I know that one can prove that for $u\in H^1(-1,1)$, the following equality hods: $$ u(x)=u(-1)+\int^x_{-1}u'(\tilde{x})d\tilde{x},\quad x\in (-1,1) $$

But can we obtain a kind of Newton-Leibniz formula over weak partial derivative in high dimension?

For an example, let $\Omega=[-1,1]\times [-1,1]\subset\mathbb{R}^2$, and $u\in H^2(\Omega)$. By embedding theorem, we have $u\in C(\Omega)$. Can we have following equality: $$ \begin{aligned} u(x,y)&=\int_{-1}^x\int_{-1}^yu_{xy}(\tilde{x},\tilde{y})d\tilde{x}\tilde{y}+\int_{-1}^xu_{x}(\tilde{x},-1)d\tilde{x}\\&+\int_{-1}^yu_y(-1,\tilde{y})d\tilde{y}+u(-1,-1) \end{aligned} $$ My try is: By Fubini's theorem, one can obtain that $\int_{-1}^1(u_{x}(\tilde{x},y))^2d\tilde{x}$ and $\int_{-1}^1(u(\tilde{x},y))^2d\tilde{x}$ is finite for almost every $y$ with $x\in (-1,1)$. Can we say that for almost every $y$, we have $u(x,y)\in H^1(-1,1)$ and then $$ u(x,y)=\int_{-1}^xu_{x}(\tilde{x},y)d\tilde{x}+u(-1,y) $$ If it holds, we can do the same on $u_x(\tilde{x},y)$ with respect to $y$. But how to deal with $u(-1,y)$ or how to prove $u(-1,y)\in H^1(-1,1)$?


In the one-dimensional case, it matters that the evaluation map $u\mapsto u(x_0, y_0)$ is continuous from $H^1$ to $\mathbb{R}$, for any fixed $(x_0, y_0)$. Think of this as the restriction of $u$ from $1$ dimension to $0$ dimensions.

In the two-dimensional case we also have the continuity of $u\mapsto u(x_0, y_0)$ from $H^2$ to $\mathbb{R}$. But we also need to restrict $u$ to one-dimensional subsets (line segments). This is what the Trace operator is for. The standard trace theorem for Sobolev functions says that Sobolev functions can be restricted to a smooth hypersurface of codimension $1$ with the loss of $1$ derivative: that is, from $H^1(\Omega)$ to $L^2(\partial \Omega)$, or (by applying the previous result to $\nabla u$) from $H^2(\Omega)$ to $H^1(\partial \Omega)$,

So, given $u\in H^2$ (with the canonical continuous representative), we get $u(\cdot, y_0)\in H^1$ and $u(x_0, \cdot)\in H^1$ for every $x_0$ and every $y_0$. (Not for "almost every", this is not a Fubini-type theorem.)

As a consequence, every term in the equation $$\begin{aligned} u(x,y)&=\int_{-1}^x\int_{-1}^yu_{xy}(\tilde{x},\tilde{y})d\tilde{x}\tilde{y}+\int_{-1}^xu_{x}(\tilde{x},-1)d\tilde{x}\\&+\int_{-1}^yu_y(-1,\tilde{y})d\tilde{y}+u(-1,-1) \end{aligned} \tag1 $$ is a continuous functional on $H^2$. Since (1) holds for smooth functions, and smooth functions are dense in $H^2$, it follows that (1) holds on all of $H^2$.

  • $\begingroup$ Thank you very much for your answer! It helps me a lot, I never thought trace theorem can be used here. $\endgroup$ – whereamI Jun 20 '18 at 11:29

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.