Relaxation checks if you can improve the distance to node by going through edge . The condition is: if dist[u] + weight(u, v) < dist[v], update dist[v] to the better value. This operation is the same in both Dijkstra and Bellman-Ford. The difference is how many times and in what order you relax edges. Dijkstra relaxes each edge once in priority order.
Bellman-Ford relaxes all edges multiple times. The power of relaxation is that it propagates distance improvements. If you improve dist[u], then all edges from might lead to improvements in their destinations. Repeated relaxation ensures these improvements reach all nodes.