In mathematics, an algebraic number field (or simply number field) is an extension field of the field of rational numbers such that the field extension has finite degree (and hence is an algebraic field extension). Thus is a field that contains and has finite dimension when considered as a vector space over .
The study of algebraic number fields, and, more generally, of algebraic extensions of the field of rational numbers, is the central topic of algebraic number theory. This study reveals hidden structures behind usual rational numbers, by using algebraic methods.
The notion of algebraic number field relies on the concept of a field. A field consists of a set of elements together with two operations, namely addition, and multiplication, and some distributivity assumptions. A prominent example of a field is the field of rational numbers, commonly denoted , together with its usual operations of addition and multiplication.
whose entries are elements of a fixed field, such as the field . Any two such sequences can be added by adding the entries one per one. Furthermore, any sequence can be multiplied by a single element c of the fixed field. These two operations known as vector addition and scalar multiplication satisfy a number of properties that serve to define vector spaces abstractly. Vector spaces are allowed to be "infinite-dimensional", that is to say that the sequences constituting the vector spaces are of infinite length. If, however, the vector space consists of finite sequences
the vector space is said to be of finite dimension, n.
Every field extension of finite degree is algebraic. (Proof: for in , simply consider – we get a linear dependence, i.e. a polynomial that is a root of.) In particular this applies to algebraic number fields, so any element of an algebraic number field can be written as a zero of a polynomial with rational coefficients. Therefore, elements of are also referred to as algebraic numbers. Given a polynomial such that , it can be arranged such that the leading coefficient is one, by dividing all coefficients by it, if necessary. A polynomial with this property is known as a monic polynomial. In general it will have rational coefficients.
If, however, its coefficients are actually all integers, is called an algebraic integer.
Any (usual) integer is an algebraic integer, as it is the zero of the linear monic polynomial:
It can be shown that any algebraic integer that is also a rational number must actually be an integer, hence the name "algebraic integer". Again using abstract algebra, specifically the notion of a finitely generated module, it can be shown that the sum and the product of any two algebraic integers is still an algebraic integer. It follows that the algebraic integers in form a ring denoted called the ring of integers of . It is a subring of (that is, a ring contained in) . A field contains no zero divisors and this property is inherited by any subring, so the ring of integers of is an integral domain. The field is the field of fractions of the integral domain . This way one can get back and forth between the algebraic number field and its ring of integers . Rings of algebraic integers have three distinctive properties: firstly, is an integral domain that is integrally closed in its field of fractions . Secondly, is a Noetherian ring. Finally, every nonzero prime ideal of is maximal or, equivalently, the Krull dimension of this ring is one. An abstract commutative ring with these three properties is called a Dedekind ring (or Dedekind domain), in honor of Richard Dedekind, who undertook a deep study of rings of algebraic integers.
For general Dedekind rings, in particular rings of integers, there is a unique factorization of ideals into a product of prime ideals. For example, the ideal in the ring of quadratic integers factors into prime ideals as
However, unlike as the ring of integers of , the ring of integers of a proper extension of need not admit unique factorization of numbers into a product of prime numbers or, more precisely, prime elements. This happens already for quadratic integers, for example in , the uniqueness of the factorization fails:
Using the norm it can be shown that these two factorization are actually inequivalent in the sense that the factors do not just differ by a unit in . Euclidean domains are unique factorization domains; for example , the ring of Gaussian integers, and , the ring of Eisenstein integers, where is a cube root of unity (unequal to 1), have this property.
The failure of unique factorization is measured by the class number, commonly denoted h, the cardinality of the so-called ideal class group. This group is always finite. The ring of integers possesses unique factorization if and only if it is a principal ring or, equivalently, if has class number 1. Given a number field, the class number is often difficult to compute. The class number problem, going back to Gauss, is concerned with the existence of imaginary quadratic number fields (i.e., ) with prescribed class number. The class number formula relates h to other fundamental invariants of . It involves the Dedekind zeta function ζ (s), a function in a complex variable s, defined by
(The product is over all prime ideals of , denotes the norm of the prime ideal or, equivalently, the (finite) number of elements in the residue field . The infinite product converges only for Re(s) > 1, in general analytic continuation and the functional equation for the zeta-function are needed to define the function for all s). The Dedekind zeta-function generalizes the Riemann zeta-function in that ζ (s) = ζ(s).
