Smoothed analysis of \(\kappa(A)\) (Q1883587): Difference between revisions

For a real \(m\) by \(m\) random matrix \(A\) with independent Gaussian entries for the common variance \(\sigma^2\) and the matrix \(M\) of expected values of \(A\), the author proves that the probability of the condition number \(\kappa_2(A)=\| A\| _2 \cdot \| A^{-1}\| _2\) of \(A\) exceeding \(m \cdot x\) for any positive \(x\) is bounded by \[ \frac{1}{x}\left( \frac{1}{4\sqrt{2\pi m}}+7\left(5+\frac{4\| M\| _2(1+\log m)}{\sigma^2 m}\right)^{1/2}\right)\;. \]