382

Chapter 8. Formal Series Revisited 8.4. Lagrange’s Inversion Theorem

383

function of the right-hand side is

Here is a quite exotic problem: suppose that ¢ & K| [[T7] and consider the
Y 3!2“7}’9)';‘ )'_*T Sk L(T_fl:’.“i’{(fl'w—m sequence a, = [T"](e"(T)). What is the generating functio
n of the sequence
n 0gk<n 5% bDk G a,? The following result answers this question in a more general
context:
Thus, by dividing by e¥7 it suffices to Theorem
prove that 8.A.8. Let X, Y be variables satisfying Y = Xe(Y), where e €
KT has nonzero constant term. Then for any F € K{[T]] we have
RS
DTFEDery) - xn = EO
k=1
But this is a consequence of theorem 8.A.5
with n>0
FT) = e*T and e(T) = T
i}
Let us apply this result to prove the follo Proof. Apply theorem 8.A.5 to f(T) = [T g,g‘)ldu (recall that this is formal
wing nice-looking identity.
Ezample 8.A.7. Prove that for all n > integration, so f(T) is the unique formal series vanishing at 0 and
1 we have such that
fu)e(u) = F(u)). Using that f' = %, we obtain
> (ff)(z: F DTG 1P = 20 ),
0,520t 5=n \' fy=3" i(Y/e(Y))" (TTTHET)T))
AMM E. 2898 >l
Proof. Take @ =1, z = 1 and ¥ =n+1in the previous theorem. We obtain
- %[T"“](F(T)e(T)"‘l)X'".
n>1

3 (t")u
idjmn ¢
PTG 1Y = (s 2, Differentiating this equality with respect to ¥ we obtain
Unfortunately, this is not really what
2. ) = S PO
we want to prove, but it shows that we
are on the right track. To get rid of daX
the extra 1 in the exponent of j + 1,
differenti ate with respect to y the equality in the previ ay’
ous theorem and then
take a =1, z 5= 1 and y = n + 1. This
time we end up with
Finally, differentiating ¥ = Xe(Y') we obtain dY = dX - e(Y)+Xe(Y)dY,
so
a2 = 3 (;’)(H DTG+ 1
idjmn dX 11— Xé(Y)
Now, observing that j = J+ 1~ 1, we can write the last sum as
vy e(y)

¥ (ari+nga
(7

igmn
1) ni-
(G 1) — ¥ (36P+ 1) v16+1) 1)
n Replacing this in the previous equality yields the desired
equality. [m)
ik jmn
Ezample 8.A.9. Find a closed form for the generating function of
Combining this with the first relation, the result follows the sequence
. O an, where ay is the constant term of (1+ X + Eol
384 Chapter 8. Formal Series Revisited 8.A. Lagrange’s Inversion Theorem 385

Proof. Note that a, = [T"}(e"(T)), where e(T") = T% + T -+ 1. So, by the vertices, so nj + ng + -+ ng = n — 1. The number of ways to distribute the
previous theorem n — 1 vertices different from the root in these k subtrees is

Z X" 1T-
=17 XXYy (n—l)'(nfl—m _ (=1}
ny np T onptngl-
- ong!
where ¥ = Xe(Y). Solving the equation in Y yields
Once such a distribution is made, we have an,an, -+ an, ways to label the

yo Lo~ XXoyl-28
- JT=2X=3X%
- 357 elements of the forest and so the contribution to the total number is
(n— ])'
nylngl - ’ama,., G
and then an easy computation shows that
The total contribution coming from all partitions of n — 1 is
EanX" = S . =
- Vi-2X -3x? {(n—1)!
nylng!- . nk!anlauz TGy
b=y 12
8.A.3 Examples from enumerative combinatorics however each configuration is counted k! times (since we do not care about
In this section we consider applications of the inversion formula in count- the order of the children of the root) and the main root can be chosen in n
ing problems. We start with an absolutely classical and beautiful theorem of ways. We finally deduce the very complicated recurrence relation
Cayley, but we need a series of definitions before stating it. Recall that a tree
is a connected graph with no cycle. A labeled tree on the set {1,2,....n} is
a tree whose set of vertices is {1,2,...,n}. A rooted tree is a tree in which
SR K
D D M T
1
it U2
(n— 1)
! )
one of the vertices (called the root) is distingnished. There is a unique (non- This simplifies drastically if we consider the exponential generating function
backtracking) path between any two vertices of a tree. The parent of a vertex T(X} =300 )iT , since
in a rooted tree is the vertex connected to it on the unique path to the root.
fyn—Ti k_ Gy- Oy,
A child of a vertex v is a vertex whose parent is v. A tree is called ordered if Xk = 3 Foet
one is given an ordering of the children of each vertex. nrdtng=n—1

Theorem 8.A.10. There are n""2? unordered labeled trees on the set Hence the recurrence relation can also be written
(12w}, glr: =@y TXF _ (X1 T ),
ki
Proof. Let a, be the number of unordered rooted labeled trees on {1,2 o, %
with the natural convention that ag== 0. It is enough to prove that a, a1,
that is Xe?(X) = T{X). Using Lagrange’s inversion formula, we obtain
as clearly the number of unordered labeled trees is an/n.
nn—l
The point is that giving such a tree is the same as giving its root and a
forest of subtrees whose roots are the children of the root. Suppose that the = = XOUT0) = e = T
root has k children and that the corresponding subtrees have ny,ng, ..., ng from where the result foliows. 0
386 Chapter 8. Formal Series Revisited 8.4. Lagrange’s Inversion Theorem 387

We end this addendum with two more difficult examples. The following If we expand eA7/2 and collect terms according to the exponent of T, we finally
beautiful result is taken from [65]. deduce that
Ezample 8. A.11. An intransitive tree on the set of vertices {1,2,...,n} is a tree
F"*n—.zmg(k)" : o
— 1 o~ (n n-1
such that for all 1 < i < j <k <n, {i,j} and {j,k} are not simultaneously
edges. Prove that the number of such trees is
Finally, a question by James Propp with a nice proof from [69].

e 1 "
=
()
/n -1 Ezample
1,2,1,2,...
8.A.12. The vertices of a polygon P with N + 2 vertices are labeled
in order (stopping when the end is reached). Let ax be the number
of tndngnlat ions of P with no monochromatic triangle. Then ay = 2"H (3"
Note that it is absolutely not clear that the above quantity is an integer! ifN=2nanday = 2 (") N =2+ 1.
Proof. Let F,, be the number of such trees and let Proof. Define ap = 1. Suppose that N = 2n+1 and call a triangulation proper
if it contains no monochromatic triangle. Consider a proper triangulation 7
= Z Fn+1% of P. Note that P has an edge labeled 1,1. This edge must be a side of a
n20 triangle with a vertex labeled 2. If this vertex is the ith vertex labeled 2,
with i > 0, then the two sides of the triangle split P into a 2i + 2-gon and a
be an associated exponential generating function. Call a vertex i left if all of 2n - 2i+ 2-gon and both of these polygons are properly triangulated by 7. By
its neighbors are greater than 4. Let L, be the number of rooted intransitive adding over all i we obtain a/;,,“= i oG2iG2n—2; and a similar argument
trees on the set of vertices {1,2 ,n}, whose root is a left vertex. Then for N even yields ag, = Z;fl) @z, 1. This suggests considering the two
Ly =1 and we clearly have L, = 2F, forn > 2 {n comes from the choice of generating functions
the root, division by 2 takes into account the fact that the probability that
the root is left is 1/2). If A(X):ZaZHXfl: B(X):Ea‘mflx"-
™
)
T)= g e
Ln— n2t n>0
Then the previous relations can be written in a compact form
is the exponential generatmg function associated to Ly, then L, = §F, for
n > 2 yields L(T)= £(1 + F(T)). But the exponential fortmula unphas that A(X)=2X(1+ AX)B(X), B(X)=(1+A(X)~%
F(T) = eMT) a5 an mtramxtlve tree on the set of vertices {1,2,...,n+ 1}is Indeed, we have
obtained from a forest of left-rooted trees on {1,2,...,n}, by connecting n+1
to each root. Thus, we obtain F(T) = eFUHF(TY) g HT) = T(1 + F(T)),
we deduce that f(T)= 7(1+e/(T¥2), An application of Lagrange’s inversion BX) =3oz X" Z(Zaz-afz(fi))
formula yields nz0 720
2
F, = [T :;[Tn—l](l F T %[an} b (z) T2 SanXt] =1+ 4(X))?
20
(n— 1)t k=0
388 Chapter 8. Formal Series Reuvisited 8.A. Logrange’s Inversion Theorem 389

where
21 2n+1
MXD = 3 AUSD - A0S SRR
Z (Z a;a0n, 1_1) X"=X-3" (Z azaznfl»z> X (S1.52,86)
n>1 220 \ i=0
the sum being taken again over ordered partitions of X. For instance, con-
=X Z (Z @2:02n 21%1) Xt+ X Z (E Aig1G2m 2:) X" sider the problem of counting the number of partitions (Sy,Ss, ..., . Sk) of
n>0 - n>0 G {1.2,...,n}, where each S; is nonempty. By taking f(ny=1ifn>1and
= 2X(1
4+ A(X))B(X). o)y= 0 we deduce that the exponential generating function for the number
of such partitions is
We deduce that A(X) = 2X(1+A(X))® and an easy application of Lagrange’s

BROO) = (X~ 1)t = Z( 12 (5) e

inversion formula finishes the proof. [m}

8.A.4 Composition of generating functions

One of the key points in the proof of Cayley’s theorem 8.A.18 is to estab- and expanding /X yields the desired number of partitions.
lish the functional equation Xe'X) = T(X) for the exponential generating Suppose now that f| (0)= 0. We would like to understand the generating
function of the number of labeled trees. We would like to give a more ab- function E, o Ey.
stract and general context for this kind of argument, which appears very often
Theorem 8.A.13. Lei f.g : N > K be sequences such that F{0) =0 and
in counting problems. We follow rather closely the wonderful book [76] and
9(0)= 1. Then Ey o Ef = Ej, where h : N = K is a sequence such that
we strongly advise the reader to take a look at the first chapter of it, which
h{(0)=1 and for ull finite sets X we have
contains an impressive number of examples and problems on this topi
Let K be a field of characteristic 0 and let f,g: N~ K be two sequences
of elements of K. Let Ef =35~ f(n) n, " be the generating function of f. We R(XD) =3 gl - £(1S31) - F(ISal) -+ FUSKD,
would like to give a combinatorial interpretation of the generating functions
Ef - Ey and Ey o E,. Note that By - By = Ej, where the sum being taken over all unordered partitions (S,Sz ..... Sz} (with arbi-
trary k) of X into nonempty subsets.
Wy =3k=0 (3) st = Proof. We clearly have b = 37,-, gk}, where
We deduce that for any finite set X we have
=SS S,
XD = 32 0sha(x - = 37 7(SiDg(1seD,
SCX {81,52)
the sum being taken over unordered partitions with & classes. Hence it is
the second sum being taken over all ordered partitions (84, 52) of X (and 8y, B
enough to prove that B, = T;‘-L But this follows from the previous discussion
Sy may be empty). By an obvious induction, we deduce that
and the fact that we are only considering unordered partitions here (thus the k
Ep o Bpeooee Ey, = En, classes may be permuted in k! ways and yield the same unordered partition). O
390 Chapter 8. Formal Series Revisited 8.A. Lagrange’s Inversion Theorem 391

Ezample 8.A.14. Let us consider again the problem of finding the functional
cycles of 0. Then theorem 8.A.13 yields®
equation for the exponential generating function of the number of unordered
rooted trees on {1,2,...,n}. Let T be this generating function. Then by the
previous theorem XeT(X) is the generating function for the number
(r, F), where 7 is a root and F is a forest of unordered rooted trees starting
of pairs B = B | a13 m
from this root. But it is clear that any unordered rooted tree arises in this
Ezample 8.A.18. For nonnegative integers ¢1,¢g,..., let an{c1,¢2,...) be the
way, so we actually have XeT™) = 7(X).
number of permutations ¢ € S, having ¢; cycles of length i for all i < n.
Ezample 8.A.15. Suppose that we want to count rooted unordered labeled Consider indeterminates X;,Xo,.... Then the previous example yields the
trees such that the number of children of each node is in a fixed set S, con- following cyele-index formula
taining 0. The argument used in the previous example yields the functional
equation . ™ ™
3 anlenen
) X{X§ e XS —=ep | Xp—
fX)=x3" fi?: 120 n! n
1,620 Cn 20 a>1
565 :
Ezample 8.A.19. Let us count the number of permutations of odd order, i.e.
Using Lagrange’s inversion formula, we obtain a formula for the number of for which all cycles have odd length. By taking X; = 1 when 7 is odd and 0
such trees,
otherwise, we deduce that the exponential generating function for this counting
Ezample 8.A.16. Let E be the generating function for the number of connected problem is
graphs with vertices 1,2,...,n. Giving a graph with vertices 1,2,. .., n is the
same as giving a family of disjoint connected graphs (its connected compo-
nents), thus by theorem 8.A.13 the generating function for the graphs with
vertices 1,2,...,n is e£. But since there are 2('2') such graphs, we deduce that

o (D01
_ log(1+7)—log(1 - T7)
wy X7™
E=1 .2
8 Z 2 7! 1+7T
nz0
“Vior
Ezample 8.A.17. Consider two sequences f,g such that £(0) = 0, g(0y =1 =(1+T)(1-T% %
and define 2(0} = 1 and )
" Note that we alo have
RIXD = 37 glky-
S - £(Cal) - FUCD, XD = 37 6®I 1SS ~ 1)t FGSel)ISl ~ 1),
aeSym(X)
the sum being taken over unordered partitions 7 of X, since the cycle decomposition of a
permutation yields a partition of X and since one can cyclically permute in (1S,] — 1)! ways
where Sym(X) is the set of permutations of X and C3,C,... ,Cr are the the elements of a class with |S:| elements.
392 Chapter 8. Formal Series Revisited 8.A. Lagrange’s Inversion Theorem
393

and an easy applica ion of the binomial formula yields the number of permuta- Cayley’s theorem is a direct consequence of the previous theorem
and of
tions f(n) = (1-3--- (n—1))% when n is even and f(n) = (1-3-----(n—2))%-n the multinomial formula: the number of trees on {L2,...,n}is
when » is odd.
Ezample 8.A.20. Let k be a positive integer and let f(n) be the number of — _ (n—2)! = (n—2)!
permutations ¢ € S, such that ¢ = 1, This is equivalent to the fact that the diredd=in- (dr—1)t- (. - 1)! it 12l
length of each cycle of o divides k. Thus by the previous examples
=141+ +1)"2=pt2
d
Ef =exp Z i(d’ Proposition 8.A.22. There are (377)-(n~1)""*~1 labeled trees with vertices
dk 1.2,....n in which vertex 1 has degree k.

8.A.5 More tree-counting problems Proof. This is 50 an easy consequence of the previous theorem. The desired
In this section we present another proof of Cayley’s theorem as well as number of trees is
some similar counting probles, all related to trees. The following general
{n—2)! n— ke
result is quite useful in problems concerning trees.
Ftdatotda=2(n—1)
k=)l (dz— 1)1 (.z,,vl)!:(kvl)'("_l) o
Theorem 8.A.21. Let dy,da,. ... dy be positive integers such that dy + da +
oo dp = 20— 2. Then the number of trees on the set {vy,va,...,va} such the second equality being a consequence of the multinomial formula.
]
that verter v; has degree d; is ca ESIEC?
Let us introduce a very useful notion in graph theory.
Proof. We will prove the result by induction on n. We may assume that
dy, = 1, by permuting the d;'s if ne: ry. Consider a tree on {v1,v2,...,v5} Definition 8.A.23. Let G be a loopless graph. A spanning
forest is a sub-
such that deg(v;) = d; and remove vertex v, and the unmique edge whose graph without cycles and having the same vertices as G. A spannin
g tree is a
endpoint is v,. We obtain a tree on {v1,v2,...,%,.1} whose degrees are connected spanning forest.
dy, ..y djea,dj — 1,djh, .. dnea if 0, is connected o v;. Conversely, any
such tree on {v1,...,¢n.1} yields a tree on {v1,...,v,} simply by connecting . Here is a nice application of Abel’s identity.
vy with vy, It follows that there are Ezample 8.A.24. There are (n —2) . n"3 spanning trees of K,, which do not
contain a fixed edge of K,,.
sy (n - 3)!
Fl‘(dl = (dj =21 o {dpyoy ~ 1)1 Proof. Call 1,2, ..., n the vertices of K, and assume without loss of generalit
y
that the fixed edge is e = 12. Let f(n) be the number of spanning
n-1 trees that
(n—3)! (n ! contain e. Such a tree appears uniquely as a result of the following
= dp—n+1] - = v process:
(; * ) =D~ TI{ds
— 1)t consider two trees T,T2 whose vertices form a partition of 1,2,...,
n with 1
a vertex of T} and 2 a vertex of 75, and then Jjoin these two trees by
the edge
such trees and the result follows. O e. If Ty has k vertices different from 1, these vertices can be chosen in (";2)
394 Chapter 8. Formal Series Revisited

ways. Once these vertices are chosen, we have (k + 1)1 possibilities for T
and (n — k ~ 1)"572 pogsibilities for T5. Thus
2
f(n):z(nk -2 )(k+1)k1-(
-
n~k—1) k3k=3 9. a3 pne8)
k)
the last equality being an easy consequence of example 8.A.7. The result
Chapter 9
follows. O

Remark 8.A.25. Here is another approach, suggested by Richard Stong. Let

X, be the probability that a randomly chosen spanning subtree of K,, contains
A Little Introduction to
the edge e. Then by symimetry it is clear that E[X.] is the same for all edges
¢. Since any spanning subtree of K, has exactly n — 1 edges we have Algebraic Number Theory
S EX]=n-1.
This rather long chapter is concerned with elementary
Therefore since all {3) terms in this sum are equal, algebraic number
theory. The techniqgues are rather diverse: basic linear algebra
, algebraic num-
bers and symmetric polynomials, cyclotomy and p-adic
analysis are some of
the topics discussed in this chapter. Since we will use the
notion of algebraic
number quite often in this chapter, we end this introdu
ction with a few rec-
Hence by Cayley’s formula, there are 2- 2"~ spanning subtrees containing e oliections. For more details and some proofs, the reader
is referred to the
and (n - 2)n* % that do not contain e. addendum 9.B.
A complex number 2 is called algebraic if it is root of some
nonzero poly-
nomial with rational coefficients. In this case, there exists
a unique monic poly-
nomial with rational coefficients, called the minimal pelyno
mial of z, which
vanishes at z and has minimal degree. The roots of this
polynomial are called
the conjugates of z. The crucial property of the minima
l polynomial is that
it is irreducible over Q and divides any polynomial with
rational coefficients
that vanishes at z. A fundamental theorem in algebraic
number theory states
that the algebraic numbers form an algebraically closed
subfield of Q, thus an
algebraic closure of Q. If 2 is an algebraic number, we let Q(z)
(or Qj2]) be
the subfield of C generated by z. It consists of all numbers of
the form f(z),
with f € Q[X] (or equivalently f € Q(X)). This is a finite
extension of Q,
of degree equal to the degree of the minimal polynomial of
z. The primitive
396 Chapter 9. A Litile Introduction to Algebraic Number Theory 9.1, Tools from linear algebra
397

element theorem ensures that all finite extensions of Q are of the form Q(z) implies the existence of integers A, B,C such that Au+ By + Cw
= 1. We
for some algebraic number z. We call such extensions number fields. We will deduce that Aua + Bva = a (mod w). Since au+bv =0 (mod w), we deduce
frequently use the notation [L : K| to denote the dimension of L as K -vector that (4b ~ Ba)v = —a (mod w), so that we can take p = Ab— Ba to get
space, as well as the fundamental tower relation [M : K] = [M : L} - [L : K| up = b (mod w). Also, we can immediately check that
for any finite extensions M/L/K.
A more refined notion is that of algebraic integer. This is a complex u(Ab— Ba) = Aub— Bua = b{Au + Bv) =b (mod w). m}
number that is killed by some monic polynomial with integer coefficients. By
Proof. We will actually prove a stronger result: for any integers a, b, ¢ and
Gauss’ lemma, we can characterize algebraic integers as those algebraic num-
auy integers u, v, w such that au + bv + cw = 0 and ged (u,v,w) = 1, there
bers whose minimal polynomial has integer coefficients. An easy but funda-
exist integers A, B, C such that a = Bw—Cv, b= Cu— Aw and ¢ = Av— Bu.
mental result is that a rational number which is also an algebraic integer is
Indeed, since ged(x,v,w) = 1, a standard application of Bézout’s lemma
necessarily a rational integer. Another important result is that the algebraic
yields the existence of integers X,Y, Z such that Xu + Yv+ Zw = 1. Let us
integers form a subring of the field of algebraic numbers.
define
A=cY ~bZ, B=aZ-cX, C=bX—ay.
9.1 Tools from linear algebra Then,
In this section we consider a few applications of linear algebra to num-
Bw-Cv=
- cX)w—
(a(bX Z
—aY)v
ber theory. These concern especially divisibility issues and linear diophantine
equations. =a(Xu+
Zw) — Yv
X{au + bv++ cw)
=a.
1. Let a,b, ¢ be relatively prime nonzero integers. Prove that for any rela-
tively prime integers u, v, w satisfying au+bv+cw = 0, there are integers Thus ¢ = Bw — Cv and similarly b = Cu — Aw and ¢ = Av — Bu.
The result
m,n, p such that follows.
]

@ - pu, b= pu— mw, ¢=my—nu.

