For simplicity, let $A$ be the interior of the unit square in $\mathbb{R}^2$: $A = \{(x,y) \in \mathbb{R}^2\mid 0 < x < 1, 0 < y < 1\}$. Intuitively, the boundary is $B = \{(x,y) \in \mathbb{R}^2\mid (x = 0 \text{ or } x = 1) \text{ and } 0 \le y \le 1, \text{ or } (y = 0 \text{ or } y = 1) \text{ and } 0 \le x \le 1\}$.
To show that every neighborhood of each point of $B$ contains points of $A$, you have to consider each of the 4 edges that comprise B. But in fact, there are only two cases: those with some coordinate equal to $0$, and those with a coordinate equal to $1$.
For example, consider the bottom edge, excluding $(1,0)$ which can be treated when considering the right edge.
First $(0,0)$: given $\epsilon > 0$, we can assume $\epsilon < 1$. Let $r = \epsilon / 2$; then $(r,r)$ is in the open $\epsilon$-ball around $(0,0)$, and $(r,r) \in A$.
For other $(x,0) \in B$, with $0 < x$: Given $\epsilon > 0$, again we can assume that $\epsilon < 1$. Let $y = \epsilon /2$; then
$d((x,y), (x,0)) = \sqrt {(\epsilon/2)^2} = \epsilon / 2 < \epsilon$, so $(x,y)$ is in the open $\epsilon$-ball around $(x,0)$, and $(x,y) \in A$.
The left edge ($x = 0$) is similar to the bottom edge ($y = 0$). The edges with one coordinate equal to $1$ are a little different in that you subtract a value to find a point that is both in $A$ and in the open $\epsilon$ ball. There are also the corner cases, literally, to consider separately.
To show that nothing else is in the boundary, two other facts need to be proved:
- $A$ is open, and
- $\mathbb{R}^2 \setminus (A \cup B)$ is open.
These are straightforward, and it should be fairly clear by now what their proofs are like.
In fact, this proves the proposition for any rectangle $R$ in $\mathbb{R}^2$. Suppose $R$ has its lower left corner at $\mathbf{a}$, and has width $w$ and height $h$. Then R is the image of the unit square under the homeomorphism $(x,y)\mapsto \mathbf{a} + (wx, hy)\colon \mathbb{R}^2 \to \mathbb{R}^2$. The result also follows for rectangles whose sides are not parallel to the axes, because any of those is the image under a rotation of a rectangle with sides parallel to the axes.
Proving the general statement for open rectangles in $\mathbb{R}^n$ is no different in principle — two basic cases, with subcases — but the notation gets heavy with subscripts.