Here r1 and r2 classically denote the number of real embeddings and pairs of complex embeddings of , respectively. Moreover, Reg is the regulator of , w the number of roots of unity in and D is the discriminant of .
Dirichlet L-functions are a more refined variant of . Both types of functions encode the arithmetic behavior of and , respectively. For example, Dirichlet's theorem asserts that in any arithmetic progression
with coprime and , there are infinitely many prime numbers. This theorem is implied by the fact that the Dirichlet -function is nonzero at . Using much more advanced techniques including algebraic K-theory and Tamagawa measures, modern number theory deals with a description, if largely conjectural (see Tamagawa number conjecture), of values of more general L-functions.
An integral basis for a number field of degree is a set
of n algebraic integers in such that every element of the ring of integers of can be written uniquely as a Z-linear combination of elements of B; that is, for any x in we have
where the mi are (ordinary) integers. It is then also the case that any element of can be written uniquely as
where now the mi are rational numbers. The algebraic integers of are then precisely those elements of where the mi are all integers.
Working locally and using tools such as the Frobenius map, it is always possible to explicitly compute such a basis, and it is now standard for computer algebra systems to have built-in programs to do this.
Let be a number field of degree . Among all possible bases of (seen as a -vector space), there are particular ones known as power bases, that are bases of the form
for some element . By the primitive element theorem, there exists such an , called a primitive element. If can be chosen in and such that is a basis of as a free Z-module, then is called a power integral basis, and the field is called a monogenic field. An example of a number field that is not monogenic was first given by Dedekind. His example is the field obtained by adjoining a root of the polynomial
Recall that any field extension has a unique -vector space structure. Using the multiplication in , an element of the field over the base field may be represented by matrices
Now this can be generalized slightly by instead considering a field extension and giving an -basis for . Then, there is an associated matrix which has trace and norm defined as the trace and determinant of the matrix .
Consider the field extension where . Then, we have a -basis given by
By definition, standard properties of traces and determinants of matrices carry over to Tr and N: Tr(x) is a linear function of x, as expressed by Tr(x + y) = Tr(x) + Tr(y), Tr(λx) = λ Tr(x), and the norm is a multiplicative homogeneous function of degree n: N(xy) = N(x) N(y), N(λx) = λn N(x). Here λ is a rational number, and x, y are any two elements of .
The matrix associated to an element x of can also be used to give other, equivalent descriptions of algebraic integers. An element x of is an algebraic integer if and only if the characteristic polynomial pA of the matrix A associated to x is a monic polynomial with integer coefficients. Suppose that the matrix A that represents an element x has integer entries in some basis e. By the Cayley–Hamilton theorem, pA(A) = 0, and it follows that pA(x) = 0, so that x is an algebraic integer. Conversely, if x is an element of which is a root of a monic polynomial with integer coefficients then the same property holds for the corresponding matrix A. In this case it can be proven that A is an integer matrix in a suitable basis of . The property of being an algebraic integer is defined in a way that is independent of a choice of a basis in .
Consider , where x satisfies x3 − 11x2 + x + 1 = 0. Then an integral basis is [1, x, 1/2(x2 + 1)], and the corresponding integral trace form is
The "3" in the upper left hand corner of this matrix is the trace of the matrix of the map defined by the first basis element (1) in the regular representation of on . This basis element induces the identity map on the 3-dimensional vector space, . The trace of the matrix of the identity map on a 3-dimensional vector space is 3.
