computation of the zeros of p-adic ¿functions. iizeros of the kubota-leopoldt p-adic l-functions...
TRANSCRIPT
MATHEMATICS OF COMPUTATIONVOLUME 62, NUMBER 205JANUARY 1994, PAGES 391-406
COMPUTATION OF THE ZEROS OF p-ADIC ¿FUNCTIONS. II
R. ERNVALL AND T. METSÄNKYLÄ
Abstract. The authors have carried out a computational study of the zeros of
Kubota-Leopoldt p-adic L-functions. Results of this study have appeared re-
cently in a previous article. The present paper is a sequel to that article, dealing
with the computation of the zeros under certain conditions that complicate the
original situation.
1. Introduction
This paper continues the authors' computational study, begun in [2], of the
zeros of the Kubota-Leopoldt p-adic L-functions Lp(s, x) ■ We will discuss
four themes, all of which came up but were left aside in [2].
In the following, the article [2] is referred to as Part I. When referring to an
individual section or proposition ofthat paper we write, say, §1.3 or Proposition
1.5. We retain the basic notation of Part I; in particular, p is an odd prime,
fe(T) = J2%oajTj denotes the Iwasawa power series determined by the first
factor 6 of x , and X = Xg stands for the A-invariant of this power series.Our first question, considered in §2, concerns the computation of the zeros of
Lp(s, x) when the second factor of x is nonprincipal. The remaining sections
deal with the computation of the zeros T\, ... , T\ of fg(T) under certain
conditions that complicate the original situation discussed in Part I. Thus, in
§3 there are two zeros Tx and Ti close to each other, §4 introduces types of
the Newton polygon of fe(T) different from our two basic types, and the final§5 shows how to compute zeros 7^ lying in wildly ramified extensions of Qp .
A sample of numerical results related to each of these sections is given in
Tables VI-X at the end of the paper.We recall that a crucial step in the computation of a zero To of fe(T) is the
determination of the extension E = QP(T0) and of an initial approximation to
of To satisfying the condition
fg(t0) = 0 (modn2^1),
where n is a prime element of &e and y = vn(fg(T0)). To accomplish this,
one first reduces the possible candidates for E to a family of relatively few fields
and expands T0 7t-adically with unknown coefficients; then E and io will t>e
Received by the editor April 30, 1992.1991 Mathematics Subject Classification. Primary 11S40, 11Y05, 11Y40, 11R20, 11R23.Key words and phrases, p-adic L-functions, computation of zeros, factorization of polynomials,
Newton's tangent method, Abelian fields, Iwasawa theory.
©1994 American Mathematical Society
0025-5718/94 $1.00+$.25 per page
391
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
392 R. ERNVALL AND T. METSÄNK.YLÄ
obtained by solving successively a sequence of approximation congruences
fe(T0) = 0 (mod O, m= 1,2,... , m0,
denoted by W(nm) below.
When describing the computation of To, we usually assume, to fix ideas,
that ûo = /e(0) does not vanish. If a0 = 0, the procedure is essentially the
same; one just excludes the zero To — 0, which is known to be simple, and then
replaces fe(T) by fg(T)/T. In this case we say that fg(T) is of type Z.
2. L-FUNCTIONS INVOLVING CHARACTERS OF THE SECOND KIND
In this section we are concerned with the zeros so of Lp(s, 6 y/n ), where y/„
is a nonprincipal character of the second kind. Recall that y/n is of order pn
and conductor p"+x (n > 1). We will discuss the questions of how to find
examples of Lp(s, 6tp„) having a zero So , and how to compute this So .
Suppose we have found a zero 7b of some Iwasawa power series fg(T).
By §1.3, if there is a corresponding zero So of some Lp(s, dy/n), then s0 =
log(l-r-7b)/log(l+rf/7) and vp(T0) = \/(p-\)pn~x. In the range of our compu-
tations, the only /7-ordinals of L0 of this form are v3(T0) - 1/2, v$(Tq) = 1/4,
and Vi(To) = 1/6, and the respective values of X are 2, 4, and 6 (or 3, 5, and
7 when f$(T) is of type Z). Since, moreover, the coefficients of fe(T) in all
examples are from %p , Proposition I.5(i) shows that To must belong to QP(ÇP")
in order that So lie in Ds, the domain of definition of Lp(s, x) ■
The case v^(To) = 1/6, which leads to a wildly ramified extension Qi(T0),
will be studied in §5. We found no example of this kind with L0 e (MCç).As for the other two cases, it follows from Proposition I.5(ii) that the con-
dition To € Qp(C/>) is even sufficient for So to be a zero of a Lp(s, dy/x). We
have the following practical criterion.
Proposition 1. Let X = p - 1 and vp(ao) = 1, so that vp(To) = l/(p-l). Then
To e Qp(Cp) if and only if aox = axo ■
Proof. The Newton polygon of fe(T) is of the "first type" discussed in §1.9.
From that discussion it is seen that 7b G Qp( p~yrp), where r is determined by
the conditions
To = xn (mod n2), 0 < x < p,
aox + aXorx"~x = 0 (mod p).
