Let $X^A$ be Cartesian coordinates on $\mathbb R^{n+1}$ and $x^\mu$ be an arbitrary coordinates on an $n$-dimensional sphere embedded in $\mathbb R^{n+1}$ in a usual way. Then $x^\mu, r = \sqrt{X_A X^A}$ would be coordinates on $\mathbb R^{n+1}$.
How can I show that in this coordinates the metric on $\mathbb R^{n+1}$ takes the form $$ ds^2 = r^2 h_{\mu\nu}(x) dx^\mu dx^\nu + dr^2, $$ where $h_{\mu\nu}$ is the metric tensor on a unit sphere induced by embedding?
OK, I can write $$ ds^2 = \eta_{AB} dX^A dX^B = \eta_{AB} \left( \frac{\partial X^A}{\partial x^\mu} dx^\mu + \frac{\partial X^A}{\partial r} dr \right) \left( \frac{\partial X^B}{\partial x^\nu} dx^\nu + \frac{\partial X^B}{\partial r} dr \right). $$ $g_{\mu\nu} = \eta_{AB} \frac{\partial X^A}{\partial x^\mu} \frac{\partial X^B}{\partial x^\nu}$ ($\eta_{AB} = diag(1,\dots,1)$) is the induced metric on a sphere $r = const$. $g_{\mu\nu}|_{r=1} = h_{\mu\nu}$. $g_{\mu\nu} = r^2 h_{\mu\nu}$ follows from dimensional considerations (but it would be nice to get a formal proof).
I'm stuck with the remaining parts of the tensor ($g_{\mu(r)} = g_{(r) \mu} \overset{?}{=} 0$ and $g_{(r)(r)} \overset{?}{=} 1$). I can differentiate $X_A X^A = r^2$ by $x^\mu$ and $r$ to get $$ X_A \frac{\partial X^A}{\partial x^\mu} = 0, \qquad X_A \frac{\partial X^A}{\partial r} = -r $$ but partial derivatives I need come in contraction with $X_A$ so it seems of no use...