#3 Gligoric-Szabo (view game online at Chessgames.com)
From an Open Ruy Lopez we arrive at the following position with White to move. As Nigel Short might say, 'Black is getting stuffed', being squeezed for space and with seemingly no good place to put his queen. Black comes up with an interesting idea to exchange queens and even arrive at a better endgame, all starting with an inaccuracy from White...
20. Rab1?! Qb5! (20. Rfb1 is suggested as more exact, since Qb5 is then refuted by 21. Nbd4, whereas now the a-pawn is underprotected)
21. Qc2 Qc4 (Bronstein gives the amazing sequence 21. Nfd4 Nexd4 22. Nxd4 Qxa5 23. Nxc6!! Qxa2 24. Nxe7+ Kh7 25. Rb4 g5 26. Bxg5 g6 27. Rh4+ Kg7 28. Bf6#..... although 21.... Ncxd4 seems safe for Black)
22. Nfd2 Qg4
23. f4 Qf5
A remarkable journey by Black's queen over the last four moves establishes equality - it took real imagination by Szabo to solve the problem of his queen on b8 to put it via b5, c4, and g4 to f5...just to trade it off!
24. QxQ PxQ
25. Nf3 Rfb8
26. Nfd4 Ncxd4
27. Nxd4 Nxd4
Bronstein: Taking with the pawn hems in the bishop still more severely, but 28. Bxd4 would result in a lost position after 28... Rb5 29. Rxb5 axb 30. Ra1 Ra6 31. Kf1 Kf8.
Amazing, but apparently true. Although to my first impression (and my computer's) having the bishop locked up by another pawn is the worse choice, this is not borne out by examination. Here's the analysis position Bronstein describes as lost for White:
Analysis position after 31...Kf8
Oddly enough White has no plan to improve his standing - if the king tries to support a kingside pawn push and wanders away from the queenside, Black can just push ...b4 at will because cxb4 Bxb4 loses the a-pawn, and ...c5 is coming as well, creating a protected passed pawn. Even with the White king on the queenside the plan is the same just slower - White has no way to push through a pawn of his own. So Gligoric's 28. cxd4 is correct, although the reason is not obvious.
28. .... Bb4
29. Ra1 Rb5
30. Ra4 Rab8
31. Rfa1 Bc3
32. Rac1 Rb1
33. Rcxb1 Rxb1+
34. Kf2 Ra1?! (...Rb5 is more exact, for a very subtle reason. Why? Buy the book!)
35. Rxa1 Bxa1
36. Ke2 Bc3
37. Kd3 Bxa5
Using threats to the a-pawn Black achieved the exchange of rooks and finally won the forlorn a-pawn and is left with an endgame where his bishop is monumentally better and he has an extra outside passed pawn....and indeed White resigned in a further few moves. (Despite the impressive-looking e5 pawn, Black can play an eventual ...f6 and fxe after which he can overload the defending White bishop from either the d- or f-pawn depending on which is remaining by a ...d4 or ...f4 pawn sacrifice and mop up the White pawns with his bishop).
Yet all is not lost! Bronstein found the saving:
When avoiding the exchange of bishops with 38....Bb6 allows White 29. Bb4! shutting Black's king out of the center of the board since ....f6 is met with 40. e6. That leaves...
38. .... Bxd2
39. Kxd2 Kf8
40. Kc3 Ke7
41. Kb4 Kd7
Analysis diagram after 42. Kc5
As a draw (!). Once again, an example of a hidden resource...never give up looking for them! (This is also why 34.... Rb5 was better - this continuation would not have been possible.)