Suppose $\Omega$ is an open set in $\Bbb{R}^N$ and $\sigma : \Omega \to \Bbb{R}^N$ is a field with all components belonging to $L^2(\Omega)$.

We say that $\sigma$ has weak divergence if there exists a function $w \in L^2(\Omega)$ such that for all $\varphi \in C_c^\infty (\Omega)$ we have

$$ \int_\Omega \sigma \cdot \nabla \varphi=-\int_\Omega w \varphi. $$

My question is:

Can we establish a result of the form: if $\sigma$ has weak divergence then each component of $\sigma$ is weakly differentiable?

The idea is that I've seen this technique in proving that if a function $u$ is $H^1(\Omega)$ and it satisfies some convenient weak condition then $\nabla u$ has a weak divergence and therefore $u \in H^2(\Omega)$. The book where I've seen this technique is aimed for engineers, and therefore it is not very rigorous. That is why I've asked this question.

  • 3
    $\begingroup$ If I understood your comment at the end what was actually shown was that a function $u\in H^1(\Omega)$ satisfied $\Delta u=f\in L^2(\Omega)$ weakly. This is a standard consequence of "elliptic regularity". What you are asking can't possibly be true since you have no control over most of the derivatives of each component of $\sigma$. $\endgroup$
    – Rbega
    Oct 23, 2012 at 17:14
  • $\begingroup$ That is what I was thinking, but I wanted another confirmation. $\endgroup$ Oct 23, 2012 at 17:35
  • 1
    $\begingroup$ as Rbega says: if it is weakly differentiable (i.e., $\sigma\in H^1(\Omega)$, then it clearly also has a weak divergence, but (generally) not viceversa. i do not even see why elliptic regularity applies here. what one can generally say is that if $u\in H^1(\Omega)$ is a weak solution of $\Delta u=f\in L^2(\Omega)$, then $u\in H^2_{loc}(\Omega)$. $\endgroup$ Oct 23, 2012 at 19:25

1 Answer 1


So $\sigma=\sum_{1\le j\le n}\sigma_j(x)\frac{\partial}{\partial x_j}$ is a vector field with distributions coefficients $\sigma_j$ and divergence in $L^2$: $$ \sum_{1\le j\le n}\frac{\partial \sigma_j}{\partial x_j}\in L^2. $$ If I understand your question correctly, you ask if this implies that each $\sigma_j$ belongs to $L^2$. Of course not since you can choose all $\sigma_j=0$ for $j\ge 2$ and $\sigma_1$ to be any distribution in the variables $(x_2,\dots, x_n)$.

Note nevertheless that when $n=2$ and $\text{div} \sigma=0$, then there exists a function $\psi(x_1,x_2)$ such that $$ \sigma_1=\frac{\partial \psi}{\partial x_2},\quad \sigma_2=-\frac{\partial \psi}{\partial x_1}. $$ If $\sigma$ happens to be continuous (resp. locally $L^2$), then $\psi$ is locally Lipschitz continuous (resp. locally $H^1)$.

For the general statement that you mention, note that the gradient operator is elliptic, which is not the case of the divergence operator.


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.