Chapter XIV.
On the Linear Group87.

W e shall now, in illustration of the general principles that have been developed in the preceding chapters, proceed to discuss and give an analysis of certain special groups. The first that we choose for this purpose is the group of isomorphisms of an Abelian group of order pn and type (1, 1,, to n units). This group has been defined and its order determined in §§ 171, 172. It is there shewn that the group is simply isomorphic with the homogeneous linear group defined by all sets of congruences

y1 a11x1 + a12x2 + + a1nxn, y2 a21x1 + a22x2 + + a2nxn, ............ yn an1x1 + an2x2 + + annxn, (modp),
whose determinants are not congruent to zero; and that its order is
N = (pn 1)(pn p)(pn pn1).
The operation given by the above set of congruences will be denoted in future by the symbol
(a11x1 + + a1nxn,a21x1 + + a2nxn,,an1x1 + + annxn).

216. It may be easily verified by direct calculation that, if D and D are the determinants of two operations S and S of the group, then DD is the determinant of the operation SS, all the numbers involved being reduced, (modp). Hence it immediately follows that those operations of the group, whose determinant is unity, form a sub-group. If this sub-group is denoted by Γ, the group itself being G, then Γ is a self-conjugate sub-group of G. For if Σ is any operation of Γ and S any operation of G whose determinant is D, the determinant of S1ΣS is D1D or unity; and therefore S1ΣS belongs to Γ. Suppose now that S is an operation88 of G whose determinant is z, a primitive root of the congruence

zp1 1(modp).
Then the determinant of every operation of the set
SrΓ
is zr; and therefore, if r and s are not congruent (modp), the two sets
SrΓ andSsΓ
can have no operation in common. Moreover, if S is any operation of G whose determinant is zr, then SrS belongs to Γ, and therefore S belongs to the set SrΓ. Hence finally, the sets
Γ,SΓ,S2Γ,,Sp2Γ
are all distinct, and they include every operation of G; so that
G = {S, Γ}.

The factor-group G Γ is therefore cyclical and of order p 1.

217. It may be very readily verified that the operations of the cyclical sub-group generated by

(zx1,zx2,,zxn)
are self-conjugate operations of G. To prove that these are the only self-conjugate operations of G, we will deal with the case n = 3: it will be seen that the method is perfectly general. Suppose then that
T = (αx1 + βx2 + γx3,αx1 + βx2 + γx3,αx1 + βx2 + γx3)
is a self-conjugate operation of G, while
S = (ax1 + bx2 + cx3,ax1 + bx2 + cx3,ax1 + bx2 + cx3)
is any operation. The relation
ST = TS
involves the nine simultaneous congruences89

aα + bα + cα αa + βa + γa, aβ + bβ + cβ αb + βb + γb, aγ + bγ + cγ αc + βc + γc,  etc.,  etc.;

and these must be satisfied for all possible values of the coefficients of S. Now

b c a 0
is a possible relation between the coefficients of S, whether regarded as an operation of G or Γ; and therefore
γ 0.
In the same way, it may be shewn that
β α γ α β 0,
and that
α β γ;
so that T is a power of the operation
(zx1,zx2,zx3).
The only self-conjugate operations of G are therefore the powers of A, where A denotes
(zx1,zx2,,zxn);
and the only self-conjugate operations of Γ are those operations of this cyclical sub-group which are contained in Γ. Now the order of A is p 1 and its determinant is zn. Hence the self-conjugate operations of Γ form a cyclical sub-group D of order d, where d is the greatest common factor of p 1 and n; and this sub-group is generated by Ap1 d .

218. To determine completely the composition-series of G, it is necessary to find whether Γ has a self-conjugate sub-group greater than and containing D. A simple calculation will shew that, from

(x1 + xs,x2,x3,,xn)
and its conjugate operations, all the operations of Γ may be generated; and hence no self-conjugate sub-group of Γ which is different from Γ itself can contain an operation of this form. If then it is shewn that any self-conjugate sub-group of Γ, distinct from D, necessarily contains operations of this form, it follows that D is a maximum self-conjugate sub-group of Γ.

We shall first deal with the case n = 2.

If p = 2, the orders of GΓ and D are 66 and 1. In this case, Γ is simply isomorphic with the symmetric group of three symbols, which has a self-conjugate sub-group of order 3. The successive factor-groups of the composition-series of G are therefore cyclical groups of orders 2 and 3.

If p = 3, the orders of GΓ and D are 4824 and 2. The factor-group Γ D has 12 for its order, and cannot therefore be a simple group. The reader will have no difficulty in verifying that, in this case, the successive factor-groups of G have orders 23, 22 and 2. We may therefore, in dealing with the case n = 2, assume that p is not less than 5.

Let us suppose now that Γ has a self-conjugate sub-group I that contains D; and let S or

(ax1 + bx2,ax1 + bx2)
be one of its operations, not contained in D.

If b is different from zero, Γ contains Σ, where Σ denotes

αax1 + αbx2,1 + α2a2 αb x1 αax2 ,
and therefore I contains Σ1SΣS, which is
α2x 1,(1 + α2)(b + aα2) bα2 x1 α2x 2 .

If b is zero, b is congruent with a1; therefore, in any case, I contains an operation S of the form

(cx1,dx1 + c1x 2).

Again, Γ contains the operation T, where T denotes

(x1,x1 + x2);
and I therefore contains ST1S1T, which is
(x1, (1 c2)x 1 + x2).

Hence unless 1 c2 0, I must coincide with Γ. Now, when p > 5, c can always be chosen so that this congruence is not satisfied. If p = 5, the square of the above operation Σ1SΣS, when unity is written for α, is

x1, 4 b(b + a)x1 + x2 ;
unless b + a 0, this again requires that I coincides with Γ. If finally, the condition b + a 0 is satisfied in S, it is not satisfied in ST1S1T, another operation belonging to I; and therefore again, in this case, I coincides with Γ.

Hence finally, if n = 2, the factor-group Γ D is simple, except when p is 2 or 3.

