< Earlier Kibitzing · PAGE 43 OF 43 ·
|Feb-22-14|| ||DcGentle: <AylerKupp>: Nevermimd, I was glad I could help him.|
By the way, the following position is not hopeless for engines to solve, and you needn't run them overnight. It's taken from a correspondence game, but I found it on a different site:
click for larger view
FEN: rn2k2r/p4p2/2p1p2b/1bNqN2p/2pP2pP/2Q5/pP2BPP1/R2R2K1 w kq - 0 1
White to move and win. Most engines have trouble, Houdini 4 finds it.
Actually there are positional elements, although tactics rules, so I was a bit surprised that many engines will fail.
|Feb-24-14|| ||cro777: Yesterday, in the ICC versus Komodo online voting match, as expected, the man (team) was tactically outplayed by the engine.|
|Feb-24-14|| ||DcGentle: <cro777>: It was not a consultation game and users were crazy enough to let themselves in for a tactical skirmish.|
Perhaps they thought that the majority would iron out single mistakes. There we have it, a majority decision can go wrong as well.
|Feb-24-14|| ||AylerKupp: <Possible implementation of "Swindle Mode" in a chess engine> (part 1 of 3)|
There has been much discussion in the Zurich Chess Challenge (2014) page about strengthening an engine's play to encourage it to play a move resulting in a line with a lesser evaluation than the top ranked move in an attempt to swindle the engine's opponent into playing a correspondingly poorer move in return, allowing the engine to change the most likely outcome of the game if the objectively most accurate move were played. In a discussion in Zurich Chess Challenge (2014) I postulated that there would not be much point in implementing this "swindle mode" in current engines since it was not likely to work against strong opponents (human or computer) and it would not be necessary against weaker opponents (again, human or computer) since the engine would likely have reached a winning position by current methods. Further, by diverting some of the available machine cycles to these swindle mode calculations, it would prevent the engine from performing some of the "standard" calculations it would perform during that time and therefore actually weaken rather than strengthen the engine. So, in order to implement a swindle mode without significantly weakening the engine's traditional computational capabilities, the number of additional calculations must be kept small.
First, what could make an engine play a move that resulted in less than the best evaluation for the given line? The minimax algorithm will select as its Principal Variation (PV) the line that represents best play by both sides among the tree of possible moves examined, i.e. the highest move <evaluations> (typically the highest positive number for White moves and the lowest negative number for Black moves). One way to modify this is to calculate the highest <expected> evaluation for each side rather than just the highest <evaluation>. The expected evaluation = (probability of opponent playing the expected response to our move) x (evaluation achieved by the move) or P(Mn) x E(Mn) (Probability of Move n being played x .Evaluation if Move n is played). This is typical of defining risk as the probability that an event occurs x cost of that event occurring. Since the standard minimax algorithm will, by definition, select the opponent's strongest responses (E(Rn), from the engine's perspective) to any move that the engine plays, the probability, P(Rn), that this will be the move played by the opponent in response to the engine's move is 1 and for other moves is 0.
But what if the probability of the opponent playing its strongest response was not 1? That is, after all, the essence of a swindle; instead of playing the best move (the move that leads to the PV representing best play by both sides), a less than best move is played that induces the opponent to play an even worse move in return. This of course assumes that the engine's opponents, human or computer, will play less than their best response to the engine's move.
But all of us have experienced the situation in which we have calculated a "winning" variation consisting of (what we thought) was a forced sequence of moves and have an opponent surprise us by playing a move that we didn't think was best and, perhaps not coincidentally, turning the supposedly winning variation into a losing one. Engines experience a similar situation; they evaluate their PV on the basis of <their> evaluation function applied both to its moves and its opponents expected moves. But different engines have different evaluation functions; just because an engine evaluates a response as best doesn't mean that the other engine will evaluate the same response at best. And, given the non-deterministic nature of chess engines, particularly multi-core engines, the same situation may arise if the engine was playing against itself.
