divisors on division algebras and error correcting codes

12
This article was downloaded by: [University of Kent] On: 18 November 2014, At: 01:54 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK Communications in Algebra Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/lagb20 Divisors on division algebras and error correcting codes Patrick J. Morandi a & B.A. Sethuraman b a Department of Mathematical Sciences , New Mexico State University , Las Cruces, NM, 88003 E-mail: b Department of Mathematics , California State University , Northridge, Northridge, CA, 91330 E-mail: Published online: 05 Jul 2007. To cite this article: Patrick J. Morandi & B.A. Sethuraman (1998) Divisors on division algebras and error correcting codes, Communications in Algebra, 26:10, 3211-3221, DOI: 10.1080/00927879808826337 To link to this article: http://dx.doi.org/10.1080/00927879808826337 PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions

Upload: ba

Post on 24-Mar-2017

213 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Divisors on division algebras and error correcting codes

This article was downloaded by: [University of Kent]On: 18 November 2014, At: 01:54Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: MortimerHouse, 37-41 Mortimer Street, London W1T 3JH, UK

Communications in AlgebraPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/lagb20

Divisors on division algebras and error correctingcodesPatrick J. Morandi a & B.A. Sethuraman ba Department of Mathematical Sciences , New Mexico State University , Las Cruces,NM, 88003 E-mail:b Department of Mathematics , California State University , Northridge, Northridge,CA, 91330 E-mail:Published online: 05 Jul 2007.

To cite this article: Patrick J. Morandi & B.A. Sethuraman (1998) Divisors on division algebras and error correctingcodes, Communications in Algebra, 26:10, 3211-3221, DOI: 10.1080/00927879808826337

To link to this article: http://dx.doi.org/10.1080/00927879808826337

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purposeof the Content. Any opinions and views expressed in this publication are the opinions and views of theauthors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should notbe relied upon and should be independently verified with primary sources of information. Taylor and Francisshall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, andother liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relationto or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Page 2: Divisors on division algebras and error correcting codes

COMMUNICATIONS IN ALGEBRA. 26(10). 32 11-3221 (1998)

Divisors on Division Algebras arid Error Cor*re<:tirig Codes

Patrick J . Morandi Department of Matherriatical Sciences

New Mexico State University Las Cruces, NM 88003 [email protected]

B. A. Sethuraman' Department of Mathematics

California State University, Northridge Northridge, CA 91330

[email protected]

1. Introduction

Recall that a (linear) code is just a k-subspace C of kn, wlwre k is some finite field. The eleirlents of C are referred to as codewords, and the ditnenszon of thr code, dinl(C), is just the dimension of C as a k-space. The length of the code is tlie tiitncnsion of the ambient space k"; that is, the length of C is n. The minimum dzslance, d(n, b) brtwee~i two codewords a - ( a l , . . . ,%) and b = (bl,. . . ,b,,) is defined by d(a,b) = j { i 1 u, f 6 , } / , and the mznzmuin dastance of the code, d(C), is defined by d(C) = min{d(a,h) I r ~ , 6 E C , u + 6 ) . The wezght, u:(n), of a codeword a = (a,, . . . ,a,) is defined by ui(a) = I{, / u, # 011. 'Thr liilearity of thc code ensures that d(G) is also equal to rnin{w(a) I a E C, a # 0).

A major goal in coding theory is to construct codes whose ~liinimuin drstance and di- rnension are large relative to their length. Goppa Codes were introduced by Col)pa in 131 (in a form dual to the one we consider here), and are effective at aclrrevirig this goal. One starts with a smooth geometrically irreducible project,ive curve X / k ( i d thus, a function field F lk ) , and a divisor E = PI +. . . + Pn, where the P, are prinw rlivisors of degree 1. One fixes a divisor G' such that the support of E and G' are disjoint (this guilrantres that any f E L(G) is auton~atically in each valuation ring Opt). 011c thr.11 coi~sidtv t l ~ r rvaluat,ion

'Supported in part by a grant from the N S.F.

Copyr~ght 8 1998 by Marcel Dekkcr, Inc

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 3: Divisors on division algebras and error correcting codes

32 12 MORANDI AND SETHURAMAN

map ev : C(G) -. k." given by 1 I- PI), . . . , fie,)), where by J(P , ) , wc mean the residue of J with rcspect to t l ~ e valuation at P,. Since ev is k-linear, the Inlage is a k-subspace of k"; this is the Goppa code associated to the curvc X / k and the tlivisors E and G.

We provide in this paper an analogous construction of codes using division algebra 1) defined over the function field Flk . The structure of such a I) is of course well-understood Crorn class field t.heory; we cxploit this structure in our constructlun. I'laces of the field F/k are now replaced by maximal ordrrs, selected to form 8 shenJ of ~ ; L X I I I I A ~ ordc:rs over X . By work of various aulhors (see I!)], [ I I ] , and (12]), one has an a~lalogous definition of divisors i r ~ this situaLion, as well as il dehiLiw of Lire spaces L(G) for. an) clivis~o~~ algebra divisor G, and a noncommutative Riemann-Roch Theorem.

Our code is actually an I-code, where 1 is a finite extension of 6. that xises naturally as the residue of D. The function field 1F is a maximal subfield of uur d~vision algebra; we prove that under certain conditions our 1-code is the same as the co~nn~utative Coppa code obtained by restricting our division algebra divisors in a suitable mallner 1.0 l F .

