Suppose "mathemathics would never prove a contradiction". We can write it out as ¬Provable(⊥). This is logically equivalent to Provable(⊥) → ⊥, and it also implies Provable(Provable(⊥) → ⊥) by the rules of provability. But Löb's theorem expresses ∀ A (Provable(Provable(A) → A) → Provable(A)), which we can instantiate to Provable(Provable(⊥) → ⊥)→ Provable(⊥), and now we can apply modus ponens and our assumption to get a Provable(⊥).