every r ∈ Z(p) , there is a unique sequence d0 , d1 , d2 , d3 , . . . ∈

P∞
{0, 1, 2, . . . , p − 1} with k=0 dk pk = r in (Q, | · |p ). The sequence {dk } is
eventually periodic.

4.47. If xn → x in (K, | · |), then |xn | → |x| in the real numbers. In fact, if | · |
is non-Archimedean and xn → x, where x ̸= 0, then |xn | = |x| for all large n.

16
Valuation Theory
4.48. Every nontrivial non-Archimedean absolute value on Q has the form
| · |cp for some prime p and some c > 0.

4.49. If | · | is an Archimedean absolute value on Q, we know from Exercise

3.34 that |2| = 2c and |3| = 3d for some real numbers c, d > 0.
Write 2n in base 3, say 2n = ϵm 3m + ϵm−1 3m−1 + · · · + ϵ0 , where each ϵi = 0, 1,
or 2 and ϵm > 0. There is a positive constant B with

2cn = |2n | ≤ B · |3|m = B · 3dm ;

e.g., B = |2|·|3| m n
|3|−1 works. Since 3 ≤ 2 and n can be taken arbitrarily large, it
must be that c ≤ d. Reversing the roles of 2 and 3 shows d ≤ c, and so c = d.

4.50. If | · | is an Archimedean absolute value on Q, then | · | = | · |c∞ , where

|2| = 2c .

Let | · | and | · |′ be absolute values on K. We say | · | and | · |′ are equivalent

absolute values if they induce the same notion of convergence, meaning that
xn → x in (K, | · |) if and only if xn → x in (K, | · |′ ).

4.51. If | · | and | · |′ are equivalent, then | · | and | · |′ are both Archimedean or

both non-Archimedean.
Suggestion. Look at convergent geometric sequences.

4.52 (Ostrowski’s Theorem). Every nontrivial absolute value on Q is

equivalent to exactly one of | · |p (p prime) or | · |∞ .

4.53. It is easy to find p distinct rational numbers equidistant with respect to

| · |p ; simply consider 0, 1, 2, . . . , p − 1. Can you find p + 1 such numbers?

‘p’s and Harmony

4.54 (Eswarathasan–Levine, Boyd). Recall our notation Hn for the nth

harmonic number, Hn = 1 + 12 + · · · + n1 . If |Hm |p ≥ 1, then |Hn |p = p|Hm |
whenever pm ≤ n < p(m + 1).

4.55. (electronic assistance recommended!) |Hn |3 and |Hn |5 tend to infinity.

Open problem: |Hn |p tends to infinity for every fixed prime p.

17
18
p-Set #5

And Introducing. . .
Let p be a prime number. Q∞A p-adic integer is an infinite tuple (a1 mod p, a2 mod
p2 , a3 mod p3 , . . . ) ∈ k=1 Z/pk that satisfies the compatibility condition

ak+1 ≡ ak (mod pk ) for all positive integers k. (*)

We write Zp for the collection of all p-adic integers.

A p-adic integer is no more and no less than an object which can be sensibly
“reduced” modulo
Q∞ an arbitrary power of p. Initially one might think that all
elements of k=1 Z/pk meet this description: any of these can be “reduced”
k
mod
Q∞ p byk projecting from the kth component. But for a general element of
k=1 Z/p , these reductions need not be compatible. “Compatible” means
commutativity of the diagram

Zp Z/pk+1

a mod pk+1 7→a mod pk

Z/pk

This is precisely what (*) buys us.

Here are some examples of elements of Z5 :
(101 mod 5, 101 mod 52 , 101 mod 53 , . . . ) = (1 mod 5, 1 mod 52 , 101 mod 53 , . . . ),
(−1 mod 5, −1 mod 52 , −1 mod 53 , . . . ) = (4 mod 5, 24 mod 52 , 124 mod 53 , . . . ),
(4 mod 5, 34 mod 52 , 334 mod 53 , . . . ) = (4 mod 5, 9 mod 52 , 84 mod 53 , . . . ).

Take a moment to convince yourself that this last example satisfies our compatibility
condition!

Q∞
5.56. Zp is a subring of k=1 Z/pk .

19
 Figure for Exercise 5.64.

5.57. Zp is an integral domain.

5.58. Zp has characteristic 0. (Thus, Z sits canonically inside Zp .)

Which elements of Q can be said to belong to Zp ? Is our third example of a
5-adic integer an element of Q? (Assume the nth component is 3 . . . 34 mod 5n ,
where 3 is repeated n − 1 times.)

5.59. u = (a1 mod p, a2 mod p2 , . . . ) ∈ Zp is a unit ⇐⇒ p ∤ a1 (in Z) ⇐⇒

p ∤ u (in Zp ).

5.60. Every nonzero element of Zp admits a unique expression in the form

pv u, where v is a nonnegative integer and u is a unit in Zp .

5.61. Zp is a principal ideal domain with pZp its only nonzero prime ideal.

5.62. The canonical inclusion Z ,→ Zp induces an isomorphism Z/pn ∼

=
Zp /pn Zp , for every positive integer n.

5.63. The definition of Zp makes sense without requiring p to be prime.

However, considering
Q Zg for composite g does not give anything essentially
new. In fact: Zg ∼
= p|g Zp for each integer g > 1. (For instance, Z10 ∼
= Z2 ×Z5 .)

5.64. What does the picture on this page — created by TEX StackExchange
user Qrrbrbirlbel∗ — have to do with Z3 ?

5.65 (Steuding). Repeat the last exercise for Salvador Dali’s painting La
Cara de la Guerra.

5.66 (Zp is uncountable). There is no map from Z+ onto Zp .

∗
https://tex.stackexchange.com/a/695157

20
Well, Color Me Impressed!
5.67 (Thomas, Monsky). Consider the following 3-coloring of the rational
plane Q2 : (x, y) is red if |x|2 < 1, |y|2 < 1, blue if |x|2 ≥ 1, |x|2 ≥ |y|2 , and
green if |y|2 ≥ 1, |y|2 > |x|2 . Show that any trio of differently colored points
forms a triangle ∆ with |Area(∆)|2 > 1.
This observation plays a key role in the proof of Monsky’s Theorem: It is impossible to dissect a
square into an odd number of triangles all of which have the same area.

21
22
p-Set #6

Enter Cauchy
Let K be a field equipped with an absolute value | · |. If {xn } converges to x
in (K, | · |), then for any real number ϵ > 0, all terms far enough out in the
sequence {xn } are within 12 ϵ of x. By the triangle inequality, all such terms
are within ϵ of each other. That is:

For each ϵ > 0, there is a positive integer N with

(C)
|xn − xm | < ϵ whenever n, m ≥ N .

Any sequence {xn } with property (C) is called a Cauchy sequence.

6.68. If | · | is non-Archimedean, then {xn } is Cauchy ⇐⇒ |xn+1 − xn | → 0.

This need not hold if | · | is Archimedean; a counterexample is provided by the partial sums of the
harmonic series in (Q, | · |∞ ).

n
6.69. The sequence {25 } is Cauchy in (Q, | · |5 ).

6.70 (a common calculus Cauchy claim). Let (K, | · |) be a valued field. If

{xn } is a Cauchy sequence in K, then {|xn |} is a bounded sequence of real
numbers.

6.71 (and another). Let (K, | · |) be a valued field. If a sequence {xn } of

elements of K is Cauchy, and some subsequence of {xn } converges to x ∈ K,
then {xn } converges to x.

If R is equipped with the usual absolute value, then every Cauchy sequence
in R converges to an element of R. This need not be the case for a general
valued field. For example, the sequence of rational numbers

23
 x1 = 1, x2 = 1.4, x3 = 1.41, x4 = 1.414,
...,
√
obtained by successively truncating the decimal expansion of √2, is Cauchy in
(Q, | · |∞ ) but does not converge to any element of Q, since 2 ∈ / Q. Coping
with this unsettling scenario was one of the motivations behind the invention
(discovery?) of the real numbers in the first place!
Disconcerting examples of this same kind can also be found when K = Q
and | · | = | · |p . Take any sequenceP{ck }k≥0 from {0, 1, . . . , p − 1} that is
n
not eventually periodic. Then xn = k=0 ck pk defines a Cauchy sequence in
(Q, | · |p ). To check the Cauchy condition (C), we may assume that m < n
(why?). Then

|xn − xm |p = |cm+1 pm+1 + cm+2 pm+2 + · · · + cn pn |p < p−m ,

which is smaller than ϵ once m > log(1/ϵ) log(1/ϵ)

log p . So (C) is satisfied for any N > log p .
But {xn } cannot converge to an element of Q, on account of Exercise 4.46.
To get out of this mess, we need to fill in the following blank: R is to (Q, | · |∞ )
as is to (Q, | · |p ). The answer here turns out to be Qp ! But . . . what
is Qp ?

The p-adic Numbers, At Last!

We define the field of p-adic numbers, denoted Qp , as the fraction field of Zp .
Since Zp has characteristic 0, so does Qp , and thus Q ⊆ Qp .

p−n Zp = Zp [1/p].
S
6.72. Qp = n≥0

6.73. Every nonzero x ∈ Qp admits a unique expression in the form pv u,

where v is an integer and u is a unit in Zp .

For x ∈ Q× p , we set vp (x) = v, where v is the integer from Problem 6.73. In

order to have vp defined on all of Qp , we let vp (0) = ∞. (Compare with the
definition of vp on Set #1.)

6.74. When x ∈ Q, we have defined vp (x) twice: once on Set #1 and again
just now, since x is also an element of Qp .
Check that when x ∈ Q our two definitions of vp (x) agree (so our sin is venial,
rather than mortal). Furthermore, if we set |x|p = p−vp (x) for x ∈ Qp , then
| · |p defines a non-Archimedean absolute value on Qp . (We continue to call vp
the p-adic valuation and | · |p the p-adic absolute value.)

24
6.75. Zp = {x ∈ Qp : |x|p ≤ 1}. (That is, Zp = D≤1 (0) in Qp .)
Also, Z×
p = {x ∈ Qp : |x|p = 1}.

6.76 (Zp is compact). Let x1 , x2 , x3 , . . . be a sequence of elements of Zp .

Infinitely many xn share the same mod p component, say a1 mod p. Among
these, infinitely many share the same mod p2 component, a2 mod p2 . Etc.
Thus: x1 , x2 , x3 , . . . contains a subsequence converging to (a1 mod p, a2 mod
p2 , a3 mod p3 , . . . ) ∈ Zp .

Curious Congruences
1
6.77 (Stern). Recall that log 1−t = t + 12 t2 + 13 t3 + . . . whenever t is a complex
2 3
number with |t| < 1 (usual absolute value). Exponentiating, eT eT /2 eT /3 · · · =
1 p
1−T , as formal power series. Expanding and comparing coefficients of T shows
1
that |1 − p! − p1 |p ≤ 1. Hence, (p − 1)! ≡ −1 (mod p) (Wilson’s theorem).

ap−1 −1
6.78. For each prime p and each a ∈ Z coprime to p, put qp (a) = p . By
Fermat’s little theorem, qp (a) ∈ Z.
Prove: If a and b are both coprime to p, then qp (ab) ≡ qp (a) + qp (b) (mod p).

25
26
 p-Set #7

You Complete Me

Why does Q want to grow to R or

Q2 or Q3 ? Its
√ heart√ has holes, for
example at 2 and 3. This is
similar to mankind; we can grow to
be big boys or big girls, but there is
still some sadness in our hearts, and
we grow to love another person.

Kazuya Kato

Let (K, | · |) be a valued field. We call K complete if every Cauchy sequence of

elements of K converges to an element of K. Intuitively, completeness means
that “everything in the world happens for a reason”∗ : the far-out terms of a
sequence only clump together when they have a compelling justification for
doing so, namely heading towards a single value.

7.79. Every Cauchy sequence in Zp has a limit belonging to Zp .

7.80. Every Cauchy sequence in Qp has a limit belonging to Qp . That is, Qp

is complete.
n n
7.81. In Q5 , lim 25 is a 4th root of 1. Is lim 25 ∈ Q?

7.82. Z is dense in Zp . In other words, every element of Zp is the limit of a

sequence of terms from Z.

7.83. Q is dense in Qp .

Suppose K and L are fields equipped with absolute values and that the absolute
value | · | on L extends the absolute value on K. If (L, | · |) is complete and K
∗
Disclaimer: “Everything in the world” means every occurrence of a sequence being
Cauchy.

27
is a dense subset of L, we call L the completion of K with respect to | · |. For
example, R is the completion of Q with respect to | · |∞ . By Exercises 7.80
and 7.83, Qp is the completion of Q with respect to | · |p .
Why do we say the completion and not a completion? Completions are unique,
up to isometric (absolute-value preserving) isomorphism.

7.84. If (L, | · |) and (L′ , | · |′ ) are two completions of the same valued field
(K, | · |0 ), then there is an isomorphism ϕ : L → L′ that fixes K and satisfies
|ϕ(x)|′ = |x| for all x ∈ L.

Hands On, Digits Out

k
P
7.85. If c0 , c1 , c2 , . . . is any sequence of integers, then k≥0 ck p converges to
an element of Zp .

Restricting the “digits” ck in Problem 7.85 to {0, 1, 2, . . . , p − 1}, we find that

every infinite base p expansion determines an element of Zp . What about
the other way around? Does every element of Zp admit an infinite base p
expansion?
Let’s suppose that x ∈ Zp can be expanded in base p and see where this leads
us. Write x = c0 + c1 p + c2 p2 + . . . , with all ci ∈ {0, 1, . . . , p − 1}. Then the
digit in the pk -place is given by

(c0 + c1 p + · · · + ck pk ) − (c0 + c1 p + · · · + ck−1 pk−1 )

ck = .
pk

The parenthesized terms in the numerator are the (least nonnegative integer)
reductions of x modulo pk+1 Zp and modulo pk Zp . (Make sure you see why!)
So we’ve determined the digits ck in any possible expansion of x. Now that we
know to try these digits, we are home free, as you are asked to check in the
next exercise.

7.86. Let x ∈ Zp , and let xk (k = 0, 1, 2, 3, . . . ) be the unique integer in the

range 0 ≤ xk < pk with x ≡ xk (mod pk Zp ). Define
xk+1 − xk
ck = for k = 0, 1, 2, . . . ,
pk

Then each ck ∈ {0, 1, 2, . . . , p − 1}, each xk = c0 + c1 p + · · · + ck−1 pk−1 , and

x = c0 + c1 p + c2 p2 + c3 p3 + . . . .

28
Thus, the (finite) base p expansions of the xk coalesce to an infinite base p expansion of x. The
representation produced in this way is not merely an infinite base p expansion of x, but — as the
lead-in to the problem establishes — its unique base p expansion.

7.87 (canonical expansions for elements of Qp ). For each x ∈ Qp , there

is a unique two-sided sequence of integers {ck }∞
k=−∞ satisfying

(i) each ck ∈ {0, 1, 2, . . . , p − 1},

(ii) ck is nonzero for only finitely many k < 0,
(iii) x = k ck pk .
P

7.88. The canonical expansion of x ∈ Qp terminates (meaning that ck = 0 for

all large enough k) ⇐⇒ x = 0 or x ∈ Q+ with denominator a power of p.

7.89. The canonical expansion of x ∈ Qp is eventually periodic ⇐⇒ x ∈ Q.

Curious Congruences
p−1 p−1
X a −1 (p − 1)! + 1
7.90. For every odd prime p: ≡ (mod p).
a=1
p p

Method of Successive Approximation

√
7.91 (computing a value of 2 in Z7 ). We can compute a square root of 2
in Z7 by determining a root in Z/7, then Z/72 , then Z/73 , . . . , being mindful
to maintain compatibility throughout the process.
To start things off, the residue class x1 = 3 mod 7 satisfies x2 = 2 in Z/7. This
solution can be lifted, uniquely, to a solution of x2 = 2 in Z/72 . To see why,
note that a generic integer congruent to 3 (mod 7) has the form 3 + 7k, and
(3 + 7k)2 = 9 + 42k + 72 k 2 ≡ 9 + 42k (mod 72 ). The congruence 9 + 42k ≡ 2
(mod 72 ) is satisfied precisely when k ≡ 1 (mod 7). For integers k ≡ 1 (mod 7),
we have 3 + 7k ≡ 10 (mod 72 ). Therefore, x2 := 10 mod 72 ∈ Z/72 is the lift
we are after.
Expanding (10 + 72 k)2 mod 73 , and reasoning analogously, will show that
x2 = 10 mod 72 lifts uniquely to x3 := 10 + 72 · 2 = 108 mod 73 .
This process can be continued Q indefinitely, uniquely determining x1 , x2 , x3 , . . . .
∞
Then x := (x1 , x2 , x3 , . . . ) ∈ k=1 Z/7k is a solution in Z7 to x2 = 2. The
canonical base 7 expansion of x begins

3 + 1 · 7 + 2 · 72 + 6 · 73 + . . . .

29
After that wall of text, you might be wondering what exactly you are being asked to do. Your
job: Check that the process can be continued indefinitely, uniquely determining all the xk , that
setting x = (x1 , x2 , x3 , . . . ) really does define a Z7 -solution to x2 = 2, and finally, that the 7-adic
expansion of x starts the way we claimed.

30
 p-Set #8

The World (of p-adic) Series

8.92. Let K be a field complete with respect to a non-Archimedean absolute
value | · | (for instance, Qp !).
P∞
(a) A Pseries k=1 ak converges in K ⇐⇒ ak → 0 in K.
∞ P∞
(b) If k=1 ak converges, then | k=1 ak | ≤ maxk=1,2,3,... |ak |.
P∞
For the next Pexercise, recall that a rearrangement of a series k=1 ak is a series
∞
of the form k=1 aσ(k) , where σ is a permutation of Z+ (a bijection of Z+
with itself).

8.93. Let P
K be a field complete with respect to a non-Archimedean absolute
∞
value. If k=1 akP is a series in K that converges to s ∈ K, then every
∞
rearrangement of k=1 ak also converges to s.
P∞ P∞
8.94. Suppose that n=0 an and n=0 bn are convergent series in the field
K, which is assumed
Pncomplete with respect to a non-ArchimedeanP∞ absolute
ak bn−k , for
value. DefinePcn = k=0P n = 0, 1, 2, . . . . Then n=0 c n converges
∞ ∞ P∞
and in fact n=0 cn = ( n=0 an )( n=0 bn ).
P∞ P∞
Important consequence: If F (T ) = k=0 ak T k , G(T ) = k=0 bk T k ∈ K[[T ]] both converge at
the point z ∈ K, so does F (T )G(T ), and F (T )G(T )|T =z = F (z)G(z).

8.95 (a doubleheader?). Let K be a field complete with respect to a non-

Archimedean absolute value | · |. Let {ai,j }i,j≥0 be a doubly-indexed sequence
of elements of K. Suppose there is a sequence of real numbers {ϵN }N ≥0 tending
to 0 with the property that
|ai,j |p ≤ ϵN whenever i ≥ N or j ≥ N .∗
P P P P
Show: The double series i j ai,j and j i ai,j both converge.
∗
This may seem a strange condition to impose on {ai,j }. In fact, it’s very natural; it’s
equivalent to asking that for each ϵ > 0, the inequality |ai,j | < ϵ fails at most finitely
often. We could have phrased the requirement this way to start with, but our more
elaborate formulation will turn out to be easier to work with in proofs.

31
8.96 (a double switch). Continue with the notation and assumptions of
Exercise 8.95. For every nonnegative integer N ,
XX N X
X XX N
XX
ai,j − ai,j ≤ ϵN +1 , ai,j − ai,j ≤ ϵN +1 ,
i j i=0 j j i j i=0

N X
X N X
X N N
XX N X
X N
ai,j − ai,j ≤ ϵN +1 , ai,j − ai,j ≤ ϵN +1 .
i=0 j i=0 j=0 j i=0 j=0 i=0
P P P P
Therefore, i j ai,j = j i ai,j .

Are You Feeling the Bern?

Recall from Set #4 that when k, n ∈ Z+ , we are writing Sk (n) = 1k + 2k +
· · · + (n − 1)k .

8.97. For p prime, k ∈ Z+ : p | (Sk (p) + 1p−1|k ).

1C denotes the indicator function of the condition C. Thus, 1p−1|k is 1 if p − 1 divides k and 0
otherwise.

1p−1|k j
k p


8.98. For p prime, k ∈ Z+ : Bk +
P
p + 0<j<k j Bk−j j+1 ∈ Z(p) .

1p−1|k
8.99. For p an odd prime, k ∈ Z+ : Bk + p ∈ Z(p) .

1
8.100. For k ∈ 2Z+ : Bk + 2 ∈ Z(2) .
X 1
8.101 (Clausen, von Staudt). For k ∈ 2Z+ : Bk + ∈ Z.
p
p prime
p−1|k

It follows from Exercise 8.101 that the denominator of Bk is the (squarefree!) product of the
primes p for which p − 1 | k.

Finding Your Roots

8.102 (Teichmüller representatives). For each u ∈ Z× p , the field Qp
contains a unique (p − 1)th root of unity congruent to u mod pZp , namely
n
ω(u) := limn→∞ up .

8.103. Find ω(2) (exactly) when p = 3. Then take p = 7 and determine the
digits c0 , c1 , c2 in the canonical expansion ω(2) = c0 + c1 · 7 + c2 · 72 + . . . .

8.104. If u ∈ Z has order 3 modulo p, then 1 + u has order 6 mod p, and

ω(1 + u) = 1 + ω(u).

32
p-Set #9

Is R special? Well, physical reality, as

we and Archimedes believe it, has no
infinitesimals. . . . Hence we say that
the field R and its absolute value
| · |∞ are archimedean. . . . pace
Archimedes, in recent years the
theoretical physicists have learned
about the p-adic fields and have
begun to wonder whether they may
not help in modeling just what
happens in the nuclei of atoms, for
instance. So much for reality.

Alf van der Poorten

Gazing at the Newfound Stars

9.105. Let p be odd. Let a ∈ Z×p , and let a1 mod p be the mod p component
of a. Then a is a square in Qp ⇐⇒ a1 mod p is a square in Z/p.
Suggestion. Adapt the method of successive approximation described in Exercise 7.91.

9.106. Let p be odd, and let n ∈ Z be a nonsquare mod p. Then 1, n are coset
representatives for Z× × 2
p /(Zp ) , while 1, n, p, np are coset representatives for
× × 2
Qp /(Qp ) .

9.107. If a ∈ Z×
2 , then a is a square in Q2 ⇐⇒ a ≡ 1 (mod 8Z2 ).

9.108. Find coset representatives for Z× × 2 × × 2

2 /(Z2 ) and Q2 /(Q2 ) .

9.109. For a ∈ Q×
p:
√
a ∈ Z×
p ⇐⇒ (some value of)
n
a ∈ Qp for infinitely many n ∈ Z+ .

33
 9.110 (a Liouville approximation theorem in Zp ). Suppose α ∈ Zp is
a root of a polynomial F (T ) ∈ Z[T ] of degree d having no integer roots. For
every nonzero n ∈ Z,

|n − α|p ≥ |F (n) − F (α)|p = |F (n)|p ≥ |F (n)|−1 −d

∞ ≥ c|n|∞ ,

where c is a positive constant depending only on F .

pk! ∈ Qp is not a root of

P
9.111 (a transcendental element of Qp ). k≥1
any nonconstant polynomial in Q[x].
n n
9.112. 24·5 → 1 in Z5 while 24·5 → 0 in Z2 . Since 0 and 1 are distinct
n
rational numbers, the Z10 -limit of 24·5 has a nonperiodic 10-adic expansion.
Since 10 is not prime, your solution should start with a sensible definition of convergence in Z10 .

Strassmann Series
an T n be a formal power series with Qp -coefficients.
P
Let F (T ) = n≥0

9.113. For x ∈ Qp : F (x) converges ⇐⇒ |an xn |p → 0.

Hence, F (x) converges for all x ∈ Zp ⇐⇒ an → 0 in Qp .

When F (x) converges for all x ∈ Zp , we call F (T ) a Strassmann (power) series.∗

In courses on complex function theory, one learns that an analytic function

has only finitely many zeros in a closed disc, unless it vanishes identically on
that disc. The next three exercises guide you through a proof of an analogous
result for zeros of Strassmann series within Zp .

9.114. Let F (T ) be a Strassmann series with a nonzero coefficient. Any nonzero

x ∈ Zp with F (x) = 0 satisfies |x|p ≥ δ, where δ > 0 is a constant depending
only on F . If F (T ) has Zp -coefficients, we can take δ = |am |p , where am is the
first nonvanishing coefficient of F (T ).
In particular: If F (T ) is a Strassmann series and F (x) = 0 for all x ∈ Zp , then
F (T ) = 0 in Qp [[T ]].

k
P
9.115 (recentering Strassmann series). Let F (T ) = k≥0 ak T be a
Strassmann power series, and let x0 ∈ Zp . For every x ∈ Zp ,
∗
Here we depart from convention; the usual terms are restricted power series or strictly
convergent power series.

34
  
X
j
X k k−j
F (x + x0 ) = bj x , where bj := ak x .
j 0
j≥0 k≥j

j
P
The recentered series j≥0 bj T is also Strassmann.

9.116. A Strassmann series with a nonvanishing coefficient has finitely many

zeros in Zp .

Curious Congruences
9.117 (Glaisher). For all primes p: p2 | (p−1)!+1 ⇐⇒ Bp−1 + p1 −1 ∈ pZ(p) .
Primes p for which p2 | (p − 1)! + 1 are known as Wilson primes. The only known examples, and
the only examples smaller than 2 · 1013 , are 5, 13, and 563.

9.118 (Johnson). Recall that when u ∈ Z× p , we are writing ω(u) for the
unique (p − 1)th root of unity congruent to u modulo pZp (see Exercise 8.102).
Show that if u ∈ Z has order 3 modulo p, and v ∈ Z satisfies v ≡ ω(u)
(mod pk Zp ), where k ∈ Z+ , then

(1 + v)p ≡ 1 + v p (mod p2k+1 ).

Use this to explain the congruence 3257 ≡ 1 + 3247 (mod 77 ).

u ω(u)
2 2 + 4 · 7 + 6 · 72 + 3 · 73 + 2 · 75 + . . .
3 3 + 4 · 7 + 6 · 72 + 3 · 73 + 2 · 75 + . . .
4 4 + 2 · 7 + 3 · 73 + 6 · 74 + 4 · 75 + . . .
5 5 + 2 · 7 + 3 · 73 + 6 · 74 + 4 · 75 + . . .

Sixth roots of unity in Q7 , omitting ω(±1) = ±1. Notice that ω(3) = 1 + ω(2)
and ω(5) = 1 + ω(4), as guaranteed by Problem 8.104.

35
36
p-Set #10

If you claim a series sums to S

Your metric you must not suppress
The danger, you can now see
Is that another may disagree
And you may both be right: what a mess!

Edward B. Burger
Thomas Struppeck

I . . . Have . . . the . . . Power. . . (Series)

One can often leverage identities from the real universe to establish correspond-
ing results in the p-adic realm. As a proof of concept, consider the following
formula you may have encountered in your study of Taylor series: For all real
numbers x with |x| < 1,
√ 1 1 1 5 4
1 + x = 1 + x − x2 + x3 − x + ...
2 8 16 128
1 1
X 1 ( − 1) · · · ( 21 − (k − 1))
1
= 2 xk , where 2 := 2 2 .
k k k!
k≥0

We will argue that the same sum on k defines a square root of 1 + x in Qp

whenever it converges.
Getting from R to Qp requires a stopover in the land of formal power series.
1

k
∈ Q[[T ]], and let C(T ) = B 12 (T )2 , the formal
P
Let B 21 (T ) = k≥0 k T
2

square of B 12 (T ). We claim that C(T ) = 1 + T .

To fashion a proof, suppose x is a real number with |x| < 1. Then B 12 (x)
converges absolutely (e.g., by the ratio test). So if we multiply B 12 (x) by
B 12 (x), we can reshuffle the terms as we please. One such regrouping gives us
C(x). On the other hand, our “real world” identity says that B 12 (x)2 = 1 + x
whenever |x| < 1. It follows that C(x) − (1 + x) = 0 when |x| < 1. But a power
series that vanishes on an open interval around 0 has all its coefficients equal
to 0. This forces C(T ) = 1 + T , as formal series.

37
Armed with this formal identity, we can head back to Qp . Suppose we have in
hand an x ∈ Qp for which B 21 (x) converges. If we multiply B 12 (x) by B 12 (x),
we can rearrange the result into C(x) — this time justifying ourselves not on
the basis of absolute convergence but by an appeal to Problem 8.94. Since
C(x) = 1 + x, this shows that B 21 (x) represents a square root of 1 + x in Qp
whenever B 21 (x) converges.

10.119. If p is odd, then B 21 (x) converges when |x|p ≤ 1/p. If p = 2, then B 12 (x)
converges when |x|2 ≤ 1/23 . Are these conditions necessary for convergence?

1
9 9 k 5
in R but to − 54 in Q3 .
P

10.120. B 12 ( 16 )= k≥0
2
k
( 16 ) converges to 4

Lifting and Embedding

10.121 (Taylor’s Formula). If F (T ) ∈ Qp [T ], and a ∈ Qp , then

X F (j) (a)
F (a + T ) = T j.
j!
j≥0

1 (j)
Furthermore: If F (T ) ∈ Zp [T ], so is j! F (T ), for all nonnegative integers j.

10.122 (p-adic Newton’s method). Let F (T ) ∈ Zp [T ]. If x ∈ Zp and

|F ′ (x)|p = 1, then x̃ := x − FF′(x) 2
(x) satisfies |F (x̃)|p ≤ |F (x)|p .

10.123 (Hensel’s Lemma). Let F (T ) ∈ Zp [T ]. Suppose x1 ∈ Zp satisfies

F (x1 ) ≡ 0 (mod pZp ) and F ′ (x1 ) ̸≡ 0 (mod pZp ). Then F has a zero x ∈ Zp
satisfying x ≡ x1 (mod pZp ).

10.124. Suppose F (T ) ∈ Z[T ] is nonconstant with all complex roots distinct.

Then F (T )Q[T ] + F ′ (T )Q[T ] = Q[T ]. Hence, F (T )Z[T ] + F ′ (T )Z[T ] contains
a nonzero integer R. Deduce: F (T ) and F ′ (T ) are coprime over Z/p for all
but finitely many p.

10.125. Every nonconstant F (T ) ∈ Z[T ] has a root in Zp for infinitely many

primes p.

10.126. Let K be a number field (a finite extension of Q). By the primitive

element theorem, K = Q(θ) for some θ ∈ K. Choose F (T ) ∈ Z[T ] irreducible
over Q and vanishing at θ. Then K embeds into Qp whenever F (T ) has a root
in Qp . Deduce: There is an embedding K ,→ Qp for infinitely many p.

38
 Don’t Ever Change
10.127. The only ring homomorphism from Qp to Qp is the identity.

10.128. There is no ring homomorphism from Qp to Qq (q prime, q ̸= p) or

to R.

There is a field embedding of Qp into C (assuming the Axiom of Choice). In fact, there are
many such embeddings, but none are canonical, and none carry convergent sequences in Qp to
convergent sequences in C. Thus (channeling Hermann Weyl∗ ), choosing such an embedding must
always be regarded as something of a brute act.

Is Sticking Together Irrational?

10.129 (Mahler). The real number 0.248163264128 . . . obtained by concate-

nating the decimal digits of powers of 2 is irrational.

∗
Weyl famously wrote “The introduction of numbers as coordinates . . . is an act of
violence.”

39
40
p-Set #11

In 1926, when he was nearly 40 years old,

Skolem obtained his doctorate. . . The
somewhat advanced age has the following
explanation. In their younger years, Viggo
Brun and Skolem agreed that neither of
them would bother to obtain the degree of
Doctor, probably feeling that, in Norway,
it served no useful function in the
education of a young scientist. But in the
middle twenties a younger generation of
Norwegian mathematician emerged. It
seems that Skolem then felt he too ought
to fulfil the formal requirement of having a
doctorate, and he “obtained permission”
from Brun to submit a thesis.

Jens Erik Fenstad

I . . . Have . . . the . . . Power. . . (Series)

11.130 (p-adic lumber theory). Recall from calculus that
X (−1)k−1
log x = log(1 + (x − 1)) = (x − 1)k whenever |x − 1| < 1.
k
k≥1

This motivates us to define, for each prime p,

X (−1)k−1
logp (T ) = (T − 1)k ∈ Qp [[T − 1]].
k
k≥1

Show: logp x converges for all x ∈ Qp with |x − 1|p < 1.

Passing the familiar identity log xy = log x + log y (valid for x, y ∈ R+ ) from
R to Qp [[X − 1, Y − 1]], and then on to Qp , one can show that

logp x + logp y = logp (xy) whenever x, y ∈ 1 + pZp . (†)

41
 Filling in the details here is a bit finicky. If you enjoy this kind of work (you
know who you are. . . ), give it a try!

11.131. Go ahead and assume (†).

∞   k
X 1 1 1 p
Show: 1 + k + k + ··· + = 0 in Qp .
2 3 (p − 1)k k
k=1
∞
2k
= 0 in Q2 .∗
X
In particular (p = 2):
k
k=1

11.132 (derive responsibly!). Critique the following proof that π is irra-

tional: Suppose that π = ab , where a and b are positive integers. Let p be an
odd prime not dividing a. Then
X (ap)2k+1
0 = sin(pbπ) = sin(ap) = (−1)k ,
(2k + 1)!
k≥0

where this last series converges in Qp . Therefore,

X (ap)2k+1 X (ap)2k+1
p−1 = | − ap|p = (−1)k − ap = (−1)k ≤ p−2 .
(2k + 1)! (2k + 1)!
k≥0 k≥1
p p

Contradiction!

Strassmann Series
11.133. Let F (T ) = 1 + T + (pT )2 + (pT )4 + (pT )8 + (pT )16 + . . . . There is
exactly one x ∈ Zp with F (x) = 0.

It will be convenient for the next exercise, and certain others afterward, to
name the coefficients of the falling factorial T (T − 1) · · · (T − (N − 1)). For
nonnegative integers N and K, we let s(N, K) be the integer defined by the
formal equality
∞
X
T (T − 1) · · · (T − (N − 1)) = s(N, K)T K .
K=0
∗
“Ordinary analysis has amassed a great stock of identities between power series. Many
of these are valid in p-adic analysis too. But here an identity between power series
yields congruences between partial sums.” — Max Zorn

42
For example, when N = 5, we have T (T − 1)(T − 2)(T − 3)(T − 4) = 24T −
50T 2 + 35T 3 − 10T 4 + T 5 , so that

s(5, 0) = 0, s(5, 1) = 24, s(5, 2) = −50, s(5, 3) = 35,

s(5, 4) = −10, s(5, 5) = 1, and s(5, K) = 0 for K > 5.

When N = 0, the product T (T − 1) · · · (T − (N − 1)) is empty and assigned

the value 1; hence, s(0, 0) = 1 and s(0, K) = 0 for K > 0. The s(N, K) are
known as Stirling numbers of the first kind. Don’t let the fancy name fool you;
while the Stirling numbers are important in combinatorics, for us they play
only a notational role.

11.134 (p-adically interpolating (1 + a)x ). Let p be an odd prime, and let

a ∈ pZp . If n is a nonnegative integer, then
∞
X ak
(1 + a)n = n(n − 1)(n − 2) · · · (n − (k − 1))
k!
k=0
 
∞ k ∞
X X ak X
=  s(k, j)nj  = Ca,j nj ,
j=0
k! j=0
k=0

where
X ak
Ca,j := s(k, j) .
k!
k≥j

Moreover, |Ca,j |p → 0 as j → ∞.
Your job: Fill in the missing details. In particular, justify the swapping of the sums on k and j.

Notation. For future use, we let

∞
X
Binom(1 + a; T ) := Ca,j T j ∈ Qp [[T ]].
j=0

As you have just shown, Binom(1 + a; T ) is a Strassmann series satisfying

Binom(1 + a; n) = (1 + a)n for all nonnegative integers n.

Zeros of Linear Recurrence Sequences

Let {xn }n≥0 be a sequence of integers satisfying a linear recurrence

xn = a1 xn−1 + a2 xn−2 + · · · + ad xn−d for n = d, d + 1, d + 2, . . . ,

where a1 , . . . , ad ∈ Z, and ad ̸= 0. Put

43
      
0 1 0 ... 0 x0 1
0 0 1 ... 0  x1
 0
     
 .. . .. .. ..  x2
A=. .. and v= , e = 0.
   
 . . .  ..  .. 
0 0 0 ... 1    . .
ad ad−1 ad−2 . . . a1 xd−1 0

11.135. xn = ⟨An v, e⟩.

