Higher Ramification Groups

COLORADO STATE UNIVERSITY MATHEMATICS

Higher Ramiﬁcation Groups

Dean Bisogno

May 24, 2016

1 ABSTRACT

Studying higher ramification groups immediately depends on some key ideas from valuation theory. With that in mind, this paper will outline the essential results from valuation theory before proceeding to higher ramification groups. Higher ramification groups arise when studying extensions of fraction fields of Dedekind rings. Ramification groups are also useful when studying wildly ramified extensions of function fields. The goal while studying ramification in a wildly ramified extension of function fields is to calculate the exponent of the different. By Hilbert’s different theorem, the exponent of the different can be calculated by finding orders of higher ramification groups and ramification jumps. Higher ramification groups can also be used to study the subgroup structure of a Galois group, as the higher ramification group filtration can contain information about Sylow p-subgroups. The aim of this paper is to understand at a general level what higher ramification groups are, and investigate a couple examples of their applications.

2 PRELIMINARIES

2.1 VALUATIONS

Though valuations need not be discrete, many results discussed in this paper either require discrete valuations or are most easily understood with respect to a discrete valuation. Thus discrete valuations will be the focus of this section.

Definition 1 (Discrete Valuation Ring (DVR)). A discrete valuation ring is a principal ideal domain O with a unique maximal ideal p 0. 6= Example 1. Consider the ring Z and prime p Z. The localization Z p is a local PID with unique maximal ∈ ( ) ideal pZ(p). Then Z(p) satisfies the definition of a discrete valuation ring. Example 2. Consider k[[x]], the formal power series of the field k. It was shown in Dummit and Foote,

1 2004, 7.3 #3, that all proper ideals of k[[x]] are of the form (xn) (this is actually a corollary to what is shown in that problem). Further (x) is maximal and contains all other proper ideals. It follows then that (x) is the unique maximal ideal in k[[x]] and all other proper ideals are principal. Thus k[[x]] is a DVR.

Definition 2 (Uniformizing Parameter). Let p be the unique maximal ideal of DVR O. Since O is a PID, there exists a prime π in O such that p (π). Such a π is called a uniformizing parameter. = Theorem 7, section 16.2 in (Dummit and Foote, 2004) shows that π is unique (up to multi- plication by a unit) and generates the unique maximal ideal of O. Further in problem 26, in section 7.1 of Dummit and Foote, 2004, (proven in math 566) it is shown that O \p is the set of units in O. Then for any a O \{0}, (a) is an ideal of O. Because (a) is a nonzero ideal of O,(π) ∈ is maximal, and (a) is in a local ring O; there exists some n Z such that (a) (π)n. Notice if ∈ = a is a unit, then n 0 because (a) O. = = Definition 3 (Discrete Valuation). The exponent n used above is the valuation of a, denoted vp(a), it is a function vp : K ∗ Z → where K is the field of fractions of O. The valuation vp can be extended to K by setting vp(0) = . Further vp satisfies the following ∞ I. vp(ab) vp(a) vp(b). = + II. vp(a b) min{vp(a),vp(b)}. + ≥ All valuations can be classified as either archimedean or nonarchimedean. A valuation v is nonarchimedean if v(n) is bounded for every n N. Proposition 3.6 in Neukirch, 1999, ∈ is often incorporated in the definition of discrete valuations, as every discrete valuation is nonarchimedean.

Proposition 1. A valuaion v(x) is nonarchimedean if and only if it satisﬁes v(x y) max{v(x),v(y)}. + ≤ Proof. ( ) The reverse direction is straight forward. Just notice ⇐ v(n) v(1 ... 1) 1 n N. = + + ≤ ∀ ∈ ( ) Now, v(n) N for all n N for some N N. Then for arbitrary x, y K , without loss ⇒ ≤ ∈ ∈ l n l ∈ of generality, consider v(x) v(y). Choose l 0, then v(x) v(y) − v(x). Now applying ≥ ≥ ≤ binomial formula yields

n n X ¡n¢ l n l n v(x y) v l v(x) v(y) − N (n 1)(v(x)) + ≤ l 0 ≤ + = taking the nth root of both sides yields

1 1 1 1 v(x y) N n (1 n) n v(x) N n (1 n) n max{v(x),v(y)} + ≤ + = + The result then follows if letting n . The conclusion is then for any discrete valuation → ∞ (which is nonarchimedean) v(x y) max{v(x),v(y)}. // + ≤

2 The following proposition is particularly useful because it provides a correspondence be- tween prime ideals and valuations in Dedeking rings.

