Clustered Sampling`s Algorithm2

[huang-jianfeng]

Hi, I am a graduate student and I am very interested in your thesis. I have a little question about Clustered Sampling: Low-Variance and Improved Representativity for Clients Selection in Federated Learning`s Algorithm2.
Why is b_i rather than u_i assigned to r_{k,i} in line 12?
(b_i equals qk+ui when a_i = 0, is the sum of q_k bigger than mM finally? )