ebook img

Computation of 2-groups of positive classes of exceptional number fields PDF

0.23 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Computation of 2-groups of positive classes of exceptional number fields

Computation of 2-groups of positive classes of exceptional number fields∗ Jean-Franc¸ois Jaulent, Sebastian Pauli, Michael E. Pohst & Florence Soriano–Gafiuk 8 0 0 2 n Abstract. We present an algorithm for computing the 2-group Cℓpos of the posi- F a tive divisor classes in case the number field F has exceptional dyadic places. As an J application, we compute the2-rank of thewild kernel WK2(F) in K2(F). 9 R´esum´e. Nous d´eveloppons un algorithme pour d´eterminer le 2-groupe CℓFpos des ] classes positives dans le cas ou` le corps de nombres consid´er´e F poss`ede des places T pairesexceptionelles. Celadonneenparticulierle2-rangdunoyausauvageWK2(F). N . h 1 Introduction t a m The logarithmicℓ-classgroupCℓ whasintroducedin[10]byJ.-F.Jaulentwho F [ used it to study the ℓ-part WK2(F) of the wild kernel in number fields: if F contains a primitive 2ℓt-th rooftof unity (t>0), there is a naturalisomorphism 1 v µ ⊗ Cℓ ≃WK (F)/WK (F)ℓt, ℓt Z F 2 2 7 so the ℓ-rank of WK (F) coincides with the ℓ-rank of the logarithmic group 6 2 f 3 CℓF. An algorithm for computing CℓF for Galois extensions F was developed 1 in [4] and later generalized and improved for arbitrary number fields in [3]. 1. f Incasetheprimeℓisodd,theassfumptionµℓ ⊂F maybeeasilypassedifone 0 considers the cyclotomic extension F(µ ) and gets back to F via the so-called ℓ 8 transfer(see[12],[15]and[17]). Howeverforℓ=2theconnectionbetweensym- 0 bols and logarithmic classes is more intricate: in the non-exceptional situation : v (i.e. when the cyclotomic Z -extension Fc contains the fourth root of unity i) 2 i X the 2-rank of WK2(F) still coincides with the 2-rank of CℓF. Even more if the number field F has no exceptional dyadic place (i.e. if one has i ∈ Fc for any r q a q|2), the same result holds if one replace the ordinary lofgarithmic class group Cℓ byanarrowversionCℓres. The algorithmicaspectofthis istreatedin[11]. F F Last in [13] the authors pass the difficulty in the remaining case by intro- fducinganew2-classgroufpsCℓpos,the2-group of positive divisor classes, which F satisfies the rank identity: rk Cℓpos =rk WK (F). 2 F 2 2 In this paper we develop an algorithm for computing both Cℓpos and Cℓpos F F in case the number field F does contain exceptional dyadic places. Weconcludewithseveralexamples. Combiningouralgorithmwiththefwork ofBelabasandGangl[1]onthecomputationofthetamekernelofK weobtain 2 the complete structure of the wild kernel in some cases. ∗Versiondetravaildu03decembre2007 1 2 Positive divisor classes of degree zero 2.1 The group of logarithmic divisor classes of degree zero Throughoutthis paper the prime number ℓ equals 2 and we let i be a primitive fourth root of unity. Let F be a number field of degree n = r+2c. According to [9], for every place p of F there exists a 2-adic valuation v which is related p to the wild 2-symbol in case the cyclotomic Z -extension of F contains i. The 2 p degreedegpofpisa2-adicintegersuchthattheimageoftheemapLog||p isthe Z -module deg(p)Z (see [10]). (By Log we mean the usual 2-adic logarithm.) 