2. Maximal Orders and Divisors

As mentioned in the introduction, we will use maximal orders i r ~ our co~~struction of codes LVc recall here the basic properties of maximal orders over discrete vuluat.iur~ rings. A good reference for maximal orders is Reiner's book [5]. Let A be an integral domain with quotient, field F and let S be an F-r:entml simple algebra. A subring B of S wit11 A > A i s M I

A-order if B is a finitely-generated A-module and if BF = S. If B is maxitnal with respect to inclusion among all A-orders in S , then B is said to be a inminc~l wrder over A.

If A is a discrete valuat.ion ring with quotient field F, then the following properties hold for ~naximal A-orders in an F-central simple algebra S:

I. Maximal A-orders exist dncl are unique up to mnjugation.

2. I f B is a maxinlal A-order in S, then the Jacobson radical J ( B ) is tlw unique mmimal ideal of B , and every ideal of B is a power of J ( B ) .

3. There is a b E R with J ( B ) = B6 = 6B.

4 . Thc residue ring = B / J ( B ) is simple and finite d~mens~onal ovi,~ 3 A/.I(A)

I'roofs of these properties can be found in 15, Theorems 18.3, 18.71. In this paper we need to make use of value functions. If A is a discrete valuation rlng

and if B is a maximal A-order In 5, then we can define a functiou u! : S - { O ) - Z in the following way: For b E B , we set w(b) = n if BbB = J ( B ) ^ . More generally, if s E S, write s =- bn-I with 6 E B and cr E A, and set w(s) = w(b) - w ( a ) . It is not I~ard to see that u! is well defined and that U J satisfies the following properties:

2. st) > w ( s ) 1 r t : ( l ) , and w(st) = w(s) + w(t ) i f s E Z(S);

3. LI = { s E S I w ( s ) > 0 } U {0) and J ( B ) = {s E S 1 w(s) > O } u ( 0 )

Moreover, ui is uniquely rlet~rmirlcd by R. We call w the value ~ I I I I ( . I I O I I O I I .S (:orr.esponCI- 111:: t o 8. We will der~olc. by 1'13 thc uulue group of B; that 15, r B I I I I ( W ~ , ) . LL. We refer t l ~ c rcarl~r to [ G , Src. 21 for more inforn~ation about value functio~~s

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 4: Divisors on division algebras and error correcting codes

DIVISORS ON DIVISION ALGEBRAS 3213

Suppose that X is a smootlr gtumet.rically irreducible project.ive curve defined over a field k, and let. F be t,he function field of A'. Let S be a central simple F-algebra, and suppose that we have a maximal Or-ordcr Bp In S for each P E S. Let wp hc the value function on S corresponding to Bp. Not.e that wp depends on both tlle point P and t,he choice of maximal order Rp. Let ep be tlrc index [Z! : wp(F*)]. We will call e p the mmijication index of S/F' wit11 respect to 1'.

We will be working wit.11 sheaves of maximal orders on X . We cirll a sheaf A of rings on X a sheaf of maximal orders 111 S i f A(U) is a maximal O(C1)-order in S for every proper open subset Cl of S. Let S be t.lre cunstarrt sheaf S (U) - S on X. \Ve recall that any sheaf of maxintal orders on S is isoinoiphic to a subsheaf of S. Here is a sketch of the argument. Let ( be the generic point of X . I f A is a sheaf of Ox-maximal orders in S , then thc structure maps pl, : A([/) - A( extend by bilinearity to a map 6" : A(U)%o(IilC)~ + A<. Now Oi - F, and for every proper open set (1 , A([/) is a maximal O(U) order, so the map A(U) i30(l~l F -+ S induced by the inclusion A((/) 2 S is an isomorphism. For every proper open set. I J , we thus have a map Oh : S --. A<. 8; is an injection as S is simple. Since any z E Ai is in pv(A(V)) for some open V, we find z E &(A(V) Bo(vr O1). We may assunle V C_ U. 'The structure map puv : A(U) -+ A(V) induces an isomorphism A(U) @o(") O( 2 A(V) @ q v 1 U( (as each side is isomorphic to S ) , so we find z E &(A(U) BO(") O() as well. Thus, A< S. We thus have injective maps ipc, : A(U) - Ac Z S for each proper open set 11 X . Usmg the sheaf property of A(X) and the fact that the structure map ipx : A(X) -. A( factors through A(U) for every proper open set U, it, is easy to see that 9 x is injective as well. Since pu = p v p u ~ for open sets V C U, we have an injective sheaf map from A to the constallt sheaf S.

Lemma 1. Suppose for eaeq P E X that there is a maximal Op-order Bp zn S, and let .wp be the value Junction on S corresponding to Bp. Then, for each s E S , we have wp(s) 2 0 for all but finitely many P zf and only if there is a sheaf A of mazimul Ox-orders on X such that for each P E A', the stalk Ap zs equal to Bp.

Proof: Suppose there is a sheaf A 2 S of maximal orders such that A p -- Bp for each P E X. Let s E S. If GI is any proper open subset of X, we may write s - bn-I with b E A(U) and a E O(U). There is a nonempty open subset V of U such that a-I E O(V). so s E A(V). Thus, for cach P E V we have w p ( s ) 2 0. Since the complement of V in S is finite, wp(s) 2 0 for all but finitely many P.

