Finding the weak derivatives:

a/ $f(x)=\left|x_1 \right|$, for all $x=(x_1,\ldots, x_n)$;

b/ $f(x)=\operatorname{sign} (x_1)$, where $\Omega=\{\left|x \right|<1\}$.


We know that $f$ has weak derivative $\partial^\alpha f=g$ if $$\int_{\Omega}g \varphi \ \mathrm{d}x=(-1)^\left|\alpha \right|\int_{\Omega}f\cdot D^\alpha \varphi \ \mathrm{d}x, \ \forall \varphi \in C_{0}^{\infty}(\Omega)$$


For Question a/ We assume that $g$ is the weak devirative of $f$.

  • We have $I:=\int_{-1}^{1}g\varphi \ \mathrm{d}x=-\int_{-1}^{1}|x_1|D\varphi \ \mathrm{d}x_1, \ \forall \varphi \in C_{0}^{\infty}(\Omega) \tag 1$

  • Whence $-I=\int_{-1}^{0}-x_1 D\varphi \ \mathrm{d}x_1+\int_{0}^{1}x_1 D\varphi \ \mathrm{d}x_1=-x_1\varphi \mid_{-1}^{0}+\int_{-1}^{0}\varphi \ \mathrm{d}x_1+x_1 \varphi \mid_{0}^{1}-\int_{0}^{1}\varphi \ \mathrm{d}x_1$

  • Hence, $$\begin{align*} -I&=\varphi(1)-\varphi(-1)-\left(\int_{0}^{1}\varphi \ \mathrm{d}x_1-\int_{-1}^{0}\varphi \ \mathrm{d}x_1 \right)\\ &=\varphi(1)-\varphi(-1)-\int_{-1}^{1}\operatorname{sign} (x_1) \varphi \ \mathrm{d}x_1 \\ &=-\int_{-1}^{1}\operatorname{sign} (x_1) \varphi \ \mathrm{d}x_1 \tag 2 \end{align*} $$

Since (1) and (2), we have $g(x)=\operatorname{sign} (x_1) \blacksquare$


Now, For Question b/ I have stuck when I trying to show that $\not \exists$$g$ is the weak devirative of $f$.

Can anyone solve it? Thanks!


For (b), just start as in (a), for $\phi \in C^\infty_c(\Omega)$, for $n=1$ (you do (a) also for $n=1$ and one can see the basic idea in this case) \begin{align*}\def\s{\mathop{\rm sign}}\def\abs#1{\left|#1\right|} -\int_{\Omega} \s{x_1}\cdot\phi'\, dx &= -\int_{-1}^1\s x_1\, \phi'\, dx_1\\ &= \int_{-1}^0 \phi'\, dx_1 - \int_0^1\phi'\, dx_1\\ &= -\phi(-1) + \phi(0) - \phi(1) + \phi(0)\\ &= 2\phi(0) \end{align*} So we are asked if there is any $g \in L^1_{\rm loc}(\Omega)$ such that, $$ \int_{-1}^1 g\phi\, dx = 2\phi(0), \quad \phi \in C^\infty_c(\Omega) $$ Suppose there were, choose $\phi_n \in C^\infty_c(\Omega)$ such that $0 \le \phi_n \le 1$, $\phi_n$ is supported in $[-\frac 1n, \frac 1n]$ and $\phi_n(0) = 1$. Now note that $g\phi_n \to 0$ almost everywhere, hence by dominated convergence, we have $$ 2 = 2\phi_n(0) = \int_{-1}^1 g\phi_n \, dx \to 0. $$ This contradiction shows that such a $g$ cannot exist.

  • $\begingroup$ Yes, I understand your ANS. Thanks! martini. $\endgroup$ – kimtahe6 Oct 25 '13 at 13:59
  • $\begingroup$ But I'm wondering Why did you have $g \phi_n \to 0$ almost everywhere? $\endgroup$ – kimtahe6 Oct 25 '13 at 14:13
  • $\begingroup$ We have $\phi_n $ to 0 almost everywhere as the support shrinks to $\{0\}$. $\endgroup$ – martini Oct 26 '13 at 7:40

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.