Abstract-Convex power flow relaxations have become popular to alleviate difficulties with embedding the non-convex AC power flow equations into optimisation models, and to provide guarantees on the quality of feasible solutions generated by heuristic approaches. However, their use has almost universally been limited to purely continuous problems. This paper extends the reach of relaxations to reconfiguration problems with binary decision variables, such as minimal power loss, load balancing and power supply restoration. This is achieved by extending the relaxations of AC power flows to bear on the on/off nature of constraints featured in reconfiguration problems. This leads to an approach producing AC feasible solutions with provable optimality gaps, and to global optimal solutions in some cases. In terms of run time, the new models are competitive with stateof-the-art approximations which lack formal guarantees.Keywords-Distribution Systems Reconfiguration, Power Supply Restoration, AC Power Flow, DC Power Flow, On/Off Constraints, Global Optimisation, Convex Relaxations, MINLP I. INTRODUCTION The steady-state alternating current (AC) power flow equations are at the core of virtually every computational problem in the field of power systems. Unfortunately, these equations form a system of non-convex constraints, raising significant challenges in any optimisation framework. When embedding these constraints into optimisation models, global non-linear programming (NLP) solvers do not scale and may not converge, even if all decision variables are continuous. These issues are exacerbated in the case of reconfiguration and power supply restoration problems, where discrete decisions are made that affect the grid topology. On the one hand, discrete variables increase the computational difficulty of the problem. On the other hand, topology changes lead to departure from the normal operating conditions that are used to hot-start non-linear solvers. Yet utilities are facing an increasing need for efficient reconfiguration methods, in order to optimise their operations to better handle the intermittent nature of distributed generation, and to quickly resupply customers in fault situations.There is a rich set of optimisation approaches in the power systems literature that tries to circumvent these problems. Black-box heuristic methods, which push power flow calculations outside the optimisation solver, have become very popular, owing to their broad applicability [1]- [3]. Yet, these approaches fail to exploit the problem structure and lack formal guarantees on the quality of the solutions returned. Another popular approach is to approximate the power flow equations, usually linearly [4]-[6], or with more accurate convex models [7]. This enables the use of a new generation of mathematical programming solvers including mixed-integer linear (MIP),