Conversely, suppose that we have a maximal Op-order Bp for each P E S, and that i f wp is the value function associated to Bp, then, for any s E S, we have wp(s) 2 0 for a11 but finitely many P. We define a sheaf A on A' by

for all open subsets li of A' witl~ strlict.ure maps being the obvious inclusions. I t is proved in (7; Ex. 3.41 that A([/) is a 111i~xi11la1 O(l1)-order for each proper open subset U , and so A C S is a sheaf of maximal orders o11 S. hlorcover, in 17, Ex. 3.41, it is shown that A(U)Op . - 13p

i f P E U. This shows t.lrizt A p -- Bp for each P. I

We now describe the divisor group on S and state a geileralizaticin of the R~e~lrairn-Rocli theorem. If S is a central s i i~~ple F-algebra, let A be a sheaf of nlaximal Ox-orders 011 A . For each P let wp be the value function on S corresponding to the maxinlal Op-odrr A!'. The dzznsor group div(.S) is thc: Irtv abelian group on the set {AP}p,x. \?;hilt> lhis groi~i) depends on the cho~ce of slivnf A , we will not deal with more thair olir slie,tf o f mnx~ind orders a t a tiine, so wc w~ll not i ~ ~ d to worry about this dependence. For E = Cb,npi1p. u,r define deg(E) - C p n p [ x p : kI arid supp(E) = {Ap I np # 0). Each s E S defines a divisor

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 5: Divisors on division algebras and error correcting codes

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 6: Divisors on division algebras and error correcting codes

DIVISORS O N DIVISION ALGEBRAS 3215

some cr E k[t] - (0). Then d E A p for all P such that P f Ib, and rr-' E Up.) With this sheaf, we verify that the divisor of 1 t j is equal to -P,, so ( 1 -4 j) 11,s negative degree. By Proposition 9 ahead (with z = j ) , we find that I and J form a stro~iglp orthogonal basis for D / k ( t ) ( & ) , so wp(l + j ) = min(wp(l), wpb) ) for all P E S. Since 1 I S a unit in O p for all P except those that correspond to t or lit, and since I - ' = 1 - ' j. we f i l d that J conjugates A p to itself for all such P. By (6. Prop. 2.61, wp(j) - 0 for all sucl~ P . As for Po and P,, the maximal orders at these points arr valuation rings, and ulp,,(j) - 1 . a ~ ~ d ,wp_(j) = -1. Thus, (1 + j ) - -P, as claimed.

We define a map Gs : div(F) -. div(S) by # s ( C p n p P ) - C, , (n t epAp) The main property of this map that we need is given in the following lemma.

Lemma 3. I f E E div(F), then deg(ds(E)) I- IS : F] tlegf(b')

Proof: If E = Cp n p P , then we have

I

If E , E' are divisors on S, we write E 2 E' if E - E' = C npAr with each np 2 0. Let L (E) = {s E S 1 (s) + E 2 01, a k-subspace of S . As a consequence of Lemma 2, we point out that if deg(E) < 0, then L(E) = 0. To see this, i f f E L(E) is nonzero, then (1) + E 2 0, so deg(( f ) + E ) = deg((f))+deg(E) 5 deg(E) < 0, a contradiction to the condition ( f ) + E 2 0. Let C be a canonical divisor on F , and set K = ds(C) + C p ( e p - l)Ap E div(S). We now state the generalization of the Riemann-Roch theorem to this sett,ing.

Theorem 4 (Riemann-Roch) Let g = max~,d,,~s){deg(E) t 1 - dim(L(E))}. Then g ezisls, and for all divisors E , we have dinl(C(E)) = deg(E) t 1 - g t dim(K - E) .

A proof of this theorem c a ~ be found in 112, Sat,z 121 or [ I 1. 'rheore~ri 3.171. We would like to thank J.-L Colliot-ThClkne for telling us about Wit t . '~ paprr. Analogous to the commutative case, we have deg(K) .= 29 - 2. By the definition of K and [ ~ n i m a 3, if g~ is the genus of F, we see that

In the next section, in order to construct sheaves of ntaximnl ortlers. we will use t,he fact that. i f IJ is a proper open subset of S, t.1ic11 Op is H I)t:tlek~~~tl domain of F. This is a st,andard result, but we give a proof here for the conveniencr of tlw reader.

Lemma 5. If U zs a proper open subset of X, then Of. 1 5 tr 1)rdoLmd doirmn o j F

Proof: Let X- U = {QI. . . . . Q,,} , let E - C, n,Q, 6 tliv(1;'). ,u~rI I d I;, ~- E-Q,, where the n, are positive integers. I f we c110ost: the n, large crrougl~. ~ I I < , I I l)y ( I I V Riemann-hch theorem, dim(C(E)) is larger than ~itclt tli~~l(L(b;,)) sincc tlcp,(I:') > tltag((.:,) for. each r . Each of these are finite-dimensional k-vector sprc,es, so C(E) propc.1.l~ c.olllrnlls the union U, L(E,).

