keypusher: Shredder dislikes 9.Na3 but doesn't think the move is as bad as it looks to me. After 9.Nd2 it thinks White could have kept a small advantage.
The computer does not think Black is anywhere close to winning after 9.Na3 Bxe2 10.Qxe2 cxd4 11.cxd4 Bxa3 12.bxa3 Nc6, but I think a strategically clear-cut game without lots of tactics in the offing was very welcome to Botvinnik. Instead of 13.Be3?!, the engine thinks it's even after 13.Qb5 Ne7 14.h5 Qa5+ 15.Qxa5 Nxa5 16.Rh3, but this was game 18, and Tal was four points behind. Equal endings were little better than losses to him by now.
13.Be3 Qa5+ 14.Kf1? (again Tal feels compelled to avoid offering the exchange of queens) 14
.Ne7 (the engine likes simply grabbing the pawn at once) 15.Rb1 Rb8 16.Bh3? (16.Qd3) 16
.Qa4?! 17.Rd1? Qxa3 (White has no compensation for the pawn) 18.Kg2 Qa6! (maybe not objectively the best, but very strong under the circumstances. White can hardly pass up the exchange of queens now, with a dreary ending ahead for him).
19.Qxa6 bxa6 20.h5 Kd7 21.Rb1 Rb6 22.Kg3 Na5 (the computer actually prefers keeping the doubled pawns with 22
.Rhb8 23.Rxb6 Rxb6 and White's a- or d-pawn will soon fall) 23.Rxb6? (the computer prefers 23.Rhc1 Rhb8 24.Ra1, but it's hard to blame Tal for not going in for that) 23
.axb6 with a great game for Black.
Most of the way this looked like Botvinnik's smoothest and easier win of the match, but things get a little rocky toward the time control. Instead of 33
.exf5 the computer likes 33
.Nd2 and if 34.Rd1 Ne4+ 35.Kh4 g5+! 36.fxg6 fxg6 and the h-pawn will soon fall.
Position after 37....f6:
.f6? Tal could have scared up drawing chances with the subtle 38.Rg1!, planning to answer
Nxa3 with Bf3 attacking d5 and g7. After 38
.Kf8 39.Bf3 Nb6 40.Re1 no win is apparent. Instead 38.Bf5?? dropped another pawn to Black's simple combination and brought Botvinnik a giant step closer to regaining the title.