11.136. If p ∤ ad , then A is invertible over Fp .

11.137. Let p be as in Problem 11.136, and let k be the order of A in GL(d, Fp ),

so that Ak = Id + pB for some integer matrix B. If n = km + r, where
r ∈ {0, 1, . . . , k − 1}, then
X m
xn = pj ⟨Ar B j v, e⟩.
j
0≤j≤m

11.138 (p-adically interpolating xkm+r ). Continue with the above notation

and assumptions but assume additionally that p is odd. For each fixed r ∈
{0, 1, . . . , k − 1}, there is a Strassmann series Fk,r (T ) with Fk,r (m) = xkm+r
for every integer m ≥ 0.

11.139. For each fixed r ∈ {0, 1, . . . , k − 1}, either xn = 0 for all nonnegative
integers n ≡ r (mod k), or xn = 0 for only finitely many n ≡ r (mod k).
Hence: The set of n with xn = 0 is the union of a finite set and a finite
collection of residue classes. (Skolem’s Theorem)

11.140. Give an example of an integer linear recurrence sequence, not identi-

cally 0, with the property that xn = 0 for infinitely many n.

11.141 (Mahler). Does Skolem’s Theorem hold for recurrence sequences over
Q? (“Over Q” means that a1 , . . . , ad and x0 , . . . , xd−1 belong to Q.) Over an
arbitrary finite extension of Q (number field)?

44
p-Set #12

Strassmann Series
12.142. Let p be an odd prime and let a ∈ pZp . The identity

Binom(1 + a; n) = (1 + a)n

proved in Exercise 11.134 for nonnegative integers n holds for all n ∈ Z, since
m
(1 + a)n = lim (1 + a)n+p = lim Binom(1 + a; n + pm ) = Binom(1 + a; n).
m→∞ m→∞

12.143 (Skolem). Let p be an odd prime. Suppose a1 , . . . , am ∈ Zp and

β1 , . . . , βm ∈ 1 + pZp . Let A ∈ Zp . There is a Strassmann series F (T ) with

F (n) = a1 β1n + a2 β2n + · · · + am βm

n
−A

for all integers n. Therefore: The equation a1 β1n + a2 β2n + · · · + am βm

n
= A is
satisfied either for all n ∈ Z or for only finitely many integers n.

An Aside on Cubic Rings

Exercise 12.143 has a lovely application to a cubic analogue of Pell’s equation.

Before proceeding to that main course, we whet our appetites with some
algebraic hors d’oeuvres.
Fix a cubefree integer D > 1. Let θ be the real cube root of D and let K = Q(θ).
(Thus, K is a subfield of R.) Since T 3 − D ∈ Q[T ] is irreducible over Q (a
cubic polynomial with no roots in the ground field), each element of K has a
unique representation in the form x + yθ + zθ2 for rational x, y, and z.
Fix a complex primitive cube root of 1, say ω. Then θ′ := ωθ and θ′′ := ω 2 θ
are the nonreal complex roots of T 3 − D. The field K is isomorphic to both
K ′ = Q(θ′ ) and K ′′ = Q(θ′′ ) — indeed, all three fields are isomorphic to
Q[T ]/(T 3 − D). We will decorate elements of K with ′ and ′′ to indicate the
images in K ′ and K ′′ under the isomorphisms sending θ to θ′ and θ′′ . So
(x + yθ + zθ2 )′ = x + yθ′ + zθ′2 and similarly for ′′ .

45
For each α ∈ K, the norm Nα of α is defined by Nα = αα′ α′′ . By a tedious
but straightforward calculation,

N(x + yθ + zθ2 ) = x3 + Dy 3 + D2 z 3 − 3Dxyz. (*)

As a consequence, Nα ∈ Q for all α ∈ K.

Since ′ and ′′ are isomorphisms, N(αβ) = Nα · Nβ for all α, β ∈ K, and Nα ̸= 0
as long as α ̸= 0. Furthermore, since α′ and α′′ are a complex conjugate pair,
Nα = αα′ α′′ = α|α′ |2 has the same sign as α.
Our application requires some understanding of the units in the pure cubic
number ring
Z[θ] = {x + yθ + zθ2 : x, y, z ∈ Z}.
Clearly, Z[θ]× = ⟨−1⟩ × U, where U = Z[θ]× ∩ R+ is the collection of positive
units in Z[θ]. Proceeding further requires characterizing U norm-theoretically.
The norm map is integer-valued on Z[θ], as one sees from (*). Now if ε is a
positive unit of Z[θ], with inverse δ ∈ Z[θ], then

1 = N(1) = N(εδ) = Nε · Nδ.

Since Nε, Nδ ∈ Z+ , it must be that Nε = 1 (and Nδ = 1). Conversely, if

ε = x + yθ + zθ2 ∈ Z[θ] with Nε = 1, then ε > 0 and

ε−1 = ε′ ε′′ = (x + yθ′ + zθ′2 )(x + yθ′′ + zθ′′2 )

= x2 − Dyz + (Dz 2 − xy)θ + (y 2 − xz)θ2 ∈ Z[θ].

Hence, ε ∈ Z[θ]× ∩ R+ = U.
Summarizing: U = {ε ∈ Z[θ] : Nε = 1}.

12.144. Let α = x + yθ + zθ2 with x, y, z ∈ Q. Then

3x = α + α′ + α′′ , 3yθ = α + ω 2 α′ + ωα′′ , 3zθ2 = α + ωα′ + ω 2 α′′ .

12.145. For each real number R > 0, there are finitely many α ∈ Z[θ] with
|α|, |α′ | ≤ R. (Here and in Problem 12.146, | · | is the usual real/complex
absolute value.)

12.146 (Dirichlet’s logarithmic embedding). The map L : U → R2 defined

by L(ε) = (log ε, 2 log |ε′ |) is an injective group homomorphism with image
L(U) contained in the subspace (line) {(x, y) ∈ R2 : x + y = 0}.

12.147. For every additive subgroup G of {(x, y) ∈ R2 : x + y = 0}, one of

the following holds:

46
 x+y =0

Subgroup, shown in pink, of {(x, y) ∈ R2 : x + y = 0}

(i) G = {0},
(ii) G is infinite cyclic,
(iii) some (Euclidean) disc centered at 0 has infinite intersection with G.

12.148. L(U) = {0} or L(U) is infinite cyclic.

Since L is an isomorphism from U onto L(U), we conclude from Problem

12.148 that U = {1} or U is infinite cyclic. With a bit more work, we could
show that the second possibility always holds. (Look up the proof of Dirichlet’s
unit theorem in your favorite algebraic number theory textbook.) For our
purposes, having U = {1} would only make life easier, so we do not worry
about eliminating that (pseudo)possibility.

A Finiteness Theorem for a Cubic Analogue of Pell’s

Equation
The next two exercises outline a proof of the following theorem.

D µ D µ
2
2 1+θ+θ 7 4 + 2θ + θ2
3 4 + 3θ + 2θ2 9 4 + 2θ + θ2
4 5 + 3θ + 2θ2 10 181 + 84θ + 39θ2
5 41 + 24θ + 14θ2 11 89 + 40θ + 18θ2
6 109 + 60θ + 33θ2 12 9073 + 3963θ + 1731θ2

Generators µ of U for the first several cubefree D > 1.

47
Theorem. Let D be a cubefree integer with D > 1. Then x3 − Dy 3 = 1 has
finitely many integer solutions x, y.

Restricting to cubefree D > 1 is not significant: Any D ∈ Z+ can be factored

as D = D0 D13 , where D0 , D1 are positive integers with D0 cubefree. Distinct
solutions to x3 − Dy 3 = 1 give rise to distinct solutions to x′3 − D0 y ′3 = 1
(take x′ = x, y ′ = D1 y). When D0 > 1, the theorem applies to show that the
latter equation has finitely many integer solutions; hence, so does the former.
Finally, if D0 = 1, then D is a cube and each solution to x3 − Dy 3 = 1 yields
a way of writing 1 as a difference of two cubes. But there is only one of these:
1 = 13 − 03 . (Why?) So the conclusion of the theorem actually holds for all
D ∈ Z+ .
This theorem, first shown by Thue in 1909, stands in sharp contrast with the
situation for the classical (quadratic) Pell equation x2 − Dy 2 = 1, which has
infinitely many integer solutions for all nonsquare D ∈ Z+ .
We continue with the notation of the last section. From our work there,

x3 − Dy 3 = 1 ⇐⇒ N(x − yθ) = 1 ⇐⇒ x − yθ ∈ U. (†)

If U = {1}, the equation x − yθ ∈ U forces x = 1, y = 0, and we are done

proving the theorem already! That was too easy, so we suppose U = ⟨µ⟩, where
µ has infinite order. We may assume that µ > 1, by replacing µ with 1/µ if
necessary.
From (†), the solutions x, y ∈ Z to x3 − Dy 3 = 1 are in one-to-one correspon-
dence with integers n for which µn has vanishing θ2 -coefficient when written
with respect to the basis 1, θ, θ2 . From Problem 12.144, if µn = X + Y θ + Zθ2 ,
then
3Zθ2 = µn + ωµ′n + ω 2 µ′′n .
So we are looking for n ∈ Z where µn + ωµ′n + ω 2 µ′′n = 0. This calls to mind
Exercise 12.143.

12.149. Let L = Q(θ, ω) (= Q(θ, θ′ , θ′′ )). Fix an odd prime p for which L
embeds into Qp . Then |θ|p = |ω|p = |µ|p = |µ′ |p = |µ′′ |p = 1. That is, all of
θ, ω, µ, µ′ , µ′′ belong to Z×
p.

Here we abuse notation slightly, using the same symbols for elements of L and corresponding
elements of Qp under our embedding.

12.150. Let L and p be as in Exercise 12.149. We would like to apply Exercise

12.143 to detect the vanishing of µn + ωµ′n + ω 2 µ′′n but there is no reason to
expect that µ, µ′ , µ′′ ∈ 1 + pZp .

48
To deal with this, set ν := µp−1 , ν ′ := µ′p−1 , ν ′′ := µ′′p−1 . These belong to
1 + pZp by Fermat’s little theorem. Writing n = (p − 1)m + r,

µn + ωµ′n + ω 2 µ′′n = µr ν m + ωµ′r ν ′m + ω 2 µ′′r ν ′′m .

For each fixed r, there is some m ∈ Z where the right-hand side is nonzero
(check this!). Hence, the RHS vanishes for only finitely many m ∈ Z (Exercise
12.143). The Theorem follows.
Much more is known. For instance, Delaunay and Nagell showed (independently) that x3 −Dy 3 = 1
has at most one integer solution ̸= (1, 0).

12.151Q(a striking corollary). Let P be a finite set of primes and let

D = { p∈P pep : each ep = 0, 1, or 2}. If n ∈ Z+ and n3 + 1 has all prime
factors from P, then (−n)3 −Dy 3 = 1 for some integer y and some D ∈ D \{1}.
Deduce: The largest prime factor of n3 + 1 tends to infinity with n.
By more elaborate methods, Siegel showed that the largest prime factor of f (n) tends to infinity
whenever f (T ) ∈ Z[T ] is nonconstant with at least two distinct complex roots.

49
50
p-Set #13

Narrator: And so we come to the last

chapter in which Christopher Robin and
Pooh come to the enchanted place and we
say goodbye.
Winnie the Pooh: Goodbye? Oh no
please can’t we go back to page one and
do it all over again?
Narrator: Sorry Pooh. But all stories
have an ending you know.
Winnie the Pooh: Oh bother.

The Many Adventures of Winnie the Pooh

Bern, Baby, Bern!

For each integer u not divisible by p, let ω(u) denote the (p − 1)th root of
unity in Qp congruent to u modulo pZp (see Problem 8.102). Define ϑ(u) ∈ Zp
by the equation ω(u) = u + pϑ(u).
Pp−1
13.152. For each k ∈ Z+ : u=1 ω(u)k = 1p−1|k (p − 1).

Bk 1p−1|k (p−1)
13.153. Let βk = k − pk . For each k ∈ Z+ :
p−1
X k − 1 pj X k − 1 pj−1 X
βk + Bk−j + uk−j θ(u)j = 0.
j−1 j(j + 1) j−1 j u=1
0<j≤k 0<j≤k

13.154. Assume p is odd. Then βk ∈ Zp for every k ∈ Z+ . As a consequence,

Bk
k ∈ Zp whenever p − 1 ∤ k (Adams).

Pp−1
13.155. Assume p ≥ 5. Then βk + u=1 uk−1 ϑ(u) ∈ pZp for each k ∈ 2Z+ .

13.156 (Kummer). If k, k ′ are even positive integers with k ≡ k ′ (mod p−1),

and p − 1 ∤ k, then Bkk ≡ Bkk′′ (mod pZp ).

51
 B4 1 B10 1 B4 B10 1
For example, 4 = − 120 and 10 = 132 are congruent mod 7Z7 . In fact, 4 − 10 = 7 · − 440 .

13.157 (Glaisher). Let p ≥ 5, and put k = φ(p3 ) − 1. Modulo p3 Zp ,

p−1
X p2 p2
Hp−1 ≡ nk ≡ k Bk−1 ≡ − Bp−3 .
n=1
2 3

(Here, as usual, Hp−1 = 1 + 12 + · · · + p−1

1
.) Therefore: p3 divides the numerator
of Hp−1 ⇐⇒ p divides the numerator of Bp−3 . (Compare with Problem 1.11.)
Primes p dividing the numerator of Bp−3 are known as Wolstenholme primes. The only examples
not exceeding 1011 are 16 843 and 2 124 679.

Strassmann Series

Let F (T ) = k≥0 ak T k ∈ Qp [[T ]] be a Strassmann series where not all ak = 0.

P
Since ak → 0 in Qp , there is a largest nonnegative integer K with

aK = max |ak |p .
k≥0

We refer√ to K as the Strassmann degree of F (T ). For example, p + T +

⌊ k⌋ k
T has Strassmann degree 1, while k≥0 k! · T k has Strassmann
P P
k≥2 p
degree p − 1.

13.158. Let F (T ) = k≥0 ak T k be a Strassmann series with a zero r ∈ Zp .

P
For all x ∈ Zp , we have F (x) = (x − r)G(x), where
X X
G(T ) = bj T j , with bj := ak rk−1−j .
j≥0 k>j

Moreover, if F (T ) has Strassmann degree K ≥ 1, then G(T ) is Strassmann

with Strassmann degree K − 1.

13.159 (Strassmann’s Theorem). Let F (T ) be a Strassmann series with

Strassmann degree K. Then F (x) = 0 for at most K distinct values of x ∈ Zp .

Ramanujan’s Conjecture Revisited

We finally return to the study of the equation x2 + 7 = 2m initiated in Exercise

3.40. By that problem, to establish Ramanujan’s conjecture√
it suffices to show
√
n
−β n
that there are no n > 13 with αα−β = ±1. Here α = 1+ 2 −7 and β = 1− 2 −7 .

52
 √
√ quadratic field Q( −7) can be viewed as a subfield of Q11 , identifying
The
−7 with the square root of −7 in Z11 that is congruent to 2 modulo 11Z11 .
By hand, or with the aid of software such as PARI/GP, one computes that
√
−7 = 2 + 8 · 11 + 8 · 112 + 7 · 113 + 10 · 114 + 1 · 115 + . . . ,

where . . . suppresses a quantity with 11-adic absolute value at most 11−6 .

Then

α = 7 + 9 · 11 + 9 · 112 + 3 · 113 + 5 · 114 + 6 · 115 + . . . ,

β = 5 + 1 · 11 + 1 · 112 + 7 · 113 + 5 · 114 + 4 · 115 + . . . .

n n
−β
We study the equation αα−β = ±1 by the method of Exercise 12.150. Put
10 10
A = α and B = β , so that A = 1 + a, B = 1 + b for

a = 7 · 11 + 1 · 112 + 1 · 113 + 7 · 114 + 5 · 115 + . . . ,

b = 9 · 11 + 9 · 112 + 9 · 113 + 3 · 114 + 5 · 115 + . . . ,

both of which belong to 11Z11 . Write n = 10m + r, where r ∈ {0, 1, . . . , 9}.

Then
αn − β n
= ±1 ⇐⇒ αn − β n = ±(α − β)
α−β
⇐⇒ αr (1 + a)m − β r (1 + b)m ∓ (α − β) = 0.

For each r ∈ {0, 1, . . . , 9} and each choice of ± sign, define

Fr,± (T ) = αr Binom(1 + a; T ) − β r Binom(1 + b; T ) ∓ (α − β),

so that Fr,± (m) = αr (1 + a)m − β r (1 + b)m ∓ (α − β) for all m ∈ Z.

Each of our 20 power series Fr,± (T ) is a Strassmann series. For every one of
these, we will apply Strassmann’s Theorem (Exercise 13.159) to bound the
number of zeros in Zp .

13.160. The constant term of Fr,± (T ) is αr − β r ∓ (α − β), which vanishes

when
(r, ±) ∈ {(1, +), (2, +), (3, −), (5, −)}
and is an 11-adic unit in the other sixteen cases. Every nonconstant coefficient
of every Fr,± (T ) is a multiple of 11. Hence, Fr,± (T ) has no zero in Z11 except
possibly for the four displayed values of (r, ±).

13.161. All of F1,+ (T ), F2,+ (T ), F5,− (T ) have their T -coefficients multiples

of 11 but not 112 . Their Strassmann degrees are all 1.

53
13.162. F3,− (T ) has a T -coefficient that vanishes mod 112 . Its T 2 -coefficient
is a multiple of 112 but not 113 . Its Strassmann degree is 2.
n n
−β
13.163. There are at most 3 · 1 + 2 = 5 integers n with αα−β = ±1. We know
five such integers — n = 1, 2, 3, 5, and 13 — so those must be all of them.

Remark. The choice to work in Qp for p = 11 was fortuitous. The next prime√after
11 for which −7 is a quadratic residue is p = 23. If we had elected to embed Q( −7)
into Q23 , we would have to bound the number of zeros of 2 · 22 = 44 Strassmann
22m+12
−β 22m+12
series. One of those would correspond to the equation α α−β
= −1. If you
approximate the coefficients sufficiently to apply Strassmann’s theorem, you’ll find
22m+12
−β 22m+12
that this series has at most one zero m ∈ Zp ; hence, α α−β
= −1 has at
most one integer solution m. But in fact (as we know after Exercise 13.163), there
are zero integers m satisfying the equation. So working in Q23 , we fail to rule out an
extraneous zero. What’s going on in this instance is that there really is an m ∈ Z23
where the associated power series vanishes — but this m does not belong to Z!

54
 Part II

Solutions, Discussion, and Extra Explorations

55
56
Solutions to Set #1

1.1
(a) First off, |1|2 = |1 · 1| = |1| and |1| > 0 (from (i)). Thus, |1| = 1. Next,
| − 1|2 = |(−1)(−1)| = |1| = 1. As | − 1| > 0, we conclude that | − 1| = 1.
(b) The proof is the same for the “standard” absolute value: By the triangle
inequality, |x| = |(x − y) + y| ≤ |x − y| + |y|. Now rearrange.
(c) Since |y −1 | · |y| = |y −1 · y| = |1| = 1, we have |y −1 | = |y|−1 . So |xy −1 | =
|x||y −1 | = |x||y|−1 , as claimed.

1.2 Property (i) in the absolute value definition is clear. Property (ii) is also
easy: When x + y = 0, the inequality is obvious. Otherwise, either x or y is
nonzero, so that |x| or |y| is 1. Hence, 1 = |x + y| ≤ |x| + |y|. To prove (iii),
take cases: If x and y are nonzero, both sides are 1, otherwise both sides are 0.
In this last step we use that fields are integral domains.

1.3 Property (i) is again clear. Of the remaining two properties, (iii) is
quicker to dispense with: If x or y is zero, both sides of (iii) vanish. Otherwise,
write x = pvp (x) ab and y = pvp (y) dc , where p does not divide any of a, b, c, d.
Then xy = pvp (x)+vp (y) ac
bd , and p does not divide either of ac or bd. Hence,
vp (xy) = vp (x) + vp (y) and |xy|p = p−vp (x) p−vp (y) = |x|p |y|p .
To prove (ii) we have to work a bit harder. If x, y, or x + y is zero, (ii)
is trivial. Otherwise, write x = pvp (x) ab and y = pvp (y) dc as above. The
symmetry of (ii) in x and y allows us to assume vp (x) ≤ vp (y). Then
x + y = pvp (x) ab + pvp (y)−vp (x) dc . Since pvp (y)−vp (x) ∈ Z, we can express

pvp (y)−vp (x) dc as a fraction with denominator d. Hence, ab + pvp (y)−vp (x) dc = bd
N
w ′
for some nonzero N ∈ Z. Write N = p N , where w is a nonnegative in-
′
teger and N ′ ∈ Z is not divisible by p. Then x + y = pvp (x)+w N bd , where
neither N ′ nor bd is divisible by p. Hence, vp (x + y) = vp (x) + w ≥ vp (x), and
|x + y|p = p−vp (x) p−w = |x|p p−w ≤ |x|p ≤ |x|p + |y|p .

1.4 This is implicit in our solution to Problem 1.3.

57
1.5 We may assume without loss of generality that |x| > |y|. A direct
application of the strong triangle inequality gives |x + y| ≤ max{|x|, |y|} = |x|.
The strong triangle inequality also implies that |x| = |(x + y) + (−y)| ≤
max{|x+y|, |−y|} = max{|x+y|, |y|}. (We use in this last step that |−y| = |y|,
which follows from | − 1| = 1.) Since |x| > |y|, the maximum here cannot be
|y|. So it must be |x + y|, yielding |x| ≤ |x + y|. Hence, |x + y| = |x|.

1.6 Since |2| ≤ 1, we see that |2e1 +· · ·+2en | ≤ |2|e1 +· · ·+|2|en ≤ 1+· · ·+1 = n.
To conclude, notice that (a) every positive integer less than 2n is a sum
of
P at most
n powers of 2 (e.g., use the binary representation), while (b)
n n
n

n
0≤k≤n k = 2 , so that k < 2 for every k.

1.7 Write x = ± ab where a and b are relatively prime positive integers. Since
a and b share no prime factors,
Y Y
Y −vp (a) Y vp (b)
|x|p = |a|p |b|−1
p = p p = |a|−1
∞ |b|∞ ,
p prime p prime p|a p|b

which is the multiplicative inverse of |a|∞ |b|−1

∞ = |x|∞ .

Extra Exploration 1 (simultaneous approximation). The product formula

suggests that | · |∞ , | · |2 , | · |3 , | · |5 , . . . “know about each other.” The situation is
very different if one considers only a finite subset of these absolute values. Make this
precise by proving the following independence statement.

Let P be a finite set of primes. Choose rational numbers xp for each prime p ∈ P,
alongside a rational number x∞ . For each ϵ > 0, there is a rational number x
satisfying

|x − xp |p < ϵ for all p ∈ P, as well as |x − x∞ |∞ < ϵ.

1.8 We begin by demonstrating a fabulously useful formula of Legendre: For

each prime p and nonnegative integer n,
X X X X X X n 
vp (n!) = vp (m) = 1= 1= .
pk
1≤m≤n k 1≤m≤n p |m k≥1 1≤m≤n k≥1
k≥1 pk |m

When n > 0, it follows that vp (n!) < k≥1 pnk = p−1

n
P
. (The inequality is surely
strict, as ⌊n/p ⌋ < n/p whenever p > n.) Therefore, |n!|p = p−vp (n!) >
k k k

p−n/(p−1) .

Extra Exploration 2 (following up on Legendre’s formula). Let p be a

prime. For each nonnegative integer n, expand n in base p: write n = n0 + n1 p +
n2 p2 + . . . , where each ni ∈ {0, 1, 2, . . . , p − 1} and all but finitely many ni = 0. Set
sp (n) = n0 + n1 + n2 + . . . .

58
 1
(a) Show that vp (m) = p−1 (sp (m − 1) − sp (m) + 1) for every positive integer m.
1
Deduce that vp (n!) = p−1 (n−sp (n)) for all nonnegative integers n (an alternative
form of Legendre’s formula).
(b) Prove that n!/(−p)vp (n!) ≡ n0 !n1 !n2 !n3 ! · · · (mod p) for all nonnegative n ∈ Z.
(Anton [1], Stickelberger [5, pp. 342–343], Hensel [2])

1.9 We interpret “grows faster” to mean that n!/C n tends to infinity. To prove
that, we may (in fact, should!)
P assume that C > 0. A (canonical) application
of the ratio test shows that n≥0 C n /n! converges. (It converges to eC but we
do not need that here.) So its terms must tend to 0. Since C n /n! is positive
for each n, we deduce that n!/C n → ∞.
On
Q the−1 hand, if there areQonly finitely many primes, then n! = |n!|∞ =
other Q
n/(p−1)
|n!| p ≤ pp = ( p p1/(p−1) )n . That is, n! ≤ C n with C :=
Qp 1/(p−1)
pp . Contradiction!

Extra Exploration 3 (Skolem [4]). Show that if p is prime and n ∈ Z+ , either

|2n − 1|p = 1 or |2n − 1|p = |2p−1 − 1|p |n|p . Deduce that if P is any fixed finite set
of primes, then 2n − 1 has a prime factor outside of P for all sufficiently large n.

1.10 Given n ∈ Z+ , let e be the largest nonnegative integer such that 2e ≤ n.

Then 2e is the unique integer in [1, n] divisible by 2e . (The next smallest
multiple of 2e is 2e+1 , but this exceeds n.) It follows that |1/m|2 = 2v2 (m) has
e
a unique maximum, among P m ∈ [1, n], at m =e 2 . Bye “survival of the greatest”
(Exercise 1.4), |Hn |2 = | m≤n 1/m|2 = |1/2 |2 = 2 > 12 n. So |Hn |2 → ∞.

Extra Exploration 4. Prove that in any nonempty set S of consecutive positive

integers, one element of S has strictlyPsmaller 2-adic absolute value than all the
others. Conclude that if S ̸= {1}, then n∈S 1/n ∈ / Z (Kürschák [3]).

1.11 The stated expression for Hp−1 is immediate upon pairing the terms
1
i and
1
. To show |Hp−1 |p ≤ p−2 , we must prove that |S|p ≤ p−1 , where
Pp−i 1
S := 0<i<p/2 i(p−i) .

We would like to evaluate S modulo p by reducing term-by-term. This is

essentially what we will do, but since it is not immediately clear what it means
to reduce a rational number mod p, we premultiply by (p − 1)!. In the field
Fp = Z/p, we have (p−1)!
i(p−i) = (p − 1)!i
−1
(p − i)−1 = −(p − 1)!i−2∗ whenever
0 < i < p/2. Therefore (still working in Fp ),
X 1 X 1 X
(p − 1)!S = −(p − 1)! i−2 = − (p − 1)! i−2 = − (p − 1)! j2.
2 ×
2 ×
0<i<p/2 i∈Fp j∈Fp

∗
The first equality is not a tautology! What is being claimed is that the integer on
the left has its mod p reduction equal to the element of Fp on the right. The −1st
powers indicate inverses in the field Fp .

59
Here we use that i−2 = (p − i)−2 and that as i runs over all the nonzero
elements of Fp , so does j = i−1 . To finish, observe that since p > 3, the group
F× ×
p has an element other than ±1. So there is an r ∈ Fp with r =
2
̸ 1. Since
×
multiplication by r permutes Fp ,
X X X
r2 j2 = (rj)2 = j2,
j∈F×
p j∈F×
p j∈F×
p

j 2 = 0 (since r2 ̸= 1). It follows that (p − 1)!S = 0 in Z/p,

P
forcing j∈F× p
and so (p − 1)!S is a multiple of p. Since p ∤ (p − 1)!, we conclude that
|S|p = |(p − 1)!|p |S|p = |(p − 1)!S|p ≤ p−1 , as desired.

Remark. It is not really necessary to premultiply by (p − 1)!. On the next problem

set, we will introduce the ring Z(p) , whose elements are the rational numbers with
denominators prime to p. Z(p) is a local ring (in fact, domain) with unique maximal
ideal pZ(p) , and Z(p) /pZ(p) ∼
= Z/p = Fp . The same argument we used to show
(p − 1)!S = 0 in Z/p will show directly that S = 0 in Z(p) /pZ(p) , and this is sufficient
to conclude that |S|p ≤ p−1 .

References

1. H. Anton, Die Elferprobe und die Proben für die Modul Neun, Dreizehn und
Hunderteins. Für Volks- und Mittelschulen. Archiv Math. Physik 49 (1869),
241–308.
2. K. Hensel, Über die arithmetischen Eigenschaften der Faktoriellen. Archiv Math.
Physik (third series) 2 (1902), 293–294.
3. J. Kürschák, A harmonikus sorról. Mat. Fiz. Lapok 27 (1918), 299–300.
4. T. Skolem, On certain exponential equations. Norske Vid. Selsk. Forh. 18 (1945),
71–74.
5. L. Stickelberger, Ueber eine Verallgemeinerung der Kreistheilung. Math. Ann.
37 (1890), 321–367.

60
Solutions to Set #2

2.12 We have to show that for any x, y, z ∈ K, at least two of |x − y|, |y − z|,
and |z−x| coincide. This is a simple consequence of “survival of the greatest”: If
|x−y| =
̸ |y −z|, then |z −x| = |x−z| = |(x−y)+(y −z)| = max{|x−y|, |y −z|}.
This argument shows that the largest side length always appears at least twice.

2.13 Suppose to start with that z ∈ D<r (x). Then |z − x0 | = |(z − x) + (x −

x0 )| ≤ max{|z − x|, |x − x0 |} < r. Thus, D<r (x) ⊆ D<r (x0 ) = D. Similarly,
if z ∈ D, then |z − x| = |(z − x0 ) + (x0 − x)| ≤ max{|z − x0 |, |x0 − x|} < r,
proving that D ⊆ D<r (x). So D = D<r (x).

2.14 Suppose that x ∈ K belongs to the intersection of the open discs

D0 = D<r0 (x0 ) and D1 = D<r1 (x1 ). By Problem 2.13, D0 = D<r0 (x) and
D1 = D<r1 (x). Then D0 ⊆ D1 or D1 ⊆ D0 according to whether r0 ≤ r1 or
vice versa.

2.15 Let R = {pv : v ∈ Z}. If r ∈/ R, then D<r (x0 ) = D≤r (x0 ), for any center
x0 . If r ∈ R, then D<r (x0 ) = D≤r/p (x0 ) while D≤r (x0 ) = D<pr (x0 ).

2.16 Suppose | · | is an absolute value on F = F2027 . Let g be a generator of

F × . Then |g|2026 = |g 2026 | = |1| = 1, and so |g| = 1. Since g generates F × , it
follows that |x| = 1 for all nonzero x in F . Therefore, the only absolute value
on F is the trivial absolute value.
This argument works for any Fp , or any finite field for that matter.

2.17 Applying the binomial theorem and the triangle inequality,

n   n  
n
X n k n−k X n
|x + y| = x y ≤ |x|k |y|n−k
k k
k=0 k=0
n    
X n n
≤ max{|x|, |y|}n ≤ max{|x|, |y|}n · (n + 1) max .
k 0≤k≤n k
k=0

61
(Moving from the first to the second line, we used that |x|k |y|n−k ≤
max{|x|, |y|}k · max{|x|, |y|}n−k = max{|x|, |y|}n .) Take nth roots.

2.18 Every absolute value on Fp (T ) restricts to an absolute value on Fp ,

hence is trivial on Fp (Exercise 2.16). It follows that every binomial coefficient
has absolute value at most 1. So by Exercise 2.17, for all x, y ∈ Fp (T ),

|x + y| ≤ (n + 1)1/n max{|x|, |y|}.

Sending n to infinity, |x + y| ≤ max{|x|, |y|}. In other words, | · | is non-

Archimedean.

2.19 We start by constructing a family of absolute values on Fp (T )

parametrized by the monic irreducibles in Fp [T ].
Fix a monic irreducible π ∈ Fp [T ]. Every x ∈ Fp (T )× can be written in the
form π v ab where a and b are elements of Fp [T ] not divisible by π. Here the
integer vπ (x) := v is uniquely determined by x (a consequence of the unique
factorization theorem for Fp [T ]). Fix your favorite constant Cπ > 1. If we
−v (x)
set |x|π = Cπ π for nonzero x ∈ Fp (T ), and set |0|π = 0, then | · |π is a
non-Archimedean absolute value on Fp (T ). The proof is more or less identical
to that for | · |p (see Exercise 1.3).
If π and π̃ are distinct monic irreducibles in Fp [T ], then |π|π = Cπ−1 < 1 while
|π̃|π = Cπ0 = 1. This settles the first half of the problem.
As for the concluding question: Yes, there is such an absolute value. Fix
C∞ > 1. For x ∈ Fp (T )× , write x = ab with a, b ∈ Fp [T ]. While a and b are
not uniquely determined by this representation, the difference deg a − deg b is
deg a−deg b
independent of the choice of a and b. We put |x|∞ = C∞ for nonzero
x ∈ Fp (T ), taking |0|∞ = 0. Since |T |∞ = C∞ > 1, we will be done if we show
| · |∞ is an absolute value on Fp (T ).
Condition (i) in the absolute value definition (see Set #1) is obvious. Condition
(iii) follows from deg uv = deg u + deg v. To prove (ii), we may assume x, y,
and x + y are nonzero. Write x = ab and y = dc . Then x + y = ad+bc bd , and

deg(ad + bc) − deg(bd) ≤ max{deg(ad), deg(bc)} − deg(bd)

= max{deg(ad) − deg(bd), deg(bc) − deg(bd)}
= max{deg(a) − deg(b), deg(c) − deg(d)}.

Since C∞ > 1, the inequality is preserved upon raising C∞ to both sides. This
gives |x + y|∞ ≤ max{|x|∞ , |y|∞ }, proving the strong triangle inequality.

2.20 To prove that D≤1 (0) is a subring it is enough to argue that 1 ∈ D≤1 (0)
and that D≤1 (0) is closed under multiplication and subtraction. The first

62
requirement is clear, since |1| = 1. Closure under multiplication follows from the
multiplicative property of |·|, as the interval [0, 1] is closed under multiplication.
Closure under subtraction is a consequence of the strong triangle inequality: If
|x|, |y| ≤ 1, then |x − y| ≤ max{|x|, | − y|} = max{|x|, |y|} ≤ 1.
By definition of the p-adic absolute value, x ∈ Z(p) if and only if x = pv a/b for
some nonnegative integer v and some a, b ∈ Z not divisible by p. This happens
precisely when the denominator of x in lowest terms is not a multiple of p.

2.21 T Write x = a/b in lowest terms, with b > 0. By Problem 2.20,

T ∈ p prime Z(p) ⇐⇒ b has no prime factors ⇐⇒ b = 1 ⇐⇒ x ∈ Z. So
x
p prime Z(p) = Z.

2.22 Let 0 < k < p. Working modulo p,

 
p−1
(k − 1)! = (p − 1)(p − 2) · · · (p − (k − 1))
k−1
≡ (−1)(−2) · · · (−(k − 1))
≡ (−1)k−1 (k − 1)!.
p−1 p−1



Since p ∤ (k − 1)!, we conclude that k−1 ≡ (−1)k−1 (mod p). Write k−1 =
k−1 p

p p−1

(−1)k−1 2r
(−1) + pr, where r ∈ Z. Then k = k k−1 = p k + p k . Hence,
working in the ring Z(p) ,

(−1)k−1
 
p
≡p (mod p2 Z(p) ).
k k

We have made sense of the claimed congruence mod p2 by interpreting it —

nay, proving it — as a congruence modulo the ideal p2 Z(p) of the ring Z(p) .

k−1
2.23 Summing the congruence kp ≡ p (−1)k


(mod p2 Z(p) ) over integers
p k−1
/k (mod p2 Z(p) ). Dividing by p,
P
0 < k < p yields 2 − 2 ≡ p 0<k<p (−1)

2p − 2 1 1 1 1
≡ 1 − + − + ··· − (mod pZ(p) ).
p 2 3 4 p−1
The left and right-hand sides of the displayed congruence have difference
smaller than 1 in terms of p-adic absolute value. So by the strong triangle
inequality, one side has absolute value < 1 if and only if the other does. The
p
solution is concluded by observing that | 2 p−2 |p < 1 ⇐⇒ p2 | 2p − 2.

2.24 Since F has finitely many complex roots, we can fix n0 ∈ Z with
F (n0 ) ̸= 0. Replacing F (T ) with F (T + n0 ), we may assume that F (T ) has
nonzero constant term a0 (say). Then F (a0 T ) = a0 G(T ) for some nonconstant
G(T ) ∈ Z[T ] with G(0) = 1.