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 7: Divisors on division algebras and error correcting codes

3216 MORANDI A N D SETHURAMAN

C'hoose L L. L ( E ) such that x I$ Lit.:,) for each i. The~t (x) -t I;' 0 hut ( J ) i E, 2j 0 for r x h 1. 'I'lris forces vp,(x) = -71, for each i , and vp(x) 2 0 for well 1' E ( 1 . C'or~sequentl~, t l ~ e valuation rings of F that contain x are precisely the Okb for /' E ( 1 . Note t.lrat x @ k, so F IS algebraic over k(x). A valuation ring of F/k conhins 2 if and only i f i t . contains k[x], so we see that the integral closure of k[z] in F is ope, Op. IIOWCVCI.. S I I I C ~ kl:rj is a Dedekind domain of k(x), the ring npEI, Up is a Dedekind t1o111ai11 of F by 15. 'Tlicwrenr 4.41. 1

3. The Construction of the Code

Ixt F/k be a function field in one variable over the fi~titc. fivltl k . Oiic 11;~s thr following scqucnce from class field theory (see the discussion in 15, pages 277 2781 for il~sta~rce)

where the sum in the middle term is over all the places of the field F'. Fir ,irry D E Br(F), D aF 1,'p is trivial almost everywhere; thus the second n ~ a p in thr sequence above, which is simply the sum of the various restriction maps, is well-defined. It IS well-known that for each P, there is a canonical map Inv : Br(Fp) - Q/Z which is an isonlorphisn~, and the third map above is just the sum of the Inv maps for each P. 'I'he ntap I I I V can I x explicitly described as follows: any division algebra D/Fp of degree T is isoniorphic to the cyclic algebra (W/Fp,n , T ' ) , where W is the unique unraniified field extension of Fp of d e g r t ~ rr, n is a uniformizer for Fp, u is a generator of the Galois group of the cyclic field extension W/Fp chosen so as to induce the Frobenius automorphism at the residue level, and 1 < i < r with gcd(i,r) = I . With this isomorphism at hand, Inv(D) is defined to be ilr. (See tShe discussion in 15, pages 263-2661 for the description of division dgebras over local fields.)

D is said t,o be mmijed at a place P if D g p F p is not split. Moreover, i f l n ~ ( U @ ~ P p ) = i / s with gcd(i,s) = 1, then s is said to be the local indrz of 1) at P. a ~ l d ind(D) is simply the lemt, common multiple of t.he various local indices (see 15. 'Tl~c:ore~~r 3'2. IS ] ) .

Now let X/k be a smooth geometrically irreducible projective curve, and 12/k the associ- at,cd fl~nctiorl field. Let P i , . . . , P,, be plwes of F/k of degree I. (:hoose Intrgers r, j l . . . . , j, such that 1 < ji < r , gcd(j,,r) = I , and CiCj,/r) = 0 in :$/L 'l'lius, ~ h c exact sequence (1) shows that there is a division algebra D / F ramified at exactly tltc placcs P,, and at each P,, D has local index T. It follows that ind(D) = T as well. Also, the residur field of Fp, is just k since the P, have degree 1, and if Ilk is the unique esteusion or k of degree r , then lFp, is the unique unraniified extension of Fpa of degree r 'Thus. 1,. t.l~e discussion in the previous paragraph, D @ p Fpa is isomorphic to (lFp,/Fp,,o,, K"), wl~crc r r , is a genrrator of Gal(1 Fp, IFp,) chosen so as to induce the Frobenius autolnorpllisrn 0 1 1 I l k .

1% first show that L = 1 F is a maximal subfield of D. K J ~ ally place Q of L, let I' be the corresponding place of F induced by Q. The completio~i of 1, i l t (2 is isonlorphi(. to tlle compositum lFp. If P is not one of the selected P,, the11 L) ::;F 1:p is already split, so ( D @p L) @[, Lq = (D @, Fp) @pP IFp is split. If P is olle of tht. selected P,, then, iw re~narked in the last paragraph, lFp is a maximal subfield of D :.;r f i n , so once again. (1) 'i3p L) OL Lg = ( D @F Fp) @ p p lFP is split. It follows ~ . l ~ a t D :)I. I, is split c~verywhere, so by our exact sequence ( I ) , L splits 0. Since [L : I.'] - r ( i~s k is iilget)raically closed i l l

F , this follows from the geometric irreducibility of S l k ) . I , 1s iittlcc,(l it ~~t i iu i rn i r l st~bficld of U / F .

We construct a sheaf of niaxinral orders A on our curw S L\S Iollows. I,(:( 4 - n Oy. wlterc the intersection runs across nll pl,wes Q of F en.cepl t 1 1 ( , ( I I O W I I ~ I < L ( ( . > 1.:. As described in I,erntuil 5 , A is a Dedekind donlain. Since 1 C D . 1A is a frw f i ~ ~ ~ t ( . l L R(t~~t~~, i r ted A rnodule.

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 8: Divisors on division algebras and error correcting codes

DlVlSORS ON DIVISION ALGEBRAS 3217

Choose a basis {Dl = 1 , b2 , . . . ,b,) of D / L , and consider l l ~ e frec A ~ ~ ~ c ~ t l u l c .bl = @:LI(lA)b,. Then M F = D , and 1A 01(M), where 01(M) = {x E D / LA^ C Al). As discussed in 15, page 1091, Ol(M) is all A-order in D, and is hence contained in a ~ ~ r a x ~ ~ n i l l A-order; call this rnax~~ual order B. Then, for each place Q of F (Q $ {PI, . . . , t',)). 13 i;?, (7y is a maximal OQ order. \Ve set Aq -- R aCiA Oy.

