Bridgeburner: PART 1

Lasker vs Schlechter, 1910 is the fourth game of the title match, and continues the very high standard of play. Quantitative mapping of this game between these players is below. Figures in brackets are the corrected engine evaluations generated on the return slide. The reverse slide smoothed out many, but not all fluctuation in the engine’s evaluations. The complexity of some variations was very likely too great to enable a fuller reconciliation from the forward slide. General methods used are described in the bio.

The evaluation values in the opening are of no great moment, but of interest is that they come at the end of a full forward slide to the end of the game and a full return slide from the last move to the first. Not flawless like game one, and not edgy like game 7, the accuracy by both players kept this game “quiet”, masking subterranean activity that is worthy of close study. Lasker held some advantage for the whole game bar one move, and came close to winniing, but tenacious defense by Schlechter and a missed opportunity by Lasker ensured the draw. Schlechter had one opportunity to gain advantage on move 8, when <8…Nxd4> would have been advantageous for Black.

<1. e4> ( 0.20) <1…e5> ( 0.26)

<2. Nf3> ( 0.26) <2…Nc6> ( 0.32)

<3. Bb5> ( 0.32) <3…a6> ( 0.32)

<4. Ba4> (=0.12) <4…Nf6> (=0.12)

<5. 0-0> (=0.12) <5…Nxe4> (=0.14)

<6. d4> (=0.14) <6…b5> (=0.14)

<7. Bb3> (=0.14) <7…d5> (=0.14)

<8. a4> (=/ 0.37)

Capablanca was unimpressed with this move and so was the engine. Black can now gain the advantage with <8…Nxd4>.

<8…Rb8> ( 0.40)

<9. axb5> ( 0.40) <9…axb5> ( 0.45)

<10. dxe5> ( 0.45) <10…Be6> ( 0.45)

<11. c3> ( 0.40) <11…Be7> ( 0.63)

<12. Nbd2> ( 0.45) <12…0-0> ( 0.45)

<13. Nd4> (=0.18) <13…Nxd4> (=0.18)

<14. cxd4> (=0.18) <Nxd2> (=0.18)

<15. Bxd2> (=0.18) <15…c5> (=0.09)

<16. Bc2> (=0.03) <16…cxd4> ( 0.30)

<17. f4> ( 0.30) <17…f5> ( 0.30)

<18. exf6> (=0.17) <18…Rxf6> (=0.17)

<19. f5> (=0.22) <19…Bf7> ( 0.31)

<20. Bf4> ( 0.31) <20…Rc8> ( 0.31)

<21. Bd3> ( 0.31) <21…Qd7> ( 0.78)

<22. Be5> (=/-0.78) <22…Rh6> ( 0.95)

Black’s last chance to swing the rook back to coordinate with the other rook on the queen side. White’s next move forces Black’s response, completing the rook’s lock out, and effectively splits Black’s forces.

<23. Qg4> ( 0.95) <23…Bf6> ( 0.95)

Other moves lose (ie: rating evaluations jump to over 1.40 ). It sidelines Black’s rook.

<24. Bxf6> ( 0.78) <24…Rxf6> ( 0.78)

<25. Qxd4> ( 0.78) <25…h6> ( 1.05)

<26. Rfc1> ( 0.58) <26…Rfc6> ( 0.70)

<27. Rf1> ( 0.70) <27…Qd8> ( 0.70)

It’s not clear why Capablanca gave this a ?. White is threatening <28.Bxb5> and <28.f6>.

<28. Ra7> ( 0.39) <28…b4> ( 0.72)

<29. Re1> (=0.13) <29…Rc1> ( 0.27)

<30. Rxc1> ( 0.27) <30…Rxc1+> ( 0.27)

<31. Kf2> ( 0.27) <31…Rc7> ( 1.10)

An evaluation shift of 0.83 representing a “bad move”, followed by slight inaccuracies by Black at moves 33 and 34 pushed the game into losing zone, something that was indirectly noted by Capablanca when he complained about White missing the winning move/strategy at move 35. The engine evaluated that <31…b3> would have maintained near equality, and that <31…Rh1> wasn’t far behind.

<32. Rxc7> ( 1.05) <32…Qxc7> ( 1.05)

<33. Qxb4> ( 1.05) <33…Qe5> ( 1.36)

Schlechter probably considered the centrality of the Queen to be more important than collecting White’s pawn at <h2>. However, <33…Qxh2> is considered by the engine to be the best move, followed by <34…h5> and <35…h4>. It’s notable that Lasker responded with <g3> safeguarding the pawn even after Schlechter had eschewed the immediate option, and that Schlechter engaged the <h5/h4> attack anyway.

<34. g3> ( 1.36) <34…h5> ( 1.56)

Should have been the losing move, as noted by Capablanca and confirmed by the engine when White failed to find the winning continuation (<35.Qc3>) on the next move.

<35. Qb6> ( 1.22) <35…h4> ( 1.22)

<36. b4> ( 0.49)

The game essentially flat lines into a draw at this point.