63
Let P be the set of primes dividing G(n) for some integer n. Every prime
dividing a value of G also divides a value of F , so it suffices to prove P is
infinite.
We mimic Euclid. Suppose p1 , . . . , pk is any finite list of primes in P. We
choose an integer m with |G(mp1 · · · pk )| > 1. (Such an m surely exists, as the
inequality excludes no more than 3 deg G values of m.) Then G(mp1 · · · pk ) is
divisible by some prime p, but

G(mp1 · · · pk ) ≡ G(0) ≡ 1 (mod pi )

for each i = 1, 2, . . . , k. So there is a prime p ∈ P not on our list. As this is

true no matter what finite list we start with, P is infinite.

Extra Exploration 5. Let P be a finite set of primes, say #P = k. Show that

there are positive constants c and x0 such that, for every real number x ≥ x0 ,

#{n ∈ Z : |n| ≤ x and p | n ⇒ p ∈ P} ≤ c(log x)k .

Use this to give another solution to Problem 2.24.

Extra Exploration 6 (Bauer [1]; see also Nagell [4, §49, pp. 168–169]).
Let F (T ) be a nonconstant polynomial with integer coefficients. Suppose that F has
a real root of odd multiplicity. Show that for each integer m ≥ 3, there are infinitely
many primes p ̸≡ 1 (mod m) for which F has a root mod p.

2.25 Suppose p is odd. If p | n4 + 1, then n4 ≡ −1 (mod p) and n8 ≡ 1

(mod p). Since −1 ̸≡ 1 (mod p), the order of n modulo p divides 8 but does
not divide 4 — so it must be precisely 8. As the order is always a divisor of
p − 1, we conclude that p ≡ 1 (mod 8).
To obtain infinitely many primes p ≡ 1 (mod 8), apply Problem 2.24.

2.26 If p | (2n + 1)2 − 2, then p is odd and 2 is a square modulo p. By

elementary number theory, p ≡ ±1 (mod 8).
For each n ∈ Z+ , the integer (2n + 1)2 − 2 is larger than 1 and thus factors as a
product of positive primes. If each prime in this factorization is congruent to 1
(mod 8), then (2n + 1)2 − 2 is also congruent to 1 (mod 8). But (2n + 1)2 − 2 ≡
1 − 2 ≡ −1 (mod 8). Thus, (2n + 1)2 − 2 must be divisible by some prime
congruent to −1 (mod 8).
Finally, suppose p1 , . . . , pk is any finite list of primes congruent to −1 (mod 8).
For each i = 1, 2, . . . , k, there are two residue classes ni mod pi for which
(2ni + 1)2 − 2 ≡ 0 (mod pi ). As each pi > 2, the Chinese Remainder Theorem
allows us to choose a positive integer n not congruent to any of the ni (mod pi ).
Then (2n + 1)2 − 2 is divisible by some prime p ≡ −1 (mod 8) but not divisible
by any of p1 , . . . , pk . Thus, there must be a prime congruent to −1 mod 8

64
that is not on the list p1 , . . . , pk . As our starting list was arbitrary, there are
infinitely many primes p ≡ −1 (mod 8).

2.27 Problem 2.25 handles the residue class 1 mod 8 while Problem 2.26
handles −1 mod 8. To take care of 5 mod 8, we argue as in Problem 2.26 with
(2n + 1)2 + 4 replacing (2n + 1)2 − 2. Every odd prime p with −4 a square mod p
is congruent to 1 or 5 (mod 8). Since (2n + 1)2 + 4 ≡ 1 + 4 ≡ 5 (mod 8), there
must be some prime congruent to 5 (mod 8) dividing (2n + 1)2 + 4. Following
the solution to Problem 2.26, we obtain infinitely many primes p ≡ 5 (mod 8)
by varying n.
To deal with 3 mod 8, use (2n + 1)2 + 2. Every odd p with −2 a square mod p
is congruent to 1 or 3 (mod 8). As (2n + 1)2 + 2 ≡ 1 + 2 ≡ 3 (mod 8), there
is always some prime congruent to 3 (mod 8) dividing (2n + 1)2 + 2. Varying
n, we obtain infinitely many primes p ≡ 3 (mod 8).

Remark. Dirichlet’s general theorem is proved by moderately sophisticated analytic

methods. By contrast, the proofs in the last few exercises are variants on Euclid’s
simple and familiar argument. This invites the question: Which other residue classes
can be shown to contain infinitely many primes by a proof in the style of Euclid’s?
For one reasonable interpretation of “in the style of Euclid’s” (which regrettably
would take us too far afield to motivate here) an elegant answer has been given by
Issai Schur and Ram Murty: There is a Euclid-style proof of the infinitude of primes
congruent to a (mod m) ⇐⇒ a2 ≡ 1 (mod m). See [2, 3, 5] for details.

References

1. M. Bauer, Über die arithmetische Reihe. J. Reine Angew. Math. 131 (1906),
265–267.
2. M. R. Murty, Primes in certain arithmetic progressions. J. Madras Univ., Section
B, 51 (1988), 161–169.
3. M. R. Murty and N. Thain, Prime numbers in certain arithmetic progressions.
Funct. Approx. Comment. Math. 35 (2006), 249–259.
4. T. Nagell, Introduction to number theory, Chelsea Publishing Co., New York,
1964.
5. I. Schur, Über die Existenz unendlich vieler Primzahlen in einigen speziellen
arithmetischen Progressionen. Sitzungber. Berliner Math. Ges. 11 (1912), 40–50.

65
66
Solutions to Set #3

3.28 The proof is the same as in calculus: Suppose for a contradiction that
xn → x and xn → x′ , where x′ ̸= x. Then ϵ := 12 |x′ − x| > 0. Since xn → x,
we can choose N ∈ Z+ with |xn − x| < ϵ for all n ≥ N . Similarly, we can
choose N ′ ∈ Z+ with |xn − x′ | < ϵ for all n ≥ N ′ . Taking n ≥ max{N, N ′ },
we find that

|x′ − x| = |(x′ − xn ) + (xn − x)| ≤ |x′ − xn | + |xn − x| < 2ϵ = |x′ − x|.

Contradiction!

3.29
(a) Let ϵ > 0. Choose N1 = 1. If n ≥ N1 , then |xn − x| = 0 < ϵ.
(b) Let ϵ > 0. Choose N1 , N2 ∈ Z+ so that |xn − x| < 12 ϵ whenever n ≥ N1
and |yn − y| < 12 ϵ whenever n ≥ N2 . For n ≥ max{N1 , N2 },

1 1
|(xn +yn )−(x+y)| = |(xn −x)+(yn −y)| ≤ |xn −x|+|yn −y| < ϵ+ ϵ = ϵ.
2 2
So xn + yn → x + y.
(c) Here we must work a bit harder. When checking the definition of conver-
gence, we can assume that 0 < ϵ < 1. (Larger values of ϵ only make life
1
easier.) Given such an ϵ, choose N1 , N2 ∈ Z+ with |xn − x| < 3(|y|+1) ϵ for
1
all n ≥ N1 and |yn − y| < 3(|x|+1) ϵ for all n ≥ N2 . Write xn = x + dn and
yn = y + en , so that xn yn = xy + xen + ydn + dn en . For n ≥ max{N1 , N2 },

|x| ϵ |y| ϵ
|xen | ≤ ϵ< and |ydn | ≤ ϵ< .
3(|x| + 1) 3 3(|y| + 1) 3
ϵ 1
For these same values of n, we have |dn |, |en | < 3 < 3. So (estimating
crudely) |dn en | ≤ |dn | < 3ϵ . Therefore,
ϵ ϵ ϵ
|xn yn − xy| = |xen + ydn + dn en | ≤ |xen | + |ydn | + |dn en | < + + = ϵ.
3 3 3

67
 3.30 Notice that 2xn = 2 + 2 · 3 + 2 · 32 + · · · + 2 · 3n , which is precisely
n+1
the ternary expansion of 3n+1 − 1. So xn = 3 2 −1 = − 12 + 12 3n+1 , and
|xn − (− 12 )|3 = | 12 3n+1 |3 = 3−(n+1) , which tends to 0. Therefore, xn → − 12 in
(Q, | · |3 ).
P∞
The series k=0 3k diverges in Qp for each prime p ̸= 3. To prove this, we
appeal to a result possessing the air of the familiar.
P∞
Lemma (kth term test for a valued field). If k=1 ak converges in (K, |·|),
then ak → 0.

To apply this in our situation, observe that if p ̸= 3, then |3k |p = 1 for every
k, and 1 does not tend to 0 !∗
P∞
Proof. Suppose k=1 ak = P x (where x ∈ K). This means that the sequence
n
{sn } with nth term sn = k=1 ak converges to x. Let ϵ > 0 and choose
N ∈ Z with the property that |sn − x| < 12 ϵ for all n ≥ N . Then for every
+

positive integer n ≥ N + 1,
1
|an | = |sn − sn−1 | = |(sn − x) + (x − sn−1 )| ≤ |sn − x| + |x − sn−1 | < 2 · ϵ = ϵ.
2
We have verified the definition of “an → 0.” ❚
PN
3.31 Some experimentation suggests that n=0 n · n! = (N + 1)! − 1, which
PN
is easily confirmed by induction. Hence, | n=0 n · n! − (−1)|2 = |(N + 1)!|2 .
Since the power of 2 in (N + 1)! tends to infinity, |(N + 1)!|2 → 0, and
PN P∞
n=0 n · n! → −1. That is, n=0 n · n! = −1.
PN PN
Next, we look at n=0 n2 · 2n . Let F (T ) = n=0 T n . Differentiating and
PN
multiplying by T gives T F ′ (T ) = n
n=0 nT . Another round of the same
process yields
N
X
n2 T n = T (T F ′ (T ))′ = T 2 F ′′ (T ) + T F ′ (T ).
n=0

1−T N +1 1 1
Substituting in F (T ) = 1−T = 1−T − T N +1 1−T and simplifying,

N
X T (T + 1) GN (T )
n2 T n = + T N +1
n=0
(1 − T )3 (1 − T )3

for some GN (T ) ∈ Z[T ].

Plugging in T = 2, we deduce that
∗
Although it does tend to “0!”

68
 N
X
n2 2n − (−6) = 2−N −1 · |GN (2)|2 ≤ 2−N −1 .
n=0 2
P∞
We send N to infinity and conclude that n=0 n2 2n = −6 in (Q, | · |2 ).

3.32 With x0 := 117 ∈ Z(3) , each step of the displayed algorithm has the form
xn = dn + 3xn+1 , where dn ∈ {0, 1, 2} and xn+1 ∈ Z(3) .
Repeated substitution reveals that for each nonnegative integer n,

x0 = d0 + 3x1
= d0 + 3(d1 + 3x2 )
..
.
= d0 + 3(d1 + 3(d2 + · · · + 3(dn + 3xn+1 )))
= d0 + 3d1 + 32 d2 + · · · + 3n dn + 3n+1 xn+1 .

Therefore,
n
X
x0 − 3k dk = 3−n−1 |xn+1 |3 ≤ 3−n−1 .
k=0 3
P∞
Sending n to infinity, x0 = k=0 3k dk .
As shown in the problem statement, x1 = x7 , implying that the “digits” di
repeat in blocks of six starting from i = 1.

3.33 We modify the algorithm of Problem 3.32. This time x0 = 27 ∈ Q(5) , and
each dn ∈ {0, 1, 2, 3, 4} is chosen so that xn = dn + 5xn+1 for an xn+1 ∈ Q(5) .
Grinding this out,
2 −1
=1+5·
7 7 −6 −4
=2+5·
−1 −3 7 7
=2+5·
7 7 −4 −5
=3+5·
−3 −2 7 7
=1+5·
7 7 −5 −1
=0+5· .
−2 −6 7 7
=4+5·
7 7
Replicating the logic of the solution to Problem 3.32, we conclude that in
(Q, | · |5 ),
2
= 1 + 2 · 5 + 1 · 52 + 4 · 53 + 2 · 54 + 3 · 55 + 0 · 56 + 2 · 57 + . . . ,
7
where the “digits” follow the eventually periodic pattern 1, 2, 1, 4, 2, 3, 0.

69
Extra Exploration 7 (cf. Burger and Struppeck [1]).
P∞ Show that there is
a sequence of rational numbers {an } with the property that n=1 an converges to 0
with respect to | · |∞ and converges to 1/p with respect to | · |p for every prime p.

3.34 If | · | is non-Archimedean, a straightforward induction shows that

|m| ≤ 1 for all positive integers m. This handles the forward direction of the
equivalence.
Turning to the reverse implication, suppose that |2| ≤ 1. If n is any positive
integer, we get from Problem 1.6 that
 
n
≤ n whenever 0 ≤ k ≤ n.
k

So by Exercise 2.17, |x + y| ≤ (n(n + 1))1/n max{|x|, |y|} for all x, y ∈ K.

Sending n to infinity in this last inequality, |x + y| ≤ max{|x|, |y|}: That is,
| · | is non-Archimedean.
We can draw the same conclusion if 3 replaces 2. More generally, suppose m ≥ 2
and |m| ≤ 1.
Let n be a positive

P integer and write each associated binomial
coefficient nk in base m: nk = j≥0 dj mj , where each dj ∈ {0, 1, 2, . . . , m−1}
and the
dj are eventually zero. If J is the largest index for which dj = ̸ 0, then
2n > nk ≥ mj ≥ 2j . Hence, j < n and
 
n X X
≤ |dj ||m|j ≤ |dj | ≤ max{|0|, |1|, . . . , |m − 1|} · n.
k
0≤j<n 0≤j<n

This bound on | nk | is a suitable substitute for that of Problem 1.6 in the

argument of the last paragraph.

In summary: If m is an integer with m ≥ 2, and |m| ≤ 1, then | · | is non-
Archimedean.

3.35 Let |·| be a nontrivial non-Archimedean absolute value on Q. As noted in

the solution to Problem 3.34, |m| ≤ 1 for all positive integers m. In particular,
|p| ≤ 1 for all primes p. If equality holds for all p, then |m| = 1 for all m ∈ Z+
(apply the Fundamental Theorem of Arithmetic). But then Exercise 1.1(a,c)
allows us to deduce |x| = 1 for all nonzero x ∈ Q, contradicting that | · | is
nontrivial.

3.36 Since | · | is non-Archimedean, |k| ≤ 1 for all integers k.

Given relatively prime integers m and n, write 1 = am + bn with a, b ∈ Z
(Bézout). Then

1 = |am + bn| ≤ max{|a||m|, |b||n|} ≤ max{|m|, |n|} ≤ 1.

70
 Hence, max{|m|, |n|} = 1, so that either |m| = 1 or |n| = 1.

3.37 The units in O are precisely the x ∈ K × satisfying both |x| ≤ 1 and
|x−1 | ≤ 1. As |x−1 | = |x|−1 , the last two inequalities are satisfied simultane-
ously precisely when |x| = 1.
Let M be the collection of nonunits in O, so that M = D<1 (0). Clearly 0 ∈ M .
If x, y ∈ M , then |x + y| ≤ max{|x|, |y|} < 1, and so x + y ∈ M . Moreover, if
x ∈ M and r ∈ O, then |rx| = |r||x| ≤ |x| < 1, so that rx ∈ M . Hence, M is
an ideal of O. Since 1 ∈/ M , the ideal M is proper.
Let I be any proper ideal of O. If x ∈ I, then x cannot be a unit in O:
Otherwise I ⊇ xO = O. Thus, x ∈ M . Since this holds for all x ∈ I, we
conclude that I ⊆ M .
Thus, M is a proper ideal of O containing all proper ideals of O. So M cannot
itself be properly contained in a proper ideal of O; that is, M is maximal.

3.38 Each nonzero x ∈ Z(p) has the form ab where a, b ∈ Z and p ∤ b. If we

′ ′
factor a = pvp (a) a′ , then x = pvp (a) ab . Here vp (a) ≥ 0 and ab ∈ Z×
(p) . So we
have at least one decomposition of the desired form.
′
Uniqueness is easy: Suppose x = pv u = pv u′ with v, v ′ nonnegative integers
′
and u, u′ ∈ Z× ′
(p) . Then |u|p = |u |p = 1, so that p
−v
= |x|p = p−v . Hence,
v = v ′ . But then pv u = pv u′ , and u = u′ .

3.39 Let I be any nonzero ideal of Z(p) and choose a nonzero x ∈ I with
vp (x) minimal among nonzero elements of I. Set v = vp (x).
Claim: I = pv Z(p) .

Since x = pv u for some u ∈ Z× v

(p) , it is immediate that I ⊇ xZ(p) = p uZ(p) =
v
p Z(p) . To prove the reverse containment, we take an arbitrary y ∈ I and show
that y ∈ pv Z(p) . Clearly y = 0 belongs to pv Z(p) . If y =
̸ 0, write y = pvp (y) w
×
where w ∈ Z(p) . Then vp (y) ≥ v, and y = pv (pvp (y)−v w) ∈ pv Z(p) , finishing
the proof of the Claim.
It follows that Z(p) is a principal ideal domain (PID), with each of its ideals
somewhere in the infinite chain
(0) ⊊ · · · ⊊ pn Z(p) ⊊ · · · ⊊ p3 Z(p) ⊊ p2 Z(p) ⊊ pZ(p) ⊊ Z(p) .
(The containments are strict since p is a nonzero, nonunit element of the
domain Z(p) .) It is clear from this linear ordering that pZ(p) is the unique
maximal ideal of Z(p) .∗
∗
We could also have proved that pZ(p) is the unique maximal ideal of Z(p) by invoking
Problem 3.37: pZ(p) = {x ∈ Z(p) : |x|p < 1}.

71
3.40 Since x2 + 7 = 2m , the integer x must be odd. Hence, 8 | x2 + 7 = 2m
and m ≥ 3.
√
Since x is an odd number, x± 2 −7 ∈ R for both choices of sign. Furthermore,
√ √
recalling that α = 1+ 2 −7 and β = 1− 2 −7 ,
√ √
x + −7 x − −7
· = 2m−2 = (αβ)m−2
2 2
= αm−2 · β m−2 . (*)

We proceed by analyzing prime factorizations. First we show that α and β are

prime in R. Since R = Z[α], it is plain that every residue class mod αR has a
representative from Z. In fact, since 2 = αβ = 0 in R/αR, every class mod αR
is represented by 0 or 1. If every class is represented by 0, then R = αR, forcing
α to be a unit and contradicting that R× = {±1}. So R/αR is (isomorphic
to) Z/2. Since Z/2 is a domain, αR is a prime ideal of R, and α is a prime
element of R. An identical argument shows that β is prime in R.
Continuing, we argue that neither α nor β is a common divisor of the left-hand
√ √ √
factors in (*). If α is common divisor, then α | x+ 2 −7 − x− 2 −7 = −7. But
√
−7 = 2α − 1 = −1 in R/αR, rather than 0. A similar argument shows that
β is not a common divisor.
The two left-hand factors in (*) are nonunits. By unique factorization, each
is a (nonempty) product of the primes α and β (up to units). We also know,
from our work in √
the last paragraph,
√
that α appears in the prime factorization
x+ −7 x− −7
of only one of 2 and 2 , and similarly for β.
√ √
How can this be? x+ 2 −7 and x− 2 −7 must match up with αm−2 and β m−2 ,
up to order and associates. That is, for some ϵ = ±1 and some units η, η ′ of R,
√ √
x + ϵ −7 x − ϵ −7
= ηαm−2 , = η ′ β m−2 .
2 2

Multiplying the last two equations, x2 + 7 = ηη ′ 2m−2 , and so ηη ′ = 1. Since

η, η ′ ∈ {−1, 1}, in fact η = η ′ , and
√ √
x + ϵ −7 x − ϵ −7 √
η(αm−2 − β m−2 ) = − = ϵ −7 = ϵ(α − β).
2 2
Therefore,
αm−2 − β m−2
um−2 = = ϵη −1 ∈ {±1}.
α−β

The converse also holds: Every m with um−2 = ±1 gives rise√to an x with x2 +
7 = 2m . If um−2 = ±1, then αm−2 − β m−2 = ±(α − β) = ± −7. On the other

72
 √
hand, if we write αm−2 = x+y2 −7 for integers x and y, then (applying complex
√ √
conjugation) β m−2 = x−y2 −7 , so that αm−2 − β m−2 = y −7. Comparing
x2 +7y 2 x2 +7
expressions, y = ±1. Therefore, 2m−2 = αm−2 β m−2 = 4 = 4 , and
x2 + 7 = 2m .

Remarks.
(i) We didn’t need to know R was a UFD to execute our solution. After proving
that α and β are prime, we could have appealed to the following result, valid in
every integral Q domain: If a product U V factors as π1 · · · πk , with all πi prime∗ ,
then U = η i∈S πi and V = η i∈S ′ πi for some units η, η ′ and some partition
′Q

of {1, 2, . . . , k} into sets S and S ′ . (We allow S or S ′ to be empty.) Try to prove

this if you haven’t seen it before!
(ii) For completeness, we include a proof that R× = {±1}. It is obvious that ±1 ∈ R× .
What requires proof is that ±1 are the only elements of R× .
For each µ ∈ R, we define the norm of µ √ by Nµ = µµ̄, where the bar denotes
complex conjugation. Thus, if µ = 12 (a+b −7), then Nµ = 14 (a2 +7b2 ). Recalling
that a ≡ b (mod 2), we deduce that (a) Nµ is a nonnegative integer for every
µ ∈ R, with Nµ = 0 only when µ = 0. Furthermore, (b) for every µ, ν ∈ R,
N(µν) = µν · µν = µµ̄ · ν ν̄ = Nµ · Nν.
Suppose ε is a unit in R with inverse ε′ . Using (b), 1 = √ N(1) = N(εε ) =
′
′ ′ 1
Nε · Nε . From (a), Nε = Nε = 1. Writing ε = 2 (e + f −7), the equation
Nε = 1 translates √ into e2 + 7f 2 = 4. This forces f = 0 and e = ±2, so that
1
ε = 2 (±2 + 0 −7) = ±1.

References

1. E. B. Burger and T. Struppeck, Does ∞ 1

P
n=0 n! really converge? Infinite series
and p-adic analysis. Amer. Math. Monthly 103 (1996), 565–577.

∗
we really do mean prime, not merely irreducible!

73
74
Solutions to Set #4

(jT )k
4.41 We have 1 + eT + e2T + · · · + e(n−1)T =
P P
0≤j<n k≥0 k! =
P 
k Tk Tk
P P
k≥0 0≤j<n j k! = k≥0 Sk (n) k! .

4.42 Summing the finite geometric series,

X X enT − 1 enT − 1 T
ejT = (eT )j = = · T .
eT − 1 T e −1
0≤j<n 0≤j<n

Therefore, by Problem 4.41,

X Tk enT − 1 T
Sk (n) = · T
k! T e −1
k≥0
  
X nr+1 T r X Ts
=  Bs 
(r + 1)! s!
r≥0 s≥0
X Tk X (r + s)! r+1
= Bs n .
k! (r + 1)!s!
k≥0 r+s=k
r,s≥0

Tk
Comparing coefficients of k! gives
X (r + s)! r+1 X Bk−r k 
Sk (n) = Bs n = nr+1 .
(r + 1)!s! r+1 r
r+s=k 0≤r≤k
r,s≥0

Remark. If you stare carefully, you will notice this argument takes for granted the
identities ejT = (eT )j (for j = 0, 1, 2, 3 . . . ). Here what is important is not that these
identities hold for real numbers T (which is familiar from calculus), but that they
hold as identities of formal power series in the indeterminate T .

To effect a proof, write

!
X 1 X Tk X k
(eT )j = T k1 +···+kj = .
k1 ! · · · kj ! k! k1 , k 2 , . . . , k j
k1 ,...,kj ≥0 k≥0 k1 +···+kj =k

75
 k
= j k . Therefore, (eT )j =
P

By the multinomial theorem, k1 +···+kj =k k1 ,k2 ,...,kj
(jT )k
= ejT .
P
k≥0 k!

We could also have proved the required identities byP leveraging our prior knowledge
of the “real world.” Expand (formally) ejT − (eT )j = k≥0 ck T k for some coefficients
ck . The exact same transformations you use to put the left side into the form of the
right will show that
X
(ex )j − ejx = ck xk for all real numbers x.
k≥0

Here all manipulations with real numbers can be justified by citing absolute con-
vergence of the relevant series. (Operations on formal power series are defined to
mirror operations that can be performed on numerical series in the presence of
sufficiently good convergence.) Since (ex )j − ejx = 0 for all real numbers x, the series
P k
k≥0 ck x converges everywhere to 0. This forces each ck = 0, which in turn shows
that ejT = (eT )j formally. While gratuitous in this instance, the principle that an
identity of real numbers can (often) be transmogrified into an identity of formal
power series is frequently useful.

4.43 All of the claimed equalities are straightforward to verify, including the
invariance of T coth T under the substitution T 7→ −T (provided we accept
that e2T e−2T = 1, which can be proved by the method of the preceding
Remark). Writing down the power series for −T coth(−T ) and T coth(T ), we
conclude that
X (−2T )k X (2T )k
−T + Bk =T+ Bk .
k! k!
k≥0 k≥0

When k is odd and larger than 1, comparing coefficients of T k on both sides

shows that Bk = 0. When k = 1, the same reasoning gives −1 − 2B1 = 1 + 2B1 ,
leading to B1 = − 12 (as asserted in our table).

k
4.44 Starting from T coth T = T + k≥0 Bk (2Tk!) , divide by T and substitute
P

B0 = 1, B1 = − 12 and Bk = 0 for odd k > 1. This provides the expansion

d
claimed for coth T . That dT coth T = 1 − coth2 T can be checked directly.
To ease notation, write
1 X
coth T = + ck T 2k−1 .
T
k≥1

As ck and B2k share the same sign, we will be done if we show that
(−1)k+1 ck > 0 for each k ∈ Z+ . Assuming this claim fails, let K be the minimal
counterexample. For later use, note that K > 1 (since c1 = 2B2 = 13 > 0).
Differentiating the last displayed equation,

76
 d 1 X
coth T = − 2 + (2k − 1)ck T 2k−2 .
dT T
k≥1

2K−2
The right-hand Laurent series has T appearing with coefficient (2K −
d
1)cK . On the other hand, the coefficient of T 2K−2 in dT coth T is the same
2
, since dT coth T = 1 − coth2 T and K > 1.
as its coefficient in − coth TP d

P coefficient is −(2cK + i+j=K, i,j≥1 ci cj ). Therefore, (2K + 1)cK =

That
− i+j=K, i,j≥1 ci cj , and

(−1)K X 1 X
(−1)K+1 cK = ci cj = (−1)i+1 ci · (−1)j+1 cj > 0.
2K + 1 2K + 1
i+j=K i+j=K
i,j≥1 i,j≥1

We use in the last step that (−1)i+1 ci > 0 and (−1)j+1 cj > 0, since i and j
are positive integers smaller than K.

Remark. A more satisfying explanation for why the even-indexed Bernoulli numbers
alternate in sign is found in Euler’s formula
(2m)!
B2k = (−1)k+1 2ζ(2k) · ,
(2π)2m
where ζ(s) := n≥1 n1s is the Euler–Riemann zeta function. This remarkable relation
P
pins down B2k rather precisely as a real number. Indeed, when conjoined to Stirling’s
estimate on factorials, it implies that
|B2k |
lim √ k 2k
= 1.
k→∞ 4 πk( πe )
Unfortunately, Euler’s formula does not contain any obvious information about the
number-theoretic properties of B2k .

We will not prove Euler’s result here. Interested readers are referred to the exquisitely
written textbook of Ireland and Rosen for a characteristically elegant treatment [2,
pp. 231–232].

4.45 Information on the first ten partial sums is collected below.

k
P
n 1≤k≤n 2 /k

1 2
2 22
3 22 · 5/3
4 22 · 8/3
5 28 · 1/15
6 25 · 13/15
5
7 2 · 151/105
8 213 · 1/105
9 29 · 83/315
10 210 · 73/315

77
Already this limited data suggests that v2 ( 1≤k≤n 2k /k) tends to infinity
P
P∞ k
with n (equivalently, that P k=1 2 /k = 0 in (Q, | · |2 )). Being a bit bolder,
we might conjecture that v2 ( 1≤k≤n 2k /k) is bounded below by a function
ever-so-slightly smaller than n. For the resolution of both conjectures, see the
solution to Problem 11.130.

4.46 We need the following lemma.

Lemma. For each x ∈ Z(p) , there is a d ∈ {0, 1, 2, . . . , p−1} with x−d ∈ pZ(p) .

Proof. Write x = a/b where a, b ∈ Z and p ∤ b. Choose B ∈ Z with Bb ≡ 1

(mod p), and select d ∈ {0, 1, . . . , p − 1} with d ≡ aB (mod p). Then db ≡
aBb ≡ a (mod p), and x − d = p (a−db)/p b ∈ pZ(p) . ❚

With the lemma in hand, we can express r in the desired form by the algorithm
of Problems 3.32 and 3.33. Specifically, let x0 = r, and for n = 0, 1, 2, 3, . . . ,
select dn ∈ {0, 1, 2, . . . , p − 1} so that xn = dn + pxn+1 for some xn+1 ∈ Z(p) .
Then r = k≥0 dk pk .
P

Having shown existence we turn to uniqueness. Suppose r = k≥0 dk pk =

P
′ k ′ ′
P
k≥0 dk p with all dk , dk ∈ {0, 1, . . . , p − 1}. Assume {dk } and {dk } do not
coincide, and let k0 be the smallest nonnegative integer with dk0 = ̸ d′k0 . Then
′ ′ k
P
0 = r − r = k≥k0 (dk − dk )p , and
X
(d′k0 − dk0 )pk0 = (dk − d′k )pk .
k>k0

− d′k )pk , then sn → (d′k0 − dk0 )pk0 in (Q, | · |p ).

P
So if we set sn = k0 <k≤n (dk
That is,
|sn − (d′k0 − dk0 )pk0 |p → 0. (*)
However, |sn |p ≤ maxk0 <k≤n |(dk −d′k )pk |p ≤p −k0 −1
, while |(d′k0 −dk0 )pk0 |p =
p−k0 . So by “survival of the greatest,”

|sn − (d′k0 − dk0 )pk0 |p = p−k0

for every n > k0 , contradicting (*).

Finally we prove {dk } is eventually periodic. It suffices to show that in the
algorithm of Problems 3.32 and 3.33, there will always be nonnegative integers
m < n with xm = xn . In that case {dk } repeats, from k = m onwards, with
(not necessarily minimal) period n − m.
Write r = a/b where p ∤ b. We will base our proof on two claims:
(a) each xn ∈ 1b Z,

78
(b) each |xn |∞ ≤ M , where M := max{2, |r|∞ }.
Claim (a) is obvious when n = 0, since x0 = r = a/b. Suppose that xn ∈ 1b Z.
Then bpxn+1 = bxn − bdn ∈ Z. As xn+1 ∈ Z(p) , we also have |bpxn+1 |p =
|bp|p |xn+1 |p ≤ p−1 . So in fact bpxn+1 ∈ pZ, and xn+1 ∈ bp 1
(pZ) = 1b Z. This
gives (a). Since x0 = r, (b) is trivial when n = 0. If |xn |∞ ≤ M , then
1 1 1 1
|xn+1 |∞ = |xn − dn |∞ < (|xn |∞ + p) ≤ |xn |∞ + 1 ≤ M + 1 ≤ M.
p p 2 2

So we have (b) as well. From (a) and (b) it is easy to conclude: [−M, M ] has
finite intersection with 1b Z, so the xn cannot all be distinct.

4.47 Observe that

|xn | − |xn − x| ≤ |xn − (xn − x)| = |x| = |xn + (x − xn )| ≤ |xn | + |x − xn |.

Hence,
−|x − xn | ≤ |x| − |xn | ≤ |x − xn |.
Since |xn − x| → 0, the squeeze theorem implies that |x| − |xn | → 0. Therefore,
|xn | converges to |x|.
Suppose now that | · | is non-Archimedean and that xn → x, where x = ̸ 0. The
limit definition guarantees that |xn − x| < |x| for all sufficiently large n. For
these n, “survival of the greatest” yields |xn | = |(xn − x) + x| = |x|.

4.48 By Problem 3.34 we can choose a prime p with |p| < 1. Then whenever
n is an integer not divisible by p, the integers p and n are relatively prime and
Problem 3.35 tells us that |n|p = 1.
Write |p| = p−c with c > 0. Given x ∈ Q× , we can express x = pvp (x) a/b where
a and b are integers not divisible by p. Then

|x| = |p|vp (x) |a||b|−1 = |p|vp (x) = p−cvp (x) = (p−vp (x) )c = |x|cp .

Of course, we also have |0| = 0 = |0|cp . So | · | = | · |cp .

4.49 From Exercise 4.45, |r| > 1 for each integer r ≥ 2. So there are indeed
real numbers c, d > 0 with |2| = 2c and |3| = 3d .
Pm
If we write 2n = i=0 ϵi 3i , with each ϵi ∈ {0, 1, 2} and ϵm > 0, then
m
X |3|m+1 − 1
2cn = |2|n = |2n | ≤ |ϵi ||3|i ≤ max{|1|, |2|}
i=0
|3| − 1
max{|1|, |2|}|3| |2| · |3| dm
≤ · |3|m = 3 ,
|3| − 1 |3| − 1

79
 |2|·|3|
proving the claimed inequality with B := |3|−1 . As ϵm > 0, we have 2n ≥ 3m ,
and thus 3 ≤ 2 . Therefore, 2 ≤ B · 2 , and 2(c−d)n ≤ B. Since n may
dm dn cn dn

be taken arbitrarily large, it follows that c ≤ d.

We could have run the argument with the roles of 2 and 3 reversed. Writing 3n
in base 2 and reasoning analogously would lead to the inequality 3dn ≤ B ′ 3cn ,
|2|
where B ′ = |2|−1 . We would then conclude that d ≤ c. Thus, c = d.

4.50 The arguments of Problem 4.49 apply equally well with 3 replaced by
an arbitrary integer r ≥ 3. Writing |2| = 2c and |r| = rd , we find that for each
positive integer n,

2cn ≤ B · 2dn while rdn ≤ B ′ · rcn

|2|
for the constants B = max{|2|,...,|r−1|}|r|
|r|−1 and B ′ = |2|−1 . These two inequalities
imply c ≤ d and d ≤ c (respectively), yielding c = d.
It follows that |r| = rc for every integer r > 1. That equality holds trivially
when r = 1, and so | · | = | · |c on all of Z+ . Writing each x ∈ Q× in the form
x = ± ab where a, b ∈ Z+ , we deduce that

|x| = |a| · |b|−1 = ac b−c = (ab−1 )c = |x|c∞ .

Of course |0| = 0 = |0|c∞ as well, and so | · | = | · |c∞ .

Extra Exploration 8. For which positive real numbers c is | · |c∞ an absolute

value on Q? Now let p be prime. For which positive real numbers c is | · |cp an absolute
value on Q?

4.51 We begin with a simple but useful observation: If (K, | · |) is any valued
field, and x ∈ K, then

xn → 0 in K ⇐⇒ |xn − 0| → 0 in R ⇐⇒ |x| < 1.

Now suppose for a contradiction that | · | and | · |′ are equivalent absolute

values on K with | · | Archimedean and | · |′ non-Archimedean. By Problem
3.34, |2| > 1, and so | 12 | < 1. Hence, ( 12 )n → 0 in (K, | · |). Since | · | and | · |′ are
equivalent, ( 12 )n → 0 in (K, | · |′ ) as well, so that | 12 |′ < 1. But then |2|′ > 1,
contradicting that |2|′ = |1 + 1|′ ≤ max{|1|′ , |1|′ } = 1.

4.52 Let | · | be a nontrivial absolute value on Q. If | · | is Archimedean, then

| · | = | · |c∞ for some c > 0 (Problem 4.50). In this case, | · | is equivalent to
| · |∞ , since

xn → x in (Q, | · |) ⇐⇒ |xn − x| → 0
⇐⇒ |xn − x|∞ → 0 ⇐⇒ xn → x in (Q, | · |∞ ).

80
If | · | is non-Archimedean, then | · | = | · |cp for some prime p and some c > 0.
In this case | · | is equivalent to | · |p (same reasoning as displayed above).
It remains to show that none of | · |∞ , | · |2 , | · |3 , | · |5 . . . are equivalent. From
Exercise 4.51, | · |∞ is not equivalent to any of the others, since | · |∞ is
Archimedean while | · |p is non-Archimedean. If p and q are distinct primes,
then pn → 0 in (Q, | · |p ) but pn ̸→ 0 in (Q, | · |q ). So | · |p and | · |q are
inequivalent.