As for t,l~e places PI , . . . . e,, there is a unique rnaxinlal or-dcr ovc,r (w11 Oq which IS in facl a valuat,ior~ nng. ('l'llis follows, for instance, from 15, ' l ' l ~ e o w ~ ~ l ~ 12 8 d11c1 11.51. WC set ApE to be Lhis n~axi~nal Op, order.

Now let wy be the value funct.ion associated with this cholcc. of I I I H X I I I I ~ orders <uld consider the sheaf of rings A given by the assignment U u ny,, /Iq. ICxact.l,y as in the argument in the proof of Lenlnta I , any s E D can be written as bn-' for sonw b E A ( [ [ ) , where U = A' - { P I , . . . . P,), and for all but finitely many points C) E 11, a-I E A y . I t follows that utQ(s) 2 0 for all hut finitely many points Q E I i . S ~ n w the coi~~ple~neul of U contains only fiuitely many points, Lemma 1 shows that A is intleetl a. sheaf of n~axi~nal orders, whose stalk at each point Q E X is AQ.

We let P be the D-div~sor Ap, + . . . + Ap,, and we let G be any I)-divisor whose support, is disjoint from the P,, and such that L(G) # {O). The condition 011 t l~e supports guarantees that any f E L(G) is autonlatically in each Ap,. Since each Ap8 is just the valuation rmg corresponding to the P,-i~dic valuation, the factor ring Ap,/J(AE). where J(Ap,) is the maximal ideal of Ap,, is just the residue field of D with respect to the & - d i c valuation. The residue of D is the same as the residue of the completion D @ F Fp,, which is just 1. We thus have, exactly as in the commutative case, a map

where we have written J(P,) for the residue of f modulo the maximal ideal J (AE) of Ap,. By our choice of maximal orders above, the field 1 is contained I I I cb\wy maxi~nal order

AQ for Q $ { P I , . . . , P,,). Since each Apt is a valuation ring, and since ally valuation on 1 [nust be trivial, 1 must be contained in each Ap, as well. It follows that f (~r trny Q, wQ(x) 2 0 for all x E 1'. Hence, for any Q, any x E I * , and any f E L(G),

so x f E L(G). Thus, besides being a k-space, L(G) is also an 1-space. Lloreover. r (P , ) is just x for any P,, as the P,-adic valuation is trivial on 1. It follows f r m t h s t t~at the map cav above is an I-linear map. (In fact, each x E 1' conjugates each Av tu itself, so by 16, Prop. 2.6 and Lemma 2.21. ulQ(2) = 0, and wQ(xf) = wQ(x) + wq(f) : u ~ ~ ( f ) . )

The image of ev is thus an I-subspace of 1"; this will be our noncor~l~r~uttrtive Coppa code associated to D I F and the divisors P and G. We denote it CD(P.C;).

We point out that while different choices in our sheaf const,ruction n~ay lead t,o different (e.g., non-isomorphic) sheaves. the ring of constants nQcx A4 IS uniquely determined up to isomorphism by t.hc division dlgebra D since k is finite, by a tlleorcl~~ of Schofield 110. Tileorem 1 . 1 1 . We are wurkit~g with a specific choice of sheaf to st^ ~llat 1 IS it sulm11g of t.he ring of constants. In fact, i t is not hard to see that. 1 - n,,,, ,Aq. so 1 is LIIC: full r111g 111' conslants of the slreaf A .

We determine in the next proposition the minimum dist.a~cc: ,u~tl tlw tlillw~slon of ~ I K ~ code C.'o(P, G) . Thc ~nethotls are analogous to those used fur wlriiliulatlve Coppa c d e x (scr 131). Note that si~~cc. - 1 for each i, we have deg(Ap,) [ I : I : ] . ,Therefore. tleg(P) = 7.71.

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 9: Divisors on division algebras and error correcting codes

3218 MORANDl A N D SETHURAMAN

Proposit ion 6. t lm(CLj(P (:)) ~ I I I I ( L ( ( ; ) ) - (hn(C((; - P)) u d c~(( '~)(P. (:)) 2 71 -

T - I tlcg(G) Moreover, ~Jdeg((:) < I 71, then dlm(Co(P, C)) - d m L(L')) 2 dcg((:) -I I - 0. where y zs the genus o j C)

Proof: The kernel of ev is { J E L(G) I vp,(J) 2 1, I = I . . . . .T I ) . which is pnwsc,ly C(G - P ) . The result 0 1 1 tht, c l~~~ lcws io~~ in~nnediately follows. As for. t,he t l ~ s t a ~ ~ c e , write d for d(C:o('P,G)) and let J E C(C), / f 0, be such t,hat w(cv(J)) - - (1. T11c.n / i s I I I

t.he maximal ideal of exactly 11 - d of the Ap, , say A q l , . . . , A p I t follows that J E

C(G - (A,;, t . . . t A,:., , ) ) . tlc~lcr. C(C; - (Apnl + - - - + A p , ~ , ) ) # 10). I t Collows that deg(G - (AP,~ t . - - t Ap,,, ,)) > 0 , since C ( H ) = (0) for a divisor 11 of ~~egat ive degree. Thus, deg(G) - r ( n - d ) > 0, or tl 2 n - r . - I deg(G).

For l l ~ e second statement, il tlrg(C:) < r.rr - deg(P), then deg(G - P) < 0, so L(G - P) {O). It follows from the peragr-npl~ abovc that, ev is then an injechve wap. Moreover, l),y the Kiemann-Roch theorem, wc have dinl(Cn(P. G')) = dim(L(G)) 2 deg(C) I 1 - g.