Since Qp(Cp) = Qp( p~\/--p), the assertion follows. D
For p = 5 our numerical material contains five examples with X = 4 and
v$(ao) = 1, and one similar example of type Z . In all these, T0 $ Qs(Cs) •
Four of the examples are exhibited in Tables IV and V of Part I; the remaining
two are (5, 5708,0) and (5, -12712, 1).Our main case is that of p = 3, X = 2, and ^3(00) = 1 • Applying the
criterion of Proposition 1 to the tables computed in Program A (Part I), one finds
numerous examples of this case which give zeros of L^(s, 6y/x). Including some
analogous cases with X = 3 , we settled 13 examples of this kind completely.
Before describing the computation, we observe that there are two characters
y/x mod 9, say ip+ and y/- , identified by the equations y/±(\ + 3d) = — \± j,
where n — \/-3- A zero Tq with 1)3(7b) = 1/2 may be written in the form
Tq = 3b + en , where b, c G Z3, c ^ 0 (mod 3). Then, by Proposition 1.3, the
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
COMPUTATION OF THE ZEROS OF p-ADIC ¿-FUNCTIONS. II 393
number so = log(l + 7b)/log(l + 3úf) corresponding to c = +\ (mod 3) is a zero
of L3(s, 8ip+), in fact, the only zero of this function (under the assumption
that X = 2). Its conjugate is of course the zero of L$(s, 8\p-).
After computing L3(s, 8), fe(T), T0 , and So in the usual way, we proceed
with the computation of the functionoo
L3(s, 0^,) = ]Tm-s'.(=0
As in §1.5, this means the approximation of u\ by rational integers congruent to
u\ modulo a prescribed power pM ; this makes sense since u\■ —> 0 as / —» oo .
Here we make use of Washington's formula with <3> = 9d in the way described
in §§1.5,1.11. A lower bound for vp(u'¡) needed in the computation is provided
by Proposition 2 below.
Recall that
(1) Lp(s,dy) = fe(p(\+dp)s-\),
where p = i//(l +dp)~x , \p = ipn . To check the correctness of our L3(s, 0i//x),
we evaluated both sides of this formula for s = 1 . As a final check it was
verified that L3(so , 6y/x) vanishes mod nM> , where (see the end of §1.11)
Mx = min ( 2M, M - 3 + min vK(u'¡) J.
By the remark at the end of the present section, Mx> M.
Example. Let 6 be the character defined by (3, -1147, 1). The followingtable presents L3(s, 8) (1st column) and fe(T) (3rd column) according to
the same principles as the Tables I-V in Part I; see §1.10 for a description.The middle column lists the coefficients u'0, ... , u'6 of L^(s, 8y/±) ; here, n =
\/^3. The coefficients u\ with i > 6 are zero to the accuracy displayed in this
column.
0.110000.011220.010200.000110.001020.000000.00011
0.0200 ±0.0122710.0111 ±0.1102ti0.0110±0.0002tc0.0001 ±0.0220tt0.0011 ±0.0000^0.0000 ±0.000l7T0.0001 ±0.0000;r
0.110000.11021.1000.202.00.
The zeros of fe(T) are TXt2 = 0.10 ± 2.071 (approximately), and the corre-
sponding numbers sx ,2 = 0.0 ± (2.)tt are the zeros of L3(s, 9y/^). (The upper
and lower signs correspond to each other.)
Table VI contains further examples of this kind. In Table Via there are two
similar examples with an additional pair of zeros of fg(T) and L$(s, 8).
We conclude this section by proving the following estimate (cf. Proposition
1.6).
Proposition 2. For any y/ - y/„ (n > 1), let Lp(s, dip) = Ylltou'is' • Then
vp(u\) > ip-\
(1 = 0, 1,
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
394 R. ERNVALL AND T. METSÄNKYLÄ
Proof. Considering the right-hand side of (1), we write
oo
p( 1 + dp)5 - 1 = p exp (slog( 1 + dp)) - 1 = ^2 d¡s',
where do = p - 1 and ú?, = p(\og(\ + dp))1 //! for / > 1 . This gives theexpansions
oo
(p(i+dPy-iy = J2dusl a = o,i,...)i=0
with i/oo = 1 > djo = 0 (/ > 1), and
du= E dh-d'i (*>o,y>i).
We show that
(2) v,(dy) > ./^ - 1) + (l - j^j\ i (i > 0, j > 0).
This is trivial for j = 0 ; so let j > 1. Denote by / the number of positive
indices in a set {tx, ... , tj} with tx H-1-1¡ = i. For simplicity of notation,
assume that the positive indices are just tx, ... , ty , so that
j fEUÄ) = U - J>p(p - i) + EMdt.)-
Since vp(p - 1) < \/(p - 1) and vp(dh) = tv - vp(tv\) >tv- (tv - \)/(p - 1)
(v = 1,...,/), we obtain the estimate
> / j \ /
EV<U>./*v(/>-i) + (i--¿—¡-IE'"-i/=i \ p j „_,This proves (2).
Equation ( 1 ) now allows us to write
oo J oo oo / J \
E «**' = J™ E °> E rfyj' = Ji™ E E aA si>i=0 7=0 i=0 i=0 \j=0 /
whenever s e L\. Restrict 5 for a moment to the subset of Ds defined by
vp(s) > 0. Then it follows, in view of (2), that
oo / oo
$>'/*' = £ E«AK1=0 1=0 \;=0 /
(see Lemma in [3, p. 53]). Hence,
CO
(3) «Î = EM« (1 = 0,1,...),,=0
and this together with (2) implies the proposition. D
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
COMPUTATION OF THE ZEROS OF p-ADIC ¿-FUNCTIONS. II 395
Remark. Let, in particular, X = p - 1 and vp(ao) = 1. Then the coefficients of
the series for Lp(s, 8ipx) satisfy the stronger inequality
vP(u'i) > i - j—¡ + 1 (i = 0,l,...).
