3 mistakes that lost the game

[Lyudmil Tsvetkov]

The 16th game of the TCEC final was another confirmation that Komodo handles closed-prone positions better.

Where did SF go wrong?

I see 3 fatal moves.

[d]r2q1rk1/pppb1nbp/2np2p1/3NpP2/2P5/3P2P1/PP1BNPBP/R2Q1RK1 b - - 0 11

Komodo has just captured ef5, whcih I think is a mistake, and SF could have been left with a better game after gf5. However, SF recaptures Bf5. Why so?
SF obviously thinks that the h pawn otherwise becomes isolated, and the bishop would have worse mobility, but in the particular situation control of center is much more important.

After Bf5, white will place a knight on e4, which should be regarded as outposted.

[d]1r3rk1/pp2qnbp/3p2p1/2pPpb2/2P1N3/1Q4PP/PP1B1PB1/4RRK1 b - - 4 19

SF now plays Qd7.

Well, black struggles to fully equalise, so why not go the straigth equalisation way, by playing Bh6, followed by Be4?
The knight on e4 should be exchanged as soon as possible, it is factually an outposted knight, a central outposted knight in a semi-closed-prone position, and as such worth more than a bishop, more than even the pair of bishops. Komodo might not perfectly be aware of that, but, as it generally values knights higher as usual, it handles this position well.

After Bh6 and Be4, the draw is simple.

[d]2r3k1/pq3nbp/r2p2p1/2pPpb2/4N2P/1PB3P1/P3QPB1/4RRK1 b - - 4 26

Here however comes the final, outright mistake, that already loses the game. Up until this point, the game could probably still have been saved, but after black's disastrous 26...c4, there are no salvation chances.

Why is this a very major positional mistake?
Very simple, after c4, black creates a passer, that is however easily stopped/blocked, meaning it is of no value at all, while at the same time, after white's b4 response, white creates a candidate on the b file, that is much more dangerous than the black c passer, as in sharp distinction, it can not be stopped/blocked.

This means that the white b candidate passer/soon to become passer is taking an active part into the game, while the black c passer is completely out of the game, not influencing the fight in any way. Therefore blocked/stopped passers should be penalised, under certain very specific conditions, of course, which might be difficult to successfully test even in a large framework with many contributors.

And indeed, the b candidate wins Komodo the game.

Let's hope for other interesting games further on.

[cdani]

Thanks!!

[Lyudmil Tsvetkov]

Thanks!!

What for?

I am urgently waiting for someone to point out why I am wrong.

[fern]

Excellent analysis, full of positional acumen.
You looks the game as I do, with the difference, perhaps, that I surely commit many more tactical mistakes.
Chess is won or lost 99% of times in that field....

Fern

[cdani]

Thanks!!

What for?

For the analysis!

[Lyudmil Tsvetkov]

Thanks!!

What for?

For the analysis!

The Spanish-speaking presence in this thread becomes menacing.

[Lyudmil Tsvetkov]

Excellent analysis, full of positional acumen.
You looks the game as I do, with the difference, perhaps, that I surely commit many more tactical mistakes.
Chess is won or lost 99% of times in that field....

Fern

I try to consider the game as you do, Fern...

[carldaman]

The 16th game of the TCEC final was another confirmation that Komodo handles closed-prone positions better.

Where did SF go wrong?

I see 3 fatal moves.

[d]r2q1rk1/pppb1nbp/2np2p1/3NpP2/2P5/3P2P1/PP1BNPBP/R2Q1RK1 b - - 0 11

Komodo has just captured ef5, whcih I think is a mistake, and SF could have been left with a better game after gf5. However, SF recaptures Bf5. Why so?
SF obviously thinks that the h pawn otherwise becomes isolated, and the bishop would have worse mobility, but in the particular situation control of center is much more important.

After Bf5, white will place a knight on e4, which should be regarded as outposted.

Hi Lyudmil,

Short on time, so I'll just comment on the 1st diagram. The thematic and usually correct recapture in KID-type positions is gxf5 for the reasons you gave. One might want to also go back to another reason you had previously brought up yourself, that being the phalanx - gxf5 produces a better (mini-) phalanx for Black! (gxf is so strong in the KID proper that White does not normally take on f5, and instead lets Black create a strong pawn chain with f5-f4)

If the phalanx concept can be better quantified and codified, such errors in engine evaluation can likely be corrected.

Regards,
CL

[M ANSARI]

I think that a lot of positions that look like the engine made a mistake for a human ... are actually not mistakes at all. You have to realize that the engine can see extremely deep and can see tactics in its search that a human will never even know exist. A tactic that changes material balance without a long term disadvantage is all it takes for a win. That is the reason why humans have a horrible time beating engines ... they can play well positionally, but then comes along an obscure tactic that screws up a totally nice and symetrical position and the game is immediately lost. In hindsight it is always easy to block out all possible moves and pick one that might have a better chance.

In TCEC there is a very nice feature which allows you to see what line the engine sees as critical. What I thought was really strange was the Houdini and to some extent SF ... would have relative mild evaluations when their end positions in the analysis window was completely lost! Komodo seemed the most sane by far it its evaluation end diagrams.

[carldaman]

I think that a lot of positions that look like the engine made a mistake for a human ... are actually not mistakes at all. You have to realize that the engine can see extremely deep and can see tactics in its search that a human will never even know exist. A tactic that changes material balance without a long term disadvantage is all it takes for a win. That is the reason why humans have a horrible time beating engines ... they can play well positionally, but then comes along an obscure tactic that screws up a totally nice and symetrical position and the game is immediately lost. In hindsight it is always easy to block out all possible moves and pick one that might have a better chance.

In TCEC there is a very nice feature which allows you to see what line the engine sees as critical. What I thought was really strange was the Houdini and to some extent SF ... would have relative mild evaluations when their end positions in the analysis window was completely lost! Komodo seemed the most sane by far it its evaluation end diagrams.

Engines have taught humans quite a few tricks as well, but here we're talking of known principles where the engines cannot give a definitive answer and accumulated human knowledge is still very valuable. In the above example, Bxf5 is playable, but gxf5 is more in the 'spirit' of the position. Even top engines can lack this spirit at times . Better eval is certainly needed in such cases.


As to your second point, the last few moves together with the end positions in the PV are often not worth looking at. The further out you go from the root, the less reliable the analysis displayed in those PVs...

Regards,
CL