A very nice and classical result is that Ticicjcn "7 is an integer for
any integers @1,az,. ... an. There are many proofs of this result, at least two
of them being presented in [3]. The following problem is a variation on
Oct n Standsild, Romanian TST 1989 this
topic.
Proof. Counsider the linear system in the variables m,n,p
2. Prove that for any integers aj, as,. .., a, the number
g = nw -~ pu, b= pu—mw, c=mv— nu
lem{ay, a,, -@n) H (4 —as)
Trivially, the determinant of this system is 0 and the rank of its associated ajag -
1027 Gn 1<i<i<n
matrix is 2. It is thus enough to solve in integers the system ¢ = nuw — pr,
b = pu~—mw. This system has integer solutions if and only if there is an integer is an integer divisible by 112!--- (n ~ 2)!. Moreover, we cannot replace
p such that vp = —a (med w) and up = b {(mod w). Now, the hypothesis 112!--- (n - 2)! by any other multiple of 112! - .- (n—2)1.
 398 Chapter 9. A Little Introduction to Algebraic Number Theory 9.1. Tools from linear algebra
399

Proof. Consider the matrix A = {a;,;}1<j<n

: with a5 = (’;‘:21) for j > 2 and
lem(ay To see that the result is optimal, simply choose a,, = (n1)? and ¢; =
(n)?+i
G,p: == e v ) We will prove that for 1 <i <n. Then

det(4) = lem(ay, az,. .-, an) . Thicicjenlts ~ ai) lem(ei,as,

.. a,)
. = nlnay - - ap_q
- ajag - ay 2l (n—2)1
because the numbers ay,, Li=12, 7 — 1 are pairwise relatively prime,
Since the entries of A are integers, it is clear that its determinant is an integer,
The result Tollows easily from this. o
from which the first part of the problem will follow.
Factoring an L = lem(ay, a2, ..., a,) out of the first column and multi- Remark 9.1. Quantities such as 1L <i “—;—:'fl and the one in the previous
plying the i-th row by a;, it follows that prob-
lem have natural combinatorial interpretations: they are the dimensio
ns of
some irreducible representations of special unitary groups. Of
course, explain-
e 1 (7 - @oh L e afa=1) - a8} ing this is beyond the scope of this modest book, but the reader should konow
B L0 e DL [ a w1 o e that these are not “just some random problems.
St amean |t : = . Remark 9.2. A similar result is proved in the beautiful paper [6]: if
() e ) 11 an anlan=1) - (%)) ag,a1,...,a, are integers, then [T, <j(a? - a}) is a multiple of %—?—"fl
and
this result is optimal. This is also related to the dimension of some
Taking out the numbers % that appear at the denominators of the binomials irreducible
representations of the symplectic group.
in each column shows that
We continue with a nice application of linear-algebraic arguments. The
1 e afar o aPlge ideas used in the following solutions are very useful in other contexts,
2 too.
L Sl az az—agz - ey -1 e
det A =~
aras - anli2l- - (1 — 251 !5 \{3. Let p be a prime and let aq, 4z, ..., ap+1 be real numbers such that
no
1oan al—an - afl4e. matter how we eliminate one of them. the remaining numbers
can be
divided into at least two nonempty pairwise disjoint subsets each having
Note that the (4,7) entry of this matrix can be written as Pj(a;}, where P;, the same arithmetic mean. Prove that @) =ay = --- = api1-
0 < j £ n—1isa monic polynomial of degree j. For each column 4 add
a suitable linear combination of the previous columns to reduce the previous Marius Ridulescu, Romanian TST 1994
determinant to
Proof. Subtracting from the a;’s their arithmetic mean (observe that the new
1 ap az a’{’]l numbers have the same property), we may assume that arFaz+
o +apy =0,
ay af; - ayo Fix some 1 < j < p+ 1 and let Cy,...,C, be the classes of a partition
of
{e1,-..,ap11} — {a;}, such that IFI,[ 3sec, @ does not depend on [. Since
ap - anll o : the
sum of the a;’s is zero and since
1 an a2 - ant
Since the last determinant is Vandermonde, the identity stated above is proved.
p2) JE
bt zeC;
i
|
|
|
!
g
 400 Chapter 9. A Little Introduction to Algebraic Number Theory 9.1. Tools from linear algebra 401

we deduce that we have a linear relation of the form Let |S;1] = m and observe that 37, [Si;| = p—m and
1 a
Iraais @agwl, + N =0,
Gl (p—-m) Zx =m Z z fiz =0 (modp)
2€8i1 TEA-Siy TEA;
with |Ci} < p.
Now, consider all such linear relations, obtained by making 7 run over all Summing over all choices of i we obtain
1,2,...,p+1. This gives us a linear system with p + 1 equations and p + 1
P+l
unknowns (the a;’s), whose matrix has }1; on the main diagonal and numbers of
N a=r (modp) = a;i=r (modp)Vie{L,2,...,p+1}
1 the form % with k < p elsewhere. But then the determinant of the matrix will =1
|
1 be of the form 47 for some rational m = 1 (mod p). Thus, the determinant
is nonzero and since the system is homogeneous, the only sclution is the trivial Thus, we can write ¢; = pb; + r for some integers b;. But clearly
one. This implies that all a;’s are zero and the conclusion follows. ]
{b1,b2,...,bpp1} also satisfy the conditions of the problem and moreover
max {b;| < maxla;|. By the inductive hypothesis all b;s are equal and so
all g;’s are equal O
Proof. First, we will reduce the problem to the case when all g; are integers.
The following method is classical and very useful in a whole variety of situa- Proof. Here's a “no formula™ proof, which uses the same kind of argument,
tions: consider the vector space spanned over Q by the ey’s. This is a finite but replaces the choice of a basis in a vector space with an approximation
dimensional Q-vector space and if we take a basis of it and write each a; as argument: first, we reduce to the case when all a; are integers in the follow-
a linéar combination with rational coefficients of the elements of the basis, ing way. By Dirichlet’s approximation theorem there exists a large integer
we easily see that the coordinates of the a;'s also satisfy the conditions of M such that all Ma; are very close to some integer 4;. The linear equations
the problem (becanse by definition the elements of the basis are linearly in- deduced from the fact that the ¢;’s satisfy the conditions of the problem be-
dependent over (2). Working coordinate by coordinate reduces therefore the come approximate linear equations for the A;’s. But there are finitely many
problem to the case when all g; are rational. Multiplying all a;’s by N! for such equations and each has rational coefficients. Thus, if at the beginning we
some sufficiently large N reduces then the problem to the case when all a; are ensured that Ma; are sufficiently close to the A;’s, the approximate equations
integers. in A; are actually exact. Thus the A;’s are integers satisfying the conditions
Assume now that all ¢; are integers and let us prove the result by induction of the problem. If we solved the problem over the integers, it follows that all
on max ja;|. The base case is obvious, so let us focus on the inductive step. A; are equal. But then any two a;’s are less than 2/M apart and since M is
Removing every element a; gives sets S;; C {a1,a2,...ap¢1} ~ {a;} = A; non- arbitrarily large, this implies that all a; are equal.
empty, pairwise disjoint so that TS«I‘]T Poe 5, independent of 7, say equal Now, let us assume that the a;’s are integers. If we remove one number,
to k. Then the common rational arithmetic mean for the rest of the numbers cannot have
p in the denominator, so the sum of all other numbers is rational with p in
the numerator and, thereby, an integer divisible by p. Hence all numbers have
the same remainder modulo p as their sum. Now continue as in the end of the
previous solution. o0
402 Chapter 9. A Little Introduction to Algebraic Number Theory 9.2. Cyclotomy 403

9.2 Cyclotomy Theorem 9.6. For all n there are infinitely many primes p=1 (mod n).

There are @(n) primitive nth roots of unity, namely 5 where k is Proof. For k > n large enough (it is actually enough to take &k > 2) we have
relatively prime to n. Hence the nth cyclotomic polynomial @n(k!) > 1 and so we can choose some Prlon(k!). Since ¢,(0) is 1 or ~1, we
have ¢ (k) = 1,—1 (mod k!), which obviously implies that ged(pg, k1) = 1.
o(X)= [l (x-¢% As k > n we get p > k > n and by the previous theorem we deduce that
1<k<n pr=1 (mod n). The result follows. [m]
god{k,n)=1
For the next three problems we will use a very useful rationality result:
if
has degree ¢(r). The splitting field Q(e%) of ¢y, is called the nth cyclotomic ~ and cos(rm) are both rational numbers, then cos(rr) € {d:l,i%,‘()}n Let us
extension of ). These polynomials snd their splitting fields play a very im- recall the argument: 2cos(rm) = €™ 4 ¢ " and the numbers ™, e~ are
portant role in mauy areas of mathematics and gave rise to a whole series of algebraic integers (they are roots of unity), so 2cos(rr) is an algebraic integer.
very deep results. Their study would require a whole book by itself, so we Thus, if it is rational, it has to be a rational integer and the result follows.
decided to focus only on some very elementary and classical applications. Before passing to the next problem, let us discuss a beautiful consequence of
Since any nth root of unity in C is primitive of order d for a unique djn, the previous observation. We will prove that the only regular n-gons all
of
we get the: whose vertices are lattice points are the squares. Indeed, let A, B, C be three
consecutive vertices of the polygon and observe that
Proposition 9.3. (Fundamental identity) We have
l+cos®™ 2r (AB®+ BC? - AC??
X7 — 1 =[] éulX). 7
din
TS T T e <@
Using the previous observation, the result follows easily. We strongly advise
This easily implies (by strong induction) that ¢(X) € Z{X] for all n. the reader to look for a geometric proof in order to appreciate the power
of
The following result is not trivial and plays an important role in many proofs algebraic numbers!
concerning cyclotomic polynomials. We'll also see that a weak form of Dirich-
4. Let A, B,C be lattice points such that the angles of triangle ABC
let’s theorem follows very easily from it. For a proof of Dirichlet’s theorem in are
rational muitiples of 7. Prove that triangle ABC is right and isosceles.
full generality, see addendum 7.A.

Theorem 9.4. Let a be an integer and let p be a prime divisor of ¢n(a). Then Proof. Note that any angle # = ZABC with A, B, and C lattice points must
cither the order of a modulo p is n (and so p =1 (mod n)) or p divides n. have tan# rational or infinity. To see this note that all Lines between lattice
points have rational or infinite slopes and if tano and tan # are rational {or
Proof. By assumption, p divides ¢,(a), and s0 if ¢ has order k {mod p) then infinite) then so is tan(a — ) = %%4 This implies that
kln. Ik < n, then p divides both a* - 1 and ‘;{Z% {the second because of the
fundamental identity and the fact that p divides ¢n(a)). As ged (ak -1, a@l) tan?fd
= sec’§ ~ 1 =
1—cos26
1+ cos28
divides by the Euclidean algorithm, p|n and we're done. 0
is rational and hence cos 26 is rational. Combining this with the discussion
Remark 9.5. Note that the proof also works for prime powers p. preceding the problem shows that cos 24, cos2B, cos2C are all equal to +1,
404 Chapter 9. A Little Introduction to Algebraic Number Theory 9.2. Cyclotomy 405

:i:% and 0. It is immediate to check that the condition tanA, tan B, tanC and we have [Q(2) : Q(z)] = 2. Indeed, 2z = 2 + 2z}, which implies that z
rational or infinite says A4, B,C must be integer multiples of =/4. Hence the satisfies a quadratic equation with coefficients in Q(z), so [Q(z) : Q(z)] < 2.
only possibility is when ABC is right and isosceles. : [m} On the other hand, we cannot have Q(z) = Q(z), because z is not a real
number. Putting these observations together, we deduce chat has degree
5. Let o be a rational number with 0 < o < 1 and “’(") Using the previous result and the fact that cos (7 —2) = sinz, we can
compute the degree of sin 2". The answer is a bit complicated: if n # 4,
cos(3ra) + 2cos(2ra) = 0. the degree of sin 22 is £4Y if 8 divides n, £ if ged(n,8) = 4 and @(n) if
ged(n,8) < 4.
Prove that o =
IMO Shortlist 1991 6. Prove that none of the numbers v/n + 1 —~ y/n for positive integers n can
be the written in the form 2cos (%Z) for some integers k,m.
Proof. Let @ = cosma and observe that the equation satisfied by « can be
Chinese Olympiad
written as

da®
4 4?32
- 2= 0 = (2z+1)(22% +z—2) =0, Proof. Flr% we will find a polynomial with # = yn+1 — f as a root.
We have 2° = 2n + 1 ~2\/n2+n and so (2% — 20 — 1)? = 4n? + 4n, from
Of course, if © = %, we must have o 3 and we are done. The difficult point where we easily find that 2% — 2{2r + 1)2% + 1 = 0. Note that the other toots
is to prove that we cannot have 2¢? +z — 2 = 0. If this is the case, then z = of the polynomial f(X) = X* —2(2n + DX% +1are 2 = =R+ 1+ /A
:h’z‘fii, because |z| < 1. We will then prove that cos(2"ra) takes infinitely Next, we will find a polynomial with roots & = 2cos 2’"’
2Z Let T, be the mth
many values as 7 runs over the positive integers. This will clearly contradict Chebyshev polynomial, defined by the equality T}, (cos x) cosmaz for all z.
the hypothesis that « is rational. But since cos(2%ra) = 2cos?(2" ') — 1, Then Ty, (§) = cos2kn= 1. Thus the numbers 2cos2% for k=0,...,m — 1
it is easy to prove that we can write are roots of g{X) = Tj, (X ) — 1. These m numbers are not d]st.mct, but
2cos %FT" 2cos gfi"’m'gk)" for 1 < k < m/2 are double roots of this polynomial
b, a2+ 1782 — 8 since g eves a local maximum at these points. Thus these are the only
cos(2an) = %, brgt = Gnbp, Gnp1 =~ —2—”7
roots of g{X).
The previous relations yield by induction that a, by, are odd integers and that
If +1—/n=2
v cos2 then f(X) and g(X) have
a common factor
in Z{X]. The only roots of f whlch Lie in the interval [—2, 2] (which contains
Gnt1 > Gn. Thus cos(2%me) takes infinitely many values. [m}
all roots of g) are vVn+1 — /n and /n — /n + 1. Therefore this common
Remark 9.7. In general, let us'choose relatively prime integers m, n thh n >2 factor is either X — (V¥ 1 ~ /1) or X2~ (vn+1 — yR)2 In either case
and find the degree of the algebraic number z = cos (?"”) Define z = ™%
e we see that (Vi + 1~ n)? = 2n+ 1 — 2y/n{n+ 1) is an integer and hence
primitive n-th root of unity. The irreducibility of the cyclotomic polynomials n(n+1) is a square. But this would make 4n(n +1) and (2n+1)? consecutive
(which is a very nontrivial theorem) implies that z has degree p(n) as an positive squares, a coniradiction. )
algebraic number. On the other hand, we have
‘We continue with a very beautiful and classical result ([50]) concerning
Q) : Q) = {Q(z) : Q)] - [Ql=) - QY linear equations in roots of unity.
 406 Chapter 9. A Little Introduction to Algebraic Number Theory 9.2. Cyclotomy 407

7. a) Suppose that ay,ag,...,ax are rational numbers and (ST O If we combine the results of the previous two paragraphs, we see that we
are 1oots of unity such that a;(; +age + - - - + agl = 0. Moreover, must have j = 1, as z is killed by the nonzero polynomial Z;:ul X", of degree
suppose that .., a;¢; % 0 for any proper subset I of {1.2,....k}. at most p — 1. But then 2 has degree p — 1 over Q(ez‘—v:!) (as follows from the
Prove that {* = " for all 4, j, where m is the product of primes
previous computation) and so 35 » 5 X" is the minimal polynomial of z over
smaller than or oqual to k.
Q(«?). As 2 is also killed by 1+ X +--- + X?~1, we deduce that these two
b) Let 2 be a complex number. Prove that there are at most 2% . k% k-
polynomials differ by a constant. In particular, all & are nonzero. So for all
suples (¢1,¢2,. .., (k) of roots of unity with the following property:
0 < < pone can find ¢ such that r; = [. Clearly, this implies that p < k and
there exist rational numbers ay, ag, ., such that z = Ef‘;l a;(;
the proof is finished.
and 2 % 37 a;¢; for any proper subset 7 of {1,2,..., k}.
b) Fix a solution z = Z:‘:l a;¢; of the equation and consider another
Mann’s theorem solution, say z = Zle b;z;. Thus
Proof. a) We may assume that a; = ¢; = 1. Let m be the least positive integer
such that (f*= 1 for all ¢ and choose a prime factor p of m. If m = pin with k k

ged(n,p) = 1, we will prove that j = 1 and p < k. This will imply that m DaGi-Y bim=0
divides ]Ipqp and the first part of the theorem will follow. =1 =1
Proving this is
however not a simple task.
but one has to be a little bit careful, as this relation does not necessarily satisfy
We start with an observation: let z = e El and let ¢ be an m-th root of the conditions of (a). However, if we fix 1 < ¢ < &, we can find 2 minimal
unity. We claim that there exists 0 < 7 < p and = such that z5 = 1 and sub-relation of the previous relation which contains z;. By hypothesis, such a
¢ == 2" -z. This is very easy: if { = > , simply choose 0 < 7 < p such that sub-relation must contain some ;. As the length of this sub-relation is at most
o= (mod p). 2k and as it clearly satisfies the hypothesis of (a), we deduce that 2[" = ("
Applying this observation to each (;, we can write o= 2% with @y, for all ¢; in this sub-relation. Here m = H <24P- So, for any i, z can take at
as above, Wc have @) = 1 and r; = 0. The equation }_“_1 aIC,=0 can be most km values and so the number of solutlons of the equation in 21,29, ..., 2,
written Y P b2t where by = Z,I:l a;z;. Note that by € Q(e’= ), On the is at most (km)*. It remains to use Erdés’s famous inequality (theorem 3.A.3)
Hp<n p < 4" 1o conclude. ]
other hand, we can compute the degree of z over Q(e?fifi). Indeed, observe
that Q(z.¢ 7 ) = @(egfvr), so that
Remark 9.8. Let a1, as, . , a, be nonzero complex numbers and consider the
2ir
equation a12; + @222+ -+ - +@nz, = 1. A non-degenerate solution is an n-tuple
Q%)
2imp
)) : Q)]
piasd
)
= [@(P ~):Q)
) (L
= p(m)
elmipy
= o (p7)
{z1,29,-..,2,) of roots of unity which satisfies the equation and such that
(!
3 icr @izi # 0 for any nonempty subset [ of {1,2,...,n}. Conway and Jones
and the last qudntxty is p—1 for j = 1 and p otherwise.
{20] improved Mann’s thmrem by proving that 1f a; € Q, then for any non-
Note that, Etg i X ¥ is not the zero polynomial, since otherwise we obtain
degenerate solution we have 2§ = 2§ = -~ = zd = I where d is the product of
the relation Er,«t ;¢ =0 for all 0 < I < p. But the hypoth yields then primes p1,po.. .., ,ps such that 377 (p;— 2) < n~— 1. Also, in [30] the author
{ilri =1} =0 or {1,2,...,k} for all L. Asry = 0, this gives r; = 0 for all ¢ proves using rather elementary and very beautiful arguments that there are
and so (;” = 1 for all 4, contradicting the minimality of m. at most (n + 13" non-degenerate solutions of the equation,
408 Chapter 9. A Little Introduction to Algebraic Number Theory 9.3. The ged trick 409

