A group with only one generator x, is the cyclic group generated by x and can now be written as gp{π₯}. To illustrate these ideas, let us once more turn to the group G(β
π3 ) of order 6 which is exhibited in the table below
1
a
b
c
d
e
1
1
a
b
c
d
e
a
a
b
1
e
c
d
b
b
1
a d
e
c
c
c
d
e
1
a
b
d
d
e
c
b
1
a
e
e
c
d
a
b
1
table 1
we find that each of the six elements can be expressed in terms of a and c thus, 1 = π 2 = (π3 ), π = π, π = π2 , π = π, π = ππ, π = ππ2 Hence in this case we can write πΊ = ππ{π, π}
(2.24)
Alternatively, it can be shown that πΊ = ππ{π, π}
(2.2.5)
For a and c and hence the whole group, can be expressed in terms of b and d, namely π = π 2 , π = ππ The generators in (2.24) or in (2.25) are certainly irredundant, because the group is non-Abelian and cannot be generated by a single element, in which case it would be cyclic and hence non-Abelian. It is important to realize that irredundant generators may nevertheless be linked by non-trivial relations. Thus, by referring to the table 1, we see that ππ = ππ2 (2.26) Which is equivalent to (ππ)2 = 1,
(2.26)β²
Because (ππ)2 = ππππ = π1π2 = π3 = 1. It is impossible to solve either of these equations so as to express one of the generators in terms of the other. An equation like (2.26) or (2.26)β² is called a defining equation. It is often convenient to specify a
particular group by a set of generators and a set of defining relations. We shall return to this principle in more detail later but mention here that in the present case the equations π3 = π2 = (ππ)2 = 1
(2.27)
Serve as a set of defining relations. Indeed, the information contained in (2.27) suffices to construct the entire multiplication table. First, we remark that the six elements 1, π , π2 , π. ππ, ππ2
(2.28)
are certainly distinct: for example, if a where equal to ππ2 ,it would follow that π β1 = π,contradicting the fact that a and c are irredundant generators. Next, we check that the system (2.28) is closed under multiplication by virtue of the relation (2.27); for example (ππ)(ππ2 ) = π(ππ)π2 = πππ2 π2 = π 2 π4 = π π2 π = π(ππ) = πππ2 = ππ4 = ππ, and so on, a factor c being systematically moved to the left by virtue of (2.26), until the product is identical with one of the elements (2.27). In this notation the complete multiplication table is as follows;
1
a
π2
c
ca
cπ2
1
1
a
π2
c
ca
cπ2
a
a
π2
1
c π2
c
ca
π2
π2
1
a
ca
cπ2
c
c
c
ca
cπ2 1
a
π2
ca
ca
cπ2
c
π2
1
a
c
ca
a
π2
1
cπ2 cπ2 Table 2
and this provides yet another version of the group first presented in table 1. If A , B ,Cβ¦ are subsets of f,the group generated by them is denoted by ππ{π΄ , π΅ , πΆ, β¦ } And is defined as a collection of all finite products in which each factor is an element of A or B or C β¦ or the inverse of such an element, taken in any order with or without repetition. This reduces to the previous concept of generators if we regard as generators of all the elements of Aβͺ π΅ βͺ πΆ βͺ β¦ We might equally well right instead ππ{π΄ βͺ π΅ βͺ πΆ βͺ β¦ }, of course, if A is a subgroup, we have that A=gp{A}
THE DIRECT PRODUCT We shall now discuss a simple method of constructing a new group out of two given groups. Let H and K be any groups and consider the set of all pairs (u, v), Where u and v range over H and K respectively. The set of these pairs is denoted by G= H x K And is called the (exterior) direct product of H and K. The set G is turned into a group by endowing it with the composition law (u,v)(π’β² , π£ β² ) = (π’π’β² , π£π£ β² ).
