algebraic number theory notes

Upload: fearlesslies

Post on 08-Jul-2018

224 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/19/2019 Algebraic Number Theory Notes

    1/173

    NOTES ON ALGEBRAIC NUMBER THEORY

    JEFFREY D. VAALER*

    1.   Algebraic number fields

    Let  Q  be the field of rational numbers, and let  Q  denote an algebraic closureof  Q. The elements of  Q   are called   algebraic numbers , and  Q   is the field of allalgebraic numbers. If  α   is an element of  Q  then  α   is algebraic over  Q. Thereforeα   is the root of a unique, monic, irreducible polynomial  mα(x) in the ring  Q[x].The polynomial mα(x) is called the minimal polynomial  of  α  over Q, (see TheoremB.4). If the minimal polynomial mα(x) belongs to the subring Z[x] ⊆ Q[x], then wesay that α  is an  algebraic integer . It is trivial that the elements of  Z  are algebraicintegers, and we sometimes refer to the elements of  Z  as  rational integers .

    Let   k   be a subfield of  Q, so that  Q   ⊆   k   ⊆  Q. We say that   k   is an   algebraic number field   if the degree [k  :  Q] of the extension  k/Q  is finite. Alternatively, if  kis a field of characteristic zero then  k   is an extension of its prime field  Q, and wesay that  k   is an  algebraic number field  if the degree [k  : Q] of the extension  k/Q  isfinite. Then an algebraic closure of  k  is also an algebraic closure of  Q  and we haveQ ⊆  k  ⊆  Q. As an algebraic number field  k  has characteristic zero,  k /Q  is a finiteseparable extension. Then it follows from Theorem H.9  that an algebraic numberfield k  is a simple  extension of  Q. That is, there exists an element α  in  k  such thatk = Q(α), and  α  is said to   generate   the field extension  k/Q.

    Yet another point of view is to begin with a field  k  such that  Q ⊆ k  ⊆ C. If thedegree [k : Q] of the extension k/Q is finite then k  is an algebraic number field, andalso a subset of the complex numbers. As  C  is algebraically closed, the set

    Q = {γ  ∈ C : γ   is algebraic over  Q}

    is an algebraic closure of Q (see Corollary C.4) and obviously contains k. Thereforewe have

    (1.1)   Q ⊆ k  ⊆ Q ⊆ C.

    As an arbitrary finite extension of  Q   can always be embedded into a specific al-gebraic closure of  Q  (see Theorem C.5), the containment (1.1) can always be ar-ranged. It is sometimes useful to work with an algebraic number field embeddedin C  because the complete metric topology and the subfield  R ⊆ C  can be used toadvantage.

    Suppose, for example, that [k : Q] = N . Then there exist N  distinct embeddingsσn   :   k   →  C, where   n   = 1, 2, . . . , N  . If  σn(k)   ⊆  R  we call   σn   a   real  embedding,otherwise we call  σn   a   complex   embedding. Let z   →  ρ(z) denote complex conju-gation on elements   z   in  C. Because complex conjugation is an automorphism of C   that fixes  R, it follows that if   α   →   σn(α) is a complex embedding of   k   then

    Date : 656, February 2, 2016.*Research supported by the National Science Foundation, DMS-06-03282.

    1

  • 8/19/2019 Algebraic Number Theory Notes

    2/173

    2 VAALER

    α  →  ρ

    σn(α)

      is also a complex embedding of  k  and is distinct from  α  → σn(α).Thus the complex embeddings of   k   into  C   come in pairs. Let  rk   be the numberof real embeddings and let 2s

    k  be the number of complex embeddings of   k   into

    C. Obviously we have   rk  + 2sk   =   N . The numbers   rk   and   sk   are examples of invariants  attached to the number field  k. If all the embeddings of  k   into  C  arereal embeddings, that is, if   rk   =  N   and 2sk  = 0, then we say that  k   is a  totally real  algebraic number field. At the other extreme, if all the embeddings of  k   intoC  are complex embeddings, that is, if  rk  = 0 and 2sk  =  N , then we say that  k  is atotally complex  algebraic number field. Further general results about algebraic fieldextensions are established in Appendix B.

    A basic object of investigation in elementary number theory is the integral do-main Z  of rational integers in  Q. In the theory of algebraic number fields we makea similar investigation of an analogous integral domain contained in an algebraicnumber field  k. The analogous integral domain is the set of all algebraic integersin  k , which we denote by  Ok. In view of our previous remarks, we have

    Ok  = α ∈  k  :  mα(x) ∈ Z[x].Our first objective is to show that  Ok   is a Noetherian domain, that is,   Ok   is anintegral domain and each ideal in  Ok   is finitely generated. Later we will prove astronger result: each integral domain Ok  is a Dedekind domain.

    Theorem 1.1.   Let  α   be an algebraic number. Then the following are equivalent,

    (i)   the minimal polynomial  mα(x)  belongs to Z[x],(ii)  there exists a monic polynomial  f (x)  in  Z[x]  such that  f (α) = 0,

    (iii)   the ring  Z[α]   is a finitely generated  Z-module.

    Proof.  It is trivial that (i) implies (ii).Assume that (ii) holds with deg f   =  M . As  f   is monic we have  M   ≥   1. It is

    obvious that the Z-module generated by

    (1.2)   {1, α , α2, . . . , αM −1}

    is a subset of  Z[α]. Let

    β  =  b0αN  + b1α

    N −1 + b2αN −2 + · · · + bN 

    be an element of  Z[α], where   b0, b1, . . . , bN   are integers and   b0   = 0. From thedivision algorithm in  Z[x] we have

    b0xN  + b1x

    N −1 + b2xN −2 + · · · + bN  = q (x)f (x) + r(x),

    with  q (x) and r(x) in Z[x], and either 0 ≤  deg r < M   or r(x) = 0. It follows that

    β  =  q (α)f (α) + r(α) =  r(α),

    and therefore   β  belongs to the  Z-module generated by the finite set (1.2). Thisverifies (iii).

    Assume that (iii) holds, and let  g1(x), g2(x), . . . , gL(x) be a finite collection of nonzero polynomials in Z[x] such that

    g1(α), g2(α), . . . , gL(α)

    generates Z[α] as a Z-module. Let N  be an integer larger than the maximum of thedegrees of the polynomials  g1(x), g2(x), . . . , gL(x). Then αN  belongs to  Z[α] andtherefore there exist integers  c1, c2, . . . , cL  such that

    αN  = c1g1(α) + c2g2(α) + · · · + cLgL(α).

  • 8/19/2019 Algebraic Number Theory Notes

    3/173

    ALGEBRAIC NUMBER THEORY 3

    it follows that α  is a root of the monic polynomial

    xN  − c1g1(x) − c2g2(x) − · · · − cLgL(x)

    which clearly belongs to Z[x]. This establishes (ii).Now assume that (ii) holds. Let f (x) be a monic polynomial in  Z[x] such that

    f (α) = 0, and let mα(x) be the minimal polynomial for α  in Q[x]. By Theorem B.4the principal ideal in  Q[x] generated by  mα(x) must contain the polynomial  f (x).Therefore there exists a monic polynomial  g (x) in Q[x] such that

    f (x) =  mα(x)g(x).

    Let d be a positive integer such that dmα(x) has relatively prime integer coefficients.Let e  be a positive integer such that  eg(x) has relatively prime integer coefficients.Then the identity

    def (x) =  {dmα(x)}{eg(x)}

    holds in  Z[x]. Now suppose that  p  is a prime such that  p|d, and let  F p  denote the

    finite field with  p  elements. Then the image of  def (x) in  F p[x] is the zero polyno-mial. However, both  dmα(x) and  eg(x) have relatively prime integer coefficients,and so the image of both  dmα(x) and  eg(x) in  F p(x) is a nonzero polynomial. AsF p[x] is an integral domain this is clearly impossible. It follows that   d  = 1 andtherefore mα(x) has integer coefficients. This establishes (i).  

    Theorem 1.2.   Let  α1, α2, . . . , αN  be algebraic integers, then  Z[α1, α2, . . . , αN ], the ring of all polynomials in  α1, α2, . . . , αN  with coefficients in Z, is a finitely generated Z-module.

    Proof.   We argue by induction on   N , the case   N   = 1 clearly follows from theprevious theorem. Assume then that Z[α1, α2, . . . , αN ] is a finitely generated  Z-module and consider the  Z-module  Z[α1, α2, . . . , αN , β ], where   β   is an algebraicinteger. Let f 1(x), f 2(x), . . . , f  L(x)be a finite collection of polynomials in Z[x1, x2, . . . , xN ], such that

    f 1(α), f 2(α), . . . , f  L(α)

    is a set of generators for  Z[α1, α2, . . . , αN ]. Let  g1(y), g2(y), . . . , gM (y) be polyno-mials in Z[y] such that

    g1(β ), g2(β ), . . . , gM (β )

    is a set of generators for  Z[β ]. We will show that the finite set

    (1.3)

    f l(α)gm(β ) : 1 ≤  l  ≤  L,  and 1 ≤  m  ≤  M 

    is a set of generators for  Z[α1, α2, . . . , αN , β ].Let

    (1.4) (α)pβ q = α p11   α p22   · · · α

     pN N   β 

    q

    be a monomial in  α1, α2, . . . , αN , β . Then there exist integers  a1, a2, . . . , aL, andintegers  b1, b2, . . . , bM , such that

    (α)p = a1f 1(α) + a2f 2(α) + · · · + aLf L(α),

    and

    β q = b1g1(β ) + b2g2(β ) + · · · + bM g(β ).

  • 8/19/2019 Algebraic Number Theory Notes

    4/173

    4 VAALER

    Now it is obvious that the monomial (1.4) belongs to the  Z-module generated bythe set (1.3). As each element of the  Z-module  Z[α1, α2, . . . , αN , β ] is a finitesum of such monomials with integer coefficients, it follows that (1.3) generatesZ[α1, α2, . . . , αN , β ]. In particular, the  Z-module  Z[α1, α2, . . . , αN , β ] is finitelygenerated. The result follows by setting β  =  αN +1.  

    Corollary 1.3.  The subset of algebraic integers in  Q  forms an integral domain. If k   is an algebraic number field, then the set  Ok   of algebraic integers in  k   forms an integral domain.

    Proof.   Let α  and  β  be algebraic integers in  Q. Then Z[α, β ] is finitely generated asa  Z-module, and therefore each of the submodules  Z[α + β ],  Z[α − β ], and  Z[αβ ]is a finitely generated  Z-module. Then it follows from Theorem  1.1  that   α +  β ,α −  β , and   αβ   are algebraic integers. Both assertions of the Corollary are nowobvious.  

    If [k : Q] = N   then k  is a vector space of degree  N  over the field Q. Hence thereexists a basis  α1, α2, . . . , αN  of elements from  k , such that

    (1.5)   k =

    b1α1 + b2α2 + · · · + bN αN   : bn ∈ Q  for each  n  = 1, 2, . . . , N  

    .

    Moreover, each element  β   in  k  has a unique  representation of the form

    (1.6)   β  =  c1α1 + c2α2 + · · · + cN αN    with   cn ∈ Q.

    If  α  is an algebraic number and

    mα(x) =  xN  + a1x

    N −1 + a2xN −2 + · · · + aN −1x + aN 

    is its minimal polynomial in Q[x], we may select a positive integer d  so that dmα(x)has relatively prime integer coefficients, as in our proof of Theorem  1.1.  If we writeeach rational coefficient as

    an =  rn

    sn,

    where rn  and  sn   are relatively prime integers and  sn   is positive, then it is easy tosee that

    d = lcm{s1, s2, . . . , sN }.

