Let $A(x) = (a^{ij}(x))_{ij}$, $i, j = 1, 2, 3$ be a smooth matrix-valued function on $\mathbb{R}^3$ such that $a^{ij} = a^{ji}$, and there exists constants $c, C$ such that $c|\zeta|^2 \le \langle A(x)\zeta, \zeta \rangle \le C|\zeta|^2$ for all $(x, \zeta) \in \mathbb{R}^3$ x $\mathbb{R}^3$.

Show that if $u$ is a smooth solution to the variable-coefficient wave equation:

$\partial_t ^2 u$ - div$(A \nabla u) = 0$,

such that $u$ and its derivatives decay rapidly as $|x| \rightarrow \infty$ (say, bounded by $|x|^{-3/2 - \epsilon}$ for some $\epsilon > 0$), then:

$E(t) := \frac{1}{2} \int_{\mathbb{R}^3} (|\partial_t u|^2 + \langle A \nabla_x u, \nabla_x u \rangle) dx$

is conserved.

This is what I've done so far:

$E(t) := \frac{1}{2} \int_{\mathbb{R}^3} (|d_t u|^2 + \langle A \nabla u, \nabla u \rangle) dx$

= $\int_{\mathbb{R}^3}(u_t u_{tt} + \frac{1}{2} [A \nabla u_t \cdot \nabla u + A \nabla u \cdot \nabla u_t])dx$

= $\int_{\mathbb{R}^3} (u_t \nabla \cdot (A \nabla u) + \frac{1}{2}[A \nabla u_t \cdot u + A \nabla u \cdot \nabla u_t])dx$

But I don't know where to go for the next step. I am honestly completely stuck on this part. Any help/guidance on where to go next would be great. Thank you.


We are given the partial differential equation

$\partial_t^2 u = \text{div}(A \nabla u) = \nabla \cdot (A \nabla u), \tag 1$

for which we define the energy at time $t$ via an integral over $\Bbb R^3$:

$E(t) = \displaystyle \dfrac{1}{2} \int_{\Bbb R^3} ((\partial_t u)^2 + \langle A\nabla u, \nabla u \rangle) \; dx; \tag 2$

we wish to show that

$\dot E(t) = \partial_t E(t) = 0. \tag 3$

Before proceeding with the demonstration of (3), a few remarks are in order: first, there is no need to take the absolute value of $\partial_t u$ when forming the term $(\partial_t u)^2$ appearing in (2), as our OP Axioms does in the text of his question. Since we are taking the square of $\partial_t u$, the absolute value signs are unnecesary: $\vert \partial_t u \vert^2 = (\partial_t u)^2$; this simplification makes taking the $t$-derivative of $\vert \partial_t u \vert^2 = (\partial_t u)^2$ somewhat more transparent, since it makes clear that we needn't be concerned with the failure of $\vert \cdot \vert$ to be differentiable when its argument vanishes; second, the hypothesis that $a^{ij}(x) = a^{ji}(x)$, i.e., that $A(x)$ is a symmetric matrix, $A^T(x) = A(x)$, for all $x \in \Bbb R^3$ implies, as is well known, that for any vectors $v, w$ we have

$\langle v, A(x) w \rangle = \langle A(x) v, w \rangle; \tag 4$


$\langle v, A(x) w \rangle = \langle A^T(x) v, w \rangle = \langle A(x) v, w \rangle; \tag 5$

this property of the coefficient matrix $A(x)$ will enter into the discussion which follows; third, there is no need to write the subscript $x$ when taking $\nabla u = \nabla_x u$, since it is understood that $\nabla$ refers only to the spatial variables $x$; bearing these remarks in mind, we proceed.

From (2) we have

$\dot E(t) = \partial_t E(t) = \displaystyle \dfrac{1}{2} \partial_t \int_{\Bbb R^3} ((\partial_t u)^2 + \langle A\nabla u, \nabla u \rangle) \; dx$ $= \displaystyle \dfrac{1}{2} \int_{\Bbb R^3} \partial_t ((\partial_t u)^2 + \langle A\nabla u, \nabla u \rangle) \; dx = \dfrac{1}{2} \int_{\Bbb R^3} (\partial_t (\partial_t u)^2 + \partial_t \langle A\nabla u, \nabla u \rangle) \; dx; \tag 6$


$\partial_t (\partial_t u)^2 = 2 \partial_t u \; \partial_t^2 u; \tag 7$


$\partial_t \langle A\nabla u, \nabla u \rangle = \langle \partial_t(A \nabla u), \nabla u \rangle + \langle A \nabla u, \partial_t \nabla u \rangle, \tag 8$