|Feb-24-14|| ||AylerKupp: <Possible implementation of "Swindle Mode" in a chess engine> (part 2 of 3)|
Another way of looking at it is interpreting P(Mn) x E(Mn) as the <expected gain>. If the engine is considering two moves to play, Ma and Mb, and looking at two opponent responses to those moves, Ra and Rb respectively, and if the evaluation of the line after Rb, E(Rb), is less favorable to the opponent than the evaluation of the line after Ra, E(Ra), then the sequence Mb + Rb will be eliminated as a candidate PV since it would not represent the opponent's best move. But if P(Rb) is sufficiently high (and hence P(Ra) is correspondingly small, then the expected gain P(Rb) x E(Rb) could be greater than P(Ra) x E(Ra) and therefore Rb will be the selected response. Granted, this is not likely to happen very often. But swindling possibilities don't happen very often either, yet that doesn't mean that one should never look for them.
So the key is evaluating the probability that Rb rather than Ra will be played by an opponent. But how could an engine do that? One way is to play the game with Ponder=ON. In this mode an engine uses its opponents thinking or calculating time (depending on whether it is playing a human or another engine) to effectively slide forward one ply and improve its PV by searching deeper. If the opponent does not play the response expected in its PV, nothing is lost; the engine will restart its calculations from that new position as though Ponder=OFF. But if the opponent does play the expected response, then the engine can either play the next move in its PV immediately and save some time on its clock for future calculations or it can continue to improve its PV, reaching a deeper search ply that it would have been able to achieve if Ponder=OFF.
But the fact that the engine's opponent did not play the expected response to the engine's last move means that, from the engine's perspective, it played an inferior move; possibly the 2nd best, 3rd best, or simply a move that the engine had not considered. Most of the top engines have an MPV mode when the engine can calculate not only its PV but its second best move/line (MPV=2), third best move/line (MPV=3), etc. So the engine can keep track of how many times its opponent plays the expected response to each of its moves (i.e. the "best" move from the engine's perspective) and, if the engine's opponent's response is in the engine's MPV=2, how many times the opponent plays the second best move; if in the engine's opponent's response is in the engine's MPV=3 the third best move, etc. and therefore calculate the probabilities of the opponent playing less than what the engine considers the best move. Of course, there may be many times that the engine's opponent plays a response that is not in the engine's MPV set, but that would just mean that if MPV is set to 5, that the opponent played no better than the 6th best response. In that case the engine would not know what E(R6) would be since the engine did not consider that move but the information that it was no better than the 6th best move may still be useful. Over the course of the game the engine can therefore determine the probability that its opponent will play the best response, the second best response, etc. and use this information to calculate P(Rn) x E(Rn). And a probability distribution that emphasizes P(R6) (in this case) may cause the engine to change its contempt mode (if it has one) to disdain draws, thus justifiably increasing its winning chances at low risk.
Of course, these probabilities would only be approximate if only the moves in the current game are considered. But if the engine knows who/what its opponent is, it can maintain a file that records the distributions and augment it each time that the engine plays that opponent. In that case the probabilities will get more accurate over time since there will be more data points used in the calculations.
|Feb-24-14|| ||AylerKupp: <Possible implementation of "Swindle Mode" in a chess engine> (part 3 of 3)|
There would likely be needed refinements. For example, if the game is in the opening phase, both engines may be using an opening book and the first few moves would be determined by the book and not by the engines' evaluation functions. The engine could not assume anything about its opponent's opening book knowledge until the opponent deviates from the engine's opening book. Of course, that doesn't necessarily mean that the engine's opponent has abandoned the opening book, it may be that the opponent's opening book is deeper than the engine's! It may be possible for the engine to infer that it's opponent has abandoned it's book if it had access to the time that the opponent takes to make its moves since there would likely be a significant increase in the opponent's thinking/calculating time once it abandons its book. It may also be necessary to separately calculate the probabilities for lines involving captures or checks since the opponent presumably has a more limited choice of responses and these would alter the probability distribution for non-forcing moves.