(2.29)
It is easy to verify that the associative law holds in G, because multiplication is associative in H and in K. The unit element in G is the pair (1π» , 1πΎ ) Where 1π» and 1πΎ are the unit elements of H and K respectively. Also (π’, π£)β1 = (π’β , π£ β1 ). If H and K are finite groups of order h and k respectively, then H x K is a group of order hk. More generally, if π»1 , π»2 , β¦ , π»π are any groups, their direct product π»1 Γ π»2 Γ β¦ Γ π»π consists of all π β π‘π’ππππ (π’1 , π’2 , β¦ , π’π ), where π’π β π»π (π = 1 ,2 , β¦ , π) and multiplication is carried out in each component of the π β π‘π’πππ. If each π»π is finite, then clearly, π
|π»1 , π»2 , β¦ , π»π | = β|π»π | π=1
it sometimes happens that a group G is isomorphic with the direct product of two of its subgroups, H and K, thus πΊ β
π» Γ πΎ,
(2.30)
Or with a slight abuse of notation πΊ = π»ΓπΎ
(2.30)β²
This situation arises in the following circumstances: (1) The subgroups H and K commute elementwise, that is, if u and v are arbitrary elements of H and K respectively, then π’π£ = π£π’ (2.31)
(2) Every element π₯ β πΊ can be expressed in the form π₯ = π’π£, or more briefly, G= π»πΎ (2.32) (3) The intersection of H and K is the unit element, that is π»β©πΎ =1 (2.33) We remark that (2) and (3) are equivalent to the single condition:(2β² ).Every element π₯ β πΊ Can be uniquely decomposed as π₯ = π’π£, where π’ β π» πππ π£ β πΎ. For assume (2) and (3) and suppose we have two decompositions π₯ = π’π£ = π’1 π£1 . Then π’1β1 π’ = π£1 π£ β1 ,
(2.34)
and the element on either side of (2.34) belongs to H and to K, whence by (3), it must be equal to 1. Thus π’ = π’1 πππ π£ = π£1 , which proves the uniqueness of the decomposition, as required in (2β² ). Conversely, assume (2β² ) and suppose that π€ β π» β© πΎ. Then π€ = 1π€ = π€1 are two decompositions of w with factors in H and K respectively, and we deduce from the uniqueness that π€ = 1. The foregoing conditions make it plain that each element π₯ β πΊ uniquely determines a pair (u,v) where π’ β π» πππ π£ β πΎ and that each such pair occurs , because (u,v) corresponds to the product π’π£ = π₯. The correspondence π₯π = (π’π£)π = (π’, π£) establishes the isomorphism (2.30), because, by (1), (π’π£)(π’β² , π£ β² )π = (π’π’β² , π£π£ β² ). Similarly, we have that πΊ β
π»1 Γ π»2 Γ β¦ Γ π»π , where π»π (π = 1 ,2 , β¦ , π) are subgroups of G, if the following conditions are satisfied: (a)any two groups π»π , π»π commute elementwise (b)Every element π₯ of G can be expressed in the form π₯ =, π’1 , π’2 , β¦ , π’π Where π’π β π»π (c)
(2.35)
π»π β© π»1 π»2 β¦ π»πβ1 π»π+1 β¦ π»π = {1}, or alternatively, in place of (2) and (3), (2β² ) The decomposition (2.35) is unique. In particular, if π’1 π’2 β¦ π’π = 1, it follows from (2β² ) that π’1 = π’2 = β― = π’π = 1, Because 1 = 11 β¦ 1 is the only decomposition of 1.
