< Earlier Kibitzing · PAGE 20 OF 21 ·
|Aug-23-15|| ||diceman: <kereru: Maybe Byrne's "premature" resignation reflects his strength as a player? Apparently the grandmasters commenting at the demonstration board didn't even know he was lost.>|
"Grandmaster who plays game, calculates
more than Grandmasters who comment on game."
|Aug-23-15|| ||AylerKupp: <Laziness, etc.> (part 1 of 2)|
<maxi> Oh, I'm just anal about computer analysis since different engines give different evaluations and suggest different moves as being best in any given position. And that evaluation changes with search depth, plus the greater the search depth the greater the confidence in the correctness of the engine's evaluation.
And I'm sure that you're aware that most analysts believe that Byrne's losing move was 14.Rfd1 instead of 14.Rad1. The point is that after 14.Rad1, 14...Nxf2 and the rest of Fischer's continuation is not possible. So, if indeed neither 14.Rfd1 and 14.Qc2 were the best moves in those positions, which one was the greater contributor to Byrne's loss? Hard to say.
I was interested in what your engine evaluated White's best 15th move would be. Fischer himself indicates (per Wade in the notes to this game) that there was hardly any defense against 15...Ne4 other than 15.Qc2. But a quick check with Stockfish 6 at d=34 show the following as White's best moves in this position:
click for larger view
1. [-0.71]: 15.Nf4 Ne4 16.Nxe4 dxe4 17.Rab1 Rc8 18.Bb4 Nxf4 19.gxf4 Qxd2 20.Rxd2 Bd3 21.Rbd1 a5 22.Bd6 Re6 23.Ba3 Bc3 24.Rxd3 exd3 25.Rxd3 Bb4 26.Bb2 Ree8 27.Bd5 Rc2 28.Bd4 Rd8 29.e4 b5 30.Be3 Kg7 31.a4 bxa4 32.bxa4 Bc5 33.Bxc5 Rxc5
click for larger view
Black is up the exchange for a pawn, has the better pawn structure, and has eliminated White's two bishops. He certainly has the advantage if not quite winning as far as Stockfish is concerned.
2. [-0.79]: 15.Nd4 Ne4 16.Nxe4 dxe4 17.Bb2 Rc8 18.a4 Bb7 19.Bf1 h5 20.a5 bxa5 21.Bc3 h4 22.Rxa5 a6 23.b4 Qc7 24.Ra3 hxg3 25.hxg3 Red8 26.Bh3 f5 27.Bf1 Be5 28.Ne6 Bxc3 29.Qxc3 Qxc3 30.Rxc3 Rxc3 31.Nxd8 Bd5 32.Bxd3
click for larger view
Initially I wasn't sure which recapture, 32...Rxd3 or 32...exd3 was better (I am after all a patzer) but restarting the analysis from this position Stockfish evaluates the resulting position as winning for Black, [-6.10], d=37, after 32...exd3. White's knight is immobilized by Black's bishop and it doesn't look like it can be saved after 33...Kf8.
3. [-0.84]: 15.Bf1 Ne4 16.Nxe4 dxe4 17.Nd4 Bxd4 18.exd4 Qd5 19.Qe3 Rad8 20.Rd2 Qxd4 21.Qxd4 Rxd4 22.Bb2 Rd7 23.Rad1 Re6 24.a4 Red6 25.Be5 Rd5 26.Bf6 h6 27.Bxd3 Bxd3 28.Rc1 Rc5 29.Rxc5 bxc5 30.Rd1 Rd6 31.Be7 Rc6 32.Rc1 c4 33.bxc4 Bxc4 34.Rd1
click for larger view
Black is a pawn up but with BOC the win might be problematic.
|Aug-23-15|| ||AylerKupp: <Laziness, etc.> (part 2 of 2)|
FWIW, Stockfish evaluates the position after your 15.Rb1 at [-1.36], d=32, just like you said, after 15... Bh6 16.f4 Rc8 17.Bh3 Rxc3 18.Nxc3 d4 19.Bf1 Rxe3 20.Bxd3 Rxd3 21.Qxd3 Bxd3 22.Rxd3 Qe8 23.Rbd1 Bf8 24.Bxf8 dxc3 25.Ba3 Kg7 26.Rxc3 Qe2 27.Rdc1 Qxa2 28.Bd6 Ng4 29.Be5+ Nxe5 30.fxe5 Qe2 31.e6 Qxe6 32.Rd1 Qe5 33.Rf3 Qc5+ 34.Kh1 Qc8 35.h4. I won’t bother to review it.