All this is, of course, conjecture. I haven't worked out all the necessary details nor do I know whether the additional calculations needed to implement this "swindle mode" will significantly reduce the time that the engine can spend doing its usual calculations. But it is, at least on the surface, not conceptually impossible, at least from my perspective. My hope is that some engine developers would read this and be sufficiently intrigued to explore it further, since the approach would likely not require significant changes to the way their current engine likely works, just a calculation of the best move probability distribution and basing the minimax algorithm on P(Mn) x E(Mn) rather than just E(Mn), although the complexity of the changes needed would likely be different from engine to engine. However, I still believe that this would be an intellectual exercise only since I doubt that the engine's playing strength would be enhanced by implementing this swindle mode. But, of course, the only way to find this out for sure is to implement it and play a sufficient number of games with this swindle mode enabled and a similar number of games with this mode disabled and compare the results.
|Feb-25-14|| ||DcGentle: <AylerKupp>: Your idea looks fine at first glance, but I have got qualms.|
One problem is, that in normal match play, the MPV-Mode is not used in order to get the engine searching deeper into the game tree.
And even if you turn MPV-Mode on, the fact, that the opponent often prefers the second best move for example, maybe totally irrelevant for the trap that should be set in the critical position, because the logic here is not known.
This is the decisive difference from the kind of engine I am working on. The objectives of the lines would be known, because the engine only calculates <maneuvers> with known goals, and it would see, well, this move would lead to stalemate under certain circumstances and this to mate always. So here it would be easy to choose the former one and not the latter.
|Feb-25-14|| ||AylerKupp: <DcGentle> Oh, this concept has many holes, I'm not claiming otherwise; it's just an idea that tries to make use of many features that already exist in a chess engine so that major redesign effort is hopefully not necessary. And, yes, setting MPV > 1 slows down the engine but it's already implemented so it would be a case of automatically enabling it at a suitable level. Besides, anything that adds capability will likely slow down the engine somewhat, including enhancements to the evaluation function. The question is, is the slow down and reduction in search depth compensated by other benefits gained? I am of the opinion that if the objective is for the engine to play better chess then it does not since this swindle mode is not likely to be needed for the reasons I've stated before. But if the goal is not necessarily to play better chess but to make the engine play more human-like, then something like this might be useful. Anyway, just a flight of fancy.|
Speaking of flights of fancy, your concept of calculating maneuvers with known goals sounds somewhat similar to an idea I had inspired by Capablanca who claimed that he first saw positions that he would like to reach and then figured out the best sequence of moves that would allow him to reach those positions. Well, the first stage sounds like fuzzy pattern recognition to me; a database of known advantageous chess patterns to which the current game position is compared to. And the second stage, once the desired position (i.e. goal) is determined, sounds like tablebase generation-like retrograde analysis to determine the best moves to get you there.
But I don't want you to confirm or deny whether your approach is somewhat similar or not, I don't want you to give away any of your secrets. ;-)
|Feb-25-14|| ||DcGentle: <AylerKupp>: I still have problems with positional play.|
But generally, this wall art is right for chess as well:
|Feb-25-14|| ||diceman: <DcGentle:
Giri set a trap by <16. Nd4>, anticipating the bad but natural move <16... Nf6>.
The mean thing is, that <16. Nd4> is a good move actually.
But would any engine go for it?>
<(16... Qf6 <was a good move here, and in the following lack might have saved the draw:>.>
<(17... Nd7 <was the last chance to avoid disaster, but it would have meant a tempo loss anyways.>>
This is the problem of “no definitions“ undefined benchmarks.
I quickly threw this on Fritz and it liked 16...Qf6 and 17...Nd7.
So we rate moves by the human falling into it.
We criticize the computer because it “cant find it.”
However, the computer see’s thru it and doesn’t consider it a good move.
The only “real” question is how good a move Nd4 is?
The end always justifies the means.
If Giri wins, even poor moves are brilliant.