This is seen from (3) and (2) upon observing that vp(Oj) > 1 for j = 0, ... ,
p - 2 and jvp(p - 1) > 1 forj>p-\. In the range of our computations,
vp(u\) attains this lower bound for most values of i.
3. TWO ZEROS CLOSE TO EACH OTHER
This section continues our study of the case X = 2, vp(ao) = 2 in which the
Newton polygon of fe(T) is of the "second type" considered in §1.9. This case
was settled under the assumption that the two zeros Tx and T2 (denoted by
7q' and T¿' in §1.9) satisfy vp(Tx - T2) < 2. We now describe how to deal with
examples in which vp(Tx - T2) is bigger. As before, it is assumed throughout
that TX^T2.
We begin with the following proposition that will also be needed in §4.
Proposition 3. If all the zeros Tx,..., Tx of fg(T) are pairwise distinct, then
x
vP(fe(Tk)) = Y, MTk -Tj) (k=l,..., X).;=im
Proof. By the /7-adic Weierstrass preparation theorem (see §1.2),
fe(T) = ue(T)we(T),
where ug(T) is an invertible power series and wg(T) a monic polynomial of
degree X having Tx, ... , Tx as its zeros. Differentiate this equation to obtain
f¡>(Tk) = ug(Tk)w'e(Tk) and, hence, vp(f'6(Tk)) = vp(w'g(Tk)). D
Going back to the above case, suppose that the rational integers x , y , and
r fulfil the conditions
0 < x < p, 0 < y < p, r = 1 or g (a primitive root of p),
TXt2 = xp±ypy/rp (mod/72).
Such numbers are found in the way presented in §1.9. In particular, if y = 0,
then vp(Tx - T2) > 2, and we may put
Tk=xp + zkp2 (mod/75/2) (k = l,2),
where zx and z2 are either in Z or in 1[^/g]. Looking at the coefficient of
p4 in the approximation congruence mod p5, we infer a condition
d0 + dxzk +a20z2k = 0 (mod/7),
where do and dx are rational integers determined by x and by the first digits
of a0, ... , a4.
If the solutions zx and z2 of the last congruence are distinct, then
vp(Tx - T2) = 2 and Krasner's lemma implies that Tk - xp G <Qp(zk). Thus,
E/Qp is unramified, and we have y = vp(fg(Tk)) = 2 by Proposition 3. In this
case the algorithm may be started with tk = xp + zkp2.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
396 R. ERNVALL AND T. METSÄNK.YLÄ
We found two examples of this kind, for the characters (3, 13564, 0) and
(3,-11188, 1), both with E = Q3(\/2). They are presented in detail in TableVII.
Secondly, suppose that zx = z2 = z. Then z G Z, and one proceeds with
the congruence
TXt2 = xp + zp2±up2yfrp (mod/73),
where w G Z, 0 <u < p . Put n = ^frp . To find u and r, one has to solve a
congruence of the form ru2 = I (mod p), where / G Z ; this is in fact obtained
from ^(tt11). If u ¿ 0, we see that E = Qp(n) and vp(Tx - T2) = 5/2,
and Proposition 3 gives y = vK(fg(Tk)) - 5. Hence, we may take ii,2 =
xp + zp2 ± up2n .
In Table VII, the example for the character (3, 13784, 0) belongs to thiscategory. Table Vila shows two similar examples of type Z .
If u = 0, i.e., vp(Tx - T2) > 3, we are in the analogous situation we started
from. There is no example of this in our material.
4. Various types of the Newton polygon
We now show how to deal with some classes of the Newton polygon of fg(T)
which differ from our two basic types. They are called Types 3-6 below; each of
them was met in the computations and will be treated in a generality sufficient
to settle our particular examples.
The examples mentioned in this section appear in Table VIII unless otherwise
stated.Type 3. vp(a0) > 2, vp(ax) = 1 , X = 2.The two zeros Tx and T2 satisfy vp(Tx) = vp(a0) - 1 > 1 and vp(T2) = 1 .
Hence they lie in Qp . This is a slight variant of our "second type", and from
§1.9 it is seen, on putting a02 = 0 , that one may start with tk = xkp (k = 1, 2),
where xx = 0 and x2 is the root of the congruence axx + a2ox = 0 (mod p).
Examples of this kind are obtained for the characters (5,3101, 0), (5,-4371,1 ), ( 11, -723, 1 ), the last two being of type Z . The example (3, -1399, 1 )in Table II (Part I) also belongs to this family.
Type 4. vp(ao) = c > 1 , vp(ak) > c(l - k/X) for k = 1, ... , X - 1 , X (> 1)prime to c and p .
The Newton polygon has one nonzero slope, and this equals -c/X. As in
the case of the "first type", the zeros To (= Tx, ... , Tx) are in fully ramified
extensions E = Qp(tfrp). Because c and p are prime to X, one easily verifies
that y = vn(XaxTQ~x) = (X - l)c. If X = 2, one has to solve for x and r the
congruence aoc + a2orcx2 = 0 (mod p), obtained from W(n2c+X), and then take
to = xnc. For X > 2 the procedure is similar but more complicated.