9.3 The ged trick and

The division algorithm shows that if K C L are fields and if f,g ¢ K[X] 9(X) = X*" —pX" +1 = (X" -2 ") (X" - 27).
are two polynomials, then their ged is the same if we see f, g as polynomials The crucial claim is that
with coefficients in K or with coefficients in L. That is, the greatest common
divisor of two polynomials is not sensitive to the field in which the coefficients ged(p.q) = X* — (2 +27)X + 1= (X - 2)(X - 27).
of these polynomials live. Combining this observation with Gauss’ lemma, we
also obtain that if f and g are monic polynomials with integer coefficients, Assuming this for a moment, we can conclude that = +z~! is an integer by
then their ged computed in Q[X] has integer coefficients. the ged trick.
This gives a very
indirect, but sometimes very useful way to prove the rationality or integrality It remains to establish the claim and for that it is enough to
prove that z
of a real number z: it is enough to exhibit X —z as the ged of two polynomials and z~! are the only common zeros of 7,9 {since clearly p, g have no
multiple
with rational coefficients (respectively of two monic polynomials with integer root). But if z is a common zero, we have 2™ = £ or ™ = z-m and similarl
y
coefficients). The next problems in this section illustrate this trick. 2" =a" or 2" = 27", We may assume (by changing z and 27!) that z™ = 2™,
so that [z} > 1. Then clearly we must have 2 = 2”. But then z/z is
atoot of
8. Let a,b be two positive rational numbers such that for some n > 2 the unity whose order divides both m and n. Since ged{m,n} = 1, it follows
that
number ¥/a -+ /& is rational. Prove that /7 is also rational. z =z and we are done.
]
Marius Cavachi, Gazeta Matematici The following problem is very similar to the previous problem,
but a bit
more difficult.
Proof. Let us write /a+ /b = c for some (positive) rational number ¢. Then
/ais aroot of X™ — a and also of (¢~ X)™ —b. The key point is that it is the 10. Let 8 € (0,7/2) be an angle such that cos4 is irrational. Suppose that
unique common root of these polynomials, Indeed, if 2 is a common root, then cos k@ and cos[(k -+ 1)6] are rational for some positive integer k. Prove
we can write z = {/az and ¢~z = {‘/EZ2 for some nth roots of unity zy, 22. that § = /6.
We deduce that ¥/a+ Vb = {3z + VB2, Since |z] = 1, the real parts of
21,2 are at most 1. Passing to real parts in the previous equality then implies USA TST 2007
that z; = 2z = 1 and the claim is proved. Now, since the two polynomials
Proof. We will actually prove more: it is enough to replace k+1
don’t have multiple roots, it follows that ged(X™ — ¢, (c— X)* - b) = X — /a. by any integer
I which is relatively prime to k. The key point is the followin
The result follows now from the ged trick. g
n)
Lemma 9.9. If coskf and coslf are rational for relatively
prime positive
9. Let m,n be relatively prime numbers and let z > 1 be a real number integers k,1, then either cos8 is rational or 8 is o rational multiple
of 7.
such that 2™ + 1.‘1-‘ and 2™+ T%, are integers. Prove that z + % is also an
integer.
Proof. 1f coskf = p and coslf = q, then ¢ is a common root of the polyno-
mials
Proof. Let aw= ™ + —t FXY=X%—0pX* 41, g(X)=X" —2¢X' + 1.
B and b = 2™ + <.
@ and consider the polynomials
On the other hand, it is not difficult to check that if 6 is not a rational
PX) = X7 — a X 1= (X — 7YX g™ of 7, then ¢ and e~ are the only common roots of f and g. Indeed,
multiple
all roots
410 Chapter 9. A Little Introduction to Algebraic Number Theory 9.4. The theorem of symm(itn"(t polynomials
411

of f are ¥+ 5 for 0 < 7 < k and all roots of g are e*?+ % with 0 < i<t Proof. We will use induction on n and
inside the induction step an induction
On the other hand, since ged(k,!) = 1, the only solution of the equation on deg(f). Forn=1 everything is clear,
so assume the result holds for n — 1.
EOT I ki R g < i1 < kand 0 < jy < (for some choices of We now prove by induction on deg(f) the assertion
of the theorem with n
signs) is j1 = ja = 0. This proves that the greatest common divisor of fandg variables. If deg(f) = 0 or 1, everything is clear. It
is clear that the pelyno-
is precisely (X ~e?)(X —e %) = X2~ 2cos X + 1, thus cos @ is rational. mial g(Xy, ..., X,q) = f(X7,... s Xn-1,0) is
still symmetric, so by (the first)
induction it is a polynomial of the form
MXy + o+ Xy, .)&1 o Xasy)
Coming back to the proof, the previous lemma shows that § is a rational for some 4 € R|Xy,... + Xn-1]. Note that
the difference '
multiple of 7. On the other hand, we saw in section 9.2 that the only rational
numbers 7 & [0, 1] such that cosrr is rational are r =0, 1,1, 2, 1. We deduce
that k@ and 16 are integer multiples of . Since ged(k 1, Bézout’s lemma
implies that 8 is an integer multiple of %. Since cos@ is irrational, we deduce vanishes when X,, = 0 and is a symm
etric polynomial. Therefore this poly-
that § = &. nomial is a multiple of X; .. -+ X,. Appl
0 ying the inductive hypothesis to t.l«w,
quotient between this polynomial and
Xi--- X, (which has degree less than
deg f). the result follows.
9.4 The theorem of symmetric ]
polynomials
Ramark 9.11. 1t is not difficult to prove
that the polynomial g is unique.
The proof of the following result is quite elementary, but the result it- This means that there are no algebraic
relations between the polynomials
self s incredibly powerful and useful. If R is a commutative ring and if T1,02,...,0n.
f € R[X1, Xz, ..., Xs} is a polynomial, we say that f is symmetric if for Remark 9.12. The theorem also implies that any symme
all permutations o of {1,2,...,n} we have tric rational function
fFeR(Xy, Xa,.... X3} is a rational function
in the a;'s. Indeed, let
F(Xn o Xn) = F(Xoys - Kom)-
T PG Xy Xn) = P00, Koy, Xgy)
Recall that the fundamental symmetric polynomials are for Pe R[Xl.,Xg, ++., Xn]. Then we can write

ok= 3 XaXy oo X,
1< <ig<-<ipSn 7P _ B
Q Ihes,oQ
for 1 <k < n. We have the equality for some polynomials P,Q, Py. Since f
is symmetric, so is P;.
The result follows from the theorem of
symumetric polynomials applied to
EF X+ X)Xy = o " o, € RIEL X, X Brand to [[,0-Q.
Remark 9.13. We rofer the reader to {66], chapter 5 for
Theorem 9.10. (Fundamental theorem of symmetric polynomials.) Let R be the proof of the
following theorem of Lagrange: let K be a field of characteri
a commutative ring and let f € R[Xy,..., Xn] be a symmetric polynomial. stic 0. If
I { K(X1,X,,.. <+ Xp), let G be the set (actually group) of those permu
Then there is g € R[Xy,..., Xy] such that f(Xa,..., Xo) = glor.09,...,05). -
tations o € S, such that (X
Xa,,X)) = f(X,(l),X,m,A4..X‘,(,,>}. I
412 Chapter 9. A Little Introduction to Algebraic Number Theory 9.4. The theorem of symmetric polynomials
413

fg € K(X1,Xy,..., Xy) satisfy Gy C Gy, then one can find a rational func- 11. Let a,b,c be integers. Define the sequence (ZTn)nzo by 2o = 4, 3 = 0,
tion h whose coeflicients are symmetric polynomials in Xj, X»,..., X, such 72 = 2¢, z3 = 3b and Tp43 = ap_; + by + cTpyy. Prove that for any
that g = A(f). prime p and any positive integer m, the number apm is divisible by p.
A very important consequence of theorem 9.10 is the following result, that Calin Popescu, Romanian TST 2004
will be constantly used in this section.
Proof. Let r1,72,73,74 be the roots of the characteristic polynomial of the
Corollary 8.14. @) Let f € Q[X1,Xs,...,X,] be a symmetric polyno- recurrence relation, namely X* — ¢X2 — bX —a. The crucial point and by far
mial and let g € Q[X] be a polynomial of degree n, with complex roots the hardest step in the proof is to realize that
203, 2 Then f(z1,22,..
.0 22) € Q.
Tn =17 +75 + 75 41
b) If f has integer coefficients and if g is monic with integer coefficients,
then f(z1,22,-..,2n) is an integer. for all n. This is suggested by 2 = 4 and by the fact that problem
creators
tend to try to be sneaky.! Proving the previous formula is immediate by
Proof. Using theorem 9/10, we can write induction, once we prove it for n = 0,1,2,3. For n = 0,1 this
is trivial, for
n = 2 follows from the identity
F(X1,-- Xn) = hlo1,00,. ., 0n)
2
er = (Zr,-) — Zngrj =2¢
for some h € Q[Xy,..., Xy} (resp Z[X1,...,Xn]). The result follows from
i<
the fact that o321, 22,...,2n) are rational {respectively integers), because the
and for n = 3 we can use the recursive relation (since it is easy to see that
coefficients of ¢ are so. ]
Yn =77 + 75 + 15 + 1} together with y_; = —b/a satisfies the same recursive
Another very useful result is the following generalization of Fermat's little relation as z,,). With this closed formula for the general term of the sequence,
theorem. we need to prove that erm is a multiple of p. Since 3"r; = 0, the
result
follows from corollary 9.15 and by induction on m.
Corollary 9.15. Let f € Z[X] be a monic polynomial with complez roots o
21,22, -1 2 (mudiiplicities counted) ond let p be a prime number. Then Let us consider now a few more or less direct applications of theorem 9.10
and of corollary 9.14.
A4+ +B=(m+zm+-+zm)f (modp).
12. a} Let P, R be polynomials with rational coefficients such that P #0.
Proof. Corollary 9.14 implies that both sides are integers. Consider the quo- Prove that there exists a non-zero polynomial Q € Q[X] such that
tient by p of the difference between the left-hand side and the right-hand side. PX)IQ(R(X)
Using the multinomial formula, it is easy to see that this quotient is a symmet- b) Let P, R be polynomials with integer coefficients and suppose that
ric polynomial with integer coefficients in 21, 2z, . . ., 2, thus the result follows P is monic. Prove that there exists a monic polynomial Qe Z[X!
from corollary 9.14. [m} such that P(X)|Q(R(X))
Here is a nice application of the previous corollary. It was one of the Iranian Olympiad 2006
difficult problems given in the Romanian IMO Team Selection Tests in 2004. *As Richard Stong kindly remarks. ..
 414 Chapter 9. A Little Introduction to Algebraic Numb
er Theory 9.4. The theorem of symmetric polynomials
415

Proof. The idea is very natur

al: the first condition that should
order to have P(X)|Q(R(X)) is that for be satisfied in Evaluating the relation fig) + fogs =1 at a; yields falai)ga(a:) =
each oot z of P we have Q(R(z)) = 0. 1. Thus
Therefore, if zy, 23, ..., 5, are
the roots of P (some of the z;'s may
then we would like to have Q(R(x;)) be equal),
= 0. The most natural choice is to take {H falas) '!Hm(as) =1L
=1 i=1

QUX) = [[(X = Ray). On the other hand, [T, f2(a;) and 1% g2(a;) are integers, by corollar
y 9.14.
i=1 These two observations are enough to conclude.
Note that it satisfies P(X)IQ(R(X)), becau b) Note that |a; —b;| = 1 for all , 7, because a;,b; are integers. It is then
se X ~ z; divides R(X) ~ R{x;)
for all 4. It remains to check that Q has immediate that we have only two possible cases:
rational (respectively integer, for the
second part of the problem) coefficients. 1) A, B are singletons of the form {a}, {a+1}or {a+1},{a}
This follows from corollary 9.14. O
\.( 13- 2) A= {a} and B = {a — 1.a+ 1} for some integer @ or B = {a} and
a) Let ajag,... am,
by, by, - b, € C be such that A={a—1l,a+1}.
Thus, by symmetry in A and B and by making a translation of the variable
A =TIx ~a). fa(x) = f[(x- b) € Z[X). X = X —a, it is enough to consider the cases when 4 — {0}, B = {1} and
w1 i=]
A={0},B = {-1,1}. In each case f; divides some X" and f2 divides some
Suppose that there exist gy, 92 € ZIX] such
that fig; + fagy = 1.
(X?—1)™ Thus f; divides X** and f» divides (X** - 1)" for k,n sufficiently
Prove that: large. It is thus enough to find Bézous relations with integral coefficients for
the polynomials X2° and (X2* — 1). But this is immediate. =]
! "o | Remark 9.16. The assumption that a; and b; are integers is useless.
HH((Z,, byl =1 Here is a
proof, due to Richard Stong. We advise the reader not familiar with
{1 =1 the notion
of resultant to read the discussion before problem 27 in chapter
b) If a;, b; are integers and 12. It is not
too difficult to check that the resultant of f; and faisTIZ, ;‘:I (bj—a;) = £1.
lm n But then the map
IHH(a, —bi) =1,
Jimt =1 # ZiXlacg( 1231 X 21X aeg( )1 ~* ZX aent 1) 4ot fo)-1
prove that there exist polynomials 91,92 defined by (g1,4) = gi(X)f1{X) + g2{X) fo(X) is invertible, thus
€ Z{X] such that fig; + we can
fagz = 1. find g1,g2 € Z{X] such that fig1 + fogo = 1.
A classical problem is to prove that
Tbero-American Olympiad
Proof. a) Note that the relation to be prove
d can also be written as
) 1
e+ V22 e
fz(lh)f =1 for any integers @, b, not both of them equal to 0. The idea is that it is not
clear how to deal with |a + bv/2| directly, but it is very easy to say someth
ing
416 Chapter 9. A Little Introduction to Algebraic Number Theory 9.4. The theorem of symmetric polynomials 417

about the product of this number and its conjugate ja — bv/2|. Indeed, this we have [[77/(1 — z;) = ¢ and so
product is a nonzero integer, thus at least 1. The result follows immmediately.
—1
With a similar idea, it is not difficult to prove the following-absolutely classical
theorem of Liouville: if z is an algebraic irrational number of degree d, then HP %) = q"HQ(A)
=1 i=1
there exists ¢ > 0 such that for all integers p and ¢ we have |z — §
That is, irrational algebraic numbers are badly approximable with ration‘a‘l But the same mgumenl as before shows that ]']'_] Q) is a nonzero integer,
numbers. A much deeper result, for which Roth won the Fields medal, is that therefore [Li P(z;)| > ¢*. This is however impossible, since by assumption
we can improve the previous inequality to ‘z - g‘ > —"%52— for all € > 0. The
we have {P(z;)] < n + 1, therefore
following problems use this trick of multiplying by conjugates and estimating g1
the conjugates, but they are much more challenging than the very simple [PE) <m+177 < g~
example discussed above. =1
The previous arguments show that P vanishes at one of the primitive roots
¥14. Let k,n be positive integers and let P(X) be a polynomial of degree n of unity of order ¢. But since the polynomial 1+ X +--- 4+ X% is irreducible
with all coefficients in the set {~1,0,1}. Supposc Lhdt (X - 1)*PX) over the rational numbers, if P vanishes at a primitive root of unity of order
and that there exists a prime ¢ such that lnq < ]n(n Fuyl Prove that the g, then it also vanishes at all the other roots. This ends the proof. a
primitive complex roots of unity of order ¢ are roots of P. One needs some gymnastics if one wants to avoid the use of Galois theory
IMC 2001 for the fellowing problem.

(ylfx Let p be a prime and let n1,ng. ..., ni be integers. Define

Proof. The problem looks rather complicated because of the strange inequality
imposed on ¢. Let us forget first about that and consider the product of 27m |
all values of P at the primitive gth roots of unity, [T7Z} P(z). This is an i
=t P
integer, by corollary 9.14. If it is not 0, then Hf,,l |P(2:)] > 1. However, by
assumption there exists a polynomial Q) (nec ily with integer coefficients) Prove that either § =0 or S >k (g) = 3
such that P(X) = (1 - X}*Q(X). Therefore
Holden Lee
1 -1 g1
Proof. Let. > = ¢ . The crucial ingredient in the proof is the following:
L L-z)-
T e Lemma 9.17. The number
1 il i1
2=k
Since N= H Z ( LI z—l-u,)
g1 =1 \s=1
is an integer. Moreover, N =0 if and only if S = 0.
418 Chapter 9. A Little Introduction to Algebraic Number Theory 9.4. The theorem of symmetric polynomials 419

Let us admit this for a moment and see how we can finish the proof. Now, assume that N = 0. Thus, there exists 1 < | < L such that
Assume that § s 0. Then [N| > 1. So F(2 + 27!} = 0. By the lemma, F is a multiple of f and so I< vanishes at
z+ 271, But this means that § = 0, a contradiction. Thus, we have proved
the crucial claim and the result follows. g
The following result is certainly classical, but it is rather difficult to find an
elementary proof in the literature. We follow one of the approaches proposed
and the conclusion follows.
in the beautiful article {10].
Now, let us prove the lemma. As is well-known (and easily proved by
induction) there are polynomials F; € Z{X| of degree j such that X7 + X~/ \J16. Let a1, a,...,a, be positive rational numbers and let ki ko, ...k, be
=1
Fy(X +X71). Let F = }:;t Fy;. Then N = Hle F(z' + 27"). The lemma integers greater than 1. If all ki +a;/k’ +-- +a:./ k2 is a rational number,
will be proved if we prove the following result: then any term of the previous sum is also a rational number.

Lemma 8.18. The minimal polynomial of z + 27" is Proof. 1t is clearly enough to prove the following result: let £ > 1 and suppose
that the positive rational mumbers ai, ..., an,b1,. .., b, satisfy
T =TIx ~ @+t = Foot 4 Foca 4+
i=1
a ¥+tb € Q
Proof. Note that Then ¥b; € Q for all 7.
Let

F0r+x71y = T E =0 =7 A; = {roots
of XF — afb;}= {we;
V5 |1 <<k},
where w is a primitive root of order k of 1. Also, let
=1
=
S= Ea; Vo
=1
and
P(X) = M s-x-z-s
22€A2,..,.
2 €An
Thus f = F = + F 5 s +--- by definition of the polynomials F;. In particular,
[ has integer coefi:( xentb and so N is an integer (by the fundamental theorem of
By theorem 9.14 we have P € Q[X]. Note that P(a; ¢/B;) = 0. Let d be the
symmetric polynomials). Moreover, f has degree l’- and vanishes at 2+ 2™1. least positive divisor of & for which {/ € Q (it exists, as ke Q). 1 we
_ o @Eee -
But manage to prove that d = 1, it will follow that {/5; € Q, so we can delete the
first term of § and conclude by induction on n. Se, let us prove that d = 1.
(2): Q+2"1] 7 By definition, we can write a; ¥/b; = ¢ with = € Q4. The crucial fact is the
so f must be the minimal polynomial of z + 2~1. O following:
420 Chapter 9. A Little Introduction to Algebraic Number Theory 9.5. Ideal theory and local methods 421

Lemma 9.19. X¢ -z is irreducible in Q{X]. 17. Suppose that (@n)n>1 is a linear recurrence sequence of integers such
that n divides a,, for all positive integers n. Prove that (%{1) is also a
Proof. If F is a monic polynomial with rational coefficients of degree between
linear recurrence sequence.
1 and d — 1 that divides X% — z, all roots of F* have absolute value {/z and so
Polya
{F(0)] = (z)¥8E) is a rational number, that is \i/ bfegw) € Q, contradicting
the minimality of d. a Proof. By the general theory of linear recurrence sequences, we can find dis-
tinct nonzero algebraic numbers 21,22, . .., 2, and polynomials fi, fa,.... fin
Since P(¢x) = 0, the previous lemma yields X¢ ~ x| P in Q[X]. Thus,
with algebraic coefficients such that
if 2 is a primitive root of order d of 1, we have P(2{/z) = 0 and so there are
(T2, ., 3) € Ag x Agx
-+ X Ap with § — 29k = mp+ -+, I d > 2, then
an = fi(n)2] + fo(n)2h + -+ fm(n)2p,
Refz) < 1, 50
for all n. We will prove that if n|a, for all n, then £(0) = 0 for all 4, from
Re(S) = § which the result follows easily.
=Re(z{Z+uz+
- +Zn) Let K be the field obtained by adjoining to Q@ all z;’s and all coefficients
of the polynomials f;. Choose a prime p which does not divide any of the
< Re(2¢)
+ Y |zl norms of the (nonzero) coefficients of fi’s or the norms of one of the z;’s. All
sufficiently large primes satisfy this property. Fix such a prime p and consider
a prime ideal I of K over p, with norm N(J) = p/. Impose the condition that
= Re(z¥x) + Z‘aiw
jp? divides ajp,r. Note that
=2
m
<YE+Y el @ =Y fi(0)2] (mod I),
=2 =1
=5
since zf’! = z {mod I} and since p € /. Thus, we must have
a coutradiction. So d = 1 and a; ¥/5; € Q. ]
m
9.5 Ideal theory and local methods S A =0 (med 1)
i=0
We strongly advise the reader not familiar with algebraic number theory for all 7 = 0,1,...,m — 1. Seeing this as a linear system in the f;{0)'s, it
to read the appendices on number fields and p-adic numbers before reading follows that f;{0) € I for all ¢, unless 7 divides the determinant of the matrix
this section, which is short but rather challenging. associated to this system. However, this is a Vandermonde determinant in
We start with a beautiful result of Polya concerning linear recurrence the 2;'s and so, if we ensure that I and H#j(z; — z;) are relatively prime, we
sequences. Recall that a sequence (a,)n lled a linear recurrence sequence will be able to conclude that f;(0) € 1. But to ensure the last property, it is
if one can find d and 2y, ..., 34 such thak tniq + T10nsda1 + -+ + Tgan =0 enough to choose a prime p which does not divide the norm of the algebraic
for all n. number Hl7é ;{2 — 2;)- Again, all sufficiently large primes have this property.
 422 Chapter 9. A Liitle Introduction to Algebraic Number Theory 9.5. Ideal theory and local methods 423

