Groups of order $pq$ have a proper normal subgroup

I am doing the following exercise from [Birkhoff and MacLane, A survey of modern algebra]:

Let $G$ be a group of order $pq$ ($p,q$ primes). Show that either $G$ is cyclic or contains an element of order $p$ (or $q$). In the second case, show that $G$ contains either $1$ normal or $q$ conjugate subgroups of order $p$. In the latter case the $pq-(p-1)q=q$ elements not of order $p$ form a normal subgroup. Infer that $G$ always has a proper normal subgroup.

I think I was able to prove $G$ has a proper normal subgroup, but without quite establishing that the number of subgroups of order $p$ is 1 or $q$.

Here is my solution:

Suppose $G$ is cyclic and $G=\langle x \rangle$. Then, $\langle x^p \rangle$ has order $q$ and is normal, so we are done. So suppose $G$ is not cyclic. Let $x \in G-1$. By Lagrange's theorem, the order of $x$ is $p$, say. Let $H:=\langle x \rangle$. $G$ acts on all subgroups of order $p$ by conjugation. In this action, the stabilizer of $H$ is the normalizer $N_G(H)$. Since $N_G(H) \supseteq H$, $|N_G(H)|$ equals $p$ or $qp$. Hence the orbit containing $H$ has length $q$ or 1 (by the orbit-stabilizer lemma). If this value is 1, then $H$ is normal, so we are done. If this value is $q$, then there are $pq-(p-1)q=q$ remaining nonidentity elements in $G$ that are not in any conjugate of $H$.

Case 1: If any of these $q$ elements, say some $y \in G$, has order $p$, then $K:=\langle y \rangle$ is either normal (so we are done) or has $q$ conjugate subgroups (in which case the conjugates of $H$ and of $K$ together contain $q(p-1)+q(p-1)+1>qp$ elements, a contradiction).

Case 2: If none of the $q$ elements has order $p$, then they all have order 1 or $q$, and hence form a cyclic subgroup $K$ of order $q$. $K$ must be normal, since otherwise $g^{-1}Kg \ne K$ for some $g \in G$, whereas $K$ already exhausted all the elements of $G$ of order $q$.

Thus, $G$ has a normal subgroup of order $p$ or $q$.

While I have shown the group is not simple, I think I haven't shown yet that the number of subgroups of order $p$ is 1 or $q$. The proof only establishes that the number of distinct conjugates of $H$ is 1 or $q$. Any suggestions on how to prove (using elementary methods) that any other subgroup of order $p$ would have to be conjugate to $H$? In other words, I need to rule out Case 1 in the proof.


In you proof (assuming $G$ is not cyclic), you have get that there exists a normal subgroup $K$ of order $q$, say. Let $H$ be a subgroup of order $p$. If $H$ is normal in $G$, then $G$ will be cyclic. We get $H$ has $q$ conjugates. But now we get $q+q(p-1)=|G|$ elements in $K $ and the conjugate of $H$. Hence there is no other subgroup of order $p$ except the conjugate of $H$. (All the proof holds under the condition $p \ne q$)


  1. You say "If this value is $q$ then there are $pq-(p-1)q=q$ remaining nonidentity elements in $G$..." Not quite. There are $q$ remaining elements. One of the remaining elements is the identity $1$.

Here are a few details that are implicit in your argument (probably too easy to need to be spelled out, but, better safe than sorry):

First, if $H$ is any subgroup with $p$ elements, note that every nonidentity element of $H$ is a generator for $H$.

If $H_1$ and $H_2$ are any two subgroups with $p$ elements each, if $H_1 \cap H_2$ contains any nonidentity element, then that element in the intersection is a generator for both $H_1$ and $H_2$. So $H_1=H_2$. Therefore, if $H_1$ and $H_2$ are any distinct $p$-element subgroups, then they are also disjoint (except for the identity element).

This is the reason why the union of the $q$ conjugate groups of $H$ contains $q(p-1)$ nonidentity elements.

  1. You want to rule out Case 1. It seems to me that your argument does this. Can you say what part worries you? If there is an element $y$ which is not in any conjugate of $H$, and $y$ has order $p$, then $K = \langle y \rangle$ is another subgroup of order $p$, and the conjugates of $K$ again contain $q(p-1)$ nonidentity elements. No conjugate of $K$ shares any nonidentity element with any conjugate of $H$, otherwise $K$ and $H$ would be conjugate. So all together these conjugates of $K$ and $H$ have $2q(p-1)$ nonidentity elements, or $2q(p-1)+1$ elements in total. But $2q(p-1)+1=pq+q(p-2)+1>pq$, as you noted. This is impossible, the union of the conjugates can't have more elements than $G$. So the element $y$ can't have order $p$. This shows that every element $y$ outside the conjugates of $H$ has order $q$.

Since $q$ is prime, every nonidentity element $K = \langle y \rangle$ has order $q$. This accounts for all the $q-1$ elements of order $q$. So, this set of $q$ elements left over from the conjugates of $H$ must be equal to $K$.

And $K$ must be normal because any conjugate of $K$ consists of order $q$ elements, but all the order $q$ elements are already in $K$.

In conclusion, your argument seems fine (modulo a minor wording thing in point (1) above), and in particular it seems to me that you already ruled out Case 1.