And, after 15.Qf2, Stockfish evaluates the resulting position at [-3.10], d=30 following Fischer's line after 15...Nxf2 16.Kxf2 Ng4+ 17.Kg1 Nxe3 18.Qd2 d4 19.Nxd4 Nxg2 etc. So clearly 15.Qf2 is a worse, nay, losing move after either 15.Nf4, 15.Nd4, 15.Bf1, 15.Rb1, and who knows how many other moves.
After 14.Rfd1 as suggested by the analysts Stockfish evaluates the resulting position at [-0.84], d=32 after 14...Nd3 and a continuation (perhaps surprisingly) into the line after 14...Nd3 15.Nd4 rather than after 14....Nd3 15.Nf4, showing how even a 1-ply change in the starting position changes the evaluation of the position and even the ranking of the moves.
But if White plays 14.Rad1 instead of 14.Rfd1, Stockfish evaluates the position at [-0.24], d=30 after 14...Qc8 15.h3 Qf5 16.Nd4 Qd7 17.Nde2 Rac8 18.Nf4 Bxf1 19.Kxf1 Qb7 20.Kg1 Ne4 21.Qxd5 Qxd5 22.Ncxd5 Nc3 23.Nxc3 Rxc3 24.Nd5 Rd3 25.Rxd3 Nxd3 26.Kf1 Nc5 27.Ke2 Rd8 28.Ne7+ Kf8 29.Nd5 Ke8 30.Bxc5 bxc5 31.h4 f5, and [0.00] after other moves.
Then the question becomes, is a move that changes the evaluation from [-0.33] (14.Rad1) to [-0.84] (14.Rfd1) considered more or less of a losing move than a move that changes the evaluation from [-0.71] (15.Nf4) to [-3.10] (15.Qf2)? I don't know. Numerically, the answer is clearly no, since the magnitude of the difference in the evaluation is much greater between [15.Nf4, 15.Qf2] than between [14.Rad1, 14.Rfd1]. But this would have to be confirmed with other engines since Stockfish's evaluations tend to be higher. And I am too lazy to do that. :-) So in practice I don't know if it really makes a difference and this might be a subjective answer.
Therefore this might be a case of, like Fischer once said, "White might play differently but in that case he just loses differently."
|Aug-24-15|| ||NeverAgain: <AylerKupp: Oh, I'm just anal about computer analysis>|
You should find this article of interest, then:
Stockfish (Computer) (kibitz #50)
<plus the greater the search depth the greater the confidence in the correctness of the engine's evaluation.>
Wish it were that simple. For each additional ply of search depth the engines have to discard an increasing range of moves (which is called pruning), as the number of possible moves increases in geometrical progression. All engines have to do this but some (like SF) are more aggressive with pruning than others. In practical terms it means that as you look deeper your field of vision narrows progressively. It's like seeing a forest in the distance - indistinct, but you can see a lot of trees. Train a telescope on it and suddenly you can see the individual cracks on the bark of a single tree - while the rest of the forest disappears.
|Aug-28-15|| ||maxi: <AylerKupp> Thanks for your analysis. About some 15 years ago I was a bit obsessed with computer analysis and its accuracy. I learned a lot about how the programs are written and eventually reached an understanding that was satisfactory to my (meager) standards. Basically I don't trust any analysis over 14 or so plys. I don't believe the advantage, as given by an engine, is meaningful as a measure of if a game is won by force or not. This brings us back to the Byrne-Fischer game.|
How did White get into this mess? The move 8.Ne2 allows Black to play e5 with tactical possibilities, but how could it ever lose?! White could also play 12.e5 instead of 12.Qd2; it seems better. Next week I will check this move with the computer at work which is kind of powerful. (It ain't the one the NSA has in Utah.) After 12.Qd2 Black has the initiative and, in practice, has good a good chance of winning. If Black has a won game perhaps you can tell me, I certainly don't know.
