In trying to understand the mathematical details of high-dimensional QKD with few mutually unbiased bases. First, prepare a state $|\psi\rangle \in H^d$ as a linear combination of a first orthonormal basis $A$. Then use a second mutually unbiased basis B as the Hamiltonian POVM operator.
What is the probability of observing the value associated to $|b_i \rangle$? The probability is $\frac{1}{d}$.
$\langle \psi | b_i \rangle \langle b_i | \psi \rangle = (\sum_s \alpha_s^* \langle a_s|b_i\rangle) (\sum_t \alpha_t \langle b_i |a_t\rangle) = \sum_j \alpha_j^* \alpha_j \langle a_j|b_i\rangle \langle b_i |a_j\rangle + \sum_{s \neq t} \alpha_s^* \alpha_t \langle a_s|b_i\rangle \langle b_i |a_t\rangle = \frac{1}{d}$.
Since $|\langle a_s|b_i\rangle|^2 = \frac{1}{d}$ (mutually unbiased bases), the first summation with the index $j$ is simply $\frac{1}{d}$.
But the problem, I do not see why the second summation with the indices $s \neq t$ is equal to $0$. The phases of each element should cancel out ($\langle a_s|b_i\rangle = e^{i\theta} \cdot 1/ \sqrt{d}$).Furthermore, can I derive this result without having to deal explicitly with the phases?