Hasse-Witt Matrices of Hyperelliptic Function Fields
Tadashi WASHio and Tetsuo KODAMA*
Department of Mathematics, Faculty of Education,
Nagasaki University, Nagasaki, Japan (852).
(Received Oct. 31, 1985)
Abstract
The Hasse-Witt matrices of some hyperelliptic function fields of positive charac- teristic are studied and a dual relation between the ranks of them attached to the
function fields of different characteristics is proved.
1 Introduction. Let A be an algebraic function field of genus gal of one variable over a perfect field K of positive characteristic p. With a saparating element x of A, every differential w of A can be written as
= (a0P+ aiPx + .+ ap_iPxP-')dx (ajeA for i 0, 1, p -1) . Then the Cartier operator C is defined by
C(w)=ap_idx.
The followings are well-known (see Eichler[1] and Kodama[4] for some related properties to the operator C) :
( i ) C(co) is independent of a choice of the separating element x.
( ii) C (YIP wi+ Y2P 692) = YIC (col) + Y2C (602),
where yl, y2€A and col and w2 are differentials of A.
(iii) If w is a holomorphic differential, then so is C(60).
Let Wg be a basis of the K-module of holomorphic differentials in A. Then the representation matrix M over K of C with respect to this basis is called the Hasse- Witt matrisx (see Hasse and Witt [3] and Manin[5]) :
t(C(w i), C(Gz0= M t((i, wg).
In this paper, we assume that A= K(x, y) is defined by y2 = P(x) over K, where aeK* K - {0} , p*2, (p,f)=1 and
x-f+ a if f =2g +1, P( x),
x(x-f+ a) if f =2g.
* Department of Mathematics, (810).
College of General Education, Kyushu University, Fukuoka, Japan
Moreover we always choose as a basis for the K‑module of holomorphic differ‑
entials,
a' dx/y, a'2=xdx/y, , . . . b,g=x hldx/y, and, if necessary, we will denote the Hasse‑Witt matrix by
M = M1 = Ml(K,, p, a) if f = 2g+ l, M2 = M2(K,p, a) if f = 2g.
It is well‑known that if g=1 then f Ml 0 if psl (mod. 6),
Ml = O if p E 5 (mod. 6), f M2 0 if p s I (mod. 4),
M2 =0 if p =3 (mod. 4) (Hasse[2] , Olson[7]).
In the present consideration, we want to extend these results to g 2 as follows.
Let K' be a perfect field of positive characteristic p'. Then, for two function fields of the same genus g,
rankMi(K, p , a) + rankMi(K', p', a') = g if p p' and p +p' O (mod. 2f), where i= 1, 2.
Moreover we will investigate two cases
rankM g and M=0.
When P(x) are trinomials, Miller[6] investigated two families of function fields with the regular Hasse‑Witt matrices.
2 . The case P(x) =x2g+1+a. Let K be a perfect field of characteristic p > 2. Let A = K(x, y) be an algebraic function field over K defined by
y2 = P(x)=x2g+1 +a (a 0, aeK), where g denotes a positive integer such that (p,2g+ 1)= 1.
It is clear that A is of genus g. Put
f=2g+1 and l=(p‑1)/2 Then
Q, x' Idx/y k 0( )al kx'+fk‑Ida/yp and x(i+fk)/p‑Id if i+fu 0 (mod. p), C(xi+fk‑Ida)‑ O
otherwise.
For a given integer i satisfying I i g, the diophantine equation
i+fr=pj (o k 21 1 J 2g)
has a unique solution (k, j) and then
f 1 j g if O k l,
g < j 2g if l<k 21.
This implies that the congruence i+fe =0 (mod. p) has a solution k in the range
O k I if and only if the congruence i: pj (mod. f) has a solution j in the range 1
j g and then such a solution is uniquely determined.
Therefore, for a given i (1 i g), if i= pj (mod. f) has a solution j in I j g, then C(G,) a,.G, with a,. ( ) dl‑k)!p 0 and k=(pj‑i)/f,
and if not, then C(Q,i)=0. This means that the Hasse‑Witt matrix M1 = Ml(K,p, a) of A has at most one non‑zero element in each row.
Moreover it is clear that for i,i' (1 i i' g),if i= pj (mod. f), i' spj'(mod. f) and 1 j, j' g, then j j'. This shows that Ml has at most one non‑zero element in each column. Summing up, we get the following theorem.
THEOREM l. ( i ) The Hasse‑Witt matrix M* has at most one non‑zero element in each row and in each column.
(ii) rankMl=#{i 1 i g, i+fu=0 (mod.p) has a solution k in O k l}.
(iii) rankM*=#{i 1 i g, i=pj (mod.f) has a solution j in 1 j g}.
We will now prepare the following lemma for the between the ranks .of Hasse‑Witt matrices.
proof of a duality relation
LEMMA 1. Let f=2g+1 and let p and p' be odd primes satisfying p p', ,,f)
= I and (p',f)=1. If p+p' O (mod. 2f), then for a given i (l i g) only one of two congruences
f iEpj (mod.f) (l J' g), (modf) (1 j' g)
.