It seems to me that you showed that 15.Qc2 is a losing move, and that the other possibilities do not quite reach equality. But do they lose? Too bad the Oracle at Delphi does not exist anymore.
|Aug-28-15|| ||AylerKupp: <NeverAgain> As you can see in my response to your Aug-23-15 post in the Stockfish page, we are like ships crossing in the night. And, yes, I am a little bit familiar with search tree pruning :-) but you had no way of knowing that. What you say about narrowing your field of vision the deeper you get into the search tree is true enough, but remember that at each ply the engine reassess its PV. And the branches that got discarded at earlier plies may not, depending on the move ordering and the search tree pruning heuristics, get discarded at later plies. Yes, there is no guarantee that the branches containing the better moves that got discarded at lower plies will be retained at later plies but there is a probability > 0 that they will. And that improves the chances that a better move will be uncovered the deeper you search. It all depends on the search heuristics and whether the branches containing the better moves were discarded before or after the alpha-beta pruning.|
|Aug-28-15|| ||AylerKupp: <<maxi> Basically I don't trust any analysis over 14 or so plys.>|
A lot depends on what you mean by "trust". Will the game proceed along the lines that the engine indicated as being best by both sides? No, or at least not likely, even if the same version of the engine is playing itself. But searching deeper does provide some insurance against the horizon effect occurring earlier than later although, as you'll note in my pompously named "AylerKupp's Corollary to Murphy's Law" described in my forum's header, you can never eliminate it completely.
FWIW, one thing that I often do if the analysis evaluation looks a little bit suspicious is to restart the analysis at the final position. I call this "leaping forward" as a pun on "sliding forward". Its objective is not the same as in sliding forward (finding better alternative moves in the Principal Variation) but just to do a quick check to see if the engine's original evaluation was reasonable. Sometimes it is, sometimes it isn't. If it isn't, then I do some backward sliding to find out where the original analysis went sub-optimal.
You might want to update your knowledge of computer analysis; a lot has improved in the last 15 years, particularly in the area of search tree pruning. That's the main reason why today's engines can search so much deeper than earlier engines, even given the improvements in computer hardware. It might increase your confidence in deeper analyses.
And I think that Oracle Corporation is missing the boat by not moving its corporate headquarters From Redwood City, CA to Delphi, Indiana. Then they could legitimately call themselves "The Oracle at Delphi". And get a corporate tax break also, 6.5% in Indiana in 2015 vs. California's 10.84% for C-corporations. :-)
|Aug-29-15|| ||NeverAgain: <AylerKupp: As you can see in my response to your Aug-23-15 post in the Stockfish page, we are like ships crossing in the night.>|
Nice analogy. ;)
<but remember that at each ply the engine reassess its PV. And the branches that got discarded at earlier plies may not, depending on the move ordering and the search tree pruning heuristics, get discarded at later plies. >
It would be hard for me to remember something I didn't know to begin with. Thank you for the reply. Yours are some of the more interesting and informative posts among the sea of net.noise.
<AylerKupp: <<maxi> Basically I don't trust any analysis over 14 or so plys.>
A lot depends on what you mean by "trust">
I think what he referred to is a well-known saying in chess literature, something to the effect that any chess analysis longer than x (5? 10? don't remember) moves is bound to have a hole in it. The engines don't seem to have solved that problem 100%.
|Sep-08-15|| ||maxi: <AylerKupp> I tried 12.e4 (instead of Byrne's 12.Qd2) and got a score of about -1. So the idea is not good and Byrne's move seems the best in the position. The line is even if White plays... yes, 14.Rad1 (instead of 14.Rfd1). I used the computer at work; it has several cores and Houdini 4 Pro x 64B. No big news here. Sorry.|
|Feb-02-16|| ||thegoodanarchist: 21. White resigns came as a disappointment:
As Fischer wrote, after Byrne resigned: <"A bitter disappointment. I'd hoped for 22.Qf2 Qh3+ 23.Kg1 Re1+!! 24.Rxe1 Bxd4 with mate to follow shortly.">
|Mar-11-16|| ||socratos: wonderful|
|Mar-11-16|| ||Howard: One thing that's always puzzled me is why Byrne would just simply assume that Bobby had seen all the winning lines. After all, 23...Re1+ was an exceptionally hard move to find, and even the grandmasters who were analyzing the game for the spectators, didn't see it---they just couldn't understand why Byrne had just resigned.|
So, why wouldn't Byrne have played on for a few more moves just to make sure?