(not to mention the probabilities of a 2749 beating a 2455)
If moves like Nd4 made it simple to defeat computers
the Kasparovs and Kramniks would be doing it all day long.
|Feb-25-14|| ||DcGentle: <<diceman>: If moves like Nd4 made it simple to defeat computers the Kasparovs and Kramniks would be doing it all day long.>|
Of course. There are no "simple" moves to beat a computer.
In order to defeat such tactical monsters like current engines, great skills and a special method is required, and the ability to apply this method, namely the positional one. On fast machines current engines are so strong, that even this might be not enough. If you go over Carlsen games, the engine sometimes finds moves for the opponent, which I really would like to see Carlsen playing against, and it could well be that the opponent missed this move or that move and Carlsen won due to these misses.
Nevertheless, even if you play the engine suggestion and try to play on with this line, the outcome often is not so clear. This shows, that Carlsen's play is essentially sound.
|Feb-26-14|| ||keypusher: Comment on computers from Nakamura (who is doing "Ask Me Anything" on Reddit).|
<Q: Hey, I was wondering how big of a role computers play in your chess life.
How much and in what ways do you use them when studying chess?
Do you ever play against them as practice?
What do you think we can learn from computer chess and what can't we?
A: Hello Jack, your question is very pertinent not only to my chess career but the very future of chess as well. I would say that nowadays, when I study, computers comprise 90% overall.
I do not play against computers anymore because it is severely depressing to lose every game without ever even having a chance!
I think mainly what can be learned from computers is a deeper understanding that almost all positions are ok with accurate play. In the past, many people assumed certain positions were automatically bad, but computers have shown that the rules and thought processes aren't always accurate!>
|Feb-26-14|| ||DcGentle: <keypusher>: Well, Nakamura's opinion may represent the majority of contemporary younger GMs, I think this is safe to claim.|
If the engine is their main tool, they won't learn the great positional mastership Carlsen displays in his games.
Carlsen has twittered (https://twitter.com/ChessQuotes/sta...):
<Contrary to many young colleagues I do believe that it makes sense to study the classics.>
And when he says classics, in his case it means the games of Capablanca.
And Kasparov said at the end of this book review (http://www.nybooks.com/articles/arc...):
<... But there were other goals as well: to develop a program that played chess by thinking like a human, perhaps even by learning the game as a human does. Surely this would be a far more fruitful avenue of investigation than creating, as we are doing, ever-faster algorithms to run on ever-faster hardware.>
It's rewarding to read the whole article, because here Garry describes his relationship to chess computers in great detail.
|Feb-27-14|| ||AylerKupp: Chess, like most things in life, requires a balance. It is as unbalanced to rely on computers exclusively as to not use them at all. Computers are a tool and, like most tools, you need to use them appropriately. And they can be a good teaching tool, but the benefits that you get from them is directly proportional to the effort that you are willing to devote to learning from them.|
|Feb-28-14|| ||DcGentle: <Learning from engines?>|
Well, from my own experience, I must admit that I myself learned a lot from engine play, while analyzing a certain opening, where I wanted to proof a hypothesis of mine. This was all tactical stuff, because I already knew, that engines are good in this area. Yes, one can claim they are better than any human today, when I used them, they were not as good, because not so advanced, but nevertheless excellent.
So all kinds of double attack I could notice, the famous Zwischenzug motif, and especially another phenomenon, the tactical cover, as I call it. This means that there is an apparently loose enemy piece to be taken, but if one thinks, 'Oh nice, I'll get it', in the very next move or only a few moves later tactical punishment will ensue and the own material is only equal or even worse. Yes, engines are masters of this kind of stuff.
But I also noticed one downside: Just from the opening engines are capable to lose their way. Say you are White and want to win. Engines are following some promising tactical operations, but gradually Black will be able to hold or become even better, because something was overlooked. This is the effect of the horizon during engine search, which can be very annoying for the analyst.
