On the Parabolic Generators of the Principal Congruence Subgroups of the Modular Group

Author(s): Emil Grosswald

Source: American Journal of Mathematics, Vol. 74, No. 2 (Apr., 1952), pp. 435-443
Published by: The Johns Hopkins University Press
Stable URL: http://www.jstor.org/stable/2372008
Accessed: 08-11-2017 00:47 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

The Johns Hopkins University Press is collaborating with JSTOR to digitize, preserve and
extend access to American Journal of Mathematics

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
 ON THE PARABOLIC GENERATORS OF THE PRINCIPAL CON-
GRUENCE SUBGROUPS OF THE MODULAR GROUP.*

By EMIL GRoSSWALD.

1. The set of matrices (a b) where the integers a, b, c, d

(1) ad -bc 1

and where we co
define r(p), the principal congruence subgroup of r modulo a prime p, by
the additional conditions a! d 1 and b = c -= 0 (unless otherwise stated,
all congruences are taken mod p). For p > 3, H. Frasch has shown ([3];
see also [7]) that r(p) can be generated as a free group by the generators

SP= (O ) and (p -l)p(p + 1)/12 other generators, depending on three

parameters, identified by the symbols (A, u, v) and defined as follows. Let

( 2 ) V = TSaTSOTSa=(_(P1) c ( 2))'

where a and , are any two primitive roots (mod p) satisfying ,83 = 1. Let
A, ,u, v be any set of integers satisfying 0 ? A ? (p - 3)/2, 0 ? K ? p -1,
1 ? v ? p -I, and define

r=indv, X* = +- (reduced mod(p -1)),

(3) a
U* ,kV2 v (reduced mod p), vv* 1 (mod p)

so that A*, p*, v* are in the same range, respectively, as A, ,u, v. Let us call
" reciprocals " two numbers related as v and v*. Finally, let

(4) R I SATSvTS-v*TS-A*

(-/4vv* + I) + v* w*(vv* + 1) -(w+i*v*) + 1

V (vv* + I) /-t*(vv* + I) -v J

Then the generators (An, , v) of r(p) are defined by

* Received January 12, 19a51.

435

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
 436 EMIL GROSSWALD.

(5) (A, u, v) - xRV-N*.

The object of this paper is to show that if the prime p satisfies a certain
condition ' to be stated later, the independent generators (A, u, v) can be
chosen in such a way that none of them is parabolic,2 so that the only parabolic
generator of r(p) is SP.
For our proof, we have to change slightly the process of elimination
indicated by Frasch [3], whose method we follow, however, closely. Through
out the paper we consider only values of p > 7 (cf. a similar situation in [7],
section I); in the other cases we would need sometimes special considerations,
which complicate needlessly the exposition, as the corresponding generators
are explicitly indicated in [3] and none of them is parabolic. A few inequali-
ties will be meaningful only for some higher values of p and will be under-
stood to apply only for them.

2. We recall some results of Frasch [3], with some of their immediate

consequences. The -p(p - 1)2 generators (A, u, v) satisfy p(p - l)(5p - 7)/12
distinct defining relations, each permitting the elimination of one generator.
We use them in order to eliminate all generators with a same value v0 of v,
unless v0 satisfies

(6) vO2 I or (6') vo(vo -)-1

The defininig relations can be arranged in groups of binary and ternary

relations. Following Frasch, we use the first group of binary relations to
eliminate the generators with v = 1. The following ternary relations permit
the elimination of v = p - 1. Using the second group of binary relations,
v - 2 and v = p - 2 are eliminated; and the corresponding ternary relations
permit the elimination of their reciprocals, v = I (p + 1). In general,
for kl> 3, we use the binary group to eliminate v = k, v = p - k, and
v {p -+- (2k - 3) }, and the corresponding ternary relations to eliminate
their reciprocals. The generators with the mentioned values of v actually
appear in the defining relations in the stated order, as explicitly shown in
[3]. The number of relations at our disposal is larger than the number of
generators on which we have centered our attention, and we may use the
remaining ones in the order suggested in [3]. The process of elimination

1 It seems likely that all primes satisfy that condition; see last footnote.
2A corresponding property holds for the generators Vk of P0 (p), defined in [9].
See [5].

