I just read a theorem on my class notes that states:

Let $f:\mathbb R^2\rightarrow \mathbb R$ be a function and $(x_0, y_0) \in \mathbb R²$ a point such that the partial derivates exist on a ball of center $(x_0,y_0)$ and both are continuous on $(x_0,y_0)$. This implies $f$ differentiable on $(x_0,y_0)$.

This raised some questions in me.

Consider the function $ f(x,y) = \begin{cases} 1 & \text{if $x=0$ or $y=0$} \\ 0 & \text{elsewere} \end{cases}$

I know differentiability on a point implies continuity on that point. But here I have that partial derivates exist and are continuous on a ball around $(0,0)$, and by theorem that would implied differentiability, but that would implied continuity and this function is not continuous on $(0,0)$.

Then consider the function $ g(x,y) = \begin{cases} |x| & \text{if $y=0$} \\ |y| & \text{if $x=0$} \\ 0 & \text{elsewere} \end{cases}$

This function is continuous on $(0,0)$, but partial derivates does not exist (because of the absolute value over the axis) so I can not use the above theorem. Now imagine the function $h$ results from rotating $g$ 45 degrees (so the absolute values would be on the lines of the domain $y=x$ and $y=-x$. The partial derivates of $h$ does exist, and are continuous on a ball around $(0,0)$. This implies by above theorem that $h$ is differentiable on $(0,0)$.

On the other hand, $h$ and $g$ are practically the same function. Why do I need partial derivates instead of just 2 different arbitrary directional derivates to ensure differentiability?

Now consider the function $ i(x,y) = \begin{cases} x² & \text{if $y=0$} \\ y² & \text{if $x=0$} \\ 1 & \text{elsewere} \end{cases}$

This function does have partial derivates and they are continuous around $(0,0)$, so that would implied differentiability but, again this function is not continuous on $(0,0$).

I absolutely convinced I'm reasoning something wrong (maybe a lot of things) but I can't see my own misconceptions, so I would appreciate some help. What I'm doing wrong? Or in the unlikely case that the theorem is wrong, why?

  • $\begingroup$ I think your main misunderstanding relates to the existence & continuity of the partial derivatives on a ball, rather than at the point. For your first function, what is $\frac{\partial f}{\partial x}\rvert_{(0, \epsilon)}$? $\endgroup$ – Sten Nov 15 '19 at 11:51
  • $\begingroup$ $\frac{\partial f}{\partial x}\rvert_{(0, \epsilon)} $would be 0? Since $f$ is constant over the axis, it does not matter how close I am the directional derivate always gives 0 since fixing $y$ let us with a single variable function that is constant. Is this wrong? $\endgroup$ – xtreyreader Nov 15 '19 at 12:00
  • $\begingroup$ I think I got it. I was thinking about fixing the $y$ variable like forcing it to be zero, while instead i must treating it like a constant but it can be whatever value I want. Understood after your answer and a gif, so thanks. $\endgroup$ – xtreyreader Nov 15 '19 at 12:16

Your main misunderstanding relates to the existence & continuity of the partial derivatives on a ball, rather than at the point. I'll look at your first function because the other cases are similar.

Let's evaluate the derivative $\frac{\partial f}{\partial x}$ at some point just off the origin, but on the $y$-axis. From the definition of the partial derivative, we have

$$\left.\frac{\partial f}{\partial x}\right\rvert_{(0, \varepsilon)} = \lim_{h \to 0} \frac{f(h, \varepsilon) - f(0, \varepsilon)}{h} = \lim_{h\to 0} \frac{0 - 1}{h} = \lim_{h\to 0} \frac{-1}{h}$$

And this limit does not exist, no matter what non-zero value of epsilon we picked. So the partial derivatives of $f$ do not exist on the entirety of any ball around (0, 0).

| cite | improve this answer | |

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.