I am studying the LoRA paper and have trouble understanding this figure. The function essentially measures how much of the subspace spanned by the top i vectors is contained in the subspace spanned by the top j vectors in the higher rank matrix. Therefore, j can not be lower than i. So when they say