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
 ON SUBGROUPS OF THE MODULAR GROUP. 437

has to be modified for p * - 1 (mod 12), whe

(6) or (6'). Let v k or v ==p-k satisfy (6). Then k2 + 1I 0, so that
1c2 -(p I 1) _ 0, and k ? (p - 1)'. If v satisfies (6'), then k(k ) - ,
so that _2 - k (p I) O k 1 + (p -) > (p- )t. In the same
way we show that if v= {p + (21 -3)} satisfies (6) or (6'), then
kc > ' (p' + 1). It follows that all values of v belonging to any one of the
three intervals

(7) I v -< pa -; 12(P Pa) + I < v < 21(p + Pa)

p - P + I < V ? p-i

are not affected by the difficulties arising from (6), (6'), and can be eliminated
from the final set of independent generators, using only the binary relations.
Whenever we reach any value satisfying (6), or (6'), we continue as in [3].
If v does not satisfy (6) or (6'), respectively, then the same is true of its
reciprocal, and the generators with values of v reciprocal to those of (7) will
also be eliminated, using the corresponding ternary relations. The values of v
in (7) and their reciprocals form a set of at most 3 (pl - 1) distinct numbers.
For the elimination of the corresponding generators we dispose (for some
fixed A and kt) of -1-p (p - ) (5p -7) /j p (p 1 ) (Sp- 7) /6 relations.
For p > 7, (Sp - 7)76 ? 3(pa - 1), proving the previous assertion that the
number of independent defining relations exceeds the number of generators
which we want to eliminate. Finally, we remark that there remain in the
generators only values of v, which satisfy, with their reciprocals, the inequalities

(8) pa- < v v* < j2(p -p) + I., or 2 (p + pa) < v v* < p pi + 1

3. From (3) and (5) it follows that

V-X(A, ,v)VX=l RV-T, if X+oa< (p I), and

(9)
V-x , v)Vx RV+(p--), if X + 4(p-1)

If (A, ju, v) is parabolic, then all its transforms M-1 (A, ,u, v) M by any modular
matrix 111 are also parabolic. In order to prove that (A, J,t v) is not para-
bolic it is, therefore, sufficient to show that RV-I and RV-ff+'(P-1) are not

parabolic. Let, generally, VT= (C Dj3 Then it follows from VT7tL

Vn. V (An+k Bn+1 that
Cn+ Dn+1

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
438 EMIL GROSSWALD.

An+i /A -(<Xi f-1)hBn Bn+i= (awe-1)An -ca(ax -2)Bn

(10)

Cn+i w 8Cn - (Xi] -L ) Dn Dn+i ( (a,B -L ) Cn -(X (a,x - 2 ) Dn.

For n 1, by (2), A1 D1= BQx(ac (/ )/(-c (aI)) and B1 + C, = O.

Assuming that

(11) An-Dn 2tBn, where t 2(a- (a -/3)/(ax -1)),

(12) Bn+Cn ?

hold for n, it follows by induction on n, using (10), that (11) and (12)
hold for n + 1; they hold, therefore, for all n. It follows from (1) and (12)
that

(13) AnDn + Bn2 1

and, as Bn is an integer,
(14) AnDn_ ? 0.
Hence, by (11) and (12),

(15) AnBn > O AnCn < 0.

For n 1, by (2), B1 > A,; assuming that, f

(16) 1 Bn I > I An 1n

it follows by induction on n, using (10) and (15), that (16) holds for n + 1
and so for all n. From (16) it obviously follows that (aq3 - 1)Bn/An - /3> 0
so that, by (10), An+,An < 0. Hence, using (15), (12) and (14), BnBn+, < 0,
CnCn+i < 0 and DnDn+ _ 0. We shall see immediately that the equality
cannot hold, so that we have, generally, with M standing for A, B, C or D,

(17) MnMn+l < 0.

Let, generally, oc3 = Kp + 1, K _ 1; then B1 =- Kp. From (10) it

follows, by (14) and (17), that j Bn+l/Bn | a (a,8 -2) - (a8 -1)An/Bn
> ac (ox -2)- ( 1) a(Kp -1) -Kp Kp -2 so that Bn ?
> j B I (Kp - 2) n-I Knpn(I - 2/Kp) n-I and, for p > 7,

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
 ON SUBGROUPS OF THE MIODULAR GROUP. 439

IHence it follows in particular that I Bn I > 1 a

and in DnDn+1 < 0 may be suppressed. We have
and (12),

(i9) AnDn < 0, BnDn < 0, CnDn > 0.

We solve now (11) and (12) for An and Dn, using (16) and (19) to determine
-the signs of the radicals and obtain

An B-n(t -(t2 l + Bn-1)')

Dn Bn(t + _2IL + Rn -2)1).

