Consider the function $$f(x) = (x_1-1)^2x_2$$

Consider the points of the form $\overline{x} = (1,x_2)$

a) Analyze the optimialty of first and second order for these points.

b) What we can say about $\overline{x}$ using these informations?

c) Use the expression of the function to obtain more conclusive informations about the characteristics of $\overline{x}$


$$\frac{\partial f}{\partial x_1} = 2(x_1-1)x_2$$ $$\frac{\partial f}{\partial x_2} = (x_1-1)^2$$

$$\nabla f(1,x_2) = 0$$

$$\frac{\partial^2 f}{\partial x_1^2} = 2x_2$$

$$\frac{\partial^2 f}{\partial x_2^2} = 0$$

So the hessian at $(1,x_2)$ is

$$H(1,x_2) = \begin{bmatrix} x_2 & 1 \\ 1 & 0 \end{bmatrix}$$

$$\begin{bmatrix} a & b \end{bmatrix}\begin{bmatrix} x_2 & 1 \\ 1 & 0 \end{bmatrix}\begin{bmatrix} a \\ b \end{bmatrix} = a^2 x_2^2 + 2ab$$

b) this matrix is not definite positive or definite negative, so I cannot say that the points of the form $(1,x_2)$ are extreme points but I also cannot say they are not.

c) I think that this item is asking me to use the function to prove $(1,x_2)$ are not extreme points. Let's analyze $(1,x_2) + \lambda (d_1,d_2)$

$$f(x) = (1+\lambda d_1 -1)^2 (x_2+\lambda d_2) = \lambda^2 d_1^2 (x_2+\lambda d_2)$$

But small variations on $\lambda$ do not change the signal of $f$


$$\frac{\partial^2 f(x_1, x_2)}{\partial x_1 \partial x_2}=2(x_1-1)$$

$$\frac{\partial^2 f(1, x_2)}{\partial x_1 \partial x_2}=0$$

$$H(1,x_2) = \begin{bmatrix} 2x_2 & 0 \\ 0 & 0 \end{bmatrix}$$

Same conclusion as you for the front part.

Now to analyze its optimality.

Consider $x_2 > 0$. Then, we can draw a ball around $(1,x_2)$ such that the $y$-coordinate is always positive. Hence $f$ is nonnegative on that ball, hence it is a local minimum.

Similarly for $x_2<0$, we can draw a ball such that the $y$-coordinate is always negative, hence $f$ is nonpositive on that neighborhood. Hence it is a local maximum.

For $(1,0)$, whenever we draw a ball, we can always find $x_1 \ne 0$ and it is possible for $x_2$ to be positive or negative, hence it is neither local maximum nor minimum.

  • $\begingroup$ For $x_2>0$, if we think about $(1+h-1)^2(x_2+h)$ I can see that there's always $h$ that makes this positive, because I can always pick $h$ such that $0<h<x_2$, and $(1+h-1)^2(x_2+h)>0$. However I'm not convinced that this is a ball. It looks like that passes through $(1,x_2)$ $\endgroup$ – Paprika Sep 10 '18 at 4:23
  • $\begingroup$ $(1+h_1-1)^2(x_2+h_2)$, pick $h_1$ and $h_2$ such that $h_1^2 +h_2^2 < x_2^2$. $\endgroup$ – Siong Thye Goh Sep 10 '18 at 4:28

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.