2 2 The construction of the 2-adic logarithmic valuations v yields p ∀α∈R :=Z ⊗ F× : v (α)deg(p) = 0, (1) F 2 Z p e p∈XPlF0 e where Pl0 denotes the set of finite places of the number field F. Setting F div(α) := v (α)p p p∈XPlF0 we obtain by Z -linearity: f e 2 deg(div(α)) = 0. (2) We define the 2-group of logarithmic divisors of degree 0 as the kernel of the f degree map deg in the direct sum Dℓ = Z p: F p∈Pl0 2 F P Dℓ := a p∈Dℓ | a deg(p)=0 ; F p∈Pl0 p F p∈Pl0 p F F n o and the subgroup of prPincipal logarithmic diPvisors as the image of the logarith- f mical map div: Pℓ := {div(α)|α∈R } . F F f Because of (2) Pℓ is clearly a subgroup of Dℓ . More ever by the so-called F F f f extended Gross conjecture, the factorgroup f f Cℓ := Dℓ /Pℓ F F F is a finite 2-group, the 2-group of logarithmic divisor classes. So, under this f f f conjecture, Cℓ is just the torsion subgroup of the group F Cℓ :=Dℓ /Pℓ f F F F of logarithmic classes (without any asumption of degree). f Remark 1. Let F+ be the set of all totally positive elements of F× (i.e. the subgroup F+ :={x∈F×|x >0 for all real p}. For p Pℓ+ := {div(α)|α∈R+ :=Z ⊗ F+} F F 2 Z the factor group f f Cℓres := Dℓ /Pℓ+ (resp. Cℓres := Dℓ /Pℓ+) F F F F F F is the 2-group of narrow logarithmic divisor classes ofthe number field F (resp. f f f f the 2-group of narrow logarithmic divisor classes of degree 0) introducedin [16] and computed in [11]. 2 2.2 Signs and places For a field F we denote by Fc, (respectively Fc[i]) the cyclotomic Z -extension 2 (resp. the maximal cyclotomic pro-2-extension) of F. We adopt the notations and definitions in this section from [13]. Definition 1 (signed places). Let F be a number field. We say that a non- complex place p of F is signed if and only if F does not contains the fourth p root i. These are the places which do not decompose in the extension F[i]/F. Wesaythatpislogarithmically signedifandonlyifthecyclotomicZ -extension 2 Fcdoesnotcontaini. ThesearetheplaceswhichdonotdecomposeinFc[i]/Fc. p Definition 2 (sets of signed places). By PS, respectively PLS, we denote the sets of signed, respectively logarithmically signed, places: PS := {p|i6∈F } , p PLS := {p|i6∈Fc} . p A finite place p ∈ PLS is called exceptional. The set of exceptional places is denoted by PE. Exceptional places are even (i.e. finite places dividing 2). These sets satisfy the following inclusions: P⊂PLS =PE∪PR⊂Pl(2)∪Pl(∞) where Pl(2), Pl(∞), PR denote the sets of even, infinite and real places of F, respectively. From this the finiteness of PLS is obvious. We recall the canonical decomposition Q× =2Z×(1+4Z )×h−1i and we 2 2 denote by ǫ the projection from Q× onto h−1i. 2 Definition 3 (sign function). For all places p we define a sign function via 1 for pcomplex sign(x) for preal sg : F× →h−1i : x7→ . p p  ǫǫ((NNKp−p/νQp2(x()x))Np−νp(x)) ffoorr pp6||22∞ These sign functions satisfy the product formula: ∀x∈F× sg(x)=1. p∈YPlF