0
$\begingroup$

What does it mean that the equation: $$ \text{div}_{x,y}(y^a\nabla_{x,y}u)=0,\quad \text{in }\mathbb{R}^n\times(0,\infty),$$ is the Euler-Lagrange equation for the functional: $$ J(u)=\int_{\mathbb{R}^n\times (0,\infty)}y^a|\nabla_{x,y}u(x,y)|^2\,dx\,dy?$$ Morover there exist a Banach space on which is defined $J$? I find this terminology in "An extension problem related to the fractional Laplacian" by Caffarelli-Silvestre at page 2.

$\endgroup$

1 Answer 1

1
$\begingroup$

It means that the functional derivative of $J(u)$ is zero (i.e. $J(u)$ has $u$ as a stationary point) if the function $u$ solves the given divergence form PDE, i.e. $$\DeclareMathOperator{\divg}{\mathrm{div}} \DeclareMathOperator{\grad}{\nabla} \bigg{[}\frac{\mathrm{d}}{\mathrm{d}\epsilon}J(u+\epsilon \varphi)\bigg{]}_{\epsilon = 0} =\lim_{\epsilon \to 0}\frac{J(u+\epsilon \varphi)-J(u)}{\epsilon} = 0\iff\divg_{x,y} \big(y^a \grad_{x,y} u\big)=0 $$ for all test function $\varphi\in C^1_c(\Bbb R^n\times [0,+\infty)\big)$, where $$ \bigg{[}\frac{\mathrm{d}}{\mathrm{d}\epsilon}J(u+\epsilon \varphi)\bigg{]}_{\epsilon = 0} = 2\int\limits_{\Bbb R^n \times [0,+\infty)} \!\!\!\!\! y^a\langle\grad_{x,y}u(x,y), \grad_{x,y}\varphi(x,y)\rangle\,\mathrm{d}x\,\mathrm{d}y\label{1}\tag{FD} $$ The given PDE is equivalent to the vanishing of the functional (as a matter of fact, Gâteaux or Frechet) derivative \eqref{1}, provided $u$ is $C^2$-smooth, has a sufficiently smooth boundary value on $y=0$, as can be seen by applying Du Boys-Reymond's lemma: this is classically known and due to Euler and Lagrange, and thus PDEs that come as necessary conditions for the vanishing of the functional derivative of a given (integral) functional are called "Euler-Lagrange equations" of the given functional.

Edit following the comments

As a follow-up of the discussion started on comments on how to prove the uniqueness of the solution of the following Dirichlet problem $$ \begin{cases} \divg_{x,y} \big(y^a \grad_{x,y} u\big)=0\\ u|_{y=0}=u_0\in \mathscr{S}(\Bbb R^n) \end{cases}\label{2}\tag{1} $$ where $\mathscr{S}(\Bbb R^n)$ is the ordinary Schwarz space, it may be noted that the functional $J(u)$ could be defined naturally on the following weighted Sobolev space $$ \begin{split} H^1_{y^a}\big(\Bbb R^n\times [0,+\infty)\big) & \equiv W^{1,2}_{y^a} \big(\Bbb R^n\times [0,+\infty)\big) \\ & \triangleq\Big\{u\in L^2\big(\Bbb R^n\times [0,+\infty)\big) : \|u_{0}\|_{L^2(\Bbb R^n)} + J(u) < +\infty\Big\} \\ \end{split} $$ (here we have put $u_0=u_{y=0}$ in order to simplify the notation) which is naturally Hilbert when equipped with the following inner product $$ \langle u , v \rangle_{H^1_{y^a}}= \langle u_0 , v_0 \rangle_{L^2(\Bbb R^n)} +\int\limits_{\Bbb R^n \times [0,+\infty)} \!\!\!\!\! y^a\langle\grad_{x,y}u(x,y), \grad_{x,y}v(x,y)\rangle\,\mathrm{d}x\,\mathrm{d}y $$ (proving the completeness of this space does not seem to be too difficult). Using this Hilbert space for studying problem \eqref{2} seems to be quite natural, since

  • $\mathscr{S}(\Bbb R^n)\subsetneq L^2(\Bbb R^n)$ so it is possible to efficiently deal with the boundary condition, and
  • $C_c^\infty(\Bbb R^n)\times C_c^\infty\big( \mathbb{R}^n\times[0,\infty)\big)$ is dense in $H^1_{y^a}$ as this spaces explicitly deals with boundary conditions and thus
  • with the Hilbert space structure and the density of it is possible to use well known methods to prove existence and uniqueness of solution $u$.
$\endgroup$
8
  • $\begingroup$ the weighted $W^1_2$ Sobolev space that you have defined is an Hilbert space ? The weighted $L^2$ norm of gradient it seems no to be a norm in $W^1_2$. Moreover, the smooth function with compact support in $\mathbb{R}^n\times [0,\infty)$ is dense in $W^1_2$? Moreover if $u\not\in W^1_2$, $u$ can’t be a critical point of $J$. $\endgroup$
    – inoc
    Dec 16, 2020 at 13:05
  • $\begingroup$ @inoc I apologize for I have put a typo: it should be $u\in L^2(\Bbb R\times [0,+\infty)$. The norm in this Sobolev space should be defined as $$ \|u\|_{W^1_2(g)} = \|u\|_{L^2} + \int\limits_{\Bbb R^n \times [0,+\infty)} \!\!\!\!\!\!g(x,y)|\nabla_{x,y}u(x,y)|^2\,\mathrm{d}x$$ as is it customary in (old?) presentations of the theory of Sobolev spaces, and this is a norm (provided the weight function $g$ 1s positive on $\Bbb R^n\times [0,+\infty)$). $\endgroup$ Dec 16, 2020 at 13:30
  • $\begingroup$ $C_c^\infty( \mathbb{R}^n\times[0,\infty))$ is dense in $W_2^1$? $\endgroup$
    – inoc
    Dec 16, 2020 at 13:41
  • $\begingroup$ As for your other questions, $C_c^1({\Bbb R^n \times [0,+\infty)})$ is not dense in $W_2^1(g)$ since it cannot approximate the boundary values of $u\in W_2^1(g)$. However, density is not required for the Du Bois-Reymond's lemma to hold: indeed it is mandatory that the test functions $\varphi$ vanish on the boundary of the given domain, also for other reasons pertaining to the boundary values you expect from $u$. $\endgroup$ Dec 16, 2020 at 13:53
  • $\begingroup$ @inoc Finally, you are also right saying $u\notin W_2^1$ implies $u$ can't be a critical point of $J$ despite being a solution of the given PDE: this is also true for Laplace's equation $\Delta u=0$. There are harmonic functions that are continuous on the boundary of their domain of definition without being critical point of the associated Dirichlet functions, as the Prym-Hadamard phenomenon shows us: however, the laplacian is nevertheless the Euler-Lagrange equation of the Dirichlet functional. $\endgroup$ Dec 16, 2020 at 14:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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