Extra Exploration 9 (classifying absolute values on Fp (T )). Fix a prime

p, and let Π denote the collection of all monic irreducible polynomials π ∈ Fp [T ].
Let | · | be an absolute value on Fp (T ).
(a) Suppose that |T | > 1. Prove that |F | = |T |deg a−deg b for all F = ab ∈ Fp (T )× .
(b) Now assume that |T | ≤ 1. Prove that |F | ≤ 1 for all F ∈ Fp [T ]. Assuming | · |
is nontrivial, show that there is a unique π ∈ Π with |π| < 1. Furthermore,
|F | = |π|vπ (F ) for all F ∈ Fp (T )× . (Recall that when F = π v ab for a, b ∈ Fp [T ]
coprime to π, we are setting vπ (F ) := v.)
(c) State and prove an Fp (T )-analogue of the assertion in Problem 4.52.

4.53 NO. Suppose x0 , . . . , xp are p + 1 (distinct) equidistant rational numbers.

Translating each by −x0 , we can assume x0 = 0. Next, scaling the xi by the
same power of p, we can assume that each xi ∈ Z(p) and that at least one of
them, say x1 , is not in pZ(p) . Then |x1 − x0 |p = |x1 |p = 1.
Each of x1 , . . . , xp is congruent modulo pZ(p) to one of 0, 1, 2, 3, . . . , p − 1 (see
the solution to Problem 4.46). Since the xi are equidistant from one another,

|xj |p = |xj − x0 |p = |x1 − x0 |p = 1

for all j = 1, 2, . . . , p. Therefore, each of x1 , . . . , xp is congruent to one of

1, 2, . . . , p − 1 (mod pZ(p) ). But then two of x1 , . . . , xp , say xi and xj , coincide
modulo pZ(p) (Pigeonhole Principle). For these two,
1
|xi − xj |p ≤ < 1 = |x1 − x0 |p .
p
Contradiction!

4.54 Notice that Hn − p1 Hm = 1≤k≤n k1 − 1≤r≤m pr 1

= 1≤k≤n, p∤k k1 ∈
P P P

Z(p) . Hence, |Hn − p1 Hm |p ≤ 1 ≤ |Hm |p < p|Hm |p = | p1 Hm |p . Now “survival

of the greatest” gives us
 
1 1 1
|Hn |p = Hm + Hn − Hm = Hm = p|Hm |p
p p p p p

4.55 Let p be a prime. Suppose we happen to have in hand a nonnegative

integer k with |Hm |p ≥ 1 for all m in the range pk ≤ m < pk+1 . Problem 4.54

81
allows us to conclude that |Hn |p ≥ p whenever pk+1 ≤ n < pk+2 . Repeating
the reasoning, |Hn |p ≥ p2 whenever pk+2 ≤ n < pk+3 . In general, |Hn |p ≥ pj
for all n ≥ pk+j (for j = 0, 1, 2, 3, . . . ). Therefore, |Hn |p → ∞.
One can check with a simple computer program that k = 2 satisfies our
hypothesis both when p = 3 and when p = 5. So |Hn |3 → ∞ and |Hn |5 → ∞.

Remark. For an in-depth examination of |Hn |p , see [1].

Several open questions persist regarding the numerators and denominators of the
harmonic numbers. Here is one that appears deceptively simple: Are there infinitely
many n for which the denominator of Hn (in lowest terms) is the least common
multiple of 1, 2, 3, . . . , n?

References

1. D. W. Boyd, A p-adic study of the partial sums of the harmonic series. Experi-
ment. Math. 3 (1994), 287–302.
2. K. Ireland and M. Rosen, A classical introduction to modern number theory,
second ed., Graduate Texts in Mathematics, vol. 84, Springer-Verlag, New York,
1990.

82
Solutions to Set #5

5.56 We check that Zp contains the multiplicative

Q∞ identity 1 = (1 mod p, 1 mod
p2 , 1 mod p3 , . . . ) of the ambient ring k=1 Z/pk and that Zp is closed under
subtraction and multiplication. The first requirement is clear. To prove the
second and third, suppose x, y ∈ Zp . Write x = (a1 mod p, a2 mod p2 , . . . )
and y = (b1 mod p, b2 mod p2 , . . . ), where ak ≡ ak+1 (mod pk ) and bk ≡
bk+1 (mod pk ) for all k = 1, 2, 3, . . . . Then ak − bk ≡ ak+1 − bk+1 (mod pk )
and ak bk ≡ ak+1 bk+1 (mod pk ), for all k = 1, 2, 3, . . . , These congruences
imply that x − y = (a1 − b1 mod p, a2 − b2 mod p2 , . . . ) and xy = (a1 b1 mod
p, a2 b2 mod p2 , . . . ) also belong to Zp , as desired.

5.57 Suppose x = (a1 mod p, a2 mod p2 , . . . ) and y = (b1 mod p, b2 mod

p2 , . . . ) are nonzero elements of Zp . Since x ̸= 0, there is an i for which pi ∤ ai .
Similarly, there is a j for which pj ∤ bj . We will prove xy ̸= 0 by showing that
the mod pi+j component of xy is nonzero, i.e., that pi+j ∤ ai+j bi+j .
The compatibility condition in the definition of Zp assures us that ai+j ≡ ai ̸≡ 0
(mod pi ). Similarly, bi+j ≡ bj ̸≡ 0 (mod pj ). Therefore,

vp (ai+j bi+j ) = vp (ai+j ) + vp (bi+j ) < i + j,

and pi+j ∤ ai+j bi+j .

5.58 For each n ∈ Z+ , the element n·1 of Zp has a nonzero mod pk component
whenever pk > n. In particular, n · 1 ̸= 0. Thus, Zp has characteristic 0, so
that we can (and henceforth will!) view Z as a subring of Zp , identifying each
integer n with n · 1 = (n mod p, n mod p2 , n mod p3 , . . . ).
The rest of the problem asks (essentially) for a description of Q ∩ Zp . It may
not be clear that this task even makes sense: To take this intersection, both Q
and Zp have to be viewed as sitting inside a common superset. What is that
set?
Fortunately, this question has an easy answer: Both Q and Zp are contained in
the fraction field of Zp . Taking this interpretation, if a, b ∈ Z with b nonzero,

83
a
b ∈ Q ∩ Zp precisely when there is a y ∈ Zp with by = a. When does this
happen?
We can assume, without loss of generality, that gcd(a, b) = 1. If p | b, there
is no hope of finding a y with by = a: The mod p component of by is 0 for
every y ∈ Zp , while the mod p component of a is nonzero (remember that
gcd(a, b) = 1, so that p cannot divide a). Suppose instead that p ∤ b. In this
case, we can choose integers Bk with bBk ≡ 1 (mod pk ), for k = 1, 2, 3, . . . .
Observing that bBk+1 ≡ 1 (mod pk+1 ), we find that

bBk+1 ≡ 1 ≡ bBk (mod pk ), and thus Bk+1 ≡ Bk (mod pk ).

Therefore y0 := (B1 mod p, B2 mod p2 , B3 mod p3 , . . . ) ∈ Zp . Furthermore,

by0 = (bB1 mod p, bB2 mod p2 , . . . ) = (1 mod p, 1 mod p2 , . . . ) = 1, so that
y = ay0 satisfies by = a.
We conclude that Q ∩ Zp is the set of rational numbers whose denominators, in
lowest terms, are not multiples of p. This is precisely our faithful and familiar
companion Z(p) . (Incidentally, this explains the term p-integral for elements
of Z(p) ; the p-integral rationals are the rational numbers that are also p-adic
integers.)
As for the final question: Yes, (4 mod 5, 34 mod 52 , 334 mod 53 , . . . ) ∈ Q ∩
Z5 . In fact, 3 · (4 mod 5, 34 mod 52 , 334 mod 53 , . . . ) = (12 mod 5, 102 mod
52 , 1002 mod 53 , . . . ) = (2 mod 5, 2 mod 52 , 2 mod 53 , . . . ) = 2.

5.59 Let u = (a1 mod p, , a2 mod p, . . . ) ∈ Zp . If u ∈ Z× p with inverse v =

(b1 mod p, b2 mod p2 , . . . ), then uv = 1 = (1 mod p, 1 mod p2 , . . . ). Comparing
mod p components, a1 b1 ≡ 1 (mod p). Thus, a1 is invertible mod p, which
implies that p ∤ a1 .
Conversely, suppose that p ∤ a1 . Since each ak ≡ ak−1 ≡ · · · ≡ a1 (mod p), all
of the ak are coprime to p. Choose integers bk with ak bk ≡ 1 (mod pk ), for
k = 1, 2, 3, . . . . Working modulo pk+1 ,

ak+1 bk+1 ≡ 1 ≡ ak bk ≡ ak+1 bk ,

so that bk+1 ≡ bk (mod pk ). Hence, v := (b1 mod p, b2 mod p2 , . . . ) ∈ Zp . It is

now straightforward to check that uv = 1, so that u ∈ Z×p (with v = u
−1
).
It remains to prove that p | a1 (in Z) ⇐⇒ p | u (in Zp ). Suppose p | u and
write u = pv, where v = (v1 mod p, v2 mod p2 , . . . ) ∈ Zp . Then a1 ≡ pv1 ≡ 0
(mod p), and p | a1 .
Conversely, suppose p | a1 . Since each ak ≡ a1 (mod p), p divides every ak .
Put vk = ak+1 /p for k = 1, 2, 3, . . . . Dividing the congruence ak+2 ≡ ak+1
(mod pk+1 ) through by p gives vk+1 ≡ vk (mod pk ), and so v := (v1 mod
p, v2 mod p2 , . . . ) ∈ Zp . Moreover,

84
 pv = (a2 mod p, a3 mod p2 , . . . ) = (a1 mod p, a2 mod p2 , . . . ) = u.

(We use here that ak+1 ≡ ak (mod pk ) for each k.) Thus, p | u.

5.60 Let x = (a1 mod p, a2 mod p2 , . . . ) be a nonzero element of Zp and

choose the nonnegative integer v minimally with av+1 ̸≡ 0 (mod pv+1 ). For
each integer j > v,

aj ≡ aj−1 ≡ · · · ≡ av+1 ≡ 0 (mod pv ).

(Consider two cases, v = 0 or v > 0.) So bk := p−v av+k ∈ Z for each k =

1, 2, 3, . . . . The congruences av+k+1 ≡ av+k (mod pv+k ) imply that bk+1 ≡ bk
(mod pk ). Therefore, u := (b1 mod p, b2 mod p2 , . . . ) ∈ Zp .
av+1
By the choice of v, we have that p ∤ pv = b1 . Thus, u ∈ Z×
p (Exercise 5.59).
Furthermore,

pv u = (0 mod p, . . . , 0 mod pv , a2v+1 mod pv+1 , a2v+2 mod pv+2 , . . . )

= (0 mod p, . . . , 0 mod pv , av+1 mod pv+1 , av+2 mod pv+2 , . . . ) = x.

(Here we use that a2v+i ≡ a2v+i−1 ≡ · · · ≡ av+i (mod pv+i ).) So we have
produced a representation x = pv u with u ∈ Z×
p.
′
To prove uniqueness, suppose x = pv u = pv u′ with v, v ′ nonnegative integers
v ′ −v ′
and u, u′ ∈ Z× ′ v
(p) . Assume v ≤ v . Canceling p from both sides, u = p u.
×
Since u ∈ Zp , to avoid a contradiction with Exercise 5.59 we must have
v ′ − v = 0, i.e., v = v ′ . Then pv u = pv u′ , leading to u = u′ .

5.61 For each nonzero x ∈ Zp , define vp (x) as the integer v appearing in the
representation of x described in Problem 5.60. Then proceed as in the solution
to Problem 3.39.

5.62 We preface the solution with some philosophical comments. With the
definition we have given of Zp , there is an obvious informal way to reduce x
modulo powers of p: Write x = (a1 mod p, a2 mod p2 , . . . ) and reduce mod pn
by extracting the nth component, i.e., singling out an mod pn . This defines
the reduction of x mod pn as an element of Z/pn . Since Zp is a ring, there is
also a canonical way to reduce x modulo pn : Take the image of x in Zp /pn Zp .
It would be reassuring if these two ways of reducing mod pn were somehow the
same. As we will see shortly, this turns out to be the case! The isomorphism
between Z/pn and Zp /pn Zp induced by the inclusion Z ,→ Zp matches up
an mod pn and x mod pn Zp . Equivalently: x mod pn Zp = an mod pn Zp .

85
 There is no justice in Heaven or Earth,
but there is certainly justice in
MATHEMATICS!

Paul Erdős

We proceed to the proof proper. Let ϕ denote the homomorphism from Z to

Zp /pn Zp induced by the inclusion Z ,→ Zp . We start by computing the kernel
of ϕ. For x ∈ Z,

x ∈ ker(ϕ) ⇐⇒ x ∈ pn Zp ⇐⇒ x/pn ∈ Zp ∩ Q ⇐⇒ x/pn ∈ Z(p) ⇐⇒ x ∈ pn Z.

(Here we have recalled that Q ∩ Zp = Z(p) .) So ker(ϕ) = pn Z, and ϕ induces

an embedding ϕ̃ : Z/pn ,→ Zp /pn Zp .

Next we show that ϕ, and hence also ϕ̃, is surjective. Let x = (a1 mod
p, a2 mod p2 , . . . ) ∈ Zp . The mod pj -component of an − x is an − aj mod pj ,
which vanishes for all j ≤ n (by the compatibility condition in the definition
of Zp ). So by our solution to Problem 5.60, either an − x = 0 or an − x = pk u
for an integer k ≥ n and a unit u ∈ Z× n
p . In either case, an − x ∈ p Zp , so that
ϕ(an ) = an mod pn Zp = x mod pn Zp . This proves ϕ is surjective, and so ϕ̃ is
an isomorphism.
The isomorphism ϕ̃ carries an mod pn to ϕ(an ) = x mod pn Zp , as we promised
in our initial remarks.

5.63 Keeping in mind

Q the Chinese Remainder Theorem, it is straightforward
to check that Zg ∼
= p|g Zpvp (g) , via the map sending

(a1 mod g, a2 mod g 2 , a3 mod g 3 , . . . )

7→ ((a1 mod pvp (g) , a2 mod p2vp (g) , a3 mod p3vp (g) , . . . ))p|g .

“Straightforward” means that the tedium of writing out the proof outweighs
the enlightenment gained from doing so. (Please walk through the argument
in your head and decide if you agree!)
To complete the problem, it suffices to show that if p is prime and v ∈ Z+ ,
then Zpv ∼= Zp . Here we can map (b1 mod pv , b2 mod p2v , b3 mod p3v , . . . ) to
(b1 mod p, . . . , b1 mod pv , b2 mod pv+1 , . . . , b2 mod p2v , . . . ), each bi repeated
v times. (Again, that this work is “straightforward.” Verify!)

5.64 The outer, red disc represents the entirety of Z3 . The salmon-colored
discs partition Z3 into three parts, based on the mod 3 component. Once the
mod 3 component is fixed, there are 3 possibilities for the mod 32 component,
depicted by the green discs. Finally, having fixed the mod 32 component, there
are three possibilities for the mod 33 component, illustrated by the purple

86
 discs. In theory this partitioning process could continue indefinitely, but at
some point our eyes would ask for a break!

5.65 Who am I to tell you how to appreciate art?

Remark. For further discussions around visualizing Zp , see [3], [9], [10, Chapter 2],
and [11, Chapter 1, §2].

5.66 Let f be any function from Z+ to Zp . We will show f cannot be onto by

adapting Cantor’s famous diagonal argument. Choose a residue class a1 mod p
different from the mod p component of f (1). The class a1 mod p lifts to p
residue class mod p2 ; choose one, say a2 mod p2 , different from the mod p2
component of f (2). This in turn lifts to p classes mod p3 ; choose one, say
a3 mod p3 , different from the mod p3 component of f (3). Continue. At the end
of the (very long) day, we find that (a1 mod p, a2 mod p2 , a3 mod p3 , . . . ) is
an element of Zp that cannot belong to the image of f , since it has a different
mod pn component from f (n) for each n = 1, 2, 3, . . . .

Extra Exploration 10. Show that Zp has the same cardinality as R. A useful
tool for this kind of proof is the Cantor–Schröder–Bernstein theorem: If there are
injections from A to B and from B to A, then there is a bijection between A and B.∗

5.67 Let (xr , yr ), (xb , yb ), and (xg , yg ) be the coordinates of the red, blue,
and green vertices of ∆. Then, by a formula well-known to those who know it
well,

xb yb 1
2 · Area(∆) = ± xg yg 1 = ±(xb yg + xr yb + xg yr − xg yb − xb yr − xr yg ).
xr yr 1

We claim that xb yg has 2-adic absolute value strictly larger than the other five
terms. By definition of the coloring, |xb yg |2 = |xb |2 |yg |2 ≥ 1 · 1 = 1. Next, we
observe that
|xr |2 |yb |2 ≤ |xr |2 |xb |2 < |xb |2 ≤ |xb |2 |yg |2 .
If yr = 0, then |xg yr |2 = 0 < |xb yg |2 ; otherwise,

|xg |2 |yr |2 < |yg |2 |yr |2 < |yg |2 ≤ |xb |2 |yg |2 .

If yb = 0, then |xg yb |2 = 0 < |xb yg |2 ; otherwise,

∗
The author cannot resist outlining Cox’s straightforward-to-verify proof of the
Cantor–Schröder–Bernstein result [2]. First, reduce to showing that if f is an injection
from a set X to a subset YS ⊆ X,(n)then there is a bijection between X and Y . To tackle
this new claim, put S = ∞ n=0 f (X \ Y ) = (X \ Y ) ∪ f (X \ Y ) ∪ f (f (X \ Y )) ∪ . . . .
Define g : X → Y by letting g(x) = f (x) for x ∈ S and g(x) = x for x ∈ / S. Argue
directly that g is well-defined, injective, and surjective.

87
 |y|2

|x|2

Coloring of Q2 in Problem 5.67.

|xg |2 |yb |2 < |yg |2 |yb |2 ≤ |xb |2 |yg |2 .

Next,
|xb |2 |yr |2 < |xb |2 ≤ |xb |2 |yg |2 .
Finally,
|xr |2 |yg |2 < |yg |2 ≤ |xb |2 |yg |2 .
Phew!
Falling back on “survival of the greatest,” we conclude that |2 · Area(∆)|2 =
|xb yg |2 ≥ 1, so that |Area(∆)|2 ≥ |2|−1
2 > 1.

Remark. Here is a quick sketch of Monsky’s proof. We can assume the square we
are trying to dissect is S = [0, 1] × [0, 1] ⊆ R2 .

Color Q2 as described in Problem 5.67. A triangle with vertices from Q2 will be

called a rainbow triangle if each vertex receives a different color.

Suppose S has been dissected into finitely many triangles. So that we can bring
the coloring of Q2 into the picture, let’s suppose that the vertices of all triangles
appearing in the dissection have rational coordinates.

Using a version of Sperner’s lemma from combinatorial geometry, along with basic
properties of our coloring, Monsky argues that the dissection must contain an odd
number of rainbow triangles. In particular, there is always at least one rainbow
triangle ∆ involved. As we showed above, |Area(∆)|2 > 1. So if we assume there are

88
n triangles involved in the dissection, each with the same area, then |1/n|2 > 1; that
is, n is even.

This is all well and good, but we are interested in arbitrary (real) dissections of S,
not merely “rational” ones. Well, here is the astounding part: The same argument
applies in the general case! At first glance this claim doesn’t seem to make sense:
Our initial coloring was defined in terms of | · |2 , and | · |2 has domain Q, not R. It
turns out, however, that it is possible to extend | · |2 to an absolute value on R. (This
is far from obvious!) With this extension in hand, everything else in the argument
goes through with zero change.

A complete proof of Monsky’s theorem can be found in [1, Chapter 22]. The discussion
in [1] sidesteps extending | · |2 to an absolute value on R; the reader interested in
seeing such an extension constructed can consult §14 of [12].

Extra Exploration 11 (cf. Hales and Straus [4]). Let K be an infinite field
that can be endowed with a nontrivial, non-Archimedean absolute value. Show that
it is possible to 3-color the affine plane K 2 , using all three colors in an interesting
way, so that every line receives points of at most two colors. In an interesting way
means that each color is assigned to three non-collinear points. (If we remove “in an
interesting way,” this can be done for every infinite field. Can you see why?)

Extra Exploration 12 (Hensel’s definition of the p-adic integers). Let

ϕ : Z[[T ]] → ∞ k 2
Q
k=1 Z/p be the map sending a0 + a1 T + a2 T + . . . to (a0 mod p, a0 +
2 2 3
a1 p mod p , a0 + a1 p + a2 p mod p , . . . ). Here the kth component of the output is
a0 + a1 p + · · · + ak−1 pk−1 mod pk .
(a) Check that ϕ is a ring homomorphism.
(b) Show that ϕ(Z[[T ]]) = Zp .
If you squint just right, you can see Hensel defining the p-adic integers in [8] as the quotient
Z[[T ]]/ ker(ϕ). This claim must be taken cum grano salis; even the notion of a “ring” was still
up in the air when [8] was written. Nevertheless, it gets at the essence of Hensel’s description.
(c) Show that ker(ϕ) = (T − p)Z[[T ]].

Remark. It seems that neither the definition in [8], nor Hensel’s earlier descriptions
in [6, 7] (similar but less refined), were viewed by Hensel’s contemporaries as entirely
rigorous. In his obituary [5] for Hensel, Hasse describes the p-adic numbers as “a
genuine creation driven by intuition and imagination, which, initially, like every
revolutionary idea, was thrown down bluntly and in raw form and which, similar to
Leibniz’s differential calculus, initially lacked a solid logical foundation.” Apprehen-
sions about the logical standing of p-adic numbers were dispelled only after Kürschák
and Ostrowski’s papers on valuation theory, which began to appear around the same
time as [8].

References

1. M. Aigner and G. M. Ziegler, Proofs from The Book, sixth edition, Springer,
Berlin, 2018.

89
 2. R. H. Cox, A proof of the Schroeder-Bernstein theorem. Amer. Math. Monthly
75 (1968), 508.
3. A. A. Cuoco, Visualizing the p-adic integers. Amer. Math. Monthly 98 (1991),
355–36.
4. A. W. Hales and E. G. Straus, Projective colorings. Pacific J. Math. 99 (1982),
31–43.
5. H. Hasse, Kurt Hensel zum Gedächtnis. J. Reine Angew. Math. 187 (1949),
1–13.
6. K. Hensel, Neue Grundlagen der Arithmetik. J. Reine Angew. Math. 127 (1904),
51–84.
7. , Theorie der algebraischen Zahlen, B. G. Teubner, Leipzig and Berlin,
1908.
8. , Zahlentheorie, G. J. Göschen, Berlin and Leipzig, 1913.
9. J. E. Holly, Pictures of ultrametric spaces, the p-adic numbers, and valued fields.
Amer. Math. Monthly 108 (2001), 721–728.
10. S. Katok, p-adic analysis compared with real, Student Mathematical Library,
vol. 37, American Mathematical Society, Providence, RI; Mathematics Advanced
Study Semesters, University Park, PA, 2007.
11. A. M. Robert, A course in p-adic analysis, Grad. Texts in Math., vol. 198,
Springer-Verlag, New York, 2000.
12. W. H. Schikhof, Ultrametric calculus: An introduction to p-adic analysis, Cam-
bridge Studies in Advanced Mathematics, vol. 4, Cambridge University Press,
Cambridge, 2006.

90
Solutions to Set #6

6.68 Suppose {xn } is Cauchy. To prove that |xn+1 − xn | → 0, let ϵ > 0

be given, and select N ∈ Z+ so that |xn − xm | < ϵ for all positive integers
n, m ≥ N . Then |xn+1 − xn | < ϵ whenever n ≥ N . (This direction of the proof
does not use that | · | is non-Archimedean.)
Conversely, suppose that |xn+1 − xn | → 0. To prove {xn } is Cauchy, let ϵ > 0,
and choose N ∈ Z+ with |xn+1 − xn | < ϵ whenever n ≥ N . If we assume that
n > m ≥ N , the strong triangle inequality yields

|xn − xm | = |(xn − xn−1 ) + (xn−1 + xn−2 ) + · · · + (xm+1 − xm )|

≤ max{|xn − xn−1 |, . . . , |xm+1 − xm |} < ϵ.

The cases where m > n ≥ N are interchangeable with these, since |xn − xm | =
|xm − xn |. Finally, the inequality |xn − xm | < ϵ is trivial for n = m.

n n+1 n
6.69 Let xn = 25 . By Euler’s theorem, xn+1 − xn = xn · (25 −5 − 1) =
n+1
xn · (2φ(5 ) − 1) ≡ xn · 0 ≡ 0 (mod 5n+1 ). Thus, |xn+1 − xn |5 ≤ 5−n−1 , which
tends to 0, and {xn } is a Cauchy sequence by Exercise 6.68.

6.70 Choose N ∈ Z+ so that |xn − xm | < 1 whenever n, m ≥ N . Then for all

n ≥ N , we have |xn − xN | < 1, so that

|xn | = |(xn − xN ) + xN | ≤ |xn − xN | + |xN | ≤ 1 + |xN |.

As a consequence, each |xn | ≤ max{|x1 |, |x2 |, . . . , |xN −1 |, 1 + |xN |}.

6.71 The argument is the same as in calculus. Suppose {xn } is Cauchy and
that the subsequence {xnk } converges to x.
Given ϵ > 0, choose N1 ∈ Z+ with the property that |xk − xℓ | < 12 ϵ whenever
k, ℓ ≥ N1 . Choose N2 ∈ Z+ so that |xnk − x| < ϵ for all k ≥ N2 . If k ≥
max{N1 , N2 }, then
1 1
|xk − x| = |(xk − xnk ) + (xnk − x)| ≤ |xk − xnk | + |xnk − x| < ϵ + ϵ = ϵ.
2 2

91
(We use here that nk ≥ k ≥ N1 .) Hence, xk → x.

6.72 Since p−1 ∈ Qp , both n≥0 p−n Zp and Zp [1/p] are contained in Qp .
S

Both also contain 0, so we will be done if we show that both contain Q×

p.

pv u
By Exercise 5.60, each x ∈ Q×
p can be written as pv ′ u ′
for some nonnegative
v−v ′ ′
integers v, v ′ and some u, u′ ∈ Z× . Then x = p uu′−1 ∈ pv−v Zp . Since
′ S p
pv−v Zp is a subset of both n≥0 p−n Zp and Zp [1/p], it follows that x is
contained in both sets.

6.73 That x has such a representation is implicit in our solution to Problem

′
6.72. To prove uniqueness, suppose x = pv u = pv u′ , with v, v ′ ∈ Z and
u, u ∈ Zp . Choose w ∈ Z large enough that both v + w and v ′ + w are
′ ×

nonnegative, and apply the already-known uniqueness statement from Problem

′
5.60 to pw x = pv+w u = pv +w u′ .

6.74 Both definitions assign vp (0) = ∞, so it is enough to prove our two

definitions of vp (x) agree for x ∈ Q× . Write x = pv ab where a and b are integers
not divisible by p. Then vp (x) = v if we go by the Set #1 definition. To show
vp (x) = v with our new definition, it suffices to prove that ab is a unit in
Zp . Thankfully, this is easy: Since p ∤ ab, we have both ab ∈ Z(p) ⊆ Zp and
( ab )−1 = ab ∈ Z(p) ⊆ Zp . (See Problem 5.58 for the containment Z(p) ⊆ Zp .)
Thus, ab ∈ Z× p.

It remains to show that | · |p is a non-Archimedean absolute value on Qp .

Property (i) in the absolute value definition (see Set #1) is clear. Property
(iii) is straightforward: We can assume x, y are nonzero. Write x = pvp (x) u
and y = pvp (y) u′ , where u, u′ ∈ Z×p . Then xy = p
vp (x)+vp (y)
uu′ , and uu′ ∈ Z× p.
−vp (x)−vp (y)
Hence, vp (xy) = vp (x)+vp (y), so that |xy|p = p = p−vp (x) p−vp (y) =
|x|p |y|p . Finally, we turn our attention to the strong triangle inequality: |x +
y|p ≤ max{|x|p , |y|p }. For the proof we may assume that x, y, and x + y
are all nonzero. As above, write x = pvp (x) u and y = pvp (y) u′ . Assuming
(as we may) that vp (x) ≤ vp (y), we find that x + y = pvp (x) w for some
w ∈ Zp . Furthermore, w is nonzero by our assumption that x + y ̸= 0. Write
w = pvp (w) u′′ . Then x + y = pvp (x)+vp (w) u′′ , and |x + y|p = p−vp (x)−vp (w) ≤
p−vp (x) = max{p−vp (x) , p−vp (y) } = max{|x|p , |y|p }.

6.75 First, we argue that for x ∈ Qp : x ∈ Zp ⇐⇒ |x|p ≤ 1.

We may assume that x ∈ Q× p . Write x = p
vp (x)
u where u ∈ Z×
p . When
vp (x) ≥ 0, it is clear that x ∈ Zp , as both pvp (x) and u belong to Zp . On the
other hand, Problem 5.60 gives vp (x) ≥ 0 for all nonzero x ∈ Zp . Therefore,

x ∈ Zp ⇐⇒ vp (x) ≥ 0 ⇐⇒ p−vp (x) ≤ 1 ⇐⇒ |x|p ≤ 1,

92
and Zp = {x ∈ Qp : |x|p ≤ 1}.
Next, observe that for each nonzero x ∈ Qp ,

x ∈ Z×
p ⇐⇒ vp (x) = 0 ⇐⇒ p
−vp (x)
= 1 ⇐⇒ |x|p = 1.

Hence, Z×
p = {x ∈ Qp : |x|p = 1}.

6.76 We address only the part after “thus”; the earlier claims in the problem are
consequences of the Pigeonhole Principle. By construction, ak+1 ≡ ak (mod pk )
for every k, so that x := (a1 mod p, a2 mod p2 , . . . ) ∈ Zp . Furthermore, for
each k, there are infinitely many n ∈ Z+ where the mod pk component of xn
is ak mod pk .
Choose n1 ∈ Z+ so that the mod p component of xn1 is a1 mod p. Then choose
n2 > n1 where the mod p2 component of xn2 is a2 mod p2 , then n3 > n2 where
the mod p3 component of xn3 is a3 mod p3 , etc.
We argue that xnk → x in (Qp , | · |p ). Let k ∈ Z+ . By construction, the mod
pk component of xnk − x vanishes. By the compatibility condition baked into
the definition of Zp , the mod p, mod p2 , . . . , mod pk−1 components also
vanish. Therefore (see the solution to Problem 5.60), either xnk − x = 0 or
xnk −x = pv u for an integer v ≥ k and a u ∈ Z×p . In either case, vp (xnk −x) ≥ k,
−k −k
and |xnk − x|p ≤ p . Since p → 0, we conclude that xnk → x, as required.

Remark. Grouchy experts may complain that Problem 6.76 establishes the sequen-
tial compactness of Zp , not compactness with today’s accepted meaning in general
topology (cf. [1]). But those same experts will know that for metric spaces — such
as Zp — the two concepts coincide. So what exactly are they complaining about?

Such grumblers may find the following Extra Exploration more to their tastes.

Extra Exploration 13 (Zp is compact, take two).

(a) A (possibly infinite) number of Ross Program counselors are stationed at points
of Zp . Each counselor is responsible for the campers within a certain positive
radius of their position, as measured by | · |p . Presently, all of Zp is monitored:
each point of Zp lies within the assigned radius of some counselor. Show that all
but finitely many of the counselors can leave to purchase talent show supplies
while Zp remains fully monitored (without the need to relocate the remaining
counselors).
(b) What happens if we replace every occurrence of Zp by Z? Assume distances are
still measured by | · |p for some prime p.

j
1
eT /j
Q
6.77 Let’s accept for the time being the identity j≥1 = 1−T in Q[[T ]].
1
The series for 1−T has T p -coefficient (in fact, every coefficient) equal to 1. To
see what the coefficient of T p looks like on the left side of our identity, we
expand

93
 ∞ 
Tj 1 T 2j 1 T 3j
Y 
1+ + 2
+ + ...
j=1
j 2! j 3! j 3
X X 1 1
= Tk .
e1 !e2 !e3 ! · · · 1 2 3e3 · · ·
e 1 e 2
k≥0 e1 +2e2 +3e3 +···=k
all ei ≥ 0

All but two of the tuples e1 , e2 , . . . with e1 + 2e2 + 3e3 + · · · = p have

1 1
e1 !e2 !e3 !··· 1e1 2e2 3e3 ··· ∈ Z(p) . The two exceptions are (1) the tuple with e1 = p
and all other ei = 0 and (2) the tuple with ep = 1 and all other ei = 0, which
1
make respective contributions of p! and p1 . Equating coefficients of T p leads to
1 1
the conclusion that 1 − ( p! + p ) ∈ Z(p) , as claimed.
1 1
Since 1 also belongs to the ring Z(p) , we infer that p! + p ∈ Z(p) . But

1 1 (p − 1)! + 1
+ = = p|(p − 1)! + 1|p .
p! p p p! p

Consequently, |(p − 1)! + 1|p ≤ 1/p. This translates to p | (p − 1)! + 1, or

(p − 1)! ≡ −1 (mod p).

Remark. It is a little tricky to write down a rigorous proof of the formal identity
Q T j /j 1
j≥1 e = 1−T . We sketch one argument, which, however, involves some ‘cheating’;
specifically we draw on the theory of complex variables, which is not a prerequisite
for the rest of the text.
j
For each J ∈ Z+ , let FJ (T ) = 1≤j≤J eT /j (as a formal power series). Reasoning as
Q
j
in the Remark to Problem 4.42, FJ (z) = 1≤j≤J ez /j for all complex z. As J → ∞,
Q
1
the functions FJ (z) converge uniformly to 1−z on every compact subset of the open
unit disc |z| < 1. So by Cauchy’s integral formula, if we write FJ (T ) = k≥0 ak,J T k ,
P
then for each fixed k,
Z J→∞
Z
1 1 1
ak,J = FJ (z)z −k−1 dz −−−→ z −k−1 dz = 1.
2πi |z|=9/10 2πi |z|=9/10 1 − z

(Here we recognized the final integral as the coefficient of T k in 1−T

1
= 1+T +T 2 +. . . .)
Since ak,J = ak,k for all J ≥ k, we conclude that ak,J = 1 for all J ≥ k. In particular:
For each fixed k, the kth coefficient of FJ (T ) eventually stabilizes at the kth coefficient
j
1
. This is precisely what it means to say that j≥1 eT /j = 1−T 1
Q
of 1−T as formal
power series.

6.78 Write ap−1 = 1 + pqp (a), bp−1 = 1 + pqp (b). Multiplying, (ab)p−1 =
1 + p(qp (a)qp (b) + pqp (a)qp (b)), so that

(ab)p−1 − 1
qp (ab) = = qp (a) + qp (b) + pqp (a)qp (b) ≡ qp (a) + qp (b) (mod p).
p

94
References

1. M. Raman-Sundström, A pedagogical history of compactness. Amer. Math.

Monthly 122 (2015), 619–635.

95
96
Solutions to Set #7

7.79 Let {xn } be a Cauchy sequence in Zp . By Problem 6.76, {xn } has a

subsequence {xnk } converging to some x ∈ Zp . By Problem 6.71, xn → x.

7.80 Let {xn } be a Cauchy sequence in Qp . We know from Exercise 6.70 that
{|xn |} is bounded. Hence, if k is sufficiently large, then each |pk xn | ≤ 1, i.e.,
{pk xn } is a sequence in Zp . Since {xn } is Cauchy, so is {pk xn }. By Problem
7.79, pk xn → x for some x ∈ Zp . Then xn → p−k x.
n
7.81 Let xn = 25 . By Problem 6.69, {xn } is a Cauchy sequence in Z5 .
Invoking Exercise 7.79, xn → x for some x ∈ Zp .
By the product rule for limits, lim x5n = (lim xn )5 = x5 . On the other hand,
x5n = xn+1 , so that lim x5n = lim xn+1 = x. Hence, x = x5 .
n 1
As shown in the solution to Problem 6.69, xn = 25 ≡ 25 ≡ 2 (mod 5) for all
n. Therefore, |xn − 2|5 ≤ 15 for all n, and (see Problem 4.47)
1
|x − 2|5 = | lim (xn − 2)|5 = lim |xn − 2|5 ≤ .
5
̸ 0, and x = x5 tells us 1 = x4 . That
Thus, x ≡ 2 (mod 5Z5 ). In particular, x =
is, x is a 4th root of 1, as claimed.
The only 4th roots of 1 belonging to Q are 1 and −1. Neither is congruent to
2 modulo 5Z5 . Therefore, x ∈
/ Q.

7.82 Let x ∈ Zp . For each k ∈ Z+ , there is an integer ak with x ≡ ak

(mod pk Zp ). (If we view x as an infinite tuple, we can take for ak any integer
representing the mod pk component of x; see the solution to Problem 5.62.)
Then |ak − x|p ≤ p−k , which tends to 0 as k → ∞. So {ak } is a sequence of
integers converging to x.