In addition we have: Proposition 1. The places p of F satisfy the following properties: (i) if p∈PLS then (sg ,v ) is surjective; p p (ii) if p∈PS\PLS then esgp( )=(−1)vep( ) and vp is surjective; (iii) if p6∈PS then sg (F×)=1 and v is surjective. p p p e Remark 2. The logarithmic valuation v is surjective in all three cases. Part p e 2 of the preceding result is often used for testing p∈PLS. e 3 2.3 The group of positive divisor classes For the introduction of that group we modify several notations from [13] in order to make them suitable for actual computations. Since PLS is finite we can fix the order of the logarithmically signed places, say PLS = {p ,··· ,p }, with PE = {p ,··· ,p } and PR = {p ,··· ,p }. 1 m 1 e e+1 m Accordingly we define vectors e=(e ,··· ,e )∈{±1}m. 1 m For each divisor a= a p, we form pairs (a,e) and put p∈Pl0 p F P m sg(a,e) := (−1)ap × e (3) i p∈PYS\PLS Yi=1 LetDℓ (PE):= a∈Dℓ a= a p betheZ -submoduleofDℓ gen- F F p∈PE p 2 F erated by the exnceptional (cid:12)dyadiPc places. Aond let DℓPFE be the factor group (cid:12) Dℓ /Dℓ (PE). Thus the group of positive divisors is the Z -module: F F 2 Dℓpos := (a,e)∈DℓPE ×{±1}m sg(a,e)=1 (4) F F n (cid:12) o For α∈RF :=Z2⊗ZF×, let div′(α) denotes the i(cid:12)(cid:12)mage of div(α) in DℓPFE and sg(α) the vector of signs (sg (α),...,sg (α)) in {±1}m. Then p1 pm f f Pℓpos := (div′(α),sg(α))∈DℓPE ×{±1}m α∈R (5) F F F n (cid:12) o is obviouslyfa submodulefof Dℓpos which is called the pri(cid:12)ncipal submodule. F (cid:12) Definition 4 (positive divisor classes). With the notations above: (i) The group of positive logarithmic divisor classes is the factor group Cℓpos = Dℓpos/Pℓpos . F F F (ii) The subgroup of positive logarithmic divifsor classes of degree zero is the kernel Cℓpos of the degree map deg in Cℓpos: F F f CℓFpos :={(a,e)+PℓFpos | deg(a)∈deg(DℓF(PE))}. Remark 3. ThefgroupCℓpos isinfifnitewheneverthenumberfieldF hasnoex- F ceptionalplaces,sinceinthiscasedeg(Cℓpos)isisomorphictoZ . Thefiniteness F 2 ofCℓpos incasePE 6=∅followsfromthe so-calledgeneralizedGrossconjecture. F For the computation of Cℓpos we need to introduce primitive divisors. F Definition 5. A divisor b of F is called a primitive divisor if deg(b) generates f the Z -module deg(Dℓ )=4[F ∩Qc :Q]Z . 2 F 2 We close this section by presenting a method for exhibiting such a divisor: Let q ,··· ,q be all dyadic primes; and p ,··· ,p be a finite set of non- 1 s 1 s dyadic primes which generates the 2-group of 2-ideal-classes Cℓ′ (i.e. the quo- F tient of the usual 2-class group by the subgroup generated by ideals above 2). Then every p ∈ {q ,··· ,q ,p ,··· ,p } with minimal 2-valuation ν (degp) 1 s 1 t 2 is primitive. 