Other engine weaknesses were not so obvious to me when I started using them as tools, and indeed they are not easy to detect. It's their positional play. The World Team game against Akobian was an eye opener here. Firstly, engines want open positions always, in order to apply their tactical play most efficiently. But doing so they miss the generation of space advantage created by pawn chains, for example. Secondly, in this game we had to transfer a rook and a knight to their optimal locations. Current Engines don't either know any optimal locations in this context, neither how to take pieces to them. Nobody knowing the engine algorithm a bit closer needs to be astonished about this lack of skill. For current engines are just trying moves, and in order to assess the outcome, they need enemy contact basically, and this was not given due to to the special pawn structure of this game. So regardless of the own piece movements, the outcome was always rather similar. But we humans could detect crucial differences, because Black could have won a pawn and turned the game, if we had chosen a different deployment order of our pieces. So generally bringing up and configuring reinforcements is a very hard task for current engines, and this is not something to gloss over, especially in the opening this is a crucial job to do. And there are more positional weaknesses, which I don't want to mention here now.
Engines are playing positional moves, but only according to general guidelines, and only those making sense on this side of their search horizon.
The conclusion must be that current engines have their special style and that their move suggestions must be critically contemplated. But more and more I read and hear people talking about "objective truth" in the context with engine play. This "truth" may be valid for certain levels only, but not absolutely. One can learn from current engines their computer style, but not more.
|Feb-28-14|| ||AylerKupp: <DcGentle> With all your discussions about the "positional" play that current engines are unable to do, I've been meaning to ask you, how do you define the difference between "positional" and "tactical" play in this context? You referred above that engines are playing positional moves but only according to general guidelines, and only for those moves making sense within their search horizon. But how to do play positional moves using anything other than general guidelines?|
You may or may not have read "Grandmaster Preparation: Positional Play" by Jacob Aagaard of which the following is an excerpt: http://www.qualitychess.co.uk/ebook.... The book itself is reviewed in http://www.chesscafe.com/text/revie.... It is the 2nd book of a planned 5-book series, and the 3rd book in the series is titled "Grandmaster Preparation: Strategic Play". So then I began to question whether I fully understood what you mean by "position play" and whether I have been confusing it with "strategic play", since I don't know GM Aagaard's definition of these terms and I have usually thought of the two as synonymous. Alas, while both books are available at amazon.com, they didn't provide a means to peek at their contents so that I couldn't find out Aagaard's definition of those two terms.
For example, in our recent discussion of Zhao Jun vs Xiu Deshun, 2011 I didn't fully appreciate your characterization of 16.Bxd6 as an "ingenious move" that was the start of a combination containing a positional move, which I would now assume to be 19.Rf6. This would be consistent with Aagaard's Positional Principle #2, Improve the position of your worst placed piece, which is certainly a well-known general principle. I would say that Zhao Jun's (or should that be Jun Zhao's) worst placed piece is the Ne2 since it has no prospects as long as White's DSB is on g3 and Black's g-pawn is on g5 since its only other move is the retreat Ne2-c1 which would, at vest, allow it to be exchanged with Black's Na5, reducing the pressure on the Pc4. But this is clearly too slow since Black is already threatening to take the pawn and can increase the pressure on it by ...Rc8 as a precursor of ...cxd4.
If viewed in this context, improving the position of the Ne2, then 16.Bxd6 certainly deserves the "!!"; it frees g3 for the Ne2 and the zwischenzug 19.Rf6 probably also deserves a "!" or two (although White really has no choice since 19.hxg3 f5 probably equalizes) since it cuts off the Black Ng3's retreat to h5, and the Ng3 is the only piece that can quickly challenge White's Rf6. And, unfortunately for Black, trying to greedily keep both pieces by 19...Nf5 leads, according to Komodo 6, to a mate in 11 after 20.Qh5 Rfd8 21.Bxf5 exf5 22.Rh6 Qxe5 23.dxe5 Kf8 24.Qxg5 Ke8 25.Rh8+ Kd7 26.Rd1+ Kc7 27.Qe7+ Kc6 28.Rh6+ f6 29.Rxf6+ Rd6 30.Rdxd6#, and many of those moves were horizon-effect driven attempts to delay the mate; Black was lost after 22.Rh6.
