[guided]This is the heart of the counting argument, and it proceeds by iterating a single principle — the [Roots-Homomorphisms Correspondence](/theorems/1256) — along a tower of simple extensions.
**Setting up the tower.** We decompose $L/K$ into a chain of simple extensions by adjoining one generator at a time:
\begin{align*}
K = K_0 \subset K_1 = K(\alpha_1) \subset K_2 = K(\alpha_1, \alpha_2) \subset \cdots \subset K_n = K(\alpha_1, \ldots, \alpha_n) = L.
\end{align*}
At each level, $K_i = K_{i-1}(\alpha_i)$ is a simple extension of $K_{i-1}$, and by hypothesis (4), the minimal polynomial $Q_i$ of $\alpha_i$ over $K_{i-1}$ is separable.
**The iteration.** We count $|\operatorname{Hom}_K(L, \bar{K})|$ by building up homomorphisms one step at a time.
*Base case ($i = 0$):* There is exactly one $K$-homomorphism from $K_0 = K$ to $\bar{K}$, namely the inclusion $\iota \colon K \hookrightarrow \bar{K}$.
*Step $i$:* Given any $K$-homomorphism $\tau \colon K_{i-1} \to \bar{K}$, we ask: in how many ways can $\tau$ be extended to a $K$-homomorphism $\tilde{\tau} \colon K_i = K_{i-1}(\alpha_i) \to \bar{K}$? The [Roots-Homomorphisms Correspondence](/theorems/1256) provides the answer: extensions of $\tau$ to $K_i$ are in bijection with roots of $\tau(Q_i)$ in $\bar{K}$, via the map $\tilde{\tau} \mapsto \tilde{\tau}(\alpha_i)$.
The polynomial $\tau(Q_i)$ is obtained by applying $\tau$ to each coefficient of $Q_i$. Since $\tau$ is a field homomorphism (hence injective), $\tau(Q_i)$ has the same degree as $Q_i$. Moreover, $Q_i$ is separable over $K_{i-1}$ by hypothesis, meaning $Q_i$ has $\deg Q_i$ distinct roots in any algebraic closure of $K_{i-1}$. The isomorphism $\tau \colon K_{i-1} \xrightarrow{\sim} \tau(K_{i-1})$ carries this splitting to $\tau(Q_i)$: concretely, if $\beta_1, \ldots, \beta_d$ are the distinct roots of $Q_i$ in some algebraic closure $\overline{K_{i-1}}$, then $\tau(\beta_1), \ldots, \tau(\beta_d)$ are distinct roots of $\tau(Q_i)$ (here $\tau$ extends to $\overline{K_{i-1}}$). Since $\bar{K}$ is algebraically closed and contains $\tau(K_{i-1})$, the polynomial $\tau(Q_i)$ has all $d = \deg Q_i$ roots in $\bar{K}$, and they are distinct.
Therefore $\tau$ extends in exactly $\deg Q_i = [K_i : K_{i-1}]$ ways to $K_i$.
**Multiplying through the tower.** At each stage $i$, every homomorphism $\tau \colon K_{i-1} \to \bar{K}$ gives rise to exactly $[K_i : K_{i-1}]$ extensions to $K_i$. Starting from the single inclusion $K \hookrightarrow \bar{K}$ and iterating:
\begin{align*}
|\operatorname{Hom}_K(L, \bar{K})| &= \prod_{i=1}^{n} [K_i : K_{i-1}] = [K_n : K_0] = [L : K],
\end{align*}
where the product telescopes by the Tower Law: $[L:K] = [K_n : K_{n-1}] \cdots [K_1 : K_0]$.
This achieves equality $|\operatorname{Hom}_K(L, \bar{K})| = [L:K]$, which is condition (1) with $E = \bar{K}$.
**Why separability is consumed here.** The argument would fail at step $i$ if $Q_i$ were inseparable: in that case, $\tau(Q_i)$ would have fewer than $\deg Q_i$ roots in $\bar{K}$ (due to repeated roots), and $\tau$ would extend in strictly fewer than $[K_i : K_{i-1}]$ ways. The product would then be strictly less than $[L:K]$, giving only the inequality $|\operatorname{Hom}_K(L, \bar{K})| < [L:K]$.[/guided]