    It follows that  dα  is a root of the monic polynomial

    xN  + (a1d)xN −1 + (a2d

    2)xN −2 + · · · + (aN −1dN −1)x + (aN d

    N ),

    and this polynomial has integer coefficients. That is, dα  is an algebraic integer. Asα  can now be expressed as a ratio of two algebraic integers:

    (1.7)   α = dα

    d  ,

    it follows easily that  k  is the field of fractions attached to the integral domain  Ok.

    Let  α1, α2, . . . , αN   be a basis for  k/Q  as in (1.5). Then we can select positiveintegers  d1, d2, . . . , dN   so that  dnαn  is in Ok  for each n  = 1, 2, . . . , N  . We find that

    {d1α1, d2α2, . . . , dN αN }

    is a collection of  Z-linearly independent points in  Ok. More generally, let  I  be anonzero ideal in  Ok  and let  β  be a nonzero element of  I. Then  dnαnβ  belongs toI  for each  n  = 1, 2, . . . , N  , and it follows that

    {d1α1β, d2α2β , . . . , dN αN β }

  • 8/19/2019 Algebraic Number Theory Notes

    5/173

    ALGEBRAIC NUMBER THEORY 5

    is a collection of  Z-linearly independent points in  I. In particular we have

    (1.8) spanZ{d1α1β, d2α2β , . . . , dN αN β } ⊆ I.

    In order to show that equality can be achieved in ( 1.8), we introduce the discrimi-nant.

    Using (L.15) we find that

    (1.9) Normk/Q :  k× → Q×

    is a homomorphism of multiplicative groups, and

    (1.10) Tracek/Q  :  k  → Q

    is a nonzero linear map from the  Q-vector space   k   into  Q. If   α   is an algebraicinteger in  k  with minimal polynomial mα(x) in  Z[x], then using (L.12) and (L.13)we find that both Normk/Q(α) and Tracek/Q(α) are rational integers. It follows that

    if  α1, α2, . . . , αN   are algebraic integers in  Ok, then their discriminant (see (L.17))

    ∆k/Q(α1, α2, . . . , αN ) = det

    Tracek/Q(αmαn)

    is a rational integer.

    Theorem 1.4.   Let  k  be an algebraic number field with  [k : Q] =  N , and let  I  be a nonzero ideal in  Ok. Then there exists a  Z-linearly independent subset 

    (1.11)   {β 1, β 2, . . . , β  N } ⊆ I,

    that generates  I  as a  Z-module. That is, every element  γ   in  I  has a unique repre-sentation of the form 

    (1.12)   γ  =  b1β 1 + b2β 2 + · · · + bN β N 

    with each  bn   in  Z.

    Proof.   Let A  be the collection of all (column) vectors  α  such that the coordinatesα1, α2, . . . , αN   are Z-linearly independent elements of the ideal  I. In view of (1.8)it is clear that A   is not empty. It follows from Theorem  L.4  that for each vector  αin A  the discriminant ∆k/Q(α) is a nonzero integer. Thus we may select a point  βin  A  such that

    (1.13)∆k/Q(β) = min∆k/Q(α) : α  ∈ A.

    We claim that the coordinates of  β  form a  Z-module basis for  I.As the coordinates β 1, β 2, . . . , β  N  are Q-linearly independent elements of  k, they

    form a basis for k  as a vector space over  Q. Thus every element  γ  in  I  has a unique

    representation as

    (1.14)   γ  =  b1β 1 + b2β 2 + · · · + bN β N 

    with each  bn   in  Q. Assume that γ   in  I  can be selected so that at least one of thecoefficients  bn   is in  Q  but not in  Z. By reordering the elements  β 1, β 2, . . . , β  N   if necessary, we may assume that  b1  is in Q  but not in Z. Then by replacing  γ  with

    γ  − c1β 1

  • 8/19/2019 Algebraic Number Theory Notes

    6/173

    6 VAALER

    if necessary, where  c1  is an integer close to  b1, we may assume that  −12   ≤ b1  <

      12 .

    Now observe that we have the matrix identity

    (1.15)

    γ 

    β 2β 3...

    β N 

    =

    b1   b2   b3   . . . bN 0 1 0   . . .   00 0 1   . . .   0...

    ......

      . . .  ...

    0 0 0   . . .   1

    β 1β 2β 3...

    β N 

    .It follows from (1.15) and (L.21) that

    ∆k/Q(γ, β 2, β 3, . . . , β  N ) = (b1)2∆k/Q(β 1, β 2, . . . , β  N ),

    and therefore

    (1.16) 1 ≤∆k/Q(γ, β 2, β 3, . . . , β  N )

    ≤   14∆k/Q(β 1, β 2, . . . , β  N )

    .

    As  γ  belongs to the ideal  I, (1.16) contradicts the minimal property (1.13) associ-

    ated to the elements β 1, β 2, . . . , β  N . Thus every element γ  in I has a representationof the form (1.14) with coefficient bn   in  Z. This proves the theorem.  

    The collection of elements (1.11), asserted to exist in Theorem 1.4, is called anintegral basis  for the nonzero ideal I in the integral domain Ok. By an integral basis for the field  k  we understand an integral basis for the ring of integers  Ok. Plainlyan integral basis for  k  is also a basis for  k  as a vector space over  Q.

    Recall (see Theorem J.1) that an integral domain in which every ideal is finitelygenerated is a Noetherian domain. Thus Theorem 1.4   implies the following result.

    Corollary 1.5.   Let   k   be an algebraic number field and   Ok   the ring of algebraic integers in  k . Then  Ok   is a Noetherian domain.

    In our proof of Theorem 1.4  we selected  β 1, β 2, . . . , β  N  in the nonzero ideal  I  soas to minimize the expression∆k/Q(β 1, β 2, . . . , β  N ) = detTracek/Q(β mβ n).We now reconsider this number.

    Lemma 1.6.   Let   k   be an algebraic number field with   [k   :   Q] =   N , and let   Ibe a nonzero ideal in   Ok. Let   α1, α2, . . . , αN   be a  Z-module basis for   I   and let β 1, β 2, . . . , β  N  be a second  Z-module basis for  I. Then we have 

    (1.17) ∆k/Q(α1, α2, . . . , αN ) = ∆k/Q(β 1, β 2, . . . , β  N ).

    Proof.   Let σ1, σ2, . . . , σN  be the distinct embeddings of  k   into an algebraic closureQ. Then define  N  × N  matrices

    (1.18)   A = σn(αm)   and   B  = σn(β m),where in both matrices  m  = 1, 2, . . . , N    indexes rows and  n  = 1, 2, . . . , N    indexescolumns. From (L.16) and (L.17) we have

    (1.19) ∆k/Q(α1, α2, . . . , αN ) = det

    AAT 

    and

    (1.20) ∆k/Q(β 1, β 2, . . . , β  N ) = det

    BBT 

    .

  • 8/19/2019 Algebraic Number Theory Notes

    7/173

    ALGEBRAIC NUMBER THEORY 7

    By hypothesis there exists an  N ×N  matrix C  = (cmn) with rational integer entriessuch that

    (1.21)   β l =

    N m=1

    clmαm   for   l = 1, 2, . . . , N .

    Applying the embedding σn to both sides of (1.19) leads to the system of identities

    σn(β l) =

    N m=1

    clmσn(αm) for   l = 1, 2, . . . , N    and n  = 1, 2, . . . , N .

    These identities can also be written as the matrix identity

    (1.22)   B  =  CA.

    In a similar manner we find that there exists an  N  × N   matrix  D  = (dmn) withrational integer entries that satisfies the identity

    (1.23)   A =  DB .

    It follows from (1.22) and (1.23) that  CD  =  DC   =  1N , where   1N   is the  N  × N identity matrix. As C  and D  have integer entires we conclude that

    (1.24) det C  = det D =  ±1.

    Now (1.17) follows immediately from (1.19), (1.20) and (1.24).  

    If  I  is a nonzero ideal in the ring  Ok, and  β 1, β 2, . . . , β  N  is an integral basis forI, we define the  discriminant of the ideal  I  to be the nonzero integer

    (1.25) Disck/Q(I) = ∆k/Q(β 1, β 2, . . . , β  N ).

    Then it follows from Lemma 1.6 that the discriminant of the nonzero ideal  I  is welldefined because it does not depend on the choice of an integral basis for  I. Thediscriminant of the improper ideal  Ok  is also called the  discriminant of the field  k ,

    and in this special case we simplify notation by writingDisc(k) = Disck/Q(Ok)

    for the discriminant of  k. The discriminant of  k  is a further example of an invariantattached to the number field  k .

    Lemma 1.7.   Let  k  be an algebraic number field, and let  I  be a nonzero ideal in Ok. Then there exists an integral basis  α1, α2, . . . , αN , for  Ok, and positive integers d1, d2, . . . , dN , such that 

    (i)   dn|dn+1   for  n = 1, 2, . . . , N   − 1, and (ii)   d1α1, d2α2, . . . , dN αN   is an integral basis for  I.

    Proof.   Let α1, α2, . . . , α

    N  be an integral basis for  Ok, and let  β 

    1, β 

    2, . . . , β  

    N  be an

    integral basis for the ideal  I. Write σ1, σ2, . . . , σN   for the  N   distinct embeddings

    σn :  k  → Q. Then letA =

    σn(α

    m)

      and   B =

    σn(β m)

    be the corresponding N  × N  matrices, where in both cases m  = 1, 2, . . . , N    indexesrows and  n  = 1, 2, . . . , N    indexes columns. As in our proof of Lemma 1.6,  thereexists a nonsingular  N  × N  matrix  C  = (cmn) with integer entries such that

    (1.26)   B = CA.

    Now, however, we cannot conclude that  C  is a matrix in  GL(N, Z).

  • 8/19/2019 Algebraic Number Theory Notes

    8/173

    8 VAALER

    Let U  =

    umn

     be a matrix in the group  GL(N, Z), so that

    αm =

    N n=1

    umnαn   for   m = 1, 2, . . . , N ,

    is a second integral basis for Ok. Similarly, let V   =

    vmn

     be a matrix in the groupGL(N, Z), so that

    β m =N n=1

    vmnβ n   for   m = 1, 2, . . . , N ,

    is a second integral basis for the ideal  I. We find that

    A =

    σn(αm)

    = U A,

    and

    B  = σn(β m) = V B.

    Therefore we get the matrix identity

    (1.27)   B =  V B =

    V CU −1

    U A

    =

    V CU −1

    A.

    By a basic diagonalization theorem for matrices with integer entries (see [4, Theo-rem 4.3, Chapter 12] or [21, Theorem 4, Chapter 10]) we can select the matrices  U and V   so that

    (1.28)   V CU −1 = [dn]

    is an   N  × N   diagonal matrix, each diagonal entry   dn   is a positive integer, anddn|dn+1   for   n   = 1, 2, . . . , N    −  1. Here (1.28)   is the   Smith normal form   of   C .Combining (1.27) and (1.28) leads to the identity

    B = [dn]A,

    which also gives the conclusion (ii). And we have already noted that the positiveintegers  dn   satisfy (i).  

    Theorem 1.8.   Let  k  be an algebraic number field, and let  I  be a nonzero ideal in Ok. Then the index  [Ok  : I]   is finite, and we have 

    (1.29) Disck/Q(I) = [Ok  : I]2 Disc(k).

