Let $f:\mathbb{R}^2 \rightarrow \mathbb{R}$ and $a \in \mathbb{R}^2$. Suppose that all $D_v (a)$ exist and that $D_{(v_1, v_2)} f (a)= 5v_1 ^2 + v_2$. Is $f$ differentiable in $a$?

I think it is differentiable. The only way I can think of solving this problem is by using the sufficient condition of differentiability:

Let $A$ be open in $\mathbb{R}^n$, $a \in A$ and $f:A \rightarrow \mathbb{R}^m$. If the partial derivatives $D_1 f(x),...,D_n f(x)$ exist in an open ball $B(a,r)$ and they are all continuous in $a$ then $f$ is differentiable in $a$.

So the values of the partial derivatives in $a$ are $D_1 f (a)= 5 \cdot 1^2+0$ and $D_2 f (a)= 5 \cdot 0^2+1 = 1$ and they exist because all directional derivatives exist by hypothesis. But I don't know how to conclude that the partial derivatives are continuous in $a$ since I don't know their general expression, only their value in $a$.

  • $\begingroup$ Hint: Examine the directional derivatives in the directions of $(1,0)$ and $(-1,0)$. $\endgroup$ – amd Aug 29 '18 at 21:05
  • $\begingroup$ @amd How can we conclude from that? $\endgroup$ – gimusi Aug 29 '18 at 21:14
  • 1
    $\begingroup$ @gimusi What do those values mean for a potential tangent plane at $a$? $\endgroup$ – amd Aug 29 '18 at 21:17
  • $\begingroup$ @amd Ah ok yes of course! We have that $D_{(v_1, v_2)} f (a)$ should be linear in $v_1$ and $v_2$ if f were differentiable. $\endgroup$ – gimusi Aug 29 '18 at 21:21
  • 1
    $\begingroup$ Not quite. You need $D_{-v}f(a) = -D_vf(a)$. More generally, a necessary condition for differentiability is that $D_v(a) = \nabla f(a)\cdot v$. $\endgroup$ – amd Aug 29 '18 at 22:33

Capturing the discussion in the comments, if $f$ is differentiable at $a$, then $D_vf(a) = \nabla f(a)\cdot v$, that is, the directional derivative is a linear function of $v$. In particular, we must have $D_{-v}f(a) = -D_vf(a)$, but for this function we have $D_{(1,0)}f(a) = D_{(-1,0)}f(a) = 5$. Coming at it from the other direction, $\nabla f(a) = \left(D_{(1,0)}f(a),D_{(0,1)}f(a)\right) = (5,1)$ and so $\nabla f(a)\cdot(-1,0) = -5 \ne 5$.

Geometrically, differentiability at a point is equivalent to there being a tangent (hyper)plane to the graph of the function at that point. The directional derivative in the direction of $v$ at $a$ is the slope in that direction of this tangent plane. There is no plane that slopes upward in opposite directions, so there can be no tangent plane at $a$.


Following suggestion given by amd in the comment, recall that for $f$ differentiable $D_{(v_1, v_2)} f (a)$ should be linear in $v_1$ and $v_2$ but in that case it is not.

Therefore $f$ is not differentiable in $a$.

  • $\begingroup$ I've not been able to find any proof about the linearity of the directional derivatives being a necessary condition for differentiability $\endgroup$ – Yagger Aug 29 '18 at 21:50
  • $\begingroup$ @Yagger We have that if $f$ is differentiable then $D_{(v_1, v_2)} f (a)$, therefore if $D_{(v_1, v_2)} f (a)$ is not linear $f$ can't be differentiable. $\endgroup$ – gimusi Aug 29 '18 at 21:52

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.