But how much of this can be considered positional play vs. tactical play? And how much of the continuation did Zhao Jun see before playing 19.Bxd6? Certainly through 19.Rf6; that's a well known pattern where Black's pieces cannot come to his king's defense even though Black is temporarily 2 pieces up. And 22.Raf1 is a natural move regardless of Black's 21st move. But did he see through 25.Qg6+ and its follow up? If so that's truly remarkable.
I realize that there is a fine line between "positional play" and "tactical play" and "positional" play often leads to a position where "tactical" play can happen with advantage. But do you have any guidelines as to what you consider positional play and what you consider tactical play?
BTW, as soon as I finished writing this I found an excerpt on "Grandmaster Preparation: Strategic Play": http://www.qualitychess.co.uk/ebook....
|Mar-01-14|| ||AylerKupp: <DcGentle> Speaking of "positional play" vs. "tactical play", here is a game that you probably haven't seen: A Sacks vs M Micayabas, 1984. Andrew Sacks (username <andrewjsacks>) is a long-time friend of mine that I knew when I was active in chess back in the 60s and 70s. The game is the usual Sicilian "castling on opposite sides and whoever gets to the other's king first wins", and only the position of White's rook on d4 is a little bit unusual. The following is the critical position after 19...Qc6:
click for larger view
Here <andrewjsacks> played what I considered the best move of the game, 20.e5 (!!). Looking back now it is an example of applying Aasgaard's Principle #1 (Attack your opponent's weaknesses, in this case the Pd6) and Principle #2 (Improve the position of your worst placed piece, in this case the Rd4 which, though centralized and seemingly well placed, is biting on granite on the Pd6 which is currently properly defended and can be reinforced if necessary by ...Rad8 or ...Rfd8. And it only has one square to move to, d2).
But 20.e5(!!) in one stroke applies the first 2 of Aasgaard's principles to give the Rd4 another axis of movement along the 4th rank to participate in the k-side attack. It lead to the following position after 20...dxe5 21.Rh4 Rac8 22.Rgh3 Rfd8:
click for larger view
Here <andrewjsacks> gave up his queen (with check, no less) by playing 23.Rxh7 but after 23...Rxd1+ 24.Ka2 it was Black who succumbed a mating attack. So I would definitely classify 23.Rxh7 as "tactical play" but perhaps 20.e5(!!) as positional play. What do you think? I think that most people would go gaga over the impressive-looking (and it was) 23.Rxh7 but I told <andrewjsacks> that in my opinion 20.e5(!!) was the most inspired move since it made everything else that followed possible. And I believe that now more than ever.
If you look at my comments on the game you'll see that at d=12 Houdini 1.5a was evaluating the position at [-0.44] after 23.Qc1, but at d=13 its evaluation jumped to [+37.86] after 23.Rxh7, the biggest jump in evaluation in one ply that I've ever seen. As a follow-up I asked myself what if instead of 21...Rac8 Black plays 21...Rfd8? Then White does not have time for 22.Rgh3 and 23.Rxh7 is not possible (well, possible yes, but good, no). But Houdini still considers that White has a won game after 21...Rfd8 22.Qh5 since White's k-side breakthrough can't be prevented, and at low ply (d=21) Houdini evaluates the position at [+7.38] after 22...f5 23.gxf6 Bxf6 24.Qxh7+ Kf7 25.Rh6 and now Houdini goes into its horizon effect-driven attempt to delay the inevitable as long as possible by 25...Qxc2+. So it's a good thing that this line is available to White, otherwise I would have given <andrewjsacks> a hard time for playing the weak move 20.e5(!!) :-)
|Mar-01-14|| ||DcGentle: <Positional and Tactical Play>|
Hi <AylerKupp>, yes, there are a lot of questions around positional play, and thanks for the links, by the way.
I'll try to answer a few of these questions.
Firstly, the difference between positional and tactical play.