4 2.4 Galois interpretations and applications to K-theory Let Flc be the locally cyclototomic 2-extension of F (i.e. the maximal abelian pro-2-extensionofF whichiscompletelysplitateveryplaceoverthecyclotomic Z -extensionFc. Thenbyℓ-adicclassfieldtheory(cf. [9]),onehasthefollowing 2 interpretations of the logarithmic class groups: Gal(Flc/F)≃Cℓ and Gal(Flc/Fc)≃Cℓ . F F Remark 4. Letusassumei∈/ Fc. Thuswemaylistthefollowingspecialcases: f (i) In case PLS =∅, the group Cℓpos ≃Z ⊕Cℓpos of positive divisor classes F 2 F hasindex2inthegroupCℓ ≃Z ⊕Cℓ oflogarithmicclassesofarbitrary F 2 F degree; as a consequence its torsion subgrfoup Cℓpos has index 2 in the F finite group Cℓ of logarithmic classfes of degree 0 yet computed in [3]. F f (ii) In case PE = ∅, the group Cℓpos ≃ Z ⊕Cℓpos has index 2 in the group f F 2 F Cℓres ≃Z ⊕Cℓres of narrow logarithmic classes of arbitrary degree; and F 2 F its torsionsubgroupCℓpos has index 2 in tfhe finite groupCℓres of narrow F F logarithmic clfasses of degree 0 introduced in [16] and computed in [11]. f f Definition 6. We adopt the following conventions from [6, 7, 13, 14]: (i) F is exceptional whenever one has i∈/ Fc (i.e. [Fc[i]:Fc]=2); (ii) F is logarithmically signed whenever one has i∈/ Flc (i.e. PLS 6=∅); (iii) F is primitive whenever one at least between the exceptional places does not split in (the first step of the cyclotomic Z -extension) Fc/F. 2 The following theorem is a consequence of the results in [6, 7, 9, 10, 13, 14]: Theorem 1. Let WK (F) (resp. K∞(F) :=∩ K2n(F)) be be the 2-part of 2 2 n≥1 2 the wild kernel (resp. the 2-subgroup of infinite height elements) in K (F). 2 (i) In case i∈Flc (i.e. in case PLS =∅), we have both: rk WK (F) = rk Cℓ = rk Cℓres. 2 2 2 F 2 F (ii) In case i∈/ Flc but F has no exceptionfal places (if.e. PE =∅), we have: rk WK (F) = rk Cℓres. 2 2 2 F (iii) In case PE 6=∅, then we have f rk WK (F) = rk Cℓpos. 2 2 2 F And in this last situation there are two subcases: (a) If F is primitive, i.e. if the set PE of exceptional dyadic places contains a primitive place, we have: K∞(F) = WK (F) . 2 2 (b) If F is imprimitive and K∞(F)=⊕n Z/2niZ, we get: 2 i=1 i. WK2(F)=Z/2n1+1Z⊕(⊕ni=2Z/2niZ) if rk2(CℓFpos)=rk2(CℓFpos); ii. WK2(F)=Z/2Z⊕(⊕ni=1Z/2niZ) if rk2(CℓFpos)<rk2(CℓFpos). f f 5 3 Computation of positive divisor classes We assume in the following that the set PE of exceptional places is not empty. 3.1 Computation of exceptional units Classicallythe groupoflogarithmicunits is the kernelin R ofthe logarithmic F valuations (see [9]): E ={x∈R |∀p:v (x)=0} F F p In order to compute poseitive divisor classes ien case PE is not empty, we ought to introduce a new group of units: Definition7. Wedefinethegroupoflogarithmic exceptionalunitsasthekernel of the non-exceptional logarihtmic valuations: Eexc ={x∈R |∀p∈/ PE :v (x)=0} (6) F F p We just know thaet is a subgroup of the 2-greoup of 2-units EF′ = Z2⊗EF′ . If we assume that there are exactly s places in F containing 2 we have, say: E′ = µ ×hε ,··· ,ε i F F 1 r+c−1+s For the calculationofEexc we use the sameprecision η as for our2-adicapprox- F imations usedinthe courseofthe calculationof Cℓ . Thenweobtaina system F of generators of Eexc byecomputing the nullspace of the matrix F f e | 2η ··· 0 B = v (ε ) | · ··· ·  pi j  | 0 ··· 2η  e  with r+c−1+s+e columns and e rows,where e is the cardinality of PE and the precision η is determined as explained in [3]. We assume that the nullspace is generated by the