4. Relation between C U ( P , G) and CL(+(P) , @(G))

We continue to use the notation of t.he previous sections: P I , . . . , P, are rational points on X , and D is a division algebra with center F , the function field of Xlk, such that the local index of D a t each P, is equal to the index r of D, and the local index of D a t any point P # Pi is 1. Recall that we have a sheaf A of maximal orders on D, that A is a subsheaf of the constant sheaf U c D, and t.hat each stalk A p contains 1. Note that since Ilk is a cyclic Galois extension and 1, F are linearly disjoint over k, the extension L / F is also cyclic Galois. l i ~ t u he a generator of Cal(L/F). Since L is a maximal subfield of D, wr may write D as a cyclic crossed product D - (I , / F, o, a ) = 9)::; Lx', where x' - a 6 F and zbz-I - u(b) for all b E L. We will make use of this description of D in a number of places below.

If f{ is a field extension of t.', recall the canonical map 3~ : div(F) - div(K) that satisfies l / 'K (P) = C eqlpQ% where the sum is over all IC-points Q lying over P , and where e ~ , p is t,he ramification index of Q over P . In terms of valuations, eQ/p is t.he ramification index of f</F relat,ive to t l ~ c vaiuatio~~ ring of K corresponding to Q. We will refer to this valuatior~ ring by C3,i,Q. I.hr our tnituin~al subfield 1F = L of D, \ye writ.e for d:L. We also view the map li, as a map fro111 div(D) to div(L) by sending Ap to ri,(P). Since / , IF ' is everywhere unrarnified, eq/p 1 for every P and every Q lying over P, and. for each of the points P,, there is a unlquc point Q, lying over P,. This follows from the fr~ndan~el~tal - - equality [L : FI := Cu,peu i r f a ip~ where fytp = (OL,Q : OP1, and from the f ~ c t that thc residue field of any point ex(cwlnig f: is 1, an extension of k = of degrec IL : FJ . As above, let P - C, AI: and let C t ) c b a divisor with supp(G)n{Ap,, . . . , Apq} -= 0 . In the uext lenima we compare our code w ~ t ll $1 Coppa code constructed from the algebraic functioit field L/1. To refer to such a Coplm code we denote it by CL(E', G') if E', C' are appropriat.ely chosen divisors on L.

Lemma 7. The space L(qi(C:)) 85 cnnonzcully rsomorphzc lo C(C) f' I,, and, wzlh the zdcn- tzjtcalton oJ C(+(G')) us n ruh\pcwe oJ C(C), the zmage ev(C($(G))) oJ C(@(G)) under thc ct~nluatron map ev L(C) - I" zs /he Coppa code CL($J(P). @(G'))

Proof: I.rt f E L s;itlsfy ( J ) I r/,(G) 2 0 in div(L). Givwl 1' E S - { I J , . . . . . I J , , ) . 11.1 QI . . . . .Q, he the e x t r ~ ~ s i o ~ ~ s r ~ f I' to I,. Let. r ~ p be the coefficient of i l l (.: ' l ' l~us, I S

the coeflirie~~t, of each (2, i l l < I : ( ( : ) , C'onscqucntly, vy,(f) t- 71,. 2 0 for each 1 . lret R I)(: CL

uniforrnizcr for Op. T I I ~ I I K I S < L ~ ~ l ~ i i o r ~ n ~ z e r for each Q, since L I P is everywhere unramifird. So. V ~ , ( J T " ~ J ) 1 0 for all I . so x"" J lit% in all of the valuation rings that extend Op. Hut.

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 10: Divisors on division algebras and error correcting codes

DIVISORS ON DIVISION ALGEBRAS 3219

this means t,hat nnP J is i n the 1nt.egra1 c4os11rt. of Of in L, rv11wt1 IS l o p . (To see this, if B is the integral closure of Up i l l L , take n basis 1 1 , . . . ,I, of Ilk, and let l',, . . . .1: he the dual basis relative to Trl/k. Then l;, . . . .1: is also an F-bass of L. Suppose that b E B, and write b = C, lia, with a, E F. 'Then 61, = C, l,lJn,, so 'IYLIF.(W,) = C, 'Iklp.(ltl:)u3 = a,. But, TrLIF(B) O P SO each ( 1 , E O p . 'I'hus. U - C / ; U p = /Up.) tlrncr, snP f E Ap. Since T

is also a unifor~nizer for A p i m l is I I I its ctlltrr, wc8 have ci:p(s""J) - wp(f) t up 2 0, SO

J E L(G). For P E { P I , . . . , f:,), lOp, is ( l ~ e u~~iquc. valuat,lon ring ext~entlir~g O r z , and I T sa,y v: denotes this valuation on L . t,lien (J) 1- @(G) > 0 implies lhat v:(J) 2 0 for each i, which implies that / E Aps for each 1 'I'hus, ulpL( f ) 2 0 for each 1. 'I'llerefore, L(dr(G)) C L(G)n L.

