Showing if $H$ is positive semi-definite there are either no minimizers or infinitely many

555 Views Asked by At

Suppose I have some $f(x) = g^{T}x + \frac{1}{2}x^{T}Hx$ where $g \in \mathbb{R}^{n}$ and $H$ is symmetric, positive semi-definite. This is the standard form for a quadratic minimization function.

It is just briefly mentioned in my textbook that unlike a positive definite $H$ which ensures $f(x)$ has a global minimiser, or an indefinite $H$ which ensures $f(x)$ is unbounded below, a positive semi-definite $H$ indicates that $f$ either has infinitely many solutions or is unbounded below. Considering that positive semi-definiteness of $H$ ensures $x^{T}Hx \geq 0 \ \forall \ x$, I don't see how this follows.

2

There are 2 best solutions below

0
On BEST ANSWER

Suppose $x^THx=0$. Then certainly $\lambda x$ for any real $\lambda$ satisfies the same equation, and provided $g$ is nonzero, we may make the quantity $f(x)$ as small as we want. Otherwise if $g$ is zero, then by the same logic we have infinitely-many zeros of $f$, which in this case are minima.

1
On

Such a function is a convex function. If two points minimize it, you can show the line connecting them does so as well.