Suppose we have n normalized vectors on an arbitrarily large Hilbert space $A_1\rangle,\dots,A_n\rangle$, $\langle A_iA_i\rangle=1$ for every i. And there're $\frac{n(n1)}{2}$ inner products $\langle A_iA_j\rangle$. But they're not mutually independent, for instance, if $\langle A_1A_2\rangle=\langle A_1A_3\rangle=1$, then it's required $\langle A_2A_3\rangle=1$. Then my question is what is the exact relations between these inner products? Or how can we compute this relation efficiently?

The matrix of inner products is the Gram matrix of your vectors. This is a positive semidefinite matrix, whose rank is the dimension of the span of the vectors. Conversely, every PSD matrix of the right rank arises as the Gram matrix (since your vectors are normalized, the diagonal elements are all $1$). 

