Consider what threats, if any, each Black piece makes. Answer: both Black rooks (in addition to being forked by the knight at f4) are trained on g2, the square in front of White’s king. This could be turned into mate if that square were not guarded by the now-familiar defensive knight, this time on f4. Removal of the piece requires the large sacrifice QxN, but where mate is available the size of the sacrifice is neither here nor there. If White recaptures QxQ, Black mates in three moves. He operates exclusively with checks: first comes Re2xg2+; White is forced to move his king to h1. Now Black plays Rg1+, and White is forced to play his king to h2. The final blow is Rg6-g2#. This is a common pattern; the idea is to get the two rooks onto adjacent ranks with the king’s progress up the board blocked (in this case by his own pawn on h3). This way the two rooks protect each other as well as sealing off the king’s escape routes. The artfulness of the sequence just described is that it manages to corner White’s king without ever putting a rook on h2 where it could get nabbed by White’s queen from f4 (or by White's remaining knight on f1).

The upshot of all this from White’s standpoint is that in reply to Black’s initial QxN, he should not recapture. He should play Qc8+, giving a check of his own (always ask whether you can take the offensive). Black probably interposes his knight at f8, and now White plays his queen to g4, trying to interrupt Black’s attack along the g-file. White’s queen is lost to RxQ, but then he can play h3xR and at least the immediate mating threat has been avoided. White’s material losses have essentially ended the game anyhow, of course.