OFDM can provide frequency diversity gain for point-to-point communications over frequency-selective slow fading channel. Recent works show that OFDM may also form a flexible and efficient multiple access method, which is often referred to OFDMA. However, the user outage probability and the optimal frequency diversity gain in OFDMA systems are not known. In this paper, random bipartite graph is used to model and analyse the multi-user subcarrier allocation problem over frequency-selective slow fading channels. Our aim is to minimize the user outage probability as well as guarantee fairness by dynamic allocating various subcarrier to each user. An optimal subcarrier allocation algorithm, which we shall refer to as the Hungarian method with random vertices rotation, is introduced to achieve these objectives. A simple but effective approximation equation for user outage probability is then derived. It is shown that the optimal frequency diversity gain in OFDMA system is the same as the point-to-point OFDM system. In particular, the frequency diversity gain does not decay as the number of users increases.