In Loring Tu's Introduction to Manifolds pg. 49, it was said that the cross in $\mathbb R^2$ 
is not a topological manifold because it is not locally Euclidean at the intersection $p$.
In the proof, it was written:
Suppose the cross is locally Euclidean of dimension $n$ at the point $p$. Then $p$ has a neighborhood $U$ homeomorphic to an open ball $B :=B(o,\epsilon) \subset\mathbb R^n$ with $p $ mapping to $0$.
But the definition of locally Euclidean is given on pg. 48 as:
A topological space $M$ is locally Euclidean of dimension $n$ if every point $p$ in $M$ has a neighborhood $U$ such that there is homeomorphism $\phi $ from $U$ onto an open subset of $\mathbb R^n$.
The definition suggests that there may be only one neighbohood $U$ containing $p$ that can be mapped onto an open subset of $\mathbb R^n$, which need not be an open ball. Why can the author assume that for the cross, there can be multiple neighborhoods $U$ containing $p$ that each maps onto an open ball of arbitrary radius $\epsilon?$
Best Answer
Locally Euclidean implies more than you think: let $p \in M$ and let $U$ be the promised neighbourhood of $p$ in the definition.
So there is a homeomorphism $h: U \to O$ where $O \subseteq \Bbb R^n$ is (Euclidean) open.
So in particular $h(p) \in O$ and so for some $r>0$ we have that $B(h(p), r) \subseteq O$ by the definition of the Euclidean topology. But if $V$ now is an open ball around $h(p)$ of radius $0 < r' \le r$ we have that
$V':=h^{-1}[V]$ is open in $U$ and $h\restriction_{V'}: V' \to V$ is also a homeomorphism (simple check using that the restriction of an open map to open sets is still open, restrictions of continuous maps remain continuous etc. All basic facts).
So we can pick the codomain set to be some open ball of small enough radius if that is convenient in the proof. This proof is an example where it's used.