[guided]The covering-interval technique is the engine that converts combinatorial data (which intervals cover which) into analytic conclusions (existence of periodic points). Let $f \colon \mathbb{R} \to \mathbb{R}$ be continuous with a $k$-cycle $\{x_1 < \cdots < x_k\}$, and define $I_j = [x_j, x_{j+1}]$ for $j = 1, \ldots, k-1$.
We say $I_j$ **covers** $I_m$ (written $I_j \to I_m$) if $I_m \subset f(I_j)$. Since $f$ is continuous, $f(I_j)$ contains every value between $f(x_j)$ and $f(x_{j+1})$ (by the Intermediate Value Theorem), so $I_j \to I_m$ whenever $I_m$ lies between $f(x_j)$ and $f(x_{j+1})$.
The key claim is: **a closed path $I_{j_0} \to I_{j_1} \to \cdots \to I_{j_{n-1}} \to I_{j_0}$ of length $n$ yields a periodic point $z$ with $f^n(z) = z$.** The proof works by pulling back through the chain of coverings.
Start at the end: $I_{j_{n-1}} \to I_{j_0}$ means $I_{j_0} \subset f(I_{j_{n-1}})$. By the [Onto Closed Intervals](/theorems/2803) lemma, there exists a closed sub-interval $C_{n-1} \subset I_{j_{n-1}}$ with $f(C_{n-1}) = I_{j_0}$. Next, $I_{j_{n-2}} \to I_{j_{n-1}}$ means $I_{j_{n-1}} \subset f(I_{j_{n-2}})$, and since $C_{n-1} \subset I_{j_{n-1}}$, the lemma gives $C_{n-2} \subset I_{j_{n-2}}$ with $f(C_{n-2}) = C_{n-1}$. Continuing this backward induction through the entire chain, we obtain nested sub-intervals $C_0 \subset I_{j_0}$, $C_1 \subset I_{j_1}$, ..., $C_{n-1} \subset I_{j_{n-1}}$ with $f(C_i) = C_{i+1}$ for $i = 0, \ldots, n-2$ and $f(C_{n-1}) = I_{j_0}$. In particular, $f^n(C_0) = I_{j_0} \supset C_0$.
Since $f^n$ maps $C_0$ over a set containing $C_0$, the function $g(x) = f^n(x) - x$ satisfies: there exist $a, b \in C_0$ with $f^n(a) = \min I_{j_0} \leq a$ (so $g(a) \leq 0$) and $f^n(b) = \max I_{j_0} \geq b$ (so $g(b) \geq 0$). By the Intermediate Value Theorem applied to $g$ on $[a,b] \subset C_0$, there exists $z \in C_0$ with $g(z) = 0$, i.e., $f^n(z) = z$. By construction, $f^i(z) \in C_i \subset I_{j_i}$ for each $i = 0, 1, \ldots, n-1$, so the orbit of $z$ visits the prescribed sequence of intervals.
The delicate part is ensuring the periodic point has *exact* period $n$ rather than a proper divisor of $n$. A fixed point of $f^n$ could have period $d | n$ with $d < n$ if the orbit revisits its starting interval prematurely. To rule this out, one chooses closed paths in the directed graph that do not return to $I_{j_0}$ before completing the full circuit of length $n$. Concretely, one selects paths where $j_1, \ldots, j_{n-1}$ are all distinct from $j_0$, or where the return pattern is incompatible with any shorter period.
For instance, in the proof that an odd $m$-cycle forces a period-$l$ cycle (for $l \triangleleft m$), one constructs a specific closed path of length $l$ that passes through a self-covering interval $I_j \to I_j$ and visits distinct intermediate intervals, ensuring that the resulting fixed point of $f^l$ cannot have period less than $l$.
The Sharkovsky ordering emerges precisely from the combinatorial constraints on which closed-path lengths are achievable for each cycle type. Odd cycles produce the richest directed graphs (with self-loops and long paths), while power-of-$2$ cycles produce the simplest graphs (with only $2$-cycles in the graph, yielding periods that are powers of $2$).[/guided]