We have two examples with c = 3 and X = 2, namely for (3, 1901, 0) and
(5, 12056, 2), and one similar example of type Z , for (5,-7816, 1). TableIII in Part I gives an example with (5, 1317,0) in which c — 2 and X = 3 .
Type 5. vp(ao) = 2 , vp(ax) > 2, X = 4.As in Type 4, there is but one nonzero slope, this time equal to -1/2. The
zeros 7|, ... , 7*4 lie in fully ramified quartic extensions E = Qp(J/rp) or, if
this is not the case, either in quadratic or biquadratic extensions of Qp (here
"biquadratic" means "a composite of two quadratic").
An example of the former kind is given by (3, 1541, 0). The computation
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
COMPUTATION OF THE ZEROS OF p-ADIC L-FUNCTIONS. II 397
is analogous to that in Type 4; note that Proposition 3 yields y — 7 (see Table
VIII). On the other hand, for the character (3, 4204, 0) one rules out the
fields Q3(v^±3) by using ^(ti9) , and it turns out that the zeros are contained
in Q3(v/3, \/2). In this case there is no zero of any L}(s, dy/„), n > 0,
corresponding to 70 ; indeed, u3(log(l + 7b)) = 1/2, and so v3(sq) - -1/2 .In these two examples, the approximate values of the zeros given in Table
VIII are easily computed by hand. A third example of the same kind, not
included in the tables, is (3, 7244, 0). In this case we have E = Q3(-v/-3).
Type 6. vp(ao) > 2 , vp(ax) = 1 , X > 3 , X - 1 prime to p .Among the zeros To - Tk , k = I, ... , X, there is one, say Tx , belonging to
Qp and satisfying vp(Tx) > vp(ao) - 1 > 1. The other zeros satisfy vp(Tk) =
\/(X - 1) (k = 2, ... , X). For T0 = Tx , Proposition 3 yields
^ = ¿^-^ = ¿^ = 1.
Thus, it is sufficient to look at &(p3), and this gives t0 = xp with x = -aoi/ax x
(mod p).The computation of the remaining zeros is analogous to that in the "first
type". In particular, E/Qp is a fully and tamely ramified extension of degree
X - 1. We find that vp(Tk - 7)) = 1/(A- 1) for j = 1, ... , X, j ¿k, and so
y = X- 1.To the first zero Tx there always corresponds a zero sx of Lp(s, 8). The
same holds true for T2, ... , Tx in the case X < p, while the situation varies
for X > p .Our examples with X = 3 < p are the characters (5, 4924, 2) and (7, -4072,
5). The character (3, -25528, 1) has X = 5 (computed first by Kobayashi),and s2, ... , s$ lie outside of Ds. So also do 52, s3 in the example (3,3512,0),where X = 3 . Compare this example with (3, 7804, 0) from Table Via: here,
too, X = 3 but 52 and s3 are zeros of Li,(s, 8y/±). Finally, the character
(3, -5051, 1) in Table V (Part I), for which fe(T) is of type Z , also fits thiscategory.
5. Wildly ramified extensions
All examples of wildly ramified extensions E/Qp were found for p - 3.
The degree of the extensions is 3 or 6.
Consider first the case of cubic extensions. Let X = 3 and v3(a0) = 1,
so that Vi(T0) = 1/3 and E/Q3 is indeed wildly ramified of degree 3. Onemay identify E (up to conjugates) by providing a cubic Eisenstein polynomial
r(X) G Z3[X] such that E = Q3(tt) with r(n) = 0. In the table below, the listof 11 such polynomials r(X) corresponds to a complete system of nonconjugate
fields E . This list is extracted from results by Amano [1], who characterizes in
this way all ramified extensions of degree p over any finite extension of Qp .
We will give a further characterization of the fields E, better suited for their
computational identification. Note that every r(X) is of the form
r(X) = X3 - 3aX2 - 3bX - 3c,
where a, b,c G Z with c = 1 (mod 3). Let 71 and E be as above. We
have 3 = e?r3, with e a unit of &e . and from r(n) = 0 it follows that
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
398 R. ERNVALL AND T. METSANKYLA
e = l/(c + bn + an2). In particular, e = 1 (mod n). Write
CO
e = \ + Y,Wv, e„G{0,±l}.i/=i
A straightforward calculation shows that our fields E can be identified by the
triple (ex, e2, e3) as indicated in the table.
AxA2
BxB2
B3
r(X)
X3 - 3X - 3
X* + 3X- 3A-3 - 3X2 - 12
X3 - 3X2 - 3
X3 - 3X2 + 6
(£i, e2.f3)
(-1.+1.-1)(+1,-1-1,+1)(0,-1,-1)(0,-1,0)
(0,-1,+1)
c,
c3
^2
^3
r(X)
X* + 1X2- 12A-3 + 3X2 - 3
X3 + 3X2 + 6
A-3- 12
A"3-3
A-3+ 6
(ei,£2>í3)
(0,-1-1,-1)(0,+l,0)(0,+l, +1)
(0,0,-1)(0,0,0)(0,0,+1)
Here the fields are labeled with Ax, A2, Bx , etc. The extension E/Q3 is
cyclic for E — Cx, C2, C3 and pure for E — Px, P2, P3.
We computed 17 examples in which the zeros belong to the fields of this table;
eight of them were selected to be exhibited in Table IX. In the entire sample of
17 fields, Ax and A2 occur five times each, supporting the natural hypothesis
that the values of ex be randomly distributed.