When a group has been expressed as the direct product of subgroups, we speak of an interior direct product. Example The least positive residues prime to the modulus 15 is 1 , 2 ,4 ,7 , 8 ,11 , 13 ,14. (2.36) They form an Abelian group of order 8 , which, as we shall see, is isomorphic with the direct product of the cyclic groups generated by the elements 2 and 11 respectively. In fact, the residue 2 generates a cyclic group of order 4, namely πΆ4 : 1, 2,4 ,8 (24 = 16 β‘ 1(πππ15). Similarly, 11 generates a cyclic group of order 2, πΆ4 : 1, 11 (114 = 121 β‘ 1(πππ15)) Since the whole group is Abelian, we have only to check conditions (2) and (3). Taking all possible products, we obtain 1 , 2 , 4 , 8 , 11 , 22, 44 , 88, Which on reduction mod.15 becomes 1 , 2 , 4 , 8 , 11 , 7, 14 , 13. As this is the complete group, condition (2) holds, and we see at once that πΆ4 β© πΆ4 = {1}. This shows that the group is isomorphic with πΆ4 Γ πΆ4 . The following simple proposition, which is of some independent interest, will be used in the next section. PROPOSITION 6. Let G be a finite group in which all elements satisfy the equation π₯2 = 1
(2.37)
That is each element, other than the unit element, is of order 2. Then G is isomorphic with an Abelian group of type πΆ2 Γ πΆ2 Γ β¦ πΆ2 , And the order of G is therefore a power of 2. Proof By the fact that A group of prime order has no proper subgroups and is necessarily cyclic, the proposition is obviously true when G is the (only) group of order 2. Suppose that G is of order greater than 2 and let a and b be distinct elements other than 1. By hypothesis π2 = π 2 = 1,
So that π = πβ1 , π = π β1 . Next, we consider the element ab. By (2.37), (ππ)2 = 1, whence ππ = (ππ)β1 = π β1 πβ1 = ππ. This shows that G is Abelian. Let π’1 π’2 β¦ π’π be a set of irredundant generators of G. Since G is Abelian, products of the generators can be collected in such a way that every element can be expressed in the βnormalβ form π
π
π
π’1 1 π’2 2 β¦ π’π π .
(2.38)
But by virtue of (2.37) the exponents in (2.28) can be restricted to take only the values of 0 and 1. Then all the products will be distinct; For an equality between two such products would lead to a relation π
π
π
π’11 π’22 β¦ π’ππ = 1, Where each ππ is either zero or unity. This would enable us to express one of the generators in terms of the others, contradicting the hypothesis that the generators were irredundant. Thus, the normal form (2.38) is unique, which amounts to saying that πΊ = ππ{π’1 } Γ ππ{π’2 } Γ β¦ Γ ππ{π’π }, and hence πΊ β
πΆ2 Γ πΆ2 Γ β¦ Γ πΆ2 (π ππππ‘πππ ).
SURVEY OF GROUPS UP TO ORDER OF 8 No successful method has yet been discovered for constructing all possible abstract groups of preassigned order, nor do we know in advance how many such groups exists, except in a few simple cases. However, the elementary means which we have so far developed, suffice to give a complete list of groups up to order 8. Since groups of prime order have already been discussed, it remains only to discuss in more detail the cases in which the order, g, is equal to 4 or 6 or 8.
There are two groups of order 4, both Abelian. For if g=4, an element, other than 1, can be of order 4 or 2 (corollary 1, p. 35). (1) If G contains an element a of order 4, this element generates G; In fact, the four elements of G are 1 , π , π2 , π3 , (π4 = 1) and we have that G= πΊ4 , the cyclic group of order 4. Then all the elements, other than 1, are of order 2 and we deduce from proposition 6 that πΊ = πΆ2 Γ πΆ2 . Thus, G is generated by two elements a and b, and the four elements of G are 1 , π , π, ππ,
(2.39)
Where π2 = π 2 = 1, ππ = ππ
(2.40)
This group is called the four-group (F. Kleinβs βVierergruppeβ) and is often denoted by V There being no other possibilities we conclude that any group of order 4 is isomorphic either with πΆ4 or with V(β
πΆ2 Γ πΆ2 ). In a different notation the multiplication tables of these groups are presented in (3)πππ (4)
1
a
b
c
1
a
b
c
1
1
a
b
c
1
1
a
b
c
a
a
1
c
b
a
a
b
c
1
b
b
c
1
a
b
b
c
1
a
c
c
b
a
1
c
c
1
a
b
table 3
table 4
There are two groups of order 6, one cyclic and one-Abelian. (1) If G possesses an element of order 6, then πΊ = ππ{π} = πΆ6 (2) Next, suppose there is no element of order 6. The order of every element, other than 1, is therefore 2 or 3 (corollary 1, p.35). Since the order of G is not a power of 2, not all its elements can satisfy (2.37). Hence there exists at least one element, a, of order 3, so that 1 , π , π2 (2.41) are three distinct elements of G and π3 = 1,
(2.42)
If c is a further element of G, the six elements 1 , π , π2 , π , ππ, ππ2
(2.43)
are distinct, as we pointed out on P.40, in connection with the elements listed in (2.28) If the elements (2.43) are to form a group of order 6, the axiom of closure must be fulfilled. In particular, πΆ 2 must be one of these elements. We cannot have an element of the form πΆ 2 = πππ (π = 0, 1 , 2), as this would imply that c belongs to the set (2.41). There remain only the following three possibilities: (πΌ)π 2 = 1,
(π½)π 2 = π
(πΎ)π 2 = π2 .
