Estimation of the confidence limits for the quadratic forms in normal variables using a simple Gaussian distribution approximation (Q2488386)

The distribution of a quadratic form \(Q=\sum_{k=1}^n d_k y_k^2\), where \(y_k\) are i.i.d. standard Gaussian variables, is approximated by \({\mathcal N}(\sum_{k=1}^n d_k, 2\sum_{k=1}^n d_k^2)\). The related approximation for quantiles of \(Q\) (one-sided confidence bounds) is analysed via simulations. E.g., it is noted that for \(n=100\) the relative error of such approximation is less then 7\%.