<36…hxg3+> ( 0.49)

<37. hxg3> ( 0.49) <37…Qb2+> ( 0.49)

<38. Kf3> ( 0.49) <38…Qc3> ( 0.49)

<39. Qd8+> (=0.01) <39…Kh7> (=0.01)

<40. Qh4+> (=0.01) <40…Kg8> (=0.00)

<41. Qd8+> (=0.00) <41…Kh7> (=0.01)

<42. Qh4+> (=0.01) <42…Kg8> (=0.01)

<43. Ke2> (=0.01) <43…Qb2+> (=0.01)

<44. Ke3> (=0.01) <44…Qc1+> (=0.24)

<45. Ke2> (=0.01) <45…Qb2+> (=0.01)

<46. Ke3> (=0.01) <46…Qc1+> (=0.24)

<47. Kf2> (=0.24) <48…Qb2+> ( 0.47)

<48. Kg1> (=0.13) <48…Qc3> (=0.13)

<49. Bf1> (=0.13) <49…d4> (=0.13)

<50. Qd8+> (=0.13) <50…Kh7> (=0.20)

<51. Qd6> (=0.01) <51…Qe3+ > (=0.01)

<52. Kh2> (=0.01) <52…Qd2+ > (=0.01)

<53. Kh3> (=0.01) <53…Qe1> ( 0.29)

<54. Ba6> (=0.02) <54…Qh1+> (=0.03)

<55. Kg4> (=0.03) <55… Qd1+> (=0.03)

<56. Kg5> (=0.03) <56…Qc1+> (=0.02)

<ANALYSIS>

<Note> The fluctuations generated in the relatively low (16 minimum) ply forward slide were smoothed out in the equivalent return slide. The corrected evaluations extracted from the return slide are used in this analysis, as they are considered more reliable than the raw evaluations generated on the initial forward slide. All moves have been evaluated on forward and return slide for completeness.

<Evaluation range>: Between < 0.37> applying to the move <8. a4> - representing a slight advantage to Black - and < 1.56> in respect of the move <34…h5> representing a winning advantage for White. However, neither player took advantage of the potential advantages offered by these moves.

<The largest evaluation shift>:

- for White was <0.73> between <35…h4> (< 1.22>) and <36.b4> (< 0.49>), coming close to the evaluation shift representing a bad move. <35.Rf7> was the engine’s preference for preserving White’s advantage.

- for Black was <0.83> between <31.Kf2> (< 0.27>) and <31…Rc7> ( 1.10>), inside the range of an evaluation shift representing a bad move. <31…b3> was the engine’s preference for maintaining equality for Black’s advantage.

<Computer statistics>:

• 89.3% of the ply in this game (100/112) coincided with engine preferences 1, 2 or 3

• 79.5% of the ply in the game (89/112) coincided with engine preferences 1 or 2

• 64.3% of the ply in the game (72/112) coincided with the engine’s first preference

• 92.9% of Schlechter’s moves (52/56) coincided with the engine preferences 1, 2 and 3

• 85.7% of Lasker’s moves (48/56) coincided engine preferences 1, 2 and 3

• 80.4% of Schlechter’s moves (45/56) coincided with engine preferences 1 and 2

• 78.6% of Lasker’s moves (44/56) coincided with engine preferences 1 and 2

• 62.5% or 35/56 of Schlechter’s moves coincided with the engine’s first preference

• 66.1% or 37/56 of Lasker’s moves coincided with the engines first preference

<The engine evaluation of the final position>:

was <=0.03> in a position agreed drawn.

<CONCLUSION>

This was another accurate game by both players. Neither player blundered and only Schlechter made a move that met the project definition of a <bad move>, namely <31…Rc7>. Lasker played one move <36. b5>, that was very close to being a defined bad move, although <35.Qb6> very likely missed the win.

End note: White’s move 35 provides a challenge for the methodology: should a move that concedes the win be considered a “bad move” for the purposes of weighting the game? The evaluation shift was only 0.33, but reduced the game evaluation below the threshold for a winning game.

While it makes notional sense to weight such a move as a "bad move", it can't be justified in terms of the quantitative method employed, as a concurrent method for re-evaluating blunders would have to be employed.

The advantage threshholds are drawn within the continuum of engine evaluations to distinguish between <equality> (-0.25 0.25), <slight advantage> (0.26-0.79), <significant advantage> (0.80-1.40) and <winning advantage> (over 1.40). The negative scale naturally applies to Black.

Consequently, any alteration to the definition of a bad move or a blunder resulting from a game evaluation crossing any of these threshholds, including the winning advantage threshhold, would need to be proportionately scaled as the evaluation shifts approach a threshhold if it were to be done consistently.

<While it makes notional sense to weight such a move as a "bad move", it can't be justified in terms of the quantitative method employed, as a concurrent method for re-evaluating blunders would have to be employed.>

It would become quite messy if you change the criteria for <bad move> now because <a concurrent method for re-evaluating blunders would have to be employed>.