As, by (11) and (18), 2 < 2t < Bn, it follows that

t- _ /2t > (t2 _ + Bn-2)' > t 1/(2t I)

so that

2t -I/ (2t -I) < I Dn/Bn I < 2t - /2t

(20)

1/2 t < An/Bn < I/ (2 t- 1).

4. Let now, for integral n> 1, RV-n= R(' An)=(a9 Dn)

If this matrix is parabolic, then we have (see [2, p. 21]) | a F + j = 2,

where, by (4),

aUn - Dn [j(vv* + 1) - v*] Cn[IW*(v1'* + 1) - (,uv + ,u*v*) + 1]

(21)
9n Bn(vv* + -) + AnL[IA(vv* + 1) -v].

If tw* -4 0, all brackets are positive, as is obvious for all except the coefficient
of Cn. Let its value be N, and apply (1) to R. It follows that S > 0, and
by (15), (18) and (19), we obtain I (tn+ in I > I Bn I (N-+vv* + I1)
> (5p/7) n(N + p) > 2, so that RV-n cannot be parabolic. If y = 0, then,
by (3), ,u* p - v. From (15) and (19) it follows that I an, +2)n I
=| Ia In I - On I I; and by (20) and (21), we have l< I aC/Bnt < L,
H > I 21)n/Bn j > h, with
1= (p-v)v*-l + [2t-1/(2t-1)]v*,

L (p- v)v* -i + (2t -1/2t)v*,

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
 440 EMIL GROSSWALD.

h = (vv* + 1) + [(p -v)(vv* + 1) - v]/2t,

11= (VV* + 1) + [(p -v)(vv* + 1) -v]/(2t -1).

Consequently,

(22) In | | |> I Bn I (h L)
and we assert that

(23) h -L> 1/2t

if

(24) either /3 < a _ v or a v -,/3 > a.

Proof. (23) is equivalent to

4t2 2t[2(v + 1/v*) p] [(p v)(v + 1/v*) -(v+1)/v* + 1] ? 0,

which is satisfied for 0 ? t ? iv, or, by (11), for a C v + (- /3)/Qx/3 -),.

proving the assertion.

By virtue of (8), (24) is always satisfied for

(25) cc?p-2

and also for

(25') a <(p + ph) -2 if 1(p + pi) - <v< p -p +.

Similarly,

(26) - I | n > I Bn H)
where

(27) 1 -H I/(2t -1) if (28) a v + 2.

Proof. (27) is equivalent to

4t2- 2t[2(v + l/v*) -p + 1]

-[(p -v - ) (v + I/v*)- (v -2)/v* + p- v + ] ?O,

which is satisfied for 2t ? v + 1/v* + 1, or, by (11), for a > v + I/v* + 1

-(a 3- f)/Aa/3 - 1) proving the assertion. In view of (8), a satisfies (28),
provided that
(29) a?p -p'+3
and also for

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
 ON SUBGROUPS OF THE MODULAR GROUP. 441

(29') o c - p ) + 3, if 21v( -P,) +I

Combining (25') and (29'), it follows that for any possible value of v either
(23) or (27) holds, in case we have simultaneously 3

(30) p) + 3 < a < 2 ( + P) 2.

We shall say that a prime p has the property

following three alternatives holds:

(a) the smallest primitive root of p does not exceed p' - 2;

(b) the largest primitive root of p is not less than p - pl + 3;

(c) there exists a primitive root of p in the interval (30).

Let us choose for a a primitive root satisfying either (a), or (b), or (c).
From (22), (23), (26), and (27) it follows now, on account of (18), that
if p has the property C, then

(31) | Ca. + n I > I Bn 1/2t > (5p/7)n/p.

By (3) ca0>==v and, by the choice of a, v#ac; consequently, u 74&1 in (9),

and it is sufficient to consider in (31) only values n ? 2. As for n> 2,
(5p/7)n/p > 2, it follows from (31) that a , + 9) I > 2, proving that
= 0, then RV-T is not parabolic. The case y* - 0 can be reduced to the
previous one by interchanging v and v*, which is possible without any further
change of the argument or of the chosen value of a, since v and v* were both
eliminated from the intervals (7). This finishes the proof that RV-I is not
parabolic.

