Hilbert-Schmidt approximation problems arising in quantum chemistry (Q762440)

Questions related to the orthogonalization of molecular orbitals lead to the following mathematical problem. Given two bases \(f=\{f_ 1,...,f_ N\}\), \(g=\{g_ 1,...,g_ N\}\) of unit vectors in \({\mathbb{C}}^ N\), with g orthonormal, and given \(0\leq b\leq 1\), find the orthonormal basis e which minimizes \[ b\sum^{N}_{j=1}\| f_ j-e_ j\|^ 2+(1- b)\sum^{N}_{j=1}\| g_ j-e_ j\|^ 2. \] This in turn leads to the problem: given \(b_ 1,...,b_ n\) nonnegative numbers with \(\sum^{n}_{j=1}b_ j=1\) and given \(N\times N\) matrices \(A_ 1,...,A_ N\), minimize \[ E(U)=\sum^{N}_{j=1}b_ j\| U-A_ j\|^ 2_{HS}, \] where the norm is the Hilbert-Schmidt norm and U ranges over the unitary \(N\times N\) matrices. This problem is solved assuming \(C=\sum^{N}_{j=1}b_ jA_ j\) is a normal operator. J. Arazy and, independently, R. Bhatia and K. K. Mukherjea have pointed out that the assumption of normality on C can be dropped.