The previous paragraph shows that we can find infinitely many primes p Proof. 1f d; is the degree of the algebraic number al, it is clear that dox > dyesr
and for each such prime an ideal I over p such that fi(0) € I for all 7. But then Q(a?")). Thus there exists an integer j and a positive
(because Q(azk“)
p will divide the norm of £;(0) for infinitely many primes p and so [0y =0
integer d such that doe = d for all k > j. Let a3 = a? | ag,...,ag denote
for all i. Tt is then clear that s js still a linearly recurrence sequence. o the conjugates of o>’ and choose a positive integer ¢ such that fo=cX -
@)+ (X — ag) € Z{X] is primitive. Then go = c(X +a1) <X +ag) € Z[X]
‘We present two approaches for the following challenging problem: a rather
is also primitive, so by an easy application of Gauss’ lemma f; = AX -
exotic elementary one and a more advanced approach which uses standard
a?)--- (X — afi) € Z[X] is also primitive. Since a? has degree d and since
facts about number fields and their p-adic completions.
deg fi = d, it follows that f; is irreducible over Q. Repeating this argument, we
— obtain that fr = ¥ (X —a}')--- (X —aj ) € Z[X] is primitive and irreducible.
¥ 18. Let as,ay,...,a, be complex numbers such that a]* +a* +- - -+ ay is an
Next, since na?’ is an algebraic integer, we have h, = n?(X — daf Yo (X -
integer for all positive integers m. Prove that (X — ay)(X — ag}--- (X -
an) € Z{X). a¥) = (nX —na) -+ (nX —nal) € Z[X], so we must have 5r € Z. Since
this happens for all sufficiently large r, it follows that ¢ = £1 and so ¥ isan
Locad ~
Michael Larsen, AMM E 2993 algebraic integer. As a result, a is an algebraic integer and we're done. a
Gloled P ode Wiks
Proof. Let
Proof. This proof uses rather heavy material, but it is much more conceptual
op = > Qg i+ 7 than the previous one. Namely, we will use a local-global principle, stating
ISdr<ig<<ipSn
that an algebraic number z is an algebraic integer if and only if v(z) > 0 for
and Py = af +af + - + af, so Newton's identities? can be written any valuation v on Q. This follows easily from the relations between a number
(for
1<k<n) field and its completions (see the addendum on number fields), but the result
is not obvious at all. Anyway, once we have this, the lemma is immediate: if
P = 01Pgy + 02Fkg — -+ + (=1)¥kay, = 0.
v is a valuation, then we know that v(n) + kv(a) > 0 for all k. Dividing by
1t follows immediately from these relations that if Py e Zfor all k, then oy €
k and making & — oo yields v(a) > 0, which is enough to ensure that ¢ is an
arZ for all 1 < k <n. In particular, (X — a3 )}{X — ag) - (X ~an) € HZ[X] algebraic integer. )
and so n} - a; are algebraic integers. Observe that if ay,ag,... ay satisfy the
conditions of the problem, then so do af.dh, ... al for all 7 > 1. We deduce O
The lemma is proven, and so we are done.
that nla is an algebraic integer for all ¥. Thé next lemma shows that all
a;’s are algebraic integers, so the coefficients of (X — a1 )(X — ag) - (X —an) case when all a;’s are rational numbers is much easier and
Remark 9.21. The
are algebraic integers. Since these coefficients are tational numbers (this has problem. In this case, the problem reduces easily to the
is a rather folklore
already been established), they must be integers and the result follows.
following: if p is a prime number and if @1, a9....,ax are integers such that
Lemma 9.20. Let n be a positive integer and a be an algebraic number. If P divides o} + a3 + --- + af for all n, then p divides all ;’s. This follows
na® is an algebraic nteger for all positive integers k, then a is an algebraic easily from Euler’s theorem, by choosing n = o(p™) with N sufficiently large.
integer. Actually, using ideal theory as in the previous problem and imitating the proof
for rational numbers, one can give yet another solution of the problem. We
“See the remark 9.22 for a proof of these. leave this as a nice exercise for the reader.
424 Chapter 9. A Little Introduction to Algebraic Number Theory 9.5, Ideal theory and local methods 425

Remark 9.22. Let us recall the proof of Newton’s relations. Let a; be elements Assume that 1o g; is a multiple of p and let z; = 2! (1 < i < ¢). Since
of a field K of characteristic 0 and define
n 2=l
fx) =Tl -ax zmxl Zr,, T ) =ud 1) >0,
et
i=l
Let P = af +af +- ak. Observe that (=
there exists o(i) € {1, -,q} such that vy(a; ¥ — z5:) > 0. Since we
S N ek have 'z:,,(
e gwazx = %ux ke . ia,?) > 0, it follows that v, (3 2,) > 0. So, if f(X) =
T X797 we have v,(f(2)) > 0. Since f(z) is an algebraic integer for all
i, it follows that v,(f(1) [T, f(2:)) > 0. Let N =[], f(z). N is an integer,
Identifying coefficients in the equality
because it is a symmetric polynomial expression with integer coefficients in

£(X) = —5(x)- Y Rx* the roots of the polynomial X97! +--. 4+ X + 1. We claim that N is nonzero.
Otherwise, there exists 2 < i < ¢ such that f(2;) = 0. The irreducibility of
the polynomial 1+ X + .-+ + X9 over the rational numbers implies that f
yields Newton’s relations is a multiple of 1+ X +--- + X¢°L. But then r = f(1) is a multiple of ¢, a
contradiction with the hvpothesls
b 4 Pibey 4 <o Pby = 0 So N is a nonzero integer and v,(rN) > 0 so that v,(N) > 0 (clearly p
does not dmde ). Thus [N] > p. On the other hand, we have |f(z)i<r, so
for L<m<n.
that [N} < 77~1 < p. This contradiction finishes the proof. jul
The following is also a very tricky problem. We use a p-adic approach
to solve it and we refer the reader to the appendices on p-adic numbers and
Proof. Here is a more elementary, but still very tricky sclution, based on
number fields for more details. the theorem of symmetric polynomxa]s Suppose that none of the a;’s is a
¢ 19, Let p, ¢ be prime numbers and let r be a positive integer such that ¢|p—1, multiple of p and let b = ¢ = , where g is a primitive root mod p. We
-
¢ does not divide r and p > 7971, Let a1, az,.. ., a, be integers such that can therefore find positive integers m; such that a,° = h™ (mod p). Let
el ity Pzl
a;® +ag® 4 ---4a® is a multiple of p. Prove that at least one of f=X™ 4+ X™ +...+ X™ and let g € Z[X] satisfy
the a;’s is a multiple of p.
FXa) f(X2) - f(Xgm1) = 901 (X1, X, o, Xg1)s oo 0n(Xn Xa -+~ Xgo1))s
J. Borosh, DA Hensleye J. Zinn, AMM 10748

2in where o; are the symmctnc fundamental sums. Let z;,2,.... zg-1 be the
Proof. Let z=¢« and let K = Q(z). This is an extension of degree g — 1 of
complex roots of X . Note that
Q. By choosing a prime dividing p in the ring of algebraic integers of K and
completing K with respect to this prime, we obtain an extension of the p-adic X7—
r
5 )
valuation v, on K. Moreover, if z is an algebraic integer in K, then vy(z) > 0. (X = h)(X = h%) oo (X =BT € Fy[X],
 3

426 Chapter 9. A Little Introduction to Algebraic Number Theory 9.6. Miscellaneous problems 427

as k%, W3 are distinet qth roots of unity in F,. This implies that and first compute the minimal polynomial of cos Z. In order to do this, we

Gilzn 2y 2e1) = by B RS (mod p) will first find a rational equation of low degree satisfied by cos FoLetz= e7,
so that 27 = —1 and
for all i. Therefore
6 5
g1 B2 42t 24241=0
T 7 = glon(h B, 080, o (R B2 hEY) Dividing this by 2% and rearranging terms yields
i=1
g-1
1 1 1
=[Ts)=0 (modp), 3 2
z +;§—(z +3)+z+—~1 0.
st
that is p divides the integer N = f(21) - -+ f(z4-1). As in the previous
1
Thus, if z =cos § = -232-1, then the previous relation gives
solution we obtain [N] < 74! < p and so N = 0. We conclude as in the
previous solution. o 8%~
6 — (422 —2)+
2z —1=0,

9.6 Miscellaneous problems that is 82° — 42% — 4z + 1 = 0. Since the polynomial

1 21 1
It is really not easy to solve the following problem without the use of
minimal polynomials. However, once the yoga of minimal polynomials is un-
FX) =X~ “x2
3 sX+g
X+ -

derstood, the argument is rather standard.

is trivially irreducible over the rational numbers (it has degree 3, so we only
have to look for rational roots), this is the minimal polynomial of z. Therefore
20. Find the least positive integer n such that cos I cannot be written in
z and = — p have degree 3 over Q.
the form p + /g + 7 with p,q,7 € Q.
But then (observe that the identity ({(\/g+ ¢/7) — /§)® = r easily implies
O. Mushkarov, N. Nikolov, Bulgaria that /g € Q(/7+ V7))
Proof. For n < 6, explicit computations show that cos % can be written in the
[Qva+¥7): Qval 1QVD) : B = [Qve + V1) : Q) =3
desired form (the argument is a bit tricky for n == 5, but note that z = s a
solution of the equation 2% ~ 2% + 22 — z 41 = 0, which can also be written as and since [Q{/4) : Q] is 1 or 2, it follows that /7 is a rational number. Thus
(24212 = (2+271) ~1=0.) The question is whether we can write cos % in z~p— /G = Y and u = p+ /g is arational number. Now, since z is
the form p+,/g-+ /7 with p,¢,7 € Q and the answer turns out to be negative, irrational, we must have 7 irrational and so X® — 7 is irreducible over the
implying that the answer to the problem is n = 7. rational numbers. Since f{u+ ¥r) = f(z) = 0. it follows that X3 — r divides
Let us assume that fu+ X) and so (for degree reasons) we must have f(u + X) = X3 —r. It
is trivial now, by identifying coefficients, to see that this is not possible. The
(:os-;:p+\/§+\3/; result follows. a
428 Chapter 9. A Little Introduction to Algebraic Number Theory 9.6. Miscellaneous problems 429

Proof. As in the previous solution it is enough to show that we cannot have Now, by assumption s,t, is an integer for all 7. But then
m
sz =p e+ Ir. 8ity + st = siti + 85t ~ (8 — 85}t — 1)
is also an integer for all 4, 7. Since sit; + s;t; and (s;t;) - (s;t:) = (sit:)(s5;)
As before we compute that cos § satisfies 82° — 42” — 4z + 1 and that this
are integers, s;f; and s;f; are algebraic integers. Since they are also rational
polynomial is irreducible since it has no rat] ional roots. Also either by noting
numbes hey must be rational integers. Thus s;f; is an integer for all 7, ;.
that the other two roots are cos 5 and cos% or by plugging in a few values,
For i = u, we obtain that all ; are integers. Let d be their greatest common
we see that this polynomial has three real roots.
divisor. Then clearly % is an integer for all . We claim that ds; is also
Now let z = €2™/% and suppose z = p+ Vit 2*/r. Then
an integer for all 4, which will solve the problem. But since d is a linear
combination with integer coefficients of some #;’s (by Bézout’s lemma)
r= (@ pF V) = (@—p)* +3g(e - p) F B — p)* + O3 and
since s;t; € Z for all 4,4, it is clear that ds; € Z for all 5. The conclusion
So follows. =)
[(@ ~p)* + 3q(e —p) — 7] = (3(x ~ p)* + g)%¢
In order to motivate the next problem, we will discuss first a very classical
Thus and nontrivial result in elementary number theory. The reader is advised to
read the addendum 9.A before reading the proof.
9(X) = [(X ~ ) +3¢(X ~p) ~ r* = (3(X —p)? + 9)*¢ € QX
Theorcm 9.23. (Lucas-Lehmer) Define a sequence by ag = 4 and a,4;=
is a sixth degree polynomial with roots p+ /g + 25 /7. Tf the equality above —2 forn > 0. Let m be an odd positive integer and let n=2" — 1. Then
holds, then this polynomial must be a multiple of f(X), the minimal polyno- n s a prime if and only if nlay,_o.
mial of cos § = p+ /G + §/r. However f{X) has three real roots and g(X)
has only two real roots (the ones with k = 0). Thus this cannot occur, ] Proof. The first difficulty is to actually find a manageable formula for the
general term of the sequence. We use the identity 22 + 22 = (z+271)" — 2
We continue with a very beautiful problem and a very elegaut solution. and set a, = zn +z;tfora sequenoe Zp > 1 (note that a, > 2, so z, exists).
Then Zp41 =22, 50 2y = 3’ and we easily conclude that
21. Let sy,8g,... and #1,%2,... be two infinite nonconstant sequences of ra-
tional numbers such that (s; — s;)(; — ¢;) Is an integer for all 4,7 > 1. =2+v3¥
+2- V37
Prove that there exists a rational number r such that (s;—s ) and &T,L
are integers for all 4, j. Suppose that n = p is a prime and m > 3. Since p = 1 (mod 3) and
P = —1 (mod 8), the quadratic reciprocity law implies that (%) = 1 and
USAMO 2009
(é) = —1. Pick some a in an algebraic closure® of F,, such that o® = 3.
Proof. We start with some useful reductions: first of all, by working with the
sequences (s; — s1); and (£; —#1);, we may assume that 81 = #; = 0. Secondly, "%One does not, need the existence of an algebraic closure to prove the existence of ot if 3
is & quadratic residue mod p, it is cleAr what we have to do; otherwise, it is easy to check
there is w such that s, # 0 and, by working with the sequences (';fl) and
“/n that Fy[X]/(X? - 3) is a field with p? elements and we can take for a the image of X in this
(8w * tn)n, we may assume that s, = 1. field.
 430 Chapter 9. A Little Introduction to Algebraic Numb
er Theory 9.6, Miscellaneous problems
431

Note that « is actually an element of Fy2 and that we can define a map
2V Ifz e F}, everything is easy,
= Fp by fla+ bv/E) = a+ ber, where & = a since then Lagrange’s theorem for this
(mod p) (seen as an subgroup yields 2°*2?|p—1 and so trivially 2#3/p21. So,
element of Fy2). Since o? = 3, it is immediate assume that z is not
to check
that this is a ring in F. Then z, y are roots of the irreducible polynomial
homomorphism. Trivially, [ vanishes on X2 —4X +1¢ Fp[X],
pZ. Let z = f2+V3) =2+a so that we must have 27 = y.
and y = f(2 - v3) = 2~ a. Thus z,y Indeed, since 2? — 4z + 1 = 0, we also have
€ Fy2 and they are nonzero, since (by raising the previous equelity to the p-th power and
zy = f(1) = 1. We want to prove that flam— by using the formula
2) = 0 or equivalently that (& +y)? = 2P + ¥, valid in fields of characteristic p)a®
22" gt 0, ie. 2" = 21 4o 41 = 0, so
Since 2z = (1+ @)?, we obtain the that z¥ is also a root of X2~ 4X + 1, which cannot be z (because otherwise
following equality in Fp:
#=gandz e F,). Thus 2¥ = y and so 27*! = 1. But then 2°+2 which
is
the order of x. must divide p + 1 and we are done
again.
2" - = (;)2 = 1
= 2% = (1P = (14 a1 +oP).
o
There is really no obvious approach to the following rather
exotic problem.
Since a? = 3, we have aP = (g) + & = -, 23. Let k be a positive integer and let aj.aq,....a;
which combined with the previous and bi.ba,. . by be
equality yields the desired result, two sequences of rational numbers with the property:
for any irrational
numbers #1,22,...,7z > 1 there exist positive
Let us prove the converse now. Suppose that integers ny, no,...7
nja,_2, we need that 1 is a and my,ma, ..., g such that
prime. It is enough to check that for all pln we have p
> /. Since p divides
@m-2, the previous arguments yield the equali a7 + aolzp?l 4+ aplel] = bl + bfe2] 4
ty (24 a)%l = ~1in IF;Q. Thus 1 Bl ]
2+ac¢ IF;g has order n + 1 and Lagrange’s theor
em yields n + Up? — 1. The Prove that a; = b; for all 4.
result follows.
o Gabriel Dospinescu, Mathlinks Contest
22. The sequence ag, a;, ay, ... is defined by Proof. The key point is the following result:
ag = 2 and gy = Qai — 1 for
k 2 0. Prove that if an odd prime p divides
a,, then 273 divides 1. Lemma 8.24. For any integer N > 2 we can find irrational numbers
a,b > 1
IMO Shortlist 2003 such that for every positive integerm we have [a™] = ~1 (mod
N) and pN =
0 {mod N).
Proof. Note that 2a, is precisely the sequence studie
d in theorem 9.23, so
Proof. We will choose a,b to be algebraic integers of degree
2. Let us show
LV how to construct a and leave to the reader the details for the
= e - va construction of
& We want to find a polynomial with integer coefficients

Letnowp > 2bea prime factor of a,, and

let o € E be such that o2 = 3. AX)=X*
+v=(+u
X -a)(X
X~ ¢
Define f,,y as in the proof of the previo
us theorem. Since Plan, we have for some irrational numbers a > 1,0 < ¢ < 1. In this
7 42" = 0, thus 22" = 1. Hence 2 has order case, since a™ + " is an
22 in the group F7, integer for all positive integers m, it follows that fa™
= a™ + ¢™ 1 for all
and so by Lagrange’s theorem 272 divides p? — 1. m. Thus, we need to ensure that ¢™ + ¢™ = 0 (mod N
Unfortunately, this is not for all m. Since
enoug h, but we are close.
@™ oy +¢7) —ofa™ 4 Yy
432 Chapter 9. A Little Introduction to Algebraic Number Theory 9.7. Notes 433

for all m, it is enough to ensure that N divides u,v. Also, to ensure that Let us prove the base case: assume that n = 1 and that
0 < e <1 we will choose v > 0 and 14+ u+ v < 0. For instance, we can take
u=—2N,v= N, yielding a = N + VN? - N, V@i
Gm = a+by/pr
Similarly for b, we will choose u = —(2N + 1) and v = N, so0
for some rational numbers a,b. Squaring this relation and using that V1 is
AN +1+VANT+1 irrational, we deduce that ab = 0. But then either 1¢2 -+ gm 0F 1G2 -+ - gmp1
b= — [}
2 is a perfect square, which is clearly not possible. Now, assume that the result
holds for n and let us prove it for 7 + 1. Let F = Q(\/B1, /P,
-- -, /Pa) and
Coming back to the proof, choose a positive integer N and a, b irrational assume that \/giG2 -~ gm = a + by/Prr1 for some a,b € F. Again, we square
numbers as in the lemma. Set z; = a and zy = -+ = z; = b. By hypothesis,
this relation to deduce that
we can find positive integers n1, ng, ..., g and my, ma, ..., my such that
2aby/Pril = Q2 Gm ~ &% — pps1b® € F.
a o]+ aglay?] 4 - 4 apfep] = bifel?] + bolz5] 4 - -+ bplaF]
However, by the inductive hypothesis we have VPrs1 € F, so we must have
By the properties of a and b we deduce that gy = b; (mod N). Since N was
ab = 0. If @ = 0, we obtain that \/F,11iGs Gm € F, contradicting the
arbitrary, it follows that a¢; = b;. Since we can do the same with the other inductive hypothesis. If & = 0, we get a similar contradiction. In all cases, the
pairs (a;, b;), the result follows. [m} inductive step is proved and the conclusion follows. [m]
The following result is really a mathematical gem, taken from [5]. It is Remark 9.25. In [58], Mordell proved the following generalization:
quite difficult and has a very elementary proof.
Theorem 9.26. Let K C L be fields of characteristic 0 and let z1,%a,...,2, be
24. Prove that if py, po, . .., py are distinct primes and if elements of L such that for alli there exists o least positive integer n; such that
z[* € K. Suppose that for all integers e1,es,..., >R e K,
a1v/P1+ azyPr+ -+ an/Pn = 0 then n; divides e; for all i. Finally, suppose that L C R or that K contains
all nith roots of unity, for all i. Then (z‘i‘ . :v;’ ----- fl')ogi, <n; 35 a linearly
for some rational numbers ay,as, ..., an, then g; = 0 for all i independent set. In particular, [K(z1,22,...,2,) : Kl =n1----- .
Besicovitch’s theorem
9.7 Notes
Proof. We will prove by induction on n the following statement: for any m > 1
and any distinct primes g1, g2, .. ., G, P12 P2, -, Pn. We have? ‘We thank the following people for providing solutions: Amol Aggarwal
(problem 18), Darij Grinberg (problem 1), Daniel Harrer (problem 9), Holden
Vo gm ¢ QVPL VP2, -5 VPr)- Lee (problems 8. 10), Thanasin Nampaisarn (problem 13), Fedja Nazarov
s an extension of fields and if @y, @2, #n € K, we let Fle1, a2, ., 5) be the (problem 3), Richard Stong (problems 4, 6, 20), Qiaochu Yuan (problems 4,
smallest subfield of K which contains ¥ and z1,%2,...,%n. It is also the set of elements of 17), Victor Wang (problems 9, 19), Gjergji Zaimi (problems 2, 3).
the form f(21,22,...,n), where f is a rational function in 7 variables with coefficients in
.
 f