$\partial_t \nabla u = \nabla \partial_t u, \tag 9$

since the $t$- and $x$-derivatives commute, and

$\partial_t A \nabla u = A \partial_t \nabla u, \tag{10}$

since $A(x)$ is constant with respect to $t$; combining (8), (9) and (10) we see that

$\partial_t \langle A\nabla u, \nabla u \rangle = \langle A \nabla \partial_t u, \nabla u \rangle + \langle A \nabla u, \nabla \partial_t u \rangle, \tag{11}$

and using the symmetric property (4) of $A$ and of the inner product ($\langle v, w \rangle = \langle w, v \rangle$),

$\partial_t \langle A\nabla u, \nabla u \rangle = \langle \nabla \partial_t u, A \nabla u \rangle + \langle \nabla \partial_t u, A \nabla u \rangle = 2\langle \nabla \partial_t u, A \nabla u \rangle; \tag{12}$

using (7) and (12) we return to (6) and write it in the form

$\dot E(t) = \partial_t E(t) = \displaystyle \dfrac{1}{2} \int_{\Bbb R^3} (\partial_t (\partial_t u)^2 + \partial_t \langle A\nabla u, \nabla u \rangle) \; dx$ $= \displaystyle \int_{\Bbb R^3} (\partial_t u \; \partial_t^2u + \langle \nabla \partial_t u, A \nabla u \rangle) \; dx, \tag{13}$

where the factors of $2$ occurring in (7) and (12) have been cancelled against the $1/2$ occurring in (6).


$\nabla \cdot (\partial_t u (A \nabla u)) = \langle \nabla \partial_t u, A \nabla u) \rangle + \partial u_t \nabla \cdot (A \nabla u), \tag{14}$

which follows from the standard formula from vector calculus

$\nabla \cdot (fX) = \langle \nabla f, X \rangle + f \nabla \cdot X \tag{15}$

applied with $f = \partial_t u$ and $X = A \nabla u$; re-arranging (14),

$\langle \nabla \partial_t u, A \nabla u) \rangle = \nabla \cdot (\partial_t u (A \nabla u)) - \partial_t u \nabla \cdot (A \nabla u); \tag{16}$

we integrate (16) over $\Bbb R^3$, as follows:

$\displaystyle \int_{\Bbb R^3} \langle \nabla \partial_t u, A \nabla u \rangle \; dx = \int_{\Bbb R^3} \nabla \cdot (\partial_t u (A \nabla u)) \; dx - \int_{\Bbb R^3} \partial_t u \nabla \cdot (A \nabla u) \; dx; \tag{17}$

we evaluate the first integral on the right of (17) by first taking it over a ball $B(R)$, of radius $R$ and centered at $(0, 0, 0)$, and then letting $R \to \infty$; we have, using the divergence theorem,

$\displaystyle \int_{B(R)} \nabla \cdot (\partial_t u (A \nabla u)) \; dx = \int_{\partial B(R)} \partial_t u (A \nabla u) \cdot n \; dA, \tag{18}$

where $n$ is the outward normal on the sphere $\partial S(R)$, the boundary of $B(R)$, and $dA$ is its area element; under the assumption that $u$ and its derivatives fall off sufficiently rapidly as $R \to \infty$, we have

$\displaystyle \int_{\Bbb R^3} \nabla \cdot (\partial_t u (A \nabla u)) \; dx$ $ = \displaystyle \lim_{R \to \infty} \int_{B(R)} \nabla \cdot (\partial_t u (A \nabla u)) \; dx = \lim_{R \to \infty} \int_{\partial B(R)} \partial_t u (A \nabla u) \cdot n \; dA = 0, \tag{19}$

so that (17) yields

$\displaystyle \int_{\Bbb R^3} \langle \nabla \partial_t u, A \nabla u \rangle \; dx = -\int_{\Bbb R^3} \partial_t u \nabla \cdot (A \nabla u) \; dx; \tag{20}$

using (20), we may re-assemble (13) to find

$\dot E(t) = \displaystyle \int_{\Bbb R^3} (\partial_t u \; \partial_t^2u - \partial u_t \nabla \cdot (A \nabla u)) \; dx = \int_{\Bbb R^3} \partial_t u (\partial_t^2u - \nabla \cdot (A \nabla u)) \; dx = 0, \tag{21}$

by virtue of (1). The conservation of the energy (2) is thus estsblished.

In closing, we note that there is nothing inherently $3$-dimensional about this derivation; everything can carried into $\Bbb R^n$ provided the conditions on $u$ and $\nabla u$ as $R \to \infty$ are appropriately modified.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.