Minimization of $J(x)=\langle Ax,x\rangle$ on the unit sphere on $\mathbb{R}^{n}$

77 Views Asked by At

I would like to show that the minimum of $J(x)=\langle Ax,x\rangle$ (where $A$ is a symmetric $n$ by $n$ matrix) on $K=\{x\in\mathbb{R}^{n} : F(x) = 1 -\lVert x \rVert^{2} = 0 \}$ is the eigenvector associated to the smallest eigenvalue of $A$.

To do this I use the Lagrange theorem, we now that $K$ is compact and $J$ is continuous so there exists a solution to this problem, let’s say $u$. Now from the Lagrange necessary condition we know that there exists $\lambda$ such that

$$ J’(u)(h) + \lambda F’(u)(h) = 0\quad\forall h\in\mathbb{R}^{n} $$

Here we have $J’(u)(h) = 2\langle Au, h\rangle$ and $F’(u)(h) = 2\langle x,h\rangle$.

Thus we have

$$ \forall h\in\mathbb{R}^{n}, \langle Au -\lambda u, h\rangle = 0 \implies Au = \lambda u $$

From this we have indeed that $u$ is the eigenvector associated to $\lambda$. However I do not see how to prove that $\lambda$ is the smallest egeinvalue of $A$ ?

If you have any hints (instead of full answer, in order to work a little bit by myself) do not hesitate to share it please.

Thank you

1

There are 1 best solutions below

2
On BEST ANSWER

Just use the fact that you minimize over the unit sphere and the connection between dot product and norm. $$\langle Au,u\rangle=\langle \lambda u,u\rangle=\lambda \langle u,u\rangle=\lambda \|u\|^2 =\lambda $$