9.A. Equations over Finite Fields 435

Addendum 9.A Equations over Proof. First, let us check that F, is a field. It is clearly stable by multiplication
and stability under addition follows from the previous proposition. Fy has ¢
Finite Fields
elements since X7 — X splits into linear factors over F, (because F, is alge-
This addendum is a modest introduction to finite fields and polynomial braically closed) and all of these linear factors are distinct (because X7 — X
equations over finite fields. There are some very beautiful and extremely deep is prime to its derivative —1).
results on the subject, which are far beyond the scope of this book. But the Let us consider now a subfield L of F—P with g elements. As L* is a group
fact that their proofs are very difficult should not be a reason for not presenting with ¢ — 1 elements, Lagrange’s theorem yields 2971 = 1 for all z € L*. Thus
them. We highly recommend the introductory text [43] for more deta z? = zforall z € L and so L C F,. A cardinality argument finishes the
To avoid spending too much time on preliminaries, we will fix a prime proof. o
number p and an algebraic closure Fy, of the field F, = Z/pZ. Recall that
this means that any z € F, is a root of some nonzero polynomial f € F, p1X] A more subtle result is the following generalization of Gauss’ classical
and that any f € F,[X] has at least one root in Fj (which actually implies theorem on primitive roots modulo prime numbers.
that it splits into linear factors over Fp). It is a rather nontrivial theorem of
Theorem 9.A.3. F is a cyclic group of order q— 1. More generally, if K is
Steinitz that any field has an algebraic closure and any two algebraic closures
any field and G is a finite subgroup of K*, then G is cyclic.
are isomorphic. We take this approach when introducing finite fields si it
is pretty rapid, though not very elegant. ..
Proof. Let d be the maximal order of the elements of G. It is a general property
Before proving the first fundamental result, let us glorify the following
of finite abelian groups that if «, € G have orders m, n, then one can find
easy result, which will be constantly used in this chapter:
2 € G with order lem(rm, n) (the reader can take this as an casy exercise).
Proposition 9.A.1. Let p be a prime and let A be a ring such thet® pa =0 Using this, we deduce that the order of any element of G divides d. Thus for
for alla € A. Then for all powers q of p and for all ay, az,...,a, € A we have all g € G we have g* = 1. But the polynomial X¢ — 1 € K[X] vanishes at all
elements of G, so d > |G|. On the other hand, d is the order of some element
(a1 +agt - +an)¥=af
+af+ . +al. of G, so d}|G| by Lagrange’s theorem. Therefore d = |G| and G is cyclic. O
Proof. By induction on n, we may assume that n = 2. Then everything follows There is 2 trap concerning finite fields: it is not true that if n > m, then
from the usual binomial formula, the hypothesis on A and the fact that (j’) =0 Fpm C Fpr. Actually, this inclusion takes place if and only if X*™~1—1 divides
{mod p) forany 1 <i<gq. [} XP"~1 1 (this follows immediately from the definition and the fact that the
If ¢ is & power of p, let roots of X¢ — X are simple} and this happens if and only if p™ — 1 divides
" ~ 1, which in turns happens if and only if m divides n.
F, = {z € Fyla x}. A fundamental object in the theory of finite fields is the Frobenius map

‘We have the following easy, but crucial result: Frg:Fgn = Fyn, Frg(z) =29,
Theorem 9.A.2. Fy is the unique field with q elements contained in F,.
an automorphism of F» which acts as identity on Fy. Moreover, any such au-
tomorphism is an iterate of the Frobenius map and there are precisely n such
436 Chapter 9. A Little Introduction to Algebraic Number Theory 9.A. Equations over Finite Fields 437

automorphisms.S All these results would be pretty hard to prove without the-
orem 9.A.3, but they become easy exercises once we have it. The following
result is fundamental. It says that if you know a root of an irreducible poly-
nomial over Fy, then the other roots are obtained by successively applying the
Frobenius map to that root.
The following result summarizes the basic properties of these maps, that will
be used in future sections.
Proposition 9.A.5. The norm and trace maps are surjective maps from Fyn
to Fy. The norm map is multiplicative and the irace map is additive.

Theorem 9.A.4. Let f € Fo[X] be a monic irreducible polynomial of degree Proof. To avoid complicated notations, write N and T for the norm, respec-
n and let ¢ € ]F‘Ap be a root of f. Then the roots of f are x,:cq,xqz, szt tively trace map. First, let us check that N(z),7(x) € F, for all z € Fyn.
In other words, £(X) = [Tg (X ~ 7). It is enough to see that N(z)7 = N(z) and T(z)? = T(z). For N(z), this is
clear since z¢" = z, while for T{(z), this follows from proposition 9.A.1 and
Proof. The key point is that z¢" = z. Iudeed, the field generated by z over
the equality 27" = z. It is clear that NV is multiplicative and proposition 9.A.1
¥, (inside F,) has ¢" elements, because = has degree n over Fy, so this field is
shows that T is additive. It remains to prove the surjectivity of these maps.
Fgn. But in Fye all elements are roots of X¢" - X. Having done this, define
Let £ be a generator of F; and let u be a generator of Fy.. There exists
the polynomial G(X) = [T (X — 27). The key point and proposition 9.A.1
a € Z such that £ = u® As €971 = 1, we have ©*9"1 = 1 and so there is
yield
n—1 n-1 an integer b such that e = b - ’}:Tl. But then £ = N(u®) and the surjectivity
G(x)7 = [T (x? = 2") = T (x% - 2¢') = q(x9). of N follows. For the trace map, this argument does not work, however we
=0 =0 note that T(az) = eT'(z) for any a € F, and any 2 € Fyn. Thus, it is enough
to prove that there exists z such that T(z) # 0. But this is clear, as the
Thus, if we write G(X) = go+ g1X + - + g X', then again by proposition
polynomial X + X9+ ---4+ X7 has at most ¢"~! roots and so it cannot
9.A1
vanish on all of Fgn. ju}
RHGEXTd g X = gk g X+ g X,
which implies that ¢f = g; for all and s0 g; € F;. Thus G ¢ Fy[X]. Since 9.A.2 Characters of finite fields
G vanishes at o and f is irreducible, we deduce that f divides G. A degree
argument finishes the proof. [m] As Fypn is an n-dimensional vector space over ¥, the choice of a basis
yields a group isomorphism Fyn =~ Fy, X -+ X F,. Now, basic properties of the
9.A.1 Norm and trace maps dual of a group discussed in section 7.A.1 yield the following result.

“onsider a finite field Fy and a finite extension Fgn. Define the norm and Proposition 9.A.6. Let g be a power of p. There is an isomorphism of groups
trace maps by a— 1, between Fy and its character group, where

n-1 n-1 = T Trmles)

Negsp,
:Fgn = Fgn,
N
zo0 ”z“
7
s Trpgsr,
Fon = Fgn,
w
zer 2
i
z? . Yal
=0 =0
Also, Fy is a cyclic group, so its group of characters is also cyclic of order
®In fancy terms, the Galois group of the extension Fou /F, is cyclic of order n and gener- ¢—1. The following result will play an important role in the following sections,
ated by Fry. when we will compute the zeta function of a diagenal hypersurface.
 438 Chapter 9. A Little Introduction to Algebraic Number Theory 9.A. Equations over Finite Fields 439

Proposition 9.A.7. Let d be a divisor of g — 1. The Map X =+ Xn, where Finally, the following result will be used in the proof of the Davenport-
Xnlz) = X(Np, /(%)) induces a bijection between characters of order d of
F, Hasse relation, to which the next section is devoted. It is by no means specific
and characters of order d of Fon
to finite fields, but the short proof we are going to give uses properties of finite
fields developed in the previous sections.
Proof. 'The fact that x, is a character of order dividing d is a consequence of
the multiplicativity of the norm map. The fact that it has order exactly d and Proposition 9.A.9. Let x € Fgn and let
that X — xn Is injective is a consequence of the surjectivity of the norm map
(proposition 9.A.5). It remains to check the surjectivity
F=X" X4 4 4 (~1)4ag € Fy[X]
of x -+ xn. Let ¥ be
a character of order d of o and let « be a generator of be its minimal polynomial over Fy. Then din and [] i In
¥, Then & =« k=y
is a generator of Fy and since §(u)7"! = 1 (because ¥ = 1 and dlg—1), there particular, Ng /s, (2) = af and Trs.z (2) = Sa1.
is a unique character y of F; such that x(€) = %(x). By construction, X = x» Proof. Since [Fy(x) : Fy] = deg(f) = d, we have F,(z) = [ (this uses theorem
and the result follows. ] 9.A.2). But then F o« C Fyn and, as we have already remarked, this implies
Just as for Dirichlet characters, it is convenient to extend the definition of that d|n. Next, for degree reasons it is enough to prove that ¢ = ;';(}(X —z#)
a multiplicative character y of ¥ to Fy, by defining x(0) = 0 if y is nontrivial has only one irreducible monic factor, namely f. But if % is such a factor, then
and x(0) = 1 if x is trivial. The following innocent-looking identity will play h has some root % . But proposition 9.A.4 implies that f also vanishes at 2%
a crucial role in future arguments and is constantly used when dealing with
(note that the cited proposition applies only for j < d, but we have z%° = z
equations over finite fields: anyway, since we have s en that Fo(z) = Fua). Thus ged(f, h) is nonconstant
and by irreducibility f h. This finishes the proof. a
Proposition 9.A.8. Let d be a divisor of g—1 and let x € Fy. The num-
ber of solutions of the equation y* = z with y € Fy, denoted N(y* = z)
9.A.3 Gauss and Jacobi sums, the Davenport-Hasse relation
is Exd:I X{(x), the sum being taken over all multiplicative characters whose
order divides d. Gauss and Jacobi surns play a fundamental role in the theory of equations
over finite fields and in number theory, in general. We give here their basic
Proof. If & = 0, this is clear, as both sides are equal to 1. Assume that properties, that we will need in the following sections. But before doing that,
z 0.
1f the equation ¢ = z has a solution in ¥y, then it has exactly d such solutions, it is convenient to define them. ..
as the equation y% = 1 has precisely d solutions in F; (because dlg - 1 and
¥,
is cyclic of order ¢ ~ 1). On the other hand, the dual group of Fy is also cyclic Definition 8.A.10. 1) If ¥ and y are characters of F,, respectively F.
the associated Gauss sum is
of order ¢ — 1, so the equation x? = 1 has d solutions and for each of them
x(®) = x(y*) = x()! = 1, so both sides of the equality we want to prove a9 = 3 x@(a)-
are equal to d and we are done. Finally, if the equation has no solution, the z€Fy
result is a consequence of the orthogonality relations (theovem 7.A.5) for the
abelian group Fg/ {a¥z 2) If x; and x2 are characters of F;, the associated Jacobi sum is
F,}, whose dual group is precisely the subgroup of
those multiplicative characters y such that x¢ =1 (actually, this argument Joxe) = Y. x@xa)
also covers the previous case. . . ). [ z.y€Fg
o ty=1
440 Chapter 9. A Little Introduction to Algebraic Number Theory 9.A. Equations over Finite Fields 441

Theorem 9.A.11. If x and 9 are nontrivial, then [g(x,¥)} = /4. where o "

Proof. The orthogonality relations (theorem 7.A.5) yield (using also the sub-
I'(z) = / e dr, Bla,y) = / 1 lar,
0 0
stitution § = ¢) the integrals being convergent for Re(z), Re(y) > 0.
Theorem 9.A.13. Jf x1,x2 are nonirivial characters of Fy such that x1 - x2
g = Y x(@/ywE
-y = > xEw-1) is nontrivial, then for all nontrivial characters ¥ of Fy we have
wyeFy tyeF;

- 90a,9)
- gxe,¥)
=3 x (Z Yyt~ 1) - 1) =3 x®(g lem - 1) T == e
teFy yEFy tefy
Proof. This is a rather tricky computation:
=g-1-3 x{t)=g- > xlt)=q
10,1 teF; Toox)gbaxad) = Y 3 @@l - 2)xe)v(y).
2€F,~{0.1} yeF}
Using the substitution @ = zy and b = y(1 — z), this becomes
Corollary 9.A.12. If x and ¢ are nontrivial, then
Y xal@e®ua+b) =900, 9k 9) - 3 xal@xa(-a).
9069 o) = x(~1)q. a.beFyatb£0 acFy
Proof. This is just a long string of obvious computations, using the previous As x1x2 is nontrivial, the orthogonality relations (theorem 7.A.5) yield the
theorem and the fact that g(x, ¥(~)) = x(~1)g(x,%) (which is immediate by desired result. )
definition and the fact that = — ~z is a permutation of F}). More precisely,
we have Here is a striking application. Assume that p=1 {mod 4} is a prime. As
¥}, is cyclic of order p — 1, there exists a nontrivial character x; of order 4 of
F;. Let xa(z) = (%) be Legendre’s symbol. The previous two theorems imply

= x(-Dalx, that |J{(x1,x2)|* = p. On the other hand, it is clear that X1 takes only the
values 0, £1, 44, thus J(x1,x2} € Zi]. In particular, [J{x1, x2){? is the sum of
the squares of two integers. We recovered thus Fermat’s celebrated theorem
o
that any prime of the form 4k + 1 is the sum of the squares of two integers.
One has the following beautiful result which connects Gauss and Jacobi ‘We end this section with a much deeper result, the famous Davenport-
sums. Note the striking similarity with Euler’s famous formula Hasse relation. This is a quite strong identity between Gauss sums, which is
crucial for the proof of Weil’s theorem 9.A.24 that we will see a bit later on.
It also has relations to the Langlands program, but that is really beyond the
scope of this book. The proof is very ingenious.
442 Chapter 9. A Little Introduction to Algebraic Number Theory 9.A. Equations over Finite Fields 443

Theorem 9.A.14. (Davenport-Hasse) Let x and v be nontrivial characters differentiating, we obtain

of Iy, respectively Fy. Then —gn(x,%) = (~g(x, ¥))", where
TL %7))
IUT
=TȎ%logL(T)
9068 = D x(Ne ./, (@) 6(Tre e, ().
#E¥ n
=3 3" deg(PyA(PyrT
s
P on>1
Proof. As Ng,,. /v, (2) and Try,, sz, (x) only depend on the minimal polynomial
of # and not on its roots, we will partition
elements over Iy. Define
Fgn into collections of conjugate
(5, oo
n d=deg(P)|n
MEE by X g (1)) = x(ba)w(h) Combining this with the key relation of the previous paragraph, we conclude
that
for any b; € Fy. It is easy to check that A(fg) = M) - Ag) for all menic
()
polynomials f,g € Fo[X]. Combining this with proposition 9.A.9 shows that 763 =
— Xn:g n 06(X 9) )T
if 2 € Fyn has minimal polynomial P = X% ~ ag X414 .- 4 (~1)%ay over Fy,
then Finally, we will show that (7)) has a very simple expression. Note that

X (N2, (@) © (Trey (@) = a9 (G ) = A(PY4. LD =1+ { 3 Ml

Summing over all conjugates of - and then over collections of conjugate ele- n>1 \deg f=n
ments in Fyn vields the crucial identity
On the other hand,

gl = 3 AP, D0 M =D MX —a) =3 xa@la) = g(x. v,

dmdeg Pln deg f=1
the sum being taken over all irreducible monic polynomials P € F [ X] whose while for n > 2
degree divides n.
To exploit this relation, consider the L-function 3 A= Y xedwte) =2 Y vlar) - 3 xlen) =0,
deg f=n a1vtn
. 1 - e
L) = Ip] 1= A(P)Tes? ;A(f)T‘i 8 by the orthogonality relations (theorem 7.A.5). We deduce that

Here the product is taken over all P € F,[X] monic irreducible, while the L(T) =1+ g(x.9)T
sum is over all f € Fy{X] monic. The second equality follows from multiplica- and the result follows immediately from this and the last equality of the pre-
tivity of A and the unique factorization theorem in Fy[X]. Taking log and vious paragraph. O
444 Chapter 9. A Little Introduction, to Algebraic Number Theory 9.A. Eguations over Finite Fields 445

We end this section by stating a very deep result of Dwork, a consequence the sum being taken over all nontrivial characters x; of Fy such that x* = 1
of his proof of the rationality of the zeta function of algebraic varieties. It is and xoxi-+ x1 = 1.
a vast and very difficult generalization of the Davenport-Hasse relation.
Proof. Let M be the number of solutions (g, z1, ..., 2) € II",‘,‘”l of the equation
Theorem 9.A.15. Let f,g € Fo[X] and let x, v be multiplicative, respectively apzg + -+ + ez = 0. Since we work with a projective variety, we have
additive characters of ¥y, If [ X (Fg)} = 251, s0 it remains to find M. Note that
Su= 30 X (NegmF@)) - (Trey
m otz M= E lagzg
4 tazp=0
#EFyn
.. 51 EFy
then there exist polynomials P,Q € C[T)| such that P(0) = Q(0) =1 and
= Z Lausp Z Lepaug = ooo- Lapoy
g, €Fg L5 Tt

= 3 NG =uo) - NG = w),
2u=0
where we wrote for simplicity a-u = 0 for agug + ayuy +--- + aju; = 0. Using
9.A.4 Diagonal equations and a theorem of Weil
proposition 9.A.8, then expanding the product and Te-arranging terms, we
Using almost everything we have done so far, we can prove the following obtain
beautiful theorem of Weil. The true beauty of the result will be revealed in
i
a next section, when we will use this result and the Davenport-Hasse relation
to compute the zeta function of a diagonal hypersurface. Before stating the M= I X ww)|= 3 ( > xo(uu)X1(u1)-~xn(w))~
theorem, we need some notation. Let ag,ay,...,4; € F, and let Xo, X1, -, Xt @u=07=0 \x7'=1 XP=1¥0<5<I \au=0
be multiplicative characters of F;. Consider the additive character y(z) = Next, note that
5T ) g let 906) = g(xi,). Finally, define
i

Wolxo: X1 -->X1) = 9(x0) .i9(x1)e2 90x0) . >

au=0
xoluo)xalu) - xelw) = TTxta)
=0
dotxon - ),
oo X) xolao) xi(a1) xilar)
Note that this quantity depends on the a;'s, but we suppress the dependence where
from the notation, as we will consider the a;’s as fixed elements, while the
Jolxo, X1.-- o x1) = b
characters x; will vary. We are now ready to state and prove: Xo(uo)x1(w1) - - xa(w).
oty
+- =0
Theorem 9.A.16. (Weil) Let ag, a1,...,a; € Fy and let X be the projective
variety defined by agaf’ + a1af* + - + az]* = 0. Then So, we end up with the pretty complicated formula
-1 i

XEN =@ +5 S Wyloxw - M= 3
XP=-=x=1 j=0
T G .x)
J=0 XOX
Lo Xt
 446 Chapter 9. A Little Introduction to Algebraic Number Theory 9.A. Equations over Finite Fields
447

1t is convenient to study in more detail these sums JolX0, X151 X2)s Lemma 9.A.18. If x0, X1, X2, -- ., X1 are nontrivial characters and
which are generalizations of the Jacobi sums discussed in section 9.A.3.
The
following lemma deals with those terms in the sum defining M for which some X0 X1 x#1,
character is trivial. then Jo(Xos
X1s-- - x1) = 0. If xo- X1+ Xt =1, then

Lemma 8.A.17. Suppose that the trivial character appears —1

X0: X155 X1 Then either x; =1
in the list Jolxo: X1, x0) = ng(Xo)g(Xl) - g0a)
for all§, in which case
Proof. Let Jo = Jo(x0,
X1, .-+ xt) and
Jolxo, X1se oo x) = ¢, or Jolxo x1,- -, x1) = 0.
S=A0mxe- o) = Y xale)xe(u)
- xaw).
Proof. If all x; == 1, it is clear that Jo(xes X10---»x0) = ¢ (don't forget that wytup
o tu=1
the trivial character evaluated at 0 yields 1 by convention). Assume that not Then
all x; are trivial, say (without loss of generality) yp = xq = = x& = 1and
x; # 1 for j > k. Then h=3 Y xelxl@)-xlz)
LEFy o1 bap=—t
Jo(xo, X1, x1) =3 xot)(x1 - x) (=)
=3 Xenlween)
- xelu) teF;
gty by =
=" 3 X lugaa)
- xalur) =0a-x)(=1) (Z X0X1'"Xt(t)) Ji
U1ty teFy
If xox1---x¢ # 1, we are done since Etel’; Xoxi---xf{t) = 0 in this case
(orthogonality relations). If xox1---x; = 1, the previous equality becomes
Jo = xo(=1){g — 1)1r. Since g(x0)9(x5") = xa(~1)g (by corollary 9.A.12), it
remains to prove that

the last equality being a consequence of the orthogonality relations (theorem 7 | == 90a)g0e)
AR gl
XD
7.A.5). [} gbaxz-x1)
Now, by definition we have
Using this, we obtain
90a)gla) - aa) = 3 xilexa(@) - xelanyilas + 25+ - +27)
' TT2oenT
M=d+ 3s TTxta)
-
Dol ..ox)-
Xt =Lixil j=0
= Jolxt x2: -, x0) + (Z'."f(t)hm»--)a(t) i
The crucial step is the following lemma, whose proof uses a generalization of =0
theorem 9.A.13. =Jo(x1, X2+, x0) + 9(xaxa - xa)i
448 Chapter 9. A Little Introduction to Algebraic Number Theory 9.A. Equations over Finite Fields 449

