We turn now to another important decomposition of an arbitrary group G into disjoint complexes. Such decompositions will play an important role in our later considerations of the Sylow Theorems.
Let
be an arbitrary group and let
and
be subgroups of
. For
, we define
Thus
given by (8.3) is indeed an equivalence relation on
. We next take a look at the equivalence classes.
For
, the equivalence class
of
as we recall, contains all
with
. By the definition given in (8.3), this means
, where
and
. Thus
. As the above statements are all ``if and only if'', we see that
, the double coset given in Definition 8.2.1. By our general theorem on equivalence relations, Theorem 1.1.4, we know that either

If
, then we simply get the right coset decomposition of
with respect to
. If
, then we have the left coset decomposition of
with respect to
. Thus the double coset decomposition of a group may be viewed as a generalization of the coset (right or left) decomposition of a group. However, the reader should be careful not to generalize all facts related to coset decompositions to the case of double coset decompositions. For example, we saw that any two cosets of a finite group have the same number of elements. We shall presently see that this is not the case with double cosets.
Let us consider the double coset
. Clearly
contains all right cosets of the form
, where
and
contains all left cosets of the form
where
. We claim, as a matter of fact, that
is a union of right or left cosets of the above form. For suppose that
Next, we wish to ascertain the number of left and right cosets in the double coset. Even though this number can be finite for an infinite double coset, we assume
. This is contained in
Proof: We first note that
is a subgroup by Proposition 7.2.1. Consider the mapping of the double coset
onto the complex
given by
. It is easy to show that this map is well-defined, 1-1, and onto (see exercise 2 for this section). Thus
. But
is the product of two subgroups
and
so by the product theorem (Theorem 4.3.6),

![\begin{displaymath} \frac{\vert a^{-1}H_1a\vert}{\vert a^{-1}H_1a\cap H_2 \vert} =[a^{-1}H_1a:a^{-1}H_1a\cdot H_2 ]. \end{displaymath}](img1713.png)
Under the same hypotheses as in Theorem 8.2.2, we use the notation
to be the number of right cosets of
in
times
(note from (8.5) that
). Thus Theorem 8.2.2 implies that