16. Bc2 by Lasker actually sacs a pawn, with positional compensation - Black doubled pawns, White Kingside space advantage and the initiative. There are no large jumps in the computer's eval, probably precisely because there was compensation?

Capa notes that 28...b4 is an attempt by Schlechter to sac the pawn back in order to attain drawing chances.

(I find it totally weird why there are comments by kibitzers in effect saying the top masters of Lasker's era did not know when or when not to sac a pawn for positional compensation or for the initiative. They were doing it all the time. Even Anderssen was doing it all the time 50 years before this.)

After 34. g3, Schlechter probably realized he should have played h5 a move before. However, in that position, what other option did the computer give aside from 34...h5? Schlechter decided to play it anyway, a move late.

Earlier, 31... Rc7 at first sight looks like Schlechter automatically banging out his moves to just beyond time control. He probably had decided to neutralize White's Rook on the 7th rank a few moves earlier, and played on inertia. However, at this point, Black's Rook on White's 1st rank is at least as active as White's Rook on the 7th rank. Schlechter should have re-evaluated the position first. I would guess that maintaining Rooks would have increased chances for drawing. Perhaps Schlechter could have used his Rook to threaten attacks and checks on the now exposed White King. I believe that the time control was 30 moves for 2 hours, and it has always been a common occurrence for inaccuracies to be made just before, and for some reason just after, time control.

After 8. a4, was Capa trying to be humorous? "Why he plays it again can hardly be understood." I had to smile on that one.

I think Lasker may have wanted to try an improvement if Schlechter had contiunued with 8... Nd4. I think that Schlechter, wary of any potential opening novelty prep by Lasker, decided to sidestep the entire line by deviating first himself. Unfortunately, it turned out to be an inferior move.

Although Capa was trying to be funny, I have seen him do the exact thing that Schlechter did, that is sidestepping possible opening preps by his opponents. At this point in time, apparently top players such as Lasker, Schlechter, Rubinstein, Capablanca were playing such high level chess that it was easy to get blown off the board if any one of them badly misplayed the opening. Schlechter deviated first because he was scared he would get massacred by a Lasker improvement. Just goes to show that the top masters of this era were already thoroughly booked and were conducting intense over-the-board opening theory debates and psychological warfare.

Schlechter as you say played superlatively in an ending that <can’t be said to be an easy draw>. We have seen many GMs live in the internet collapsing in difficult endings. This was a Queen ending, the most difficult ending to play for the human eye, but Schlechter went through it accurately without an error and survived.

See Capa's note on move 35 above. According to Bridgeburner's computer analysis:

<34. g3> ( 1.36) <34…h5> ( 1.56)

Should have been the losing move, as noted by Capablanca and confirmed by the engine when White failed to find the winning continuation (<35.Qc3>) on the next move.

<34. g3> ( 1.36) <34…h5> ( 1.56)

Should have been the losing move, as noted by Capablanca and confirmed by the engine when White failed to find the winning continuation (<35.Qc3>) on the next move.

<35. Qb6> ( 1.22) <35…h4> ( 1.22)

<36. b4> ( 0.49)

Then

Lasker:

1 blunder

<35. Qb6> missing the winning <35.Qc3>

Schlechter:

1 bad move/ error

<31…Rc7> (0.27) to ( 1.10) eval shift of 0.83. For purposes of this study, a <bad move> or an <error> is defined as a move that causes an engine evaluation shift of between 0.80 and 1.20

1 blunder

<34…h5> pushing the position evaluation to greater than 1.40

------

Therefore, the corected results are:

Under weighting method A, weighting is 5.0 (no bad moves and 1 blunder by Lasker and 1 bad move and 1 blunder by Schlechter).

Just a minor suggestion. All these numbers are practically meaningless without giving the depth of search and the engine used. The engine could well see a at a depth of 15 but at a depth of 20. Rybka will value the position different than Shredder or Fritz.

Although there is a certain fascination with these kind of evaluations, not much can be learned from them. The student is better served by playing through the variations and asking questions about the moves.

The background of the study is in Luis Ramirez de Lucena

There are also some discussions in Arpad Elo

If you would take a quick peek at User: Bridgeburner, you could see that as of now 6 games of the Lasker-Schlechter World Championship Match (1910) and 3 games of the Anand-Kramnik World Championship Match (2008) have been analyzed.

If the trend for error and blunder rates from the 6 analyzed games continue, it seems that Lasker and Schlechter were playing just as well as Anand and Kramnik; but the the study is not finished yet. We will draw partial conclusions after the study is completed.

Thanks for the heads up. I see the link to Bridgeburner in your 7/29 post. He/she has taken the time to do a more thorough analysis than we usually see from engine jockeys. Since you devoted so many lines to this, I think explaining the engine setup would have been reasonable.

Those interested in engines will find all this interesting but the value for the student is minimal. Capa's brief comments are worth more to the student than all these numbers. Perhaps a link to Bridgeburner's forum with some results attached to Capa's comments would be more appropriate for a game page.

[1] Emanuel Lasker, 'Pester Lloyd', 1910.01.20, p. 6