219. When n is greater than 2, it will be found that it is sufficient to deal in detail with the case n = 3, as the method will apply equally well for any greater value of n. Suppose here again that Γ has a self-conjugate sub-group I which contains D; and let S, denoting

(ax1 + bx2 + cx3,ax1 + bx2 + cx3,ax1 + bx2 + cx3),
be one of the operations of I which is not contained in D. S cannot be permutable with all operations of the form (x1,x2,x3 + x1), as it would then be permutable with every operation of Γ. We may therefore suppose without loss of generality that S and T are not permutable, T denoting (x1,x2,x3 + x1). Then T1STS1 is an operation, distinct from identity, belonging to I. Now a simple calculation shews that this operation, say U, is of the form
x1 cX,x2 cX,Ax1 + Bx2 + Cx3 ,
where X is the symbol with which S1 replaces x1.

If c and c are both different from zero, Γ will contain an operation V of the form

x1 c cx2,x2,x3 ;
and I contains V 1UV , which is of the form
(x1,αx1 + βx2 + γx3,αx1 + βx2 + γx3).

Moreover, if either c or c is zero, the operation U itself leaves one symbol unaltered. Hence I always contains operations by which one symbol is unaltered.

This process may now be repeated to shew that I necessarily contains operations of the form

(x1,x2,αx1 + βx2 + γx3);
and, since the determinant of the operation is unity, γ is necessarily congruent to unity. But it has been seen that the group Γ is generated from the last operation and the operations conjugate to it. Hence finally, if n is greater than 2, the factor-group  Γ D is simple for all values of p.

220. The composition-series of G is now, except as regards the constitution of the simple group  Γ D, perfectly definite. It has, in fact, been seen that G Γ  and D are cyclical groups of orders p 1 and d; and therefore if αβ, γ, … are distinct primes whose product is p 1, and if αβ, γ, … are distinct primes whose product is d; the successive factor-groups of G are first, a series of simple groups of prime orders αβ, γ, …: then a simple group of composite order  N (p 1)d: and lastly, a series of simple groups of prime orders αβ, γ, ….

The sequence in which the set of simple groups of orders αβ, γ, … are taken in the composition-series may be clearly any whatever, and the same is true of the set of factor-groups of orders αβ, γ, …; but it is to be noticed that, when d is not equal to p 1, the composition-series is capable of further modifications. In this case, {A, Γ} is a self-conjugate sub-group of G of order N d , which has a maximum self-conjugate sub-group {A} of order p 1. The successive composition-factors of G may therefore be taken in the sequence

α,β,γ, : N (p 1)d : α,β,γ, :
and their arrangement may be yet further changed by considering the self-conjugate sub-group {Am, Γ}, where m is a factor of p 1 less than p 1 d .

221. For every value of pn, except 22 and 32, it thus appears that the linear group may be regarded as defining a simple group of composite order. We shall now proceed to a discussion of the constitution of the simple groups thus defined when n = 2, p being greater than 390. In this case, the group Γ is defined by the congruences

y1 αx1 + βx2, y2 γx1 + δx2, αδ βγ 1, (modp);
and since p 1 is divisible by 2 when p is an odd prime, d is equal to 2. Hence the self-conjugate operations of Γ are
(x1,x2) and(x1,x2).

The order of Γ is p(p2 1), and therefore the order of the simple group, H, which it defines is 1 2p(p2 1). Suppose now, if possible, that Γ contains a sub-group g simply isomorphic with H. If S is any operation of Γ, not contained in g, the whole of the operations of Γ are contained in the two sets

g,Sg.

Now (x2,x1), whose square (x1,x2) is a self-conjugate operation, cannot be contained in the simple group g. Hence both (x2,x1) and (x1,x2) are contained in Sg, an obvious contradiction. Therefore Γ contains no sub-group simply isomorphic with H.

For a discussion of the properties of H, some concrete representation of the group itself is necessary; this may be obtained in the following way. Instead of the pair of homogeneous congruences that define each operation of Γ, let us, as in § 113, consider the single non-homogeneous congruence

y αx + β γx + δ ,(modp),
where
αδ βγ 1.
Corresponding to every operation
(αx1 + βx2,γx1 + δx2)
of Γ, there will be a single operation of this new set; namely that in which αβ, γδ have respectively the same values. But since the operations
y αx + β γx + δ  andy αx β γx δ
are identical, two operations
(αx1 + βx2,γx1 + δx2) and(αx1 βx2, γx1 δx2)
of Γ will correspond to each operation
y αx + β γx + δ
of the new set; the two self-conjugate operations
(x1,x2) and(x1,x2),
in particular, corresponding to the identical operation of the new set. Moreover, direct calculation immediately verifies that, to the product of any two operations of Γ, corresponds the product of the two corresponding operations of the new set. Hence the new set of operations forms a group of order 1 2p(p2 1), with which Γ is multiply isomorphic; the group of order 2 formed by the self-conjugate operations of Γ corresponding to the identical operation of the new group.

The simple group H, of order 1 2p(p2 1), which we propose to discuss, can therefore be represented by the set of operations

y αx + β γx + δ ,(modp);
where
αδ βγ 1,
αβ, γδ being integers reduced to modulus p.

222. Since the order of H is divisible by p and not by p2, the group must contain a single conjugate set of sub-groups of order p. Now the operation

y x + 1,
or (x + 1) as we will write it in future, is clearly an operation of order p: for its nth power is (x + n), and p is the smallest value of n for which this is the identical operation. If (x + 1) and αx + β γx + δ are represented by P and S, then
S1PS = (1 αγ)x + α2 γ2x + 1 + αγ.

This is identical with P, only if

γ 0,α2 1;
and therefore P is permutable with no operations except its own powers. On the other hand, if
γ 0,
then
S1PS = Pa2 ;
and therefore every operation, for which γ 0, transforms the sub-group {P} into itself. These operations therefore form a sub-group: a result that may also be easily verified directly. The order of this sub-group is the number of distinct operations αx + β δ for which αδ 1. The ratio α δ must be a quadratic residue, while β may have any value whatever. Hence the order of the sub-group is 1 2(p 1)p; and H therefore contains p + 1 sub-groups of order p. Since H is a simple group, it follows (§ 125) that it can be represented as a transitive substitution group of degree p + 1.

This representation of the group can be directly derived, as in § 113, from the congruences already used to define it. Thus if, in

y αx + β γx + δ ,
we write for x successively 0, 1, 2, …, p 1, the p + 1 values obtained for y, when reduced (modp), will be the same p + 1 symbols in some other sequence. For if
αx1 + β γx1 + δ αx2 + β γx2 + δ ,
then
(αδ βγ)(x1 x2) 0,
and therefore x1 x2.

