I'm trying to solve the following exercise:

Let $f:\mathbb R^2 \to \mathbb R$ a continuous function whose partials exist everywhere in $\mathbb R^2$. Show that $f$ is ($\mathbb R$-)differentiable when $\frac{\partial f}{\partial x}$ is continuous.

I know from a theorem that $f: \mathbb R^n \to \mathbb R$ is differentiable if every partial is continuous. Now it seems to be the case that I only need one partial to be continuous.

My attempt was at first to show that $\frac{\partial f}{\partial y}$ is continuous, however I don't know if that is necessary or even possible (since $f$ can be differentiable without both partials being continuous).

My second attempt was to show that a $\mathbb R$-linear function $L$ exists with $$\lim\limits_{h\to 0}\frac{f(\zeta + h) - f(\zeta) - L(h)}{||h||_2} =0$$ Since this function must be $\nabla f$, I proved $\nabla f$ is linear and now want to show that $$\lim\limits_{h\to 0}\frac{f(\zeta + h) - f(\zeta) -(\frac{\partial f(\zeta)}{\partial x}h_1+\frac{\partial f(\zeta)}{\partial y}h_2)}{||h||_2} =0$$

This is where I am stuck, I doubt that this is the correct way to prove it but don't know what to do else. Would appreciate some help.

Edit: This is what I've tried to do next with help of Davids hints: Define $$\nabla f(0) = (\frac{\partial f}{\partial x}(0), \frac{\partial f}{\partial y}(0)) =: (\alpha, \beta)$$ It now holds that
$$\begin{align*} & f(h) - f(0) - (\alpha h_1 + \beta h_2) \\ &=f(h) - f(h_2 e_2)) - \alpha h_1 + f(h_2 e_2) - f(0) - \beta h_2\end{align*}$$ If we plug that into the definition we get $$\lim \limits_{h\to0}\frac{f(h) - f(h_2 e_2) - \alpha h_1 + f(h_2 e_2) - f(0) - \beta h_2}{||h||_2} \\ =\lim \limits_{h\to0}\frac{f(h) - f(h_2 e_2) - \alpha h_1}{||h||_2} + \lim \limits_{h\to0}\frac{f(h_2 e_2) - f(0) - \beta h_2}{||h||_2} \\ =\lim \limits_{h\to0}\frac{f(h) - f(h_2 e_2)}{||h||_2} - \lim \limits_{h\to0}\frac{\alpha h_1}{||h||_2} + \lim \limits_{h\to0}\frac{f(h_2 e_2) - f(0)}{||h||_2}- \lim \limits_{h\to0}\frac{\alpha h_1}{||h||_2} \\ =\lim \limits_{h\to0}\frac{f(h) - f(h_2 e_2)}{||h||_2} - \lim \limits_{h\to0}\frac{\alpha h_1}{||h||_2} + \frac{\partial f}{\partial y}(0)- \lim \limits_{h\to0}\frac{\alpha h_1}{||h||_2}$$ This is where I am stuck. I don't know the $\alpha$ and $\beta$ terms tend to 0 for $h\to0$ or not and how to use the continuity of $\partial_x f$.


Hint: $$f(\zeta+h)-f(\zeta)=[f(\zeta+h)-f(\zeta+(0,h_2)]+[f(\zeta+(0,h_2))-f(\zeta)].$$The continuity of $\partial f/\partial x$ lets you handle $f(\zeta+h)-f(\zeta+(0,h_2)$, and then you can use existence of $\partial f(\zeta)/\partial y$ for $f(\zeta+(0,h_2))-f(\zeta)$.

Slightly more detailed hint: Say $\nabla f(\zeta)=(\alpha,\beta)$. The triangle inequality shows that

$$|f(\zeta+h)-f(\zeta)-(\alpha h_1+\beta h_2)|\le |f(\zeta+h)-f(\zeta+(0,h_2)-\alpha h_1|+|f(\zeta+(0,h_2))-f(\zeta)-\beta h_2|.$$

Giving up, posting solution: I see you've switched to $\zeta=0$, great. So let $\nabla f(0)=(\alpha,\beta)$.

The definition of $\partial f(0)/\partial y=\beta$ says $$\lim_{h_2\to0}\frac{f(0,h_2)-f(0)}{h_2}=\beta, $$or $$\lim_{h_2\to0}\frac{f(0,h_2)-f(0)-\beta h_2}{h_2}=0. $$ Since $||h||\ge |h_2|$ this shows that $$\lim_{h\to0}\frac{f(0,h_2)-f(0)-\beta h_2}{||h||}=0. $$ Note that only used the existence of $\partial f/\partial y$ at the origin.

Now let $\epsilon>0$. Choose $\delta>0$ so $$\left|\frac{\partial f}{\partial x}(x,y)-\alpha\right|<\epsilon\quad(||(x,y)||<\delta).$$ Suppose $||h||<\delta$ and $h_1\ne0$. Just to make the notation work out, assume $h_1>0$; the case $h_1<0$ is similar. The Mean Value Theorem shows that there exists $t\in(0,h_1)$ with $$\frac{f(h)-f(0,h_2)}{h_1}=\frac{\partial f}{\partial x}(t,h_2).$$Since $||(t,h_2)||<\delta$ this shows that $$\left|\frac{f(h)-f(0,h_2)}{h_1}-\alpha\right|<\epsilon.$$So $$\frac{|f(h)-f(0,h_2)-\alpha h_1|}{||h||} \le \frac{|f(h)-f(0,h_2)-\alpha h_1|}{|h_1|}<\epsilon.$$

Now you're done; the triangle inequality shows that if $||h||$ is small enough then $$\frac{|f(h)-f(0)-\alpha h_1-\beta h_2|}{||h||}<2\epsilon.$$

  • $\begingroup$ Did you confuse signs in your equation? Why is it $-f(\zeta - h)$ and not $+f(\zeta - h)$? $\endgroup$ – Staki42 Jun 29 '16 at 15:01
  • $\begingroup$ @lappen68 Aargh. Yes, it's all screwed up. Thanks. $\endgroup$ – David C. Ullrich Jun 29 '16 at 15:06
  • $\begingroup$ You're welcome. I'll try using what you said. $\endgroup$ – Staki42 Jun 29 '16 at 15:10
  • $\begingroup$ @lappen68 I think I have it straight now. Anyway it sounds like you got the idea... $\endgroup$ – David C. Ullrich Jun 29 '16 at 15:15
  • $\begingroup$ I tried to do some stuff with your hint now, but I can't seem to find the right answer. Do I even have to use my $\nabla f$ being the $\mathbb R$-linear function or do I follow the wrong approach? $\endgroup$ – Staki42 Jun 29 '16 at 18:50

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.