Proposition 2 (Neukirch, 1999). Let O be a noetherian integral domain. O is a Dedekind domain if and only if, for all prime ideals p 0, the localizations Op are discrete valuation rings. 6= When the previous proposition is applied to Z and the fraction ﬁeld Q is consider, the following proposition is gained.

Proposition 3 (Neukirch, 1999). Every valuation of Q is equivalent to one of νp (x) (the nonarchimedean valuations) or ν (x) ∞ (the archimedean valuations).

Finally, since ramiﬁcation occurs via extensions or covers, it will be useful to know how valuations extend.

Deﬁnition 4 (Prolongation of a valuation). If A B are rings with B integral over A, p a prime in A, and q prime in B dividing p with ⊂ ramiﬁcation index eq then vq(x) eqvp(x) is a prolongation of vp with index eq (Serre, 1995). =

2.2 FIELD COMPLETION

With a valuation ν on a field K then for any real number a (0,1) an absolute value on K can ∈ be defined by ( aν(x) x 0 x 6= || || = 0 x 0 = which satisfies the usual conditions of a metric as outlined by Serre, 1995, II.1. A topology is then induced on K via the absolute value metric, and the completion of K with respect to the valuation ν is denoted by Kb. Note also that the metrics induced by different choices of a are topologically equivalent, so the completion is dependent only on ν. Further, ν extends in the completion of K to a valuation (which will continue to be called ν) on Kb (Serre, 1995). The Galois group of Lb/Kb relates to the decomposition group of L/K in the following way.

Proposition 4. For L/K Galois with group G, νp(x) a valuation on K and ωq(x) a prolongation of νp, let L,b Kb be the completions of L and K with respect to ωq and νp respectively. If Dq(L/K ) is the decomposition group of q over p, then Lb/Kb is Galois with Galois group D(L/K ) (Serre, 1995).

Proof. This is corollary 4 to theorem 1 in Serre, 1995, II.3, which shows that [Lb : Kb] e f . As = proven in class D(L/K ) e f . // | | =

3 HIGHER RAMIFICATION GROUPS

Consider now K , the fraction ﬁeld of a Dedekind ring OK , with valuation ν and uniformizing parameter πK . For Galois extension L/K with Galois group G; prolongation of ν, ω, on L;

3 uniformizing parameter πL of ω; and discrete valuation ring OL. The definitions of the higher ramification groups generalizes the definition of the inertia group.

Definition 5 (Higher Ramification Groups, Serre, 1995). For every real number i 1 define the i th ramification group of L/K with respect to ω by ≥ − G G (L/K ) {σ G ω(σ(a) a) i 1 a O }. i = i = ∈ | − ≥ + ∀ ∈ L Note that because ν corresponds to a prime p and ω corresponds to a prime q p, the rami- | fication groups are generalizations of the decomposition and inertia groups. The following is a proposition which helps characterize the higher ramification groups with respect to the presentation of valuations above.

Proposition 5. i 1 For any a O and σ G, ω(σ(a) a) i 1 if and only if σ(a) a mod (π ) + . ∈ L ∈ − ≥ + ≡ L Proof. ( ) ⇒ t x νL(σ(a) a) i 1 σ(a) a π − ≥ + =⇒ − = L y where t i 1 and x, y,πL OL all relatively prime. This follows from the definition of the ≥ + ∈ i 1 valuation in the first section. The equation above then implies σ(a) a 0 mod (π ) + . − ≡ L ( ) ⇐ i 1 t x σ(a) a 0 mod π + σ(a) a π − ≡ L =⇒ − = L y with the same restrictions as above. It follows then that ν (σ(a) a) i 1. // L − ≥ + The condition that ν(σ(a) a) i 1 clearly induces the following filtration on G − ≥ +

{Gi (L/K )}i 1 G 1 G0 G1 .... ≥− = − ⊇ ⊇ ⊇ This ﬁltration can be investigated further due to the fact that i, G /G. ∀ i i 1 i 1 Proof. Consider τ G. By membership of G, ω(τ(a) a) 0 a OL, i.e. τ((πL) + ) (πL) + ∈ − ≥ ∀i ∈1 = for all i. Thus there is an induced automorphism τ¯ of OL/(πL+ ). Now consider φi 1 : G i 1 + → aut(OL/(π + ) such that φi 1(τ) τ¯. This is a group homomorphism with ker (φi 1) Gi , so L + = + = Gi /G for all i. //