The determinant of this is 1304 = 23·163, the field discriminant; in comparison the root discriminant, or discriminant of the polynomial, is 5216 = 25·163.
Mathematicians of the nineteenth century assumed that algebraic numbers were a type of complex number. This situation changed with the discovery of p-adic numbers by Hensel in 1897; and now it is standard to consider all of the various possible embeddings of a number field into its various topological completions at once.
A place of a number field is an equivalence class of absolute values on pg 9. Essentially, an absolute value is a notion to measure the size of elements of . Two such absolute values are considered equivalent if they give rise to the same notion of smallness (or proximity). The equivalence relation between absolute values is given by some such that
In general, the types of places fall into three regimes. Firstly (and mostly irrelevant), the trivial absolute value | |0, which takes the value on all non-zero . The second and third classes are Archimedean places and non-Archimedean (or ultrametric) places. The completion of with respect to a place is given in both cases by taking Cauchy sequences in and dividing out null sequences, that is, sequences such that
For , the following non-trivial norms occur (Ostrowski's theorem): the (usual) absolute value, sometimes denoted which gives rise to the complete topological field of the real numbers . On the other hand, for any prime number , the p-adic absolute value is defined by
It is used to construct the -adic numbers . In contrast to the usual absolute value, the p-adic absolute value gets smaller when q is multiplied by p, leading to quite different behavior of as compared to .
Note the general situation typically considered is taking a number field and considering a prime ideal for its associated ring of algebraic numbers . Then, there will be a unique place called a non-Archimedean place. In addition, for every embedding there will be a place called an Archimedean place, denoted . This statement is a theorem also called Ostrowski's theorem.
The field for where is a fixed 6th root of unity, provides a rich example for constructing explicit real and complex Archimedean embeddings, and non-Archimedean embeddings as wellpg 15-16.
Here we use the standard notation and for the number of real and complex embeddings used, respectively (see below).
Calculating the archimedean places of a number field is done as follows: let be a primitive element of , with minimal polynomial (over ). Over , will generally no longer be irreducible, but its irreducible (real) factors are either of degree one or two. Since there are no repeated roots, there are no repeated factors. The roots of factors of degree one are necessarily real, and replacing by gives an embedding of into ; the number of such embeddings is equal to the number of real roots of . Restricting the standard absolute value on to gives an archimedean absolute value on ; such an absolute value is also referred to as a real place of . On the other hand, the roots of factors of degree two are pairs of conjugate complex numbers, which allows for two conjugate embeddings into . Either one of this pair of embeddings can be used to define an absolute value on , which is the same for both embeddings since they are conjugate. This absolute value is called a complex place of .
To find the non-Archimedean places, let again and be as above. In , splits in factors of various degrees, none of which are repeated, and the degrees of which add up to , the degree of . For each of these -adically irreducible factors , we may suppose that satisfies and obtain an embedding of into an algebraic extension of finite degree over . Such a local field behaves in many ways like a number field, and the -adic numbers may similarly play the role of the rationals; in particular, we can define the norm and trace in exactly the same way, now giving functions mapping to . By using this -adic norm map for the place , we may define an absolute value corresponding to a given -adically irreducible factor of degree by
For any ultrametric place v we have that |x|v ≤ 1 for any x in , since the minimal polynomial for x has integer factors, and hence its p-adic factorization has factors in Zp. Consequently, the norm term (constant term) for each factor is a p-adic integer, and one of these is the integer used for defining the absolute value for v.
For an ultrametric place v, the subset of defined by |x|v < 1 is an ideal of . This relies on the ultrametricity of v: given x and y in , then
Actually, is even a prime ideal.
Conversely, given a prime ideal of , a discrete valuation can be defined by setting where n is the biggest integer such that , the n-fold power of the ideal. This valuation can be turned into an ultrametric place. Under this correspondence, (equivalence classes) of ultrametric places of correspond to prime ideals of . For , this gives back Ostrowski's theorem: any prime ideal in Z (which is necessarily by a single prime number) corresponds to a non-Archimedean place and vice versa. However, for more general number fields, the situation becomes more involved, as will be explained below.