|Mar-11-16|| ||TheFocus: Byrne knew that if he could find it, Bobby would too.|
|Jul-21-16|| ||Howard: Sounds rather presumptuous to me---it would hardly have hurt for Byrne to have played on for only another 3-4 moves at the most, just to be sure.|
|Jul-21-16|| ||keypusher: <Howard: Sounds rather presumptuous to me---it would hardly have hurt for Byrne to have played on for only another 3-4 moves at the most, just to be sure.>|
No, it's the opposite of presumptuous. What would be presumptuous would be thinking that Fischer had sacrificed a piece for no reason.
Anyway, Byrne has written about this game, and is quoted here. After ...Nxg2, he realized what Fischer was up to, though I don't know when he worked out the full combination.
Everyone from Fischer on down is sorry he didn't keep playing, not because it would or wouldn't be presumptuous but because you just don't see moves like ...Re1+ over the board very often.
|Jul-26-16|| ||Howard: If I recall correctly, an article in British Chess Magazine about 15 years ago called "...Re1+" one of the greatest moves every played...|
....even though it didn't actually appear over-the-board!
|Feb-22-17|| ||Jimmy720: memorize|
|Mar-09-17|| ||RookFile: Usually when you play a King's fianchetto in these type of openings, it's a little safer for defense of white's king compared to more classical approaches. That's just one of the things that makes this game so remarkable.|
|Mar-09-17|| ||kevin86: Fischer was a brilliant player, though lacking in other facets of his life.|
|Mar-09-17|| ||offramp: In order to fully comprehend this game you have to read Fischer's comment at A Reinhard vs Fischer, 1963.|
|Mar-09-17|| ||MissScarlett: I believe the last word on this game was uttered by Elliott C Winslow.|
|Mar-09-17|| ||AylerKupp: <Howard> Maybe what British Chess Magazine actually said (or meant) was that ...Re1+ was one of the greatest moves never played ... :-)|
I have a magazine/book titled "The 100 Most Influential People Who Never Lived". People (characters) like Sherlock Holmes, James Bond, etc. Quite a fun read.
|Mar-09-17|| ||Petrosianic: In order to fully comprehend this game you have to read Fischer's comment at A Reinhard vs Fischer, 1963.|
That will only mislead you. That was made at a particularly unobjective time of Fischer's career, when he was saying things like that the King's Gambit was busted. Black isn't really better in that position, nor will that position help you to understand this game.
But if you'd really like to understand this game better, have a look at a few Black victories in the Tarrasch Defense.
|Mar-09-17|| ||Petrosianic: <Howard> <So, why wouldn't Byrne have played on for a few more moves just to make sure?>|
If it makes you feel any better, if Byrne HAD played on, you probably wouldn't have seen Re1 anyway. Since his resignation shows that Byrne DID see Re1, then he wouldn't have played that line.
Like, for example 22. Nf3 Qh3+ 23. Qg2 Qf5 24. Nd5 Bxd5 25. Rxd5 Qxd5, and White's an exchange down and totally busted, but it's not as flashy as the Re1 line.
|Mar-09-17|| ||sudoplatov: While there is always a "horizon effect" is always a problem, increasing what computer programmers term depth of search does not entail any narrowing of search trees. The alpha-beta always returns the value of the game tree that a full search would (it's provable). Problems creep in through traps deeper than a search or just an incorrect evaluation function of a leaf position.|
Leaf positions are supposed to be quiescent; no checks or captures or promotions (or a few other big types of moves). These operations are part of the evaluation function from a programming POV. Search depth refers to the depth before the quiescent postion.
I have always thought that grandmasters (ok Capablanca and a few others) see effectively infinitely far. They do not always look for long forcing lines, but rather look how to rearrange their pieces (and those of their opponent) to achieve a goal.
< Earlier Kibitzing · PAGE 20 OF 21 ·
Now on DVD