For the reverse inclusion, t.ake f E L(G) n L. Then, for P E S - { P I , . . . , P,), wp(f) t np 2 0. If rr E Op is a uniformizer for Op and A r , then J - T - " ~ U for some u E Ap. But this puts u in Ap n L, wl~ich is integral over O f . Thus, 11 is in every valuation ring of L extending Op; in other words, tru(J) > -np for every valt~ation ring of L extending Op. For P E { P I , . . . , P,), we have urq( J) 2 0, so J E Ap, n L. As before, Apa is a valuation ring, so Ap, n L is the (unique) valuation ring of L that extends Opt. Thus, up,(/) 2 0 for each i. Thus, we have proven C($:(G)) = C(G) n L.

Consider ev : L(G) -+ l", where ev( f) = (f (PI) , . . . , f (P,)). I f / E L($(G)), we claim that f (QJ = f (PJ for each 1 , and so ev(L(d4G))) = {( / (&I) , . . . , j(Qn)) I j E L($(G))}. This is CL($(P),$(G)), finishing the proof of the lemma once we prove the c l a i m A d o this, note that if r n ~ , ~ , is the maximal ideal of OL,Q,, then j(Q,) = J + r n ~ , ~ , E O L . ~ . = - 1 = Ap<. Also, viewing J E D, we have J(P,) = f t J(Ap,). But, Ap, contains O L , ~ , and J(A.1 ~ O L . Q , = r n ~ . ~ , . so J t ~ L , Q . = J + J(Ap,). In other words, J(Q,) = f(P,). t

We say that an L-basis d l , . . . . d, of D is an orlhogonal baszs with respect to a maximal or- der Ap E X with associated value functjon wp, provided t,hat wp(C,l,d,) = min,{wp(l,d,)} for all 1, E L. If , in addition. urp(C,l,d,) = min, {wp(l,) + wp(dt)}, we will call the L-basis d l , . . . , d, a strongly orthogonal basis with respect to Ap.

Lemma 8. With D = @:,: LJ:' us tn the beginning of thts section,

I . I] D is splzt a1 P . lel d l . . . . .d, he unils in A p such that &, . . . ,h is a basis for over Ap fl L. Then d l . . . ,d, as u strongly orthogonul basts for D with respect to Ap.

2. Ij P = P,, lhen 1 ,s.. . . . Y"' is u stronyly orlhogonal basis for D with respect to A r

Proof: Suppose t l~at d l , . . . , tl, are u~~ i t , s in Ar such that &. . . . ,dT is a basis for over Ap C1 L. Note that u:p(d,) : 0 for eac11 1. and t.hat since 11, E A;, we have ,wp(ed,) = urp(e) for all e E D by the dcfi~~ition of the value ful~ction ,u~p (or by 16, Lemma 2.21). Suppose that min. {wp(l,d,)} = u!p(lk). Then wp(lt) = t for some t E Z, so if a is a uniformizer for Op, then wp(a-'1,) 2 0 for all i , and wp(s-'lt) = 0. These facts hold since a E F = Z(D). Thus, we may as sum that wp(1,) 2 0 for all i and wp(1k) == 0. Then C,l ,d, = C,cz f 0 since 5; # 0, so wp ( C , l,d,) - 0, as desired.

For the second part, suppose that P 7 P, is one of the rational points at which the local index of D is equal to r. - ~ n d ( D ) . Let Inv(Dp) r= k/r t Z with gcd(k,r) = 1. Then Dp =

(lFp/Fp,o, x k ) for some unik)rniizer s of Op. Since Y = s k u for some unit 11 E O;,. and since wplF = rvp if imp is the ~lorlnitlized valuat.ioli o n F wit11 vAuation ring Op, we see that wp(xl) = ik. Mor~owr. L/k' is ~~~lrdniifit 'd at P, so t.hc: valur group of L is TZ. ' ~ h u s , since gcd(r, k) = 1, t,he va111c.s c~.p(.r') iwc distinct ~notlulo the vitlue group of L. If x,1,:1:' E 11. each term 1,x' has dist111c.t v,tlur~, so / u p (C, I,.?) -~ lnin, /I)~(/ , .E') = n~in, {,wp(l,) t ,up(x')]: the latter equality I~ol(ls si~lc.v WJ, 1s ;I valuntion OII 11. 'I'hls shows that l , z , . . . ,zr-I is a strongly ort,hogonal basis of U over L w~t,li respect to Ap.

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 11: Divisors on division algebras and error correcting codes

3220 MORANDI A N D SETHURAMAN

Proposit ion 9. 7'he set { 1 , ~ . . . . , xV-I} zs an ortlro!loncd hnsi .~ u~zlh rr..?pccl 1 0 vm:h A r f w PEA'.