7.83 Let x ∈ Qp . Choose k ∈ Z with pk x ∈ Zp . By Problem 7.82, we can

find a sequence of integers ak converging to pk x. Then p−k ak is a sequence of
rational numbers converging to x.

97
7.84 We define a candidate isomorphism ϕ : L → L′ as follows. Since K is
dense in L, for each x ∈ L there is a sequence {xn } in K such that xn → x in
L. Since {xn } converges in L, {xn } is Cauchy in L. Then {xn } is also Cauchy
in L′ , as each xn ∈ K and | · | and | · |′ extend the same absolute value on K.
′
Since (L′ , | · |′ ) is complete, {xn } converges in L′ . We define ϕ(x) = lim(L ) xn ,
where the superscript indicates that the limit is taken in L′ .
Honor demands we check that ϕ(x) depends only on x and not on the particular
{xn }. Suppose {xn } and {x̃n } are two sequences in K both converging to x in
L. Then xn − x̃n → 0 in L. As each xn − x̃n ∈ K, and | · | and | · |′ extend the
same absolute value on K, it must be that xn − x̃n → 0 in L′ . Consequently,
′ ′
lim(L ) xn = lim(L ) x̃n . Vindication!
If x ∈ K, we can compute ϕ(x) by taking each xn = x. This shows that
′
ϕ(x) = lim(L ) x = x. So ϕ fixes K.
Let x, y ∈ L and choose sequences {xn }, {yn } in K such that xn → x in L
and yn → y in L. Then xn + yn → x + y in L, and
′ ′ ′
ϕ(x + y) = lim(L ) (xn + yn ) = lim(L ) xn + lim(L ) yn = ϕ(x) + ϕ(y).

Similarly, ϕ(xy) = ϕ(x)ϕ(y), confirming that ϕ is a homomorphism.

A homomorphism between fields is always injective. To establish surjectivity,
take any x′ ∈ L′ . Since K is dense in L′ , there is a sequence {xn } in K
converging to x′ in L′ . Since {xn } converges in L′ , it is Cauchy in L′ and hence
′
in L as well. Define x ∈ L by x = lim(L) xn . Then ϕ(x) = lim(L ) xn = x′ .
Thus far we have shown ϕ is an isomorphism. To prove ϕ is isometric, let
x ∈ L, and consider a sequence {xn } in K converging to x in L. Then {xn }
converges to ϕ(x) in L′ . Given ϵ > 0, we choose N ∈ Z+ such that, for all
n ≥ N,
|xn − x| < ϵ and |xn − ϕ(x)|′ < ϵ.
Then, for n ≥ N ,

|x| ≤ |xn | + ϵ = |xn |′ + ϵ < |ϕ(x)|′ + 2ϵ,

and
|x| ≥ |xn | − ϵ = |xn |′ − ϵ ≥ |ϕ(x)|′ − 2ϵ.
Hence |x| and |ϕ(x)|′ are within 2ϵ of each other. Since this holds for each
ϵ > 0, it must be that |x| = |ϕ(x)|′ .

Extra Exploration 14. Show that the map ϕ described above is the unique
isometric isomorphism from L to L′ .

98
Remark. Every valued field (K, | · |) admits a completion (which we have just seen
is then unique up to isometric isomorphism). The usual way to show a completion
exists is to mimic one of the standard constructions of R from Q, due to Cantor and
Méray: Take the ring of Cauchy sequences in K and quotient by the maximal ideal
of sequences tending to 0. For details, see for instance [3, §1.3]. Those with more
exotic tastes might enjoy a recent, very different argument by Kionke [4].

Pn
7.85 Let sn = k=0 ck pk . Then |sn+1 − sn |p = |cn+1 pn+1 |p ≤ p−n−1 . As
p−n−1 → 0, the sequence {sn } is Cauchy (see Exercise 6.68). By Problem
P∞ 7.79,
sn has a limit in Zp . This is precisely what it means to say that k=0 ck pk
converges to an element of Zp .

7.86 Our selection process guarantees that xk+1 ≡ x ≡ xk (mod pk Zp ) and

thus xk+1 ≡ xk (mod pk ) (this last congruence holding in Z). It also ensures
that xk is the least nonnegative representative of its residue class mod pk .
Therefore, xk+1 ≥ xk , and ck = p1k (xk+1 − xk ) ≥ 0. Since xk+1 < pk+1 , we
also have ck < pk+1 /pk = p. Hence, ck ∈ {0, 1, . . . , p − 1}.
Clearly, xk → x, since |xk − x|p ≤ p−k . Moreover,

c0 +c1 p+· · ·+ck−1 pk−1 = (x1 −x0 )+(x2 −x1 )+· · ·+(xk −x1 ) = xk −x0 = xk

for every k ∈ Z+ . Sending k to infinity, c0 + c1 p + c2 p2 + · · · = lim xk = x.

Extra Exploration 15. Show that

P if p is an odd prime, then every x ∈ Zp has a
unique representation in the form ∞ k
k=0 dk p , where the dk are integers from the
interval (−p/2, p/2). For which x is this expansion terminating? For which x is it
eventually periodic?

Extra Exploration 16 (Knopfmacher and Knopfmacher [5]). Prove that

every x ∈ 1 + pZp has a unique product representation x = (1 + p)e1 (1 + p2 )e2 (1 +
p3 )e3 · · · , where e1 , e2 , e3 , . . . ∈ {0, 1, 2, . . . , p − 1}.

Extra Exploration 17 (Pigeons hard at work; Mahler [6]). Let x ∈ Zp .

In this problem we look for prescribed patterns of digits in the base p expansions of
the integer multiples of x.
(a) Let m, n ∈ Z+ . Prove that there are integers A, B with A ̸= 0, −pm ≤ A ≤ pm ,
and 0 ≤ B < pn satisfying Ax ≡ B (mod pm+n Zp ).
Suggestion. Consider Ax − B mod pm+n Zp for 1 ≤ A ≤ pm and 0 ≤ B < pn . If 0 mod
pm+n Zp is represented, you are done. Otherwise, start stuffing pigeons in holes.
(b) Taking m = 1 in (a), show that one of the 2p p-adic integers ±x, ±2x, . . . , ±px
has infinitely many ‘0’ digits in its base p expansion. More generally, for each
m ∈ Z+ , one of the 2pm p-adic integers ±x, ±2x, . . . , ±pm x has infinitely many
runs of m consecutive zeros in its base p expansion.
(c) Assume x is an irrational element of Zp . (That is, x ∈ Zp \ Z(p) .) Let m ∈ Z+ ,
and let d0 , d1 , d2 , . . . , dm−1 ∈ {0, 1, . . . , p − 1}. Show that there is a nonzero

99
 integer k such that the following holds: The digits d0 , d1 , . . . , dm−1 appear, in
that order, infinitely often in the base p expansion of kx.
cj pj , then cj = d0 , cj+1 = d1 , . . . , cj+m−1 =
P
The last bit means that if we expand kx = j≥0
dm−1 for infinitely many nonnegative integers j. For the proof, consider multiples of k0 x,
where k0 x has long runs of zeros as in (b).
(d) Assume x is an irrational element of Zp . Let m ∈ Z+ . Show that there is a
positive integer k′ such that every sequence of m base p digits appears infinitely
often in the base p expansion of k′ x.
Hint. Can you do this with a not-necessarily-positive (but nonzero) k′ ? If so, what happens
when you replace k′ with −k′ ?

7.87 Start with any x ∈ Qp and select k ∈ Z with pk x ∈ Zp . Take the infinite
base p expansion of pk x constructed in Problem 7.86 and scale by p−k to
obtain an expansion of x satisfying (i)–(iii).
If there were two expansions of x possessing properties (i)—(iii), scaling both
by the same sufficiently large power of p would produce an element of Zp with
two different base p expansions.

7.88 The forward direction is clear. Now suppose x = 0 or x ∈ Q+ with

denominator a power of p. Write x = a/pk , where a and k are nonnegative
integers. Then the canonical expansion of x is obtained by taking the usual
base p expansion of a and scaling by p−k ; this is obviously terminating.

7.89 We record a simple observation for later use: For each ℓ ∈ Z+ , the
series 1 + pℓ + p2ℓ + . . . converges to 1/(1 − pℓ ) in (Qp , | · |). We omit the
straightforward proof (compare with Exercise 3.30).
Suppose x has an eventually periodic canonical expansion with (not necessarily
minimal) period ℓ. Scaling x by aPsuitable power of p (which does not affect
rationality), we can assume x = k≥0 ck pk where ck = ck+ℓ for all k ≥ k0 .
k k k
P P P
Then x = 0≤k<k0 ck p + k≥k0 ck p . Clearly, 0≤k<k0 ck p ∈ Q. Less
trivially,
X X
ck pk = lim ck pk
m→∞
k≥k0 k0 ≤k<k0 +mℓ
X
= lim ck pk (1 + pℓ + · · · + p(m−1)ℓ )
m→∞
k0 ≤k<k0 +ℓ
X
= ck pk (1 + pℓ + p2ℓ + . . . )
k0 ≤k<k0 +ℓ
X pk
= ck ∈ Q.
1 − pℓ
k0 ≤k<k0 +ℓ

Therefore, x ∈ Q.

100
Turning to the converse: We have already shown that each x ∈ Z(p) has
an eventually periodic base p expansion. To obtain an eventually periodic
canonical expansion for an arbitrary x ∈ Q, scale x by pk to place pk x ∈ Z(p) ,
then rescale by p−k .
P∞
Open problem: Is the p-adic number k=0 k! rational for some prime p?

Extra Exploration 18 (Casacuberta [1]). Show that the p-adic number

∞
pvp (k!) has a canonical expansion containing arbitrarily long (but finite) runs
P
k=0
of zeros, for every prime p. Deduce that ∞ vp (k!)
P
k=0 p ∈
/ Q.

P∞
Extra Exploration 19 (Dragovich [2]). Disprove: k=0 k! converges to the
same rational number in Qp for all primes p.

Pp−1 (p−1)!p−1 −1
7.90 By Problem 6.78, a=1 qp (a) ≡ qp ((p − 1)!) ≡ p (mod p). So
(p−1)!p−1 −(p−1)!−2
the claimed congruence is equivalent to p ≡0
(mod p), or
(p − 1)!p−1 − (p − 1)! ≡ 2 (mod p2 ). Using Wilson’s theorem, write (p − 1)! =
−1 + pr with r ∈ Z. Then, working modulo p2 ,
p−1  
X p−1
(p − 1)!p−1 = (−1)p−1−j (pr)j ≡ 1 + (p − 1)(−1)pr ≡ 1 + pr,
j=0
j

so that (p − 1)!p−1 − (p − 1)! ≡ (1 + pr) − (−1 + pr) ≡ 2, as desired.

7.91 Let us argue that whenever we have a solution to x2 = 2 in Z/7k , say

ak mod 7k , we can lift it uniquely — by the process described — to a solution
ak+1 mod 7k+1 in Z/7k+1 . Expanding, (ak +7k q)2 ≡ a2k +2·7k ak q (mod 7k+1 ).
This last expression is congruent to 2 modulo 7k+1 precisely when

2 · 7k ak q ≡ 2 − a2k (mod 7k+1 ).

By assumption, 7k | 2 − a2k . So the preceding congruence is equivalent to

2 − a2k
2ak q ≡ (mod 7).
7k
Both 2 and ak are coprime to 7. (Since a2k ≡ 2 (mod 7), we cannot have ak ≡ 0
(mod 7).) Thus, the last displayed congruence uniquely determines the residue
class of q modulo 7. Picking any q from this class and defining ak+1 = ak + 7k q
yields our desired lift. Note that since q is uniquely determined mod 7, our lift
is uniquely determined as a residue class mod 7k+1 .
Assume now that a1 mod 7, a2 mod 72 , a3 mod 73 , . . . have been determined
by the above procedure. Let x = (a1 mod 7, a2 mod 72 , a3 mod 73 , . . . ). Then

101
 x ∈ Z7 as each ak+1 is a lift of ak . By construction, x2 = (2 mod 7, 2 mod
72 , 2 mod 73 , . . . ) = 2.
We still need to verify that the canonical expansion of x is as stated. The
initial steps of the algorithm, starting with 3 mod 7, are described in the
problem statement. We lifted 3 mod 7 to 3 + 1 · 7 mod 72 , and we lifted that
to 3 + 1 · 7 + 2 · 72 mod 73 . If we run the algorithm one more round, we arrive
at 3 + 1 · 7 + 2 · 72 + 6 · 73 mod 74 (check this!). We can read off from here that
the 7-adic expansion of x starts as 3 + 1 · 7 + 2 · 72 + 6 · 73 + . . . .

Remark. Gauss left us in 1855, four decades before Hensel’s first publication on
the p-adic numbers. It is therefore rather remarkable that one finds in his Nachlass
(manuscripts left behind at death) the claim that
√
5 (mod 11∞ ) = 9.0.4.10.4.4.

This seems to be Gauss’s way of expressing that in Z11 ,

√
5 = 4 + 4 · 11 + 10 · 112 + 4 · 113 + 0 · 114 + 9 · 115 + . . . .

I owe this historical nugget to math StackExchange user user2554.∗

References

1. S. Casacuberta, Irrationality of the sum of a p-adic series. Unpublished.

arXiv:1710.11484 [math.NT]
2. B. Dragovich, On p-adic power series. In: p-adic functional analysis (Poznań,
1998), Lecture Notes in Pure and Appl. Math., vol. 207, Dekker, New York, 1999,
pp. 65–75.
3. S. Katok, p-adic analysis compared with real, Student Mathematical Library,
vol. 37, American Mathematical Society, Providence, RI; Mathematics Advanced
Study Semesters, University Park, PA, 2007.
4. S. Kionke, Constructing the completion of a field using Quasimorphisms, p-Adic
Numbers Ultrametric Anal. Appl. 11 (2019), 335–337.
5. A. Knopfmacher and J. Knopfmacher, A binomial product representation for
p-adic numbers, Arch. Math. (Basel) 52 (1989), 333–336.
6. K. Mahler, On the digits of the multiples of an irrational p-adic number. Proc.
Cambridge Philos. Soc. 76 (1974), 417–422.

∗
https://math.stackexchange.com/a/4877205

102
Page from Gauss’s Nachlass; originally scanned by the Göttinger Digital-
isierungszentrum.
103
104
Solutions to Set #8

8.92
(a) In any complete valued field, a sequence {sn } converges ⇐⇒ {sn } is
Cauchy. The forward implication was noted on Set #6 (for every valued
field, not necessarily complete). The backward implication is the definition
of completeness.
P∞
Let k=1 ak be a series in K, a field assumed to P be complete withPrespect
n ∞
to a non-Archimedean absolute value. Put sn = k=1 ak . Then k=1 ak
converges ⇐⇒ {sn } converges ⇐⇒ {sn } is Cauchy ⇐⇒ sn+1 − sn → 0
⇐⇒ an+1 → 0 ⇐⇒ an → 0.
P∞
(b) If we assume k=1 ak converges to s, then |sn | → |s| (see Problem
4.47). By the strong inequality, each |sn | ≤ max{|a1 |, |a2 |, . . . , |an |} ≤
maxk=1,2,3,... |ak |. (That the last max exists is ensured by |ak | tending to
0.) Hence, |s| = lim |sn | ≤ maxk=1,2,3,... |ak |.

Extra Exploration 20 (convergence of infinite products). Let K be a

field complete with respect to a non-Archimedean absolute value. Let {ak } be a
sequence in K.
(a) Suppose that ak → 0. Show that ∞
Q
k=1 (1 + ak ) is a well-defined element of
K
Q∞ (meaning that that the partial products have a limit in K). Furthermore,
k=1 (1 + ak ) = 0 if and only if some ak = −1.
(b) Conversely, prove that if ∞
Q
k=1 (1 + ak ) defines a nonzero element of K, then
ak → 0.

P∞ P∞
8.93 Let k=1 bk be any rearrangement of Pk=1 ak ; say bk = P aσ(k) for all
n n
k = 1, 2, 3, . . . , where σ ∈ Sym(Z+ ). Put sn = k=1 ak and tn = k=1 bk . By
assumption, sn → s, while our task is to prove that tn → s.
Given ϵ > 0, select N0 ∈ Z+ with |an | < ϵ whenever n > N0 . Then choose
N ∈ Z+ as the maximum of σ −1 (1), . . . , σ −1 (N0 ). Let us argue that |tn −s| < ϵ
whenever n ≥ N .
Let n ≥ N . By our choice of N , all of a1 , a2 , . . . , aN0 appear among
b1 , b2 , . . . , bN , so that

105
 X X X
tn − sN0 = bk − ak = aσ(k) ,
1≤k≤n 1≤k≤N0 1≤k≤n, σ(k)>N0

and
|tn − sN0 | ≤ max{|aσ(k) | : 1 ≤ k ≤ n, σ(k) > N0 } < ϵ.
Moreover,
X
|s − sN0 | = ak ≤ max |ak | < ϵ.
k>N0
k>N0

Hence,

|tn − s| = |(tn − sN0 ) − (s − sN0 )| ≤ max{|tn − sN0 |, |s − sN0 |} < ϵ,

as desired.
Pn Pn Pn
8.94 Put sn = k=0 ak , tn = k=0 bk , and un = k=0 ck . Let s = lim sn
and t = lim tn be the infinite sums of the ak and bk , respectively. We must
show that un → st.
Observe that X X
sn tn = ak bℓ + ak bℓ
0≤k,ℓ≤n 0≤k,ℓ≤n
k+ℓ≤n k+ℓ>n

and that the first of the two right-hand sums satisfies

X X X X X X
ak bℓ = ak bℓ = ak br−k = cr = un .
0≤k,ℓ≤n 0≤r≤n k+ℓ=r 0≤r≤n 0≤k≤r 0≤r≤n
k+ℓ≤n k,ℓ≥0
P
Let en be the previously neglected sum, en = 0≤k,ℓ≤n, k+ℓ>n ak bℓ . Using
A and B to denote upper bounds on {|ak |} and {|bk |} respectively (which
certainly exist, since ak , bk → 0), we find that

|en | ≤ max |ak bℓ | ≤ B max |ak | + A max |bk |,

0≤k,ℓ≤n, k+ℓ>n n/2<k≤n n/2<ℓ≤n

which tends to 0. Hence, en → 0, and un = sn tn − en → st, as desired.

P
8.95 For each fixed P j ai,j converges, since |ai,j | ≤ ϵj and ϵj → 0
i, the series P
as j → ∞. Put si = j ai,j . Then i si also converges, since |si | ≤ max{|ai,j | :
j ≥ 0} ≤ ϵi , and
P P ϵi → 0 as i → ∞. Thus, we have obtained convergence of the
double
P P series i j ai,j . A symmetric argument demonstrates the convergence
of j i ai,j .

8.96 We start by noting that all of the double series appearing here converge or
involve finitely many terms. P
This follows from Problem 8.95. ToPtake
P one exam-
N P
ple (the others are similar), i=0 j ai,j can be rewritten as i j ai,j 1i≤N .

106
Since |ai,j 1i≤N | ≤ |ai,j |, the sufficient condition for convergence furnished by
Problem 8.95 is satisfied with the same sequence {ϵN }.
We can now establish our four inequalities. To attack the first of these, notice
that

XX N X
X ∞
X X X
ai,j − ai,j = ai,j ≤ max ai,j .
i≥N +1
i j i=0 j i=N +1 j j

Each term ai,j appearing P in the final sum on j has i ≥ N + 1, and thus
|ai,j | ≤ ϵN +1P
. Hence, | j ai,j | ≤ maxj |ai,j | ≤ ϵN +1 for each i ≥ N + 1, and
maxi≥N +1 | j ai,j | ≤ ϵN +1 . So we have the first inequality.
Analogous reasoning shows that

XX N
XX ∞
X X ∞
X
ai,j − ai,j = ai,j ≤ max ai,j ≤ ϵN +1 ,
j
j i j i=0 j i=N +1 i=N +1

N X
X N X
X N N
X ∞
X ∞
X
ai,j − ai,j = ai,j ≤ max ai,j ≤ ϵN +1 ,
0≤i≤N
i=0 j i=0 j=0 i=0 j=N +1 j=N +1

and
N
XX N X
X N ∞
X N
X N
X
ai,j − ai,j = ai,j ≤ max ai,j ≤ ϵN +1 .
j≥N +1
j i=0 j=0 i=0 j=N +1 i=0 i=0

Now look at the first column of inequalities in the problem statement. These
inequalities, in conjunction with the strong triangle inequality, imply that

XX N X
X N
ai,j − ai,j ≤ ϵN +1 .
i j i=0 j=0

Similarly, we obtain from the second column of inequalities that

XX N X
X N
ai,j − ai,j ≤ ϵN +1 .
j i j=0 i=0

PN PN PN PN
Since i=0 j=0 ai,j = j=0 i=0 ai,j (it is always OK to swap finite sums!),
a final application of the strong triangle inequality yields

XX XX
ai,j − ai,j ≤ ϵN +1 .
i j j i

107
 P P P P
Send N to infinity to conclude that i j ai,j = j i ai,j .

8.97 If p − 1 | k, then nk ≡ 1 (mod p) for all of n = 1, 2, . . . , p − 1, so that

Sk (p) ≡ p − 1 (mod p). Thus, p | Sk (p) + 1 = Sk (p) + 1p−1|k .
If p − 1 ∤ k, choose an integer g that generates the multiplicative group mod p.
Working modulo p,
X X X
g k Sk (p) = g k nk ≡ (gn)k ≡ mk ≡ Sk (p),
0≤n<p 0≤n<p m mod p

using that multiplication by g permutes the residue classes modulo p. As

g k ̸≡ 1 (mod p), we infer that Sk (p) ≡ 0 (mod p). Therefore, p | Sk (p) =
Sk (p) + 1p−1|k .

pk pj
8.98 By Faulhaber’s formula, Skp(p) = k+1 + Bk + 0<j<k kj Bk−j j+1
P

.
Rearranging,
1p−1|k X k  pj Sk (p) + 1p−1|k pk
Bk + + Bk−j = − .
p j j+1 p k+1
0<j<k

It will suffice to argue that both right-hand terms belong to Z(p) .

The first is in Z (Problem 8.97), so certainly also in Z(p) . To handle the second,
notice that pk ≥ 2k ≥ k + 1 for each k ∈ Z+ . Therefore, vp (k + 1) ≤ k, and
pk pk
vp ( k+1 ) = k − vp (k + 1) ≥ 0. That is, k+1 ∈ Zp .

8.99 Suppose the claim is false and let k be the smallest positive integer for
which Bk + p−1 1p−1|k ∈ / Z(p) . Let S be the sum on j appearing in Problem
8.98, so that Bk + p−1 1p−1|k + S ∈ Z(p) . For use momentarily, we rewrite
X k  pj−1
S= pBk−j .
j j+1
0<j<k

By the minimality of k, we have Bk−j + p−1 1p−1|k−j ∈ Z(p) for all j in the
j−1
range 0 < j < k. Hence, pBk−j ∈ Z(p) for all these j. Furthermore, pj+1 ∈ Z(p)
in this same range of j: To see why, note that pj ≥ 3j > j + 1, so that
j−1
vp (j + 1) < j. Since vp (j + 1) is an integer, vp (j + 1) ≤ j − 1, and vp ( pj+1 ) ≥ 0.
It follows that each term in our rewritten expression for S belongs to Z(p) , so
that S itself belongs to Z(p) . But if S ∈ Z(p) and Bk + p−1 1p−1|k + S ∈ Z(p) ,
then Bk + p−1 1p−1|k ∈ Z(p) . This contradicts the choice of k.

8.100 This is similar to Problem 8.99. Assuming the claim fails, let k be the
smallest even positive integer for which Bk + 12 ∈
/ Z(2) . From Exercise 8.98,
1
Bk + 2 + S ∈ Z(2) , where

108
 X k  2j−1
S= 2Bk−j .
j j+1
0<j<k

Let us argue that each term in this expression for S is in Z(2) (and thus,
S ∈ Z2 ). Let 0 < j < k. If j is odd, then Bk−j = 0 unless j = k − 1. The
k−2
k
2B1 2 k = −2k−2 , which is indeed in Z(2) . (We


j = k − 1 term of S is k−1
recalled here that B1 = − 12 .) If j is even, the minimality of k ensures that
j−1
2Bk−j ∈ Z(2) . As 2j+1 is also in Z(2) (clear, as j + 1 is odd), the jth term is
2-integral in the even case as well.
Since S ∈ Z(2) and Bk + 12 + S ∈ Z(2) , we are forced to have Bk + 1
2 ∈ Z(2) .
This contradicts the choice of k.

8.101 Let k be a positive even integer. From the last two problems we have
that for every prime p,
1p−1|k
Bk + ∈ Z(p) .
p
Put X 1
B̂k := Bk + .
p
p: p−1|k

Then for every prime p,

 
1p−1|k X 1
B̂k = Bk + + ∈ Z(p) .
p ℓ
ℓ prime, ℓ̸=p
ℓ−1|k
T
Hence, B̂k ∈ p Z(p) = Z.

Remark. Let Dk denote the denominator of Bk in lowest terms. The Clausen–von

Staudt theorem implies that when k is a positive even integer, Dk is the product of
the primes p for which p − 1 | k. (So in particular Dk is a multiple of 6 for all even
k > 0.) It was realized by Erdős and Wagstaff [2] that this characterization of Dk
allows one to establish strong statistical results. For example, enlisting methods from
analytic number theory — specifically, the field known as the “anatomy of integers”
— they showed that if D = Dk for some positive even k, then the limit
#{even positive n ≤ x: Dn = D}
pD := lim
x→∞ #{even positive n ≤ x}
exists and is positive. That is, any D appearing as a denominator of an even-indexed
Bernoulli number actually appears with a well-defined, positive limiting frequency.
Furthermore, X
pD = 1.
D≥1

More recent work on the distribution of the pD can be found in the article [3] of
Pomerance and Wagstaff; among other things, they re-prove (in stronger form) a

109
theorem of Sunseri that 6 is the most popular denominator among even-indexed
Bernoulli numbers.

8.102 Since u is a unit in Zp , it is also a unit modulo every power of pZp .

n
Setting xn = up , Euler’s theorem yields
n+1
−pn n+1
xn+1 − xn = xn (up − 1) = xn (uφ(p )
− 1) ≡ 0 (mod pn+1 Zp ),

for each n = 1, 2, 3, . . . . (We may appeal to Euler’s theorem here since the
unit groups of Zp mod powers of pZp are “the same” as the unit groups of Z
mod powers of p, by Problem 5.62.) Hence, |xn+1 − xn |p ≤ p−n−1 for each n,
and {xn } is a Cauchy sequence in Zp .
Let x = lim xn . Since lim xn+1 = lim xpn = (lim xn )p = xp , and lim xn+1 =
lim xn = x, it follows that xp = x.
n n−1
By Fermat’s little theorem, each xn = up ≡ up ≡ · · · ≡ up ≡ u (mod pZp ),
Therefore, x ≡ u (mod pZp ) (cf. the solution to Problem 7.81), and in partic-
ular, x ̸= 0. Thus, we have shown that ω(u) := x is a solution to ω(u)p−1 = 1
with ω(u) ≡ u mod pZp .
To prove ω(u) is the unique (p−1)th root of unity congruent to u mod pZp , it is
enough to argue that no residue class mod pZp contains two different (p − 1)th
roots of unity. This is easy: The (p − 1)th roots of unity ω(1), . . . , ω(p − 1)
belong to different residue classes mod pZp . And these are all of the (p − 1)th
roots of unity, since the degree p − 1 polynomial xp−1 − 1 cannot have more
than p − 1 roots in Qp .

8.103 In Q3 , the element −1 is a (3 − 1)th root of unity congruent to 2

modulo 3Z3 . Therefore, ω(2) = −1.
n+1 n
Let p = 7. As seen in the solution to Problem 8.102, 27 ≡ 27 (mod 7n+1 Z7 )
for each n ∈ Z+ . Hence, for all integers k ≥ 3,
k k−1 2
27 ≡ 27 ≡ · · · ≡ 27 (mod 73 Z7 ),

from which it follows that

k 2
ω(2) = lim 27 ≡ 27 (mod 73 Z7 ).

This information is enough to compute c0 , c1 , and c2 , provided we are willing

to get our hands (or computing devices) a little dirty: 27 ≡ 128 (mod 73 ), and
2
27 ≡ 1287 ≡ 324 (mod 73 ). So the base 7 expansion of ω(2) begins the same
way as that of 324 = 2 + 4 · 7 + 6 · 72 . In other words, the desired digits are
c0 = 2, c1 = 4, and c2 = 6.

8.104 We begin with a lemma valid over an arbitrary field of characteristic

not equal to 2.

110
 Lemma. Let F be a field of characteristic not equal to 2. If ζ is an element
of order 3 in F × , then 1 + ζ is an element of order 6.

Proof. By assumption, ζ is a zero of T 3 − 1 but not T − 1. Hence, ζ is a root of

T 3 −1 2 2
T −1 = T + T + 1 ∈ F [T ], and 1 + ζ = −ζ . Since char(F ) ̸= 2, the element
−1 ∈ F × has order 2, while ζ 2 has order 3. Since 2 and 3 are relatively prime,
1 + ζ = (−1) · ζ 2 has order 2 · 3 = 6. ❚

Now assume that u ∈ Z has order 3 mod p. There are no elements of order 3
in F×
2 . Hence, p is odd, and we can invoke our lemma to deduce that 1 + u
has order 6 in F×
p . In particular, p ≡ 1 (mod 6), as the order of each element
in F×
p necessarily divides p − 1; this congruence will be needed shortly.

Continuing, observe that

n n
ω(u)3 = ( lim up )3 = lim (u3 )p = ω(u3 ).
n→∞ n→∞

Here ω(u3 ) is the unique (p − 1)th root of unity congruent to u3 mod pZp . But
1 is a (p − 1)th root of unity, and 1 ≡ u3 (mod pZp ). Thus, 1 = ω(u3 ) = ω(u)3 .
If ω(u) = 1, then u ≡ ω(u) ≡ 1 (mod pZp ), contradicting that u has order 3
modulo p. So ω(u) is an element of order 3 in Q× p and, applying the lemma a
second time, 1 + ω(u) is an element of order 6.
In particular, 1 + ω(u) is a (p − 1)th root of unity (since 6 | p − 1). Furthermore,
1 + ω(u) ≡ 1 + u (mod pZp ). Therefore, 1 + ω(u) = ω(1 + u).

Extra Exploration 21 ((Z/p)× is cyclic, revisited, or sledgehammer

swats fly ). For this exercise (only), you are asked to “forget” the general result
that finite subgroups of multiplicative groups of fields are always cyclic.∗
(a) Show that the mapping u mod p 7→ ω(u) sets up an isomorphism between (Z/p)×
and the group µp−1 of (p − 1)th roots of unity in Qp . This map is known as the
Teichmüller lift.
(b) Let ζ = e2πi/(p−1) ∈ C. Explain why the minimal polynomial of ζ over Q divides
T p−1 − 1. Then deduce from (a) that said minimal polynomial splits over Qp .
Conclude that Q(ζ) embeds into Qp .
(c) Prove that the image of ζ under any of the embeddings in (b) generates µp−1 .
Hence, µp−1 and (Z/p)× are cyclic.
This unusual argument is due to Matt Baker [1].
∗
“Please forget everything you have learned at school, because you haven’t learned
it. Please keep in mind everywhere the corresponding portions of your school work,
because you haven’t actually forgotten them.” — Edmund Landau

111
References

1. M. Baker, Primitive roots, discrete logarithms, and p-adic numbers. URL: https:
//mattbaker.blog/2013/11/07/primitive-roots-discrete-logarithms-and
-the-interplay-between-p-adic-and-complex-numbers/
2. P. Erdős and S. S. Wagstaff, Jr., The fractional parts of the Bernoulli numbers.
Illinois J. Math. 24 (1980), 104—112.
3. C. Pomerance and S. S. Wagstaff, Jr., The denominators of the Bernoulli numbers.
Acta Arith. 209 (2023), 1–15.

112
Solutions to Set #9

9.105 Let a ∈ Z× 2 2
p . If a = x for some x ∈ Qp , then |x|p = |a|p = 1. Therefore,
|x|p = 1; in particular, x ∈ Zp . Reducing the equation x2 = a modulo the ideal
pZp shows that a mod pZp is a square in Zp /pZp . Now recall from Exercise
5.62 that the inclusion of Z into Zp induces an isomorphism between Z/p and
Zp /pZp and that this isomorphism identifies a1 mod p with a mod pZp . Since
a mod pZp is a square in Zp /pZp , it follows that a1 mod p is a square in Z/p.
Conversely, suppose a ∈ Z× p and that a1 mod p is a square in Z/p. Then
a mod pZp is a square in Zp /pZp and we can choose r1 ∈ Zp with r12 ≡ a
(mod pZp ). We construct a Zp -solution to x2 = a following the iterative
procedure of Problem 7.91.
Let k ∈ Z+ . Suppose we have a mod pk Zp -solution rk mod pk Zp to x2 = a;
we lift this to a mod pk+1 Zp solution. Expanding, (rk + pk q)2 ≡ rk2 + 2pk rk q
(mod pk+1 Zp ). The right-hand side is congruent to a mod pk+1 Zp precisely
when
2pk rk q ≡ a − rk2 (mod pk+1 Zp ).
By construction, a − rk2 ∈ pk Zp , so we can rewrite the last congruence as

a − rk2
2rk q ≡ (mod pZp ).
pk

Since p is odd, 2 is invertible in Zp . As rk2 ≡ a (mod pZp ) and a ∈ / pZp , we

have rk ∈
/ pZp . So rk is invertible in Zp , and the last displayed congruence is
equivalent to
a − rk2
q≡ mod pZp .
2pk rk
a−r 2 a−r 2
So if we put rk+1 = rk + pk 2pk rkk = rk + 2rkk , then rk+1 mod pk+1 Zp is a lift
of rk mod pk Zp to a mod pk+1 Zp solution of x2 = a.
Assume r1 , r2 , r3 , . . . have been determined by the above procedure. Since
rk+1 mod pk+1 Zp is a lift of rk mod pk Zp , we have |rk+1 − rk |p ≤ p−k . Thus,
{rk } is a Cauchy sequence of elements of Zp . Let x = lim rk , which belongs to

113
Zp . By construction, |rk2 − a|p ≤ p−k . Thus, rk2 → a. Since rk2 also tends to x2 ,
we conclude that x2 = a.

Remark. By a small modification of the argument, we can choose our mod pk Zp

solutions rk to all be rational integers. Then their limit x is simply (r1 mod p, r2 mod
p2 , r3 mod p3 , . . . ) ∈ Zp .

a1
9.106 By Problem 9.105, the map ϕ : Z×


p → {±1} that sends a to p is a
homomorphism with kernel (Z× 2
p ) . This homomorphism is onto, as 1 is sent
to 1 and any n ∈ Z with np = −1 is sent to −1. Therefore, the two cosets

of (Z× 2 ×
p ) in Zp are ϕ
−1
(1) = (Z× 2
p ) and ϕ
−1
(−1) = n(Z× 2
p ) , establishing the
first claim.
Every x ∈ Q× v
p has a unique representation in the form p u where v ∈ Z and
u ∈ Zp . Identifying p u with (v, u) sets up an isomorphism Q×
× v ∼ ×
p = Z × Zp ,
× × 2 ∼ × × 2
which after quotienting by squares becomes Qp /(Qp ) = Z/2 × Zp /(Zp ) .
In this last isomorphism, 1, p, n, np are sent to the four distinct elements of
Z/2×Z× × 2 × × 2
p /(Zp ) . Hence, 1, p, n, and np are coset representatives for Qp /(Qp ) .

9.107 Assume a ∈ Z× 2 . Then a is a square in Q2 ⇐⇒ a is a square in Z2 (cf.

the solution to Problem 9.105).
If a = x2 with x ∈ Z2 , and a ∈ Z× ×
2 , then x ∈ Z2 . Hence, x ≡ 1, 3, 5, or 7
2
(mod 8Z2 ). In each of these cases, x ≡ 1 (mod 8Z2 ), and so a ≡ 1 (mod 8Z2 ).
Conversely, suppose that a ≡ 1 (mod 8Z2 ) and write a = 1 + 8b. Any square
root of a in Z2 must have the form 1 + 2y, for some y ∈ Z2 . As (1 + 2y)2 = a
⇐⇒ y 2 + y = 2b, it suffices to prove that y 2 + y = 2b has a solution y ∈ Z2 .
This we do following the method of Problem 9.105.
To get started, the residue class 0 mod 2Z2 is a mod 2Z2 -solution to y 2 +y = 2b.
Suppose we have already found a mod 2k Z2 solution rk mod 2k Z2 to y 2 +y = 2b,
where k ∈ Z+ . We proceed to lift this to a mod 2k+1 Z2 solution. Expanding,
(rk + 2k q)2 + (rk + 2k q) ≡ rk2 + rk + 2k q (mod 2k+1 Z2 ). The right-hand side
is congruent to 2b modulo 2k+1 Z2 precisely when

2k q ≡ 2b − (rk2 + rk ).

