Let $G$ be a finite nilpotent group and $G'$ its commutator subgroup. Show that if $G/G'$ is cyclic then $G$ is cyclic.

Let $G$ be finite and nilpotent. Then $G' subseteq Phi(G)$, the Frattini subgroup. So, if $G/G'$ is cyclic, then $G/Phi(G)$ is cyclic. Hence, there is a $g in G$ with $G/Phi(G)=langle overline{g} rangle$. But then $G=langle g ranglePhi(G)=langle g rangle$, since the elements of the Frattini subgroup are non-generators.

Derek's answer in the comments is the fastest. Induction on $|G|$. Since $G$ is nilpotent $Z(G)>1$. The commutator subgroup of $G/Z(G)$ must contains (the image of) the commutator subgroup of $G$, whence $G/Z(G)$ satisfies the induction hypothesis. Thus $G/Z(G)$ is cyclic, whence $G$ is abelian. Thus $G$ is cyclic and $G$ is generated by one element.

---

The 'correct' way to do this is to work with maximal subgroups. It suffices to show this for $p$-groups. Notice that $G$ is cyclic means it is $1$-generator, so let $x_1,dots,x_n$ be a minimal generating set for $G$. Let $M$ be a maximal subgroup of $G$, necessarily normal and of index $p$. Then $G/M$ is cyclic of order $p$. But $G/G'$ is cyclic, so $M$ must be the unique maximal subgroup of $G$. By assumption, removing any single generator from the set results in a proper subgroup, whence is contained in $M$. But then this means that (unless $n=1$) all $x_i$ lie in $M$, a contradiction.

This generalizes into the Burnisde basis theorem, one of the foundational results of the theory of $p$-groups.

Let $$G=gamma_0(G)>[G,G]=gamma_1(G)>...>gamma_n(G)=1$$ be the lower central series of $G$. We know that $G/gamma_1(G)$ is cyclic. Let $k>1$ be the maximal number such that $G/gamma_k(G)$ is cyclic. If $k=n$, we are done. So suppose $k<n$. Then $G_k/G_{k+1}$ is central in $G/G_{k+1}$. Since $G/G_kequiv (G/G_{k+1})/(G_k/G_{k+1})$ is cyclic, we have that $G/G_{k+1}$ is abelian, so $gamma_1(G)le gamma_{k+1}(G)$, a contradiction. QED