Similarly, let RV1 a (7 . ) . Assuming that this matrix is para-

bolic, we obtain, proceeding as before,

(25), (29) and (30) may be improved to a<p2, p -a ?p- and [ a-p
respectively, by sharper estimates in (20). We remark also that (30) is meaningful
only for p > 29. For all smaller primes, however, either (25), or (29) hold. See also
remark at the end of 1.
4 Although it seems likely, that every prime has the property C, the best that is
known at present is a weaker result of Loo-keng Hua [61 ; see also [1], [4], and [8].
In [7] J. Nielsen considers only primes p, admitting as primitive roots either 2, or
p - 2 (consequently also I (p ? 1)). Property C represents a weakened Nielsen condi-
tion and every prime, satisfying Nielsen's condition, obviously has the property C.

12

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
 442 EMIL GROSSWALD.

1 aIn + -'n I - An[jI

(21') + Cn[ pX*(vv* + 1) - (ptv + II*v*) + ]
Bn(vv* + 1) + Dnpu * (vv* + 1) - v] = 2.

We observe that, except for its sign, (21') is obtained from the values (21),
interchanging the couples (pu, v) and (IA*, v*). Consequently, i an + ) , I; 2
implies I C'n + I) I #+ 2 contrary to (21'), and RV,& is not parabolic, unless
n= 1. From (p + 1) -u a 1 would follow, however, by (3), a = 0 and,
further, v a= a 0 1, contradicting (7). Therefore, in (9), the exponent of
1 is different from one and RVT(P+1)- is not parabolic. By (9) we conclude
that no generator (A, tt, v) is parabolic, so that the only parabolic generator
of r(p) is SP, q. e. d.

Remark. A separate consideration of the cases p = ? 1, + 5 (mod 12)

yields some additional information and suggests that the above result holds
for all primes, whether they have property C or not, with at most a finite
number of exceptions. As an example, let p - - 1 (mod 12). Then no
relations (6), (6') exist, and all values of v of the form v k7, v p - 7k,
1 ? k ? [(5p - 7) /24], and their reciprocals can be eliminated from the
generators. Erdos has shown [1] that every prime p has a primitive root
a < (log p) 17p2; and the considerations of 4 hold, provided a __ v -1, or

(32) (logp)17pC [(5p -7)/24] -,

which is true for all sufficiently large p. In the same way, results of FFueter
[4] could be used to improve our results in the cases p - + 1, and p = + 5
(mod 12). The interest of such considerations is, however, considerably
diminshed by the fact that " sufficiently large " means, for (32), p > e200.
Therefore, the additional results obtained do not seem to justify a detailed
case distinction (mod 12). It is rather to be expected that the known results
concerning the smallest primitive root (mod p) will be improved sufficiently
(see [8]) to show that all primes have the property C.

UNIVERSITY OF SASKATCHEWAN.

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms
 ON SUBGROUPS OF THE MODULAR GROUP. 443

BIBLIOGRAPHY.

[1] P. Erd6s, " On the least primitive root of a prime," Bulletin of the American Mathe-
matical Society, vol. 51 (1945), pp. 131-132.
[2] L. R. Ford, Automorphic Functions, New York, 1929.
[3] H. Frasch, " Die Erzeugenden der Hauptkongruenzgruppen fur Primzahlstufen,"
Mathematische Annalen, vol. 108 (1933), pp. 229-252.
[4] R. Fueter, " Ueber primitive Wurzeln von Primzahlen," Commentarii Mathematicae
Belv., vol. 18 (1946), pp. 217-223.
[5] E. Grosswald, " On the genus of the fundamental region of some subgroups of the
modular group," American Journal of Mathematics, vol. 74 (1952), pp.
86-88.

[6] L. K. Hua, "On the least primitive root of a prime," Bulletin of the American
Mathematical Society, vol. 48 (1942), pp. 726-730.
[7] J. Nielsen, "A study concerning the congruence subgroups of the modular group,"
Det Kgl. Danske VTidenskabernes Selskab. Matematisk-Fysiske Meddelelsen,
vol. 25, no. 18 (1950).

[8] S. Pillai, " On the smallest primitive root of a prime," Journal of the Indian
Mathematical Society (New Series), vol. 8 (1944), pp. 14-17.
[9] H. Rademacher, "Ueber die Erzeugenden von Kongruenzuntergruppen der Modul-
gruppe," Abhandlungen aus dem Seminar der Haamburgischen Universitdt,
vol. 7 (1929), pp. 134-138.

This content downloaded from 177.244.76.242 on Wed, 08 Nov 2017 00:47:51 UTC
All use subject to http://about.jstor.org/terms