DrMAL: Game here is identical to Fischer vs Spassky, 1972 up to move 17 for black, see commentary on 14...a6?! versus 14...Qb7! there. Post by <Some call me Tim> about preparation is incorrect, maybe there was misprint. Making move 16...Qb7 does not give any advantage in fact there are number of better moves.
Geller's first opportunity to prove 14...Qb7 move that he gave to Spassky prior to WC match was in Timman vs Geller, 1973 and he played it there.
Game here was first game where 14.Bb5 was played and, being as Geller lost, he looked for better response.
Back to line played here and two years later in WC match, here is computer eval of alternatives with starting lines for black on move 17.
Houdini_20_x64: 29/65 2:07:35 80,891,715,277
-0.11 17. ... Rac7 18.b3 Bf5 19.Rfd1 d4
-0.11 17. ... Qf8 18.Qc3 Rac7 19.Rfd1 Nd7
-0.14 17. ... a5 18.Rc3 Nd7 19.Nd4 Qd8
-0.16 17. ... Nd7 18.Nd4 Nf6 19.Nxe6 fxe6
-0.17 17. ... Rcc7 18.Rc3 Nd7 19.Nd4 Qd6
-0.18 17. ... Rb7 18.b3 Rb6 19.Nd4 Nd7
As several GMs later pointed out, particularly Kasparov, Spassky's 17...Nd7 was also not improvement over 17...a5 played here. They could have transposed if Spassky played 18...a5 but his choice of 18...Qf8 was unremarkable as well. Geller's 19...Re8?! was poor here, best was probably 19...Qd8 but computer shows 19...Rb8 and 19...Qf6 and 19...Bg4 as also good, several others such as 19...Bg4 were OK as well. It was new position and common reaction is to make "safer" passive moves. After 20.Bb5! pinning K that protects pawn, 20...c4 was clearly best and 20...Bg4?! was second error.
However, Furman did not play 21.Rxc5! to win pawn (e.g., 21...Nxc5 22.Bxe8 and 22...Bxf3 or 22...Qxe8 is met by 23.Qxc5). Instead he tried to complicate a bit with 21.Nd2 and was left with only small edge after 21...d4! and exchanges that followed that simplified game instead. After 24.Rc8+ best was 24...Nf8 to later re-position N but 24...Kh7 only other plausible move was OK, particularly after 25.Nb3 innocuous move instead of 25.Rd8! try for attack. I guess Geller had 25...Ne5 in mind earlier but here instead of 25...Be2 is was another error, and again Furman did not take proper advantage of it, playing 26.Rd8 to put R behind P instead of 26...R8c5!
With 26...d3 Geller had even chances but game seemed likely draw but intead he sought to further simplify with 26...Rac7?! Again, 28.Rd5 similar to move 26 may have been best but this time 28.f4 was OK but 29.Bxd7 was stronger two choices, both win pawn. Game looked probably drawn from here especially if 30...Rc5 was finally played (R had allergy to fifth rank?). Geller's next two moves were inaccurate but game was far from decisive, I guess his flag either fell or was about to. Here is eval at end of game, there were at least 3 moves to survive.
Houdini_20_x64: 30/74 21:22 14,288,271,463
-0.56 32. ... Rf1+ 33.Kg3 Bd3 34.Rd7 Be4 35.Ne2
-0.61 32. ... Bd3 33.a4 Be4 34.b3 g6 35.g3
-0.66 32. ... Rb1 33.b3 Bd3 34.a4 Rb2+ 35.Ke3