(2.44)
Under the assumption (π½) πππ (πΎ), The element c cannot be of order 2 and must therefore be of order 3. But on multiplying (π½) πππ (πΎ) on the left by c we should obtain 1 = ππ πππ 1 = ππ2 respectively, neither of which can be true. Hence, we conclude that (πΌ) must hold, that is π2 = 1
(2.45)
Next, consider ac. It must be amongst the elements (2.43). As it cannot be equal to c or to a power of a, we are left with the alternatives ππ = ππ,
or ππ = ππ2 .
(2.46)
The first of these renders the group Abelian. Let us find the order of ac in this case. Thus (ππ)2 = π2 π 2 = π2 β 1, (ππ)3 = π3 π 3 = π 3 = π β 1, and the element ac would have to be of order 6, contrary to our initial assumption. Hence the second equation (2.46) must hold, that is ππ = ππ2
or equivalently, (ππ)2 = 1,
See (2.26) and (2.26)β² . To summarize we can state that if there is a group G of order 6 other πΆ6 , then πΊ = ππ{π, π} Subject to the relations π3 = π 2 = (ππ)2 = 1. This does not prove that such a group exists. But we happen to know that this is the case; indeed, its multiplication table is exhibited on table 2. Thus, there are precisely two groups of order 6. There are five groups of order 8, of which three are Abelian and two are non-Abelian. Three Abelian groups of order 8 are written down, namely
1) πΆ8 = ππ{π}, where π8 = 1(π‘ππππ 5) 2) πΆ4 Γ πΆ2 = ππ{π} Γ ππ{π}, π€βπππ π4 = π 2 = 1, ππ = ππ (π‘ππππ 6) 3) πΆ2 Γ πΆ2 Γ πΆ2 = ππ{π} Γ ππ{π} Γ ππ{π}, π€βπππ π2 = π 2 = π 2 = 1, ππ = ππ, ππ = ππ, ππ = ππ(π‘ππππ 7) From the general theory, which we shall develop later, it will follow that these are all positive Abelian groups of order 8, but we shall here derive this result from first principles. If the group contains an element of order 8, it must be the group πΆ8 , and if all elements, other than 1, are of order 2, then it is isomorphic with the group (3). Henceforth we shall assume that each element, other than 1, is either of order 4 or of order 2 and that there is at least one element of order 4, say a, where π4 = 1, π2 β 1.
(2.47)
If b is an element not contained in ππ{π}, then the eight elements 1, π , π2 , π3 , π, ππ , π2 π, π3 π
(2.48)
Are distinct and therefore constitute the whole group, if such a group exists. Now π 2 must be one of these elements and in fact must be one of the first four, since b is not a power of a. The equation π 2 = π or π 2 = π3 have to be ruled out, since they would imply that the order of b is eight. Thus, there remain the possibilities (πΌ)π 2 = 1 or (π½)π 2 = π2
(2.49)
(πΌ): Assume that π 2 = 1. The product ba must be one of the last three elements in (2.48). (πΌ, π): If ba=ab, the group is Abelian and is the group listed under (2). (πΌ, ππ): If ba=π2 π, we could deduce that π β1 π2 π = π, whence (π β1 π2 π)2 = π β1 π4 π = π β1 π2 = π, which is impossible. Hence, we must conclude that (πΌ, πππ): ab=π3 π or equivalently, (ππ)2 = 1. The group defined by the relations π2 = π 2 = (ππ)2 = 1
(2.50)
Does in fact exist. It is denoted by π·4 and is called the dihedral group of order 8 (π‘ππππ 8) It belongs to a class of groups which will be discussed later, when the associative law will be confirmed.