Each operation of H gives therefore a distinct substitution performed on the symbols 0, 1, …, p 1; and the complete set of substitutions thus obtained gives the representation of H as a transitive substitution group of degree p + 1. Since H contains operations of order p, this substitution group must be doubly transitive. That this is the case may also be shewn directly. Thus

y a y b mx a x b
is an operation changing a into a and b into b. This operation may be written
y k(bm a)x + k(ab mab) k(m 1)x + k(b ma) ,
and its determinant is
k2m(b a)(b a).

If now (b a)(b a) is a quadratic residue (or non-residue) (modp), m may be any quadratic residue (or non-residue); and k can always be chosen so that the determinant is unity. There are therefore 1 2(p 1) substitutions in the group, changing any two symbols ab into any other two given symbols ab. Further, if the operation αx + β γx + δ keeps x unchanged in the substitution group, x must satisfy the congruence

x αx + β γx + δ ,
that is
γx2 + (δ a)x β 0.
Such a congruence cannot have more than two roots; and therefore every substitution displaces all, all but one, or all but two, of the p + 1 symbols.

223. The substitutions, which keep either one or two symbols fixed, must therefore be regular in the remaining p or p 1 symbols. Hence the order of every substitution which keeps just one symbol fixed must be p; and the order of every substitution that keeps two symbols fixed must be equal to or be a factor of p 1. Now it was seen in the last paragraph that the order of the sub-group that keeps two symbols fixed is 1 2(p 1). Moreover, if z is a primitive root (modp), the sub-group that keeps a and b fixed contains the operation

y a y b = z2x a x b,
and the order of this operation is 1 2(p 1). Hence, the sub-group that keeps any two symbols fixed is a cyclical group of order 1 2(p 1); and every operation that keeps two symbols fixed is some power of an operation of order 1 2(p 1). Since the group is a doubly transitive group of degree p + 1, there must be 1 2(p + 1)p sub-groups which keep two symbols fixed; and these must form a conjugate set. Each is therefore self-conjugate in a sub-group of order p 1. To determine the type of this sub-group, we may consider the sub-group keeping 0 and  fixed: this is generated by Q, where Q denotes zx z1. If αx + β γx + δ is represented by S, then
S1QS = (zαδ z1βγ)x (z z1)αβ (z z1)γδx + z1αδ zβγ ,
which can be a power of Q only if
αβ 0,γδ 0.

Hence either

β γ 0,
in which case S is a power of Q: or
α δ 0,γ β1.

In the latter case, we have

S = β β1x,
which is an operation of order 2; and then
S1QS = z1x z = Q1.

The group of order p 1, which contains self-conjugately a cyclical sub-group of order 1 2(p 1) that keeps two symbols fixed, is therefore a group of dihedral (§ 202) type. Moreover, if t is any factor of p 1, this investigation shews that {S,Q} is the greatest sub-group that contains {Qt} self-conjugately.

224. A substitution that changes all the symbols must either be regular in the p + 1 symbols, or must be such that one of its powers keeps two symbols fixed. The latter case however cannot occur; for we have just seen that, if Q is an operation, of order 1 2(p 1), which keeps two symbols fixed, the only operations permutable with Qr are the powers of Q. Hence the substitutions that change all the symbols must be regular in the p + 1 symbols, and their orders must be equal to or be factors of p + 1.

Suppose now that i is a primitive root of the congruence

ip21 1 0(modp),
so that i and ip are the roots of a quadratic congruence with real coefficients; and consider the operation K, denoting
y ik y ikp i2(p1) x ik x ikp,
where k is not a multiple of p + 1. On solving with respect to y, K is expressed in the form
y ik 2+1(p1) ik 2+1(p1) ik 2(p1) ik 2(p1) x i(p1) i(p1) ik 2(p1) ik 2(p1)ik 2(p+1) i(p1) i(p1) ik 2(p1) ik 2(p1)xik 2(p+1) + ik 21(p1) ik 21(p1) ik 2(p1) ik 2(p1) ,
an operation of determinant unity. It will be found, on writing ip for i in the coefficients of this operation, that they remain unaltered; therefore, since they are symmetric functions of i and ip, they must be real numbers. The operation therefore belongs to H. The nth power of this operation is given by
y ik y ikp i2n(p1) x ik x ikp,
and therefore, since the first power of i which is congruent to unity, (modp), is the (p2 1)th, the order of the operation is 1 2(p + 1). If we write kp for k in the operation K, the new operation is K1; but if k is replaced by any other number k, which is not a multiple of p + 1, the new operation K, given by
y ik y ikp i2(p1) x ik x ikp,
generates a new sub-group of order 1 2(p + 1), which has no operation except identity in common with {K}. Now there are p2 p numbers less than p2 1 which are not multiples of p + 1; therefore H contains 1 2(p2 p) cyclical sub-groups of order 1 2(p + 1), no two of which have a common operation except identity. The corresponding substitutions displace all the symbols.

225. A simple enumeration shews that the operations of the cyclical sub-groups of orders 1 2(p 1)p and 1 2(p + 1), exhaust all the operations of the group. Thus there are, omitting identity from each sub-group:

(i) 1 2p(p + 1) sub-groups of order 1 2(p 1), containing 1 4p(p2 1) 1 2p2 1 2p distinct operations;

(ii) 1 2p(p 1) sub-groups of order 1 2(p + 1), containing 1 4p(p2 1) 1 2p2 + 1 2p distinct operations;

(iii) p + 1 sub-groups of order p, containing p2 1 distinct operations;
and the sum of these numbers, with 1 for the identical operation, gives 1 2p(p2 1), which is the order of the group.

Every operation that displaces all the symbols is therefore the power of an operation of order 1 2(p + 1).

226. We shall now further shew that the 1 2p(p 1) sub-groups of order 1 2(p + 1) form a single conjugate set, and that each is contained self-conjugately in a dihedral group of order p + 1. Let S be any operation of H, which is permutable with {K} and replaces ik by some other symbol j. Then S1KS is an operation which leaves j unaltered; it may therefore be expressed in the form

y j y j = m x j x j.

