(Chain rule) Assume $F : \mathbb{R} \to \mathbb{R}$ is $C^1$, with $F'$ bounded. Suppose $U$ is bounded and $u \in W^{1,p}(U)$ for some $1 \le p \le \infty$. Show $$v :=F(u) \in W^{1,p}(U) \quad \text{and} \quad v_{x_i}=F'(u)u_{x_i}.$$

From PDE Evans, 2nd edition: Chapter 5, Exercise 17.

Here is what I understand conceptually so far:

Since $u \in W^{1,p}(U)$, it follows $Du=u'$ exists, with $$\int_U u \phi' dx = -\int_U Du \phi \, dx.$$

I need to show that $D(F(u))=F'(u)Du$ exists, with $$\int_U F(u) \phi' dx = -\int_U D(F(u)) \phi \, dx.$$ Then, I can conclude that $F(u) \in W^{1,p}(U)$.

This is all I know so far; how can I go about making the connection?


2 Answers 2


Assume first that $1\leq p <\infty$

If $u\in C^\infty(\bar{U})$ then clearly $v=F(u)\in C^1(\bar{U})$ and $\nabla v=F'(u)\nabla u$.

Now if $u$ is a general $W^{1,p}$ function then take a sequence $u_k \to u$ in $W^{1,p}$ with $u_k\in C^\infty(\bar{U})$ and such that $u_k\to u$ and $\nabla u_k \to \nabla u$ pointwise a.e. in $U$. Then $$ |F(u)-F(u_k)| \leq M|u-u_k|, $$ with $M=\| F'\|_\infty$. On the other hand we also have $$ F'(u_k(x))\nabla u_k(x) \to F'(u(x))\nabla u(x), \qquad \text{ for a.e. } x\in U, $$ and moreover $| F'(u_k)\nabla u_k|\leq M|\nabla u_k|$. Since $\nabla u_k \to \nabla u$ in $L^p$, by the dominated convegence theorem, $F'(u_k)\nabla u_k \to F'(u)\nabla u$ in $L^p(U)$. Combining this with the first estimate we get that $F(u)\in W^{1,p}(U)$ and $\nabla F(u)= F'(u)\nabla u$.

If $p=\infty$ then we can simply note that $W^{1,\infty}(U)$ is the space of Lipschitz continuous functions in $U$, so take $u$ a Lipschitz function with Lipschitz constant $N$, and $M$ the Lipschitz constant of $F$ as before, then $$ |F(u(x))-F(u(y))|\leq M|u(x)-u(y)| \leq NM|x-y|, \qquad \forall x,\ y\in U. $$ Therefore $F(u)\in W^{1,\infty}(U)$.

Edit: As an extra exercise try to see that the condition $U$ being a $C^1$ domain is not needed when $1\leq p<\infty$ (try to prove that $F(u)\in W^{1,p}(U)$ whenever $u\in C^\infty(U)\cap W^{1,p}(U)$).

  • 2
    $\begingroup$ I understand how we have $Du_k \to Du$ in $L^p(U)$. Can you elaborate on how the dominated convergence theorem suggests $F'(u_k) Du_k \to F'(u) Du$ in $L^p(U)$? The DCT is suggesting (in an appendix section of my textbook) that, since $F'(u_k) \to F(u)$ a.e. and $|F'(u_k)| \le M|Du_k|$ a.e., we have $$\int_U F_k(u) \, dx \to \int_U F(u) \, dx.$$ Perhaps this implies $F'(u_k)Du_k \to F'(u) Du$ in $L^p(U)$ as you wrote? $\endgroup$
    – Cookie
    Jan 20, 2015 at 8:23
  • $\begingroup$ It's a variant of the DCT that says $|h_k|\leq |H_k|$, $h_k\to h$ a.e. and $H_k\to H$ in $L^p$, then $h_k\to h$ in $L^p$. In this case $h_k=F'(u_k)\nabla u_k$ and $H_k=M\nabla u_k$. The "first estimate" refers to $|F(u_k)-F(u)|\leq M|u_u-u|$, which ensures that $F(u_k)\to F(u)$ in $L^p$. $\endgroup$
    – Jose27
    Jan 20, 2015 at 9:07
  • $\begingroup$ When I try to combine $F'(u_k) Du_k \to F'(u) Du$ to $|F(u)-F(u_k)| \le M|u-u_k|$, how exactly can I manipulate the estimate (and finally conclude that $F(u) \in W^{1,p}(U)$ for finite $p$)? The infinite case is a lot easier to understand since $W^{1,\infty}$ contains only Lipschitz continuous functions. $\endgroup$
    – Cookie
    Jan 20, 2015 at 22:46
  • $\begingroup$ Well, you have that $v_k\to v$ and $\nabla v_k \to w$ in $L^p$. It's an easy exercise to prove that in this situation $v\in W^{1,p}$ and $w=\nabla v$ (obviously in this case $v=F(u)$). $\endgroup$
    – Jose27
    Jan 21, 2015 at 0:40

As a modification to Jose27 I would argue that you may suppose U (path) connected (otherwise we work on connected components separately). Then you approch $\nabla u$ by $C^\infty$ functions and "take primitives" by some kind of path integral. That way you ensure $u_k \to u$ pointwise and $\nabla u_k \to \nabla u$ in $L_p$ norm right from the construction.

Jose27's last comment is actually the fact that $\nabla$ is a closed 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.