Is one continuous partial derivative and other existing enough to imply diffrentiability of $R^2 \to R$ function

47 Views Asked by At

For $f: R^2 \to R$ to be diffrentiable in point $p = (p_x;p_y)$ it has to satisfy for aribitrary $\theta$

$$1)\space\space\lim_{r \to 0} \frac{f(p_x + rcos(\theta);p_y + rsin(\theta)) - f(p_x;p_y) - \cfrac{df}{dx}(p_x;p_y)rcos(\theta) - \cfrac{df}{dy}(p_x;p_y)rsin(\theta)}{r} = 0$$

We can estimate $f(p_x + rcos(\theta);p_y + rsin(\theta))$ using value $f(p_x;p_y)$ and partial derivatives of $f$ in a following way.

$$2)\space\space f(p_x + rcos(\theta);p_y + rsin(\theta)) = f(p_x;p_y) + \frac{df}{dx}(p_x;p_y)rcos(\theta) + \frac{df}{dy}(p_x + rcos(\theta);p_y)rsin(\theta) + E_x + E_y$$ Where $E_x$ and $E_y$ are errors of approximation and based on definition of derivative $$\lim_{r \to 0}{\frac{E_x}{r}} =0 $$ and $$\lim_{r \to 0}{\frac{E_y}{r}} =0 $$ Substituting 2 to 1 and I end up with $$\lim_{r \to 0} \frac{rsin(\theta)(\cfrac{df}{dy}(p_x+rcos(\theta);p_y)-\cfrac{df}{dy}(p_x;p_y))}{r} + \lim_{r \to 0}{\frac{E_x}{r}} + \lim_{r \to 0}{\frac{E_y}{r}}$$ Now if $\frac{df}{dy}$ is continuous at p the first limes is 0 and limit of whole expression is also 0 implying $f$ is frechet diffrentiable at p. However I only used continuity of one partial derivative (other just has to exist), while known theorem states that all partial derivatives should be continous to imply diffrentiability. Do I make a mistake somewhere along the way? It would be great if someone could provide couterexample of non diffrentiable function with one continous partial derivative. Thank you in advance.

1

There are 1 best solutions below

3
On BEST ANSWER

No, you're not making a mistake

It is a general result that if $f: \mathbb R^n \to \mathbb R$ has only $n-1$ continuous partial derivatives at $x \in \mathbb R^n$, then $f$ is Fréchet differentiable at $x$. Unfortunately this stronger result is rarely mentioned in the litterature.