This can belong to the sub-group generated by K, only if j and j are the same pair as ik and ikp. Hence j must be either ik or ikp; and similarly, if S replaces ikp by j, the latter must be either ikp or ik. Hence either S must keep both the symbols ik and ikp unchanged or it must interchange them; and conversely, every operation which either keeps both the symbols unchanged or interchanges them, must transform {K} into itself. If S keeps both of them unchanged, it is a power of K. If S interchanges them, it is of the form

y ik y ikp mx ikp x ik ;
and a simple calculation shews that
S1KS = K1.
If we take m = 1, S becomes
x + y ik + ikp,
an operation belonging to H. Hence the cyclical sub-group {K} is contained self-conjugately in the sub-group of {S,K} which is of dihedral type. If there were any other operation S, not contained in {S,K}, which transformed K into its inverse, then SS would be an operation permutable with K and not contained in {K}. It has just been seen that no such operation exists. Hence {S,K}, of order p + 1, is the greatest sub-group that contains {K} self-conjugately; and {K} must be one of 1 2p(p 1) conjugate sub-groups.

227. The distribution of the operations of H in conjugate sets is now known. A sub-group of order p is contained self-conjugately in a group of order 1 2p(p 1), while an operation of order p is permutable only with its own powers. There are therefore two conjugate sets of operations of order p, each set containing 1 2(p2 1) operations. Again, each of the operations of a cyclical sub-group of order 1 2(p 1) or 1 2(p + 1) is conjugate to its own inverse and to no other of its powers. Hence if 1 2(p + 1) is even and therefore 1 2(p 1) odd, there are 1 4(p 3) conjugate sets of operations whose orders are factors of 1 2(p 1), each set containing p2 + p operations; 1 4(p 3) conjugate sets of operations whose orders are factors of 1 2(p + 1), other than the factor 2, each set containing p2 p operations; and a single set of operations of order 2, containing 1 2(p2 p) operations. If 1 2(p 1) is even and 1 2(p + 1) odd, there are 1 4(p 1) conjugate sets of operations whose orders are factors of 1 2(p + 1), each containing p2 p operations; 1 4(p 5) conjugate sets whose orders are factors of 1 2(p 1), other than the factor 2, each set containing p2 + p operations; and a single set of 1 2(p2 + p) conjugate operations of order 2. In either case, the group contains, exclusive of identity, 1 2(p + 3) conjugate sets of operations.

228. Since p 1 and p + 1 can have no common factor except 2, it follows that, if qm denote the highest power of an odd prime, other than p, which divides the order of H, qm must be a factor of 1 2(p 1) or of 1 2(p + 1); and the sub-groups of order qm must be cyclical. Moreover, since no two cyclical sub-groups of order 1 2(p 1), or 1 2(p + 1), have a common operation except identity, the same must be true of the sub-groups of order qm.

If 2m is the highest power of 2 that divides 1 2(p 1) or 1 2(p + 1), 2m+1 will be the highest power of 2 that divides the order of H. Moreover, a sub-group of order 2m+1 must contain a cyclical sub-group of order 2m self-conjugately, and it must contain an operation of order 2 that transforms every operation of this cyclical sub-group into its own inverse; in other words, the sub-groups of order 2m+1 are of dihedral type.

Suppose now that two sub-groups of order 2m+1 (m > 1) have a common sub-group of order 2r (r > 2). Such a sub-group must be either cyclical or dihedral: in the latter case, it contains self-conjugately a single cyclical sub-group of order 2r1. Hence, on the supposition made, a cyclical sub-group of order 4 at least would be contained self-conjugately in two distinct cyclical sub-groups of order 2m. It has been seen that this is not the case; and therefore the greatest sub-group, that two sub-groups of order 2m+1 can have in common, must be a sub-group of order 4, whose operations, except identity, are all of order 2. Now every group of order 2m+1 contains one self-conjugate operation of order 2, and 2m operations of order 2 falling into 2 conjugate sets of 2m1 each. Moreover, the group of order p ± 1, which has a cyclical sub-group of order 2m and contains the operation A of order 2 of this cyclical group self-conjugately, has 1 2(p ± 1) other operations of order 2; and therefore it contains p ± 1 2m+1 sub-groups of order 2m+1, each of which has A for its self-conjugate operation. If now B is any operation of order 2 of this sub-group of order p ± 1, and if it is distinct from A, then B enters into a sub-group of order 2m+1 that contains A self-conjugately. But since A is permutable with B, A must belong to the sub-group of order p ± 1, which contains B self-conjugately; hence A enters into a sub-group of order 2m+1 which contains B self-conjugately. The sub-group {A,B} is therefore common to two distinct sub-groups of order 2m+1. Now no group of order 2r (r > 2) can be common to two sub-groups of order 2m+1; and therefore {A,B} must (§ 80) be permutable with some operation S whose order s is prime to 2. If s is not 3, S must be permutable with A and B: and then {A,S} and {B,S} would be two distinct sub-groups of orders 2s, whose operations are permutable with each other. It has been seen that H does not contain such sub-groups. Hence s = 3; and S transforms AB and AB cyclically, or {S,A,B} is a sub-group of tetrahedral type (§ 202).

The number of quadratic91 sub-groups contained in H may be directly enumerated. A group of order 2m+1 contains 2m1 such sub-groups, which fall into 2 conjugate sets of 2m2 each; a single group of order 8 containing each quadratic group self-conjugately. The quadratic groups, contained in the p ± 1 2m+1 sub-groups of order 2m+1 of a sub-group of order p ± 1, are clearly all distinct, and each quadratic group belongs to just 3 groups of order p ± 1; thus {A,B} belongs to the 3 groups which contain AB and AB respectively as self-conjugate operations. Hence the total number of quadratic groups contained in H is

1 2p(p 1)p ± 1 2m+12m11 3 = 1 2p(p2 1) 12 .

229. The greatest sub-group of a group of order 2m+1, that contains a quadratic group self-conjugately, is a group of order 8 and dihedral type; and it has been shewn that 3 is the only factor, prime to 2, that occurs in the order of the sub-group containing a quadratic group self-conjugately. Hence finally, the order of the greatest group containing a quadratic group self-conjugately is 24, and the 1 2p(p2 1) 12 quadratic groups fall into two conjugate sets of 1 2p(p2 1) 24 each. The group of order 24, that contains a quadratic group self-conjugately, contains also a self-conjugate tetrahedral sub-group, while the sub-groups of order 8 are dihedral. Hence (§ 84) this group must be of octohedral type.

