The core operation of Bellman-Ford is edge relaxation. For each directed edge $(u, v)$ with weight $w(u, v)$, the algorithm checks whether going from the source to $u$ and then taking the edge to $v$ yields a shorter path: $d[u] + w(u, v) < d[v]$. If so, it updates $d[v]$ to $d[u] + w(u, v)$, which is exactly captured by $d[v] = \min(d[v], d[u] + w(u, v))$.