The first two ramification groups, G 1 and G0, are already familiar. For i 1 from the def- − = − inition it follows G 1 {σ G ω(σ(a) a) 0 a OL} which is the definition of the decom- − = ∈ | − ≥ ∀ ∈ position group Dπ (L/K ). If πK is fully ramified in the extension then G 1 G. For i 0, G0 K − = = equals the inertia group I(L/K ) by noticing that φ1 is the homomorphism used to originally define the inertia group. Occasionally it is useful to denote the first index of the ramification group in which an element σ G ceases to appear. ∈

4 Definition 6. Let x O such that O O [x]. Such an element is assured to exist by Neukirch, 1999, II.10.4. ∈ L L = K Then for σ G define ∈ i (σ) ω(σ(x) x). L/K = − 1 Noting the argument above which showed G /G, it follows σ,τ G, i (τ− στ) i (σ). i ∀ ∈ L/K = L/K Another simple conclusion which follows from the definition of i is i (σ) i 1 if and L/K L/K ≥ + only if σ G . ∈ i Consider now H G a subgroup, and let K 0 be the fixed field of H so that H Gal(L/K 0). The ⊂ = following relations follow from the above propositions.

Proposition 6. For every σ H, i (σ) i (σ), and H G H (Serre, 1995). ∈ L/K 0 = L/K i = i ∩ i 1 i 1 Proof. Let φi 1 : G Aut(OL/(πL) + ) and ψi 1 : H Aut(OL/(πL) + ) such that φi 1(σ) σ¯ + → + → + = and ψi 1 deﬁned similarly. As shown above, Gi ker (φi 1) and Hi ker (ψi 1). Thus the + = + = + conclusions H H G and i (σ) i (σ) follow. // i = ∪ i L/K 0 = L/K

4 THE UPPER NUMBERINGAND RAMIFICATION JUMP

There is a renumbering of the ramiﬁcation groups which lends itself to other computations. The ordering used above is called the lower numbering while the renumbering introduced below is called the upper numbering. The upper numbering is useful when working with quotient group computations, while the lower numbering works best for subgroup computations (Neukirch, 1999).

Definition 7 (Upper Numbering of the Higher Ramification groups). Consider the function Z s dx t ϕ(s) = = 0 [G0 : Gx ] called the Herbrand function which has inverse map ψ. Then for any real s 1 let Gs G s ≥ − = d e and renumber the ramification groups by Gt (L/K ) G (L/K ) where s ψ(t). = s = As this new numbering allows for noninteger indices, the index at which the ramification group changes is given a name.

Definition 8 (Ramification Jump). t t ² If G (L/K ) G + (L/K ) for any ² 0 then t is called a ramification jump. 6= ≥ These two concepts are important for the final theorem of this paper. Which assures that in certain circumstances, ramification groups only change at integer indexes.

Theorem 1 (Hasse-Arf). i For a ﬁnite abelian extension L/K , the jumps of the ﬁltration {G (L/K )}i 1 are rational inte- ≥− gers.

5 Serre gives a useful interpretation of the Hasse-Arf theorem, namely that if Gi Gi 1 then 6= + ϕ(i) is an integer. This interpretation is helpful in the following example (Serre, 1995).

Example 3. Suppose G is a cyclic group of order pn where p is the characteristic of K¯ . Let G(i) be the sub- n i group of G with order p − . Then there exist integers i0,...,in 1 such that all ramiﬁcation − groups are identiﬁed as follows:

G G G G0 Gi0 0 = i0 = = =

i0 1 i0 i1 Gi 1 ... Gi pi G(1) G + ... G + 0+ = = 0+ 1 = = = = i0 i1 1 i0 i1 i2 2 Gi0 pi1 1 ... Gi pi p i G(2) G + + ... G + + + + = = 0+ 1+ 2 = = = = . .

i0 in 1 1 G 2 n 1 1 G − . i0 pi1 p i2 p − in 1 1 +···+ + + + +···+ − + = = Proof. Because G Gal(L¯/K¯) is cyclic with order pn, it follows that L¯/K¯ is an extension of = finite fields, i.e. L F¯pn and K F¯p . Further recall that G is generated by fröbenius σ (shown =∼ =∼ in Math 567). By Euler’s totient there are (p 1)pn elements in G with order relatively prime n n n 1 − n 1 to p , consequently since p (p 1)p − p − , there are n 1 distinct proper subgroups − − = − of G. This places n as an upper bound on the number of distinct ramification groups. Consider Stitchenoth, 1993, III.8.6, which states that if char K¯ p 0, then Gi 1 / Gi for = > + all i. Then since each Gi is a subgroup of a cyclic group, for each ramification group there exists a σl G which generate G . As argued above there are only n 1 such σl generating ∈ i − distinct proper subgroups of G. This places n as the lower bound on the number of distinct l ramification groups. Thus there are n 1 ramification jumps and Gi (σ ) where 0 l n. l − = n l ≤ < Finally (σ ) Gal(F¯ n /F¯ l ) which is isomorphic to the cyclic group of order p − . // = p p