Since every tetrahedral sub-group of H contains a quadratic sub-group self-conjugately, and every octohedral sub-group contains a tetrahedral sub-group self-conjugately, there must also be two conjugate sets of tetrahedral sub-groups and two conjugate sets of octohedral sub-groups, the number in each set being 1 2p(p2 1) 24 .

230. We have hitherto supposed m > 1, or what is the same thing, p ±1(mod 8). If now m = 1, so that p ±3(mod 8), the highest power of 2 that divides the order of H is 22; and, since 22 is not a factor of 1 2(p ± 1), the sub-groups of order 22 are quadratic. Moreover, since 22 is the highest power of 2 dividing the order of H, the quadratic sub-groups form a single conjugate set. Each sub-group of order p ± 1, which has a self-conjugate operation of order 2, contains 1 4(p ± 1) sub-groups of order 4, and each of the latter belongs to 3 of the former. The total number is as before

1 2p(p2 1) 12 ,
and since they form a single conjugate set, each quadratic group is self-conjugate in a group of order 12. Also, for the same reason as in the previous case, this sub-group is of tetrahedral type.

Finally, since every sub-group of H of tetrahedral type must contain a quadratic sub-group self-conjugately, H must contain a single conjugate set of 1 2p(p2 1) 12 tetrahedral sub-groups. In this case, the order of H is not divisible by 24, and therefore the question of octohedral sub-groups does not arise.

231. The group H always contains tetrahedral sub-groups; when its order is divisible by 24, it contains also octohedral sub-groups. Now if p ±1(mod 5), the order of H is divisible by 60: it may be shewn as follows that, in these cases, H contains sub-groups of icosahedral type.

Let us suppose, first, that p 1(mod 5); and let j be a primitive root of the congruence

j5 1(modp).
Then jx j1, which we will denote by A, is an operation of order 5. The operations of order 2 of H are all of the form B, where B denotes αx + β γx α, since each is its own inverse. Now
AB = αjx + βj1 γjx αj1 ;
and (§ 203) if A and B generate an icosahedral group,
(AB)3 = 1.
A simple calculation shews that, if this condition is satisfied, then
α2(j j1)2 1.
Also, since the determinant of B is unity,
α2 + βγ 1.
These two congruences have just p 1 distinct solutions, the solutions αβγ and α β γ being regarded as identical. There are therefore p 1 operations of order two in H, namely the operations
1 j j1x + β γx 1 j j1 ,
where
βγ 1 1 (j j1)2,
which with A generate an icosahedral sub-group.

The group generated by

jx j1  and 1 j j1x + β0 γ0x 1 j j1 ,
contains 5 of the p 1 operations of order 2 of the form
1 j j1x + β γx 1 j j1 ,
viz. those for which
β β0jn,γ γ 0jn,(n = 0, 1, 2, 3, 4).
Hence the sub-group {A}, of order 5, belongs to 1 5(p 1) distinct icosahedral sub-groups. Now each icosahedral sub-group has 6 sub-groups of order 5; and H contains 1 2p(p + 1) sub-groups of order 5 forming a single conjugate set. The number of icosahedral sub-groups in H is therefore
1 6 p 1 5 1 2p(p + 1) = 1 2p(p2 1) 30 .
The group of isomorphisms of the icosahedral group is the symmetric group of degree 5 (§ 173). Now H can contain no sub-group simply isomorphic with the symmetric group of degree 5. For if it contained such a sub-group, an operation of order 5 would be conjugate to its own square; and this is not the case.

Hence (§ 165), if an icosahedral sub-group K of H is contained self-conjugately in a greater sub-group L, then L must be the direct product of K and some other sub-group. This also is impossible; for the greatest sub-group of H in which any cyclical sub-group, except those of order p, is contained self-conjugately, is of dihedral type. Hence L must coincide with K, and K must be one of 1 2p(p2 1) 60 conjugate sub-groups. The icosahedral sub-groups of H therefore fall into two conjugate sets of 1 2p(p2 1) 60 each.

In a similar manner, when p 1(mod 5), we may take as a typical operation A, of order 5,

y i y ip ip21 5 x i x ip;
and it may be shewn, the calculation being rather more cumbrous than in the previous case, that there are just p + 1 operations B, of the form αx + β γx α, such that
(AB)3 = 1,
and that five of these belong to the icosahedral group generated by A and any one of them. It follows, exactly as in the previous case, that H contains 1 2p(p2 1) 30 icosahedral sub-groups, which fall into two conjugate sets, each set containing 1 2p(p2 1) 60 groups.

232. Finally, we proceed to shew that H has no other sub-groups than those which have been already determined. Suppose, first, that a sub-group h of H contains two distinct sub-groups of order p. These must, by Sylow’s theorem, form part of a set of kp + 1 sub-groups of order p conjugate within h. Now H contains only p + 1 sub-groups of order p, and therefore k must be unity and h must contain all the sub-groups of order p; or since H is simple, h must contain and therefore coincide with H. Hence the only sub-groups of H, whose orders are divisible by p, are those that contain a sub-group of order p self-conjugately. They are of known types.

Suppose next that g is a sub-group of H, whose order n is not divisible by p, and let S1 be an operation of g whose order q1 is not less than the order of any other operation of g. In H the sub-group {S1} is self-conjugate in a dihedral group of order p ± 1; and the greatest sub-group of this group, which contains no operation of order greater than q1, is a dihedral group of order 2q1. Hence in g the sub-group {S1} is self-conjugate in a group of order q1 or 2q1, and therefore it forms one of  n q1 or of  n 2q1 conjugate sub-groups. Moreover, no two of these sub-groups contain a common operation except identity; and they therefore contain, excluding identity, n(q1 1) 𝜖1q1 distinct operations, where 𝜖1 is either 1 or 2.