The functions L^(s, 8) and fg(T) were computed with the parameter n =
16. This implies, by Propositions 1.12 and 1.7, that the coefficients a¡ are
obtained mod 38 J (j —
us to determine the field
0, ... , 7). It is seen below that this accuracy allows
E in all examples but one, and to find the zeros
T0 = Y,xvnv (xve{0,±l},xx¿0)
to 1-4 Xy-places. We computed 7b by hand, thus avoiding the rather tedious
implementation of the arithmetic of the fields in question. The results were
checked by computing fe(To). From §1.3 it follows that the corresponding
numbers Sq are not zeros of any L3(s, 8y/n); hence they are ignored.
To find E and T0, start with the approximation congruence ©
reduces to the congruences
«01+030^1=0, aoxex +axxxx + a40 = 0 (mod 3),
(rc5). This
which give us xx and ex . If ex = -1 or +1 , the field is Ax or A2, respectively,
and the coefficients e2, e3, ... are determined by the equation e = 1/(1 ± n).
Better approximations for T0 then follow easily from W(nm) for bigger expo-
nents m.
In Table IX, the characters (3, -3592, 1) and (3, 1781, 0) are examplesgiving the field Ax . The latter requires slightly different techniques since v3(ao)
= 2. In both examples, the zeros Tx , T2, T3 lie in conjugate extensions
generated by the roots n = nk of r(X) = 0 (k = \ ,2,3).Similarly, the characters (3, 281, 0), (3, -311, l),and (3, -2132, 1) yield
examples in which E — A2. The first of these is also mentioned in §1.10, with
another normalization of r(X). The second example is of type Z , while the
third shows a function fg(T) with two zeros (one trivial) in Q3.
Here are the other examples of Ax and A2 computed by us.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
COMPUTATION OF THE ZEROS OF p-ADIC L-FUNCTIONS. II 399
Example 1. For nk a zero of X3 - 3X - 3 (/c = 1,2,3),
(3, 1397,0)(3, 5368,0)(3,6712,0)
Tk = nk-n2k- n\ (mod n5k),
Tk = -nk -n2k-n3k + n4k (mod n5k),
Tk = -n\ + n\ (mod n\).
Example 2. (3,-2920, 1): nk zero of X3 + 3X - 3 (k= 1,2,3),
Tk = nk + n¡ + n¡ (mod n¡).
Example 3. (3,-4184, 1): Tx=0,sx=0,
nk zero of X3 + 3X - 3 (k = 2, 3, 4),
Tk = nk-n\ + n\ (mod n\).
If ex = 0, we compute e2 from the congruence
-a0xe2 = a50xx + a2x (mod 3),
which is a consequence of W(n6) .
Let first e2 = 1 , so that £/Q3 is cyclic. Then ^(n1) gives e3 and so fixes
E. When computing further places for To, one should observe that E now
contains all the three zeros of fg(T).
We have two examples of cyclic fields, in fact, of E — Cx . One, included in
Table IX, is for (3, 2504, 0), the other is given below.
Example 4. (3, 5624, 0) : tt zero of X3 + 3X2 - 12,
Tx = -n - n3, T2 = -n + n2 - n3, T$ = -it - n2 (mod 7i4).
Secondly, suppose that e2 = 0. Then E is a pure extension of Q3 and one
finds e3 and x2 by solving simultaneously a pair of congruences produced byg?V).
In Table IX, this case is represented by the characters (3, 4172, 0) and
(3, -1144, 1), which lead to the fields Q^v'Tl!) and (Qbt-v^ó), respectively.The next two examples give our other characters of this kind. Example 6 in-
troduces a case with X = 4 ; here the ultimate identification of E must be left
open.
Example 5. For nk a zero of X3 + 6 (k = 1, 2, 3),
(3,401,0): Tk =-nk + n2 (mod n3),
(3, 4472, 0) : Tk = nk + n\ (mod tt3).
Example 6. (3,6856,0): 7*, = 0.2201, sx = 2.000,
nk zero of X3 - a (k = 2, 3, 4) with a = 4, 1,
or - 2,
Tk = nk (modn2k).
After this, it is also clear how to deal with e2 = -1 corresponding to the
fields Bx, B2, B?,. We have no example of these fields.
Now let us turn to the case of sextic extensions. The basic case is the one
with X = 6 and v^(ao) = 1 . As usual, put E = Q3(70) = Q3(jt) .
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
400 R. ERNVALL AND T. METSÄNKYLÄ
Proposition 4. For a fully ramified sextic extension E = Qi(n),let t be the unit
of the ring &E defined by the equation 3 — en6. The field E contains the subfield
Q3(\/3) or <Q>3(\/^3) according as t = 1 ore = -1 (mod n), respectively.
Proof. The polynomial q(X) — X2 ■+ 3 satisfies vn(q'(n3)) = 3 and q(n3) =
n6(l +e) = 0 (mod tt7) provided that e = ±1 (mod n). D
It follows that E is a ramified cubic extension of Q3(a/3) or Q3(\/^3).
Hence, it can be classified in the way shown by Amano in [1]. We do not write
down the complete classification but, rather, show how to handle the specific
examples we have of this case. They are the three examples presented in Table
X, two of them with X = 6 and one with X = 1 and fe(T) of type Z . In theseexamples we chose the truncation parameter r\ = 18 or 20 according as X = 6
or 7, respectively. This enables us to identify E and compute To mod tt3. As
mentioned in §2, we never have E = Q3(Í9) ; hence, we do not compute So.
