Perfect play

[corres]

[/quote]These things certainly can be refined in definition. Suppose we have the first ever perfect player: 32 men bases. It plays against the present day paradigm engines of say 3200, 3400, 3600, ... , etc ELO level. 32 men bases are not used smartly to fool around, say they are DTZ50 and just that. Then the ELO of the perfect engine will be probably close to what Maurizio shows.[/quote]
I think that if one would have a database of 32 man then the idea of perfect play and the toy of chess would be meaningless.
The quantum CPU and a quantum-based memory may be the tools to calculate and to store this database in the very far future. Until that days we and our chess engines can play chess "imperfectly".
In practice the perfect play is a mode of play what conduces to success.
So to be perfect is always a relative thing.
All world champions (human and engine) played perfect in their time.

[Thomas Zipproth]

Suppose we have the first ever perfect player: 32 men bases.

I wonder how 32 men bases can play chess.

If we assume that the starting position is draw, the 32 men bases would show the same score for 1.a3 and 1.e4: 0.00.
So we would need an additional score or criteria, how good the winning chances are in different 0.00 positions.

A possible way would perhaps be to let a standard engine play, and only avoid loosing lines and always play the winning line, if it exists.

[Ozymandias]

A possible way would perhaps be to let a standard engine play, and only avoid loosing lines and always play the winning line, if it exists.

That's what engines do, when they probe syzygy bases and are already down to six pieces.

[Henk]

Has this game already been solved ?

[d] 4k3/pppppppp/8/8/8/8/PPPPPPPP/4K3 w - - 0 1

[Laskos]

A possible way would perhaps be to let a standard engine play, and only avoid loosing lines and always play the winning line, if it exists.

That's what engines do, when they probe syzygy bases and are already down to six pieces.

Exactly, and this morning I had an idea to test, to dispel some myths about vagueness what perfect engine plays, what its elo would be, how it is rated. I simply became annoyed at Uri's continual tries to confuse things with issues which in fact need simplification and more clearness, making examples with extreme cases for the sake of saying "we don't know anything".

Solved chess: 5-men chess

I took 5-men theoretically drawn positions as start positions, in analogy with chess standard opening position which with the best guess is a theoretical draw.
The perfect player is SF Syzygy at 20k nodes. His adversaries are non perfect (no Syzygy) SF at 20,40,80k Nodes. Round-Robin:

Code: Select all

Rank Name                              ELO     +/-   Games   Score   Draws
   1 SF 80k                             6       1    6000     51%     98%
   2 SF 20k + Syzygy                    2       1    6000     50%     99%
   3 SF 40k                            -1       1    6000     50%     99%
   4 SF 20k                            -7       1    6000     49%     98%



   # PLAYER                : RATING  ERROR    POINTS  PLAYED     (%)
   1 SF 80k                :    9.7    0.9    3050.5    6000    50.9
   2 SF 20k + Syzygy       :    6.7    0.8    3017.5    6000    50.3
   3 SF 40k                :    4.6    0.8    2992.0    6000    49.9
   4 SF 20k                :    0.0    0.9    2940.0    6000    49.0

White advantage = 5.11 +/- 0.40
Draw rate (equal opponents) = 100.00 % +/- 0.12

Non-perfect Stockfish at 80k nodes per move is higher on the rating list than the perfect player SF 20k nodes per move + Syzygy. But looking at individual statistics, we see that the perfect player didn't lose a single game, and has beaten every opponent:

Code: Select all

  SF 20k + Syzygy                : 6000 (+ 35,=5965,-  0), 50.3 %

SF 20k                           : 2000 (+ 26,=1974,-  0), 50.6 %
SF 80k                           : 2000 (+  1,=1999,-  0), 50.0 %
SF 40k                           : 2000 (+  8,=1992,-  0), 50.2 %

It's just that SF 80k nodes has beaten weaker engines more decisively, probably exploiting by deeper search possible swings from draw to win.

So, aberrations amounting to the conclusion that perfect player has no ELO rating are debunked. Not only that, but the perfect player used here has a rating below some less-than-perfect players. This kind of perfect player will be close to the top, though. What Maurizio (an me) have derived, that perfect player has a rating between 3800 and 5000 CCRL ELO points, should be taken with these premises about perfect player and the current paradigm of engines.

[Lyudmil Tsvetkov]

While the average white advantage might be some 15cps for normal openings, I tend to believe lately that in a range of e4 and c4 lines, not mainstream though, with perfect play white edge might be as high as 25-30cps.

So it is a bit murky to me if in those lines the game outcome is closer to a draw or white win. Too many unknowns to conclude anything as of now.

With perfect play, there will be a single playable PV line, ending somewhere at move 294 or 365, with either a draw or white win.

We will have many perfect play (PP) engines: PP Stockfish, PP Komodo, and so forth, playing a single PP line. We will live in a PP world, unless we get bunker-bombed in the meantime...

[syzygy]

So, aberrations amounting to the conclusion that perfect player has no ELO rating are debunked.

Wasn't it Uri's point that "perfect play" does not equate to a particular Elo rating?

One "perfect player" is not the other "perfect player".

Assuming the opening position is a draw, lots of GM games will have been played perfectly by white until say move 30. But some of these games are then dead draws impossible to lose by a 2000-rated player and other games require someone stronger than Carlsen to hold the draw.

Playing perfect is nice, but chess is really about luring your opponent into making a mistake (or, if you are the weaker side, keeping the game under control so you minimise the danger of making a mistake). 32-men TBs will be of only limited help to the "better" player. (Of course they kill the game if the weaker player has access to them.)

[Laskos]

So, aberrations amounting to the conclusion that perfect player has no ELO rating are debunked.

Wasn't it Uri's point that "perfect play" does not equate to a particular Elo rating?

One "perfect player" is not the other "perfect player".

One particular perfect player can be defined, that was my point. That swindling, fooling around and who knows what perfect player would do, can have very different ELO remains true, I did not contest that. I just tried to exemplify to what particular perfect player our results for ELO of the perfect player may apply. It seemed to me that Uri tried to state with some outlandish analogies that any perfect player is something hard to define ELO-wise (and by that the conclusion that Maurizio's results are meaningless).

Assuming the opening position is a draw, lots of GM games will have been played perfectly by white until say move 30. But some of these games are then dead draws impossible to lose by a 2000-rated player and other games require someone stronger than Carlsen to hold the draw.

Playing perfect is nice, but chess is really about luring your opponent into making a mistake (or, if you are the weaker side, keeping the game under control so you minimise the danger of making a mistake). 32-men TBs will be of only limited help to the "better" player. (Of course they kill the game if the weaker player has access to them.)