There are some sides on the net giving an overview about tactical motifs, the best is http://www.chesstactics.org/, but unfortunately it's so slow, at least for me, that I have been looking for alternatives. This site (http://www.chess.com/article/view/c...) is not bad, but also has some definitions that I would rather move elsewhere, namely everything that has to do with checkmate and the point "Defense", which can be by tactical or positional means. For tactics essentially is concerned with winning material and not much more. It's the material gain, that has the biggest influence on the engine eval, this is telling, and this is the reason why engines are so good in this area of chess.
Positional play, on the other hand, usually serves to prepare the tactical action, because only with a positionally superior setup any side can hope for a successful attack, for example. I said "usually", because there is a method in chess, I call it "purely positional", which doesn't aim at tactics at all. This is new to me too, I must concede, and very few players employ it, the protagonist here is Carlsen, and maybe Capablanca was also ready to play this way. I only got the idea, because I noticed, that Carlsen a few times ignored tactical options in his games, although they had guaranteed him a nice win. And I thought to myself, "Wait a moment, what's going on here?" The short answer is, that Carlsen has a different view on the chessboard than most players, he only wants to improve his position. And interestingly this can be done until the very end, checkmating is a positional process, it means restricting the king, as we all know.
I read the book review of Aagaard's "Grandmaster Preparation: Positional Play", and I was not so much enthused. For example you cannot reduce all positional play to the three questions
1. Where are the weaknesses?
2. Which is the worst placed piece?
3. What is your opponent's idea?
Of course these questions are helpful, but there is much more to positional play. In the meanwhile I also read a bit from the excerpt of this book, and I only got assured about my view, that I need a more fundamental and systematical work about this area, with fewer exampels, but then with a better reasoning.
I already wrote a small article <Essential components of positional play.> (DcGentle chessforum) and here there are more thoughts about this topic:
Positional play always tries to improve the own position. A List of building blocks of positional play could look like this:
o improving pawn structure,
o improving pieces and their collaboration,
o centralizing pieces,
o improving king safety,
o providing space advantage,
o bringing up and configuring reinforcements,
o controlling strategic squares,
The last point can only be carried through, if the answer to the question "What is your opponent's idea?" has been found.
|Mar-01-14|| ||DcGentle: ... continued
Ok, let's look at the position after move 15 of Black in Zhao Jun vs Xiu Deshun, 2011:
click for larger view
Black to play moved <16. Bxd6!!>
I was wondering how a player could get this idea, which seems counterintuitive at first glance. Why take this pawn, which is not obstructing White in any way?. Ok, a human player might think: "I could open a line to the white king for my bishop on d3 playing <16. e5 dxe5 17. Bxe5> but Black would counter with <17... f6>. Then f6 is covered not only by Nh5 but also by Rf8 and Qd8 and even worse, Black has got a new flight square for his king on f7. Not good." So in order to prevent a possible escape of the black king it would be very convenient, if White could hold off Black from playing f6, and the best way to accomplish this is by occupying this square with an own piece, in this situation with rook f1. But Nh5 is covering f6, so maybe this knight could be diverted? But how? The piece to attack Nh5 is the knight on e2 and the best square to move this piece to is g3, but this is occupied by a white bishop. So if g3 could be cleared in a way that would force Black to react and prevent him from moving f6 or strengthening his king position in other ways, that would be the move to go for. And so the idea of <16. Bxd6> could have ripened starting a kingside attack that inevitably ends in checkmate.
As we see, a human has a certain imagination what they want to reach and ponders any means being able to make his wish come true. Of course White didn't calculate all lines right to checkmate here, this is impossible. So yes, <19. Rf6> is a positional move, but not only because he improves the rook, it also has a prophylactic component, preventing an escape of Black's king, and is improving the own space advantage by blocking an enemy pawn. The nice side of this move is also, that it's a goal of tactical play essentially, because the first sac of the bishop was only made to clear a square for the knight, which then could divert Black's knight from an important task, namely guarding f6. It's rare that we have a tactical combination with a positional goal, which is only the prerequisite for the ensuing checkmating. Interestingly material plays only a secondary role here, because White's heavy pieces are not concerned and will do the rest, together with some of White's pawns.
