In chapter 4 of "Measure Theory and Fine Properties of Functions" by Evans and Gariepy, part (ii) of theorem 4.4 states:
If $f \in W^{1,p}(U)$ and $F \in C^{1}(\mathbb{R})$, $F' \in L^{\infty}(\mathbb{R})$, $F(0)=0$, then $F(f) \in W^{1,p}(U)$ and $F(f)_{x_i}=F'(f)f_{x_i}$ $\mathcal{L}^n$-a.e. for $i=1,..,n$.
With the remark at the end of the proof that:
If $\mathcal{L}^n(U) < \infty$, then the condition $F(0)=0$ for (ii) is unnecessary.
The proof given in the book is as follows: if $\phi \in C^1_c(U)$ with $spt(\phi)\subset V \subset \subset U$ and $f^{\epsilon}$ is the "standard mollification", that is mollification with: $\eta(x) = ce^{\frac{1}{x^2-1}}$ when $|x|<1$ and $\eta(x)=0$ for $|x|>1$, $c$ chosen so that it has integral $1$, then:
$\begin{align*} \int_UF(f)\phi_{x_i}dx = \int_VF(f)\phi_{x_i}dx \\= \lim_{\epsilon \rightarrow 0} \int_V F(f^{\epsilon})\phi_{x_i}dx \\ = -\lim_{\epsilon \rightarrow 0} \int_V F'(f^{\epsilon})f_{x_i}^\epsilon \phi dx \\=-\int_VF'(f)f_{x_i}\phi dx \\= -\int_UF'(f)f_{x_i}\phi dx \end{align*}$
I am having trouble seeing where the assumption that $F(0)=0$ is used in this proof, and also why it would be unnecessary in the case the $U$ has finite Lebesgue measure. Does anybody have a hint?
Related question: Sobolev inequality and chain rule