(π½). assume that π 2 = π2 . In this case both a and b are of order 4. Again, ba must be one of the last three elements of (2.48), which we shall consider in turn: (π½, π): If ba=ab, the group is Abelian. The element π = ππ β1 is of order 2 because (ππ β1 )2 = π2 π β1 = 1, and since π = π β1 π, the generator b may be replaced by c. The eight elements may therefore be written as in (2.48), but with c in place of b. Once again, we arrive at the group (2). (π½, ππ):The relation ba= π2 π is impossible as it would imply that ba=π 2 π, π‘βππ‘ ππ π = π 2 , which is inadmissible. (π½, πππ): The only remaining alternative, namely ba=π3 π, is feasible and, as we shall see, leads to a group defined by the relations π4 = 1, π2 = π 2 , ba=π3 π.
(2.51)
In order to demonstrate that such a group does not in fact exist, we construct a faithful matrix representation. Let 0 ], π΄ = [ββ1 0 βββ1
π΅=[
0 β1 ] 1 0
The reader will have no difficulty in verifying that these matrices satisfy the relations (2.51) with the appropriate change of notation and the eight matrices πΌ, π΄ , π΄2 , π΄3 , π΅, π΄π΅ , π΄2 π΅ , π΄3 π΅ are distinct and therefore constitute a multiplicative matrix group, which is isomorphic with the group envisaged under (π½, πππ). This group is known as the quaternion group (table 9). We recall that a quaternion is a hypercomplex number. π0 1 + π1 π + π2 π + π3 π Where the coefficients π0 , π1 , π2 , π3 are real numbers and the symbols 1 , π , π , π satisfy the equations π 2 = π 2 = β1, ππ = βππ = π
or equivalently, π 4 = 1, π 2 = π 2 , ππ = π 3 π,
Which agrees with (2.51) apart from the notation. To summarize our discussion of groups of order 8 we append the complete multiplication tables of the five possible abstract groups of this order:
πΆ8 = ππ{π}, π8 = 1 1
a
π2
π3
π4
π5
π6
π7
1
1
a
π2
π3
π4
π5
π6
π7
a
a
π2
π3
π4
π5
π6
π7
1
π2
π2
π3
π4
π5
π6
π7
1
π3
π3
π4
π5
π6
π7
1
π4
π4
π5
π6
π7
π5
π5
π6
π7
1
a
π2
π3
π4
π6
π6
π7
1
a
π2
π3
π4
π5
π7
π7
1
a
π2
π3
π4
π5
π6
1
a
a
a
π2
π2
π3
Table 5
πΆ4 Γ πΆ2 = ππ{π} Γ ππ{π}, π4 = π 2 = 1.
