click for larger view

White: Kb1, Qc6, Be3, Re6, Rh1; Pawns - b3, c2, e4, & h2.

Black: Kf7, Qa5, Nf6, Rb8, Rc8; Pawns - a6, c3, e5, g6, & g7.

I spent 5-10 minutes on this problem, however, I did NOT get the solution! (I found the first couple of moves, but went astray later.)

I failed to appreciate the subtle Rd1! (The purpose of this move is to move Black's Rook off c8. Otherwise, Black will have ...RxP/b3+!; followed by ...c2+; winning.)

It looks like Black was forced to play 35...Rd8. (At least, I thought maybe he should.) 35...d8; 36.e7+ g8; (Or 36...h8; 37.xf6+ h7; 38.f7+ h8; 39.g1, winning for White.) 37.g1 g5; 38.xg5!

Like many others, you failed to consider 32...Rxb3+; and 33...c2+. (Read a few of the previous posts.)

Hiarcs evaluated 35. ... Rd8 as >6.0 within a second or two, and after a while, with a little of forward/backward analysis, settles on (+25.26).

I verified that the line found in your posts (indeed, initially it appears to be best for the black):

36.Qe7+ Kg8 37.Rg1 g5 38.Bxg5

resulits in a forced mate in 9. Here's the mainline:

Since many others mentioned 36 ... Rc7 defense, I did some analysis, coming up with this line that is pretty much best for both sides. The way I do this is run the engine in multi-PV mode, picking the substantially stronger lines, and branching on moves where several similar alternatives occur - then backtrack to mainline after exploring branches and pulling their valuations along during the backtrack. With sufficient memory, this accelerates the exploration of the position immensely, provided that a forcing line exists.

35. Rd1 Rc7 36. Rd7+ Rxd7 37. Qxd7+ Kh8 38. Qh3+ Kg8 39. Qe6+ Kg7 40. Qe7+ Kh8 41. Qxf6+ Kg8 42. Qxg6+ Kh8 43. Bg5 Rb6 44. Bf6+ Rxf6 45. Qxf6+

which ends up in this position (black to move):

click for larger view

Exactly my point. (I don't think like a machine ... and have never bothered to pretend that I do.) I only analyzed the continuation in question, because I was interested in it ... and therefore, it was possibly that someone else might be as well.

To be honest, unless someone bothers to explain the ideas behind a line, or gives analysis as meticulous as (say) <RandomVisitor>, I find such lengthy analysis boring.

I followed the Candidates Matches quite closely, and even the best players in the world failed to discover the "machine move" at many points ... no one made a big deal out of it.

Rd8 was one move (of several) that I failed to consider in my first (computer-free) post on page 1, so after you and <Peligroso> have mentioned it as winning for white, I decided to look at the continuation of the line you both mentioned, and had Hiarcs find for me a forced mate (AKA "horribly losing").

I suppose both you and I need to stop looking for criticism where none is implied.

Another point I want to make is many of the lines that the computer wants to look at are not playable, at least, not by (normal) human standards.

For example, when the computer starts playing "suicide chess" to prevent a mate ... the average human will resign before doing this. OR ... let's say the computer wants to give up its Queen as the only defense to a threat ... once more humans won't consider this. (It won't matter if there is even a five-point difference between two lines, any normal human isn't going to consider that for long.)

Another one of my pet peeves is when annotators discover an improvement ... usually with the computer ... and then they assign the old move a question mark. (The "Modern School" of annotation.) To me, this is silly. (A good example in this game was that Browne missed an improvement on his 26th move in this game, see my earlier posts.)

I have annotated hundreds of "older" (pre-computer) games, [not all of them get posted on the Internet]; and the machine has found dozens of improvements. Maybe Browne saw 26.Nxe7 (followed) by Rxd6), but did not analyze it deep enough to see the win.

To me, none of this matters. I much prefer the way humans play chess. Just about every GM game I have examined in recent times - Fritz has found an improvement at some point.

Of course, the search for truth is an entirely different matter. For example, when "RandomVisitor" discovers a new move ... in a heavily analyzed position, where many GM's have looked at the game before ... and discovers a new move, this can be exciting. Its also amazing to see how the computer finds shots a human won't even look at, (unless your name is Kasparov, Fischer or Tal).