[step:Handle the boundary term in the whole-space case]Now suppose $U=\mathbb{R}^n$. For $R>0$, define the truncated ball $B_R := B(0,R)$ and define the truncated energy $E_R : [0,T] \to [0,\infty)$ by
\begin{align*}
E_R(t) = \frac{1}{2}\int_{B_R} |\partial_t u(t,x)|^2\, d\mathcal{L}^n(x) + \frac{c^2}{2}\int_{B_R} |\nabla u(t,x)|^2\, d\mathcal{L}^n(x).
\end{align*}
Since $B_R$ is bounded and $u$ is $C^2$ on $[0,T]\times \overline{B_R}$, differentiating under the integral sign on $B_R$ is justified and gives
\begin{align*}
E_R'(t) = \int_{B_R} \partial_t u(t,x)\,\partial_t^2 u(t,x)\, d\mathcal{L}^n(x) + c^2\int_{B_R} \nabla u(t,x)\cdot \nabla\partial_t u(t,x)\, d\mathcal{L}^n(x).
\end{align*}
Let $\nu_R : \partial B_R \to \mathbb{R}^n$ denote the outward unit normal field. Applying Green's first identity on $B_R$, namely the classical integration-by-parts formula on the smooth ball $B_R$, to the $C^1$ functions $x\mapsto u(t,x)$ and $x\mapsto \partial_tu(t,x)$ gives
\begin{align*}
E_R'(t) = \int_{B_R} \partial_t u(t,x)\left(\partial_t^2u(t,x)-c^2\Delta u(t,x)\right)\, d\mathcal{L}^n(x) + c^2\int_{\partial B_R} \partial_t u(t,x)\,\nabla u(t,x)\cdot\nu_R(x)\, d\mathcal{H}^{n-1}(x).
\end{align*}
The wave equation cancels the interior integral, so
\begin{align*}
E_R'(t) = c^2\int_{\partial B_R} \partial_t u(t,x)\,\nabla u(t,x)\cdot\nu_R(x)\, d\mathcal{H}^{n-1}(x).
\end{align*}
The global energy $E_u : [0,T]\to[0,\infty)$ is differentiable by the $L^2$-valued $C^1$ hypotheses from the first step, and its derivative is
\begin{align*}
E_u'(t) = \int_{\mathbb{R}^n} \partial_t u(t,x)\,\partial_t^2 u(t,x)\, d\mathcal{L}^n(x) + c^2\int_{\mathbb{R}^n} \nabla u(t,x)\cdot \nabla\partial_t u(t,x)\, d\mathcal{L}^n(x).
\end{align*}
We use the pointwise inequality
\begin{align*}
|ab|\leq \frac{a^2+b^2}{2}
\end{align*}
with $a=\partial_tu(t,x)$ and $b=\partial_t^2u(t,x)$, and the analogous vector inequality
\begin{align*}
|v\cdot w|\leq \frac{|v|^2+|w|^2}{2}
\end{align*}
with $v=\nabla u(t,x)$ and $w=\nabla\partial_tu(t,x)$. The $C^1$ Hilbert-space hypotheses imply that the functions $x\mapsto \partial_tu(t,x)\,\partial_t^2u(t,x)$ and $x\mapsto \nabla u(t,x)\cdot\nabla\partial_tu(t,x)$ belong to $L^1(\mathbb{R}^n)$. By the absolute continuity of the [Lebesgue integral](/page/Lebesgue%20Integral) applied to these two $L^1$ functions, the integrals over $B_R$ converge to the corresponding integrals over $\mathbb{R}^n$ as $R\to\infty$, and hence $E_R'(t)\to E_u'(t)$.
Let $\operatorname{supp}(f)$ denote the closure of the set where a function $f$ is nonzero. If $u(t,\cdot)$ has compact support, choose $R_0>0$ such that $\operatorname{supp}(u(t,\cdot))\subset B(0,R_0)$. Since the function $x\mapsto u(t,x)$ is $C^1$ and vanishes on the [open set](/page/Open%20Set) $\mathbb{R}^n\setminus \operatorname{supp}(u(t,\cdot))$, its spatial gradient satisfies $\nabla u(t,x)=0$ on $\mathbb{R}^n\setminus \operatorname{supp}(u(t,\cdot))$. Hence, for every $R>R_0$, the boundary flux integrand vanishes on $\partial B_R$, and the boundary integral is zero. Under the stated flux decay assumption, the boundary integral also tends to zero, since
\begin{align*}
\left|\int_{\partial B_R} \partial_t u(t,x)\,\nabla u(t,x)\cdot \nu_R(x)\, d\mathcal{H}^{n-1}(x)\right|\leq\int_{\partial B_R} |\partial_t u(t,x)|\,|\nabla u(t,x)|\, d\mathcal{H}^{n-1}(x).
\end{align*}
Therefore $E_u'(t)=0$ for every $t\in(0,T)$ in the whole-space case.[/step]