πΆ8 = ππ{π}, π8 = 1 1
a
π2
π3
b
ab
π2 π
π3 π
1
1
a
π2
π3
b
ab
π2 π
π3 π
a
a
π2
π3
1
ab
π2 π
π3 π
b
π2
π2
π3
1
a
π2 π
π3 π
b
ab
π3
π3
1
a
π2
π3 π
b
ab
π2 π
b
b
ab
π2 π
1
a
π2
π3
ab
ab
π2 π
a
π2
π2 π π2 π
π3 π
π3 π π3 π
b
Table 6
π3 π
π3 π b ab
π2
π3
ab π2 π
π3
1
b
π3 1 a
1 a π2
πΆ2 Γ πΆ2 Γ πΆ2 = ππ{π} Γ ππ{π} Γ ππ{π}, π€βπππ π2 = π 2 = π 2 = 1
1
a
b
c
ab
ac
bc
abc
1
1
a
b
c
ab
ac
bc
abc
a
a
1
ab
ac
b
c
abc
bc
b
b
ab
1
bc
a
abc
c
ac
c
c
ac
bc
1
abc
a
b
ab
ab ab
b
a
abc
1
bc
ac
c
ac ac
c
abc
a
bc
1
ab
b
ππ bc
abc
c
b
ac
ab
1
a
c
b
a
1
abc abc bc ac ππ Table 7
Dihedral group: π2 = π 2 = (ππ)2 = 1
1
a
π2
π3
b
ab
π2 π
π3 π
1
1
a
π2
π3
b
ab
π2 π
π3 π
a
a
π2
π3
1
ab
π2 π
π3 π
b
π2
π2
π3
1
a
π2 π
π3 π
b
π3
π3
1
a
π2
π3 π
b
b
b
π3 π
ab
ab
b
π3 π
ab
b
π2 π π2 π
π2 π ab
π3 π π3 π π2 π ab Table 8
ab
ab π2 π
1
π3
π2
π2 π
a
1
π3
π2
π3 π
π2
a
1
π3
b
π3
π2
a
1
a
Quaternion group: π4 = 1, π2 = π 2 , ba=π3 π
1
a
π2
π3
b
ab
π2 π
π3 π
1
1
a
π2
π3
b
ab
π2 π
π3 π
a
a
π2
π3
1
ab
π2 π
π3 π
b
π2
π2
π3
1
a
π2 π
π3 π
b
ab
π3
π3
1
a
π2
π3 π
b
ab
π2 π
b
b
π3 π π2 π
ab
π2
a
1
π3
ab
ab
b
π3 π
π2 π
π3
π2
a
1
ab
b
π3 π
1
π3
π2
a
b
a
1
π3
π2
π2 π π2 π
π3 π π3 π π2 π ab Table 9
THE PRODUCT THEOREM At the beginning of this chapter, we defined the product of two subsets. We shall now examine the case in which both these subsets are subgroups of a group. It will appear that the product of two subgroups is not always a subgroup, but that explicit information can be obtained in the finite case about the number of elements in the product. THEOREM 5 (Product theorem). (i) Let A and B be subgroups. Then the subsets AB is a group if and only if AB=BA
(2.52)
(ii) In the case of finite subgroups, let |π΄| = π, |π΅| = π, |π΄ β© π΅| = π.Then, irrespective of whether (2.52) holds; |π΄π΅| = |π΅π΄| = ππ/π. Proof. (i) Since A and B are groups, we have that π΄2 = π΄ πππ π΅ 2 = π΅. First, suppose that (2.52) is satisfied and put H=AB. Then π» 2 = π΄π΅π΄π΅ = π΄2 π΅ 2 = π΄π΅ = π», Which proves the closure of H. Obviously 1β π», since 1β π΄ and 1β π΅. Finally if a and b are arbitrary elements of A and B respectively, then π β1 πβ1 β π΅π΄; and hence, by (2.52) π β1 πβ1 β π΄π΅ = π»;that is (ππ)β1 β π»,which completes the proof that H is a group.
Conversely, suppose that H=AB is a group. Hence if a and b are arbitrary elements of A and B respectively, ππ β π», πβ1 π β1 β π», and also (πβ1 π β1 )β1 β π», that is ππ β π». This means π΅π΄ β π΄π΅. In particular, πβ1 π β1 = π1 π1, where π1 and π1 are certain elements of A and B respectively. Hence (π β1 πβ1 )β1 = ππ = π1β1 π1β1 , that is π΄π΅ β π΅π΄ Hence, we conclude that AB=BA. (ii) Let D=π΄ β© π΅. Since D is a subgroup of B, we may decompose B into cosets with respect to D, say π΅ = π·π‘1 βͺ π·π‘2 βͺ β¦ βͺ π·π‘π,
(2.53)
Where π·π‘π β π·π‘π ππ π β π
(2.54)
And n=b/d
(2.55)
multiplying (2.53) on the left by A and observing that AD=A because Dβ π΄, we obtain that AB= π΄π‘1 βͺ π΄π‘2 βͺ β¦ βͺ π΄π‘π .
