In endings with a K+2B’s or K+B+N it’s well-known that the superior side can deliver mate if he knows the technique. In the case of 2N’s mate cannot be forced against the correct defense (which is really very easy).
However, if the lone K has a Pawn then in some cases a mate can be forced. Sometimes it can require over 100 moves to do so though. The reason a mate can be forced is that the defender’s P gives him a move and deprives him of a stalemate defense.
The technique is to block the P with one N and use the K and other N to force the opposing K into a corner or near the other N. Then when the block on the defender’s P is removed, the N can be used to checkmate
The fact that 2N’s can sometimes win against one or more P’s was known at least as early as 1780. In 1851 Horwitz and Kling published three positions where the N’s win against one pawn and two positions where they win against two pawns. Then Troitsky started studying the endgame in the early 20th century and published his extensive analysis in 1937. Modern computer analysis found it to be very accurate.
Games with this ending are rare – Troitzky knew of only six when he published his analysis in 1937. In the first four (from c. 1890 to 1913), the weaker side brought about the ending to obtain a draw with an opponent who did not know how to win. The first master game with a win was in 1931 when Adolf Seitz beat Eugene Znosko-Borovsky. And in more recent times Karpov once lost an ending with a P vs. 2N’s to Topalov in a rapid play event.
I recently entered a correspondence tournament and in the process of checking out a few games by one of my opponents I came across this very interesting mate!