This is log-sum-inv. In essence, the same as Log( {convex} ) . Make modifications, such as having log(B) rather than 0 as 1st element in vector argument of log_sum_exp,. no H^2 term, and A_i in place of GAMMA.
Note: edited to correct log(B) to log(1/B).
Note: reverted to original., i.e., log(B), not log(1/B)