In the context of semidefinite programming (SDP), under what conditions is it preferable to formulate and solve an LMI-based SDP rather than an equivalent standard form SDP?
I have been told that solving an LMI-based SDP is computationally very expensive. If that is the case, then why use LMIs?