Using Lagrange multipliers to maximize directional derivative

431 Views Asked by At

enter image description here

I've tried to use Lagrange's multipliers to solve this question, but I'm unable to find an answer that matches the choices. I know that the gradient, i.e. $\triangledown f \langle x, y, z \rangle$ would be $\langle -1, 2, 3 \rangle$, and that the equation of the tangent plane would be $x + z -2 =0$, but I don't know how to use the constraint. I know that the unit vector that I multiply the gradient with must lie on this plane, but what does that mean quantitatively?

I've taken the unit vector to be $\langle a, b, c\rangle$, so that tells me that $-a + 2b + 3c = 0$.

I also know that since $\langle a, b, c\rangle$ lies on the plane $x+z=2$, this means that $a+c=2$, and since it is a unit vector, $| \langle a, b, c\rangle| = 1$, which means that $a^2 + b^2 + c^2 =1$. I've tried to solve the double constraint, but my answer does not match the answer choices.

1

There are 1 best solutions below

0
On

The normal vector to the tangent plane $T$ is $\parallel OP$, i.e., ${\bf n}=(1,0,1)$. We have to find two unit vectors in $T$ that are orthogonal ${\bf n}$ and to each other. These are for instance the vectors $${\bf a}=(0,1,0), \qquad {\bf b}=\left({1\over\sqrt{2}},0,-{1\over\sqrt{2}}\right)$$ (found by guessing), so that the general admissible vector ${\bf u}\in T$ is given by $${\bf u}=\cos\phi\> {\bf a}+\sin\phi\> {\bf b}\ .$$ Now $$D_{\bf u}f(P)=\langle \nabla f(P),{\bf u}\rangle=\left\langle(-1,2,3),\left({\sin\phi\over\sqrt{2}},\cos\phi,-{\sin\phi\over\sqrt{2}}\right)\right\rangle\ .$$ It follows that $$D_{\bf u}f(P)=2\cos\phi-2\sqrt{2}\>\sin\phi\ ,$$ so that the maximum is $$\sqrt{2^2+\bigl(2\sqrt{2}\bigr)^2}=2\sqrt{3}\ .$$ If you want to use Lagrange multipliers then you have to find a vector ${\bf u}=(\xi,\eta,\zeta)$ satisfying $$\langle{\bf u},{\bf n}\rangle=\xi+\zeta=0,\qquad\xi^2+\eta^2+\zeta^2=1$$ such that $$\langle\nabla f(P),{\bf u}\rangle=-\xi+2\eta+3\zeta$$ is maximal. This will lead to a lot of equations.