As xix2+--xi # 1, we have Jo(x1,x2:.-.,x1) = O by the first part of the

is exp (Enz 1 %T"), as this time two solutions that differ by a nonzero
proposition and the conclusion follows, (]
element of Fyn are the same element of the projective space.
Finally, using the previous lemma, we obtain Remark 9.A.20. Suppose that

XE =+ 25 4ot —ud = — ]
1

Mg+12 T Tt ™),
Xt Xt G0
for all n > 1 and some complex numbers z;,u; (as we will see, this always
the sum being taken over all nontrivial characters y; such that X" =1 and happens, but this is very difficult to prove). Then
X0+ xi = 1. The result follows. [m}

Zx(T) = (- Tu)(1 = Tug)

- (1 - Tue)
9.A.5 The zeta function of an algebraic variety A= Te)i —To)(1-T2)’
Tn essence, an affine variety over a field k is the locus in k¥ of a bunch essentially by definition of Zx(T') and by the equality of formal series
of polynomial equations in N variables with coefficients in k. A projective
variety is the locus in the projective space P*(k) of a bunch of homogeneous S 5T = —log(1 - al).
polynomial equations in n + 1 variables and coefficients in k. If X is an n>1 n
algebraic variety over Fy, it is natural to consider the number of points of X
over the various finite extensions Fg. The zeta function of the variety X is For instance, if X = P”, the projective space, then
(up to a convenient normalization) the generating function of the sequence
obtained in this way, i.e. X = +g g 41,
Zx (T} = exp Z ‘)S(l-:f-)—‘T" -
1
nz1 Zx(T) = I l P
J_:‘JI—qJT
Clearly, Zx(T) is a formal series in T with rational coefficients.
Another trivial example is the variety defined by the equation ryz = 1 in
Ezample 9.A.19. Consider finitely many polynomials fi. fo,..., f in k vari- three-dimensional affine space. Then clearly | X(F)| = (¢* — 1)?, so that
ables with coefficients in ¥,, and let @, be the number of solutions in Fj. of
the system of equations (4 —qgTy?
Zx(T) =
flzn
o me) = foler, o) =0 = filer,.
. ap) =00
1-n(1-£1)
Let Z[[T]] be the set of formal series in T with integer coefficients. Note
The zeta function of the affine variety defined by the polynomials fy,..., fo is
that in all previous examples we have Zx(T) € Z[[T]. It turns out that
exp (anl “ALn‘T"). On the other hand, if fy, ..., f; are homogeneous polyno-
this is always the case. The following result discusses more generally when
mials , the zeta function of the projective variety defined by these polynomials exp (32, &T") e Z{[T1).
n
450 Chapter 9. A Little Introduction to Algebraic Number Theory 9.A. Equations over Finite Fields 451

Proposition 9.A.21. Let a,, be a sequence of integers. There exist unique Considering the coefficient of 7" in the left-hand side of the previous equality,
sequences of rational numbers by, and ¢, such that we obtain that ¢, € Z. Using also the explicit formula of the ¢,’s in terms of
the a,’s yields the last statement of the proposition and finishes the proof. OO
exp ( anTn) — 1+anTn H — e, Let X be an algebraic variety over F,, say defined by some polynomials
nzl nz1 nzl fi, 2, fa in n variables with coefficients in F,. If

Moreover, all by, are integers if and only if all c,, are integers, if and only if n T={21,...,2a)
€ X(Fgm),
divides 3y, 14 (5) aq for all n (where yu is Mébius’ function).
define
Proof. The existence and uniqueness of by, is clear. As for ¢, the key point is Fr(s) = (o, 2., 02).
to consider It is easy to see that this is again an element of X| (Fgn). Let f be the
smallest positive integer such that z € X (Fyr) and associate to z the cy-
logH (1 =Ty = ZC" Jog(1 ~ cle (z,Fr{(z),...,Fr/"1(z)), of length f. Then X (Fy=) is the disjoint union
n>1
of the cycles of length dividing m, so if a, is the number of cycles of length
n, then [ X (Fgr)] = 304, d - as. Combining this and the previous proposition
yields
Zx(T) = [Ja— 17y e 21y
n1
Remark 9.A.22. The previous proposition is powerful in other contexts, too.
For instance, it immediately yields the equality of formal series
Thus we need to find the sequence ¢, for which 3, din deg = —ay, for all n. But
the M&bius inversion formula yields the explieit form exp(X) = [[(1 - x7)52,
a>1

where 4 is M6bius® function. The proposition also easily implies the following
equality

showing the existence and uniqueness of cp.

exp (Z);: ) = H (1—-X™~ “(")
It is clear that if e, € Z, then b, € Z for all n. The converse is proved by
n>1
induction. Since ¢; = —by, we have ¢; €Z. Suppose that ¢;,...,¢p.; € Z and ged(n,pj=1
observe that f =[], (1 T")% is invertible in Z{[T1]}, as its constant term is
showing that the Artin-Hasse exponential exp (Z">o
)
r has coefficients in
1. Thus
Zy. This is absolutely not clear from the definition and plays a major role in P
nyea (] .- Ty
TAFYOE
L 1 Y, b adic analysis and also in Dwork’s proof of the rationality of the zeta functions
(L—T7)(1 7 € Z(7]}. attached to algebraic varieties over finite fields.
452 Chapter 8. A Little Introduction to Algebraic Number Theory 9.A. Eguations over Finite Fields 453

In general, it is a very deep problem to compute the zeta function of a This last estimate was obtained by Lang and Weil in 1954 for arbitrary va-
given variety. Yet even without computing the zeta function, one can say a rieties, before the proof of the previous deep theorem. For an even more
great deal of things about it! This was conjectured by Weil in the wonderful concrete example, consider integers a,b and a prime p > 3. The condition
paper {84] and proved after a gigantic work by Deligne and Grothendieck. The that the curve y2 = f(z) be non-singular is that p does not divide 4a® + 2752,
following theorem is réally one of the most difficult and beautiful results of In this case the curve y? = f(z) has genus 1 and one point at infinity. Hence
modern mathematics:
Theorem 8.A.23. (Deligne-Grothendieck) Let X be a non-singular projective
[X(Fp)l =1+ |{(z.y) €Fp x Fyly® = 2* + ax + b}
variety of dimension n over Fy. Then the z¢ta function of X is a rational and the bound above becomes
function, More precisely, there are polynomials Py, Py, ...,
P € LT} such
that [{(z.9) € By x Fply® = 2® + ax+b} — p| < 2/
_ PUT)PY(T)
Zx () = B Py - PPl(T) This reproves a famous theorem of Hasse (there are however easier proofs, but
they reguire a good knowledge of the theory of elliptic curves and quite a lot
and
of algebraic geometry).
a) Po(T) =1~ T and Pou(T) = 1 ~ ¢"T. Dwork’s p-adic proof (1960) of the rationality of zeta functions works
for affine or projective varieties, be they non-singular or not. For a non-
b) We can write P; = Hi”_fl(l —wyyT), where wy; are algebraic integers such singular projective hypersurface defined by a polynomial f € Fo{Xy, ..., Xal,
that {wij| = ¢/% for all 4,5, homogeneous of degree d, Dwork proved that its zeta function is of the form
plletes) i O )
o) If x = Zfio(--l)'bz, then the zeta function satisfies the functional equa- 2(t) 1%;:23 for some P € 1+¢Z[t] of degree (@-1 +(a D@D We will prove
tion s in the next section, in the much easier case of diagonal hypersurfaces (a
1 n - famous theorem of Weil).
Zx (qT,> = (2
T Zx(T)
for some sign k. 9.A.6 Zeta function of diagonal hypersurfaces
If X is a smooth projective curve of genus g {this is an important invariant In this part, we show how to compute the zeta function of a diagonal
attached to curves; in the notations of the theorem, we have by = 2g) over IFg, hypersurface. This beautiful result, due to Weil was also the starting point of
Weil proved in 1940 that its zeta function can be written in the form % the famous Weil conjectures.

for some polynomial P &€ 1+ tZ[t]. Moreover, the roots of P have absolute Theorem 9.A.24. (Weil) Let I > 1, m|g — 1 and let X be the projective
value 1/,/4. So in this case by = by = 1, by = 2¢ and moreover we can write hypersurface of equation agzy' + a1z+ -+ ax = 0. There ezists P € Z[T]
of degree d = M—‘B’(J‘lfifllfi such that
X ()l =1 +¢" = (W +wf -+ k)
. pry-ut
with jwi}
i = /3. In P particular, we obtain the very nontrivial estimate ) Zx(T) = g eng= -
X (Fgr) = (T4 ¢ < 296" b) If P(z) = 0, then 1/z is an algebraic integer of absolute value q“‘z‘l,
 454 Chapter 9. A Little Introduction to Algebraic Number Theory
9.A. Equations over Finite Fields 455

¢) Therg exists an explicit integer x such that

Note that Wy(xo....,x:) # 0 for all (xo,...,x1) € S, as the Gauss sum
associated to a nontrivial character is nonzero. Thus deg(P) = |S]. It remains
Zx (y%) =+ (¢F7) 221, to find this number. See |S| as a function f() of I. Let g{l) be the number
of I + 1-tuples (xo,...,X:) of nontrivial characters such that x7* = 1 and
Proof. Recall that by theorem 9.A.16 we have for any q Xo---xt # L. Clearly, (1) = f({ +1). But f(I) + g(I) is just the number of
an equality
tuples of nontrivial characters such that x{* = 1. As there are m — 1 nontrivial
-1

XEN =T 042 3 Woloxu characters of order dividing m, we deduce that f(I) + g(I) = (m — 1)**1. One
immediately deduces that
=0 X0 XXt
the sum being taken (m - 1) 4 (1)(m - 1)
fy = T,
over all nontrivial characters y; of F; such that
x™ =
1 and xox:+--xt = 1. The main point is to study how the numbers
Wen(xo0:---,x1) vary and this is accomplished by the Davenport-Hasse
re- finishing therefore the computation of deg(P).
lation. More precisely, recall that for a character X of F; we have a charact
er By definition of Wy(xo.-.-,x:) and by the fact that lg(x:)| = /g, we
Xnlz) = X(qu,‘ /¥, (x)) and that x ~ X, induces a bijecti 1
on between charac-
ters of I, of a given order and characters of Fjn of the same deduce that {Wy(xo,.--,x:)l = ¢ Z . This yields part 2) of the theorem.
order (proposition
9.A.7). So, if § is the set of [ 4 1-tuples (xo,...,x:) of nontriv Next, the fact that g(x,-)g(x‘-’l) = xi(—1)g and xo---x; = 1 implies that
ial characters
of Fy such that x* = 1 and Xox1---xi = 1, then q".‘l
Wolg'seoxi ) =
-1
- Wolxo,---»x1)
X () = 3 g™ + 721 WelmmXam
o xin):
F=
(Xo:X3 1o x0)ES As clearly the set S is stable by inversion, we deduce that the map z — ?:‘;; is
a permutation of the roots of P. From here, it is an easy but tedious exercise
On the other hand, by the Davenport-Hasse relation (theor
em 9.A.14) we can to deduce the third part of the theorem.
write
Finally, it remains to prove that P € Z{T]. As Zx(T} € Q[[T]], we
must have P € Q[T]. We will prove that the coefficients of P are algebraic
War(Xo.n: X1.n5 i) = (DT W, ) integers, which will be enough to conclude. It is enough (taking into account
We deduce that the definition of P} to check that Wy(xq.....x:)/q is an algebraic integer. As
=1 1 o xi{a:) are roots of unity, it will therefore suffice to check that £X2-900) 15 ap
X(Fe)l =3 q" = (-1 3~ ((;])LW«;(XO«..-,X:)) algebraic integer. But this is an obvious consequence of lemma 9.A.18. This
=0 x5 X0 ES a finally proves the theorem! O
and so by the previous paragraph we finally obtain
the first part of the theorem,
with

roy= ] (1= Wil

. 71 i+1

(X0, x2)ES
7).
 9.B. A Glimpse of Algebraic Number Theory 457

Addendum 9.B A Glimpse of Algebraic There is a very nice characterization of prime and maximal ideals of a
Number Theory ring in terms of quotient rings. The proof is essentially trivial unwinding of
definitions, but the result is crucial:
This addendum recalls the basic properties of number fields. Of course, Proposition 9.B.2. 1} An ideal I of R is mazimal if and only if R/I is o
one would need a whole book (and actually much more. .. ) to properly develop field.
the theory of number fields, as even proving the basic properties requires a
lot of commutative algebra. We will try to as elementary as possible, 2} Anideal I of R is prime if and only if R/I has no zero divisors.
while still giving some proofs. We warn the reader that a long part of this
As a field has no zero divisors, this proposition implies that any maximal
addendums is very abstract. To see the power of the notions and theorems ideal is a prime ideal. There are however prime ideals which are not maximal:
discussed, we advise the reader to start with the last part of the addendum, the ideal (2) in Z[X] is prime and not maximal, as the quotient ring is Fa]X],
which discusses applications to problems with very elementary statements and which has no nonzero zero divisors but is not a field.
very non-elementary solutions. ..
There are natural operations on ideals: if I,.J are ideals of a ring R, one
defines their sum 7+ J = {i + j|i € I,j € J}. Tt is easy to check that this is
9.B.1 Ideals and quotient rings an ideal. The analogous definition for multiplication would fail {in general) to
yield an ideal, so one defines the product of ideals I, J as the ideal generated
Let R be a commutative ring. An ideal of R is a nonempty subset 7 of R by all products ¢ -7 with (2,5} € I x J.
which is stable under addition and such that ez € I foralla € Rand z € I.
Note that this is far stronger than the stability of I under multiplication. It 9.B.2 Field extensions
is fairly easy to construct ideals of B: if z1,72,...,2, € R then
We say that L is a field extension of K if both K, L are fields and K C L.
(1,29,
.. ®n) = {@121 + -~ + anTs|e; € R} The extension” L/K is called finite if L is a finite dimensional K-vector space.
In this case, we define the degree of the extension to be
is obviously an ideal, called the ideal generated by z1,22,..., Zp.
Once we have an ideal T in a ring R, we can naturally construct a quotient
[L: K] = dimg(L).
ring, whose elements are coset classes & = o + I with « € R and addition, For instance, the extension C/R has degree 2, as 1,7 is a basis of C over R.
multiplication are defined by @+b=a+ b and @-b = ab. It is an easy exercise On the other hand, the extension C/Q is infinite (for example, because C
to check that it is well-defined (the issue is that we may have a even if is uncountable and € is countable). We will mostly be interested in finite
a # o and one needs to check that if @ = and b =¥, thena + b=o + 1, extensions, for which the following result is of constant use:
similarly for multiplication}.
Proposition 9.B.3. Let L/K and M/L be finite extensions of fields. Then
Definition 9.B.1. 1) An ideal I of R is called maximal if I % R and if T M/K is finite and
is not contained in any ideal different from I and R. [M:K}=[M:L}-[L:K].
"This notation should not be confused with the quotient ring previously discussed, simply
2) Anideal I of R is called prime if I # Rand ab € R~1I forany a,b € R—1. because K is not an ideal in L unless K = L.
458 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 459

Proof. One can easily check that if (2); is a basis of M as L-vector space then (the classes of) 1, X,...,
X% ! form a K-basis of K{X]/(m) and
and (y;); is a basis of L as K-vector space, then (z;y;): ; is a basis of M as so [K[X]/(m) : K] = d. Combining this with the previous isomorphism
K-vector space. [m] finishes the proof.

2) This is clear: if d = [K(l) : K], then 1,1,I%,..1%

. € K(l) cannot be
9.B.3 Algebraic numbers and algebraic integers
linearly independent over K. This forces a nonzero polynomial equation
If L/K is an extension of fields and if [ € L, we say that [ is algebraic with coefficients in K satisfied by { and so [ is algebraic over K.
over K if there is a nonzero polynomial f € K[X] such that f(I) = 0. In ]
this case, there is a unique monic polynomial 7; € K{X] of least degree which
vanishes at I. It is called the minimal polynomial of [ and it is irreducible in Combining the previous results, we can now prove the following nontrivial
K[X], by minimality. The division algorithm shows that the only polynomials result (which can also be obtained using the theorem: on symmetric polynomi-
f € K{X] such that f(I) = 0 are the multiples of m. Recall that K(l) is the als):
smallest field containing K and [ and it can also be described as Theorem 9.B.5. Let L/K be any field extension. The set of elements of L
which are algebraic over K forms a subfield of L. This subfield is equal to L
1) = {E1.0 < K1X).50 0} = K101 (7011 € ). if L/K is finite.

Proof. If 11,13 € L are algebraic over K, then by proposition 9.B.4 K(h)/K

To prove this equality, it suffices to show that if f € K[X] does not vanish
and K(l)(l2)/K(l;) are finite extensions (note that Iy is also algebraic over
at I, then —1, is of the form A(l) for some A € K[X]. But since f(I) # 0
K (1)) Thus by proposition 9.B.3, K(1)(l5)/K is finite. But K(13){l2) con-
and m; is trreducible, f and m; are relatively prime, so there are polynomials
tains K (li+l), K(lilz) and K (I /Iy). We deduce thatif z € {li+l, iz 1 15},
A, B € K{X] such that Af + Bm = 1. Evaluation at { yields the result. The
then K(z)/K is finite and the result follows by proposition 9.B.4. The second
following proposition is easy, but fundamental.
part is also a trivial consequence of proposition 9.B.4. a
Proposition 9.B.4, Let L/K be any extension of fields.
Definition 9.B.6. 1) A number 2 € C is called algebraic if it is algebraic
1) Let I € L be algebraic over K. Then there is an isomorphism of K- over . It is called an algebraic integer if its minimal polynomial over @
algebras® between K (1) and K[X]/(m). Moreover, (K(1) : K] = degm < has integer coefficients. By Gauss’ lemma, this is equivalent to the fact
. that 2 is root of some monic polynomial with integer coefficients.

2) Conversely, ifl € L and [K(1) : K] < oo, then l is algebraic over K. 2) We denote by @ (respectively Z) the set of algebraic numbers {respec-
tively algebraic integers).
Proof. 1) Consider the maep sending f € K[X] to f({). It is a map of
K-algebras, vanishing precisely on the ideal (m), by definition of #;. The following result is an easy consequence of the theorem of symmetric
It is easy to check that it induces an isomorphism between K{X]/{m) polynomials 9.10.
and K{l], obtained by sending f + (m) to f(I). Next, if d = degm,
Thegrem 9.B.7. Q is an algebraically closed field and Z is a ring. For any
#This means an isomorphism of rings which is K-linear. = € Q there evists n > 1 such that nx € Z.
460 Chapter 9. A Litile Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 461

Proof. The previous theorem shows that @ is a field. Suppose that z € € 2) Let (. be a primitive nth root of unity in C and consider K = Q(¢n)-
satisfies 2™ + ap_12" 1+ 4 ag = 0 for some o; € . We want to prove The irreducibility of cyclotomic polynomials (a fairly nontrivial theorem)
that z € Q. Let afi;fi be the conjugates of ax (i.e. the Toots of the minimal implies that the nth cyclotomic polynomial is the minimal polynomial
polynomial of ay, including ag). The theorem of symmetric polynomials easily of Gn. so that K has degree ¢(n) over Q. One can prove with quite a
implies that lot of effort that Og = Z[(,), i.e. there are no algebraic integers in K
except for the obvious ones.
Fao I Xt 4alrrdxmt g g o)
bo ket 9.B.4 Factorization in Ok, the fundamental theorems
has rational coeflicients and vanishes at z, from where the result follows. To Since the proofs of the theorems stated in this section are rather long and
prove that Z is a ring, consider for instance z,y € Z and let #y, ...z, and technical, we will simply state them without proof, referring the reader to any
Y1,42,- -+ Ym be all roots of the minimal polynomials of z and y. Another basic number theory book. We prefer to focus on their arithmetic applications.
application of the theorem of symmetric polynomials shows that Hw (X — Let K be a number field of degree d over @ and let Ox he the subring of K
xi = yj), respectively [T, ;(X — x; - y;) have integer coefficients and vanish at consisting of algebraic integers.
x +y, respectively « - y. Finally, to prove the last statement, take z € Q and
Theorem 9.B.10. If I is a nonzero ideal of Ok, then Ok/1 is a finite ring.
choose integers ag, a1, ..., a, such that apa™ + ay_147" 1 4 - + ag = 0 and
an # 0. Then Remark 9.B.11. Actually, one can prove that if [K : Q] = d, then there exist
Z1,22,-.-,%q € O such that the map Z% — O sending {n1,n2,...,04) to
(@)™ 4 ot (an)™™ -+ agal = 0, 73%1 + -+ +na%g is a bijection. This easily implies the previous theorem: let
« € I be nonzero, then the norm n of is again in I and is nonzero. Hence
50 Gy -3 € Le [m] I contains nOx. But the previous resuit shows that Ok /nOk is in bijection
Let us introduce now the main object of this addendum.
with Z4/nZ?, which is finite, with [n}¢ elements.

