Least Squares Problems in Orthornormalization

Abstract

For any $n$-tuple $(\alpha_1,...,\alpha_n)$ of linearly independent vectors in Hilbert space $H$, we construct a unique orthonormal basis $(\epsilon_1,...,\epsilon_n)$ of $span\{\alpha_1,...,\alpha_n\}$ satisfying: $$\sum_{i=1}^n\|\epsilon_i-\alpha_i\|^2\le\sum_{i=1}^n\|\beta_i-\alpha_i\|^2$$ for all orthonormal basis $(\beta_1,...,\beta_n)$ of $span\{\alpha_1,...,\alpha_n\}$. We study the stability of the orthornormalization and give some applications and examples.