Yet another, equivalent way of describing ultrametric places is by means of localizations of . Given an ultrametric place on a number field , the corresponding localization is the subring of of all elements such that | x |v ≤ 1. By the ultrametric property is a ring. Moreover, it contains . For every element x of , at least one of x or x−1 is contained in . Actually, since K×/T× can be shown to be isomorphic to the integers, is a discrete valuation ring, in particular a local ring. Actually, is just the localization of at the prime ideal , so . Conversely, is the maximal ideal of .
Altogether, there is a three-way equivalence between ultrametric absolute values, prime ideals, and localizations on a number field.
One of the basic theorems in algebraic number theory are the going up and going down theorems, which describe the behavior of some prime ideal when it is extended as an ideal in for some field extension . We say that an ideal lies over if . Then, one incarnation of the theorem states a prime ideal in lies over , hence there is always a surjective map
Ramification, generally speaking, describes a geometric phenomenon that can occur with finite-to-one maps (that is, maps such that the preimages of all points y in Y consist only of finitely many points): the cardinality of the fibers f−1(y) will generally have the same number of points, but it occurs that, in special points y, this number drops. For example, the map
has n points in each fiber over t, namely the n (complex) roots of t, except in t = 0, where the fiber consists of only one element, z = 0. One says that the map is "ramified" in zero. This is an example of a branched covering of Riemann surfaces. This intuition also serves to define ramification in algebraic number theory. Given a (necessarily finite) extension of number fields , a prime ideal p of generates the ideal pOK of . This ideal may or may not be a prime ideal, but, according to the Lasker–Noether theorem (see above), always is given by
with uniquely determined prime ideals qi of and numbers (called ramification indices) ei. Whenever one ramification index is bigger than one, the prime p is said to ramify in .
The connection between this definition and the geometric situation is delivered by the map of spectra of rings . In fact, unramified morphisms of schemes in algebraic geometry are a direct generalization of unramified extensions of number fields.
Ramification is a purely local property, i.e., depends only on the completions around the primes p and qi. The inertia group measures the difference between the local Galois groups at some place and the Galois groups of the involved finite residue fields.
The following example illustrates the notions introduced above. In order to compute the ramification index of , where
at 23, it suffices to consider the field extension . Up to 529 = 232 (i.e., modulo 529) f can be factored as
Substituting x = y + 10 in the first factor g modulo 529 yields y + 191, so the valuation | y |g for y given by g is | −191 |23 = 1. On the other hand, the same substitution in h yields y2 − 161y − 161 modulo 529. Since 161 = 7 × 23,
Since possible values for the absolute value of the place defined by the factor h are not confined to integer powers of 23, but instead are integer powers of the square root of 23, the ramification index of the field extension at 23 is two.
The valuations of any element of can be computed in this way using resultants. If, for example y = x2 − x − 1, using the resultant to eliminate x between this relationship and f = x3 − x − 1 = 0 gives y3 − 5y2 + 4y − 1 = 0. If instead we eliminate with respect to the factors g and h of f, we obtain the corresponding factors for the polynomial for y, and then the 23-adic valuation applied to the constant (norm) term allows us to compute the valuations of y for g and h (which are both 1 in this instance.)
Much of the significance of the discriminant lies in the fact that ramified ultrametric places are all places obtained from factorizations in where p divides the discriminant. This is even true of the polynomial discriminant; however the converse is also true, that if a prime p divides the discriminant, then there is a p-place which ramifies. For this converse the field discriminant is needed. This is the Dedekind discriminant theorem. In the example above, the discriminant of the number field with x3 − x − 1 = 0 is −23, and as we have seen the 23-adic place ramifies. The Dedekind discriminant tells us it is the only ultrametric place which does. The other ramified place comes from the absolute value on the complex embedding of .