Definition 9.B.8. A number field is a finite extension of Q. If X is a number Corollary 9.B.12. If p is a nonzero prime ideal, then Ok/g is a finite field
field, then we let Oy = K NZ. end 0 g is a mazimal ideal.

Erample 9.B.9. 1) Let d # £1 be a squarefree integer and consider K = Proof. The ring R = Ok/p is finite with no zerc divisors. Let # 0 be any
Q(Vd) (as usual, if d < 0, vd = iv/=d). Then K has degree 2 over @, element of R. As R is finite, there must be i < j such that ' = 27, Then
but the structure of Og depends on the residue class modulo 4 of d: if 2H(z?7" — 1) = 0 and as there are no zero divisors in R, we obtain 27~% = 1.
d= 1 (mod 4), then Ox = Z [1/2], while if d = 2 or 3 mod 4, then Thus < is a unit. This proves that R is a field and the result follows. a
O = Z[vd. This is not difficult to prove: imagine that 7 € O and Definition 9.B.13. If ] is a nonzero ideal of I, we define its norm
write z = a-+bvd. If b = 0, then a must be an integer (since it is rational
and algebraic integer), so we are done. Otherwise, the conjugates of z
N(I) = |0k /1.
are z and y = a — bv/d. Thus 2a and ¢® — db* must be integers, which
easily implies the desired result. This is an integer which lies in 7, by Lagrange’s theorem in the group Og /1.
 462 Chapter 9. A Little Introduction to
Algebraic Number T "heory
9.B. A Glimpse of Algebraic Number Theory 463
We cannot emphasize enc
ugh the importance of
the theory of algebraic the following result for
number fields. Suffice it which is easily seen to be a nonzero ideal of 0. Moreover, Oy # d()L, as
same role as the fundament to say that it plays exactly the
al the orem of arithmetic (i.e. otherwise we would have 1 = ayx; + --- + anTy for some a; € ;;l an szx;i
ization theorem) and we the unique factor-
will leave the reader recall z; € Oy, and by taking norms, we would get 1 € p. Hence, by t] t=. t;;lrev m
of elementary number the that basically all results
ory follow from it. theorem, there exists a prime £ of O diwd.mg pOL.. By the S«::m(. eorcn : )
Theorem 9.B.14, this situply means that ¢ C 8. Note that 3 is not unique, but there are :)imz
(Kummer, Dedekind) Let K be o num
per Jield. finitely many possibilities for it. On the other hand, given a nonlzell':eg thc,
) Any ideal of Oy, different from 0 and Ok, ideal 4 of Op, there is a unique prime g of O(( such that o C B. n; ed, b
mutation) written in the can be uniquely (up to per- existence is obtained by taking p = BNOx (it is an easy exercise for t e re;a( 0
Jorm gy ‘P2 o for somen
prime ideals p;, not nec 21 and some to check that this is indeed a nonzero prime -idea.l of OK)_ and t:he ;n;:m(.n(x;
essarily distinct,
follows from the fact that any nonzero prime ideal of O is maxim: ( em:CeA,H
b) If I and J are two non
zero ideals of Oy, then o1 C 3 and @y C S for some different prin?% p; of O, then 1 € py +p2
Moreover, I ¢ J if and only N(I-J)y=N(I). N(J). C 3,
if there exists an ideal J' stc a contradiction).
h that I = J. J’
Remark 9.8.15. One can also Definition 9.B.18. Let L/K be a finite extension of nu'n:nbef~ fields. aul(]i let ng
prove that for all € Og we have
be a nonzero prime ideal of Ox. A prime ideal 3 of Oy is said to lie above p
N(zOk) = [N(z)|. if p C B. We also write Blp in this case.
Remark 9.B.16. Suppose We can resume the previous discussion by saying that any prime_ Bof 0,%
that K has degree d over
prime. Let pOy = w5 @Q and that pis a rational has a unique prime p of Og below it, nml.]ely AnNO0g _&nd ?any ;})lniuief §|0
- 57 50;9 be the factorization
As pOy has norm 2% (by of the ideal pOy. O has at least one (but finitely many) prime 8 above ft‘ Note tha
the first part of the remark b /;gs',
%= Nig)o ), theoremn 9.B.14 yields then the inclusion Ox C Oy induces an injective morphism
... Npg)®e and so there are Ok /p — OL/5,
N{gi) = pf. We have the hon neg ati ve integers f; such that realizing therefore Or /3 as a finite extension of O /p.
fundamental relation
Definition 9.B.19. Let 8lp be as above.
elf]+e;f2+-~~+egf =[K:q].
1) The residual degree of 8/p is defined by f(8/p) = [0/8 Ok /gl
Remark 9.B.17. Let K be a number field,
let p be a nonzero prime
Ok and let 2 O be prime to @ (ie. ideal of
factorization of 20k or, p does not appear in
the prime
equivalently, 0y + =
with N(gp) clements, we obt Og). Since Ok/p is a fiel
ain from Lagrange’s theore d
analogue of Fermat’s little m the following useful Note that by definition we have
theorem: ¢Me)-1 = ¢ (mo
d p).
Consider now a finite ext 0L
ension L/K of number = H peare),
nonzero prime ideal of fields and let p be a
Og. Let
Ble
001 = {a1z1 + ayzs + b
anTnln > 10 € g € O}, Using this and proposition 9.B.3, we easily obtf;in the following u:zfu{li E)l:iosperty
of the ramification index and residual degree in a tower of number fields.
464 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 465

Proposition 9.B.20. Let M/L/K be a tower of finite extensions of number Proof. Lift arbitrarily f; to some monic polynomials f; € Z{X]. The key
fields and let p|Blp be primes of On, Or and Og. Then point is to prove that the natural map Z{X]/(p, f;) — Ziz}/(p, fi(z)) sending
(the class of) h to (the class of) h(z) is an isomorphism of rings. Assume
elp/B) - e(B/0) = elp/p). flo/B)- f(B/w) = Flp/p). for a moment that we proved this. Let a = |Ok/Z[z]|, so ged(p,a) = 1
and aOk C Zlz]. A standard argument using Bézout’s lemma shows that
Definition 9.B.21. Let L/K be a finite extension of number fields. A prime
Ok = pOx + Z{z], thus Og/gp; is naturally isomorphic to Zz]/(p, fi(z)).
o of Ok is called
Since Z[X]/(p, f:(X)) is clearly isomorphic to Fp{X]/(f;), which is a field with
a) unramified in L if e(8/gp) = 1 for all Blp in L. pde8/i elements, it follows from the previous discussion that the p; are different
prime ideals with N(p;) = pd8%:. Since ¢ C pOx + fi(z)" Ok, f(z) =0
b) totally split in L if e(3/p) = 1 and f(B/p) = 1 for all Blp in L. This and f = [i=1
T, f{* (mod pZ[X]), it follows that [JL; ¢f' C pOx and so by
is equivalent? to: Oy is the product of L : K] different prime ideals of theorem 9.B.14 we have pOjc| [T, ©§*. So, if pOx = [[Z_; ¢i*, we must have
Qp. i < e;. We conclude by noting that both sums 3, e;-degf; and 3, s; -degf;
are equal to [K : Q] (see remark 9.B.16}, so we must have s; = e; for all 7.
The following theorem gives a practical way to factor a prime in a number
Let us prove now that Z[X]/(p, f;) — Zlz]/(p, fi(z)) is bijective. Since
field. The precise statement is a bit complicated, but the message is very surjectivity is clear, it remains to prove the following assertion: if b € Z[X]
simple: for most primes p, factoring pOx in Ok comes down to factoring satisfies h(z) € pZiz] + filz)Ziz], then h € pZ{X] + fi(X) - Z[X]. Write
F & Fy[X], where £ is the minimal polynomial of any primitive element of K h(z) = pA(z) + fi(z)B(z) for some A, B € Z[X]. Since f is the minimal
that lives in Og.
polynomial of «, there is r € Z[X] such that A = pA + £;B +rf. It suffices to
Theorem 9.B.22. (Dedekind, Kummer) Let K = Q(z) be a number field, use again that f € pZ[X] + f; - Z[X] to finish the proof. a
where © € Og has minimal polynomial f. Let p be a prime which does not
divide’® |Ox /Zfz}| and let Remark 9.B.23. It is not very difficult to prove that if p does not divide the
discriminant of f, then p does not divide |Ox/Z{z]}, so the theorem can be
applied.

be the prime factorization of f € Fp[X]. Then

9.B.5 Two classical examples
g
pOx =] ot Consider a squarefree integer d # £1 and let X = Q(v/d). We saw that
i=1 Ok = Ziz}, where z = liéfi if d = 1 {mod 4} and & = V/d otherwise. The
minimal polynomial of z is X2 — X + %4 in the first case and X2 —d in the
where p; = pOx + fi(e)Os are different prime ideals and N{p;) = po57.
second case. Theorem 9.B.22 shows that in order to understand the prime
We have 3., e(8/8) 1(8/9) = I : K, by an argument similar o that used in remark factorization of pOk, we need to understand the prime factorization of these
9.B.16, polynomials modulo p. Since a quadratic polynomial modulo p is irreducible
“"Note that the result stated in vemark 9.B.11 shows that Ox /Z[z} is a finite set. if and only if it has a root in Fy, we easily deduce the following
466 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 467

Proposition 9.B.24. Let d # 1 be a squarefree integer and let K = Q(V/d). b) Ifpln andn = p-m with ged(m,p) = 1, then pOx = (g1~ )P 77",
Let p be a prime. where s = m@f% and each p; is of degree ord(p mod m).

a) If p > 2, then pOg is a prime ideal if (g) = —1, a product of two

9.B.6 The primitive element theorem and embeddings of
different prime ideals if (g) =1 and the square of a prime ideal if pld. number fields
b} Ifp = 2, then pOy is the square of a prime ideal if d ,6,3,7 (mod 8), ‘When working with subfields of C, the following result is very handy. We
a prime ideal ifd = 5 (mod 8) and a product of two nt prime ideals will use it constantly to shorten proofs of results which actually hold in much
fd=1 (mod 8). greater generality.

Consider now # > 1 and let K = Q((), where {, is a primitive nth Theorem 9.B.26. (primitive element theorem) Let L/ K be a finite extension
root of unity. As we have already said, it is rather difficult to prove that of subfields of C. Then there exists | € L such that L = K(I}.
O = Z(,] and we will take this for granted (actually, it is easier to use
remark 9.B.23). In this case, theorem 9.B.22 reduces the prime factorization Proof. As L/K is finite, there are elements z,...,z, € L such that L =
of pOx to that of ¢, € F,{X], where ¢, is the nth cyclotomic polynomial K(z1,23,...,2,) (for instance, the elements of a K-basis of L over K). Thus,
{whose roots are precisely the primitive nth roots of unity). Assume that p by induction on n it is enough to prove that if z, y are algebraic over K, then
does not divide z and let g be an irreducible factor of degree [ of ¢, € Fp[X]. there exists { € L = K(z,y) such that L = K(I).
Let = be a root of ¢ in an algebraic closure of Fy. Theorem 9.A.4 shows that Let f,g € K[X] be the minimal polynomials of x, y respectively and let
9(X) = HI/:(,l (X —2”). Let d be the order of p modulo n. Since n is the least Ty = Z,%2,..-, %, and Y1 = §, %2, ..., Ym be their roots. Clearly, there exists
positive power of ¢ equal to 1 (because x is a root of ¢,), it follows that the c € K such that 2; +cy; # & + cy for all (i,5) # (1,1) (each of the previous
sequence =¥ is periodic with period d, so we must have f = d. That is, ¢, linear equations in ¢ has at most one solution in X and K is infinite). We
factors mod p as a product of irreducible polynomials of degree d, the order of claim that I = x + ¢y works. Clearly K(I) C L, so it is enough to check
p module 7. Since X —1 is squarefree modulo p (because the hypothesis that that z,y € K(I). As! = 2 + cy, it is enough to do it for y. Now, we
p does not divide n implies that X™ — 1 is relatively prime to its derivative}, know f(l —cy) = 0, so the polynomial f({ — cX} € K(I){X] has & common
we deduce that in theorem 9.B.22 we have e = eg = --- =1land fy=d root y with g. But by construction this is the only common root of these
polynomials. Moreover, it is a simple root, as g is irreducible over @, so it has
for all , so g = @ Assume now that pln and let n = p/ - m for some m
simple roots. Finally, we conclude that the greatest common divisor of these
relatively prime to p. It is easy to see that ¢n{X) = ¢pn(X f’k) [ m(X w* 1), 80
two polynomials is X —y. As these two polynomials have coefficients in K ({),
modulo p we have ¢, = (,}f: - + This reduces the problem of factoring pOx s0 does their greatest common divisor and so y € K(I). The result follows. O
to the previous case. All in all, we have the following useful result:
‘We will use the primitive element theorem to prove some basic results on
Proposition 9.B.25. Let n be an integer greater than 1, let K = Q(() and
letp be a prime.
the structure of number fields. The first one concerns the embeddings of a
number field in C. Note that Q(v/2) has two such embeddings, namely the
a) If ged(n,p) = 1, then pOk is a product of Ord@%m different prime identity map and a + by2 = @ ~ bv/2. Tt turns out that a number field with
ideals, each of degree ord(p mod n). degree d has exactly d embeddings in C.
468 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 469

Theorem 9.B.27. Let L/K be an estension of number fields (L/K is au- between the z;’s and the same argument shows that the numbers f(z;) must
tomatically finite). Then any embedding K — C estends to exactly (L : K} satisfy these relations. Thus, it is a fairly delicate issue to understand these
embeddings L — C. K-morphisms. This is the content of Galois theory. Let us make an important
definition first:
Proof. Fix an embedding o : K — C and use the primitive element theorem
to write L = K(I) for some ! & L. Then [ is algebraic over K, of degree Definition 9.B.29. Let L be an extension of a field K. The Galois group of
d = [L : K}, with minimal polynomial f € K[X]|. Suppose that ¢’ : I
L over K, denoted Gal(L/K} is the set of bijective K-morphisms f: L — L.
> C
is an embedding that extends o, so o/(z) = o{z) for all z € K. Thus, if ‘We can now prove one basic result of Galois theory, which is also of
g € K[X], then o’(g(I)) = g°(0’()) (where g7 is the polynomial obtained from constant use:
g by applying ¢ to its coefficients) and so ¢’ is determined by o'(1), which has
Theorem 9.B.30. Let L/K be an extension of number fields. Then
o be a root of f7 (use the previous equality with g = f). Conversely, if I''is
a root of f7, we can define an embedding o' (g(1)) = g°(I') for all g € KiX], [Gal(L/K) < [L: K],
well-defined because any two polynomials ¢ and h with g(1) = (1} differ by a
with equality if and only if L is the splitting field of some polynomial f € K[X].
multiple of f. Thus the embeddings of L into C that extend o are in bijection
with roots of f7 and the result follows. (W] Proof. Let L = K(l) for some ! € L. Each element o € Gal(L/K) is uniquely
determined by ¢{}, which must be a conjugate of /. Thus we have a natural
Taking K = @ in the previous theorem, we deduce that any number field injection of Gal(Z/K) in the set of conjugates of I, which has |L : K| elements.
L of degree d over ) embeds in exactly d ways in C. The inequality follows.
Suppose that we have equality and let f € K[X] be the minimal poly-
9.B.7 A bit of Galois theory nomial of {. We will prove that L is the splitting field of f. It is enough
to prove that if z is a conjugate of I, then = € L. But the first paragraph
Let L, M be two extensions of a field K and suppose for simplicity that
and the equality case implies that if [;,...,1, are the conjugates of I, then
they are contained in C. A K-morphism L -+ M is a K-linear map from L to
{o{)lo € Gal(L/K}} = {h,...,1,}. Thus there exists o € Gal(L/K)} such
M which is alse a ring homomorphism. Stated differently (but equivalently),
that = o({). Since o(L} C L, we have =z € L and we are done.
it is an additive and multiplicative map f : L — M such that f(z) =z for all
Conversely, suppose that is the splitting field of some f € K[X], with
wce K.
T00tS Z1,%2,-..,%n. Theorem 9.B.27 applied to the natural inclusion map
Definition 9.B.28. Let f € K{X|. The splitting field of f is the field K — C shows that there are {L : K] K-linear morphisms of rings L — C.
K(z1,%2,...,Ta), where z; € C are the roots of f. But for any such morphism ¢ : L — C we have o(z;) € L, as o(;) is just
some root of f and L contains all these roots. Thus the image of any such
Note that a K-morphism f : K(z1,....,4n) = M (where M/K is an
morphism is a subset of L, i.e. any such morphism is an element of Gal{L/K).
extension) is uniquely determined by the values f(z:), as any element of
The result follows. o
K(z1,...,Zq) is a polynomial with coefficients in K in the z;’s. However,
F(x;) cannot be any element of M, since if P € K[X] kills 2;, then P also Definition 9.B.31. We say that a finite extension L/K of number fields is
kills f(z:): note that P(f{x;)) = f(P(z:)) = 0, as f is K-linear and multi- Galois if 1Gal(L/K)} = [L : K}, or, equivalently, if L is the splitting field of a
plicative. Moreover, there might be algebraic relations with coefficients in K polynomial with coefficients in K.
470 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory ) 471

We are now able to prove the main theorem of Galois theory for number is the subgroup of Gal(L/K) generated by Hy and H. Hence L = LH2 and,
fields. The result holds in much greater generality, but we will not need it as we have seen, this forces H = Hy and so Hy C Ho. O
and we prefer to use all the extra data in order to shorten the proof rather
dramatically. Remark 9.B.33. Using similar arguments, it is not difficult to show that H is
normal in Gal{L/K) (i.e. gHg™" = H for any g € Gal(L/K)) if and only if
Theorem 9.B.32. Let L/K be a finite Galois extension of number fields. LH/K is Galois.
Sending a subficld M of L containing K to Gal(L/M) yields a bijection between
subfields of I which contain K and subgroups of Gal(L/K). The inverse of
this bijection is the map sending the subgroup H of Gal(L/K ) to
9.B.8 Prime factorization in a Galois extension
The results in this section will be crucially used in the applications that
= {z e Llo(z) = 2,50 € H}. will be presented at the end of this addendum. They are absolutely funda-
mental in algebraic number theory.
Moreover, we have Hy C Hy if and only if L#* contains LHz.
Theorem 9.B.34. Let L/K be a Galois extension of number fields, with
Proof. Let us prove first that LE®E/M) — Af for any intermediate field M G = Gal{L/K). Let p be a nonzero prime ideal of Og and let 3y and By be
between L and K. Note that L/M is again finite and Galois (as L is the two prime ideals of Or, which lie over p. Then there exists o € G such that'!
splitting field of a polynomial with coefficients in K, thus also in M ). Write B2 = oB1).
L = M(l) for'some primitive element | & L, with conjugates Iy,la,...,1, over
M. Suppose that z € LO¥E/M) 5 not in M. Thus, we can find fe M{X) Proof. Suppose that 8 # o(B) for all ¢ € G. Using the general version!?
nonconstant of degree less than n and such that = = f(1). We saw in the proof of the Chinese Remainder Theorem, we obtain the existence of a € 3z such
of the previous theorem that {o(I)|o € Gal(L/M)} = {l1,...,1,}. So, for any that a ¢ {8} for all 0 € G. Then [], 0{a) € OxN Bz = p C B and this
i we can find 0; € Gal(L/M) such that oy(l) = ;. Then & = oy(z) = f(I;). contradicts the fact that 8 is a prime ideal and a € 67 1(5;) for any 0 € G.
Hence f(l1) = f(lz) = ---= f(ln), contradicting the fact that f is nonconstant The result follows. =]
of degree less than n. The result follows.
Next, we need to prove that if H is any subgroup of Gal(L/K ), then Corollary 9.B.35. Let L/K be a Galois extension of number fields and let
Gal(L/L¥) = H. © be e nonzerc prime ideal of Ox. Then all primes above p have the same
By the very definition of L7 we have an inclusion H ¢
Gal(L/L). Tt is thus enough to check hat 1Gal(L/LM)| < |H} and, using the ramification index and the same residual degree.
previous theorem, it is enough to check that [L : L#] < {H|. Write L = L# [}
Proof. For the residual degrees, note that any ¢ € G induces a bijection be-
for some I € L. Consider f(X) = [[,eu(X ~ o(l)). This is a polynomial of
tween Oz /5 and Og /a(53), so these two sets have the same number of elements. -
degree |H| whose coefficients are clearly in L¥, Hence ! has degree at most
‘We conclude by the previous theorem. Next, if e is a positive integer such that
|H| over L¥ and the result follows.
It remains to check that Hy ¢ Hy if and only if L7 contains L2, Tt is HNote that by definition o{f;) is the set of all o(z), with = € f;. It is easy to check that
clear that if Hy C Ha, then L5 contains M2 {any element of L fixed by Hy this is again a prime ideal of Oy and that (%) 1 Ok = p.
is also fixed by H)). Assume that L' contains LH2, so LHi y LH2 — [H2,
This is stated as follows: let A be a commutative ring and let Iy, Iz, .., be ideals
such that L + I; = A for all ¢ # j (which is satisfied if I; are different maximal ideals of 4,
But it is clear that the left-hand side of this equality is simply L¥ | where H for instance). Then for any z1,. .., € A there exists = € A such that = —, € L for all i.
472 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 473