columns of the matrix C   B′ =  − − −         D          where C has r+c−1+s and D exactly e rows. It suffices to consider C. Each column (n ,··· ,n )tr of C corresponds to a unit 1 r+c−1+s r+c−1+s εni ∈EexcR2η i F F i=1 Y e 6 so that we can choose r+c−1+s ε:= εni i i=1 Y asanapproximationforanexceeptionalunit. Thisprocedureyieldsk ≥r+c+e exceptional units, say: ε ,··· ,ε . By the so-called generalized conjecture of 1 k Grosswewouldhaveexactlyr+c+esuchunits. Soweassumeinthefollowing that the procedure does give k = r + c + e (otherwise we would refute the e e conjecture). Hence, from now on we may assume that we have determined exactly r+c+e generators ε ,··· ,ε of Eexc, and we write: 1 r+c+1 F EeFxc =e h−1i×ehε1,··· ,eεr+c−1+ei Definition 8. The kerenel of the canoneical maep RF → DℓFpos is the subgroup of positive logarithmic units: Epos = {ε∈Eexc |∀p∈PLS sg (ε)=+1} F F p The subgroupeEFpos hasefinitee index in the group EeeFxc of exceptional units. 3.2 The algoerithm for computing Cℓpose F We assumePE 6=∅ andthatthe logarithmic2-classgroupCℓ is isomorphicto F the direct sum Cℓ ∼= ⊕ν Z/2niZ f F i=1 subject to 1≤n ≤···≤n . Leta (1≤i≤ν) be fixedrepresentativesofthe 1 ν i f ν generating divisor classes. Then any divisor a of Dℓ can be written as F ν a = a a +λb+div(α) i i i=1 X f with suitable integers a ∈ Z , a primitive divisor b, λ = deg(a) and an appro- i 2 deg(b) priate element α of R . With each divisor a we associate a vector F i e := (sg(a ,1),1,··· ,1)∈{±1}m , i i where m again denotes the number of divisors in PLS. Clearly, that represen- tation then satisfies sg(a ,e ) = 1, hence the element (a ,e ) belongs to Dℓpos. i i i i F Setting e =(sg(b,1),1,··· ,1) as above and writing b ν e′ := sg(α)× eai ×e×eλ i b i=1 Y for abbreviation any element (a,e) of Dℓpos can then be written in the form F ν ν (a,e) = a a +λb+div(α),e′× eai ×sg(α)×eλ i i i b ! i=1 i=1 X Y ν f = a (a ,e )+λ(b,e )+(0,e′)+(div(α),sg(α)) . i i i b i=1 X f 7 The multiplications are carried out coordinatewise. The vector e′ is therefore contained in the Z -module generated by g ∈ Zm (1 ≤ i ≤ m) with g = 2 i 1 (1,··· ,1), whereas g has first and i-th coordinate -1, all other coordinates 1 i for i>1. As a consequence, the set {(a ,e )|1≤j ≤ν}∪{(0,g )|2≤i≤m}∪{(b,e} j j i contains a system of generators of Cℓpos ( note that (0,g ) is trivial in Cℓpos). F 1 F We still need to expose the relations among those. But the latter are easy to characterize. We must have ν m a (a ,e )+ b (0,g )+λ(b,e ) ≡ 0 mod Pℓpos , j j j i i b F j=1 i=2 X X ν m f a (a ,e )+ b (0,g )+λ(b,e ) = (div(α),sg(α))+ (d p,1) j j j i i b p j=1 i=2 p∈PE X X X f with indeterminates a ,b ,d from Z . Considering the two components sepa- j i p 2 rately, we obtain the conditions ν a a +λb ≡ d p mod Pℓ (7) j j p F j=1 p∈PE X X f and ν m eaj × gbi ×eλ = sg(α) . (8) j i b j=1 i=2 Y Y Let us recall that we have already ordered PLS so that exactly the first e ele- ments p ,··· ,p belong to PE. Then the first one of the conditions above is 1 e tantamount to ν e degp a a ≡ d p − ib mod Pℓ . j j pi i degb F j=1 i=1 (cid:18) (cid:19) X X f The divisors degp p − ib i degb on the right-hand side can again be expressed by the a . For 1≤i≤e we let j ν degp div(α ) + p − ib = c a . i i degb ij j j=1 X f The calculation of the α ,c is described in [15]. i ij Consequently, the coefficient vectors (a ,··· ,a ,λ) can be chosen as Z - 1 ν 2 linear combinations of the rows of the following matrix A∈Z(ν+e)×(ν+1): 2 8 2n1 0 ··· 0 0 | 0 0 2n2 ··· 0 0 | 0   · · ··· · · | ·  · · ··· · · | ·     0 0 ··· 2nν−1 0 | 0    A =  0 0 ··· 0 2nν | 0     −− −− −−− −− −− −−−     | deg(p1)   deg(b)   .   c | ..   ij   | deg(pe)   deg(b)    Eachrow(a ,··· ,a ,λ) ofA correspondsto alinear combinationsatisfying 1 ν ν a a +λb ≡ div(α) mod Dℓ (PE) . (9) j j F j=1 X f Condition (8) gives m ν gbi = sg(α)× eaj ×eλ . (10) i j b i=2 j=1 Y Y Obviously, the family (g ) is free over F implying that the exponents i 2≤i≤m 2 b are uniquely defined. Consequently, if the k-th coordinate of the product i sg(α)× ν eaj×eλis−1wemusthaveb =1,otherwiseb =0for2≤k ≤m. j=1 j b k k (Wenotethattheproductoverallcoordinatesisalways1.) Therefore,wedenote Q by b ,··· ,b the exponents of the relation belonging to the j-th column of 2,j m,j A for j =1,··· ,ν+e. Unfortunately,theelementsαareonlygivenuptoexceptionalunits. Hence, we must additionally consider the signs of the exceptional units of F. For Eexc = h−1i×hε ,··· ,ε i (11) F 1 r+c−1+e we put: e me e sg(ε )= gbi,j+v+e . (12) j i i=1 Y Using the notations of (11)eand (12) the rows of the following matrix A′ ∈ Z(ν+e+r+c)×(ν+m) generate all relations for the (a ,e ), (b,e ), (0,g ). 2 j j b i | b2,1 ··· bm,1 0 1 | · ··· · B A | · ··· · C B C B | · ··· · C BB | b2,ν+e ··· bm,ν+e CC A′ = BB − −−− − | − −−− − CC . BBB | b2,ν+e+1 ··· bm,ν+e+1 CCC B | · ··· · C B C B O | · ··· · C B C B | · ··· · C B C @ | b2,ν+e+r+c ··· bm,ν+e+r+c A 9 3.3 The algorithm for computing Cℓpos F We assume that PE = {p ,··· ,p } 6= ∅ is ordered by increasing 2-valuations 1 e v (degp ); that the group Cℓpos of positive diviseor classes is isomorphic to the 2 i F direct sum Cℓpos ∼= ⊕w Z/2miZ; F i=1 andthatweknowafullsetofrepresentatives(b ,f ) (1≤i≤w)forallclasses. i i Then each (b,f)∈Dℓpos satisfies deg(b)∈deg(Dℓ (PE)) and F F w f b ≡ b b mod(Dℓ (PE)+Pℓ ) . i i F F i=1 X f Obviously, we obtain w 0 ≡ deg(b) ≡ b deg(b ) mod deg(Dℓ (PE)) . i i F i=1 X We reorder the b if necessary so that i v (deg(b )) ≤ v (deg(b )) (2≤i≤w) 2 1 2 i is fulfilled. We put t := max(min({v (deg(p))|p∈Dℓ (PE)})−v (deg(b )),0) 2 F 2 1 = max(v (deg(p ))−v (deg(b ),0) 2 1 2 1 and w deg(b ) i δ := b + b . 1 deg(b ) i 1 i=2 X Then we get: w deg(b ) b ≡ b b − i b +δb mod (Dℓ (PE)+Pℓ ) i i deg(b ) 1 1 F F i=2 (cid:18) 1 (cid:19) X f and so degb≡0≡ b ×0+δdegb moddegDℓ (PE). i 1 F From this it is immediateXthat a full set of representatives of the elements of Cℓpos is given by F f b − deg(bi)b ,f ×f−deg(bi)/deg(b1) for 2≤i≤w i deg(b ) 1 i 1 (cid:18) 1 (cid:19) and degb (b′ :=2tb −2t 1p ,f2t) . 1 1 degp 1 1 1 Let us denote the class of (c,f) in Cℓpos by [c,f]. F f 10

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.