Doubts about discontinuous $\to $ existence of derivatives.

81 Views Asked by At

Maybe it's trivial, but I need certaintes about this. Consider

$$f(x, y) = \begin{cases} \dfrac{2xy^2}{x^2+y^2} & (x, y) \neq (0, 0) \\\\ 0 & (x, y) = (0, 0) \end{cases}$$

I showed that the function is not continuous at the origin. Indeed by restricting to the straight lines though the origin $y = mx$ I get $0$ as a limit ($x\to 0)$.

Instead, taking the sequences $(x_n, y_n) = \left(\frac{1}{n^2}, \frac{1}{n}\right)$, I observed that

$$\lim_{n\to +\infty} f(x_n, y_n) = \frac{2}{3}$$

Hence the statement.

Besides this, can I still talk about existence of the derivatives? I'm asking because I managed to show that by using the definition:

$$f'_x(0, 0) = 0$$ $$f'_y(0, 0) = 0$$

Hence the derivatives exist everywhere, but they are not continuous since

$$\lim_{(x, y) \to (0, 0)} f'_x (x, y) = -2m^2$$

when restricting to $y = mx$ again, hence the limit does not exist. Similarly for $f'_y$.

After that, I also showed that $f(x, y)$ is not differentiable at $(0, 0)$, indeed

$$\lim_{(h, k)\to (0, 0)} \frac{f(h, k) - f(0, 0) - h\cdot f'_x(0, 0) - k\cdot f'_y(0, 0)}{\sqrt{h^2 +k^2}} = \sqrt{2}$$

  • Question: is this all right? Did I make any error?

  • Can I conclude that even if $f$ is not continuous at a point, its partial derivatives (along the axis) could exist (everywhere)?

1

There are 1 best solutions below

4
On

$f(x, y) = \begin{cases} \dfrac{2xy^2}{x^2+y^2} & (x, y) \neq (0, 0) \\\\ 0 & (x, y) = (0, 0) \end{cases}$

is continuous everywhere.

Indeed by restricting to the straight lines though the origin $y=mx$ I get $0$ as a limit $(x\to 0)$.

Why this contradict that the limit doesn't exist?

If the limit exists then it would be independent of path.To prove non existence of a limit, we choose a path(cleverly) such that the limit depends on the parameter because for two distinct values of parameter we get two different limit.

For an example : $f(x, y) = \begin{cases} \dfrac{2y^2}{x^2+y^2} & (x, y) \neq (0, 0) \\\\ 0 & (x, y) = (0, 0) \end{cases}$

Now choose the path $y=mx$

Then $\lim_{x\to 0}f(x, mx) =\frac{m^2}{1+m^2}$

So along two different path $y=x$ and $y=2x$ we get two different value, so the limit at the origin doesn't exist.

Note: In case of your example, you haven't proved the dependency of limit on the parameter.

$(x_n, y_n) = \left(\frac{1}{n^2}, \frac{1}{n}\right)$ $\lim_{n\to +\infty} f(x_n, y_n) = \frac{2}{3} $

Again wrong.

$f(x_n, y_n) =\frac{1}{n^2+1}\to 0 =f(0,0)$


Both the above approach can't discard the discontinuity of $f$ at $(0, 0) $.

Infact $f$ is continuous at $(0, 0) $

$\begin{align}|f(x, y) -f(0, 0) |&=\left|\frac{2xy^2}{x^2+y^2}\right|\\&\le 2|x|\\&\le 2\sqrt{x^2+y^2}\end{align}$

Hence $\forall \epsilon>0$ , choose $\delta \in(0, \frac{\epsilon}{2}) $ such that $\forall (x, y) \in N_{\delta}((0,0))$ we have $|f(x, y) -f(0, 0) |<\epsilon$


Are the partial derivatives exist at $(0,0)$?

We can prove more than that "directional derivatives of $f$ at $(0, 0) $ along the direction of every unit vector exists and $0$"

Proof:

Let $u=(u_1, u_2) $ be a unit vector i.e $\|u\|=\sqrt{a^2+b^2}=1$

Then $\begin{align}D_{u}f(0, 0) &=\lim_{t\to 0}\frac{ f(tu_1, tu_2) -f(0, 0) }{t}\\&=\lim_{t\to 0} \frac{2tuv^2}{u^2+v^2}\\&=0\end{align}$

Let $e_1=(1, 0) $ and $e_2=(0,1)$ , then $\frac{\partial{f}}{\partial{x}}=D_{e_1}f(0, 0)=0$ and $\frac{\partial{f}}{\partial{y}}=D_{e_2}f(0, 0)=0$


I'm asking because I managed to show that by using the definition: $f′_x(0,0)=0$ and $f′_y(0,0)=0$

Hence the derivatives exist everywhere

This is again false. Existence of partial derivatives (even directional derivatives along every unit vector) at a point doesn't imply differentiability at that point.

Your example is good to enough to emphasize that fact.


Is $f$ differentiable at $(0, 0) $?

Ans: No

Proof: $\begin{align}&\frac{f(h,k)-f(0,0)-h\frac{\partial{f}}{\partial{x}}(0,0)-k\frac{\partial{f}}{\partial{y}}(0,0)}{\sqrt{h^2+k^2}}\\&=\frac{2hk^2}{(h^2+k^2)^{\frac{3}{2}}}\end{align}$

$\lim_{(h, k) \to (0, 0)} \frac{2hk^2}{(h^2+k^2)^{\frac{3}{2}}}$ doesn't exists.

Choose the path $k=mh$.