5 APPLICATIONS

5.1 CYCLOTOMIC EXTENSIONSOF Qp In this example consider the p-adic completion of Q with respect to a valuation (and associ- m th ated metric) νp . Let n p , and ζ be a primitive n root of unity and the extension Qp (ζ)/Qp = m 1 with Galois group G. Recall that [Qp (ζ): Qp ] φ(n) (p 1)p − and G ∼ (Z/n)∗. v = = − = Now, if 0 v m then let G be the subgroup of G isomorphic to the subgroup H (Z/n)∗ ≤ ≤ v v ⊂ such that a 1 mod p for every a H. Then Gal(Qp (ζpm )/Qp (ζpv )) ∼ G because Gal(Qp (ζpv )/Qp ) ∼ v ≡ ∈ = = (Z/p )∗. Using this fact it is straight forward to ﬁnd all ramiﬁcation groups Gi of G (Serre, 1995).

6 Proposition 7. The ramiﬁcation groups Gi of G are  G u 0  =  1 G 1 u p 1  ≤ ≤ − G G2 p u p2 1 u = ≤ ≤ − . .   m 1 {1} p − u ≤

Proof. Consider an element x 1 in G(n) (Z/n)∗ and σx the corresponding element in G. 6= =∼ Let i Z be maximal such that x 1 mod pi . By deﬁnition of G(n)i and maximality of i, ∈ i i 1 ≡ x G(n) but x G(n) + . Now consider ∈ 6∈ q q 1 iQ Q (σ ) ω(σ (ζ) ζ) ω(ζ ζ) ω(ζ − 1). p (ζ)/ p x = x − = − = − q 1 m i q 1 Because ζ − is a primitive p − root of unity, ζ − 1 is a uniformizer of the valuation on − Q (ζ m i ) (this is Serre, 1995, prop. IV.17.iii). Then it follows p p −

m i i iQ ( )/Q (σx ) [Qp (ζpm ): Qp (ζ m i )] #(G(n) − ) p . p ζ p = p − = = k 1 k i If p − u p 1, then σ G if and only if i k. Consequently G G(n) . // ≤ ≤ − x ∈ u ≥ u =

5.2 ARTIN-SCHREIER EXTENSIONSIN POSITIVE CHARACTERISTIC

What follows is the above theory of ramification groups, applied to extensions of function field. Rather than speaking of primes, the theory is rephrased in terms of places. For the purposes here a place πF in the extension F /K is the maximal ideal of a valuation ring O of F /K . Note that in number fields primes correspond to places (Stitchenoth, 1993). The set of all places of an extension is denoted PF . Another important tool is the generalized Hurwitz Genus formula defined in terms of Weil differentials (which are beyond the scope of this paper). Such a generalization yields the following (Stitchenoth, 1993).

Theorem 2 (Generalized Hurwitz Genus Formula). Consider F /K an algebraic function field of genus g and F 0/F a finite separable extension, K 0 denotes the constant field of F 0, and g 0 the genus of F 0/K 0.

[F 0 : F ] 2g 0 2 (2g 2) deg(Di f f (F 0/F )). − = [K 0 : K ] − +

Then the different exponent over a place deg(Di f f (πF 0 /πF )) can be computed using Hilbert’s different formula (Stitchenoth, 1993).

Deﬁnition 9 (Hilbert’s Different Formula).

Consider a Galois extension F 0/F of algebraic function ﬁelds. A place πF of F , and a place πF 0 of F 0 such that π π . Then the different exponent can be calculated F 0 | F X deg(Di f f (πF 0 πF )) ( Gi 1). | = i 0 | | − ≥

7 The different exponent of the extension is then the sum over all places of the extension X X deg(Di f f (F 0/F )) ( Gi 1). = P P i 0 | | − ∈ F 0 ≥ Putting the above together (along with Dedekind’s different formula (Stitchenoth, 1993)) results in the following Wild Riemann-Hurwitz formula.

Definition 10 (Wild Riemann-Hurwitz, Pries, n.d.). For algebraically closed field F which has characteristic p and p e; where e is the ramification | index of place, π , under a degree d covering map φ : X Y of curves over F , and g , g are F → X Y the genera of X and Y respectively.

