Modern global optimization algorithms involve the construction of convex underestimating programs, termed convex relaxations. For dynamic optimization problems, this relaxation procedure poses unique challenges and has been the subject of several recent articles. In this talk, we review some existing relaxation procedures for dynamic problems and investigate the use of a convergence metric for evaluating their usefulness. Using numerical results and recent developments in convergence analysis, we argue that standard metrics for nonlinear programs may not be sufficient; there is a complicating factor, related to time (the independent variable of the dynamic system), that is unique to dynamic problems. This observation is corroborated by recent advances in the analysis of the so-called cluster effect. Combined, these observations suggest a new design goal for dynamic relaxation procedures.