$\newcommand{\R}{\mathbb{R}}$ $\newcommand{\al}{\alpha}$

Let $\Phi:\R^2 \to \R$ be a function which has partial derivatives at the point $(x_0,y_0)$. Let $\al:I \to \R^2$ be a path satisfying:

$\al(0)=(x_0,y_0),\dot \al(0)=v=(v_1,v_2)\neq (0,0)$.

Assume $\Phi \circ \al$ is differentiable at $t=0$ and the directional derivative of $\Phi$ at direction $v$ ($d_v \Phi(x_0,y_0)$) exists.

Is it true that $(\Phi \circ \al)'(0)=d_v \Phi(x_0,y_0)$?


As noted by user251257, if $\Phi$ is Lipschitz then both derivatives agree. So, in order to find a counter-example (where they differ) we need to search for non-Lipschitz functions, which still has partial derivatives at the point $(x_0,y_0)$.

This is not trivial, since a function with bounded partial derivatives is Lipschitz (as proved here). So, our search is narrowed down to functions which do not have partial derivatives in a neighbourhood of $(x_0,y_0)$, or that there partial derivatives are unbounded in some neighbourhood of it.


In general, it can happen that either one of $(\Phi \circ \al)'(0),d_v \Phi(x_0,y_0)$ exists without the other.

Example 1: $d_v \Phi(x_0,y_0)$ exists ,$(\Phi \circ \al)'(0)$ does not. (essentially taken from here)

$(x_0,y_0)=(0,0), \Phi(x,y) = \begin{cases} \frac{x^2y}{x^4 +y^2}, & \text{if $(x,y) \neq 0$}\\ 0, & \text{if $(x,y) = (0,0)$} \\ \end{cases}, y(x)=x^2 $

$\Phi$ has directional derivatives at every direction at $(0,0)$, but for $\al(x)=(x,x^2)=(x,y(x))$ we get: $F(x)=\Phi \circ \al(x)=\begin{cases} \frac{1}{2}, & \text{if $(x,y) \neq 0$}\\ 0, & \text{if $(x,y) = (0,0)$} \\ \end{cases} $ is not differentiable at $x_0=0$.

Example 2: $(\Phi \circ \al)'(0)$ exists ,$d_v \Phi(x_0,y_0)$ does not.

Update: this exmaple is wrong! I am not sure yet if a suitable example exists

Take $(x_0,y_0)=(0,0),\al(t)=(t,t^3+t),\Phi(x,y)=\sqrt{|xy|}$, $\Phi$ has partial derivatives ($\Phi_x(0,0)=\Phi_y(0,0)=0)$. $\Phi(\al(t))=\sqrt{t^4+t^2}$,so $(\Phi \circ \al)'(0)=1$. However, the directional derivative at direction $v=(1,1)$ does not exist.

  • $\begingroup$ So you want to know if $\frac{d}{dt}\Phi(\alpha(0) + t \dot\alpha(0))|_{t=0} = \frac{d}{dt}\Phi(\alpha(t))|_{t=0}$? $\endgroup$ – user251257 Feb 27 '16 at 13:52
  • $\begingroup$ @user251257 Yes, exactly. $\endgroup$ – Asaf Shachar Feb 27 '16 at 14:39
  • $\begingroup$ I'm pretty sure that this is true for all paths exactly when $\Phi$ is differentiable. (This is more or less the definition.) Unless you consider a function which are differentiable along some paths and not along others (like the two examples). $\endgroup$ – tomasz Feb 27 '16 at 16:29
  • $\begingroup$ @tomasz: If $\Phi$ is differentiable this is indeed immediate. If it is not differentiable, I want to know, if for every path along $\Phi$ is differentiable, the derivative must be equal to the directional derivative, i.e to the derivative along the linear path. $\endgroup$ – Asaf Shachar Feb 27 '16 at 19:00

If $\Phi$ is Lipschitz continuous on a neighborhood of $\alpha(0)$, then both derivatives do agree.

Updated counter example:

Consider following example: \begin{align} \Phi(x,y) &= \sqrt[3]{xy}, \\ \alpha(t) &= (t(1-t), t^2). \end{align} Then, we have $\Phi(\alpha(t)) = t\sqrt[3]{1-t}$. On the other hand we have $\alpha'(0) = (1,0)$ and $\Phi(\alpha(0) + t\alpha'(0)) = \Phi(t, 0) = 0$. So both derivatives differ.

  • $\begingroup$ Your example is interesting. I think I have learned something from it. Thanks. However, your function has a directional derivative in only one direction, $(0,1)$ (up to scaling) except zero. I am interested to know if this situation is possible for a function which has directional derivatives at two different directions. (That's why I specified the function must have partial derivatives). $\endgroup$ – Asaf Shachar Feb 28 '16 at 8:45
  • $\begingroup$ @AsafShachar: Ah. I missed that $\Phi$ should have partial derivatives. I will improve the answer later $\endgroup$ – user251257 Feb 28 '16 at 14:36
  • $\begingroup$ I have updated the question, to direct future attempts. In particular I have specified some limitations on the nature of potential candidates, which follow from your observation bout Lipschitzity. (You might find it helpful to have a look). $\endgroup$ – Asaf Shachar Feb 28 '16 at 15:48
  • $\begingroup$ @AsafShachar I updated the counter example $\endgroup$ – user251257 Feb 28 '16 at 17:34

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.