Edit: I managed to rephrase my proof in a way that does not resort to coset multiplication. I think the resulting proof is better. I've added it as an answer below, while preserving the original question to avoid wrecking the context.
In his book Finite Group Theory, section 1.D, Isaacs mentions without proof the following result:
Proposition 1: If $G$ is a nilpotent group and $H \lhd G$, then $G/H$ is nilpotent.
I think I have a proof but would appreciate confirmation and any suggestions for improvement.
Isaacs' definition is that $G$ is nilpotent if it has a central series:
$G$ is nilpotent if it has a central series, meaning that there are subgroups $N_i \lhd G$ such that $1 = N_0 \leq N_1 \leq \cdots \leq N_r = G$ and $N_i / N_{i-1} \leq Z(G / N_{i-1})$.
I would like to work directly with this definition rather than with upper or lower central series. The reasons for this: (1) Isaacs' treatment of upper central series assumes that the group is finite, and uses the result I am trying to prove in order to establish the equivalence between "$G$ has a central series" and "$G$ has an upper central series which reaches $G$"; (2) lower central series are not covered until a later chapter.
Initially I tried to prove this by defining $M_i = N_i H / H$ and attempting to show that this results in a central series. I think I made it work using a somewhat messy argument involving the isomorphism and correspondence theorems, but then it occurred to me that instead it suffices to prove the following result:
Proposition 2: If $G$ is nilpotent and $\phi : G \to H$ is an epimorphism, then $H$ is nilpotent.
Then we can apply this to the canonical epimorphism $\phi : G \to G/H$ given by $\phi(g) = gH$, which gives the desired result.
Proof: We have $N_i \lhd G$ and $1 = N_0 \leq N_1 \leq \cdots \leq N_r = G$, with $N_i / N_{i-1} \leq Z(G / N_{i-1})$. The latter condition can be stated equivalently as follows: if $n \in N_i$ and $g \in G$, then $ngN_{i-1} = gnN_{i-1}$.
Now define $M_i = \phi(N_i)$ for every $0 \leq i \leq r$. Note that $M_0 = \phi(N_0) = \phi(1) = 1$, and $M_r = \phi(N_r) = \phi(G) = H$. Also, $M_i \lhd H$ by the correspondence theorem since $N_i \lhd G$.
Now, let $m \in M_i$ and $h \in H$. To show that $M_i / M_{i-1} \leq Z(H / M_{i-1})$, it suffices to show that $mhM_{i-1} = hmM_{i-1}$. We have $m = \phi(n)$ and $h = \phi(g)$ for some $n \in N_i$ and $g \in G$. Then $mhM_{i-1} = \phi(ngN_{i-1}) = \phi(gnN_{i-1}) = hmM_{i-1}$, where the second equality holds because $N_{i} / N_{i-1} \leq Z(G / N_{i-1})$. We conclude that $1 = M_0 \leq M_1 \leq \cdots\leq M_r = H$ is a central series for $H$, so $H$ is nilpotent.
Best Answer
I cleaned up the proof so that it no longer uses the somewhat unprofessional "coset multiplication" argument. I'm happy with it now, so I'll post it as an answer in case it is of interest to anyone.
Recall that the goal is to prove the following:
First, we prove a simple lemma:
Proof: Let $z \in Z(G)$ and $h \in H$. Then $h = \theta(g)$ for some $g \in G$, and $\theta(z)h = \theta(zg) = \theta(gz) = h\theta(z)$. Therefore, $\theta(z) \in Z(H)$.
Now we prove the following proposition:
Proof: Since $G$ is nilpotent, we have $N_i \lhd G$ and $1 = N_0 \leq N_1 \leq \cdots \leq N_r = G$, with $N_i / N_{i-1} \leq Z(G / N_{i-1})$.
Define $M_i = \phi(N_i)$. We claim that $1 = M_0 \leq M_1 \leq \cdots \leq M_r = H$ is a central series for $H$.
First, we observe that $M_0 = \phi(N_0) = \phi(1) = 1$ and $M_r = \phi(N_r) = \phi(G) = H$. Also, $M_i \lhd H$ by the correspondence theorem, since $N_i \lhd G$.
Now define $\theta : G / N_{i-1} \to H / M_{i-1}$ by $\theta(gN_{i-1}) = \phi(g)M_{i-1}$. It is routine to verify that $\theta$ is a well-defined epimorphism. We can therefore apply the lemma to conclude that $\theta(Z(G/N_{i-1})) \leq Z(H / M_{i-1})$.
Note also that $\theta(N_i / N_{i-1}) = \phi(N_i) / M_{i-1} = M_i / M_{i-1}$.
Since $G$ is nilpotent, we have $N_i / N_{i-1} \leq Z(G/N_{i-1})$. Consequently, $M_i / M_{i-1} = \theta(N_i / N_{i-1}) \leq \theta(Z(G/N_{i-1})) \leq Z(H / M_{i-1})$, as required.