It's been explained numerous times. Aronian picked the move that would require White to spend the maximum time picking off his pawns. He must have thought that White's King would be too far away afterwards for it to matter that the king was closer after g6 than it would have been after h6. It very nearly was, as can be seen by the trap at Move 57.

<Even I bloody patzer saw that it is a mistake.>

If we're going to just GM moves by whether or not a patzer would play them, then I'm afraid we'll have to throw most of them out. You may not be grasping just how dangerous White's queenside pawns were.

The picking off itself takes three moves in both cases: Kg5-Kxh5-Kxg6 in the game, g6-Kxg5-Kxh6 if 50.h6. But in the former case it might seem that White is gaining time - especially if one thinks about it as a pawn race - because the black king ends up blocking his g-pawn. This is confirmed by what Kramnik himself said in that bit of interview quoted above by <polarmis>: <But g6... Actually, what's the idea? Levon also couldn't manage and got confused - he told me after the game that he couldn't calculate it to the end and played g6 with the idea that when I took the pawns the king would slow down the g-pawn.>

Exactly. I may not have explained it fully, but after g6, it takes White an extra move to pick off the pawns AND clear the way for his g pawn to start moving than it would have if he'd played h6.

click for larger view

It is indeed a draw after 47...Kd7 as you indicated but I thought that after 47...Bh7+ that it might not be. White has 2 options:

(1) Keep the king on the k-side and force Black to give up his bishop to prevent the h-pawn from queening. But I then thought that Black had enough time to pick off White's q-side pawns. But after 47...Bh7+ 48.Kf6 Kd7 49.Kg7 Bb1 (or any other bishop move along the b1-h7 diagonal other than 49...Bg6) 50.h7 Bxh7 51.Kxh7 and Black mates in 18 moves per the 6-piece Nalimov tablebases.

(2) Move the king to the q-side and protect his q-side pawns. This works out better after 47...Bh7+ 48.Ke6 Kc7 49.Kd5 Bb1 50.Kd4 Kc6 51.h7 Bxh7 52.c4 (the key move!). And now if either 52...bxc4 53.Kxc4 or 52...b4 53.axb4 it's a draw per the 6-piece Nalimov tablebases.

Black's problem is that after 52.c4 bxc4 53.Kxc4 his bishop does not control the a-pawn's queening square, so even if White lost his a-pawn the position would still be a draw. And after 52.c4 b4 53.axb4 a3 54.Kc3 White's king can get to a1 so even if White loses both his remaining pawns the position is still a draw.

I want to let you (and others) know about a nifty program developed by Pedro Pérez Romero called FinalGen. It is basically a tablebase generator for positions consisting of kings, one piece for each side, and any number of pawns. So it could theoretically determine the results of positions containing up to 20 pieces; two kings, two pieces, and 8 pawns per side.

I said "theoretically" because there are positions that meet its basic requirements but are too complex for the program to analyze. Other positions can theoretically be analyzed but are not practically possible because they will require weeks of analysis and 10s of TB of storage. It all depends on the mobility of the pawns; if the pawns are far advanced or blocked then the number of possible pawn moves is much smaller than if the pawns were near their original squares and had full freedom of movement. Still, for many positions the program can come up with definitive results beyond the capabilities of the 6-piece Nalimov tablebases and 7-piece Lomonosov tablebases.

These were the results of its analysis of the position after 47.gxh6. It took FinalGen 00:29:23 minutes and required about 28 GB of disk space (I added the move numbers myself):

47...Ba2 Draw
47...Bb3 Draw
47...Bc4 Draw
47...Bh7+ Draw
47...Kd7 Draw
47...Kb7 Draw
47...Kc7 Draw
47...Kb8 Draw
47...Kd8 Draw
47...b4 White wins in 28
47...Bd5 White wins in 3
47...Be6+ White wins in 3
47...Bf7 White wins in 3

Of course, even I as Black would not play any of the last 3 moves but it does show that the program is thorough. But for the other moves you can then ask the program to display the lines that make up the principal variation for each move, and you can explore the side variations.

No, Fischer's mistake was 29...Bxh2, not later, and his game was lost against best play by Spassky from that point forward. See my analysis (with a "little help" from Rybka 4.1) starting at Spassky vs Fischer, 1972 and continuing through most of p.17. If you can find any errors in it or other lines that might allow Fischer to salvage a draw after his mistake, please let me know. I would really like to see them.

So, Fischer's mistake wasn't in overlooking 30. g3, it was in overlooking 35. Bd2. So, they were right in saying that Bxh2 was weak, but didn't really understand why it was.

It wasn’t a mistake.

Fischer had already anticipated winning games
3,5,6,8,10,13.

He threw the game to Spassky not wanting him to go winless in a legendary match.

No, instead I think he played the Steinitz gambit. Fischer had studied Steinitz's game thoroughly. And in both his World Championship matches against Tchigorin, Steinitz lost the first game of their match, lulling his opponent into a false sense of security, and then went on to win the match.

click for larger view
It is indeed a draw after 47...Kd7 as you indicated but I thought that after 47...Bh7+ that it might not be (snip)

I had a look to this position with shredder 12

click for larger view

Of course no final proof, but on d=40 shredder gives 0,00 after

48. Ke6 and 48. Ke5.

Even if it was a draw Fischer didn't find it and I don't know if he admitted after the game that he miscalculated but he clearly missed Spassky's defence.

So, all players, no matter their level either make "blunders" (outright "howlers") or they miscalculate. This is caused by the tension of playing real chess.

I overlooked a mate in one in the NZ Correspondence Championship in 1985 I think it was. Why?

If I knew that and we all knew why we all made all out "inexplicable" bad moves or "blunders" we would all be about 200 points or more rating points higher than what we are or were!