The gradient $\nabla f$ of a differentiable function $f(x,y)$ points in the direction of steepest ascent at a given point $(x_0,y_0)$. The slope of this ascent is the magnitude of the gradient $\|\nabla f(x_0,y_0)\|$. It seems possible that there could be a non-parallel vector $u$ such that the (directional) derivative $\nabla_{u}f (x_0,y_0)$ in the direction of $u$ is also equal to $\|\nabla f(x_0,y_0)\|$. Thus the direction of steepest ascent would not be unique - the gradient just happens to equal one of directions of steepest ascent. Can this happen?

Is there a function $f(x,y)$ and a point $(x_0,y_0)$ such that

  1. $f$ is differentiable at $(x_0,y_0)$,
  2. There are two unit vectors $u\neq v$ such that $$\nabla_{u}f (x_0,y_0)=\nabla_{v}f (x_0,y_0)=\|\nabla f(x_0,y_0)\|\neq 0?$$

In the cases I've thought of where there are obviously two methods of steepest ascent (like a hyperbolic paraboloid) the gradient is zero.


If the gradient is not zero, then the steepest ascent direction is unique. This follows from the Cauchy-Schwarz inequality: $|(u,v)| \leq \| u \| \| v \|$ with equality if and only if $u=cv$ for some scalar $c$. Applying that here, the directional derivative $\nabla_u f$ is $(u,\nabla f)$, so $|\nabla_u f| \leq \| u \| \| \nabla f \|$ with equality if and only if $u=c \nabla f$. For $c>0$ you are following the steepest ascent direction, for $c<0$ you are following the steepest descent direction.


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.