I fully realize that my earlier deliberations concerning positional play are not enough for any computer algorithm, and I ponder refinements. "What exactly does this mean, improvement of a piece?" this is only one question to be answered. Or even more difficult: "What is a strategical square?" In the respective position, that is. Tactical concepts like a skewer or a fork are much easier to describe, but the perspective is a different one as well, if you think about it. Positional concepts most often deal with square properties in the frist place, at least most of them. Deployment of pieces and prophylaxis have also to do with possible piece actions, but the immediate effect often is invisible.
Prophylactic moves often are the quiet ones, a typical example was <6... h6> of the team game against <GingerGM>, neither understood by many team members nor by the engines. The engine eval dropped, so this move must be bad. Oh yes, but nevertheless it gave Black the superior game control, because knight f6 could not be chased away by White and so Black's strong center remained an asset during the whole game.
I will now look at your game against <andrewjsacks>.
|Mar-01-14|| ||DcGentle: Ok, I went over A Sacks vs M Micayabas, 1984, and <AylerKupp>, you are right, <20. e5> is a positional move, preparing the ensuing attack. In hindsight <22... h6> would have been better, but then White would not have sacced his queen. Oh well. ;-)|
Nice game, really.
BTW, one of the best games I ever analyzed is this one: DcGentle chessforum. Well, I used an engine, but it was Fritz 8 only, I guess. Nevertheless, even with today's Houdini the analysis seems to be ok.
Well, it's not an OTB game, but at that time I had to figure quite some moves myself. This game features a positional queen sac, resulting in a huge space advantage for White, which is achieved by heavy pieces and not by any pawn chains. I can imagine that the latest Houdini will find the queen sac too.
|Mar-08-14|| ||diceman: <AylerKupp:
I asked myself what if instead of 21...Rac8 Black plays 21...Rfd8? Then White does not have time for 22.Rgh3 and 23.Rxh7 is not possible (well, possible yes, but good, no). But Houdini still considers that White has a won game after 21...Rfd8 22.Qh5 since White's k-side breakthrough can't be prevented>
Do you know if this is a win in all lines?
Fritz likes 21...Rad8.
...and in the game 22...h6 instead of
...of course this isnt going as deep as a Houdini would.
<So I would definitely classify 23.Rxh7 as "tactical play" but perhaps 20.e5(!!) as positional play. What do you think?>
As far as e5!, I think its all tactics.
Its only purpose is to swing the rook over.
The pawn on d6 is simply "dots" we humans connect which isnt really relevent.
If you look at concepts like the
open d file, black chasing the queen,
(with the black queen eyeing the back rank) black controlling d4, f4, and the pawn on b4 gaining protection from the bishop. There really isnt a lot positionally to recommend e5. Its real purpose is kingside attack.
|Mar-13-14|| ||Check It Out: <AylerKupp> Thank you for your response at the WC Candidates page about matching up countries early to prevent collusion. We've come to very thin conclusions that it would actually prevent collusion, or that collusion is actually happening. |
It seems to be a subject that is very difficult to get a grasp on, though Fischer was able to strangle it quite a bit :)
|Mar-20-14|| ||Sally Simpson: Hi AylerKupp,
Just to prove I do give computers their due.
Here I do give credit for Fritz for finding a bust to a puzzle 50+ years after it was published.
It's a good find.
|Mar-24-14|| ||AylerKupp: <Sally Simpson> What can I say? Every dog must have its day. :-)|
|Apr-19-14|| ||morfishine: Tal vs Benko, 1959|
BTW: Those were some interesting posts on "swindle mode". I had a question on your comment: <...the first few moves would be determined by the book and not by the engines' evaluation functions> But this wouldn't stop the engine from evaluating the position anyways, right? Or is it invariably 0.00 for the first few "book" moves?
< Earlier Kibitzing · PAGE 43 OF 43 ·
Take the Premium Membership Tour