Of the remaining operations of g, let S2 be one whose order q2 is not less than that of any of the others. The operation S2 cannot be permutable with any of the n(q1 1) 𝜖1q1 operations already accounted for, since S2 is not a power of any one of these operations. Hence, exactly as before, {S2} must form one of  n 𝜖2q2 conjugate sub-groups in g, 𝜖2 being either 1 or 2; and these sub-groups contain n(q2 1) 𝜖2q2 operations which are distinct from identity, from each other, and from those of the previous set. This process may be continued till the identical operation only remains. Hence, finally, n being the total number of operations of g, we must have

n = 1 + νn(qν 1) 𝜖νqν
or
1 n = 1 νqν 1 𝜖νqν .

233. In this equation, let r of the 𝜖’s be 1 and s of them be 2, so that r + s is their total number, say m. Then

1 n = 1 1 1 qλ 1 2 1 2qμ = 1 r + 1 qλ 1 2s + 1 2qμ 1 r + 1 2r 1 2s + 1 4s 1 1 2r 1 4s 1 1 4m.

Hence, since n is a positive integer, there cannot be more than three terms under the sign of summation. Moreover, since

1 n 1 1 2r 1 4s,
r cannot be greater than 1, and therefore not more than one of the 𝜖’s can be unity. Also, when one of the 𝜖’s is unity, we have

1 n = 1 q 1 2s + 1 2qμ 1 2 1 2s + 1 4s 1 4(2 s),

so that, in this case, s cannot be greater than unity. The solutions are now easily obtained by trial.

(i) For one term in the sum, the only possible solution is

𝜖1 = 1,n = q1,
and the corresponding group is cyclical.

(ii) For two terms in the sum, the solutions are

 (α)𝜖1 = 𝜖2 = 2,n = 2q1q2 q1 + q2;  (β)𝜖1 = 2,𝜖2 = 1,q2 = 2,n = 2q1;  (γ)𝜖1 = 1,𝜖2 = 2,q1 = 3,q2 = 2,n = 12.

To the solution (α) there corresponds no sub-group; for n < 2q1, and the values q1 = q1, 𝜖1 = 2 imply that g has a sub-group of order 2q1.

To the solution (β) correspond the sub-groups of order 2q1 of dihedral type, for which q1 is odd, so that the operations of order 2 form a single conjugate set.

To the solution (γ) corresponds a sub-group of order 12 containing 8 operations of order 3 and 3 operations of order 2, i.e. a tetrahedral sub-group.

(iii) For three terms in the sum, the solutions are

 (α) 𝜖1 = 𝜖2 = 𝜖3 = 2, q2 = 2, q3 = 2, n = 2q1;  (β) q1 = 3, q2 = 3, q3 = 2, n = 12;  (γ) q1 = 4, q2 = 3, q3 = 2, n = 24;  (δ) q1 = 5, q2 = 3, q3 = 2, n = 60.

To the solution (α) correspond the sub-groups of order 2q1 of dihedral type, in which q1 is even, so that the operations of order 2, which do not belong to the cyclical sub-group of order q1, fall into two distinct conjugate sets.

To the solution (β) would correspond a group of order 12 containing 3 operations of order 2 and 4 sub-groups of order 3 which fall into two conjugate sets of 2 each. Sylow’s theorem shews that such a group cannot exist; and therefore there is no sub-group of H corresponding to this solution.

Solution (γ) gives a group of order 24, with 3 conjugate cyclical sub-groups of order 4, 4 conjugate cyclical sub-groups of order 3, and 6 other operations of order 2 forming a single conjugate set. No operation of this group is permutable with each of the 4 sub-groups of order 3; and therefore, if the group exists, it can be represented as a transitive group of 4 symbols. On the other hand, the order of the symmetric group of 4 symbols, which (§ 203) is simply isomorphic with the octohedral group, is 24; and its cyclical sub-groups are distributed as above. Hence to this solution there correspond the octohedral sub-groups of H.

Solution (δ) gives a group of order 60, with 6 conjugate sub-groups of order 5, 10 conjugate sub-groups of order 3, and a conjugate set of 15 operations of order 2. It has been shewn, in § 85, that there is only one type of group of order 60 that has 6 sub-groups of order 5; viz. the alternating group of degree 5: and that, in this group, the distribution of sub-groups in conjugate sets agrees with that just given. Moreover, the alternating group of degree 5 is simply isomorphic with the icosahedral group. Hence to this solution there correspond the icosahedral sub-groups of H.

234. When p > 11, then 1 2p(p 1) > 60; and, when p > 3, 1 2p(p 1) > p + 1. Hence when p > 11, the order of the greatest sub-group of H is 1 2p(p 1), and the least number of symbols in which H can be expressed as a transitive substitution group is p + 1.

When p is 57 or 11, however, H can be expressed as a transitive substitution group of p symbols92.

For, when p = 5, H contains a tetrahedral sub-group of order 12, forming one of 5 conjugate sub-groups; therefore H can be expressed as a transitive group of 5 symbols. It is to be noticed that in this case H is an icosahedral group.

When p = 7, H contains an octohedral sub-group of order 24, which is one of 7 conjugate sub-groups; and H can therefore be expressed as a transitive group of 7 symbols. Similarly, when p = 11, H contains an icosahedral sub-group of order 60, which is one of 11 conjugate sub-groups; and the group can be expressed transitively in 11 symbols.

235. The simple groups, of the class we have been discussing in the foregoing sections, are self-conjugate sub-groups of the triply transitive groups of degree p + 1, defined by

y αx + β γx + δ ,(modp),
the existence of which was demonstrated in § 113. In fact, since αx + β γx + δ and kαx + kβ kγx + kδ represent the same transformation, the determinant, αδ βγ, of any transformation may always be taken as either unity or a given non-residue; and it follows at once that the transformations of determinant unity form a self-conjugate sub-group of the whole group of transformations.

If, as in § 113, αβ, γδ, are powers of i, where i is a primitive root of the congruence

ipn1 1,(modp),
the triply transitive group G of degree pn + 1, which is defined by the transformations, has again, when p is an odd prime, a self-conjugate sub-group H of order 1 2pn(p2n 1), which is given by the transformations of determinant unity. It follows from Theorem IX, § 134, that G, being a triply transitive group of degree pn + 1, must have, as a self-conjugate sub-group, a doubly transitive simple group; and it is easy to shew that H is this sub-group.

