I reached the following position in an unrated game.

I analyzed mate in six (6) at most and in five (5) the least.

The computer gave a question mark "?" to my first move which would have led to mate in six (or 5).  Guess what Black's first move was.

REPLAY THE DIAGRAM AT THE LEFT TO SEE MATE IN SIX.

I had the game analysed by the Chess.Com computer.

REPLAY THE DIAGRAM AT THE LEFT TO SEE HOW THE COMPUTER ANALYSED THE POSITION.

NOTE:  The computer did not go for mate in 3 beginning with 6...Qf3 but instead went pawn grabbing with 6....Qxa4.

Chess.com does not have enough CPU to spend on extremely thorough analysis of all the games. However, since processing power doubles every 18 months, you can expect the analysis to be much better by their next upgrade.

Thanks for sharing that example, Matalino.  I've often gotten the feeling that the computer analysis people talk about (and get) here is a bit...well, goofy.  Your solution is neat, and quite amusing, but it would hardly require Hydra to work out all its ramifications; once you see that "box" for the White king, the rest becomes a given (indeed, I'm surprised that any chess analysis software would--or could--miss something that clear-cut).

my engine took just 15 secs to find this

Analysis by Rybka 2.3.2a 32-bit :

1. -+  (-#7): 1...c4 2.g3 c3 3.Kg2 c2 4.h5 c1Q 5.Kf3 Qe3+ 6.Kg2 Qxg3+ 7.Kf1 Qf2#
Not mate in 6 but very close.

{Deep Fritz 11:} 1... Bg3 2. h5 Ne3 3. Kg1 c4 4. h6 gxh6 5. Kh1 c3 6. Kg1 c2 7.
Kh1 c1=Q# {-#7/12}.

Found immediately. My CPU is Core 2 Duo E8200 2.66 GHz.

Odd that the computer did miss that and even more odd is the fact that he plays 8... Bd4 when it is mate in one with Qh2.

And odd that even monster crunchers like Rybka and Deep Fritz aren't coming up with the quickest mate.

I wonder if any of the top 10 human players would fail to find the quickest mate in this position ? I seriously doubt it.

It's not too surprising that the computers fail to get the mate in 6. They are distracted half-way there by the two pawn swing of taking the h-pawn rather than letting the g-pawn be taken. The heuristics simply cut out the line allowing the pawn capture on g7 since taking on h6 is +2 better.

I first started to doubt the abilities of the site machine, when the following occurred:

I played a game of 5 min against solid opposition. Three moves gave me trouble and I took the time to think. My opponent dropped a piece and resigned immediately. I analysed the game through the site machine. It said that my opponent (although in the position that my opponent blundered, was clearly worse... machine and I concur) had played perfectly. I had made three inaccuracies... all the moves I had thought about, and the very ones I believe had led to the strong position.

I started to read the site analysis more carefully (especially the bit where it says "game may have continued..."), often finding that although my final move was strong enough to force immediate resignation, the machine preffered the aquistion of material. All in all, not a very satisfying or sometimes believable facility.

yes that bothered me as well. my run was only 15 secs reaching just 15 ply. so to satisfy my own curiosity i will rerun it on my big machine and leave it run and see how long it takes to find that mate in 6.

CM 9000 on my lappy took all of 7 seconds to find the mate in 6, although it decided to put the knight at h2 instead.

rybka found the mate in 6 after reaching 19 ply overnight.

this bears out my thought that under 21 ply rybka is not to be totally trusted to find the best answer

That's really strange, my chess engine (Glaurung) found the mate in 7 instantly, and switched to the mate in 6 after 1 second 

Also, this shows the problem with fixed depth analysis. With forcing moves present, it is very easy to reach higher depth and find the mate, but chess.com cuts off the depth (at 10 ply for gold I believe). The analysis tool is helpful to spot obvious errors, but don't but all your faith in it as always correct.

Interesting thread. thanks for sharing. By no means is our computer analysis going to be right 100% of the time. Its a continuous process for us as we try to improve it and look for better/faster engines. You also have to keep in mind that we are analyzing an ENTIRE game, every position, so the deeper ply we configure our analysis to go, the longer it will take to get results back on your games. Also I noticed you are a gold member, so you only get 10 ply analysis, whereas platinum/diamond would get 12 ply analysis. that will probably affect the outcome as well.

i have Chessmaster 11 on my laptop (intel core Duo processor,512mb ddr2 sdram) and it only took it like 2-3 seconds to find mate in 6.it was almost identical to your mate except it put the knight on h2 instead.GO CHESSMASTER.

Interesting that Rybka has to search to a depth of 19 ply to find a 12 ply mate. At least according to PerfectGent Rybka didn't find the 12 ply mate at a 15 ply depth but did find it at 19 ply!

even chessmaster 10th ed. found mate in 6 , in less than 2 sec.

So, this entire thread is proof positive that the best human players are still better than the monster machines in some aspects of the game ! At least, its proof enough for me. Oh, I put the position on Shredder and it found the mate in 6 but took all of 6 seconds and it also put the N on h2 instead of e3 but the result is the same.

Crafty 20.14 found it in about 30 seconds, on a PC with AMD Athlon 64 X2 Dual Core Processor 3800+ 2.00 GHz and 2 GB RAM.  I wonder what engine they use here?

At any rate, this isn't exactly a news flash -- computer analysis should always be taken with a grain of salt.  Contrary to popular opinion, computers are not infallible chess players; their long-term strategic vision is and has always been poor compared to strong human players.  What they are excellent at is pointing out tactical opportunities, both played and missed.  Don't let their shortcomings cause you to discard a very useful learning tool.