    Proof.   By Lemma   1.7   there exists an integral basis   α1, α2, . . . , αN   for   Ok, andpositive integers d1, d2, . . . , dN , such that  d1α1, d2α2, . . . , dN αN  is an integral basisfor the ideal  I. Then every element  γ   in  Ok  has a unique representation as

    (1.30)   γ  =  b1α1 + b2α2 + · · · + bN αN ,   where   bn ∈ Z.

    Define an additive group homomorphism

    ψ :  Ok  → Z/d1Z   ⊕   Z/d2Z   ⊕ · · · ⊕   Z/dN Z

    by

    ψ(γ ) =  ψ

    b1α1 + b2α2 + · · · + bN αN ) = (b1, b2, . . . , bN ).

    Obviously the map  ψ   is surjective. An element  γ  given by (1.30) is in the kernel of ψ  if and only if 

    bn ≡  0 (mod  dn) for each   n = 1, 2, . . . , N .

  • 8/19/2019 Algebraic Number Theory Notes

    9/173

    ALGEBRAIC NUMBER THEORY 9

    That is,  γ   is in the kernel of  ψ   if and only if  γ   belongs to the ideal  I. It followsthat

    (1.31) [Ok  : I] = d1d2 · · · dN  = det[dn].As in the proof of Lemma 1.7, we have

    B =

    σn(β m)

    = [dn]

    σn(αm)

    = [dn]A,

    and therefore

    Disck/Q(I) = det

    BBT 

    =

    det[dn]2

    det

    AAT )

    = (d1d2 · · · dN )2 Disc(k).

    (1.32)

    The result follows from (1.31) and (1.32).  

    Let k  is an algebraic number field and let  α1, α2, . . . , αN  be an integral basis for

    the ring of algebraic integers  Ok. If  β  is a nonzero element of  Ok, then we can formthe nonzero principal ideal

    β  =  βOk

    =

    β (c1α1 + c2α2 + · · · + cN αN ) :  cn ∈ Z

    = spanZ{βα1, βα2, . . . , β αN }.

    As  β α1, βα2, . . . , β αN  is clearly an integral basis for the ideal  β , we find that

    Disck/Q

    β 

    = det

    Tracek/Q(β 2αmαn)

    =   N l=1

    σl(β 2)

    det

    Tracek/Q(αmαn)

    = Normk/Q(β )2 Disck/Q(α1, α2, . . . , αN )

    = Normk/Q(β )2 Disc(k),

    (1.33)

    where  σ1, σ2, . . . , σN  are the distinct embeddings of  k   into  Q. Now Theorem 1.8and (1.33) immediately imply the following result.

    Corollary 1.9.   Let  k  be an algebraic number field and  β  a nonzero element of  Ok.Then the discriminant and index of the nonzero principal ideal  β   satisfy 

    (1.34) Disck/Q

    β 

    = Normk/Q(β )2 Disc(k),

    and 

    (1.35)

    Ok  :  β 

    =Normk/Q(β ).

    Moreover, β  is a unit in  Ok  if and only if 

    (1.36) Normk/Q(β ) =  ±1.

    We recall that a maximal ideal in a commutative ring with an identity element,is also a prime ideal. But in a general commutative ring a prime ideal need not bemaximal. In the ring Ok  of algebraic integers in a number field  k, we find that eachnonzero prime ideal is also a maximal ideal.

    Corollary 1.10.   Let  k  be an algebraic number field and  P   a nonzero prime ideal in  Ok. Then  P  is a maximal ideal.

  • 8/19/2019 Algebraic Number Theory Notes

    10/173

    10 VAALER

    Proof.   As P is a nonzero prime ideal, it follows from Theorem  I.2 that the quotientring   Ok/P   is an integral domain. By Theorem  1.8  the integral domain  Ok/P   isfinite, but it is not the zero ring because P is a proper subset of  O

    k. From Theorem

    I.1  we learn that   Ok/P  is a field, and then a second application of Theorem  I.2shows that  P  is a maximal ideal.  

    Let (S, +, ×) be a commutative ring with an identity element and let (R, +, ×)be a subring such that 1R  = 1S . We recall that an element  s  in  S   is  integral  overR  if there exists a monic polynomial

    f (x) =  xN  + a1xN −1 + a2x

    N −2 + · · · + aN −1x + aN 

    in  R[x] such that  f (s) = 0. Clearly each element of  R   is integral over  R. We saythat S   is  integral  over  R  if every element  s  in  S   is integral over  R. We say that  Ris  integrally closed  in  S  if the only elements of  S  which are integral over  R  are theelements in  R. For example, if  Ok  is the ring of algebraic integers in the algebraicnumber field   k, then   Ok   is integral over  Z. The following result shows that   Okis integrally closed in the field  k, and completes the proof that  Ok   is a Dedekinddomain.

    Theorem 1.11.   Let   k  be an algebraic number field and   Ok   the ring of algebraic integers in  k . Then  Ok  is a Dedekind domain.

    Proof.   The integral domain  Ok  is Noetherian by Corollary  1.5.   And each nonzeroprime ideal  P   in  Ok  was shown to be a maximal ideal in Corollary  1.10. Thus itremains only to show that Ok  is integrally closed in its field of fractions  k .

    Let α  be an element of  k  which is integral over  Ok. As  Ok   is obviously integralover  Z, it follows from Corollary  I.9 that  α   is integral over  Z. That is,  α  belongsto Ok. This shows that Ok  is integrally closed in its field of fractions  k . Hence Okis a Dedekind domain.  

    Exercises

    1.1.  Recall that an integer  m = 0 is   square free   if  m   is not divisible in  Z  by thesquare of a prime number. Thus the set of square free integers is

    · · · − 10, −7, −6, −5, −3, −2, −1, 1, 2, 3, 5, 6, 7, 10, . . .

    .

    Let k  be an algebraic number field such that [k  : Q] = 2. Prove that there exists analgebraic integer α  in  k  and a unique square free integer  m  = 1, such that  α2 = mand k  = Q(α).

    1.2.   Let   k   =  Q(α) and   α2 =   m   be as in Exercise   1.1. Let   Ok   be the ring of algebraic integers in  k . Prove that

    Ok  = spanZ{1, α},   if  m  ≡ 1 (mod 4),

    and

    Ok  = spanZ

    1,

     1 + α

    2

    ,   if  m  ≡  1 (mod 4).

  • 8/19/2019 Algebraic Number Theory Notes

    11/173

    ALGEBRAIC NUMBER THEORY 11

    1.3.   Let k  = Q(α) and α2 = m  be as in Exercise 1.1.  Prove that

    Disc(k) = 4m   if  m  ≡ 1 (mod 4),m   if  m  ≡  1 (mod 4).1.4.   Let k  be an algebraic number field and  Ok  the integral domain of algebraic

    integers in k. Use the representation (1.7) to show that  k  is the field of fractions of the integral domain Ok.

    1.5.   Let k  be an algebraic number field and Disc(k) its discriminant. Prove that

    (−1)s Disc(k) ≥  1,

    where   r   is the number of real embeddings of   k   into  C, and 2s   is the number of complex (and not real) embeddings of  k   into C.

    1.6.   Let k  be an algebraic number field and Disc(k) its discriminant. Prove thateitherDisc(k) ≡  0 mod 4,   or Disc(k) ≡  1 mod 4.

    1.7.   Let   k   and   l   be algebraic number fields such that  Q   ⊆   k   ⊆   l. Prove thatDisc(k)| Disc(l).

    1.8.   Let α  in  Q  be a root of the polynomial

    mα(x) =  x3 − x2 − 2x − 8.

    Show that mα(x) is in fact irreducible in  Q[x], and show that

    (1.37)   β  = α2 + α

    2

    is an algebraic integer in the field  k  = Q(α). (Hint: find  mβ(x) and observe that itbelongs to Z[x].)

    1.9.   Let α,  β , and k  be as in Exercise  1.8. Show that {1, α , β  } is an integral basisfor Ok, and conclude that

    Disc(k) =  −503.

    1.10.   Let α,  β , and k  be as in Exercise 1.8  and Exercise 1.9. Prove that if  γ  is analgebraic integer in  Ok  then

    Disck/Q

    γ 

    ≡ 0 mod 4.

    Conclude that  Ok   does  not  have an integral basis of the form  {1, γ , γ  2}.

  • 8/19/2019 Algebraic Number Theory Notes

    12/173

    12 VAALER

    2.   Theorems of Blichfeldt and Minkowski

    In this section we prove some classical results from the geometry of numbers.

    We apply these results to further our understanding of ideals in the ring   Ok   of algebraic integers in an algebraic number field  k. As this subject has many furtherapplications, we take a somewhat general point of view.

    We will work in the  N  dimensional real vector space  RN , and write

    x =

    x1x2...

    xN 

    ,for a (column) vector in RN . By a lattice  in RN  we understand an additive subgroupΛ  ⊆ RN  generated as a free  Z-module by  N   vectors  ξ1, ξ2, . . . ,ξN , which are  R-linearly independent. Thus the lattice Λ generated by ξ1, ξ2, . . . ,ξN , is the set

    (2.1) Λ = m1ξ1 + m2ξ2 + · · · + mN ξN   : m1, m2, . . . , mN   in Z.We say that the R-linearly independent vectors  ξ1, ξ2, . . . ,ξN , form a basis  for thelattice Λ. If  λ   is a point in Λ then it follows from the  R-linear independence of ξ1, ξ2, . . . ,ξN , that there exist  unique   integers  m1, m2, . . . , mN   such that

    λ =  m1ξ1 + m2ξ2 + · · · + mN ξN .

    Let

    (2.2)   e1  =

    10...0

    ,   e2  =

    01...0

    ,   · · ·  ,   eN  =

    00...1

    .

    denote the standard basis vectors in  RN . Plainly these vectors generate the latticeZN . More generally, write

    (2.3)   X  =ξ1   ξ2   ξ3   · · ·   ξN 

    for the N  × N   real, nonsingular matrix having  ξn  as its nth column. From (2.1) itis clear that

    (2.4) Λ =

    X m :  m ∈ ZN 

    .

    Thus the lattice Λ is the image of the lattice  ZN  under the map  m  → X m.Let GL(N, R) denote the general linear group of all nonsingular  N × N  matrices

    with entries in R. Write GL(N, Z) for the subgroup of  N × N  matrices with entriesin  Z  and determinant equal to  ±1. If AutZN  is the group of all  Z-module auto-morphisms of  ZN , then each matrix  A  in GL(N, Z) determines the automorphismϕA : ZN  → ZN  in Aut

    ZN 

     given by

    ϕA(m) =  Am.

    It is easy to verify that A  →  ϕA  is an isomorphism of the group GL(N, Z) onto thegroup Aut

    ZN 

    . In particular, we have

    ϕA

    ϕB(m)

    = ϕAB(m)

  • 8/19/2019 Algebraic Number Theory Notes

    13/173

    ALGEBRAIC NUMBER THEORY 13

    for all  A  and  B   in GL(N, Z), and all points  m  in  ZN . Alternatively, GL(N, Z) isexactly the set of  N  × N  matrices A  in GL(N, R) such that

    ZN 

    = {Am :  m ∈ ZN 

    }.If Λ is the lattice in  RN  determined by (2.4) and   ψ   : Λ   →   Λ is a  Z-module

    automorphism of Λ, then the map

    m → X m → ψ

    X m

    → X −1ψ

    X m

    is an automorphism of  ZN . It follows from our previous remarks that there existsa unique matrix  A  in GL(N, Z) such that

    X −1ψ

    X m

    = Am,

    and thenψ(λ) =  X AX −1λ

    for all points  λ  in Λ.