X X∞ 2g 2 d(2g 2) ¡ G (x) 1¢. Y − = X − + | i | − πF X i 0 ∈ = p t Consider a smooth algebraic curve Xp,t : x x y , where t ap 1 and a Z+. Further 1 − − = − ∈ 1 consider a covering map ψ : Xp,t AF such that ψ(x, y) y. Since ψ is unramified over AF, 1 → = consider now ψ : Xp,t P . Then ψ is only ramified over P . → F ∞ From Hilbert’s different formula it follows that in order to compute the different exponent it is necessary to understand the Galois group of ψ and the ramification groups of P . To do so ∞ consider Aut(X ). Let τ(x) x 1 then if xp x y t 0 it follows = + − − = τ(x)p τ(x) y t (x 1)p (x 1) y t xp 1 x 1 y t xp x y t 0 − − = + − + − = + − − − = − − = because char (F) p. Consequently τ Aut(Xp,t ) and τ Aut(Xp,t ) Z/p which agrees = ∈ < >= =∼ with the conclusion of Stitchenoth, 1993, III.7.10. Then the Galois group of the cover is G = Aut(ψ). Applying the change of variables y¯ 1 , x¯ xy¯a results in a compactification of X where = y = p,t P (0,0). The change of variables yields: ∞ = p t p ap 1 p ap 1 x x y 0 x x y − 0 x x y y− 0 − − = =⇒ − − = =⇒ − − = p ap 1 a p a a(p 1) 1 x y− x y ap y− 0 (x y− ) x y− y− − y− 0 =⇒ − − − = =⇒ − − = p a(p 1) x¯ x¯y¯ − y¯ 0. =⇒ − − = p a(p 1) Thus Xp•,t : x¯ x¯y¯ − y¯ which is ramified over P (0,0). − − ∞ = Notice now that ν (x¯) 1 because x¯ vanishes with order 1 at P so π x¯ is a uniformizing ∞ = ∞ ∞ = parameter over P . Further applying τ, the generator of G, to x¯ results in the following: ∞ τ(x¯) (x 1)y¯a x¯ y¯a = + = + τ(π) π τ(x¯) x¯ y¯a. =⇒ − = − = Consequently it is possible to compute iG (τ) (recall this is the index of the first ramification group in which τ no longer appears):

ν (τ(π) π) ν (y¯a) aν (y¯) ∞ − = ∞ = ∞

8 p p a(p 1) aν (y¯) aν (x¯ x¯ya¯ (p 1)) a min{ν (x¯ ),ν (x¯y¯ − )} ∞ = ∞ − − = · ∞ ∞ ν (τ(π) π) ν (y¯a) a ν (x¯p ) ap. ∞ − = ∞ = · ∞ = What has been calculated is the ramification jump tP ap 1 meaning G G0 ... ∞ = − = = = Gap 1 . Gap Gap 1 ... 1. Because Z/p is simple and Gap / G is a strict inclusion; − = + = = this implies that G {1} and all previous ramification groups are isomorphic to Z/p. Con- ap = sequently it is now possible to calculate the degree of the different with Hilbert’s different formula X X∞ X G 1 (p 1)(ap). | i | − = − P Pψ i 0 P Pψ ∈ = ∈ Where Z/p p and ap is the number of nontrivial ramification groups. Because P is the | | = ∞ only ramified place, the above is simply (p 1)(ap) as all other places have trivial ramification − groups. Thus the genus of X can be accurately calculated (using ap t 1): p,t = + 2p (p 1)(t 1) 2 (p 1)(t 1) 1 2gXp,t 2 p(2gP 2) (p 1)(ap) gXp,t − − − + + − + . − = F − − − =⇒ = 2 = 2 This result can be confirmed using the standard Riemann-Hurwitz formula. For the covering map φ : X A1 where φ(x, y) x which is ramified when x Z/p and y 0 with ramifica- p,t → F = ∈ = tion index t ap 1 (not divisible by p): = − p( 2) p(t 1) 2 (p 1)(t 1) gX − + − + − + . p,t = 2 = 2 Which agrees with the previous computation.

9 REFERENCES

Dummit, D. & Foote, R. (2004). Abstract algebra. John Wiley & Sons, Inc. Neukirch, J. (1999). Algebraic number theory. Springer-Verlag Berlin Heidelberg. Pries, R. (n.d.). Introduction to hurwitz theory in positive characteristic. Unpublished manuscript. Serre, J.-P.(1995). Local ﬁelds. Springer-Verlag New York. Stitchenoth, H. (1993). Algebraic function ﬁelds and codes. Springer-Verlag Berlin Heidelberg.