Generally in abstract algebra, field extensions K / L can be studied by examining the Galois group Gal(K / L), consisting of field automorphisms of leaving elementwise fixed. As an example, the Galois group of the cyclotomic field extension of degree n (see above) is given by (Z/nZ)×, the group of invertible elements in Z/nZ. This is the first stepstone into Iwasawa theory.
In order to include all possible extensions having certain properties, the Galois group concept is commonly applied to the (infinite) field extension K / K of the algebraic closure, leading to the absolute Galois group G := Gal(K / K) or just Gal(K), and to the extension . The fundamental theorem of Galois theory links fields in between and its algebraic closure and closed subgroups of Gal(K). For example, the abelianization (the biggest abelian quotient) Gab of G corresponds to a field referred to as the maximal abelian extension Kab (called so since any further extension is not abelian, i.e., does not have an abelian Galois group). By the Kronecker–Weber theorem, the maximal abelian extension of is the extension generated by all roots of unity. For more general number fields, class field theory, specifically the Artin reciprocity law gives an answer by describing Gab in terms of the idele class group. Also notable is the Hilbert class field, the maximal abelian unramified field extension of . It can be shown to be finite over , its Galois group over is isomorphic to the class group of , in particular its degree equals the class number h of (see above).
In certain situations, the Galois group acts on other mathematical objects, for example a group. Such a group is then also referred to as a Galois module. This enables the use of group cohomology for the Galois group Gal(K), also known as Galois cohomology, which in the first place measures the failure of exactness of taking Gal(K)-invariants, but offers deeper insights (and questions) as well. For example, the Galois group G of a field extension L / K acts on L×, the nonzero elements of L. This Galois module plays a significant role in many arithmetic dualities, such as Poitou-Tate duality. The Brauer group of , originally conceived to classify division algebras over , can be recast as a cohomology group, namely H2(Gal (K, K×).
Generally speaking, the term "local to global" refers to the idea that a global problem is first done at a local level, which tends to simplify the questions. Then, of course, the information gained in the local analysis has to be put together to get back to some global statement. For example, the notion of sheaves reifies that idea in topology and geometry.
Number fields share a great deal of similarity with another class of fields much used in algebraic geometry known as function fields of algebraic curves over finite fields. An example is Kp(T). They are similar in many respects, for example in that number rings are one-dimensional regular rings, as are the coordinate rings (the quotient fields of which is the function field in question) of curves. Therefore, both types of field are called global fields. In accordance with the philosophy laid out above, they can be studied at a local level first, that is to say, by looking at the corresponding local fields. For number fields , the local fields are the completions of at all places, including the archimedean ones (see local analysis). For function fields, the local fields are completions of the local rings at all points of the curve for function fields.
Many results valid for function fields also hold, at least if reformulated properly, for number fields. However, the study of number fields often poses difficulties and phenomena not encountered in function fields. For example, in function fields, there is no dichotomy into non-archimedean and archimedean places. Nonetheless, function fields often serves as a source of intuition what should be expected in the number field case.
A prototypical question, posed at a global level, is whether some polynomial equation has a solution in . If this is the case, this solution is also a solution in all completions. The local-global principle or Hasse principle asserts that for quadratic equations, the converse holds, as well. Thereby, checking whether such an equation has a solution can be done on all the completions of , which is often easier, since analytic methods (classical analytic tools such as intermediate value theorem at the archimedean places and p-adic analysis at the nonarchimedean places) can be used. This implication does not hold, however, for more general types of equations. However, the idea of passing from local data to global ones proves fruitful in class field theory, for example, where local class field theory is used to obtain global insights mentioned above. This is also related to the fact that the Galois groups of the completions Kv can be explicitly determined, whereas the Galois groups of global fields, even of are far less understood.
To Dedekind, then, fields were subsets of the complex numbers.
Empiricism sprang from the 19th-century view of mathematics as almost coterminal with theoretical physics.