This question might sound stupid but I want to confirm an answer from it.

I saw somewhere online that it means that when the directional derivative of function $f$ along the none zero vector $v$ at certain point is equal to $0$, it means that the function $f$ is constant in that direction. But what does "constant in direction" mean? can anyone give me an example of it such as $f(x,y)$ to explain this?


  • 5
    $\begingroup$ I think it simply means that the value of the function does not change as you move along the direction given by that vector. I sounds redundant, I guess. Hope this helps $\endgroup$
    – Vishesh
    Sep 13 '13 at 12:38
  • $\begingroup$ As an afterthought, this happens when you look at tangent vectors to level curves of the given function. Also if you know $ D_{v}f = \nabla f . v $, where $v$ is your tangent vector. The gradient is always normal to a level curve of a function. $\endgroup$
    – Vishesh
    Sep 13 '13 at 12:43
  • $\begingroup$ But if the value long curve of that direction doesn't change, doesn't it imply that the curve is a constant? But hardly can I imagine a concrete function like this. $\endgroup$
    – Cancan
    Sep 13 '13 at 12:54
  • $\begingroup$ Other better answers have already been given. Anyway for the sake of completion, just take a look at $ f(x,y)= 2e^{x}+3e^{y}$ at $(0,0)$ Then use the vector $ (-3,2) $. Cheers $\endgroup$
    – Vishesh
    Sep 13 '13 at 13:05
  • $\begingroup$ What do you mean by saying a curve is a constant??? $\endgroup$
    – Vishesh
    Sep 13 '13 at 13:06

If $D_v f(x_0)=0$, then $f$ is constant to first order in the direction $v$ - that is, if you consider the values of $f$ along the line in the direction $v$, you find that there is no linear-order term:

$$ f(x_0+tv) = f(x_0) + t D_vf(x_0) + o(t) = f(x_0) + o(t). $$

(Here $o(t)$ is some function such that $o(t)/t \to 0$ as $t \to 0$, which is what we mean by zero to first order.)

Instead of considering a line, you can consider the level set $\{x : f(x) = f(x_0)\}$. So long as $x_0$ isn't a critical point of $f$, this level set will (at least in some neighbourhood of $x_0$) be a curve. Thus $f$ is constant along a curve in the direction $v$.


Let's start with the simplest case first. $f: \mathbb{R} \to \mathbb{R}$ given by $f(x) = x^2$. Then the directional derivative

$$\frac{\partial f}{\partial x} = \frac{df}{dx} = 2x$$

which is zero at the origin. If you look at a graph of a parabola, you see that the closer you zoom in on the origin, the more flat the graph looks (check this yourself on an online graph calculator).

Now take $g: \mathbb{R}^2 \to \mathbb{R}$ to be $g(x,y) = x^2 + y$, which is a parabolic cylinder, as can be seen here: http://www.wolframalpha.com/share/clip?f=d41d8cd98f00b204e9800998ecf8427empntaacb4q.

If you take the directional derivative along $y$ (also known as the partial derivative with respect to $y$), you get $1$. Thus, this function is increasing as you fix $x$ and increase $y$. However, the directional derivative along $x$ is $2x$, which is zero at the origin. Thus, if you zoom in really close to the origin, what you will see is something that looks like a tilted plane. As you run along the $x$ axis, the plane stays at the same height (constant), but it changes as you run along the $y$ axis.


Let's break our understanding in several pieces and glue them together to understand the puzzle:

  1. $f(x,y)$ is a number.
  2. $\nabla f(x,y)$ is a vector.
  3. $\vec{v}$ is also a vector.
  4. The directional derivative is a number that measures increase or decrease if you consider points in the direction given by $\vec{v}$.
  5. Therefore if $\nabla f(x,y) \cdot \vec{v} = 0$ then nothing happens. The function does not increase (nor decrease) when you consider points in the direction of $\vec{v}$.

Suppose that $f$ is differentiable and that the directional derivative of $f$ along the vector $v=(a_1,a_2,...,a_n)$ is zero i.e. $\nabla f\cdot v=0$ or $a_1f_1+a_2f_2+...+a_nf_n=0$.

Claim: $f$ is constant along any line having direction $v$

$L$ be any such line which passes through $p=(b_1,b_2,...,b_n)$. Now the parametric equations of $L$ are: $x_1(s)=b_1+a_1s,x_2(s)=b_2+a_2s,...,x_n(s)=b_n+a_ns$. Keeping this in mind, we define $F(s)=f(x_1(s),x_2(s),...,x_n(s))$. The derivative $F'(s)=f_1\dot x_1(s)+f_2\dot x_2(s)+...+f_n\dot x_n(s)=f_1a_1+f_2a_2+...+f_na_n=0$. So $F$ must be a constant function. This shows that $f$ is constant along $L$.

  • $\begingroup$ Why does $F'(s) = 0$ implies that $F$ is constant? $\endgroup$
    – John Mars
    Apr 7 at 20:49
  • $\begingroup$ @JohnMars By Lagrange's mean value theorem, $F(x)-F(y)=F'(z)(x-y)=0$ for all $x,y$. So $F$ is constant. $\endgroup$
    – Hrit Roy
    Apr 8 at 23:13

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.