In fact, if a simple group h is a self-conjugate sub-group of G it must be contained in H. Also, since h is a doubly transitive group of degree pn + 1, it must contain every operation of order p that occurs in G. Now we may shew that these operations generate H. Thus 2x + 1 x and (x + 2 i i1) are operations of order p belonging to G. Therefore (i + i1)x + 1 x belongs to h. But this operation is transformed into ix i1 by x + i1 x + i . Hence ix i1 belongs to h; and a sub-group of h which keeps one symbol unchanged is the group of order 1 2pn(pn 1) generated by (x + 1) and  ix i1. The order of h therefore is not less than 1 2pn(p2n 1); in other words h is identical with H.

When p = 2, every power of i is a quadratic residue, and the determinant of every transformation is unity. In this case it may be shewn, by an argument similar to the above, that the group G of order 2n(22n 1) is itself a simple group.

We are thus led to recognize the existence of a doubly-infinite series of simple groups of orders 2n(22n 1) and 1 2pn(p2n 1), which are closely analogous to the groups of order 1 2p(p2 1) already discussed. For an independent proof of the existence of these simple groups and for an investigation of their properties, the reader is referred to the memoirs mentioned below93.

236. We will now return to the linear homogeneous group G of transformations of n symbols, taken to a prime modulus p; and consider it more directly as the group of isomorphisms of an Abelian group of order pn and type (1, 1,, to n units). As in § 156, it may be expressed in the form of a substitution group performed on the pn 1 symbols of the operations, other than identity, of the Abelian group. In this form it is clearly transitive, since there are isomorphisms changing any operation of the Abelian group into any other operation. If P is any operation of the Abelian group, an isomorphism which changes any one of the p 1 operations

P,P2,,Pp1,
into any other, will certainly interchange the set among themselves. Hence, when expressed as a group of degree pn 1, G is imprimitive; and the symbols forming an imprimitive system are those of the operations, other than identity, of any sub-group of order p of the Abelian group. If
P1,P2,,Pn
are a set of generating operations of the Abelian group, an isomorphism, which changes each of the sub-groups
{P1},{P2},,{Pn}
into itself, must be of the form
P1 , P2 ,, Pn P1α1,P2α2,,Pnαn.
This isomorphism changes P1P2 into (P1P2α2 α1 )α1; therefore it will only transform the sub-group {P1P2} into itself when α1 α2,(modp). If then the given isomorphism changes every sub-group of order p into itself, we must have
α1 α2 αn,(modp).

Hence the only operations of G, which interchange the symbols of each imprimitive system among themselves, are those given by the powers of

P1, P2,,Pn P1α,P2α,,Pnα,
where α is a primitive root of p. This operation is the same as that denoted by A in § 217. It follows immediately that the factor-group G {A} can be represented as a transitive group in pn 1 p 1 symbols. In fact, the operations of {A} are the only operations of G which transform each of the pn 1 p 1 sub-groups of order p in itself; and these pn 1 p 1 sub-groups must be permuted among themselves by every operation of G. The substitution group thus obtained is doubly transitive; for if P and P are any two operations of the Abelian group such that P is not a power of P, and if Q and Q are any other two operations of the Abelian group subject to the same condition, there certainly exists an isomorphism of the form
P,P, Q,Q,,
and this isomorphism changes the sub-groups {P} and {P} into the sub-groups {Q} and {Q}.

These results will still hold if, instead of considering G the total group of isomorphisms, we take Γ the group of isomorphisms of determinant unity. Thus the determinant of

P ,P, Qα,Q,
is α times the determinant of
P,P, Q,Q,.
It is therefore possible always to choose α so that the determinant of
P ,P, Qα,Q,
shall be unity; and this isomorphism still changes the sub-groups {P} and {P} into {Q} and {Q} respectively.

The lowest power of A contained in Γ is (§ 217) Ap1 d . Hence the group Γ {Ap1 d } can be represented as a doubly transitive group of degree pn 1 p 1 . This group is (§ 220) simply isomorphic with the simple group of order N (p 1)d, which is defined by the composition-series of G.

We may sum up these results as follows:—

THEOREM. The homogeneous linear group of order

N = (pn 1)(pn p)(pn pn1)
when pn is neither 22 nor 32, defines, by its composition-series, a simple group of order  N (p 1)d, where d is the greatest common factor of p 1 and n. This simple group can be represented as a doubly transitive group of degree pn1 + pn2 + + p + 1.

237. The pn 1 p 1 symbols, permuted by one of these doubly transitive simple groups, may be regarded as the sub-groups of order p of an Abelian group of order pn and type (1, 1,, to n units). Now every pair of sub-groups of such an Abelian group enters in one, and only in one, sub-group of order p2; and every sub-group of order p2 contains p + 1 sub-groups of order p. Hence from the pn 1 p 1 symbols permuted by the doubly transitive group, pn 1 pn1 1 p 1 p2 1 sets of p + 1 symbols each may be formed, such that every pair of symbols occurs in one set and no pair in more than one set, while the sets are permuted transitively by the operations of the group. These groups therefore belong to the class of groups referred to in § 148. The sub-group, that leaves two symbols unchanged, permutes the remaining symbols in two transitive systems of p 1 and pn1 + pn2 + + p2; and the sub-group, that leaves unchanged each of the symbols of one of the sets of p + 1, is contained self-conjugately in a sub-group whose order is (p + 1)p times that of a sub-group leaving two symbols unchanged. This latter sub-group permutes the symbols in two transitive systems of p + 1 and pn1 + pn2 + + p2. It may be pointed out that, when n is 3, such a sub-group is simply isomorphic with, but is not conjugate to, the sub-groups that leave one symbol unchanged: this may be seen at once by noticing that an Abelian group, of order p3 and type (1, 1, 1), has the same number of sub-groups of orders p and p2.

238. Some special cases may be noticed. First, when p = 2, both p 1 and d are unity, and the homogeneous linear group is itself a simple group.

If n = 3, then N = 168; so that the group of isomorphisms of a group of order 8, whose operations are all of order 2, is the simple group of order 168 (§ 146).

If n = 4, then N = 26 32 5 7. This is the order of the alternating group of 8 symbols; and it may be shewn that this group is simply isomorphic with the group of isomorphisms.

