For $(x,y)^2\in \mathbb{R}^2$, let

$f(x,y)=\begin{cases} [(2x^2-y)(y-x^2)]^{1/4}&x^2\leq y \leq 2x^2\\ 0& \text{otherwise}\\ \end{cases} $

show that all directional derivative of $f$ exist at $(0,0)$, but $f$ is not differentiable at $(0,0)$.

My attempt: Firstly, I observed that the curve become linear when it approaches to zero.

Let $u=(u_1,u_2)\in \mathbb{R}^2$ be a unit vector.

$$D_uf(0,0)=\lim_{t \rightarrow 0}\frac{f(tu_1,tu_2)-f(0,0)}{t}=\lim_{t \rightarrow 0}\frac{0-0}{t}=0.$$

This implies all the directional derivatives of $f$ exist at $(0,0)$.

I want to improve the more justification, why $f(tu_1,tu_2)=0$. I understand with graph of the curve. Can anyone suggest me how I improve my justification in this question.

  • $\begingroup$ Is it possible you miss something in conditions or task itself? $\endgroup$ – zkutch Jul 7 '20 at 6:27
  • $\begingroup$ Which condition? $\endgroup$ – User124356 Jul 7 '20 at 6:33
  • $\begingroup$ Hello likely IU student. If you are preparing for Tier 1, feel free to email me for the resources I used for it. $\endgroup$ – Alfred Yerger Jul 7 '20 at 7:16

Let $g(t)=(t,\frac32t^2)$. Then $g(t)$ is differentiable at $t=0$ and $g(0)=(0,0)$. If $f(x,y)$ were differentiable at $(0,0)$ then its derivative there would be $0$ and by the chain rule: $$\frac{{\rm d}f(g(t))}{{\rm d}t\qquad}\left|_{t=0}\right.=\nabla f(0,0)\cdot g'(0)=0$$

However if you work out $\frac{{\rm d}f(g(t))}{{\rm d}t\quad}\left|_{t=0}\right.$, it will not be $0$.

  • $\begingroup$ This is same as my answer except that you write $1.5$ as $\frac 3 2$. What is the point in repeating an answer. $\endgroup$ – Kavi Rama Murthy Jul 7 '20 at 11:59
  • $\begingroup$ Its a slightly different perspective - exploiting the differential as a linear map, rather than for its implications for the speed of growth at the origin. I found it appealing to write it this way. $\endgroup$ – tkf Jul 7 '20 at 12:22

Hint: The inequalities $t^{2}u^{2}\leq tv \leq 2t^{2} u^{2}$ will automatically fail for $|t|$ sufficiently small and hence $f(tu,tv)=0$ for such $t$. [I will add more details if you cannot justify this].

Since the partial derivatives vanish at the origin the only candidate for the derivative at that point is $0$. So, is $f$is differentiable at $(0,0)$ then we must have $\frac {f(x,y)} {\sqrt {x^{2}+y^{2}}} \to 0$ as $(x,y) \to 0$. Get a contradiction by taking $y=(1.5)x^{2}$ and letting $x \to 0$.

  • $\begingroup$ Yes. I got it. Can you give me some hint for 2nd part. $\endgroup$ – User124356 Jul 7 '20 at 6:34
  • $\begingroup$ @User124356 I have edited my answer. $\endgroup$ – Kavi Rama Murthy Jul 7 '20 at 7:07

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.