During my analysis course my teacher explained the difference between partial derivatives and directional derivatives using the notion that a partial derivatives looks at the function as approaching a point along the axes (in case of of the plane), and a directional derivative as approaching a point from any direction in the plane. He also explained that the existence of directional derivatives is a stronger notion than the existence of partial derivatives exists: if all directional derivatives exist, then the partial derivatives exist too.

I am to show (not necessarily prove) a case where the partial derivatives exist, but not all directional derivatives exist (hence, f is not differentiable).

  • $\begingroup$ note that even if all directional derivatives exists, $f$ need not to be differentiable :-) $\endgroup$ – Ant Apr 19 '16 at 15:32
  • $\begingroup$ Thank you. I'm meant to show this in a later exercise too. For now I only want to show an example I think. I can't come up with a function, however. $\endgroup$ – falidoro Apr 19 '16 at 15:35

Consider for instance $$f(x,y)=\begin{cases} \frac{xy}{x^2+y^2}&\text{if }(x,y)\neq(0,0),\\ 0&\text{otherwise.}\end{cases}$$ The partial derivatives exist everywhere. Away from the origin this is clear. At the origin we need to calculate the partials from first principles: $$\frac{f(0+h,0)-f(0,0)}h=\frac{0-0}h=0$$ so letting $h\to0$ we find $f_x(0,0)=0$. By symmetry we also have $f_y(0,0)=0$. However, not all directional derivatives exist at the origin. For example, let $\mathbf v=(\frac1{\sqrt{2}},\frac1{\sqrt{2}})$. Then $$\frac{f(\mathbf0+t\mathbf v)-f(\mathbf0)}{t}=\frac{\frac{t^2}{t^2+t^2}-0}t=\frac1{2t}$$ which does not have a limit as $t\to0$.

| cite | improve this answer | |
  • $\begingroup$ Thanks a lot! Do you happen to know a systematic way of finding such a function with this property, or is it usually found as an educated guess? $\endgroup$ – falidoro Apr 19 '16 at 18:17
  • 1
    $\begingroup$ Usually functions of the form $f(x,y)=\frac{x^jy^k}{(x^2+y^2)^n}$ for $(x,y)\neq(0,0)$ are good for these kinds of questions. Tweak the exponents depending on whether you're looking at continuity or differentiability. $\endgroup$ – Jason Apr 19 '16 at 18:40

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.