    Lemma 2.1.  Let   Λ   be the lattice in  R

    determined by   (2.4). Assume that the columns of a second matrix  Y   in  GL(N, R)  also form a basis for  Λ, so that 

    (2.5) Λ = {Y m :  m ∈ ZN }.

    Then there exists a unique matrix  B   in  GL(N, Z)  such that  Y   = X B.

    Proof.  The mapm → Y m → X −1Y m

    is an automorphism of ZN . Hence there exists a unique matrix B  in GL(N, Z) suchthat

    (2.6)   X −1Y m =  Bm

    for all  m   in  ZN . In particular, (2.6) holds for each standard basis vector, andtherefore Y   = X B.  

    It follows from Lemma 2.1  that there is a bijection between the collection of alllattices Λ  ⊆  RN  and the collection of all left cosets of the subgroup GL( N, Z) inGL(N, R). More precisely, if Λ is determined by the matrix X  as in (2.4), then theset of all matrices  Y   in RN  such that (2.5) holds, is precisely the left coset

    XB  :  B  ∈  GL(N, Z)

    .

    If  E  ⊆ RN  is a Borel subset, we write VolN (E ) for the N -dimensional Lebesguemeasure of  E , but we often refer to this quantity as the  N -dimensional volume  of E , or simply the  volume   of   E . Each matrix   A   in GL(N, R) determines a linearhomeomorphism x   →  Ax  from  RN  onto  RN . If  E   ⊆  RN  is a Borel set then theimage

    AE  =  {Ax :  x  ∈  E }

    is also a Borel set, and we have the basic identity

    (2.7) VolN (AE ) =  | det A| VolN (E ).

    If Λ is determined by (2.3) and (2.4), then the corresponding  fundamental par-allelepiped  is the subset

    (2.8)   P X  =

    θ1ξ1 + θ2ξ2 + · · · + θN ξN   : 0 ≤  θn <  1 for each  n  = 1, 2, . . . , N  

    .

    The subset  P X   ⊆ RN  is clearly a Borel set, and is a complete set of distinct cosetrepresentatives for the quotient group  RN /Λ. That is, if  x  is a point in  RN , then

  • 8/19/2019 Algebraic Number Theory Notes

    14/173

    14 VAALER

    there exists a unique point  λ  in Λ such that  x − λ  belongs to  P X . Thus  P X   is anexample of a   fundamental domain   for the quotient group  RN /Λ. We write

    (2.9)   C N  = θ ∈ RN  : 0  ≤  θn <  1for the fundamental parallelepiped associated to the standard basis for  ZN . ThenP X  is clearly the image of  C N  under the linear map  θ  → X θ. It follows from (2.7)that the volume of the fundamental parallelepiped  P X  is given by

    (2.10) VolN (P X) =  | det X | VolN (C N ) =  | det X |.

    If Λ is also determined by a second matrix  Y  as in (2.5), then  Y   = X B, where  Bis in GL(N, Z), and it is obvious that

    (2.11)   | det Y | =  | det X || det B| =  | det X |.

    The positive real number   | det X |   is called the   determinant of the lattice . Theidentity (2.11) shows that this number does not depend on the choice of basis, andit is equal to the volume of a fundamental parallelepiped for the lattice. We writed(Λ) = | det X |  for the determinant of the lattice Λ.

    A basic problem in the geometry of numbers is to identify subsets  K  ⊆ RN , andlattices Λ  ⊆  RN , for which we can draw conclusions about the nature of the setK  ∩ Λ. In many cases it is trivial that  K  ∩ Λ contains the point   0, and we seekresults which assert that   K  ∩  Λ either does or does not contain further nonzeropoints. For example, K  could be the set of points in  RN  that satisfy a system of inequalities, and we may wish to show that there are nonzero points in a certainlattice Λ that satisfy the system of inequalities. We begin to establish results of this sort by proving the following theorem of Blichfeldt [9].

    Theorem 2.2.   Let   E   ⊆  RN 

    be a Borel set,   Λ  ⊆  RN 

    a lattice, and   L   a positive integer. If 

    (2.12)   Ld(Λ) <  VolN (E ),

    then there exist   L + 1   distinct points   y0,y1, . . . ,yL, in   E   such that each of the differences  ym − yl  belongs to  Λ.

    Proof.  We may assume that the columns of the  N  × N  matrix  X  form a basis forΛ, as in (2.4). Let χE (x) denote the characteristic function of the set  E , and writeP X   for the fundamental parallelepiped (2.8). Then it follows that

    (2.13)

    λ∈ΛP X + λ

    = RN 

    is a disjoint union of sets. Next we define the Borel measurable function

    F   : RN  → {0, 1, 2, . . . , ∞}

    by

    F (x) =λ∈Λ

    χE (x + λ).

  • 8/19/2019 Algebraic Number Theory Notes

    15/173

    ALGEBRAIC NUMBER THEORY 15

    From (2.13) and the monotone convergence theorem we find that

     P X F (x) dx = λ∈Λ P X χE (x + λ) dx=λ∈Λ

     P X+λ

    χE (x) dx

    =

     RN 

    χE (x) dx

    = VolN (E ) > L VolN (P X).

    (2.14)

    It follows that there exists a point  p   in  P X   such that  F ( p) ≥  L  + 1. Hence thereexist L + 1 distinct points λ0,λ1, . . . ,λL, in Λ such that  y l  =  p + λl  belongs to E for each l  = 0, 1, 2, . . . , L. Then each of the differences  ym − yl =  λm −λl  belongsto Λ. This proves the theorem.  

    It is often convenient to have the following refinement of Theorem 2.2, in whichthe set E  is assumed to be compact.

    Corollary 2.3.   Let  K  ⊆ RN  be a compact set, Λ  ⊆ RN  a lattice, and  L  a positive integer. If 

    (2.15)   Ld(Λ) ≤  VolN (K ),

    then there exist   L + 1   distinct points   y0,y1, . . . ,yL   in   K   such that each of the differences  ym − yl  belongs to  Λ.

    Proof.   Let 1 ≥  r1  > r2  >  · · · > rj  > · · · >  0 be a monotone decreasing sequence of positive numbers such that

    limj→∞

    rj  = 0.

    Then let Bj  = {x ∈ RN  : |x|2  ≤  rj}

    denote the closed ball in  RN  centered at  0  and having radius  rj . Write

    E j  = Bj + K  =  {x+ y :  x  ∈  Bj   and  y  ∈  K }

    for the Minkowski sum of the two compact sets  Bj   and  K . It follows that each of the sets E j   is compact,

    (2.16)   E 1  ⊇  E 2  ⊇ · · · ⊇ E j  ⊇ · · · ⊇ K,

    and therefore we have a decreasing sequence of nonnegative integers

    (2.17)E 1 ∩ Λ

    ≥E 2 ∩ Λ

    ≥ · · · ≥E j ∩ Λ

    ≥ · · · ≥K  ∩ Λ

    .

    Hence there exists a positive integer J  such that j  → E j ∩ Λ is constant for j  ≥  J .As  K  is compact and(2.18)

    ∞j=J 

    E j  = K,

    we find that E j ∩ Λ =  K ∩ Λ for j  ≥  J . By the Brunn-Minkowski theorem (see [29,Theorem 3.2.41]) we have

    VolN (Bj  + K )1/N  ≥ VolN (Bj)

    1/N + VolN (K )1/N ,

  • 8/19/2019 Algebraic Number Theory Notes

    16/173

    16 VAALER

    and thereforeVolN (Bj + K ) >  VolN (K ) ≥  Ld(Λ).

    Thus we can apply Theorem  2.2  to the sets  E j  = Bj  +  K   for j  ≥  J . We concludethat for each  j  ≥  J   there exsit  L  + 1 distinct points  x

    (j)0   ,x

    (j)1   , . . . ,x

    (j)L   in  E j   such

    that each of the differences  x(j)m   − x

    (j)l   belongs to Λ.

    As there are finitely many infinite sequences x(j)l   and all points of these sequences

    are contained in the compact set  E J , it follows that there exists a subsequence

    J  ≤ j1  < j2  <  · · · < ji  <  · · ·

    such thatlimi→∞

    x(ji)l   = yl

    exists for each   l  = 0, 1, 2, . . . , L. In view of (2.18) we conclude that  yl  belongs to

    K   for each  l. Now each of the differences  x(ji)m   − x

    (ji)l   belongs to Λ. Because Λ is

    a discrete set, the limiting values

    limi→∞

    x(ji)m   − x(ji)

    l   = ym − yl

    also belong to Λ for each l  and m. This is exactly the statement of the corollary.  

    We say that a subset  K   ⊆ RN  is  symmetric   if the map  x → −x  sends  K   ontoitself. Alternatively, K  is symmetric if  K  =  −K . We say that K  is  convex  wheneverit has the following property: if  u and v  are points in K  then the point (1−θ)u+θvis in K  for all real numbers  θ  such that 0  ≤  θ  ≤  1. Thus K  is convex if for any pairof points  u  and  v   in  K , each point of the line segment from  u  to  v  is also in  K .

    The following fundamental theorem was proved by Minkowski [54] in case L  = 1.The more general result for  L  ≥  1 was first established by van der Corput.

    Theorem 2.4.   (Minkowski’s convex body Theorem)   Let   K   be a convex,symmetric subset of  RN , Λ  a lattice in  RN , and let  L  be a positive integer. Assume 

    that either 

    (2.19)   L2N d(Λ) <  VolN (K ),

    or that  K   is compact and 

    (2.20)   L2N d(Λ) ≤  VolN (K ).

    Then   K  contains at least   L  pairs of distinct, nonzero points   ±λ1, ±λ2, . . . , ±λL from the lattice  Λ.

    Proof.   Assume that   K   satisfies (2.19). Then the volume of   12 K   satisfies the in-equality

    Ld(Λ) <

    12

    N VolN (K ) = VolN 

    12

    .

    By Theorem 2.2  the set   12 K   contains L + 1 distinct points y0,y1, . . . ,yL  such that

    each of the differences  ym − yl   belongs to Λ. Because  1

    2 K  is symmetric, both  ymand −yl  belong to

      12

    K . Because   12 K  is convex, each of the differences

    12ym −

      12yl =

      12ym +

      12 (−yl)

    belongs to   12 K . Therefore each of the differences  ym − yl  belongs to  K , and so toK  ∩ Λ.

    Now consider the subspaces

    S (l, m) =w ∈ RN  : w(ym − yl) = 0

      where   l = m.

  • 8/19/2019 Algebraic Number Theory Notes

    17/173

    ALGEBRAIC NUMBER THEORY 17

    As RN  is not the union of these subspaces, there exists a point  z  in the complementof their union. Let  ϕ   :  RN  →  R   denote the linear form defined by   ϕ(x) =   zx.Then the restriction of  ϕ  to the subset y0,y1, . . . ,yL is injective. By reindexingif necessary, we may assume that

    ϕ(y0) < ϕ(y1) <  · · · < ϕ(yL).

    Set  λm   =  ym −  y0   for each  m  = 1, 2, . . . , L. Then the points  λ1,λ2, . . . ,λL   aredistinct, hence the points  −λ1, −λ2, . . . , −λL   are also distinct. And, an identity−λl  =  λm  is impossible because

    ϕ(−λl) =  −ϕ(λl) <  0  < ϕ(λm).