/3¢ divides pOy, then o(8)° divides o(pOr) = pOy. The result follows again exists o € Gal(L/K) such that o(a) = b. We claim that ¢ is in Dg and that
easily from the previous theorem. [w] it induces the automorphism = — V) of Oy /8. First, if () # B, then
o71(8) # B and so by our choice of a we have a € o~}(8), i.e. o(a) € 8. But
Definition 9.B.36. Let L/K be a Galois extension of number fields, with
ola) — V@) € 3, forcing a™¥®) € 3 and then a € 3, a contradiction with a
Galois group G. Let 8 be a prime of Q. The decomposition group of 8 is
(mod 8) = a # 0. Thus ¢ € Dg. The automorphism induced by o on Or/8
Dy = {o € Glo(B) = B}. is uniquely determined by its action on « (generator of (Or/8)*) and by our
choice this action is a™®). The result follows. o
Note that Dg is indeed a group and that any ¢ € Dy induces an auto-
morphism ¢ : Op/B — O/8, which is trivial on Ok/p for ¢ = 81 Ok. Let us fix a prime 8 above p in L. By theorem 9.B.34, all primes above p
Hence, we have a natural map Dy — Gal((01/8)/(Ok/p)). Note that are of the form o(f) with o € Gal(L/K), i.e. Gal(L/K) permutes transitively
Gal({0/8)/(Ox/g)) is a cyclic group, generated by the automorphism the different primes 51,.. ., 3, over p. Since there are precisely | D3| elements
x — aN®), since (Or/B)/(Ok/p) is a finite extension of finite fields (see of Gal(L/K) which fix 8, we obtain [L : K] = |Gal(L/K)| = g - |Dy|. But we
the addendum 9.A for the structure of finite fields). The following result is also have (L : K] = e(3/p)- f(B/g)-g, since all ramification indices and residual
rather tricky, but fundamental. degrees of the 5;’s are the same. Hence [Dg| = e(3/p)- f(8/p). In particular,
if g is unramified in L, ie. if e(8/p) = 1, then |Dg| = F(8/p), which is
Theorem 9.B.37. With the previous notations, the map the same as the cardinality of Gal({0./8)/(Ox/g)). We obtain therefore the
following crucial result:
D = Gal((01/8)/(Ox
/)
Theorem 9.B.38. Let L/K be a Galois extension of number fields and let p
is surjective. be a prime of Ox unramified in Oy. Let Blp be a prime of Oy, over p. Then
the map D — Gal({OL/8)/(Ok/)} is a bijection. In particular, there exists
Proof. We may assume that Op/8 # Og/p, as otherwise the statement is
a unique (8, L/K) € Dg such that (8, L/K){(z) = 2V for all z € Op. We
trivial. Let a be a generator of the cyclic group (Or/8)*, so that clearly
call (8, L/K) the Frobenius substitution of § in L/K.
O1/8 = (O /p)e]. Using the general form of the Chinese Remainder The-
orem (see the proof of theorem 9.B.34), we can find @ € Of such that a Remark 9.B.39. Assume that we are only given p. The choice of some 8 over p
(mod 8) = a and a € § for any B’ # S above p. Let F be the minimal yields a Frobenius substitution (8, L/K), but this depends on the choice of .
polynomial of ¢ over K. Then F € Ox|{X] and F{a hence Fla) = 0 However, any other prime above @ is of the form o(3) for some & € Gal(L/K)
in Op/A. But then'® F(a™(®)) = 0, so that F(aV®)) € 8. So, we can find a and it is immediate to check that (¢(8),L/K) =00 (8. L/K) oo™ ! (indeed,
conjugate b of a so that b — a¥®) ¢ 8. It is not difficult!? to see that there the right-hand side is in D), which is trivially equal to 0 Dgo™!, and satisfies
the congruence that uniquely characterizes it). Hence, the conjugacy class of
s
F(X)?, 50 if @ is a root of f in some extension of ¥,, then a7 is also & a0t of £, (8,L/K) in Gal(L/K) does not depend on the choice of 8 and is called the
8ince I/K is Galols, L coutains all conjugates of @ over K, ie. all roots of F. Let M Frobenius conjugacy class of p. Note that if Gal(L/K) is abelian, then this
be the field generated over K by these conjugates, Le. the splitting field of K. Then M/K conjugacy class is reduced to an element, which we denote (g, L/K): it is
is 2 Galois subextension of L/K, thus any clement of Gal(M/K) extends to an element of (8, L/K) for any 8 over p.
Gal(L/K). But the isomorphisms K[X]/(F) — M sending X to a and b respectively yield
o € Gal(M/K) such that o(a) = b. ‘We leave to the reader to check the following easy results:
474 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 475

Proposition 9.B.40. Let M/L/K be a tower of Galois extensions of number ‘We will also need the following easy consequence of Chebotarev’s density
fields and let p|Blp be a tower of prime ideals. Assume that p is unramified theorem:
in M (thus 8 is unramified in M and @ is unramified in L ). Then we have
the following relations. Proposition 9.B.43. Let 0 € Gal(L/K). There are infinitely many prime
ideals o of degree 1 ofK such that one can find Blp in L with (B, L/K) = 0.
a) (p, M/L) = (p, MJKYB/9}. Actually, this relation still holds if M/K is
nof necessarily Galots. Proof. Take a Galois extension E over which contains L. Since L/K is
== Galois, we can extend ¢ to an automorphism again denoted ¢ of Gal(E/K).
o) (o M/K)lp = (8. L/K). By Chebotarev, we can fiud infinitely many primes p such that there is plp
in £ with (p, E/Q) = 0. Then D, C Gal(E/K}, so p = pN K has degree 1
9.B.9 Bauer’s theorem and Chebotarev’s density theorem (lemma 9.B.49). Let 3= pNOy. Then

Before stating the following fundamental result, we need one more defini-
tion. If S is a set of primes, its Dirichlet density is
(8, L/K) = (p. B/K)], = (0, E/Q)l, = o
and we are done. o
d(8)= lim }-mebl
<&~
:-»12‘179 Definition 9.B.44. If K is a number field, let Pi(K) be the set of rational
primes p for which pOx has at least one prime ideal factor g of residual deg)
if the limit exists.The next very deep theorem was conjectured (and proved 1 (i.e. such that O /g is the field with p elements).
in some special cases} by Frobenius. It is a vast generalization of Dirichlet’s
The following result explains why P;(K) is an interesting object:
theorern.
Theorem 9.B.41. (Chebotarev) Let K be a finite Galois extension of Q of Proposition 9.B.45. Let f € Z[X] be a monic irreducible polynomial, let 6
degreen and let g € Gall K/Q). Let S be the set ofpn'mesp such that (p, K/Q) be a root of f and let K = Q(6). Then there is ¢ such that
is conjugate to g for all prime divisors o of p. Then d(S) = L-i{{hgh~!|h € G}|.
Pi(K)N[e.00) = {p > c|3z € Zsuch that p|f(z)}.
Here is a typical application of this deep theorem.
Proof. If p is a sufficiently large prime, then the residual degrees of the prime
Ezample 9.B.42. Let be
{ a pnme and consider the set S of those primes p= 1
divisors of pOy are given by the degrees of the irreducible factors of f € F,[X],
(mod I} such that 2P = 1 (mod p). These two conditions are equivalent
by theorem 9.B.14. Hence, for such p we have p € Py(K) if and only if f has a
to the statement that X' — 2 splits into distinct linear factors in ]F,.[X] {as
linear factor in Fp[X]. i.e. if and only if there exists z € Z such that p{f(z}. O
25 = (mod p) is equivalent to the existence of y € F,, such that ¢ = 2).
also oqmvalent to p being a product of different primes in O, where Here is a rather nice application of this proposition:
K = Q(e ¥ ,\}F is the splitting field of X' — 2. Using ava.m Chebotarev’s Theorem 9.B.46. (Nagell) If f € Z[X], let P(f) be the set of primes
theorem, we deduce that d(S) = T_—QT But K contains Qe . ) and Q(V/2), p for which the congruence f(z) = 0 (mod p) has at least one solution.
which have degrees | — 1, respectively I. Thus [K : Q] is a mulnplc of I(l—1) If fi.fo,-.., fx are nonconstant polynomials with integer coefficients, then
and so necessarily {K : Q] = I({ ~ 1) and d(8) = fi P(f1) Y P(f2) N+ 0 P(fz) is infinite.
 476 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 477

Proof. The case k = 1 is a classical result due to Schur, but for the reader’s elements. That is, [H 1 Dg| = f(3/g). But this equals also 28/2) — _[Dsl
convenience let us recall the proof. If f1(0) = 0, everything is clear. Consider The result follows. flely = J/r ”E
the numbers z, = MZ?)!'I@—). We have |z,| > 1 and z, = 1 (mod n!) if n is
Lemma 9.B.49.
Assume that L and K are as in the previous lemma and let
large enough, so there exists p, |z, with p, > n and the result follows.
p be a prime which is unramified in L. Then
The general case is more difficult. We may assume that f; are irreducible
monic. Let z; be roots of f; and let K; = Q(z). Let K be the least number 1) p has a prime factorp in K with f(p/p) = 1 if and only if there is Blp
field containing all Ky’s and write K = Q(z) for some z € Og with minimal in L such that Dg C H.
polynomial f. Applying the case k& = 1 to f and using the previous proposition,
we see that we can find infinitely many p which have a prime p of degree 1 in 2) pis completely split in K if and only if for all Blp in L, we have DgCH.
K. Let f; = 9N Ok,. Since f(p/p) = 1, we also have f(5;/p) = 1. Applying
Proof. 1) is an immediate consequence of the previous lemma. For 2), it is
once more theorem 9.B.22 (this time for each K) we see that all but finitely
clear that if p is completely split in X, then for any Sip we have Ds C H (asif
many primes p (among the infinitely many we have just found) belong to
© = BN Ox, then necessarily f(p/p) = 1). Conversely, if Dg C H for any 3|p,
NE 1 P(fi), which is therefore infinite. !
then the previous lemma implies that f(p/p) = 1 for any plpin K. But note
The following beautiful theorem due to Bauer was discovered before Cheb- that p is unramified in K, as it is already in L. Thus p must be completely
otarev proved his theorem. It is however more conceptual nowadays to see split in K. O
Bauer's theorem as a consequence of Chebotarev’s density theorem, which is
Let us prove now the theorem. Choose® a finite Galois extension L of
what we will do.
Q containing K; and K and let H; = Gal(L/Kj;). By the main theorem of
4 Theorem 9.8B.47. (Bauer) Let K1 be a number field and let Ky be o number Galois theory (theorem 9.B.32), it is enough to prove that H; C Hy. Choose
field which is Galois over Q. Suppose that there exists a set of primes S of any o € Hy. By Chebotarev’s theorem, there is p ¢ S such that p has a prime
Dirichlet density 0, with the following property: if p ¢ S is in Pi(Ky), then p factor 8 in L with Dg = {0) (cyclic subgroup of H; generated by o). By
is completely split in Ky. Then K2 C K. lemma 9.B.49, we have p € Pi{K) and so pis totally split in K3, which forces,
by the same lemma, D C H,. Hence o € Hy and we are done. O
Proof. We start with two easy lemimas, which express the condition p € P1(K)
in a group-theoretic way. They are also results of independent interest and Remark 9.B.50. Here is another useful consequence of lemma 9.B.49. Let
will be used in other applications. 7 > 1 be an integer and let H be a subgroup of {Z/nZ)". We claim that a
prime p not dividing n is totally split in Q(¢,)” if and only if p (mod n) € H.
Lemma 9.8.48. Let L be a number field which is Galois over Q, with group
Indeed, such a prime is unramified in Q((,) and by lemma 9.B.49 it is totally
G. Let Q C K C L be a subfield and let H = Gal(L/K). Suppose that Blplp split in Q(¢x) if and only if for any Blp in Q((n) we have D3 € H. Since
is a chain of prime ideals in L,
K, Q such that p is unvamified in L. Then
Gal(Q((,)/Q) is abelian, this is also equivalent to (p, Q()/Q) € H, which
flo/p) =1 if end only if Dg C H.
is saying that p (mod n) € H (we naturally identified H with a subgroup of
Proof. Note that the decomposition group of 8 with respect to the extension Gal(Q(G,)/Q))-
L/K is simply Dg 1 H. But since p is unramified in L, p is unramified in *For instance, if K1 = Qfz) and K2 = Q(y), consider the extension generated over Qby
L/K and so the decomposition group of § with respect to L/K has f(8/¢) all conjugates of z and y.
 478 Chapter 9. A Little Introduction to Algebraic Numb
er Theory 9.B. A Glimpse of Algebraic Number Theory 479

9.B.10 Finally, a reward: applications to and Schur, which generalizes Bertrand’s postulate and for a proof
“elementary—looking” problems of which we
refer the reader to [26]. ]
In this section we show how the deep results in the previous secti
can be combined to yield some fairly nontr ons FTheorem 9.B.52. Ifn > 2k, then (%) has a prime divisor greater than k.
ivial theorems with “elementary”
aspects. The first one uses a generaliza
tion of Bertrand’s postulate to obtain Let us consider now the second application.
the following difficult irreducibility result
.
Theorem 9.B.53. (Davenport, Lewis, Schinzel) Let f QIX] be a polyno-
T Theorem 9.B.51. (Schur) If n > 2 and 1,42
- .-, ,
Gn_y are integers, then mial with the following property: any arithmetic progression P C Z contains
the polynomial
an integer x such that f(z) is the sum of squares of two rational numbers.
2
Then there exist polynomials g, h € Q[X] ] such that f = g% + h2.
oo ug»%-{»alX +1
Proof. We may assume that f is not constant. Using Gauss’ lemma, we can
is dirreducible in (. write f =¢ f‘ ----- o for some ¢ € @, some primitive, distinct, irreducible
polynemials f; € Z[X] and some ¢; € N*. Fix an index J for which ¢; is an
Proof. 1t is enough to prove that f=X odd number and let 6 be a root of f;. We claim that L = Q(6) contains Q(i).
"4na, X7 g pnlis irreducible
in Z{X]. Suppose that this is not the Using part a) of Bauer’s theorem 9.B.47, it is enough to prove that if ¢ is
case and let g be a monic irreducible
factor of f of degree m, with m < #. a sufficiently large prime in P (Q(8)), then ¢ € Py (Q(#)). We will need the
Choose any prime divisor pof n(n —
1)+ (n—m+1) and consider the reduction f of mod p. The choic following standard argument:
implies that X™ "+ e of »
divides f. If =gk it follows that X"*~™*1
7+ b and since degh = n —m, we must have divides Lemma 8.B.54. There ezists a prime qo with the Jollowing property: for all
X|[g and so pg(0). @ € Py(L) N [go, 00}, one can find an integer x such that vy(f;(z)) = 1 and
Let 2 be a root of g and let K = Ql2], & numbe
r field of degree m. Since Vol fx(2)) =0 for all k # ;.
2lg(0), it follows that p divides N{z-Og)
and s0 we can choose a prime @ of
Ok dividing 20y and such that plp. Let
¢ = e(p/p) < [K : Q] = m be the Proof. First, choose a prime ¢; > [0y, : Z[]]. Take a prime number g greater
ramification index of . Since F(z) = 0, we
obtain (set ay, = 1) than all prime factors of (the numerator or denominator of) ¢ and such that
g does not divide ged(f;(z), fi(z) - Miz; fu(@)) for any 2 € Z and any ¢ > gz.
vp(nl) n! ! To prove the existence of go, note thet [y is relatively prime to - Hkfi
2 min v,p | [y 2t
At >_1S2< )| Frr
write a Bézout relation over Q for f; and £ nkfi Jir and clear denominators.
s0 we can find ¢ such that We claim that go = max(qy.gz) works. Indeed, let ¢ € P{L) N go,oc]. By
Dedekind-Kummer’s theorem 9.B.22, we can find y € Z such that g divides
(i) > iuy(z) = fi(y). Then ¢ does not divide fily) or H,#j Fi(y), s0 one of y ot y+ ¢ satisfies
the desired conditions. [mj
Using the inequalities wp(il) < ;f] and
e < m, we deduce that p<m.
We have therefore proved that all prime facto Now, fix go as in the previous lemma and let ¢ € Pi(L) 1 [go,oc) and
rs of n(n—1)... . {(n—m+1) % as in the lemma. By hypothesis we can find y =
are less than or equal to m. This contradict (mod ¢%) such that
s a famous theorem of Sylvester f(y) = a® + b for some rational numbers a,b. Note that ve(f5()) = 1 and
 480 Chapter 9. A Little Introduction to Algebraic Number Theory 9.B. A Glimpse of Algebraic Number Theory 481

Ue(fe(y)) = 0 for k # j. Then e; = vo(f(y)) = vy(a® ++?) is odd by our greater than ¢ and n, then p divides some f(x), so p = 1 (mod n) and p is
choice of y. We deduce that ¢ = 1 (mod 4) and so g is split in Q(¢}, hence completely split in Q((,).
q € P(Q(¢)). This proves the claim made in the previous paragraph and we Now, write ., = u(#) for some u € Q[X]. Then, since § is irreducible and
conclude that ~1 & L. Hence we can write i = h(¢) for some k € Q[X]. ¢n{u(8)) = 0, the polynomial f divides ¢, o u in Q[X] (recall that ¢, is the
Then f; must divide A2 + 1. If G = ged{h — 1, f;) € Q(i){X], it is casy to see nth cyclotomic polynomial, minimal polynomial of ¢, over Q). For j relatively
that Nogyg(G) = ged(h — i, f3) - ged(h + 4, f;) is a polynomial with rational prite to n, define f; = ged(f,u — %) € Q(¢a)[X]. Let o € Gal(Q((a)/Q) be
coefficients dividing f;. We deduce that f; is a constant times No/o(G). the automorphism sending ¢, to (3. Then clearly f; = a;(f1). so
and this last polynomial is obviously the sum of squares of two polynomials
with rational coefficients.
Applying the previous arguments for each f; such that e; is odd, we
I fi= Mool
ged(jn)=1
deduce that f is of the form c(u? + v?) for a constant ¢ and some u,v € Q[X].
But using once more the hypothesis on ¢ we deduce that ¢ is the sum of squares Clearly, the polynomials f; are pairwise relatively prime, so their product
of two rational numbers and the result follows. [} divides f. But, as we have seen, their product has rational coefficients, hence
by irreducibility of f we must have f = a - Ng,)/q(f1) for some constant a.
Remark 9.B.55. Actually, Davenport, Lewis and Schinzel prove a more general
The result follows. ]
resuli: let K/Q be a Galois extension of degree n and let vi,va,...,v, be
such that Og = Zv; + Zvg + -+ + Zu,. Suppose that f € Q[X] has the Remark 9.B.57. It is much easier to check that the converse of the theorem
property that any arithmetic progression of integers contains an integer z for alse holds.
which the equation f(z) = Nyp(uivr + wgvg + -+ + wnvy) is solvable in
(ug,u2, ..., us) € Q. If either Gal(K/Q) is a cyclic group or the multiplicity YTheorem 9.B.58. (Murty) Let f € Z{X] be a polynomial and let L, n be
of any zero of f is prime to n, then one can find 41, u, ...,up € Q[X] such relatively prime positive integers. Let S be the set of primes p for which the
that congruence f{z) =0 (mod p) has solutions in Z. Suppose that there exists ¢
such that any p € S greater than ¢ satisfies p=1 (mod n) or p =1 (mod n).
Also, suppose that there are infinitely many primes p =1 (mod 1) in S. Then
The proof follows precisely the same ideas, but is a bit more technical.
=1 {mod n).
¥ Theorem 9.B.56. (Schinzel) Let f € Z|X] be an irreducible polynomial over
Q and let n be an integer greater than 1. Suppose that there is ¢ with the Proof. Let 6 be a root of f and let K = Q{f) and L = K((,), so L/K is a
Jollowing property: if p > c divides f(x) for some @ € Z, then p =1 (mod n). finite Galois extension (splitting field of X™ — 1). Let H be the subgroup of
Then there are a € Q and a polynomial g € Q(¢,)[X] such that Gal(Q(¢n)/Q) generated by the automorphism o7, which sends ¢, to ¢4, Note
that the hypothesis combined with Bauer’s theorem and with remark 9.B.50
F(X) = a- Ny, mle(X)). yield an inclusion M C K, where M = Q((,)}”. Here is the crucial technical
result:
Proof. Let 6 be a root of f an let K = Q{f). We claim that Q(¢,) < K. By
Bauer’s theorem, it is enough to prove that if p is a large prime having a factor Lemma 9.B.59. Restriction to Q((n) yields an isomorphism between
of degree 1 in K, then p is completely split in Q(¢,). But if p is such a prime, Gal(L/K) and H.