Proo[, I,et, 1' c ,Y. I S tJ I>,, t h t , h ~ J:' f 0 1 I I I :L SI I O I I ~ I ~ or1 I I O ~ O I I ~ I)P& with r ~ p ~ t , t.0 A p by Lellllna 8. Next., suppoac. t,hN I ) is splil il l P Iirbc.l~ll fro^^^ t l ~ v ~)~.uof or 1x111n1a 7 tirat 10,. is t.lle irltc.&r.al closllrc. ol 0,. i l l 1, I"I.OIII 1111s. wtx S(Y ' 1 1 1 d 1 11 /C)/$ is ~ ~ ~ n 1 ; ~ i n e d in the stalk Ap. By 14, Prop. 1.31, t.liere are 2, E A p such r h k t At, - r i i ; ' , j 112, with : , /I - u'(6)q for b E L, and z,z, = f (o',uJ),-,+, for some normalized rocycle f w~th vducs in H . Since 1) is split at P, the maxi~nal order A r 1s Azuntaya over 01,. 'l'l~us, J(A18) - r n ~ d ~ b - -ii,::Jrnp~;,, wher-e mp is the n~aximal ideal of Op. HOWCW. h j 14. t'rop. X I / . ./(A,') (I)::,: 1,2,, where 1, IS the product of the ~naxinial ideals iZ/I of A for wh~ch f ( u ' . a ') 81 hl . Since mpU is the product of all thr niaxin~d ideals or H , U I I I ~ I I V fiwtori~.llio~~ of i( l(ds i l l Ll force^ f (u', o-') @ M for every n~axirrral ideal M of H , so f(u' , o- ' ) E 11' Sor ( W I I r . 'L'her~, front the cocycle condition, we see that o'(J(o-', oJ) ) f ((n', n-" ' ) j ( u E 3 u-') j ( I , uJ) € B* , so all cocycle values are in fact in Be. Consequently, the residue r i ~ ~ g 2 is equal to lhe crossed -- - product (B/Op, o, f ) = @::;%, so Ihe f are linearly i~~tlepentlent over z. Therefore, the 2, form a strongly orthogonal basis with respect to Ap by Lemn~a 8. Note also that ~ ~ ( 2 , ) = O for each i from this description of the residue ring 7, To see that the powers of x: form an orthogonal basis with respect to A P , write x' = a,;, with a, E 1,. Wt. h,lvt.

since 2 0 , . . . ,z,-, is an orthogonal basis of D w~t h respect to A p Tl~us, l l ~ c I' form an orthogonal basis. I

T h e o r e m 10. IJ G is a divzsor on D with deg(G) < rn and supp(C) n {A p,, . . . . AP- } = 0, then C(G) C L.

ProoL. Let f = x , l i x ' E L(G), so wQ(f) t wy(C;) 2 O for all Q. However. .uiy(f) = ~nin,(w~(l,x ')} 5 1 u ~ ( 1 , ) by Proposition 9, so lo E L(G) . I,(*( 7 -- C::: I,.? - j-10 E L(C;). Then wp, (f') 2 0 for all j . Since uip, (f') = rni~~,,, (w i ; (1 , ) I- i lc4p, ( . r ) } . again by Proposition 9, each wp,(l,) + iwp,(z) >_ 0. But iwp,(x) f O in l'A,5/~~p,n~. for cac:li 1 # 0 , so in fact lurJ(l,) + iwp,(x) > 0 for each 1. Consequently ubp,(f') > O for cach J . T l ~ r coefficie~~t of AP, in (f') i G is then at least 1. The degrw or A,; is I . s i ~ w t l ~ t s residue ri t~p of Ap, is 1. This, together with (f') t G 2 0, forces dep((jl) I G) 2 r.71, while cieg((J') t C) - deg((f')) + deg(G) < deg(G) < rn , a contradictio~~ unless f ' - 0. 'I'lwreforc.. j T. 1" E 1,. I

Corollary 11. Let G be a divisor on D with supp(C;) n { A p , , . . , A/:, } -- 0 , and suppose that deg(G) < m = deg(P). I j $ is the naturnl 7nq1 11, : tiiv(1,') tliv(1,). then the code C o ( P , G) is qua1 to the code CL(ll.(P),*(C:)).

References

[2] E. I~ormanek, The Polynoir~ial Identitzes nnd liivcmc~rrls o/ r t x 1 1 Alakriccs. CBMS Notes, Vol. 78, Amer. Math. Soc. Providence. 1991.

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14

Page 12: Divisors on division algebras and error correcting codes

DIVISORS ON DIVISION ALGEBRAS 322 1

131 V. D. Coppa, Codes 011 i~lgelnaic. curves, Sornel h!ull~ Dokl . 24 (1981), 170- 172

141 D. Haile, Crossed-product orders over dlscrew va lua tm~ r~ngs. J Al!leb~u 105 (1987), 116-148.

151 1. R e i n ~ r , Mcrimul Orders. Academic Press, [ondon. 1975

161 P. M o r a ~ d i , Vdut' fu~~ct ions on cent,ral simple algebr<~s, 7hii.s A7n~7. Mnth Soc 315. 1989, 605-622.

[7] P. Mora~idi. Nonco~~~mritiitive Priifer rings. J. Algebru 161. (I!N:I). 324 311

181 M. Orzeclt and C. SIIIHII, 7he Rtauer Group of Coin~tnulutzoc I<zngs. Marcel Llckker. Inc., New York, 1975.

191 M. Van den Berg11 and J. Van Geel, A duality theorem Sor orders in central s i ~ n p l t algebras over function fields. J. Pure Appl. Algebm 31 (1984), 227239.

1101 M. Van den Bergh and J . Van Geel, Algebraic elements in divis~or~ algebras over function fields of curves, Israel J. hlath. 52 (1985), 33-45.

1111 J . Van Geel, Places ulld Vuluulions in Noncommututzve Hzrrg 'I'twr:y, Marcel Dckker. Inc., New York, 1981.

1121 E. Witt , Rien1a1111-Rocl~ster sat.z und 2-funcktion in1 hyprrko~nplexen. Math. Ann. 110 (1934), 12-28.

Received: June 1997

Dow

nloa

ded

by [

Uni

vers

ity o

f K

ent]

at 0

1:54

18

Nov

embe

r 20

14