By construction, 2b − (rk2 + rk ) ∈ 2k Z2 , and so this last congruence can be

rewritten as
2b − (rk2 + rk )
q≡ (mod 2Z2 ).
2k
2
2b−(rk +rk )
So if we put rk+1 = rk + 2k 2k
= 2b − rk2 , then rk+1 (mod 2k+1 Z2 ) is
a suitable lift.
The rest of the solution is essentially the same as that of Problem 9.105.

114
9.108 By Problem 9.107, the homomorphism from Z× ×
2 to (Z2 /8Z2 ) send-
× 2
ing a to a mod 8Z2 has kernel (Z2 ) . As this homomorphism is easily seen
× 2 ∼
to be surjective, Z×2 /(Z2 ) = (Z2 /8Z2 )
×
= {1 mod 8Z2 , 3 mod 8Z2 , 5 mod
8Z2 , 7 mod 8Z2 }. Hence, 1, 3, 5, and 7 are coset representatives for (Z2 /8Z2 )× .
Noting that each element of Q× 2 has a unique representation in the form 2 u,
v
×
with v ∈ Z and u ∈ Z2 , we conclude that

1, 3, 5, 7, 2 · 1, 2 · 3, 2 · 5, 2 · 7

are coset representatives for Q× × 2

2 /(Q2 ) . (See the solution to Problem 9.106
for further details on this last step.)

Extra Exploration 22. Characterize the elements of Zp that can be written as

x2 − y 2 for x, y ∈ Zp . Then do the same for x2 + y 2 (harder). For the latter problem,
start by showing that every element of Z/p is a sum of two squares.

9.109 Let a ∈ Q× v × n
p and write a = p u with u ∈ Zp . If a = x for some
x ∈ Qp , then n | nvp (x) = vp (a) = v. For this to hold for infinitely many n
requires
√ v = 0, so that a = pv u = u ∈ Z× p . This proves the “⇐=” implication:
If n a ∈ Qp for infinitely many n, then a ∈ Z× p.

Now suppose that a ∈ Z× p . It suffices to show that if ℓ is any prime not dividing
p(p − 1), then a is an ℓth power in Qp .
Since ℓ ∤ p − 1 = #(Zp /pZp )× , the ℓth power map is an automorphism
of (Zp /pZp )× . So we can find an r1 ∈ Z× ℓ
p with r1 ≡ a mod pZp . We take
r1 mod pZp as our starting point in the method of successive approximation.
Assume we know a mod pk Zp -solution to xℓ = a, say rk mod pk Zp . We seek a
mod pk+1 Zp -solution rk+1 mod pk+1 Zp . Expanding, (rk +pk q)ℓ ≡ rkℓ +ℓrkℓ−1 pk q
(mod pk+1 Zp ). For the right-hand expression to agree with a mod pk+1 , we
require that ℓrkℓ−1 pk q ≡ a − rkℓ (mod pk+1 Zp ), or equivalently

a − rkℓ
ℓrkℓ−1 q ≡ (mod pZp ).
pk

Both rk and ℓ belong to Z× ×

p . (We use here that a ∈ Zp and that ℓ is a prime
ℓ
a−rk
distinct from p.) So we can satisfy this congruence with any q ≡ ℓ−1
pk ℓrk
modulo
ℓ ℓ
a−rk a−rk
pZp . In particular, rk+1 = rk + pk pk ℓrℓ−1 = rk + ℓ−1
ℓrk
yields a suitable lift.
k

Assume r1 , r2 , r3 , . . . are chosen according to the above procedure. Each |rk+1 −

rk | ≤ p−k−1 , so that the {rk } form a Cauchy sequence in Zp with a limit
x ∈ Zp . Since rk → x, and taking ℓth powers preserves limits, rkℓ → xℓ . On
the other hand, each |rkℓ − a| ≤ p−k , so that rkℓ → a. Therefore, xℓ = a.

9.110 Let R be any commutative ring. If F (T ) ∈ R[T ], then

115
 F (X) − F (Y ) ≡ F (Y ) − F (Y ) ≡ 0 (mod (X − Y )R[X, Y ]).

So we can write F (X) − F (Y ) = (X − Y )G(X, Y ) for some G(X, Y ) ∈ R[X, Y ].

If G(X, Y ) ∈ Z[X, Y ] is the polynomial corresponding to our F (T ) ∈ Z[T ],
then
|F (n) − F (α)|p = |(n − α)G(n, α)|p ≤ |n − α|p .
(To make the last estimate we use that G has Zp -coefficients and that both
n, α ∈ Zp .) This establishes the first claim of the problem.
Since α is a root of F , it is clear that |F (n) − F (α)|p = |F (n)|p .
As F has Qno integer zeros, F (n) ̸= 0, so that by the product formula, |F (n)|p =
−1
|F (n)|−1
∞
−1
prime ℓ̸=p |F (n)|ℓ ≥ |F (n)|∞ .
Pd
Write F (T ) = k=0 ak T k . Then
d d
!
X X
k
|F (n)|∞ = ak n ≤ |ak |∞ |n|d∞ .
k=0 ∞ k=0

Therefore, |F (n)|−1 −d
|ak |∞ )−1 . (The definition of c makes
P
∞ ≥ c|n|∞ for c := ( k
sense since the ak are not all zero.)

Remark. When d ≥ 2 our inequality |n−α|p ≥ c|n|−d ∞ can be improved substantially.

A theorem of Mahler [3, Theorem (5,I), p. 159] allows us to replace −d with −1 − ϵ,
for any fixed ϵ > 0. In this new statement, the coefficient c in front of |n|−1−ϵ
∞ is now
allowed to depend on both F and ϵ. Mahler’s theorem is a p-adic variant of a deep
result of Klaus Roth, for which Roth was awarded the Fields Medal in 1958.

The exponent −1 − ϵ in Mahler’s theorem is essentially best possible, since approxi-

mations to within |n|−1∞ are thick on the ground. Indeed, each α ∈ Zp has a canonical
expansion of the form c0 + c1 p + c2 p2 + . . . . If we choose n = c0 + c1 p + · · · + ck−1 pk−1 ,
then |n − α|p ≤ p−k < |n|−1 ∞ (if n ̸= 0). Provided that α is not a nonnegative integer,
varying k yields infinitely many distinct positive integers n. By a similar argument, as
long as α is not a nonpositive integer, |n−α|p < |n|−1 ∞ has infinitely many solutions in
negative integers n. Therefore, |n − α|p < |n|−1 ∞ has infinitely many integer solutions
n whenever α is a nonzero element of Zp .

Pd
Extra Exploration 23. Let A(T ) = ak T k ∈ Zp [T ]. It is easy to prove (and
k=0
we essentially did this in our solution to Problem 9.110) that |A(x+h)−A(x)|p ≤ |h|p
whenever x, h ∈ Zp . Show that for x ∈ Zp and h ∈ pZp , this bound can be refined to

|A(x + h) − A(x)|p ≤ K|h|p ,

where K = maxk≥0 |kak |p is the largest absolute value of any coefficient of A′ (T ).

This is a p-adic cousin of the usual mean value theorem (cf. Robert [4]).

116
9.111 That α := k≥1 pk! ∈ Zp follows from Problem 8.92. Suppose for a
P
contradiction that α is a root of the nonconstant polynomial F (T ) ∈ Q[T ].
We can assume F (T ) is irreducible over Q and, by clearing denominators, that
F (T ) ∈ Z[T ].
If F has an integer root n0 , irreducibility over Q forces F to be linear with n0
as its only root. In that case, α = n0 ∈ Z. But this is absurd: α’s canonical
expansion is not eventually periodic, so that α is not even rational, let alone a
rational integer.
We can therefore apply the result of Exercise 9.110. Let Pcn be the constant
k!
associated with F in that problem. For each n, let rn = k=1 p ∈ Z. Then
|rn − α|p = | k>n pk! |p ≤ p−(n+1)! . Also, |rn |∞ ≤ 2pn! ; here we use that the
P

largest term in the sum defining rn is pn! and that each term in that sum is at
least twice the preceding one. Therefore,

p−(n+1)! ≥ |rn − α|p ≥ c|rn |−d

∞ ≥c·2
−d −dn!
p .

Rearranging,
pn!(d−(n+1)) ≥ c · 2−d .
But the right-hand side is a positive quantity independent of n, while the
left-hand side tends to 0 as n tends to infinity. Contradiction!

9.112 Given x ∈ Z10 , define the Z2 and Z5 -reductions of x as the first and
second components of the image of x under the isomorphism Z10 ∼ = Z2 ×
Z5 described in the solution to Problem 5.63. Concretely, if x = (a1 mod
10, a2 mod 102 , . . . ), its Z2 -reduction is (a1 mod 2, a2 mod 22 , . . . ), and its Z5 -
reduction is (a1 mod 5, a1 mod 52 , . . . ).
Convergence in Z10 can then be defined in terms of convergence in Q2 and Q5 .
If {xn } is a sequence of elements of Z10 , and x ∈ Z10 , we say xn → x in Z10 if
the Z2 and Z5 -reductions of the xn converge to the Z2 and Z5 -reductions of x
(in Q2 and Q5 ).
With this definition of convergence in place, your work on Set #7 can be
adapted toPshow that every element of Z10 has a unique, convergent 10-adic
expansion k≥0 dk · 10k with each dk ∈ {0, 1, 2, . . . , 9}. (Check this!)
n n
Clearly, 24·5 → 0 in Z2 . By Exercise 7.81, 24·5 → 1 in Z5 . So the Z10 limit
n
x of 24·5 corresponds, underP our isomorphism, to (0, 1) ∈ Z2 × Z5 . Write the
10-adic expansion of x as k≥0 dk · 10k and assume for a contradiction that
the sequence {dk } is eventually periodic, say dk = dk+ℓ for all k ≥ k0 . Working
in Z2 , we find that the Z2 -reduction of x is

117
 X X
10k + dk · 10k (1 + 10ℓ + 102ℓ + . . . )
0≤k<k0 k0 ≤k<k0 +ℓ
X X dk · 10k
= 10k + .
1 − 10ℓ
0≤k<k0 k0 ≤k<k0 +ℓ

Let r be the rational number defined by the right-hand side. The exact
samePcalculation shows that the Z5 -reduction of x is also r. It follows that
x = k≥0 dk ·10k is the element of Z10 corresponding under our isomorphism to
(r, r) ∈ Z2 ×Z5 . But we saw already that x corresponds to (0, 1). Contradiction!

Extra Exploration 24. Say that the infinite sequence of decimal digits d0 , d1 ,
d2 , d3 , . . . is self-squaring if, for every k ∈ Z+ ,

(d0 + d1 · 10 + · · · + dk−1 · 10k−1 )2 ≡ d0 + d1 · 10 + · · · + dk−1 · 10k−1 (mod 10k ).

For example, the two sequences 0, 0, 0, 0, . . . (all zeros) and 1, 0, 0, 0, . . . (all zeros
past the first term) are trivially self-squaring.
(a) Prove that there is a unique self-squaring sequence starting with d0 = 6.
(b) Show that the sequence in (a) begins 6, 7, 3, 9, 0, 1, 7, 8, 7, 1. (As a spot check,
762 = 5776, while 1093762 = 11963109376.)
(c) How many self-squaring sequences are there?

Extra Exploration 25 (Shapiro and Shapiro [6]). Let a1 , a2 , a3 , . . . be an

arbitrary sequence of positive integers. Show that for every g ∈ Z+ , the sequence
a
a 3
a1 , aa1 2 , a1 2 , . . . stabilizes modulo g. Deduce that for each g > 1, this same sequence
converges in Zg . (Define convergence in Zg by mimicking what we did for Z10 above.)

Extra Exploration 26 (continuation). Let a be a positive integer not divisible

by 10. By Extra Exploration 25, for each k ∈ Z+ the residue class mod 10k of
a
a, aa , aa , . . . eventually stabilizes. Let xk be the least nonnegative integer in the
limiting residue class mod 10k , so that 0 ≤ xk < 10k .
(a) Show that for each k ≥ 2, we have axk ≡ xk (mod 10k ).
(b) Let a = 73. Compute x33 , by hook or by crook, and thereby show that

73990485815519399724778909194186633 = . . . 990485815519399724778909194186633.
For more on this theme, see the papers of Jiménez Urroz & Yebra [2] and Germain [1].

9.113 The first equivalence is immediate from Problem 8.92.

Turning
P to the second: If x ∈ Zp , then |an xn |p ≤ |an |p for all n. So if |an |p → 0,
n
then n≥0 an x converges for all x ∈ Zp .
Conversely, suppose that n≥0 an xn converges for all x ∈ Zp . Then n≥0 an
P P
converges (the case x = 1), so that |an |p → 0.

118
Extra Exploration 27. Check that the Strassmann series form a subring of
Qp [[T ]]. This ring is commonly denoted Qp ⟨T ⟩ and referred to as the Tate algebra in
one variable over Qp . Show that for each z ∈ Zp , the evaluation map evalz : Qp ⟨T ⟩ →
Qp sending F (T ) to F (z) is a ring homomorphism.

9.114 Since F (T ) is a Strassmann series, its coefficients an tend to 0. In

particular, {|an |p } is bounded, and pk F (T ) ∈ Zp [[T ]] for a certain integer
k. Since F and pk F have the same zeros, we can (and will) assume that
F (T ) ∈ Zp [[T ]] to start with.
Let m be the smallest nonnegative integer for which am = ̸ 0 and let δ = |am |p .
Note that δ ≤ 1, since F (T ) ∈ Zp [[T ]]. If 0 < |x|p < δ, then

X m
X
an xn ≤ max |an xn |p ≤ |x|m+1
p < δ|x|m m
p = |am x |p = an xn .
n>m
n>m p n=0 p

Therefore,
m
X X m
X X
|F (x)|p = an xn + a n xn ≥ an xn − an xn > 0,
n=0 n>m n=0 p n>m p
p

so that F (x) ̸= 0.
As a consequence, if F (T ) is Strassmann and not the zero series in Qp [[T ]],
then F (pm ) is nonzero for all sufficiently large m (specifically, for any m with
p−m < δ). Hence, a Strassmann series that vanishes everywhere on Zp must
be the zero series.

9.115 Substituting and applying the binomial theorem, we find that whenever
x, x0 ∈ Zp ,
 
X XX k j k−j
F (x + x0 ) = ak (x + x0 )k = 1k≥j ak x x0 .
j
k≥0 k≥0 j≥0

We would like to reverse the order of summation. To justify

this we appeal
to the criterion of Exercise 8.96. Put uk,j = 1k≥j ak kj xj xk−j
0 . If we set
ϵN = maxk≥N |ak |p , then |uk,j |p ≤ ϵN whenever k ≥ N or j ≥ N . Since F (T )
is a Strassmann series, ϵN → 0 as N → ∞, and so the conditions of Exercise
8.96 are satisfied. Therefore,
 
XX XX X X k  k−j
F (x + x0 ) = uk,j = uk,j =  ak x0  xj .
j j
j
k k j≥0 k≥j

Here the coefficient of xj is precisely the claimed coefficient bj .

119
 j j
P P
Since j≥0 bj x converges for all x ∈ Zp (to F (x + x0 )), j≥0 bj T is a
Strassmann series.

9.116 Let F (T ) be a Strassmann series, F (T ) ̸= 0. Suppose for a contradiction

that F has an infinite number of zeros in Zp , and let x1 , x2 , x3 , . . . be a sequence
of distinct zeros. As Zp is compact (Problem 6.76), there is an x0 ∈ Zp such
that some subsequence of {xn } converges to x0 .
Problem 9.115 describes how to construct a Strassmann series G(T ) with
G(x) = F (x + x0 ) for every x ∈ Zp . By the choice of x0 , every open disc
centered at x0 contains infinitely many zeros of F . Hence, every open disc
centered at 0 contains infinitely many zeros of G. This contradicts Exercise
9.114 unless G(T ) = 0 in Qp [[T ]]. But that’s impossible: If G(T ) = 0, then
F (x) = G(x − x0 ) = 0 for all x ∈ Zp . But a nonzero Strassmann series such as
F (T ) cannot vanish on all of Zp (Exercise 9.115).

Extra Exploration 28. We assume familiarity with infinite products, as defined

in Extra Exploration 20. Let c1 , c2 , c3 , . . . be a sequence
Q of elements of Qp tending
to 0. By Extra Exploration 20, the infinite product ∞ j=1 (1 − cj x) determines a
well-defined element of Qp for all x ∈ Q p . Write down a power series F (T ) ∈ Qp [[T ]]
with the property that F (x) = ∞
Q
j=1 (1 − cj x) for all x ∈ Qp .

Extra Exploration 29 (continuation; Schöbe [5, p. 38]).

Q∞ Let e1 , je2 , p−1
e3 , . . .
ej
be a sequence of positive integers. For each x ∈ Qp , let A(x) = j=1 (1 − (p x) ) .
(a) Prove that A(x) is a well-defined function from Qp to Qp and that A(x) can be
represented by an everywhere convergent power series with Qp -coefficients.
(b) Show that |A(x)|p = 1 if x ∈ Zp .
r
Q |x|(r−j)(p−1)e
(c) Suppose p = p , where r is a positive integer. Show that |A(x)|p ≤
p−er r−1
j=1 p j
.
(d) Describe a method of choosing e1 , e2 , e3 , . . . that guarantees |A(x)|p ≤ |x|−1
p
whenever x ∈ Qp \ Zp .
(e) Deduce from (a)–(d) that there is an power series over Qp that converges
everywhere and induces a bounded but nonconstant function from Qp to Qp .
For those who have taken a complex variables course: Explain how this suggests
Qp is more analogous to R than to C.

9.117 When p = 2, neither statement holds. So assume p is odd.

Pp−1
By Problem 7.90, (p−1)!+1
p ≡ a=1 qp (a) (mod p). So it suffices to show that
Pp−1 1
a=1 qp (a) − (Bp−1 + p − 1) ∈ pZ(p) .

By Faulhaber’s formula,
p−1 p−1 
pj

X Sp−1 (p) p − 1 1 X p−1
qp (a) = − = Bp−1 + − 1 + Bp−1−j .
a=1
p p p j=1
j j+1

120
For each j = 1, 2, 3, . . . , p−1, the Bernoulli number Bp−1−j ∈ Z(p) ; this is clear
for j = p − 1 (recall that B0 = 1) and for j < p − 1 it follows from Exercise
8.99. In this same range of j, we have pj ≥ 3j > j + 1. Thus, vp (j + 1) < j,
pj
and vp ( j+1 ) = j − vp (j + 1) > 0. Consequently, each term in our sum on j
belongs to pZ(p) , implying that the sum itself belongs to pZ(p) . But that sum
Pp−1
is precisely a=1 qp (a) − (Bp−1 + p1 − 1).

9.118 By Problem 8.104 and its solution, p ≡ 1 (mod 6) and ω(1 + u)
=

1 + ω(u). Write v = ω(u) + pk q, where q ∈ Zp . Since p is odd and p | kp for
all k between 0 and p,

v p = (ω(u) + pk q)p ≡ ω(u)p + ω(u)p−1 pk+1 q (mod p2k+1 pZp )

≡ ω(u) + pk+1 q (mod p2k+1 pZp ).

On the other hand, 1 + v = 1 + ω(u) + pk q = ω(u + 1) + pk q, and (by analogous

reasoning to what is displayed above)

(1 + v)p ≡ ω(u + 1) + pk+1 q (mod p2k+1 pZp ).

Using once more that 1 + ω(u) = ω(u + 1), we deduce that (1 + v)p ≡ 1 + v p
(mod p2k+1 Zp ). Since both (1 + v)p and 1 + v p lie in Z, this last congruence
in fact holds modulo (p2k+1 Zp ) ∩ Z = p2k+1 Z.
Let p = 7. In the course of solving Problem 8.103, we computed that ω(2) ≡ 324
(mod 73 Z7 ). Taking v = 324 and k = 3 in the result of the last paragraph
“explains” the given example.

References

1. J. Germain, On the equation ax ≡ x (mod b). Integers 9 (2009), A47, 629–638.

2. J. Jiménez Urroz and J. L. A. Yebra, On the equation ax ≡ x (mod bn ). J. Integer
Seq. 12 (2009), Article 09.8.8, 8 pp.
3. K. Mahler, Lectures on Diophantine approximations. Part I: g-adic numbers and
Roth’s theorem, Notre Dame Mathematical Lectures, vol. 7, University of Notre
Dame Press, Notre Dame, 1961.
4. A. Robert, A note on the numerators of the Bernoulli numbers. Exposition.
Math. 9 (1991), 189–191.
5. W. Schöbe, Beiträge zur Funktionentheorie in nichtarchimedisch bewerteten
Körpern. Münster: Diss. Math. Univ. Münster, Universitas-Archiv 42, Math.
Abteilung Nr. 2 (1930), 61 pp.
6. D. B. Shapiro and S. D. Shapiro, Iterated exponents in number theory. Integers 7
(2007), A23, 19 pp.

121
122
Solutions to Set #10

10.119 We begin by establishing some useful arithmetic properties of the

1

generalized binomial coefficients k2 for nonnegative integers k.
1
Fix k. Let ℓ be an odd prime, and let A be a nonnegative integer with 2 ≡A
(mod ℓvℓ (k!) Zℓ ); e.g., A = 21 (ℓvℓ (k!) + 1). Since A


k ∈ Z,
 
vℓ (k!) A
ℓ | k! | k! = A(A − 1)(A − 2) · · · (A − (k − 1)),
k

where the divisibility relations are being asserted in the ring Z. Naturally,
these same relations also hold in Zℓ , so that working in Zℓ modulo ℓvℓ (k!) Zℓ ,
   
1 1 1
− 1 ··· − (k − 1) ≡ A(A − 1) · · · (A − (k − 1)) ≡ 0.
2 2 2
1

It follows that vℓ ( 12 ( 12 − 1) · · · ( 12 − (k − 1))) ≥ vℓ (k!), so that vℓ ( k2 ) ≥ 0. As
1

this holds for all odd primes ℓ, the rational number k2 has denominator a
power of 2.
To determine which power of 2, we look 2-adically. Expanding out,
1
2 1 · (1 − 2) · (1 − 2 · 2) · · · (1 − 2(k − 1))
= .
k 2k k!

The numerator on the right is odd, and so the power of 2 in the denominator
1

of k2 is 2k+v2 (k!) .
1

Collecting what we know so far: k2 is a rational number with denominator
2k+v2 (k!) .
We are now well-positioned to decide when B 12 (x) converges — equivalently,
1

when | k2 xk |p → 0 (as k → ∞). For use momentarily, we recall that v2 (k!) ≤ k,
so that k + v2 (k!) ≤ 2k.

123
 1
1

Suppose p is odd. Then | k2 |p ≤ 1, and | k2 xk |p ≤ |x|kp . If |x|p ≤ 1/p, then
|x|kp → 0 as k → ∞. Thus, B 12 (x) converges for these x.
1

Suppose instead that p = 2. If x ∈ Q2 , then | k2 xk |2 = 2k+v2 (k!) |x|k2 ≤ 22k |x|k2 .
This last expression tends to 0 if |x|2 ≤ 1/23 .
These conditions on x turn out to be necessary for convergence.

Lemma. Let F (T ) ∈ Qp [[T ]]. If F (x0 ) converges (x0 ∈ Qp ), then F (x)

converges for all x ∈ Qp with |x|p ≤ |x0 |p .

ak T k . If |ak xk0 |p → 0, and |x|p ≤ |x0 |p , then

P
Proof. Write F (T ) = k≥0
|ak xk |p → 0. ❚

Seeking a contradiction, suppose p is odd and that B 12 (x0 ) converges for a

value of x0 with |x0 |p > 1/p. Then |x0 |p ≥ 1. Invoking the lemma, B 12 (x)
converges whenever |x|p ≤ 1, i.e., on all of Zp . But whenever B 12 (x) converges,
it converges to a square root of 1 + x. We infer that that every element of Zp
has a square root in Qp — absurd!
Similarly, if B 12 (x) converges for an x0 ∈ Q2 with |x0 |2 > 1/23 , then it
converges whenever |x|2 ≤ 1/22 . But then B 12 (4) converges to a square root of
5 in Q2 , whereas 5 ∈ / (Q× 2
2) !

Extra Exploration 30. Establish the following claims.

(a) If x ∈ Zp and k is a nonnegative integer, then xk := x(x−1)···(x−(k−1))


k!
∈ Zp .
(b) If x ∈ Q and k is a positive integer, then every prime appearing in the denomi-
nator of xk appears in the denominator of x, and vice versa.

Extra Exploration 31. (continuation of Extra Exploration 30) Let m ∈ Z+ .

1

P k
(a) Show that if p is a prime not dividing m and x ∈ pZp , then k≥0 k x
m
√
m
converges to (one value of) 1 + x in Qp .
(b) Now assume p | m. Show that the conclusion of (a) holds if either p is odd and
vp (x) ≥ vp (m) + 1, or p = 2 and v2 (x) ≥ v2 (m) + 2.

10.120 When x is real and |x| < 1, the series B 12 (x) converges to the positive
9
square root of 1 + x. So over the real numbers, B 12 ( 16 ) = 54 .
1

Suppose now that p is an odd prime and that x ∈ pZp . Recalling that k2 ∈ Zp
1

for each k = 0, 1, 2, . . . , we find that |B 12 (x) − 1|p ≤ maxk≥1 | k2 xk |p ≤ 1/p.
Thus, B 21 (x) ∈ 1 + pZp .
9
When p = 3 and x = 16 , the (unique) square root of 1 + x belonging to 1 + 3Z3
is − 54 .

124
10.121 These results are not particular to Zp and Qp . Let D be any domain of
characteristic 0 with fraction field K. (For example, we could take D = Zp and
K = Qp .) Then Taylor’s formula holds for polynomials over K: If F (T ) ∈ K[T ]
and a ∈ K, then
X F (j) (a)
F (a + T ) = T j.
j!
j≥0

For the proof, fix a ∈ K. Consider K[T ] as a K-vector space and observe that
both sides of the claimed equation represent K-linear functions of F (T ) ∈ K[T ].
So the identity will be established if it is shown for F (T ) = 1, T, T 2 , . . . (a
K-basis for K[T ]). When F (T ) = T n , the left-hand side is (a + T )n while the
right is
X n(n − 1)(n − 2) . . . (n − (j − 1)) n−j j X n
1n≥j a T = an−j T j .
j! j
j≥0 0≤j≤n

This last expression is of course the binomial expansion of (a + T )n .

Next, we show that if F (T ) ∈ D[T ], then j!1 F (j) (T ) ∈ D[T ] for each non-
negative integer j. We fix j and check the claim for F (T ) = 1, T, T 2 , . . . (a
D-basis for D[T ]). This is straightforward: If F (T ) = T n , then j!1 F (j) (T ) =
1n≥j nj T n−j ∈ Z[T ] ⊆ D[T ].

10.122 Taylor’s formula gives F (x̃) = j≥0 j!1 F (j) (x)(−F (x)/F ′ (x))j . The
P
terms of the sum corresponding to j = 0 and j = 1 cancel each other out
(being F (x) and −F (x), respectively), so that

X F (j) (x)  j j
F (j) (x)

F (x) F (x)
|F (x̃)|p = − ′ ≤ max − ′ .
j! F (x) j≥2 j! F (x)
j≥2 p
p

Viewing F (j) (x)/j! as the evaluation of the polynomial F (j) (T )/j! ∈ Zp [T ] at

the point x ∈ Zp , it is clear that |F (j) (x)/j!|p ≤ 1 for every j. Furthermore,
|F ′ (x)|p = 1 while |F (x)|p ≤ 1, so that |(−F (x)/F ′ (x))j |p = |F (x)|jp ≤ |F (x)|2p
for each j ≥ 2. Hence, the above maximum does not exceed |F (x)|2p .

10.123 We iteratively apply Exercise 10.122 in order to construct a Cauchy

sequence of elements of Zp converging to a root of F .
Let n ∈ Z+ . Suppose that xn ∈ Zp satisfies both

|F (xn )|p < 1 and |F ′ (xn )|p = 1.

F (xn )
Let xn+1 = xn − F ′ (xn ) . By Exercise 10.122,

|F (xn+1 )|p ≤ |F (xn )|2p < 1.

125
Furthermore, xn+1 ≡ xn (mod pZp ), implying F ′ (xn+1 ) ≡ F ′ (xn ) ̸≡ 0
(mod pZp ), so that
|F ′ (xn+1 )|p = 1.
Thus, the hypotheses we originally assumed for xn hold for xn+1 , allowing us
to reboot the procedure with xn+1 replacing xn .
Starting from the given x1 , we produce in this way a sequence {xn } of elements
of Zp satisfying
n−1
|F (xn )|p ≤ |F (xn−1 )|2p ≤ · · · ≤ |F (x1 )|2p (*)

for each n = 1, 2, 3, . . . . As |F (x1 )|p < 1, (*) guarantees that F (xn ) converges
(rapidly!) to 0.
Observe that |xn+1 − xn |p = |F (xn )/F ′ (xn )|p = |F (xn )|p , for every n. There-
fore, {xn } is Cauchy, and xn → x for some x ∈ Zp . Hence, F (x) = F (lim xn ) =
lim F (xn ) = 0. (We use here that polynomials preserve limits, which follows
from Exercise 3.29.)
argue that x ≡ x1 (mod pZp ). This is immediate from the
It remains only toP
identity x − x1 = k≥1 (xk+1 − xk ) expressing x − x1 as a sum of terms from
pZp .

Extra Exploration 32 (a stronger version of Hensel’s Lemma). Let

F (T ) ∈ Zp [T ], and suppose that a ∈ Zp satisfies |F (a)|p < |F ′ (a)|2p . Prove that
starting from a, Newton’s method converges to a root x of F with |x − a|p =
| FF′(a) | < |F ′ (a)|p .
(a) p

10.124 Express the Q[T ]-ideal F (T )Q[T ] + F ′ (T )Q[T ] as D(T )Q[T ], where
D(T ) ∈ Q[T ]. Since F (T ), F ′ (T ) ∈ D(T )Q[T ], each complex root of D(T ) is a
common zero of F (T ) and F ′ (T ) — in other words, a multiple root of F (T ).
We are given that F (T ) has distinct complex roots. Therefore, D(T ) has
no complex roots, meaning that D(T ) is a nonzero constant. As a result,
F (T )Q[T ] + F ′ (T )Q[T ] = D(T )Q[T ] = Q[T ], and there are X(T ), Y (T ) ∈
Q[T ] with F (T )X(T ) + F ′ (T )Y (T ) = 1. Choose R ∈ Z+ so that X̂(T ) :=
RX(T ), Ŷ (T ) := RY (T ) ∈ Z[T ]. Then

F (T )X̂(T ) + F ′ (T )Ŷ (T ) = R.

Taking this last equation mod p, we find that the mod p reductions of F (T )
and F ′ (T ) generate the unit ideal in (Z/p)[T ] whenever p ∤ R. Hence, F (T )
and F ′ (T ) are coprime in (Z/p)[T ] for each prime p not dividing R.

10.125 According to Problem 2.24, there are infinitely many primes p for
which F has a root in Fp . By Problem 10.124, there are only finitely many

126
primes p for which F and F ′ have a common root in Fp . So for infinitely many
p, we can find an x1 ∈ Z with F (x1 ) ≡ 0 (mod p) and F ′ (x1 ) ̸≡ 0 (mod p).
For each of these primes, F has a root in Zp by Problem 10.123.

Extra Exploration 33 (bounding the number of modular roots of a

polynomial). Suppose F (T ) ∈ Z[T ] is a nonconstant polynomial with distinct
complex roots, and let R be a nonzero integer belonging to F (T )Z[T ] + F ′ (T )Z[T ]
(as in Problem 10.124).
(a) Show that if K ≥ 2vp (R) + 1, and a is an integer with F (a) ≡ 0 (mod pK ), then
there is a root x ∈ Zp of F with x ≡ a (mod pK−vp (R) Zp ).
(b) Continue to assume K ≥ 2vp (R) + 1. Deduce from F having at most deg F roots
in (the domain) Zp that F has at most (deg F )pvp (R) roots in Z/pK .
(c) Finally, show that for every m ∈ Z+ , the number of roots of F in Z/m is at
most (deg F )ν(m) R2 , where ν(m) is the count of distinct primes dividing m.
This bound, which finds many applications in analytic number theory, was proved
independently by Nagell [2] and Ore [3] in 1921.

10.126 If F has a root θ′ ∈ Qp , let K ′ = Q(θ′ ) be the field generated by θ′

over the copy of Q within Qp . By standard field theory, both K and K ′ are
isomorphic to Q[T ]/(F (T )), via isomorphisms identifying θ and θ′ with the
∼
class of T mod F (T ). Daisy chain the isomorphism K − → Q[T ]/(F (T )) with
∼
the isomorphism Q[T ]/(F (T )) − → K ′ to determine an embedding of K into
Qp .
Such a θ′ exists for infinitely many p by Exercise 10.125.

10.127 Let ϕ : Qp → Qp be a homomorphism. As every ring homomorphism

sends n · 1 to n · 1 (for all n ∈ Z), it is immediate that ϕ fixes Z. Furthermore,
for any a, b ∈ Z with b ̸= 0, we have
a a
ϕ b=ϕ ϕ(b) = ϕ(a) = a.
b b
Thus, ϕ ab = ab , meaning that ϕ in fact fixes all of Q.

Now let x be an arbitrary element of Qp . Since Q is dense in Qp , there is

a sequence of rational numbers {xn } for which xn → x. We will show that
ϕ(xn ) → ϕ(x). Since each ϕ(xn ) = xn , and xn → x, it must be that ϕ(x) = x.
As x was arbitrary, ϕ is the identity map.
The algebraic characterization of Z×p from Exercise 9.109 implies that ϕ maps
Z× ×
p into Zp . To use this, suppose xn − x ̸= 0 (where n is a positive integer
index). Then xn − x = pvp (xn −x) un for some un ∈ Z×p , and

ϕ(xn ) − ϕ(x) = ϕ(xn − x) = ϕ(pvp (xn −x) )ϕ(un ) = pvp (xn −x) u′n

for some u′n ∈ Z×

p . Hence,

127
 |ϕ(xn ) − ϕ(x)|p = p−vp (xn −x) = |xn − x|p .

Of course, the equation |ϕ(xn ) − ϕ(x)|p = |xn − x|p also holds when xn − x = 0.
We are assuming xn → x. Therefore, |ϕ(xn ) − ϕ(x)|p = |xn − x|p → 0. Hence,
ϕ(xn ) → ϕ(x), as desired.

10.128 Suppose first that p and q are odd primes with q ̸= p. Select n ∈ Z with
n


q = −1, and choose a ∈ Z with a ≡ 1 (mod p) and a ≡ n (mod q). Then a
is a square in Qp but not a square in Qq (see Problem 9.105). Since a ∈ Z, any
homomorphism from Qp to Qq must send a to a. However, homomorphisms
always send squares to squares, so no homomorphism from Qp to Qq can exist.
If p = 2 and q is odd, the same argument works if we select a ≡ 1 (mod 8)
and a ≡ n (mod q) (see Problem 9.107). If p is odd and q = 2, choose a ≡ 1
(mod p) and a ≡ 3 (mod 8).
The same argument works to demonstrate the impossibility of a homomorphism
from Qp to R. If p is odd, select a with a ≡ 1 (mod p) and a < 0. If p = 2,
pick a ≡ 1 (mod 8) with a < 0.

10.129 Suppose for a contradiction that α := 0.2481632 · · · ∈ Q. Then the

decimal expansion of α is eventually periodic with period ℓ, say.
n
Let β ∈ Z10 be the 10-adic limit of 24·5 (see Problem 9.112). Write the 10-adic
expansion of β as (. . . d5 d4 d3 d2 d1 d0 )10 , representing d0 + d1 · 10 + d2 · 102 + . . . .
n
For each fixed N and all large n, the decimal expansion of 24·5 terminates
with the string dN dN −1 . . . d1 d0 . Therefore, this digit string appears infinitely
often in the expansion of α. Fixing any nonnegative integer n0 , and choosing
N ≥ n0 + ℓ, we deduce from the periodicity of α’s expansion that dn0 = dn0 +ℓ .
Since this holds for each n0 , the 10-adic expansion of β is (purely) periodic
with period ℓ, contradicting the result of Problem 9.112.