Put p = y/(-l)z3, where z = 0 or 1. By Proposition 4, E = Q3(/?, it),where n is a zero of an Eisenstein polynomial r(X) e Qi(p)[X]. From [1] we
find that it suffices to consider the polynomials
r(X) = X3 - apX2 - bpX - cp,
where a, b, c G Z and c = 1 (mod p). Let k be the unit in Q3(/>) such that
p = K7T3. A similar argument as before yields k = l/(c + bn + an2), and we
may write
CO
K = 1 +YKvn" ' K„G{0,±1}.i/=l
With the usual notation To - YlTLi xv71" (where xx ^ 0) we get from W(n8)
the congruences
(4) (-l)za0i+«60 = 0, -2a0XKX = (axx + (-l)za10)xx (mod 3).
The former determines p, and the latter tells us, first, whether or not kx = 0.
If kx ^ 0—as is the case in our examples—then b ^ 0, and we arrive, by [1],
at the polynomials
r(X) = X3 + pX-p,
which generate six ramified extensions of Q3(/>). The six zeros of f$(T) are
contained in these fields, one in each. It also follows that k = 1/(1 ± tt) and
so, in particular, k2 = 1. Hence the approximation congruence ^(tt9) enables
us to compute x2.In case one is interested just in knowing whether or not E — Q3(Í9), it is
worth noting that (4) yields the following necessary conditions for this equality:
«oi = «60 ; «it = «70 • Indeed, we have z = 1 and kx =0 in this case.In our examples obtained for the characters (3, -7108, 1 ) and (3, -20692,
1), the first congruence in (4) gives the result z = 0 and z = 1, respectively,
and the second gives kx = xx . For the third character (3, -1832, 1) we find
that z = 0 and kx = -xx . See Table X for the full result.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
COMPUTATION OF THE ZEROS OF p-ADIC L-FUNCTIONS. II 401
Table VI. X-2, ao^O, sx¡2 zeros of Lp(s. 8y/±)
(3, 653, 0)
0.211000.001020.020010.000220.002200.000020.00020
0.02000.00220.02220.00020.00210.00000.0002
± O.OOOOti±0.111171±0.0011 ti±0.021 In± 0.000271± 0.000171± O.OOOOti
(3, -379, 1)
0.200000.001110.021220.001120.002000.000020.00021
0.00000.00020.02000.00120.00220.00000.0002
±0.110071±0.2212ti± 0.00227t± 0.01017t± 0.000271± 0.00027t± O.OOOOti
0.211000.02022.2121.012.02.
0.200000.01212.2102.211.10.
(3, 1153, 0)
0.110000.002100.010020.000120.001210.000010.00012
0.0110 ± O.OOOOti0.0010 ±0.101l7t0.0111 ±0.0010710.0001 ±0.0212ti0.0010 ±0.0002710.0000 ±0.00017t0.0001 ± O.OOOOti
r,,2 0.12 ±2.27t
:0.0±(l.)7t
(3, -1336, 1)
0.220000.001020.022120.001220.002100.000000.00022
0.01000.00200.02100.00100.00200.00000.0002
±0.210071±0.122Û7t±0.000l7i± 0.02007t± 0.OOO27I±0.0001n± O.OOOOti
71
Tl,2
■Î1 .2
0.20 ± 2.071
1.0±(0.)7t
0.110000.02021.2022.010.12.
0.220000.01002.2002.000.10.
Table Via. X = 3, 52,3 zeros of Lp(s, 8y/±)
00.0122220.0012010.0000210.0001200.0000100.0000110.000001
r, =Tí
Tl.3
*2,3
(3, -827, 1)
0.0200 ±0.01227t0.0112 ±0.02227f0.0010 ± 0.0021710.0000 ±0.0010ti0.0001 ±0.000l7t
0, i, =0
■■ 0.0000 ±2.210tt
: 1.011 ±0.11tc
00.2000110.020112.21000.0222.021.21.
0.011000.012200.000000.000200.000210.000000.00001
(3, 7804, 0)
0.0020 ±0.0212ti0.0120 ±0.0222tiO.OOOO ±0.000 Itc0.0002 ±0.00217!0.0002 ± O.OOOOti
T¡ =0.2100,
Tt
Í2.3
j, = 2.001
= v/=3
= 0.1 ± 2.07t
= (0.)±(1.)TI
0.011000.10100.1121.212.22.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
402 R. ERNVALL AND T. METSÄNKYLÄ
Table VIL X = 2, a0¿0, vp(Tx -T2)>2
(3, 13564, 0)
0.0110000.0101110.0120110.0011020.0012100.0000100.0001010.0000000.0000020.0000020.000002
0.01100000.1122101.211011.11011.2211.021.10.
(
Tl.2
*1.2
V2: 0.10201 ±0.01110í
: 1.0111 ± 0.1120Í
(3, 13784, 0)
0.0220000.0111100.0202210.0012120.0020010.0000100.0002210.0000000.0000010.0000020.000001
0.02200000.2221002.111122.02120.1210.022.22.
n = n/3r,.2 =0.10122 ±0.011271
j,,2 = 2.1221 ± 0.20171
(3, -11188, 1)
0.0110000.0110000.0100000.0021120.0010210.0000100.0001120.0000020.0000020.0000010.000002
0.011000000.12222011.0120212.121011.20002.1000.120.10.