(2.56)
We claim that no two of the cosets on the right of (2.56) have an element in common; for if not, we should have an equation of the form π’1 π‘π = π’2 π‘π Where π’1 , π’1 β π΄ πππ π β π. Thus π‘π π‘π β1 = π’1 β1 π’2. Now the element on the left belongs to B by (2.53), and the element on the right lies in A. Hence either side denotes an element of D. But π‘π π‘π β1 β π· implies that π·π‘π = π·π‘π , which contradicts (2.54). Thus, the cosets in (2.56) are disjoint, and since each consists of a elements, we have that |π΄π΅| = ππ = ππ/π. The argument is clearly symmetric in A and B, so that also |π΅π΄| = ππ/π.
DOUBLE COSET We saw earlier that the decomposition of a group into cosets relative to a subgroup maybe regarded as an instance of dividing a set into equivalent classes with respect to a suitably defined equivalence relation. Following Frobenius we shall now discuss a different equivalence relation, which involves two subgroups. Let A and B be subgroups of G, which need not be distinct, and term two elements π₯, π¦ β πΊ equivalent, written xβΌ π¦, if there exist elements π’ β π΄ and π£ β π΅ such that π¦ = π’π₯π£
(2.57)
It is easy to check that this is an equivalence relation on G. For (i) xβΌ π¦, since we may take u=1 , π£ = 1, (ii) if π₯ βΌ π¦,then π¦ βΌ π₯, because (2.57) implies that π₯ = π’β1 π¦π£ β1 , (iii) if π₯ βΌ π¦, π¦ βΌ π§,that is π¦ = π’π₯π£, π§ = π’β² π¦π£ β² , where π’β² β π΄, π£ β² β π΅, then π§ = (π’β² π’)π₯(π£π£ β² ), so that π₯ βΌ π§. Thus, the set G may be divided into the equivalence classes which stem from this definition of equivalence. The equivalence class containing π₯ is the complex A π₯π΅, which is called a double coset of G with respect to A and B. We choose a representative from each class and obtain the decomposition πΊ = β π΄π‘π πβπΌ
(2.58)
Where πͺ is a, possibly infinite, index set in one-to one correspondence with the set of double cosets. It is clear that (2.58) is a generalization of the left or right coset decomposition, as is seen by taking A or B to be the trivial group {1}. In contrast to single coset decompositions, the double cosets in (2.58) are not, in general, of the same cardinal. We will pursue the matter further when G is a finite group. Let |πΊ| = π, |π΄| = π, |π΅| = π. First, we observe that the subsets π΄π‘π π΅ and (π‘πβ1 π΄π‘π )π΅ have the same cardinal, because their elements may be put into one-to-one correspondence by matching π’π‘π π£ with π‘πβ1 (π’π‘π π£). Thus |π΄π‘π π΅ | = |(π‘πβ1 π΄π‘π )π΅ |.
Now π‘πβ1 π΄π‘π is a subgroup and |π‘πβ1 π΄π‘π | = π΄ = π On applying theorem 5 to the subgroups π‘πβ1 π΄π‘π and B we find that |π΄π‘π π΅| = ππ/ππ , Where ππ = |π‘πβ1 π΄π‘π β© π΅|. Collecting these results, we obtain the following theorem. THEOREM 6 (Frobenius). Let G be a finite group of order g and let A and B be subgroups of order a and b respectively. Then there exist elements π‘1 , π‘2 , β¦ , π‘π such that G is the disjoint union of double cosets, namely G= π΄π‘1 π΅ βͺ π΄π‘2 π΅ βͺ β¦ π΄π‘π π΅. The number of elements in π΄π‘π π΅ is ab/ππ , Where ππ = |π‘πβ1 π΄π‘π β© π΅|, And consequently π
π = ππ β ππβ1 π=1
(2.59)