    This proves the theorem under the hypothesis (2.19). If   K   is compact andsatisfies the inequality (2.20), the argument is essentially the same, but in this casewe appeal to Corollary 2.3.  

    Corollary 2.5.   Let  Λ   be a lattice in  RN . Then there exists a point  λ  =  0   in  Λ

    such that (2.21)   |λ|∞ = max

    λ1, λ2, . . . , λN  ≤ d(Λ)1/N .Proof.  We apply the theorem to the compact, convex, symmetric set

    K  =x ∈ RN  : |x|∞ ≤  d(Λ)

    1/N 

    .

    The result follows because VolN (K ) = 2N d(λ).  

    Theorem 2.6.   (Minkowski’s Theorem on linear forms)  Let   Λ  be a lattice in  RN , and let 

    (2.22)   Lm(x) =N n=1

    amnxn,   where    m = 1, 2, . . . , N ,

    be a system of  N  linear forms in  N  variables with real coefficients. Write  A  = (amn) for the corresponding  N ×N  matrix, and let  η1, η2, . . . , ηN , be positive real numbers.If 

    (2.23)   d(Λ)| det A| ≤  η1η2 · · · ηN ,

    then there exists a point  λ = 0  in  Λ   such that 

    (2.24)Lm(λ) = N 

    n=1

    amnλn

    ≤ ηm   for each    m = 1, 2, . . . , N .Proof.   Write  H   = [ηn] for the  N  × N  diagonal matrix having  ηn   in the  nth rowand nth column. Then define the closed, convex, symmetric set

    K  =

    x ∈ RN  :

    H −1Ax

    ∞ ≤  1

    .

    If det A = 0 then  K  is compact, and we find thatVolN (K ) = 2

    η1η2 · · · ηN 

    | det A|−1 ≥ 2N d(Λ).

    If det A = 0 then the interior of  K  is unbounded, and in this case we have

    VolN (K ) =  ∞.

    In either case there exists a point  λ  = 0  such thatH −1Aλ∞

     ≤  1.

  • 8/19/2019 Algebraic Number Theory Notes

    18/173

    18 VAALER

    That is,

    η−1mN 

    n=1 amnλn ≤ 1 for each   m = 1, 2, . . . , N ,and this proves the theorem.  

    It is useful to have a variant of Theorem 2.6 in which the linear forms are allowedto have complex coefficients, and the norm   | |∞   is replaced by   | |1. Suppose thatamong the forms (2.22) there are exactly r  forms with real coefficients, there are 2sforms with complex coefficients, and further assume that the complex forms occur incomplex conjugate pairs. Obviously we have r +2s =  N . By reindexing if necessary,we may assume that the forms  Lm(x) have real coefficients for m  = 1, 2, . . . , r, thatthe forms Lm(x) have complex (but not real) coefficients for  m =  r +1, r+2, . . . , r+s, and that

    Ls+m(x) =

    n=1amnxn

    for m  =  r + 1, r + 2, . . . , r + s. Here z  denotes the complex conjugate of the complexnumber  z .

    Theorem 2.7.   Let  Lm(x), where  m  = 1, 2, . . . , N  , be a collection of  N  linear forms in   N   variables with complex coefficients. Assume that there are   r   real forms,   2scomplex (but not real) forms, and that the complex forms occur in complex conjugate pairs. Write   A  = (amn)   for the matrix of coefficients. Then there exists a point ξ = 0   in the integer lattice  ZN  such that 

    (2.25)N 

    m=1

    Lm(ξ) ≤ 4/πsN !| det A|1/N .Proof.  We may assume, as in our previous remarks, that the forms  Ll(x) are real

    for l  = 1, 2, . . . , r, and that  Ls+m(x) and  Lm(x) form a pair of complex conjugateforms for  m  =  r  + 1, r + 2, . . . , r + s. Let  W   denote the  N  × N   matrix organizedinto blocks as

    2W   =

    21r   0 00 1s   1s0   −i1s   i1s

    ,where   1r   and   1s   are   r  ×  r   and   s ×  s   identity matrices, respectively. It is easyto verify that det W   = (−2i)−s. Because of the way we have arranged the linearforms, the matrix  W A  has real entries.

    Next we define

    K  =y ∈ RN  :

    rl=1

    |yl| + 2s

    m=1

    y2m + y

    2s+m

    1/2≤ 1

    .

    It follows that  K   is a compact, convex, symmetric subset of  RN , andVolN (K ) = 2

    r−sπs

    N !−1

    .

    Hence the linear image

    (W A)−1K  =  {x ∈  K  :  W Ax ∈  K }

    is also a compact, convex, symmetric subset of  RN , and

    VolN 

    (W A)−1K 

    = | det W A|−1 VolN (K ) = 2rπs

    N !| det A|−1

    .

  • 8/19/2019 Algebraic Number Theory Notes

    19/173

    ALGEBRAIC NUMBER THEORY 19

    Now let  t  be a positive number. Then we have

    tK  = y ∈ RN  :r

    l=1 |yl| + 2s

    m=1y2m + y2s+m1/2 ≤ t,and

    VolN (tK ) =  tN VolN (K ).

    We select t  so that

    (2.26)   tN  =

    4/πs

    N !| det A|,

    and therefore

    VolN 

    (W A)−1tK 

    = 2N .

    By Minkowski’s Theorem 2.2, there exists a point  ξ =  0  in the integer lattice  ZN 

    such that W Aξ  belongs to  tK . From the definition of  W A  and  tK , we find that

    (2.27)N 

    n=1Ln(ξ) =r

    l=1Ll(ξ)+ 2r+s

    m=rLm(ξ) ≤ t.The result follows by combining (2.26) and (2.27).  

    Let

    P (x) =  a0xN  + a1x

    N −1 + a2xN −2 + · · · + aN −1x + aN 

    = a0

    N n=1

    (x − αn)

    = a0

    N n=0

    (−1)N −nen(α)xN −n

    (2.28)

    be a polynomial in  Z[x] with degree  N   ≥  1. Here  α1, α2, . . . , αN , are the roots of 

    P (x), which we regard as elements of  C. And we write

    e0(α) = 1,   and   en(α) =

    1≤i1

  • 8/19/2019 Algebraic Number Theory Notes

    20/173

    20 VAALER

    As an application of Theorem 2.7, we give a lower bound for the discriminant of an irreducible polynomial in Z[x].

    Theorem 2.8.   Let  P (x)  be an irreducible polynomial in  Z[x]   with degree  N   ≥  1.If  P (x)  has  r  real roots and  2s  complex (but not real) roots in  C, then 

    (2.30)

    N N 

    N !

    π4

    s2≤ | Disc(P )|.

    Proof.   Let A  denote the  N  × N   complex matrix

    (2.31)   A =

    1   α1   α

    21   . . . α

    N −11

    1   α2   α22   . . . α

    N −12

    ......

    ...  . . .

      ...

    1   αN    α2N    . . . α

    N −1N 

    .From the van der Monde determinant identity we have

    (2.32) det A = 1≤m

  • 8/19/2019 Algebraic Number Theory Notes

    21/173

    ALGEBRAIC NUMBER THEORY 21

    As each monomial in  R(x1, x2, . . . , xN ) has total degree at most  N  − 1, it followsthat

    (2.36)   aN −10   Re1(α), e2(α), . . . , eN (α) = aN −10 N m=1

    β m

    is a nonzero integer. Combining (2.32), (2.35), and (2.36), leads to the inequality

    1 ≤aN −10 N 

    m=1

    β m

    4/πs

    N −N N !aN −10   det A

    =

    4/πs

    N −N N !Disc(P )1/2.

    (2.37)

    Then (2.37) clearly implies (2.30).  

    Next we give an application of the geometry of numbers to algebraic number

    theory. In particular, we prove that a nonzero ideal in an algebraic number fieldcontains a nonzero element with small norm.

    Theorem 2.9.   Let  k  be an algebraic number field with  r  =  rk   real embeddings and 2s = 2sk   complex embeddings. Write   [k  :  Q] = r  + 2s =  N , and let  I  be a nonzeroideal in  Ok. Then there exists a nonzero point  γ   in  I  such that 

    (2.38)Normk/Q(γ ) ≤ 4πs N !N N Disck/Q(I) 12 .

    Proof.   Let β 1, β 2, . . . , β  N  be an integral basis for the nonzero ideal  I  in  Ok. Thenlet σ1, σ2, . . . , σN  denote the distinct embeddings σn :  k  → Q. Here we take Q ⊆ C,and we assume that   r   =   rk   embeddings are real and 2s   = 2sk   embeddings arecomplex. Let

    B = σn(β m)be the corresponding   N   ×  N   matrix, where   m   = 1, 2, . . . , N    indexes rows andn = 1, 2, . . . , N   indexes columns. Each column of  B  determines a linear form

    Ln(x) =N 

    m=1

    σn(β m)xm,   where   n = 1, 2, . . . , N .

    Clearly there are  r  real forms, 2s  complex forms, and the complex forms occur incomplex conjugate pairs. By Theorem   2.7   there exists a nonzero point   ξ   in theinteger lattice ZN  such that

    (2.39)N n=1

    Ln(ξ) ≤ 4/πsN !| det B|1/N .Let

    γ  =  ξ 1β 1 + ξ 2β 2 + · · · + ξ N β N ,so that  γ  is a nonzero point in  I, and

    (2.40)   Ln(ξ) =N 

    m=1

    σn(β m)ξ m =  σn(γ ).

    From (1.20) and (1.25) we have

    (2.41) det

    BBT 

    = Disck/Q(I).

  • 8/19/2019 Algebraic Number Theory Notes

    22/173

    22 VAALER

    By combining (2.39), (2.40), and (2.41), we find that

    (2.42) n=1σn(γ ) ≤ 4/πs

    N !Disck/Q(I)1

    21/N 

    .

    Then from the arithmetic-geometric mean inequality and (2.42) we conclude thatNormk/Q(γ ) = N n=1

    σn(γ )≤

    N −1N n=1

    σn(γ )N ≤ 4

    π

    s N !N N 

    Disck/Q(I) 12 .(2.43)

    This proves the theorem.  

    If  x  is a real number we write

    x = min{|x − n| :  n  ∈ Z}

    for the distance from   x   to the nearest integer. Clearly     :   R   →

    0,   12

      is acontinuous function with period 1. Hence it is well defined on the cosets of thequotient group  R/Z, and we my regard it as a function     :  R/Z  →

    0,  12

    . Wenote that (x, y)   → x −  y  defines a metric on  R/Z, and this metric induces itsquotient topology. The function  x  → x  plays a fundamental role in Diophantineapproximation. As a further application of Minkowski’s convex body theorem, wederive a general form of Dirichlet’s Theorem on rational approximation.

    Theorem 2.10.  (Dirichlet’s Theorem on linear forms)  Let  A = (αmn)  be an  M × N  matrix with real entries, and let  Q  be a positive integer. Then there exist integers  q 1, q 2, . . . , q  N , not all zero, such that 

    (2.44)   |q n| ≤  Q   for each    n = 1, 2, . . . , N ,

    and 

    (2.45)   αm1q 1 + αm2q 2 + · · · + αmN q N  ≤  (Q + 1)−N/M 

     for each  m  = 1, 2, . . . , M  .

    Proof.   Let  B   denote the (M  + N ) × (M  + N ) real matrix that is organized intoblocks as

    B  =

    1M    A

    0 1N 

    .

