The spectrum norm of difference of two matrix converging to 0

53 Views Asked by At

Let $\Theta$ be a positive-definite matrix, and $\hat\Theta$ be a matrix with the same dimension as $\Theta$. If the spectrum norm of $\Theta - \hat\Theta$ converges to 0 with large probability, why can we say that $\hat\Theta$ is also positive-definite with large probability?

This fact is used in many papers, e.g. this one says Theorem 1 implies that are $\hat\Theta$ and $\hat\Omega$ are positive-definite with large probability. Why does this hold? What is the intuition behind it? Will the same result hold if we replace the spectrum norm with any other norms?

1

There are 1 best solutions below

0
On BEST ANSWER

Matrix perturbation theory might explain it. The intuition is that two close matrix also have close eigenvalue.

And since all matrix norms are equivalent up to a constant multiplier, this conclusion should also hold under other norms.