í7),2
s\.2
■ y/20.102122 ±0.012001í
1.01000 ±0.12121{
Table Vila. X = 3, ao = 0, vp(T2 - T3) > 2
(3, -2564, 1)
o0.002011200.002002010.002111010.000011120.000001100.000012220.000001120.000000100.000001000.000000010.000000000.00000001
o0.01222200.2112101.001011.20002.0121.20
1.21.
7Ï = O , si
J2,3
*2,3
O
= \/2^3= 0.20112 ±0.010l7t
= 1.2022 ± 0.210a
(3, -8804, 1)
O0.001012020.002200210.001110210.000211210.000021210.000020220.000001110.000000100.000002120.000000120.000000010.00000002
O0.02111220.2221102.111020.01000.1200.211.00.
7*1=0,
n
T2,i
s2,i
sx = O= v/3
= 0.10001 ±0.0102tt
= 2.2121 ±0.220tt
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
COMPUTATION OF THE ZEROS OF p-ADIC /.-FUNCTIONS. II 403
Table VIII. Various types of the Newton polygon
(5, 3101, 0)
0.0040120.0401130.0311100.0044030.0003420.0000130.0000420.000003
0.004010.423303.210121.41141.2341.314.20.
7, = 0.040243
5, =0.42140
7*2 =0.224113
s2 =2.12443
(3, 1901, 0)
0.00200.00120.01200.00100.00110.00000.0001
0.00200.0221.111.00.
= n/3,2 = 0.020 ± 0.1 lTt
2 = 0.12 ± 2.071
(3, 3512, 0)
0.011000.010220.001110.001220.000120.000000.00001
0.011000.21010.0101.120.22.
7',
S\
n
7-2.3
*2,3
= 0.1120
= 2.121
= v/3
= 0.1 ± 1.271
= (1.)±(12.);
(7, -4072, 5)
0.00150400.03343320.00210230.00504450.00060130.00004240.00000140.0000006
0.00150400.2454600.364542.04240.2600.123.02.
r,S\
n7*2.3
*2,3
= 0.0352103= 0.230602
= 0.101 ±3.45ti
= 0.163 ±22.271
-4371, 1)
00.0003030.0032340.0033020.0000030.0000210.0000030.000001
00.0032040.313013.11423.4041.304.10.
7Ï
Si
r,S3
0, s, =0
0.041020.4240
0.443404.2101
(11 , -723, 1)
0.0009aaa0.008a 1090.00629060.00084a30.00006570.00000280.0000006
00.00826790.7457591.473206.76830.1168.701.3
7',
7*2
Î2
7*3
S}
= 0, s, = 0= 0.0246697
= 0.33543a= 0.42a5014
= 6.7852a0
(5, 12056, 2)
0.00424120.00042210.01342440.00213040.00002100.00000020.00003020.0000043
0.00424120.0040011.221401.11100.1121.214.21.
71 = VI7*1,2 = 0.000443 ± 0.104207t
5,,2 = 0.02023 ± 1.222Itt
(3, -25528, 1)
0.022000.022210.000000.001200.000020.00000
0.022000.22100.2100.220.11.
7',
S\
It
37*2.3
32*2,3
n1
37*4,5
32-*4*. 5
(3, 1541, 0)
01100002120022100001000020000100001
0.011000.01210.2100.201.20.
71 = ^=3
7*|,2 =0.1 ±(0.)7t + (l.)Tr ±(l.)7t3
Si ,2 = ±l/7t (mod 7t)
.. ' r . 4/ ^71 = Í4 " V -3
7*3.4 = 0.1 ± (O.)tt' + (I.)tt'2 ± (I.)ti'3
S3.4 = ±1/ti' (mod n')
(5, -7816, 1)
00.00020220.00013110.00302110.00034430.00003200.00000110.00000130.0000003
00.0020440.012323.00301.1321.302.33.
'i 0 , s, = 0
n = \/5
7*2 3 =0.04202 ±0.121l7t
s2,3 = 0.4414 ± 1.23071
= 0.2120= 2.111
= v/3
= 0.0±0.l7t + (0.)7t2 ±(0.)7t3
= 0.0 ± 0.27t + (0.)7t2 ±(0.)7t3
= Í4-v-3
= 0.0±0.l7t' + (0.)7t'2 ±(0.)7t'3
= 0.0±0.2ti' + (0.)7t'2 ±(0.)ti'3
(5, 4924, 2)
0.03133210.03120320.00200040.00241020.00010130.00001340.00000200.00000440.0000001
0.03133210.2212130.120023.04304.0043.133.20.
7*1
S|
n7*2.3
•52,3
0.110124
4.43041
v-5:0.22± 1.2271
: 1.3 ±44.3ti
(3, 4204, 0)
0.011000.001210.000120.000220.000120.000000.00002
0.011000.01010.0220.021.21.
7*1.2
7*3,4
\/3. ( = V20.0+((l.)±(l.){)7t
0.0-((l.)±(l.)f)7t
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
404 R. ERNVALL AND T. METSÄNK.YLÄ
Table IX. Zeros in wildly ramified cubic extensions
(3, -3592, 1)
0.22000000.02211200.00200120.00022210.00002010.00000000.00001210.00000020.00000010.0000001
0.22000000.2121020.122012.10121.2011.202.00.