Extra Exploration 34 (Mahler [1]). Let g be an integer at least 2. Show that

the real number 0.(g)(g 2 )(g 3 ) . . . obtained by concatenating the decimal expansions
of g, g 2 , g 3 , . . . is irrational.

References

1. K. Mahler, On some irrational decimal fractions. J. Number Theory 13 (1981),

268–269.
2. T. Nagell, Généralisation d’un théorème de Tchebycheff. J. Math. Pures Appl. 8
(1921), 343–356.
3. O. Ore, Anzahl der Wurzeln höherer Kongruenzen. Norsk Mat. Tidsskr. 3 (1921),
63–66.

128
Solutions to Set #11

11.130 We assume |x − 1|p < 1 and argue that the terms of the series defining
logp x tend to
Q 0. For each positive integer k, the product formula yields
|k|−1
p = |k|∞ prime q̸=p |k|q ≤ |k|∞ = k. Therefore,

(x − 1)k
(−1)k−1 = |x − 1|kp · |k|−1 k
p ≤ k · |x − 1|p ,
k p

which tends to 0 as k tends to infinity.

11.131 We start by noting that if x ∈ 1 + pZp , then x−1 ∈ 1 + pZp . This is not
difficult to prove directly, futzing about with absolute values, but a more elegant
approach is to observe that (1 + rp)−1 = 1 − rp + r2 p2 − · · · ∈ 1 + pZp whenever
r ∈ Zp . Utilizing the addition rule (†), we deduce that for all x ∈ 1 + pZp ,
∞
X (−1)k−1
1
logp x + logp = logp 1 = (1 − 1)k = 0.
x k
k=1

1
(That logp x + logp x = 0 should not be surprising, but it ought to be comfort-
ing!)
If j ∈ {1, 2, . . . , p − 1}, then
∞ −1 !
1 pk
  
X p p j
= − logp 1 − = logp 1− = logp .
jk k j j j−p
k=1

Therefore,
 
p−1 p−1 X
∞ ∞
! p−1
k
X X 1 p =
X 1 pk X j

k k
= logp
j=1
j k j=1
j k j=1
j−p
k=1 k=1
Y j
= logp = logp ((−1)p−1 ).
0<j<p
j − p

129
Here the interchange of the sums on j and k is routine to justify, for example
by setting ak,j = 10<j<p j −k pk k −1 and applying the result of Problem 8.96.
If p is odd, then (−1)p−1 = 1, and logp 1 = 0 gives the desired result. To
finish off, we must show log2 (−1) = 0. But this is easy: log2 (−1) + log2 (−1) =
log2 ((−1)2 ) = log2 (1) = 0.

Remarks.
(i) You may have conjectured in Problem 4.45 thatPv2 ( n k
P
k=1 2 /k) is bounded below
n k
by n − (something small). Once we know that k=1 2 /k = 0, this becomes easy
to show. Indeed,
n ∞
X 2k X 2k 2k k n+1
= − ≤ max ≤ max = n+1 .
k k k≥n+1 k 2
k≥n+1 2k 2
k=1 2 k=n+1 2

(To estimate |2k /k|2 we used the displayed inequality in the solution to Problem
11.130, for x = −1.) Converting this from a statement about absolute values to
log(n+1)
one about valuations, v2 ( n k
P
k=1 2 /k) ≥ (n + 1) − log 2
. A slight modification
of this argument will show that equality holds when n + 1 is a power of 2.
(ii) In the next Extra Exploration, we outline one proof of the addition law (†).
Arguments for (†) based on different principles can be found in the books of
Gouvêa [2, §5.7] and Robert [10, Chapter 5, §4].
Our proof of (†) is motivated by the following formula from (real) calculus: For
each real number x > 0,
√
log x = lim k( k x − 1).
k→∞

d t
(Check this yourself, possibly starting from dt x t=0 = log x.) This identity is
surprisingly powerful; Landau shows in [5] that all of the familiar theory of the
natural logarithm can be developed taking the right-hand side as the definition
of log x. We will establish and apply a p-adic analogue of this relation.

Extra Exploration 35 (Leopoldt [6]). Fix a prime number p. For x ∈ 1+pZp

and m ∈ Z+ , put
m
xp −1
Lp,m (x) = .
pm
Justify each of the following assertions.
(a) For x ∈ 1 + pZp and m ∈ Z+ : |Lp,m+1 (x) − Lp,m (x)|p ≤ p−m−2 .
(b) For each x ∈ 1 + pZp , the limit Lp (x) := limm→∞ Lp,m (x) exists in Qp .
(c) For all x, y ∈ 1 + pZp , and all m ∈ Z+ : |Lp,m (xy) − (Lp,m (x) + Lp,m (y))|p ≤
p−m−2 . Hence, Lp (xy) = Lp (x) + Lp (y).

The rest of this problem is devoted to showing that Lp and logp coincide; once this
is known, (†) follows from (c).

1 pm
(d) For x ∈ 1 + pZp and m ∈ Z+ : (x − 1)k .
P

Lp,m (x) = k≥1 pm k

130
 1 p m (−1)k−1 pm
(e) For m, k ∈ Z+ :

Q

pm k
= k 1≤j<k 1− j
. Therefore, Lp,m (x) =
logp (x) + ep,m (x), where
 
X (−1)k−1 Y  p m

ep,m (x) =  1− − 1 (x − 1)k .
k j
k≥2 1≤j<k

(f) For each x ∈ 1 + pZp , we have limm→∞ ep,m (x) = 0. Thus, Lp (x) = logp x.
− pm /j) is congruent to 1 modulo a large
Q
Hint. If k is small in terms of m, then 1≤j<k (1
power of pZp . On the other hand, when k is large, vp ((x − 1)k /k) is also large.

11.132 Quite a lot of what is claimed here is true — just none of the important
bits!
T 2k+1
Let p be an odd prime and define sinp (T ) = k≥0 (−1)k (2k+1)!
P
∈ Qp [[T ]]. It is
straightforward to check that sinp x converges when |x|p ≤ 1/p; consequently,
sinp (pa) is a well-defined element of Qp for all a ∈ Z. It is also true that if
a is an integer not divisible by p, then sinp (ap) ̸= 0. Otherwise the display
following “Therefore” indeed lays out a contradiction.
Unfortunately, none of this has much to do with the real number π! Let’s
assume π = ab , P
with a, b ∈ Z, b ̸= 0. Choose an odd prime p not dividing a,
Then the series k≥0 (−1)k (ap)k /(2k + 1)! converges to sin(pbπ) = 0 in R. So
far, so good. But the fact that a series converges to 0 in R and converges to
something in Qp — something that could (misleadingly?) be labeled sinp (pbπ)
— doesn’t imply it converges to 0 in Qp . (Cf. Exercise 10.120.)

Remark. An error of a similar nature crept into the work of Hensel himself. In 1905,
Hensel claimed to prove that [Qp (e) : Q] = p for each odd prime p, from which he
derived the corollary that e is transcendental over Q [3]. While the transcendence of
e had already been shown by Hermite in 1873, Hensel’s reasoning was much shorter
and simpler; his proof has an unmistakable air of elegance. Unfortunately, it is also
fundamentally flawed.∗ See [12] and [9, §5.6] for discussion (and compare with [4,
Exercise 9, p. 84]).

11.133 We apply again the method of successive approximation. If x1 = −1,

then F (x1 ) ≡ 0 (mod p). Suppose we have found xk ∈ Zp with F (xk ) ≡ 0
(mod pk ); we demonstrate how to find xk+1 ∈ Zp with xk+1 ≡ xk mod pk Zp
and F (xk+1 ) ≡ 0 (mod pk+1 Zp ).
To enforce the congruence xk+1 ≡ xk (mod pk Zp ), we look for xk+1 of the
form xk+1 = xk + pk h, with h ∈ Zp . For each h ∈ Zp ,
X j j j
F (xk + pk h) − F (xk ) = pk h + p2 ((xk + pk h)2 − x2k ).
j≥1

∗
“For every complex problem there is an answer that is clear, simple, and wrong.” —
H. L. Mencken

131
 j j j
Since pk | ((xk + pk h)2 − x2k ) and p | p2 , the sum on j belongs to pk+1 Zp .
Hence, F (xk + pk h) ≡ F (xk ) + pk h (mod pk+1 ), and F (xk + pk h) ≡ 0
(mod pk+1 ) provided we choose h ≡ −F (xk )/pk (mod pZp).
If x1 , x2 , x3 , . . . are constructed as above, then the {xn } form a Cauchy se-
quence in Zp , so that xn → x for some x ∈ Zp . For each n,

|F (x)|p ≤ |F (x) − F (xn )|p + |F (xn )|p .

By construction, |F (xn )|p → 0. Moreover, F (x) − F (xn ) = (x − xn )(1 +

2j 2j −1 j
+ · · · + x2n −1 )). Thus, |F (x) − F (xn )|p ≤ |x − xn |p , a quantity
P
j≥1 p (x
that also tends to 0. So taking n to infinity, we deduce that F (x) = 0.
We could prove uniqueness similarly, showing that if x ∈ Zp is any zero of
F , then x is uniquely determined mod pZp , then mod p2 Zp , mod p3 Zp , etc.
We choose a different path. We will show that F assumes every value at most
once; that is, F is injective as a function from Zp to Zp . Suppose x, x′ ∈ Zp
with F (x′ ) = F (x). Rearranging,
X j j j
x′ − x = −(x′ − x) p2 (x′2 −1 + · · · + x2 −1 ).
j≥1

Since the right-hand sum is divisible by p, the only way the p-adic absolute
values of both sides can agree is if x′ − x = 0, so that x′ = x.

11.133 (yes, déjà vu all over again!) We will factor F (T ) so as to make obvious
that F (x) = 0 for a unique x ∈ Zp . The factors are constructed by a variant
of the method of successive approximation.
The only property of F (T ) we will use is that F (T ) = 1 + T + pG(T ), where
G(T ) ∈ Zp [[T ]] is a Strassmann series. Since G(T ) is Strassmann, for any
k ∈ Z+ the power series F (T ) ∈ Zp [[T ]] is congruent modulo pk Zp [[T ]] to a
polynomial with Zp coefficients. This will be crucial.
We begin by locating r1 ∈ Zp and q1 (T ) ∈ Zp [T ] with

(1 + T + p · r1 )(1 + p · q1 (T )) ≡ F (T ) (mod p2 Zp [[T ]]).

The left-hand side is congruent to (1 + T ) + p((1 + T )q1 (T ) + r1 ) modulo

p2 Zp [[T ]], and this matches the right mod p2 Zp [[T ]] if

F (T ) − (1 + T )
(1 + T )q1 (T ) + r1 ≡ (mod pZp [[T ]]). (*)
p

Choose F1 (T ) ∈ Zp [T ] with F (T )−(1+T p

)
≡ F1 (T ) (mod pZp [[T ]]). We can
find q1 (T ) and r1 (T ) satisfying (*) by performing long division in the ring
(Zp /pZp )[T ]. Specifically, if F̃1 (T ) is the mod pZp -reduction of F1 (T ), long
division furnishes us with q̃1 (T ) ∈ (Zp /pZp )[T ] and r̃1 ∈ Zp /pZp with F̃1 (T ) =

132
 (1 + T )q̃1 (T ) + r̃1 in (Zp /pZp )[T ]. Then (*) holds if we choose q1 (T ) ∈ Zp [T ]
and r1 ∈ Zp to lift q̃1 (T ) and r̃1 .
Continuing, suppose we have already found r1 , r2 , . . . , rk ∈ Zp and q1 (T ),
q2 (T ), . . . , qk (T ) ∈ Zp [T ] with
 k
X  k
X 
1+T + pj rj 1+ pj qj (T ) ≡ F (T ) (mod pk+1 Zp [[T ]]).
j=1 j=1

We determine rk+1 ∈ Zp and qk+1 (T ) ∈ Zp [T ] such that the analogous

congruence holds with kPreplaced everywhere by k +P 1. To ease notation,
k k
put Uk (T ) = 1 + T + j=1 pj rj and Vk (T ) = 1 + j=1 pj qj (T ). (Thus,
k+1
Uk (T )Vk (T ) ≡ F (T ) (mod p Zp [[T ]]).) Then

(Uk (T ) + pk+1 rk+1 )(Vk (T ) + pk+1 qk+1 (T ))

≡ Uk (T )Vk (T ) + pk+1 (Uk (T )qk+1 (T ) + rk+1 Vk (T )) (mod pk+2 Zp [[T ]]).

Hence, we would like to choose qk+1 (T ) and rk+1 to satisfy

F (T ) − Uk (T )Vk (T )
Uk (T )qk+1 (T ) + rk+1 Vk (T ) ≡ (mod pZp [[T ]]).
pk+1
Equivalently, as Uk (T ) ≡ 1+T (mod pZp [[T ]]) and Vk (T ) ≡ 1 (mod pZp [[T ]]),
we want
F (T ) − Uk (T )Vk (T )
(1 + T )qk+1 (T ) + rk+1 ≡ (mod pZp [[T ]]).
pk+1
The right-hand side is congruent to a polynomial mod pZp [[T ]] and so we are
in a similar situation as before; we can find qk+1 (T ) and rk+1 by performing
long division in (Zp /pZp )[T ] and taking lifts.
Suppose we have chosen r1 , r2 , r3 , . . . ∈ Zp and q1 (T ), q2 (T ), q3 (T ), . . . ∈ Zp [T ]
by the above procedure. We might then expect that

F (T ) = U (T )V (T ),
∞
X ∞
X
where U (T ) := 1 + T + pj rj , V (T ) := 1 + pj qj (T ).
j=1 j=1
P∞
There’s one problem with this proposal. While j=1 pj rj is a well-defined
element ofPZp (the sum being obviously convergent), it is not clear what is
meant by j=1 pj qj (T ), an infinite sum of elements of Zp [T ].∗ Fortunately
∞

∗
This should be a power series over Zp . But its precise interpretation requires some
care: Convergence in power series rings is usually defined by the requirement that
the coefficient on each fixed power of T stabilizes, i.e., is eventually constant. But
that is not the intended meaning here.

133
this difficulty is not so serious: For each fixed nonnegative integer r, the sum
r
of the TP∞coefficients of pj qj (TP
) converges, to γr ∈ Zp (say), and we simply
j r
define j=1 p qj (T ) to mean r≥0 γr T . Having made this definition, the
factorization F (T ) = U (T )V (T ) follows easily (check that the two sides are
congruent modulo pk Zp [[T ]], for every k, and convince yourself that this implies
equality in Zp [[T ]]).
So far we have shown F (T ) = U (T )V (T ), as formal power series.
P∞ As each qj (T )
is a polynomial, a moment’s thought shows that V (T ) = 1 + j=1 pj qj (T ) is
a Strassmann series. Invoking Problem 8.94, F (x) = U (x)V (x) for all x ∈ Zp .
Since V (x) ∈ 1 + pZp for each x ∈ Zp , the Zp -zeros of F are precisely the
P∞ of U . But it is obvious U (x) = 0 for a unique x ∈ Zp , namely
same as those
x = −1 − j=1 pj rj .

Remark. This took quite a bit longer than our first approach! However, if you
understand this argument, you are well on your way to proving (one form of) the
Weierstrass preparation theorem for Qp . See the remark after the solution to Problem
13.159.

Extra Exploration 36. Let F (T ) be a Strassmann series with Zp -coefficients. Re-

ducing the coefficients of F (T ) modulo pZp yields a polynomial F̃ (T ) ∈ (Zp /pZp )[T ].
Show that if F̃ (T ) has degree 1, then there is exactly one x ∈ Zp with F (x) = 0.

Extra Exploration 37. Let p be an odd prime.

(a) Show that logp defines a group isomorphism between 1 + pZp (under multiplica-
tion) and pZp (under addition). This is analogous to how the usual log map sets
up an isomorphism between R+ and R.
Hint. One proof of injectivity goes by establishing | logp x − logp y|p = |x − y|p for all
x, y ∈ 1 + pZp . For surjectivity, change variables and apply Extra Exploration 36.
(b) Prove that every element of Z× p has a unique representation in the form ζu,
where ζ is a (p − 1)th root of unity and u ∈ 1 + pZp .
(c) Let µp−1 denote the group of (p − 1)th roots of unity in Qp . Deduce that
Z× ∼ ∼ × ∼
p = µp−1 × (1 + pZp ) = Z/(p − 1) × Zp and that Qp = Z × Z/(p − 1) × Zp .

Extra Exploration 38 (the p-adic exponential). We assume familiarity

with Extra PExplorations 35 and 37. Let p be an odd prime. For y ∈ pZp , define
1 k
expp (y) = k≥0 k! y . Show that expp maps pZp into 1 + pZp , and that in fact
∼
expp : pZp → 1 + pZp is the inverse of the isomorphism logp : 1 + pZp −
→ pZp .
m
One approach is to introduce Ep (y) := limm→∞ (1 + pm y)1/p . (Here raising to the
power 1/pm is defined by the Newton binomial expansion, as in Extra Exploration 31.)
Show that Ep maps pZp into 1+pZp and is actually the inverse of Lp : 1+pZp → pZp .
Then argue that Ep (y) = expp (y). The identity Ep (y) = expp (y) is the p-adic
analogue of the well-known formula exp(x) = limn→∞ (1 + n1 x)n .

11.134 We explain how to transform the double sum on k and j into the
k
claimed sum on j. Put uk,j = 1k≥j s(k, j)nj ak! . Since p is odd and |a|p ≤ 1/p,

134
 ak
1k≥j s(k, j)nj ≤ 1k≥j p−k |k!|−1
p ≤ 1k≥j p
−k k/(p−1)
p ≤ 1k≥j p−k/2 .
k! p

k
So if we set ϵN := p−N/2 , then |1k≥j s(k, j) ak! |p ≤ ϵN whenever k ≥ N or
j ≥ N . By Problem 8.96,
 
∞ k ∞ X ∞ ∞ X ∞
X X ak X X
 s(k, j)nj  = uk,j = uk,j
j=0
k!
k=0 k=0 j=0 j=0 k=0
 
∞ k ∞
X X a X
= nj  s(k, j)  = Ca,j nj .
j=0
k! j=0
k≥j

P∞ P∞
When n = 1, this argument shows that j=0 Ca,j = 1 + a. Since j=0 Ca,j
converges, |Ca,j |p → 0 as j → ∞.

11.135 Put vn = [xn , xn+1 , . . . , xn+d−1 ]T , so that v = v0 . According to our

recurrence relation, xn+d = a1 xn+d−1 + a2 xn+d−2 + · · · + ad xn , and so
    
0 1 0 ... 0 xn xn+1
0 0 1 ... 0    xn+1  xn+2 
   

 .. .
.. .
.. . . x x
. . ..   n+2  =  n+3 
Avn =  .  = vn+1
   
   ..   .. 
0 0 0 ... 1   .   . 
ad ad−1 ad−2 . . . a1 xn+d−1 xn+d

for each nonnegative integer n. Iterating, An v = An v0 = vn , and ⟨An v, e⟩ =

⟨vn , e⟩ = xn .

11.136 Computing the determinant by expanding along the first column gives
det(A) = ±ad . Therefore, A is invertible over any ring in which a−1
d exists,
such as Fp when p ∤ ad .

11.137 We have

xn = ⟨An v, e⟩ = ⟨Ar (Ak )m v, e⟩ = ⟨Ar (Id + pB)m v, e⟩

m   m  
r
X m j j X m j r j
= ⟨A p B v, e⟩ = p ⟨A B v, e⟩.
j=0
j j=0
j

11.138 Let s(N, K) be the Stirling numbers of the first kind. For each non-
negative integer m, Problem 11.137 gives

135
 ∞
X m(m − 1)(m − 2) · · · (m − (j − 1))
xkm+r = pj ⟨Ar B j v, e⟩
j=0
j!
∞ j
!
X pj X
= s(j, ℓ)mℓ ⟨Ar B j v, e⟩.
j=0
j!
ℓ=0

j
If we put uj,ℓ = 1ℓ≤j pj! s(j, ℓ)mℓ ⟨Ar B j v, e⟩, then |uj,ℓ |p ≤ 1ℓ≤j |j!|−1
p ·p
−j
≤
j/(p−1) −j −j/2 −N/2
1ℓ≤j p p ≤ 1ℓ≤j p . Hence, if ϵN := p , then |uj,ℓ |p ≤ ϵN when-
ever j ≥ N or ℓ ≥ N . Because ϵN → 0 as N → ∞, Exercise 8.96 allows us to
write
∞ X
X ∞ ∞ X
X ∞ ∞
X X pj
xkm+r = uj,ℓ = uj,ℓ = mℓ s(j, ℓ)⟨Ar B j v, e⟩.
j=0 ℓ=0
j!
ℓ=0 j=0 ℓ=0 j≥ℓ

So if we define
 
∞
X X pj
Fk,r (T ) =  s(j, ℓ)⟨Ar B j v, e⟩ T ℓ ,
j!
ℓ=0 j≥ℓ

then Fk,r (m) converges to xkm+r for every nonnegative integer m. In particular,
Fk,r (1) converges, so that the coefficients of T ℓ tend to 0 as ℓ → ∞. Thus,
Fk,r (T ) is a Strassmann series.

11.139 Let r ∈ {0, 1, . . . , k − 1}. Suppose xn ≠ 0 for some nonnegative integer

n ≡ r (mod k). If we write n = km + r, then Fk,r (m) = xkm+r = xn ̸= 0.
Therefore, Fk,r (T ) is not the zero series. Since Fk,r (T ) is Strassmann, Exercise
9.116 shows that Fk,r has finitely many zeros in Zp . Consequently, there are
finitely many nonnegative integers m with Fk,r (m) = 0. Equivalently, there
are finitely many nonnegative integers n ≡ r (mod k) with xn = 0.
Summarizing: Either xn vanishes for all nonnegative integers n ≡ r (mod k)
or xn vanishes for only finitely many nonnegative integers n ≡ r (mod k). The
final claim of the problem (“Hence, . . . ”) follows immediately.

Remarks.
(i) Imagine you are presented with an integer linear recurrence sequence, specified
by a list of coefficients a1 , . . . , ad ∈ Z (where ad ̸= 0) and a collection of initial
values x0 , . . . , xd−1 ∈ Z. The Skolem problem asks whether xn = 0 for some
nonnegative integer n. Frustratingly, while we have algorithms to decide this for
any sequence with d ≤ 4 (“algorithm” here meaning a procedure that is proven
to always terminate), no general algorithm is known for d ≥ 5. See [8] for a
discussion.
(ii) The Monthly article [7] of Myerson and Van der Poorten provides a highly
entertaining account of the Skolem–Mahler–Lech theorem, addressing several
aspects not discussed here.

136
Extra Exploration 39 (Shapiro [11], Myerson and Van der Poorten
[7]). Let {xn }n≥0 be an integer linear recurrence sequence. Call m ∈ Z a repeat
offender (relative to {xn }) if xn = m for infinitely many nonnegative integers n. Show
that each integer linear recurrence sequence is associated with only finitely many
repeat offenders.

11.140 These are easy to find once you go looking. For instance, let x0 = 0,
x1 = 1, and set xn = −xn−2 for all n ≥ 2. Then xn = 0 precisely when n is
even.
Here is a more compelling example, yoinked from [7]. Consider the integer
recurrence sequence {xn } satisfying

xn = 6xn−2 − 12xn−4 + 8xn−6 for all n ≥ 6,

with initial conditions x0 = 8, x1 = 0, x2 = 9, x3 = 0, x4 = 8, and x5 = 0.

Then xn = 0 when n is odd, while xn = (n − 8)2 2(n−6)/2 when n is even.
Hence, xn = 0 ⇐⇒ n is odd or n = 8.

11.141 Yes, the same result holds for recurrence sequences defined over an
arbitrary number field K.
The proof is almost the same as that given for integer recurrences. An obvious
complication is that when K = ̸ Q, a general element of K does not carry
a preassigned meaning as an element of Qp . This obstacle is by no means
insurmountable; it is resolved by embedding K into Qp , which we know is
possible for infinitely many primes p (Exercise 10.126). Let’s suppose we have
not just an embedding but an embedding+ (TM), meaning an embedding of
K into Qp , p odd, where (identifying elements of K with their images in Qp )
a1 , . . . , ad , x0 , . . . , xd−1 ∈ Zp , and ad ∈ Z×
p . Then it is not hard to convince
yourself that our proof of Skolem’s theorem goes through nearly verbatim.
(Check this!)
We know there are always embeddings, but is there always an embedding+ ?
Yes! Since K embeds into Qp for infinitely many p, it will suffice to prove the
following lemma.

Lemma. Let K be a number field and let α be a nonzero element of K. There

are only finitely many primes p for which there exists an embedding of K into
Qp with |α|p ̸= 1.

(As usual, we abuse notation and identify α with its image in Qp .)

Proof. Let m(T ) be the minimal polynomial of α over Q, scaled to have integer
coefficients. Write m(T ) = cd T d + cd−1 T d−1 + · · · + c1 T + c0 . If c0 = 0, then
T divides m(T ), and the irreducibility of m(T ) over Q forces m(T ) to be a
constant multiple of T . But then α = 0, contrary to hypothesis. So c0 ̸= 0.

137
Assume now that K has been embedded into Qp , where p is a prime not
dividing cd c0 . We will show that |α|p = 1. Since only finitely many primes
divide cd c0 , the lemma follows.
Suppose instead that |α|p > 1. Since m(α) = 0, we have cd αd = −(cd−1 αd−1 +
· · · + c0 ). However (keeping in mind that p ∤ cd ),

|cd αd |p = |α|dp > |α|d−1

p ≥ max{|cd−1 αd−1 |p , . . . , |c1 α|p , |c0 |p }
d−1
X
≥ cj α j .
j=0 p

If |α|p < 1, we argue similarly, exploiting that 1/α is a root of the reciprocal
polynomial T d m(T −1 ) = c0 T d + c1 T d−1 + · · · + cd−1 T + cd . In this situation,
c0 (1/α)d = −(c1 (1/α)d−1 + · · · + cd−1 (1/α) + cd ), and (keeping in mind that
p ∤ c0 )
 d d d−1
1 1 1
≥ max |c1 (1/α)d−1 |p , . . . , |cd−1 (1/α)|p , |cd |p

c0 = >
α p α p α p
d−1  j
X 1
≥ cd−j .
j=0
α p

Again we have a contradiction. ❚

Remark. Lech has shown that Skolem’s theorem holds for recurrence sequences
defined over arbitrary fields of characteristic 0. Today one usually refers to this
general statement as the Skolem–Mahler–Lech theorem. In this setting the embedding
required to make the proof work is guaranteed by a lemma of Cassels (see [1, Chapter
5]): Let K be a finitely generated extension of Q and let S be a finite set of nonzero
elements of K. For infinitely many primes p, there is an embedding of K into Qp
with the property that |x|p = 1 for all x ∈ S. To prove the Skolem–Mahler–Lech
Theorem, apply this with K = Q(a1 , . . . , ad , x0 , . . . , xd−1 ) (details left to you!).

References

1. J. W. S. Cassels, Local fields, London Mathematical Society Student Texts, vol. 3,

Cambridge University Press, Cambridge, 1986.
2. F. Q. Gouvêa, p-adic numbers: An introduction, third ed., Universitext, Springer,
Cham, 2020.
3. K. Hensel, Über die arithmetischen Eigenschaften der algebraischen und tran-
szendenten Zahlen. Jahresber. Dtsch. Math.-Ver. 14 (1905), 545–558.
4. N. Koblitz, p-adic numbers, p-adic analysis, and zeta-functions, second ed.,
Graduate Texts in Mathematics, vol. 58, Springer-Verlag, New York, 1984.
5. E. Landau, Differential and integral calculus, third ed., AMS Chelsea Publishing,
Providence, RI, 2001.

138
 6. H.-W. Leopoldt, Zur Approximation des p-adischen Logarithmus. Abh. Math.
Sem. Univ. Hamburg 25 (1961), 77–81.
7. G. Myerson and A. van der Poorten, Some problems concerning recurrence
sequences. Amer. Math. Monthly 102 (1995), 698–705.
8. J. Ouaknine and J. Worrell, Decision problems for linear recurrence sequences.
In: Reachability Problems, Lecture Notes in Comput. Sci., vol. 7550, Springer,
Heidelberg, 2012, pp. 21–28.
9. B. Petri, Perioden, Elementarteiler, Transzendenz. Kurt Hensels Weg zu den
p-adischen Zahlen, Dr. Hut, München, 2011. URL: http://tuprints.ulb.tu-d
armstadt.de/2785/1/DissPetri.pdf
10. A. M. Robert, A course in p-adic analysis, Grad. Texts in Math., vol. 198,
Springer-Verlag, New York, 2000.
11. H. N. Shapiro, On a theorem concerning exponential polynomials. Comm. Pure
Appl. Math. 12 (1959), 487–500.
12. P. Ullrich, Der Henselsche Beweisversuch für die Transzendenz von e, in: Mathe-
matik im Wandel, bd. 1, Franzbecker, Hildesheim, 1998, pp. 320–330.

139
140
Solutions to Set #12

12.142 We begin with a simple lemma.

Lemma. If x ∈ 1 + pk Zp , where k ∈ Z+ , then xp ∈ 1 + pk+1 Zp .

p


Proof. Write x = 1+pk r, where r ∈ Zp . Then xp = 1+pk+1 r+ pjk rj ≡
P
j≥2 j
1 (mod pk+1 Zp ), as desired. ❚

Now we return to the problem at hand. Since a ∈ pZp , applying the lemma m
m m
times will show (1 + a)p ∈ 1 + pm+1 Zp . Hence, |(1 + a)p − 1|p ≤ p−m−1 , so
pm
that (1 + a) → 1. Therefore, for any n ∈ Z,
m m
(1 + a)n = (1 + a)n lim (1 + a)p = lim (1 + a)n+p .
m→∞ m→∞

This proves the first equality claimed in the problem.

m
Whenever n + pm ≥ 0, we have (1 + a)n+p = Binom(1 + a; n + pm ), by
Exercise 11.134. This explains the second equality.
Write Binom(1 + a; n + pm ) − Binom(1 + a; n) = j≥0 Ca,j ((n + pm )j − nj ).
P
Then

|Binom(1 + a; n + pm ) − Binom(1 + a; n)|p ≤ max |Ca,j |p |(n + pm )j − nj |p

j≥0

≤ p−m max |Ca,j |p ,

j≥0

which tends to 0 as m → ∞. This yields the final equality.

Extra Exploration 40. Let p be an odd prime, and let a ∈ pZp . You have just
demonstrated one way to extend (1 + a)x from a function of x defined on Z to one
defined on Zp : Set (1 + a)x := Binom(1 + a; x). But there is another approach you
could take to extend the domain of (1 + a)x to Zp . Namely, you might define (1 + a)x
as the limit of (1 + a)xn , where xn is any sequence of integers converging to x in Qp .
This should remind you of how exponentiation of real numbers is defined when the
exponent is irrational.

141
Check that this idea works (i.e., leads to a well-defined value of (1+a)x , independent of
the particular sequence xn ) but doesn’t give anything new: (1+a)x = Binom(1+a; x),
again. What happens when p = 2?
Pm
12.143 Put F (T ) = i=1 ai ·Binom(βi ; T )−A. The results of Problems
Pm 11.134
and 12.142 imply that F (T ) is a Strassmann series with F (n) = i=1 ai βin −A
for every n ∈ Z.
Every n ∈ Z satisfying a1 β1n + · · · + am βmn
= A is a zero of F in Zp . So if
this equation has infinitely many integer solutions, then F has infinitely many
zeros in Zp . In that case, F (T ) = 0 in Qp [[T ]] (by Exercise 9.116). Then
A = F (n) + A = a1 β1n + · · · + am βm
n
for all n ∈ Z.
Pm n
Remark. It is often easy to rule out that i=1 ai βi = A for all n ∈ Z. Let’s
assume, as is natural to do, that no ai = 0.

Suppose to start off that A = 0. If m n

P
i=1 ai βi = 0 for all n ∈ Z (or even just the
nonnegative n ∈ Z), then we have a formal identity
m
! m m
X X X X X ai
0= ai βin T n = ai (βi T )n = .
i=1 i=1 i=1
1 − βi T
n≥0 n≥0
Qm
Multiplying through by j=1 (1 − βj T ) and replacing T with 1/β1 shows that
0 = a1 m
Q
j=2 (1 − βj /β1 ). Hence, β1 = βj for some j = 2, 3, . . . , m. That is, the term
β1 is repeated in the list β1 , . . . , βm . But our setup is symmetric, and so every βi
must appear at least twice in that P list. Turning things around, if some βi appears
only once, then we cannot have n i=1 ai βi
n
= 0 for all n ∈ Z. For an application, see
the Remark following the solution of Problem 12.150.

̸ 0. If m n
P
Similar reasoning can be applied Pwhen A= i=1 ai βi = A for all nonnegative
A m ai
integers n, we find that 1−T = i=1 1−βi T . This leads to the conclusion that every
element in the list 1, β1 , . . . , βm is repeated.

12.144 If α = x + yθ + zθ2 , then α′ = x + yθ′ + zθ′2 = x + yωθ + zω 2 θ2 , and

α′′ = x + yθ′′ + zθ′′2 = x + yω 2 θ + zωθ2 . Hence,
α + α′ + α′′ = 3x + yθ(1 + ω + ω 2 ) + zθ2 (1 + ω 2 + ω) = 3x.
Similarly,
α + ω 2 α′ + ωα′′ = (x + yθ + zθ2 ) + (xω 2 + yθ + zωθ2 ) + (xω + yθ + zω 2 θ2 )
= x(1 + ω 2 + ω) + 3yθ + zθ2 (1 + ω + ω 2 )
= 3yθ,
and
α + ωα′ + ω 2 α′′ = (x + yθ + zθ2 ) + (xω + yω 2 θ + zθ2 ) + (xω 2 + yωθ + zθ2 )
= x(1 + ω + ω 2 ) + yθ(1 + ω 2 + ω) + 3zθ2
= 3zθ2 .

142
12.145 Suppose |α|, |α′ | ≤ R. Since α′′ is the complex conjugate of α′ ,
we also have |α′′ | ≤ R. By Exercise 12.144, |x| ≤ 13 (|α| + |α′ | + |α′′ |) ≤ R,
1
|y| ≤ 3|θ| (|α|+|ω 2 α′ |+|ωα′′ |) ≤ |θ|
R 1
, and |z| ≤ 3|θ| ′ 2 ′′ R
2 (|α|+|ωα |+|ω α |) ≤ |θ|2 .

So there are finitely many possibilities for the integers x, y, and z, and therefore
finitely many possibilities for α = x + yθ + zθ2 .

12.146 For all ε1 , ε2 ∈ U,

L(ε1 ε2 ) = (log ε1 ε2 , 2 log |(ε1 ε2 )′ |)

= (log ε1 , 2 log |ε′1 |) + (log ε2 , 2 log |ε′2 |) = L(ε1 ) + L(ε2 ).

So L is a homomorphism. If L(ε) = 0, then log ε = 0, which implies ε = 1.

Therefore, ker(L) is trivial and L is injective. Finally, summing the x and y
coordinates of L(ε) gives log ε + 2 log |ε′ | = log(ε|ε′ |2 ) = log Nε = log 1 = 0. So
the image of L is contained in the subspace x + y = 0.

x0

Figure corresponding to the solution of Problem 12.147.

12.147 We show that if (i) and (iii) both fail, then (ii) must hold. Since (i)
fails, we may choose an x0 ∈ G with x0 ̸= 0. Let D denote the closed disc
about 0 passing through x0 . The set (G \ {0}) ∩ D is nonempty (containing,
for example, x0 ). Since (iii) fails, this set is also finite. Let g be any element
of (G \ {0}) ∩ D minimizing the Euclidean distance to 0 (among elements of
(G \ {0}) ∩ D).

143
 As g is a nonzero vector belonging to the one-dimensional subspace V :=
{(x, y) : x + y = 0} of R2 , we have that V = Rg. We leverage this observation
to show that g generates G. (Hence, (ii) holds!) It is enough to show that an
arbitrarily chosen x ∈ G belongs to ⟨g⟩. Since G ⊆ V , we may write x = tg,
where t ∈ R. Then {t}g = x − ⌊t⌋g ∈ G. (Here {t} = t − ⌊t⌋ denotes the
fractional part of t.) Since 0 ≤ {t} < 1, the vector {t}g is an element of G
closer to the origin than g. This contradicts the choice of g unless x − ⌊t⌋g = 0.
But then x = ⌊t⌋g ∈ ⟨g⟩.