    Let 0 < θ

  • 8/19/2019 Algebraic Number Theory Notes

    23/173

    ALGEBRAIC NUMBER THEORY 23

    If   q   =   0   then the inequality on the left of (2.46) implies that   p   =   0, which isimpossible. It follows that  q  = 0.

    Because the coordinates of   p  are integers, the inequality on the left of (2.46)implies that

    (2.47)   αm1q 1 + αm2q 2 + · · · + αmN q N  ≤  (Q + θ)−N/M 

    for each m  = 1, 2, . . . , M  . Because the coordinates of  q  are integers, the inequalityon the right of  (2.46) implies that

    (2.48)   |q|∞ ≤  Q.

    Thus for each  θ, 0 < θ <  1, we have established the existence of a point  q  =  0   inZN  that satisfies (2.47) and (2.48). But the set of points  q  = 0  in  ZN  that satisfy(2.48) is finite, and it follows that there exists one which satisfies ( 2.47) for valuesof  θ  that are arbitrarily close to 1. This proves the theorem.  

    Let   M   and   N   be positive integers and write   G1   = (R/Z)MN . We write the

    elements of the additive group G1  as  M  × N  matrices A  = (αmn) with entries αmnin  R/Z. Then the group operation in  G1  corresponds to addition of matrices. Wenote that in the statement of Theorem 2, we may assume that the matrix  A  belongsto the group  G1, because the conclusion (2.45) depends only on the image of eachentry in R/Z.

    Exercises

    2.1.   Let K  ⊆ RN  be a convex set such that the interior of  K   is unbounded. Provethat VolN (K ) =  ∞.

    2.2.  Prove that if 2 ≤  N  then GL(N, Z) is not a normal subgroup of GL(N, R).

    2.3.   Let Γ  ⊆  RN  and Λ  ⊆ RN  be lattices such that Γ  ⊆  Λ. In this case we saythat Γ is a  sublattice  of Λ. Prove that the group Γ has finite index in the group Λ.Then show that the ratio  d(Γ)/d(Λ) is an integer, and

    d(Γ) = d(Λ)[Λ : Γ].

    2.4.  Let Λ ⊆ RN  be a lattice, and suppose that F   ⊆ RN  is a fundamental domainfor the quotient group  RN /Λ. That is,   F   is a Borel subset of  RN , and   F   is acomplete set of distinct coset representatives for  RN /Λ. Prove that VolN (F ) =d(Λ).

    2.5.  Prove the following more general form of Theorem  2.10.   Let  A = (αmn) be

    an  M  × N   matrix with real entries, let  Q1, Q2, . . . , QN , be nonnegative integers,and let 0 < m ≤  12  be real numbers for  m  = 1, 2, . . . , M  . Prove that if 

    1 ≤  12 · · · M (Q1 + 1)(Q2 + 1) · · · (QN  + 1),

    then there exists an integer vector  ξ  = 0  in  ZN  such that

    αm1ξ 1 + αm2ξ 2 + · · · + αmN ξ N  ≤  m   for m  = 1, 2, . . . , M ,

    and|ξ n| ≤  Qn   for n  = 1, 2, . . . , N .

  • 8/19/2019 Algebraic Number Theory Notes

    24/173

  • 8/19/2019 Algebraic Number Theory Notes

    25/173

    ALGEBRAIC NUMBER THEORY 25

    3.  The Ideal Class Group

    Let   k   be an algebraic number field and   Ok   its ring of integers. As   Ok   is a

    Dedekind domain, we may speak of the multiplicative group  I (k) of fractional idealsin  k  and the subgroup  P (k)  ⊆   I (k) of principal fractional ideals in  k. It followsfrom Corollary K.2 that the group I (k) is a free abelian group on the set of all primeideals in  Ok. In this section we consider the  ideal class group  H (k) =  I (k)/P (k).The main result is that  H (k) is a finite group. Then we define the  class number  of the field  k  to be the cardinality

    hk  =  |H (k)|

    of the ideal class group. Obviously   hk   is an invariant of  k, but it is not easy tocompute hk, even if  k  has relatively low degree.

    Because I (k) is a free abelian group generated by the set of nonzero prime idealsin  Ok, we may define a unique homomorphism

    (3.1) normk/Q  :  I (k) → Q× ∩ (0, ∞)

    by specifying the value of normk/Q  at each generator. If  P  is a nonzero prime idealin  Ok, we set

    (3.2) normk/Q(P) =

    Ok  : P

    .

    If  A  is an integral ideal then

    (3.3) normk/Q(A) =

    Ok  : A

    is a positive integer, and equal to the index of  A in the ring Ok. If  γ  is an integralprincipal ideal, then by (1.35) we have

    (3.4) normk/Q

    γ 

    = [Ok  :  γ ] =Normk/Q(γ ).

    Lemma 3.1.   Let  k  be an algebraic number field,  Ok   its ring of integers, and  m  a positive integer. Then there are at most finitely many integral ideals  A  in  Ok   such 

    that  normk/Q(A) =  m.

    Proof.   Let   A   be an integral ideal in   Ok   such that normk/Q(A) =   m. Then thequotient ring  Ok/A has cardinality  m, and we find that

    m +A = m(1 + A) = 0 + A.

    This shows that   m   belongs to  A. We conclude that   m ⊆  A, and therefore byTheorem K.3  we have  A|m. Let

    m = Pe11  Pe22   · · ·P

    eN N 

    be the factorization of  m into a product of distinct prime ideals in  Ok  to positiveinteger powers. Because  A|m, the factorization of  A must have the form

    A = Pd11  Pd22   · · ·P

    dN N    where 0 ≤  dn ≤  en.

    Thus there are only finitely many possible integral ideals with normk/Q(A) =  m.   Lemma 3.2.   Let  k   be an algebraic number field,  Ok   its ring of integers, and  A  a 

     fractional ideal in  k. Then there exists a nonzero element  α  in  k  such that 

    (3.5)   αA = αA

    is an integral ideal, and 

    (3.6) normk/Q

    αA

    ≤ 4

    π

    s N !N N 

    Disc(k) 12 ,

  • 8/19/2019 Algebraic Number Theory Notes

    26/173

    26 VAALER

    where  2s   is the number of complex embeddings of  k   into C, and  N  = [k : Q].

    Proof.   As  A−1 is also a fractional ideal in  k, there exists a nonzero element  β   in

    Ok  so thatβ A−1 = β A−1 = B

    is an integral ideal in  Ok. Then by Theorem 1.8, Theorem 2.9 and (3.4) there existsa nonzero element  γ   in  B  such that

    normk/Q

    γ 

    =Normk/Q(γ )

    ≤ 4

    π

    s N !N N 

    Disck/Q(B) 12= 4

    π

    s N !N N 

     normk/Q(B)Disc(k) 12 .

    (3.7)

    Let α  =  β −1γ . Because γ  belongs to  B we have

    γ  ⊆ β A−1,

    and thereforeγ β −1A = αA ⊆ Ok.

    This shows that αA is an integral ideal. Then from (3.7) we find that

    normk/Q

    αA

    = normk/Q

    γ 

    normk/QB−1

    ≤ 4

    π

    s N !N N 

    Disc(k) 12 .This proves the lemma.  

    We are now able to prove the following fundamental result about algebraic num-ber fields.

    Theorem 3.3.   Let  k  be an algebraic number field and  H (k)   the ideal class groupof  k . Then  H (k)  is a finite group.

    Proof.  It follows from Lemma 3.1 that the collection of all integral ideals  I  in  Oksuch that

    (3.8) normk/Q(I) ≤ 4

    π

    s N !N N 

    Disc(k) 12 ,is a finite set. Then Lemma  3.2  asserts that every element  A  in  I (k) has an equiv-alent coset representative

    αA = I

    from the quotient group H (k) =  I (k)/P (k) such that  I  is an integral ideal and thenorm of  I  satisfies the bound (3.8). Hence the number of distinct cosets in H (k) isfinite.  

  • 8/19/2019 Algebraic Number Theory Notes

    27/173

    ALGEBRAIC NUMBER THEORY 27

    4.  Factorization of primes in an algebraic number field

    Let k  be an algebraic number field and let  I  be a proper ideal in the ring  Ok  of 

    algebraic integers in k. Because Ok  is a Dedekind domain, it follows from TheoremK.1 that  I has a factorization

    (4.1)   I = P1P2 · · ·PM ,

    where each  Pm   is a prime ideal in   Ok. Theorem   K.1   also asserts that the fac-torization (4.1) is unique up to a permutation of the prime ideal factors. For ourpurposes it will be convenient to reorganize the right hand side of (4.1) as

    (4.2)   I = Pe11  Pe22   · · ·P

    eLL   ,

    where P1,P2, . . . ,PL  are  distinct  prime ideals in  Ok  and e1, e2, . . . , eL  are positiveintegers. In this section we consider the special case where the ideal I is the principalideal   p in  Ok  generated by a rational prime number  p.

    Theorem 4.1.   Let  k  be an algebraic number field, and let  P  be a nonzero prime 

    ideal in  Ok. Then  P   contains a unique rational prime number  p. Moreover, there exists a positive integer  f  such that 

    (4.3) normk/Q(P) = [Ok  : P] =  pf ,   and    1 ≤  f  ≤ N,

    where  N  = [k : Q].

    Proof.  The prime ideal P is also a maximal ideal. Therefore Ok/P is a finite field of characteristic p  for some rational prime number p. The homomorphsim m → m +Pfrom Z   into Ok/P, maps Z  onto the prime subfield represented by

    0 + P, 1 + P, 2 + P, · · ·  , ( p − 1) + P

    .

    In particular we have0 + P = p + P,

    and therefore p  belongs to P. If  q  is also a rational prime contained in P and q  = p,then there exist rational integers  a  and  b  so that ap + bq  = 1. But this implies that1 belongs to  P, which is impossible. Hence   p   is the  unique   rational prime thatbelongs to  P.

    Let  p denote the principal ideal in  Ok  generated by the rational prime number p. Clearly we have   p ⊆ P  and therefore, by Theorem K.3, we have  P| p. Hencethere exists an ideal  Q  in  Ok  such that

    (4.4)    p = PQ.

    Then (3.2), (3.4), and (4.4), imply that

     pN  =Normk/Q( p) = normk/Q p = normk/Q(P) normk/Q(Q).

    Using (3.3) we find that

    normk/Q(P) = [Ok  : P] and normk/Q(Q) = [Ok  : Q]are positive integers, and the conclusion (4.3) plainly follows.  

    If P is a nonzero prime ideal in the ring  Ok  containing the unique rational primenumber   p, then the positive integer   f   that occurs in the identity (4.3)   is calledthe   inertial degree  (or more simply the  degree ) of  P. As is clear from the proof of Theorem 4.1, the map

    (4.5)   P → Char

    Ok/P

  • 8/19/2019 Algebraic Number Theory Notes

    28/173

    28 VAALER

    sends each nonzero prime ideal  P  into the unique rational prime p  that belongs toP.

    Now suppose that p  is a rational prime number and the principal ideal  p in  Okfactors into prime ideals as

    (4.6)    p = Pe11  Pe22   · · ·P

    eLL   ,

    where  P1,P2, . . . ,PL  are distinct prime ideals in  Ok, and  e1, e2, . . . , eL, are pos-itive integers. Obviously (4.6) implies that  Pl| p   for each integer   l  = 1, 2, . . . , L.Therefore  p ⊆ Pl, and p  is the unique rational prime contained in Pl. In this casethe integer  L   is called the  decomposition  number of  p   in  k. The exponent el   thatoccurs on the prime ideal  Pl   in the factorization (4.6), is called the   ramification index  of the prime ideal  Pl.