Ttfc zero of A-3 -
(A: = 1,2,3),
Tk = -T-k - "Í
(mod 7t|)
ÎX
(3, 2504, 0)
0.22000000.01200120.00122010.00100200.00021020.00002200.00000010.00000120.00000020.00000220.0000001
0.22000000.2202020.000201.11121.1121.022.21.
7i zero of X} + 3X2 - 12,
T, = Tt - Tt3,
7*2 = 71 + 7t2 + 7I3 ,
7*3 = 7t - K2 - 7i3 (mod 7i4)
(3, 1781, 0)
0.01100000.00110210.00202210.00202100.00020120.00000110.00002220.00000100.00000010.00000110.0000001
0.01100000.0211200.220011.21000.2011.200.21.
7t,i zero of X1 -
(k = 1, 2, 3),T- — 2 3Tk = -nlk - n'k
(mod nl)
IX
(3, 4172, 0)
0.12100000.02001110.00110200.00222210.00000220.00002120.00002210.00000010.00000020.0000011
0.12100000.1101010.211102.00222.2121.122.02.
nk zero of X1
(k= 1, 2, 3),
Tk = nk + k\
(mod Tt3.)
12
(3, 281, 0)
0.20000000.02001220.00100100.00010110.00010020.00002210.00001210.00000210.00000020.00000010.0000002
0.20000000.1001110.202001.10110.0201.221.02.
Ttt zero of X3 + 3X
(£ = 1,2,3),
Tk = Kk + n¡ + n*k
(mod 7t^)
(3, -1144, 1)
0.22000000.02102000.00020020.00012220.00020100.00000100.00001110.00000010.00000000.00000020.0000001
0.22000000.2010010.212212.11202.1112.111.12.
nk zero of X3 + 6
(it = 1, 2, 3),
Tk = -K* + "k
(mod Tt3.)
(3, -311, 1)
00.02121210.00210220.00122110.00011200.00002020.00000200.00000110.00000010.00000200.0000002
00.1020100.022000.22122.1022.221.00.
r\ = o, s, = o
nk zero of X3 + 3X ■
(/c = 2,3,4),
Tk=nk + n¡ + %\
(mod n*)
(3, -2132, 1)
00.00110010.00222020.00010120.00021120.00002220.00002210.00000210.00000010.00000010.0000001
00.0211010.211020.21210.0011.112.00.
T, = 0, s, = 0
7*2 = 0.211, s2 = 1.12
nk zero of X3 + IX - 3
(jfc = 3,4, 5),
Tk = %k (mod n\)
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
COMPUTATION OF THE ZEROS OF p-ADIC L-FUNCTIONS. II 405
Table X. Zeros in wildly ramified sextic extensions
(3, -7108, 1)
0.121000000.020100220.001110220.001121100.000122110.000020120.000010020.000001000.000000220.000002110.000000220.000000010.00000001
0.121000000.21220120.0022120.212200.02100.2012.122.11.
kx , Tii, it-i zeros of A"3 - VÏX - %/3
7*! s -Tti + n2 (mod 7C3)
Ti = -ni + n\ (mod n\)
7*j = -7T3 + n2 (mod n\)
iti,, Tt5 , Ttf, zeros of A-3 + \ThX - VÎ
T4 = Tt4 + n\ (mod 7C3)
7*5 = tc5 + it2 (mod tt.3,)
T6 = 7t6 + xj (mod n\)
(3, -1832, 1)
00.0111010220.0011222110.0020002110.0001122110.0000101220.0000120110.0000022220.0000002110.0000012010.0000002020.0000000220.0000000100.000000002
00.221101100.01011010.1101010.210010.11120.2211.001.01
7*, = 0 , sx = 0
it2 . ni. ^4 zeros of A-3 - \fi>X - V3
T2 = 7i2 + n2 (mod n\)
7*3 = 7Î3 + n2 (mod 7t|)
7*4 =: 7t4 + tt^ (mod 7rJ)
7i5, 7t6 , 7i7 zeros of A"3 + v^A" - v^3
T¡ = -n¡ + n2 (mod 7r|)
7*6 = -7t6 + tt| (mod 7T3)
7*7 =i -7t7 + tí 2 (mod nl)
(3, -20692, 1)
0.121000000.000111010.000022200.000002010.000002010.000000110.000001210.000000200.000000010.00000001
0.121000000.00120000.0001020.012010.00200.0101.002.02.
7tx , iti, ni zeros of X3 -3X
Tx = -n 7T, (mod n
T2 = -iti - n2 (mod n\)
7*3 = -tit, - it2 (mod n\)
7t4 , 7r5 , 7t6 zeros of A-3 + v/-3A" -
7*4 = 714- it\ (mod 7T3)
75 s ne, - Tt2 (mod 7i|)
T6 = n6- n\ (mod tc3)
^3
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
406 R. ERNVALL AND T. METSÄNKYLÄ
Bibliography
1. S. Amano, Eisenstein equations of degree p in a p-adic field, J. Fac. Sei. Univ. Tokyo Sect.
lAMath. 18(1971), 1-21.
2. R. Ernvall and T. Metsänkylä, Computation of the zeros of p-adic L-functions, Math. Comp.
58 (1992), 815-830; Supplement, S37-S53.
3. L. C. Washington, Introduction to cyclotomic fields, Springer, Berlin and New York, 1982.
Department of Mathematics, University of Turku, SF-20500 Turku, Finland
E-mail address: taumetsQsara.cc.utu.fi
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use