12.148 In view of Problem 12.147, it is enough to show that each closed disc
about 0 intersects L(U) in a finite set.
Let D be the closed disc of radius R centered at 0. Every element of D ∩ L(U)
can be written as L(ε) where 1 ≤ ε ≤ eR and |ε′ | ≤ eR/2 . As a consequence of
Problem 12.145, there are only finitely many possibilities for ε and hence only
finitely many possibilities for L(ε).

12.149 Since θ3 = D and D is cubefree, 0 ≤ 3vp (θ) = vp (D) < 3. Hence,

vp (θ) = 0 and |θ|p = 1. Also, |ω|3p = |ω 3 |p = |1|p = 1p , so that |ω|p = 1.
Each of µ, µ′ , µ′′ ∈ Z[θ, ω]. Since Z ⊆ Zp and θ, ω ∈ Z× p , we have Z[θ, ω] ⊆ Zp .
Hence, |µ|p , |µ′ |p , |µ′′ |p ≤ 1. But |µ|p |µ′ |p |µ′′ |p = |µµ′ µ′′ |p = |1|p = 1. So it
must be that |µ|p , |µ′ |p , |µ′′ |p = 1.

12.150 Fix r ∈ Z. For each m ∈ Z, set E(m) = µr ν m + ωµ′r ν ′m + ω 2 µ′′r ν ′′m .

The equation E(m) = 0 is an equation involving only elements of L and field
operations in L. At present, our attention is fixed on the embedded copy of
L inside Qp , but if E(m) = 0 holds in that copy of L, it holds in the OG∗
version of L, which is a subfield of C.
Back in the complex numbers, |µ| = µ > 1. (Here and below | · | is the usual
complex absolute value.) Moreover, 1 = µ|µ′ µ′′ | = µ|µ′ |2 , so that |µ′ | = |µ′′ | =
1/µ < 1. It follows that |ν| = ν > 1 while |ν ′ | = |ν ′′ | < 1. As a consequence,

|ωµ′r ν ′m + ω 2 µ′′r ν ′′m | ≤ |µ′r | · |ν ′ |m + |µ′′r | · |ν ′′ |m < 1

for all large enough positive integers m, and

|E(m)| ≥ |µr ν m | − |ωµ′r ν ′m + ω 2 µ′′r ν ′′m | ≥ |µr |ν m − 1 > 0

for all sufficiently large m.

Therefore, if we pick m as a large enough positive integer, then E(m) ̸= 0 (and
this holds whether we are thinking of L as inside C or inside Qp ). So from
the Strassmann series machinery that we have built up, E(m) = 0 for only
finitely many integers m. (Note that this finiteness does not follow from the
∗
Original Gauss

144
bounds over C that we derived above. Those bounds imply that E(m) = 0
for only finitely many positive integers m, but they say nothing useful about
negative integers m.) Finally, letting r range from 0 to p − 1, we see that
µn + ωµ′n + ω 2 µ′′n = 0 for only finitely many n ∈ Z.

Remark. We can shorten the argument using the Remark following the solution
to Problem 12.143. If E(m) = 0 for all m ∈ Z, then each term in the list ν, ν ′ , ν ′′ is
repeated. But, working again in C,

|ν| = ν > 1 > |ν ′ | = |ν ′′ |,

and so ν is not repeated.

12.151 Suppose n3 +1 has all prime factors from P. Then n3 +1 = p∈P pvp for
Q
some nonnegative integers vp . Writing each vp = 3ep + rp , where rp ∈ {0, 1, 2},
Y Y
n3 + 1 = (−D)y 3 for D= p rp , y = − pep .
p∈P p∈P

Hence, (−n, y) is an integer solution to X 3 − DY 3 = 1.

By construction, D ∈ D. If D = 1, then n3 and n3 + 1 are positive perfect
cubes differing by 1. But the smallest difference between positive perfect cubes
is 23 − 13 = 7. Thus, D ∈ D \ {1}.
By Problem 12.150, each equation X 3 − DY 3 = 1, with D ∈ D \ {1}, has
finitely many integer solutions. Since #D is finite, we conclude that there are
finitely many n for which n3 + 1 has all prime factors from P. To deduce that
the largest prime factor of n3 + 1 tends to infinity, choose P as the set of
primes up to an arbitrarily prescribed bound.

Extra Exploration 41. Keep the same assumptions and notation from the proof
of the theorem. In this exercise, you will show that for each nonzero integer k, the
equation x3 − Dy 3 = k has finitely many integer solutions x, y. Equivalently, there
are finitely many pairs of integers x, y with N(x − yθ) = k.
(a) Show that the ring Z[θ]/kZ[θ] is finite, and that in fact #Z[θ]/kZ[θ] = |k|3∞ .
(b) Prove that if κ ∈ Z[θ] has Nκ = k, then κZ[θ] ⊇ kZ[θ]. Deduce from (a) that
there are finitely many possibilities for κ, up to associates.
(c) Suppose κ ∈ Z[θ] has Nκ = k. Prove that there are finitely many u ∈ U (same
meaning as above — positive units in Z[θ]) for which uκ has the form x − yθ for
some x, y ∈ Z. Conclude!

Extra Exploration 42 (continuation). Fix a finite set of primes P and a

nonzero integer d. Show that there are finitely many n ∈ Z such that both n and
n + d have all of their prime factors belonging to P.

145
Remark. As stated on Set #12, Delaunay and Nagell showed that x3 − Dy 3 = 1
has at most one integer solution with y = ̸ 0. It was discovered by Skolem [2] that
their theorem can be established by p-adic methods. Skolem’s proof uses arithmetic
in extensions of Q3 ; a more elementary p-adic proof can be found in lecture notes of
Ljunggren [1].

References

1. W. Ljunggren, Diophantine equations: a p-adic approach. Notes by R. R. Laxton

from a 1968 lecture course at the University of Nottingham.
2. T. Skolem, Anvendelse av 3-adisk analyse og ” bikropper“ til bevis for noen satser
angående visse kubiske ubestemte ligninger. Norsk Mat. Tidsskr. 34 (1952), 45–51.

146
Solutions to Set #13

Pp−1
13.152 When p − 1 | k, every term in the sum is 1, and u=1 ω(u)k = p − 1 =
1p−1|k (p − 1).
Now suppose that p − 1 ∤ k. Recall that every finite subgroup of the multi-
plicative group of a field is cyclic. Thus, we may choose a generator ζ for the
group µp−1 = {ω(1), . . . , ω(p − 1)} of (p − 1)th roots of unity in Qp . Since
ζ (p−1)k = 1 and ζ k ̸= 1,
p−1
X p−1
X p−1
X
ω(u)k = (ζ j )k = (ζ k )j
u=1 j=1 j=1

1 − ζ (p−1)k
= ζ k (1 + ζ k + · · · + ζ (p−2)k ) = ζ k = 0,
1 − ζk
which agrees with 1p−1|k (p − 1) in this case.

13.153 By Faulhaber’s formula and Exercise 13.152,

p−1
X
1p−1|k (p − 1) = (u + pϑ(u))k
u=1
p−1
X X k  X p−1
= uk + pj uk−j ϑ(u)j
u=1
j u=1
0<j≤k
p−1
X k  pj+1 X k  X
= Bk p + Bk−j + pj uk−j ϑ(u)j .
j j+1 j u=1
0<j≤k 0<j≤k

Shifting 1p−1|k (p − 1) to the right-hand side and dividing by pk,

p−1
X 1 k  pj X 1 k  X
βk + Bk−j + pj−1 uk−j ϑ(u)j = 0.
k j j+1 k j u=1
0<j≤k 0<j≤k

This becomes the relation claimed in the problem statement after the substi-
tution kj = kj k−1



j−1 .

147
13.154 Suppose the claim is false and let k be the smallest positive integer
with βk ∈
/ Zp . We will derive a contradiction from the relation
p−1
X k − 1 pj X k − 1 pj−1 X
βk + Bk−j + uk−j ϑ(u)j = 0 (*)
j−1 j(j + 1) j−1 j u=1
0<j≤k 0<j≤k

established in Problem 13.153.

Let j be an integer in the range 0 < j ≤ k. We argue below that the jth term
in both of the above sums is p-adically integral. It is then immediate from (*)
that βk ∈ Zp , contrary to our choice of k.
We start with the first of the two sums. Observe that
pj
 
vp = j − vp (j(j + 1)) ≥ j − vp ((j + 1)!)
j(j + 1)
j+1 j+1 j−1
>j− ≥j− = ≥ 0.
p−1 2 2

Hence, pj /j(j + 1) ∈ pZp . Since k−1

j−1 ∈ Z, to complete the proof that
k−1

j
j−1 Bk−j p /j(j + 1) ∈ Zp it is enough to establish that pBk−j ∈ Zp . But this
follows from our choice of k: If 0 < j < k, then k −j is a positive integer smaller
than k, so that βk−j ∈ Zp . Hence, pBk−j −1p−1|k−j (p−1) = p(k−j)βk−j ∈ pZp ,
and pBk−j ∈ Zp . If j = k, then pBk−j = pB0 = p, which is also in Zp .
The second sum is easier. There it is clear that the jth term lies in Zp as
long as pj−1 /j is p-adically integral. Integrality is obvious when j = 1. When
j−1

j ≥ 2, we can argue as follows: vp p j = j − 1 − vp (j) ≥ j − 1 − vp (j!) >
j j
j − 1 − p−1 ≥ j − 1 − 2 ≥ 0. (This argument actually shows a bit more: If
j ≥ 2, then pj−1 /j ∈ pZp .)

1p−1|k (p−1) 1p−1|k

Remark. Note that kβk = Bk − p
= Bk + p
− 1p−1|k . Thus, βk ∈ Zp
1p−1|k
implies Bk + p
∈ Zp , recovering the result of Exercise 8.99.

13.155 Let p ≥ 5 and let k ∈ 2Z+ . We refine the analysis

Pappearing
P in the
solution to Problem 13.154. Label the two sums in (*) as 1 and 2 .
P
We will show thatPthe jth term of 1 lies in pZp whenever 0 < j ≤ k and that
the jth term of 2 belongs to pZp whenever 1 < j ≤ k. Taking (*) modulo
Pp−1
pZp , we deduce that βk + u=1 uk−1 ϑ(u) ∈ pZp , as required.
For the second sum there is almost nothing to do. We noted at the end of
j−1
the solution to Problem 13.154 that p j ∈ pZp once j ≥ 2. (There we only
P
needed p ≥P3.) As the other factors in the terms of 2 are p-adic integers, our
claim for 2 follows.

148
 P
Since k is even, the only odd integer j which contributes to 1 is j = k − 1.
j k−1 k−1
p p 1p
For this j, we have k−1


j−1 Bk−j j(j+1) = (k − 1)B1 k(k−1) = − 2 k . We already
k−1
observed
P that p k ∈ pZp , while clearly − 12 ∈ Zp (p is odd). Hence, the term
of 1 corresponding to j = k − 1 belongs to pZp . Suppose now that j is even
and 0 < j ≤ k. If 0 < j < k, the p-integrality of βk−j established in Problem
13.154 implies that

pBk−j = p(k − j)βk−j + 1p−1|k−j (p − 1) ∈ Zp .

If j = k, then pBk−j = pB0 = p, and this also belongs to Zp . Thus, it will

pj−1
suffice to show that j(j+1) ∈ pZp . This follows upon observing that

pj−1
 
vp = j − 1 − vp (j(j + 1)) ≥ j − 1 − vp ((j + 1)!)
j(j + 1)
j+1 j+1
>j−1− ≥j−1− > 0,
p−1 4
keeping in mind for the last step that j ≥ 2.

13.156 We are assuming that p − 1 does not divide the even integer k, so
that p ≥ 5. Since k and k ′ are congruent mod p − 1, the integer k ′ is also not
divisible by p − 1. By Problem 13.155,
p−1
Bk X
= βk ≡ − uk−1 ϑ(u) (mod pZp ),
k u=1
p−1
Bk ′ X ′

′
= β k ′ ≡ − uk −1 ϑ(u) (mod pZp ).
k u=1

Suppose without loss of generality that k ′ ≥ k and write k ′ = k +(p−1)q. Then

′
uk −1 = uk−1 u(p−1)q ≡ uk−1 (mod pZp ), by Fermat’s little theorem. (To apply
Fermat, we identify Zp /pZp with Z/p, invoking Problem 5.62.) Substituting
above, Bkk ≡ Bkk′′ (mod pZp ).

Remark. Call an odd prime p regular if p does not divide the numerator of any of
B2 , B4 , . . . , Bp−3 . In the middle of the 19th century, Kummer showed that Fermat’s
last theorem for the exponent p holds for all regular primes p. That is, if p is a
regular prime, then xp + y p = z p has no solutions in integers x, y, z with xyz ̸= 0
(see the books of Ribenboim [8] and Washington [12] for accounts of this work).

All primes smaller than 37 are regular, while 37 is not: B32 = −37 · 683·305065927
2·3·5·17
.
Via Kummer’s congruence (the result of Problem 13.156), it is possible to show that
there are infinitely many irregular primes (odd primes that are not regular). We
sketch an argument for this due to Carlitz [2], which nicely illustrates how the facts
we have built up about Bernoulli numbers can be put to use.

149
 Looking at the Remark following the solution to Problem 4.44, we see that |B2m |∞
tends to infinity faster than any power of m. In particular, as we will use momentarily,
| B2m
2m
|∞ > 1 for all large values of m.

Let p1 , . . . , pr be any finite list of irregular primes, and let

N = 2 lcm[p1 − 1, . . . , pr − 1].
We let k run over the positive multiples of N and consider the ratios Bk /k. From the
last paragraph, we can choose k with |Bk /k|∞ > 1. Then there is a prime p dividing
the numerator of Bk /k. Since pi − 1 divides k for each i = 1, 2, . . . , r, each of our
primes pi appears in the denominator of Bk , and so also in the denominator of Bk /k
(by the Clausen–von Staudt theorem, Exercise 8.101). Thus, p is not any of p1 , . . . , pr .
A similar argument shows that p − 1 does not divide k (note that this implies p ̸= 2).
If we let k′ denote the reduction of k modulo p − 1, then k′ ∈ {2, 4, 6, . . . , p − 3}, and
by Kummer’s congruence,
Bk′ Bk
≡ ≡0 (mod pZp ).
k′ k
Therefore, p divides the numerator of Bk′ , implying that p is an irregular prime not
on our initial list. At this point in the proof Euclid is smiling down from heaven.

From the perspective of progress towards Fermat’s Last Theorem, it would certainly
be more encouraging to know that there are infinitely many regular primes. Unfor-
tunately, this question remains wide open! Of course, the urgency of the problem
has diminished somewhat in the wake of the full proof of Fermat’s Last Theorem by
Wiles and Taylor–Wiles.∗

13.157 Let p ≥ 5 be prime, and let k = φ(p3 ) − 1.

The group (Zp /p3 Zp )× can be identified with (Z/p3 )× , which has order φ(p3 ).
φ(p3 )
Hence, each x ∈ Z×p satisfies x ≡ 1 (mod p3 Zp ), and xk ≡ x1 (mod p3 Zp ).
Therefore, working in Zp modulo p3 Zp ,
p−1 p−1
X 1 X p2 X k  pj+1
Hp−1 = ≡ nk = k Bk−1 + Bk−j .
n=1
n n=1 2 j j+1
2≤j≤k

In this last expression, the term pBk has been dropped from Faulhaber’s
formula, which is harmless as k is odd and larger than 1.
Continuing, we argue that every term of the sum on j belongs to p3 Zp . Suppose
j ≥ 3. Since pBk−j ∈ Zp (as follows from Problem 8.99 or 13.154),
pj+1 pj
     
k k
vp Bk−j = vp pBk−j ≥ j − vp (j + 1)
j j+1 j j+1
j+1 j+1
≥ j − vp ((j + 1)!) > j − ≥j− ≥ 2.
p−1 4
∗
The story of which is gloriously recounted in Joshua Rosenblum and Joanne Sydney
Lessner’s 2000 musical Fermat’s Last Tango.

150
This shows that each term with j ≥ 3 belongs to p3 Zp . The remaining term,
3
corresponding to j = 2, is k2 Bk−2 p3 . As k − 2 ≡ p − 4 ̸≡ 0 (mod p − 1), we


3
have that Bk−2 ∈ Zp . Hence, k2 Bk−2 p3 ∈ p3 Zp .

2
Collecting our results so far, Hp−1 ≡ k p2 Bk−1 (mod p3 Zp ). Kummer, as
incarnated in Problem 13.156, now steps in to tell us that
Bk−1 Bp−3
≡ (mod pZp ).
k−1 p−3
Therefore,
Bp−3 2
Bk−1 ≡ (k − 1) ≡ Bp−3 (mod pZp ),
p−3 3
and
p2 p2 p2
Hp−1 ≡ k Bk−1 ≡ k Bp−3 ≡ − Bp−3 (mod p3 Zp ).
2 3 3

From this last congruence for Hp−1 , we have Hp−1 ∈ p3 Zp ⇐⇒ Bp−3 ∈ pZp .
This equivalence is the concluding assertion of the problem statement.

Remark. This problem brings to a close our study of the Bernoulli numbers. For
everything you ever wanted to know about this subject but were afraid to ask, see
Chapter 15 in the book of Ireland and Rosen [6], Chapter 9 in Cohen’s volume [4], and
the recent book [1] by T. Arakawa, T. Ibukiyama, and M. Kaneko. A bibliography
with ≈ 3000 books and papers related to Bernoulli numbers has been compiled by
Karl Dilcher, Ladislav Skula, and Ilja Sh. Slavutskii [5].

ak (xk − rk ) =
P
13.158 Since F (r) = 0, we have F (x) = F (x) − F (r) = k≥0
P Pk−1 j k−1−j
k≥0 ak (x − r) j=0 x r .

Let uk,j = 10≤j<k · ak xj rk−1−j . If we put ϵN := maxn≥N |an |p , then

|uk,j |p ≤ 10≤j<k |ak |p ≤ ϵN whenever j ≥ N or k ≥ N .

Moreover,
P P ϵN → 0P(asPF is a Strassmann series). So by Exercises 8.95 and 8.96,
k j k,j and
u j k uk,j both converge, and to the same value. Hence,
XX XX
F (x) = (x − r) uk,j = (x − r) uk,j
k j j k
X X
= (x − r) xj ak rk−1−j .
j≥0 k>j

Here thePfinal sum on k is exactly what we called bj . Therefore, if we set

G(T ) = j≥0 bj T j , we have shown that G(x) converges for all x ∈ Zp (i.e.,
G(T ) is Strassmann) and that F (x) = (x − r)G(x) for all such x.

151
Now we suppose that the Strassmann degree K of F (T ) is at least 1 and prove
that the Strassmann degree of G(T ) is K − 1. If j < K − 1, then

|bj |p ≤ max |ak rk−1−j |p ≤ max |ak |p = |aK |p .

k>j k

When j = K − 1,
X
|bK−1 |p = aK + ak rk−K .
k>K p

This last sum on k satisfies

X
ak rk−K ≤ max |ak rk−K |p ≤ max |ak |p < |aK |p .
p k>K k>K
k>K

So by survival of the greatest, |bK−1 |p = |aK |p . Finally, when j ≥ K,

|bj |p ≤ max |ak rk−1−j |p ≤ max |ak |p < |aK |p .

k>K k>K

Hence, the maximum value of |bj |p is |aK |p , and this maximum is attained for
the last time at j = K − 1. Therefore, G(T ) has Strassmann degree K − 1.

Extra Exploration 43. Recall our notation Qp ⟨T ⟩ for the ring of Strassmann
series. Problem 13.158 establishes a version of the Root-Factor theorem for Strass-
mann series viewed as functions on Zp . Show that the Root-Factor theorem is also
valid at the level of formal power series. More precisely, show that if r ∈ Zp is a root
of F (T ) ∈ Qp ⟨T ⟩, then F (T ) = (T − r)G(T ), where G(T ) ∈ Qp ⟨T ⟩ is the power
series constructed in the preceding solution.

Extra Exploration 44.

(a) Prove that if F (T ), G(T ) are nonzero elements of Qp ⟨T ⟩, then the Strassmann
degree of F G is the sum of the Strassmann degrees of F and G.
(b) Show that if X(T ) is a Strassmann series with all coefficients from Zp , then
1 + pT · X(T ) is a unit in Qp ⟨T ⟩, with inverse 1 − pT · X(T ) + p2 T 2 · X(T )2 −
p3 T 3 · X(T )3 + . . . , for an appropriate interpretation of the infinite series.
(c) Establish that the units in Qp ⟨T ⟩ are precisely the elements of Strassmann
degree 0.

13.159 Suppose to start off that F (T ) = k≥0 ak T k has Strassmann degree

P
KP= 0. Then |a0 |p > |ak |p for every k ≥ 1. Hence, for every x ∈ Zp , we have
| k≥1 ak xk |p ≤ maxk≥1 |ak |p < |a0 |p and

X X
|F (x)|p = a0 + ak xk ≥ |a0 |p − ak xk > 0.
k≥1 p k≥1 p

Therefore, F has no zeros in Zp (and no zeros is “at most K = 0 zeros”),

152
 Assuming the general claim fails, choose a counterexample F (T ) whose Strass-
mann degree K is as small as possible. Then K ≥ 1. By assumption, F has
more than K distinct zeros in Zp ; pick one of these and call it r. By Exercise
13.158, we can find a Strassmann series G(T ) of Strassmann degree K − 1
with F (x) = (x − r)G(x) for all x ∈ Zp . As K − 1 < K, we know that G has
at most K − 1 distinct zeros in Zp . But each zero of F , other than (possibly)
r, is a zero of G. Thus, there are at most 1 + (K − 1) = K zeros of F in Zp ,
contradicting the choice of F (T ).

Remark. It is enlightening to view Strassmann’s theorem through the lens of

the Weierstrass preparation theorem for Qp : Every Strassmann series F (T ) with
Strassmann degree K admits a factorization F (T ) = U (T )V (T ) where U (T ) ∈ Qp [T ]
is polynomial whose degree and Strassmann degree are both K and V (T ) = 1 + pṼ (T )
for a Strassmann series Ṽ (T ) ∈ Zp [[T ]]. You might try to prove this theorem yourself
by imitating our second solution to Problem 11.133. If you get stuck, look at [7,
pp. 54–55] or [9, pp. 166–167].

Suppose we have factored F (T ) = U (T )V (T ) as in the Weierstrass preparation

theorem. Then V (x) ∈ 1 + pZp for all x ∈ Zp ; in particular, V (x) ̸= 0. Since
F (x) = U (x)V (x) for all x ∈ Zp , we deduce that F and U have the same zeros in
Zp . But U , as a polynomial of degree K, has at most K zeros in Qp , and a fortiori
at most K zeros in Zp .∗ Thus, the Weierstrass preparation theorem can be viewed
as “explaining” Strassmann’s theorem. This reasoning is very close to Strassmann’s
original proof; the Weierstrass preparation theorem as stated here follows from the
assertions “1”, “1′ ”, and “3” appearing on page 21 of [11].

13.160 Referring back to Set #11, for each A ∈ pZp the constant term CA,0
of Binom(1 + A, T ) is
X Ak A0
CA,0 = s(k, 0) = s(0, 0) = 1.
k! 0!
k≥0

Therefore, the constant term of

Fr,± (T ) = αr Binom(1 + a; T ) − β r Binom(1 + b; T ) ∓ (α − β)

αr −β r
is αr − β r ∓ (α − β). This vanishes precisely when α−β = ±1. Inspecting the
r
−β r
table on Set #3, we see that when 0 ≤ r ≤ 9, we have αα−β = ±1 if and only
if (r, ±) ∈ {(1, +), (2, +), (3, −), (5, −)}.
For all other values of (r, ±) one computes directly (see the tables at the top
of this page) that Fr,± (0) is an 11-adic unit.
∗
Actually, each of its Qp -zeros is a Zp -zero. Since U has the same degree as Strass-
mann degree, scaling by an appropriate power of p turns U into a polynomial
with Zp -coefficients and leading coefficient a p-adic unit. Each Qp -root of such a
polynomial belongs to Zp . Cf. the proof of the Lemma in the solution to Problem
11.141.

153
 r 0 1 2 3 4 5 6 7 8 9
Fr,+ (0) mod 11Z11 9 0 0 7 3 7 8 1 3 8

r 0 1 2 3 4 5 6 7 8 9
Fr,− (0) mod 11Z11 2 4 4 0 7 0 1 5 7 1

Constant terms of Fr,± (T ) modulo 11Z11 .

Now let j ≥ 1. The T j -coefficient of Fr,± (T ) is given by

ak bk
X  
αr Ca,j − β r Cb,j = s(k, j) αr − βr .
k! k!
k≥j

k
For each k ≥ j, we have v11 (ak /k!), v11 (bk /k!) ≥ k − v11 (k!) > k − 10 = 0.9k ≥
0.9j. Therefore,
k k
 
ra rb
r r
|α Ca,j − β Cb,j |11 ≤ max s(k, j) α −β ≤ 11−0.9j .
k≥j k! k! 11

In particular, every nonconstant coefficient of Fr,± (T ) has 11-adic absolute

value less than 1.
We conclude that, apart from the four specified values of (r, ±), the series
Fr,± (T ) has constant term an 11-adic unit and all other terms belonging to
11Z11 . Hence, Fr,± (T ) has Strassmann degree 0 and no zeros in Zp (by Exercise
13.159).

13.161 The T -coefficient of Fr,± (T ) is

k k
 
ra rb
X
r r
α Ca,1 − β Cb,1 = s(k, 1) α −β .
k! k!
k≥1

Reasoning as in the solution to Problem 13.160, the terms of the right-hand

sum with k ≥ 2 make a contribution bounded in 11-adic absolute value by
11−0.9·2 , and hence also bounded by 11−2 (since no absolute value is strictly
between 11−1 and 11−2 ). Keeping in mind that s(1, 1) = 1, we conclude that

αr Ca,1 − β r Cb,1 ≡ αr a − β r b (mod 112 Z11 ). (*)

Plugging our approximations of α and a into (*), we find that the T -coefficient

of F1,+ is ≡ 4 · 11 (mod 112 Z11 ),

of F2,+ is ≡ 8 · 11 (mod 112 Z11 ),
of F3,− is ≡ 0 (mod 112 Z11 ), and
of F5,− is ≡ 6 · 11 (mod 112 Z11 ).

154
So the T -coefficients are divisible by 11 but not 112 for (r, ±) = (1, +), (2, +),
and (5, −).
Let’s take stock of what we’ve shown in this exercise and the last. If (r, ±)
is any of (1, +), (2, +), (5, −), then the constant term of Fr,± (T ) is 0. The T -
coefficient has 11-adic absolute value 11−1 . And the T j coefficient has 11-adic
absolute value at most 11−0.9·2 < 11−1 , for each j ≥ 2. It follows that Fr,± (T )
has Strassmann degree 1.

13.162 We saw already in the solution to Problem 13.161 that F3,− (T ) has
T -coefficient divisible by 112 .
The T 2 -coefficient of F3,− (T ) is given by
k k
 
3a 3b
X
3 3
α Ca,2 − β Cb,2 = s(k, 2) α −β .
k! k!
k≥2

Here the terms of the right-hand sum corresponding to k ≥ 3 make a contribu-

tion bounded in 11-adic absolute value by 11−0.9·3 , and hence also by 11−3 .
Since s(2, 2) = 1 (s(2, 2) is the coefficient of T 2 in T (T − 1)),
a2 b2
α3 Ca,2 − β 3 Cb,2 ≡ α3
− β3 (mod 113 Z11 ).
2 2
Working with the right-hand side, we find that
α3 Ca,2 − β 3 Cb,2 ≡ 8 · 112 (mod 113 Z11 ).
Hence, the T 2 -coefficient of F3,− (T ) is divisible by 112 but not 113 .
For every j ≥ 3, the T j coefficient of F3,− (T ) is bounded in absolute value by
11−0.9·3 < 11−2 .
Summing up: F3,− (T ) has vanishing constant term, T -coefficient bounded
in absolute value by 11−2 , T 2 -coefficient with absolute value equal to 11−2 ,
and T j coefficients with absolute value strictly smaller than 11−2 for j ≥ 3.
Therefore, F3,− (T ) has Strassmann degree 2.
n n
−β
13.163 If αα−β = ±1, then Fr,± (n) = 0, where r is the remainder when n is
divided by 10. In Exercise 13.160, we found that Fr,± has no zeros in Zp (let
alone in Z!) except possibly if (r, ±) ∈ {(1, +), (2, +), (3, −), (5, −)}.
In Exercise 13.161, we showed that Fr,± (T ) has Strassmann degree 1 for each of
(r, ±) = (1, +), (2, +), and (5, −). By Strassmann’s Theorem (Exercise 13.159),
each corresponding series Fr,± (T ) has at most one zero in Zp . Similarly, the
result of Exercise 13.162 shows that F3,− (T ) has at most 2 zeros in Zp .
Consequently, there are at most 1 + 1 + 1 + 2 = 5 integers n for which
αn −β n
α−β = ±1. We know five such integers already: n = 1, 2, 3, 5, 13. (See the
table on Set #3.) Thus, there can be no others.

155
Extra Exploration 45 (Skolem, Chowla, and Lewis [10]). Prove that
n n
−β
each integer appears at most three times in the sequence { αα−β }n≥0 (for the same
α, β as above).

Extra Exploration 46 (Cohen and Ljunggren [3]). Find all integer solu-
tions to x2 + 11 = 3m .

Extra Exploration 47. What are all of the positive integer solutions to 2x2 +1 =
3m ? Equivalently: Which squares have ternary expansions consisting entirely of the
digit 1?

References

1. T. Arakawa, T. Ibukiyama, and M. Kaneko, Bernoulli numbers and zeta functions,

Springer Monogr. Math., Springer, Tokyo, 2014.
2. L. Carlitz, Note on irregular primes. Proc. Amer. Math. Soc. 5 (1954), 329–331.
3. E. L. Cohen, Sur l’équation diophantienne x2 + 11 = 3k . C. R. Acad. Sci. Paris
Sér. A-B 275 (1972), A5–A7.
4. H. Cohen, Number theory. Vol. II. Analytic and modern tools, Graduate Texts
in Mathematics, vol. 240, Springer, New York, 2007.
5. K. Dilcher, L. Skula, and I. Sh. Slavutskii, A bibliography of Bernoulli numbers.
Online resource. URL: https://www.mscs.dal.ca/~dilcher/bernoulli.html
6. K. Ireland and M. Rosen, A classical introduction to modern number theory,
second ed., Graduate Texts in Mathematics, vol. 84, Springer-Verlag, New York,
1990.
7. D. J. Lewis, Diophantine equations: p-adic methods. In: Studies in Number
Theory, MAA Stud. Math., vol. 6, pp. 25–75. Math. Assoc. America, Buffalo,
NY, 1969.
8. P. Ribenboim, 13 lectures on Fermat’s last theorem, Springer-Verlag, New York-
Heidelberg, 1979.
9. T. Skolem, Ein Verfahren zur Behandlung gewisser exponentialer Gleichun-
gen und diophantischer Gleichungen. Comptes Rendus Congr. Math. Scand.
(Stockholm, 1934), 163–188.
10. T. Skolem, S. Chowla, and D. J. Lewis, The diophantine equation 2n+2 − 7 = x2
and related problems. Proc. Amer. Math. Soc. 10 (1959), 663–669.
11. R. Strassmann, Über den Wertevorrat von Potenzreihen im Gebiet der p-adischen
Zahlen. J. Reine Angew. Math. 159 (1928), 13–28.
12. L. C. Washington, Introduction to cyclotomic fields, second ed., Graduate Texts
in Mathematics, vol. 83, Springer-Verlag, New York, 1997.

156
Index

Symbols structure of Z× p 134

structure of Z× p /(Z × 2
p ) 33, 114
1C , indicator function of C 32 Z(p) , ring of p-integral rationals 8, 62,
Bk see Bernoulli numbers 84
Hn see harmonic number expp see p-adic exponential
Sk (n), sum of kth powers of first n logp see p-adic logarithm
nonnegative integers 15 ω(u) see Teichmüller representatives
see also Faulhaber’s formula s(N, K), Stirling numbers of the first
Binom(1 + a; T ) see Strassmann series kind 43
representing (1 + a)n vp (n) see p-adic valuation
Qp , field of p-adic numbers |x|∞ , familiar (Archimedean) abs. value
base p expansions of elements 29, 4
100 |x|p see p-adic absolute value
convergence of series 31, 105 x3 − Dy 3 = 1, finiteness of solutions
element has eventually periodic base p 49, 144
expansion iff rational 16, 29, 78, x3 − Dy 3 = k, finiteness of solutions
100 145
has Q as a dense subset 27, 97
introduction and basic properties 24
is complete 27, 97 A
structure of Q× p 134
structure of Q× × 2
p /(Qp ) 33, 114
Qp ⟨T ⟩, ring of Strassmann series (Tate absolute value
algebra) 119 p-adic 3, 24
(Z/p)× is cyclic, proof using Teichmüller Archimedean 4
classification of all abs. values on
√ lift 111
Z[ 3 D], arithmetic of 45, 142 Fp (T ) 81
Zg , ring of g-adic integers classification of all abs. values on Q
Q
decomposition as p|g Zp 20, 86 (Ostrowski’s theorem) 17, 80
example of Z10 34, 117, 128 definition 3
Zp , ring of p-adic integers equivalence 17, 80
base p expansions of elements 28, 99 non-Archimedean 4
has Z as a dense subset 27, 97 of factorials 4, 58
Hensel’s definition 89 product formula on Q 4, 58
introduction and basic properties 19, survival of the greatest 4
83 trivial absolute value 3
is compact 25, 93 Archimedean absolute value 4

157
B Fermat’s last theorem 149

Bernoulli numbers H
Adams’ theorem 51, 148
harmonic number 5, 17, 52, 59, 81, 150
alternate in sign 16, 76
Hensel’s lemma 38, 125, 126
associated characterization of Wilson
primes 35, 120 I
associated congruence for Hp−1 52,
150 irregular prime 149
Clausen–von Staudt theorem 32, 109
definition 15 L
distribution of denominators 109
La Cara de la Guerra 20
Faulhaber’s formula 16, 75, 108, 120,
Legendre’s formula for vp (n!) 58
147
limits (sequential) 11, 16, 67, 79
in terms of ζ-values 77
Liouville’s approximation theorem in Zp
Kummer’s congruence 51, 149
34, 115
vanish for odd indices > 1 16, 76
logarithm, on Qp see p-adic logarithm
C M
canonical expansions of elements of Qp method of successive approximation
29, 100 29, 33, 101, 113, 131
Cantor–Schröder–Bernstein theorem Monsky’s theorem 21, 88
87
Cauchy sequence 23, 91 N
Clausen–von Staudt theorem 32, 109
Newton’s method 38, 125
completeness of a valued field 27
non-Archimedean absolute value 4
completion of a valued field 28
is unique 28, 98 O
convergence of infinite product 105
convergence of infinite series 12, 31, 68, Ostrowski’s theorem 17, 80
105
convergence of sequences see limits P
(sequential)
p-adic
cubic Pellian equation 49, 144
absolute value 3, 24
valuation 3, 24
D
p-adic exponential 134
discs, open and closed 7, 61 p-adic integer see Zp , ring of p-adic
integers
E p-adic logarithm
definition and convergence on 1 + pZp
equivalent absolute values 17, 80 41, 129
exponential function, on Qp see p-adic proof of additivity 130
exponential sets up isomorphism between 1 + pZp
and pZp 134
F p-adic number see Qp , field of p-adic
numbers
Faulhaber’s formula 16, 75, 108, 120, p-integral rational number see Z(p) ,
147 ring of p-integral rationals

158
primes in arithmetic progressions 9, 65 Strassmann degree of 52
product formula for absolute values on Strassmann’s theorem 52, 152
Q 4, 58 strictly convergent power series see
Strassmann series
R strong triangle inequality 4
extension to infinite series in Qp 31,
Ramanujan–Nagell equation 13, 52, 72,
105
155
survival of the greatest 4
regular prime 149
respectable mathematician 11
T
non-example, see bottom of 32
restricted power series see Strassmann
Taylor’s formula 38, 125
series
Teichmüller representatives 32, 35, 51,
S 110, 121, 147
trivial absolute value 3
sadness in our hearts 27
simultaneous approximation 58 V
Skolem–Mahler–Lech theorem 44, 136
Strassmann degree 52 valued field 3
Strassmann series
definition 34 W
form a subring of Qp [[T ]] 119
if nonzero has finitely many zeros in Weierstrass preparation theorem for Qp
Zp 35, 120 134, 153
number of zeros bounded by Wieferich prime 8
Strassmann degree 52, 152 Wilson prime 35
representing (1 + a)n 43, 45, 134, 141 Wilson’s theorem 25, 94, 101
representing terms of a linear Wolstenholme prime 52
recurrence 44, 135 Wolstenholme’s theorem 5, 59

159