The Abelian group of order 16 contains 35 sub-groups of order 4; and it may be shewn that, from these 35 sub-groups, sets of 5 can be formed in 56 distinct ways, so that each set of 5 contains every operation of order 2 of the Abelian group once, and once only. If

P1,P2,P3,P4
are a set of generating operations of the Abelian group, such a set of 5 sub-groups of order 4 is given by
{P1,P2},{P3,P4},{P1P3,P2P4},{P1P4,P1P2P3},{P2P3,P1P3P4}.
Now
P1, P2 ,P3, P4 P1,P2P3,P4,P2P4
is an isomorphism of order 7 of the Abelian group, and P1 is the only operation of the group, except identity, which is left unchanged by this isomorphism. It may be directly verified that the 7 sets of 5 groups of order 4, into which the given set is transformed by the powers of this isomorphism, contain every sub-group of order 4 of the Abelian group. The 7 sets, being interchanged among themselves by this isomorphism of order 7 which leaves only P1 unchanged, must be interchanged among themselves by isomorphisms of order 7 which leave any other single operation of the Abelian group unchanged. There are therefore at least 15 isomorphisms of order 7 which interchange the 7 sets among themselves. Now the isomorphisms, which interchange the 7 sets among themselves, form a sub-group of the group of isomorphisms, which is isomorphic with a group of degree 7; and the only groups of degree 7, which contain at least 15 operations of order 7, are the symmetric and the alternating groups. The group of isomorphisms must therefore contain a sub-group which is isomorphic with the symmetric or with the alternating group of degree 7. Hence at once, since the group of isomorphisms is simple, it must contain a sub-group which is simply isomorphic with the alternating group of degree 7. Since this must be one of 8 conjugate sub-groups, the group of substitutions itself is simply isomorphic with the alternating group of degree 8.

If pn = 33, then pn1 + + p + 1 = 13, d = 1, and N = 24 33 13. There is therefore a doubly transitive simple group of degree 13 and order 24 33 13 (§§ 145, 149).

239. The homogeneous linear group may be generalized by taking for the coefficients powers of a primitive root of

ipν1 1,(modp),
instead of powers of a primitive root of
ip1 1,(modp).
When the coefficients are thus chosen, the order of the group Gp,n,ν, defined by all sets of transformations
x11 a 11x1 + a12x2 + + a1nxn, x21 a 21x1 + a22x2 + + a2nxn, ................................. xn1 a n1x1 + an2x2 + + annxn, (modp),
whose determinant differs from zero, may be shewn, as in § 172, to be
N = (pnν 1)(pnν pν)(pnν p2ν)(pnν p(n1)ν);
and the order of the sub-group Γ, formed of those transformations whose determinant is unity, is  N pν 1. The only self-conjugate operations of Γ are the operations of the sub-group generated by (ix1,ix2,,ixn), which are contained in Γ. If δ is the greatest common factor of pν 1 and n, these self-conjugate operations of Γ form a cyclical sub-group γ of order δ. Finally, the argument of § 219 maybe repeated to shew that Γ γ is a simple group.

The homogeneous linear group Gp,n,ν, when values of ν greater than unity are admitted, thus defines a triply infinite system of simple groups; it may be proved that these groups can, for all values of ν, be expressed as doubly transitive groups of degree pnν 1 pν 1 .

240. We may shew, in conclusion, that the group Gp,n,ν is simply isomorphic with a sub-group of Gp,nν,1. For this purpose, we consider the group defined by

x11 x 1 + ir1,x 21 x 2 + ir2,,x n1 x n + irn, (r1,r2,,rn = 0, 1, 2,,pν 1);

the congruences being taken to modulus p. This is an Abelian group of order pnν and type (1, 1,, to nν units). Moreover, the operation

x11 a 11x1 + + a1nxn, ......................................................................... xn1 a n1x1 + + annxn,

of Gp,n,ν transforms the given operation of the Abelian group into

x11 x 1 + is1,x 21 x 2 + is2,,x n1 x n + isn;
where

is1 a 11ir1 + a 12ir2 + + a 1nirn, ......................................................................... isn a n1ir1 + a n2ir2 + + a nnirn.

Every operation of Gp,n,ν as defined in § 239, is therefore permutable with the Abelian group, and gives a distinct isomorphism of it; or in other words, as stated above, Gp,n,ν is simply isomorphic with a sub-group of Gp,nν,1.

Further, the sub-group

x11 x 1 + ir,x 21 x 2,,xn1 x n, (r = 0, 1, 2,,pν 1),

is transformed by the given operation of Gp,n,ν into the sub-group

x11 x 1 + a11ir,x 21 x 2 + a21ir,,x n1 x n + an1ir, (r = 0, 1, 2,,pν 1).

If

a21 a31 an1 0,
the two sub-groups are identical; but if these conditions are not satisfied, they have no operation in common except identity. Moreover,
a11,a21,,an1
may each have any value from 0 to ipν1 , simultaneous zero values alone excluded. Hence the sub-group of order pν defined by

x11 x 1 + ir,x 21 x 2,,xn1 x n, (r = 0, 1,,pν 1),

is one of pnν 1 pν 1 conjugate sub-groups in the group formed by combining the Abelian group with Gp,n,ν; and no two of these conjugate sub-groups have a common operation except identity.

The pnν 1 operations, other than identity, of an Abelian group of order pnν and type (1,1,, to nν units), can therefore be divided into pnν 1 pν 1 sets of pν 1 each, such that each set, with identity, forms a sub-group of order pν; and the group Gp,n,ν is isomorphic with a group of isomorphisms of the Abelian group which permutes among themselves such a set of pnν 1 pν 1 sub-groups of order pν.

Ex. 1. Shew that the pnν 1 pnν ppnν pν p 1 p2 1pν 1 sub-groups of order pν of an Abelian group of order pnν and type (1,1,, to nν units) can be divided into sets of pnν 1 pν 1 each, such that each set contains every operation of the group, other than identity, once and once only; and discuss in how many distinct ways such a division may be carried out.

Ex. 2. Shew that the simple group, defined by the group of isomorphisms of an Abelian group of order pn and type (1,1,,1), admits a class of contragredient isomorphisms, which change the operations of the simple group, that correspond to isomorphisms leaving a sub-group of order p of the Abelian group unaltered, into operations that correspond to isomorphisms leaving a sub-group of order pn1 of the Abelian group unaltered.