    Alternatively, let  P   be a nonzero prime ideal in   Ok, and let   p  be the uniquerational prime such that  P| p. Then the  ramification index   of  P   is the uniquepositive integer  e  such that

    Pe| p   and   Pe+1    p.

    There is a similar terminology that applies to the rational prime  p. Assume thatthe principal ideal  p in  Ok  factors into prime ideals as in (4.6). If  e1  =  e2  =  · · · =eL = 1 then we say that the rational prime  p  is  unramified  in  k . If 2 ≤  el   for someindex l, then we say that the rational prime  p  is  ramified  in  k .

    We note that these remarks also show that the map (4.5) from nonzero primeideals in Ok  to rational prime numbers is surjective. And the inverse image of eachrational prime p  is exactly the finite set of distinct nonzero prime ideals that occuron the right hand side of the factorization (4.6).

    We now show that the inertial degrees and ramification indices attached to thenonzero prime ideals in the factorization (4.6) satisfy a simple identity.

    Theorem 4.2.   Let   k  be an algebraic number field with   N   = [k   :  Q]. Let   p   be a rational prime and assume that the principal ideal    p   in  Ok  has the factorization (4.6). Then the the ramification indices  el  attached to each prime ideal  Pl, and the inertial degrees  f l  attached to each prime ideal  Pl, satisfy 

    (4.7)   N  =Ll=1

    elf l.

    Proof.   From (4.6) we obtain the identity

    (4.8)   pN  =Normk/Q( p) = normk/Q p = L

    l=1

    normk/QPell

    .

    Then using (4.3) we have

    (4.9) normk/QPell = normk/Q(Pl)el =  pf lel .The identity (4.7) follows by combining (4.8) and (4.9).  

    Next we describe a method due to Dedekind [24] for determining the factorization(4.6). To begin with we suppose that k  is generated over Q  by an algebraic integerβ . Then it follows that

    (4.10)   Z[β ] =

    a0 + a1β  + a2β 2 + · · · + aN −1β 

    N −1 : an ∈ Z

    ⊆ Ok

  • 8/19/2019 Algebraic Number Theory Notes

    29/173

    ALGEBRAIC NUMBER THEORY 29

    is a subring of   Ok. For each rational prime   p   there exists a unique, surjectivehomomorphism

    ψ p  : Z → F p,where  F p   is the finite field with cardinality   p. Then   ψ p   extends to a surjectivehomomorphism

    ψ p  : Z[x] → F p[x].

    In particular, if   mβ(x) in  Z[x] is the minimal polynomial for   β , then its imageψ p(mβ)(x) in  F p[x] factors into a product of irreducible factors in F p[x]. Thus wewrite

    (4.11)   ψ p(mβ)(x) =  g1(x)e1g2(x)

    e2 · · · gL(x)eL ,

    where g1(x), g2(x), . . . , gL(x), are distinct, monic, irreducible polynomials in F p[x],and e1, e2, . . . , eL, are positive integers. The following theorem of Dedekind showsthat for all but finitely many primes p, the factorization (4.11) provides informationabout the factorization of the ideal   p into prime ideals in  Ok.

    Theorem 4.3.   (Dedekind)  Let  k  = Q(β )  be an algebraic number field. Assume that  β  an algebraic integer,  mβ(x)   in  Z[x]  is the minimal polynomial for  β , and 

    N  = [k : Q] = deg mβ.

    Assume that  Z[β ]  is the subring defined by  (4.10), p  is a rational prime such that  pdoes not divide the index 

    Ok   : Z(β )

    , and  (4.11)   is the factorization of the image 

    ψ p

    (x)   into irreducible polynomials in  F p[x]. For each integer   l  = 1, 2, . . . , L,let  Gl(x)  be a monic polynomial in  Z[x]   such that 

    ψ p

    Gl

    (x) =  gl(x).

    Then for each integer   l = 1, 2, . . . , L, the ideal 

    (4.12)   Pl =   p, Gl(β )is a prime ideal in   Ok. Moreover, the prime ideals  P1,P2, . . . ,PL   are distinct,they satisfy 

    (4.13) normk/QPl

    =

    Ok  : Pl

    = pdeg gl = pdegGl ,

    and the ideal   p  factors into powers of distinct prime ideals as 

    (4.14)    p = Pe11  Pe22   · · ·P

    eLL   .

    Proof.  For each integer  l  = 1, 2, . . . , L the monic polynomial  gl(x) is irreducible inF p[x]. Let γ l  be a root of  gl(x) in an algebraic closure  F p. Then

    F p(γ l) ∼= F p[x]/gl(x)

    is a finite extension of F p  of degree deg gl. Hence F p(γ l) is a finite field of cardinalty

     pdeg gl . Letϕl : Z[β ] → F p(γ l)

    be the surjective ring homomorphism defined by

    (4.15)   ϕl

    f (β )

    = ψ p(f )(γ l).

    It is obvious that  ϕl( p) = 0, and

    ϕl

    Gl(β )

    = ψ p

    Gl

    (γ l) =  gl(γ l) = 0.

  • 8/19/2019 Algebraic Number Theory Notes

    30/173

    30 VAALER

    Therefore both  p  and  Gl(β ) belong to the kernel of  ϕl, and we get

    (4.16)   Pl =   p, Gl(β ) ⊆  ker(ϕl).

    On the other hand, if  f (β ) belongs to ker(ϕl) then  ψ p(f )(γ l) = 0, and we findthat ψ p(f )(x) belongs to the principle ideal gl(x) in F p[x] generated by the monicirreducible polynomial  gl(x). Hence there exists a polynomial H (x) in  Z[x] suchthat

    ψ p(f )(x) =  ψ p(Gl)(x)ψ p(H )(x) =  ψ p

    GlH 

    (x).

    That is, each coefficient of the polynomial

    f (x) − Gl(x)H (x)

    is divisible by p. Therefore we have

    f (β ) =  f (β ) − Gl(β )H (β ) + Gl(β )H (β )

    ⊆  p + Gl(β )

    =  p, Gl(β ).

    (4.17)

    Now (4.16) and (4.17) imply that

    (4.18)   Pl = ker(ϕl).

    As  ϕl   induces an isomorphsim

    ϕl  : Z[β ]/ ker(ϕl) ∼= F p(γ l),

    it follows that  Pl = ker(ϕl) is a maximal ideal, and therefore a prime ideal in  Z[β ].Because Gl(x) in Z[x] is monic and  ψ p

    Gl

    = gl, we also have deg Gl = deg gl.

    Exercises

    4.1.   Let  k  be an algebraic number field and  P  a nonzero prime ideal in the ring

    Ok. Prove that  P ∩ Z  is a nonzero prime ideal in the ring  Z.

  • 8/19/2019 Algebraic Number Theory Notes

    31/173

    ALGEBRAIC NUMBER THEORY 31

    5.  Absolute values

    Let K  be a field. An absolute value  on  K  is a map

    | | :  K  →  [0, ∞)

    from K   into the nonnegative real numbers that satisfies the following three condi-tions:

    (i)   |x| = 0 if and only if  x  = 0,(ii)   |xy| =  |x||y|  for all  x  and  y   in  K ,

    (iii)   |x + y| ≤ |x| + |y|  for all  x  and  y   in  K .

    The inequality (iii) is called the triangle inequality . For some absolute values definedon a field   K   it may happen that an inequality stronger than (iii) holds. Moreprecisely, it may happen that an absolute value on a field  K   satisfies the condition

    (iv)   |x + y| ≤  max{|x|, |y|} for all  x  and  y   in  K .

    If | | satisfies (iv) then it is obvious that | | must also satisfy (iii). The inequality (iv)is called the   strong triangle inequality  or the  ultrametric inequality . An absolutevalue   | |  on a field  K   that satisfies the conditions (i), (ii), and (iii), but does notsatisfy (iv), is said to be an archimedean  absolute value. An absolute value | | on  K that satisfies the conditions (i), (ii), and (iv), is said to be a  non-archimedean  abso-lute value or an  ultrametric  absolute value. Later we will apply these designationsto certain equivalence classes of absolute values.

    The first results about a field with an absolute value are immediate consequencesof the definition. Every field  K  has at least one absolute value which we denotehere by  | |0. This is defined by

    |x|0  =

    0 if  x  = 0K ,

    1 if  x  = 0K .

    The absolute value  | |0   is called the  trivial  or  improper  absolute value. Any otherabsolute values on K  will be called nontrivial  or  proper . When working in a field  K the trivial absolute value is usually not considered. However, there is at least onesituation in which the trivial absolute value may naturally occur. Suppose that Lis an extension of  K   and  | |  is an absolute value on  L. Then  | |  restricted to  K   isobviously an absolute value on  K . It may happen that  | |  is a nontrivial absolutevalue on  L, but the restriction of  | |  to  K  is trivial.

    Let  K × denote the multiplicative group of nonzero elements in  K . If   | |   is anabsolute value on   K   then the restriction of   | |   to   K × is a homomorphism fromthe multiplicative group K × into the multiplicative group of positive real numbers.It follows that   |1K |  = 1. More generally, the multiplicative group of positive realnumbers is torsion free. It follows that if  ζ N   is a primitive  N -th root of unity in

    K ×

    , then  |ζ N | = 1. We also get| − x| =  |x|   for all x  ∈  K,   and   |x−1| =  |x|−1 for all  x  ∈  K ×.

    If  x  is in K × and |x| = 1 then x  must have infinite order in the multiplicative groupK ×. This shows that the only absolute value on a finite field is the trivial absolutevalue | |0.

    When verifying that a function   | | :  K  → [0, ∞) is an absolute value, it is oftenuseful to be able to check an inequality that is weaker than the triangle inequality.

  • 8/19/2019 Algebraic Number Theory Notes

    32/173

    32 VAALER

    Lemma 5.1.   Assume that   K   is a field and   | |   :   K   →   [0, ∞)   satisfies the three conditions 

    (i)   |x| = 0  if and only if  x = 0K ,(ii)   |xy| =  |x||y|  for all  x  and  y   in  K ,(v)   |x + y| ≤  2 max{|x|, |y|}  for all  x  and  y   in  K .

    Then   | |   satisfies the triangle inequality  (iii)   in the definition of an absolute value,and therefore   | |  is an absolute value on  K .

    Proof.   If  M   = 2m for some positive integer  m, then it follows easily from (v), byinduction on  m, that

    (5.1)   |x1 + x2 + · · · + xM | ≤  2m max{|x1|, |x2|, . . . , |xM |}

    for all x1, x2, . . . , xM   in K . If  N  is a positive integer and 2m−1 < N  ≤ 2m, then by

    selecting xN +1  =  xN +2  =  · · · =  xM  = 0K   in (5.1), we find that

    |x1 + x2 + · · · + xN | ≤  2m max{|x1|, |x2|, . . . , |xN |}

    ≤ 2N  max{|x1|, |x2|, . . . , |xN |}(5.2)

    for all   x1, x2, . . . , xN   in   K . In particular, if   x1   =  x2   =   · · ·  =   xN   = 1K , then itfollows that

    (5.3)   |N 1K | ≤  2N.

    Now let x  and  y  be elements of  K , and let L  be a positive integer. Then using (5.2)and (5.3) we get

    |x + y|L = |(x + y)L|

    = Ll=0

    L

    l

    xlyL−l

    ≤ 2(L + 1)maxLl1K |x|l|y|L−l : 0 ≤  l  ≤  L≤ 4(L + 1)max

    L

    l

    |x|l|y|L−l : 0 ≤  l  ≤  L

    ≤ 4(L + 1)

    Ll=0

    L

    l

    |x|l|y|L−l

    = 4(L + 1)

    |x| + |y|L

    ,

    and therefore

    (5.4)   |x + y| ≤

    4(L + 1)1/L

    |x| + |y|

    .

    The triangle inequality (iii) follows by letting   L   → ∞  on the right hand side of (5.4).  

    If  | |  is an absolute value on  K  then the map (x, y) → |x − y|  from  K  × K   into[0, ∞) is a metric, and therefore the absolute value induces a metric topology inK . Of course the distance from x  to  y   is   |x − y|. Let  AK  denote the collection of all absolute values on  K . We say that two absolute values in   AK   are   equivalent if they induce the same metric topology. It is easy to verify that this is in factan equivalence relation in  AK . The trivial absolute value  | |0  clearly induces thediscrete topology in  K , and it is the unique absolute value in its equivalence class.

  • 8/19/2019 Algebraic Number Theory Notes

    33/173

    ALGEBRAIC NUMBER THEORY 33

    We will show that an equivalence class determined by a nontrivial absolute valuecontains infinitely many distinct, nontrivial absolute values. An equivalence classdetermined by a nontrivial absolute value in  A

    K   is called a  place  of  K . Thus we

    may speak of the metric topology in  K  induced by a place of  K , because all theabsolute values in that place induce the same topology.

    Equivalent absolute values on  K  can be characterized in a simple way.

    Theorem 5.2.   Let   | |1   and   | |2  be two absolute values on  K . Then the following assertions are equivalent:

    (i)   | |1   and  | |2   induce the same metric topology in  K ,(ii)   {x ∈  K  :  |x|1  <  1} =  {x ∈  K  :  |x|2  <  1},

    (iii)   there exists a positive number  θ  such that  |x|θ1  =  |x|2  for all  x   in  K .

    Proof.  Assume that (i) holds and let

    U j  = {x ∈  K  :  |x|j  

  • 8/19/2019 Algebraic Number Theory Notes

    34/173

    34 VAALER

    we further assume that  | |  is nontrivial, then   |z| = 1 for some point  z   in  K ×, andwe conclude that  | |θ and | |  are equivalent but not equal. In particular, this showsthat a place determined by a nontrivial absolute value in   A

    K   contains infinitely

    many distinct but equivalent absolute values.Let | |  be an absolute value on  K . Then define

    Θ(| |) =  {θ > 0 : x  → |x|θ is an absolute value in  AK }.

    It follows from the definition of an absolute value that Θ( | |) is a closed, nonemptysubset of (0, ∞). By our previous remarks, if  τ   is in Θ(| |) then  θτ   is in Θ(| |) forall  θ  such that 0 < θ ≤  1. This shows that either Θ(| |) = (0, ∞), or Θ(| |) = (0, τ ]for some real number τ  ≥ 1. We define a second function on absolute values in AK by setting

    Φ(| |) = sup{|x + 1| :  x  ∈  k  and |x| ≤  1}.

    Obviously we have 1 ≤  Φ(| |) ≤  2. Also, the inequality

    (5.5)   |x + y| ≤  Φ(| |)max{|x|, |y|}

    holds for all  x  and  y  in  K . To verify (5.5) assume that 0 <  |x| ≤ |y|. Then we have

    |x + y| =  |(xy−1 + 1)y| =  |xy−1 + 1||y| ≤  Φ(| |)|y| = Φ(| |)max{|x|, |y|}.

    Note that if Φ(| |) = 1 then   (5.5) is the strong triangle inequality (iv) in thedefinition of an absolute value.

    Theorem 5.3.   Let   | |   be an absolute value on  K . Then the following assertions are equivalent:

    (i) Θ(| |) = (0, ∞),(ii)   |x + y| ≤  max{|x|, |y|}  for all  x  and  y   in  K ,

    (iii) Φ(| |) = 1.

    Moreover, if  1 <  Φ(| |) ≤  2, then  Θ(| |) = (0, τ ]   for some real number  τ  ≥ 1, and 

    (5.6) Φ(| |)τ  = 2.

    Proof.  Assume that Θ(| |) = (0, ∞). Then the map  x  → |x|n is an absolute valuefor all positive integers  n. It follows that

    |x + y| ≤

    |x|n + |y|n1/n

    for all  x  and  y   in  K . We let  n  → ∞ and conclude that

    |x + y| ≤  max{|x|, |y|}.

    This shows that (i) implies (ii).It is trivial that (ii) implies (iii).Assume that Φ(| |) = 1. In view of (5.5) we find that (ii) is satisfied. Then we

    get

    |x + y|θ ≤ max{|x|, |y|}θ= max{|x|θ, |y|θ}

    ≤ |x|θ + |y|θ

    (5.7)

    for all x  and  y   in K  and for all positive  θ . This shows that the map  x  → |x|θ is anabsolute value on K  for all positive θ, and so verifies that Θ(| |) = (0, ∞). We haveshown that (iii) implies (i).

  • 8/19/2019 Algebraic Number Theory Notes

    35/173

    ALGEBRAIC NUMBER THEORY 35

    Assume that 1 <  Φ(| |) ≤  2. By our previous remarks we have Θ(| |) = (0, τ ] forsome real number  τ  ≥ 1. Because  | |τ  is an absolute value, we find that

    Φ(| |)τ  = Φ| |τ = sup{|x + 1|τ  : x  ∈  k  and |x|τ  ≤ 1}

    ≤ 2.

    (5.8)

    If in fact Φ(| |)τ  τ   such that

    Φ

    | |θ

    = Φ(| |)θ ≤ 2.

    It follows from (5.5) that

    (5.9)   |x + y|θ ≤ 2 max{|x|θ, |y|θ}

    for all  x  and  y   in  K . From Lemma 5.1 and (5.9) we conclude that  x  → |x|θ is anabsolute value on  K . As  θ > τ , this contradicts the definition of Θ(| |). We haveshown that if 1 <  Φ(| |) ≤  2, then (5.6) holds.  

    We are now in position to distinguish between different types of absolute values.If  | |  is an absolute value on  K  then by Theorem 5.3, Φ(| |) = 1 if and only if theabsolute value satisfies the inequality

    (5.10)   |x + y| ≤  max{|x|, |y|}

    for all  x  and  y   in  K . An absolute value that satisfies (5.10) is said to be a  non-archimedean  absolute value, or to be an  ultrametric  absolute value. Then (5.10) iscalled the   strong triangle   inequality, or the  ultrametric   inequality. If   | |   is a non-archimedean absolute value then the equivalence class it represents in  AK   is theset

    {| |θ : 0 < θ  1, and   |α|n <  1   for  2  ≤  n  ≤  N .

  • 8/19/2019 Algebraic Number Theory Notes

    36/173

    36 VAALER

    Proof.   Suppose that N  = 2. As  | |1  and  | |2  are inequivalent, by Theorem 5.2  thereexists β   in  K  with

    |β |1  <  1 and 1 ≤ |β |2,and there exists  γ   in  K   with

    1 ≤ |γ |1   and   |γ |2  <  1.

    In this case we take  α  =  β −1γ .We continue now using induction on N . By the inductive hypothesis there exists

    β   in  K   with|β |1  >  1 and   |β |n <  1 for 2 ≤  n  ≤  N  − 1,

    and γ   in  K  with1 ≤ |γ |1   and   |γ |N     0   there exists a point  y   in  K  such that 

    |xn − y|n <   for each  n  = 1, 2, . . . , N .

    Proof.  For each integer  n  with 1  ≤  n  ≤  N  we apply Lemma  5.4,  but with   | |n   inplace of   | |1. In this way we determine a collection of points  α1, α2, . . . , αN   in  K such that

    |αm|n >  1 if  m  =  n,   and  |αm|n <  1 if  m  = n.

    Then we find that

    liml→∞

    αlm1 + αlm

    = 1 in the | |n-metric if  m  =  n,

    and

    liml→∞

    αlm1 + αlm

    = 0 in the | |n-metric if  m  = n.

    It follows that for each  n, 1 ≤  n  ≤  N , we have

    liml→∞

    N m=1

    xmαlm

    1 + αlm= xn   in the | |n-metric.

    Hence we may take

    y =N 

    m=1

    xmαlm

    1 + αlm

    with  l  a sufficiently large positive integer that depends on  .  

  • 8/19/2019 Algebraic Number Theory Notes

    37/173

    ALGEBRAIC NUMBER THEORY 37

    Given a field  K  it is an interesting problem to determine all the nontrivial ab-solute values on  K . The following result is useful when the field  K   is the field of fractions associated to an integral domain.

    Lemma 5.6.   Let  R  be a nonzero integral domain, and let  K  be its field of fractions.Assume that  x  → x is a map from  R  into [0, ∞) that satisfies the three conditions 

    (i)   x = 0   if and only if  x = 0,(ii)   xy =  xy  for all  x  and  y   in  R,

    (iii)   x + y ≤ x + y  for all  x  and  y   in  R.

    Then there exists a unique absolute value   | |  :  K   → [0, ∞)  such that   |x|  =  x   for all  x  in  R. If    on  R   satisfies the strong triangle inequality 

    (iv)   x + y ≤  max

    x, y

     for all  x  and  y   in  R,

    then  | |  on  K   is a non-archimedean absolute value. Moreover, if  a  and  b  = 0  are in R, and  a/b  is a point in  K , then these maps satisfy the identity 

    (5.11) ab =   ab .Proof.  Suppose that  a/b =  c/d   in  K . That is,  a,  b  = 0,  c  and  d  = 0 belong to  Rand ad  =  bc. It follows that  ad =  bc and

    a

    b  =

      c

    d

    in [0, ∞). Therefore we define a map  | | :  K  →  [0, ∞) byab

    =   ab

    .

    Our previous remarks show that this is well defined. Using (i), (ii) and (iii) it iseasy to verify that   | |   on  K  satisfies the corresponding conditions required of an

    absolute value. For example, if  a,  b  = 0,  c  and  d  = 0 belong to  R, thenab

     +  c

    d

    = ad + bcbd

    =   ad + bcbd

    ≤ ad + bc

    bd  =

      a

    b +

      c

    d

    =a

    b

    + cd

    .(5.12)

    If (iv) holds this inequality becomesab

     +  c

    d

    = ad + bcbd

    =   ad + bcbd

    ≤ max

    ad, bc

    bd   = maxa

    b

    , c

    d= max

    ab

    , cd

    ,(5.13)

    and we conclude that the absolute value | |  is non-archimedean.We note that (ii) implies that  y =  1K y =  1K y and therefore  1K  = 1. If 

    a  belongs to  R  then

    |a| =   a

    1K 

    =   a1K 

     = a,

  • 8/19/2019 Algebraic Number Theory Notes

    38/173

    38 VAALER

    and this shows that   | |  on  K   is an extension of    on  R. If   | |1   and   | |2  are bothabsolute values on  K  that extend the map   on  R, then we have

    ab

    1=   |a|1

    |b|1=   a

    b  =   |a|2

    |b|2= a

    b

    2.

    This shows that   | |1   and   | |2  are equal on  K . Thus the absolute value defined by(5.11) is the unique extension of    on  R.  

    Exercises

    5.1.   Let | | be an absolute value on the field K . Prove that | | is non-archimedeanif and