07 March 2021

"Developing New Skills and Habits Using Root Cause Analysis" - article

Worth a read is the "Developing New Skills and Habits Using Root Cause Analysis" article at Chess.com by NM Hans Schut. Despite the didactic title, it contains some very practical counsel and examples focusing on adult improvers. The central idea is to be aware of your repeated problems and their root causes, through analyzing your own games. According to Schut,

Some of the top root causes identified in the lessons with my students are:

- I do not know the plan behind the opening that I play, I feel lost after the opening;

- I reduced tension by trading instead of building the activity of my pieces into an attack;

- I do not know how to play this opening;

- I do not use the blunder check (capture, checks, threats) at every move;

- I do not calculate variations, I play moves based on the characteristics of the position and my gut feel;

- I am overlooking the possibilities of my opponent;

- Endgame: lack of calculation skills and knowledge of general endgames principles including rook activity, king activity, pawn breakthroughs;

- Get into time trouble and blunder (this can have different root causes).

Any of these sound familiar? My top current ones are still the lack of knowledge of opening (really early middlegame) plans, failure to consistently apply blunder check/CCT, overlooking my opponent's possibilities, and endgame weakness. On the other hand, I would say I've been aware of all of these areas for improvement and have actually improved my performance in them, although not to where they need to be. In the past, reducing tension by exchanging pieces was something I noted repeatedly in analysis, and now pay much more attention to this in my games, so it's no longer an issue.

The main takeaway is therefore to know why you are losing and work on not repeating the same mistakes over and over. This often requires replacing bad habits with new ones, including deliberate modifications of your thinking process. Simply reducing the frequency of common mistakes will up your game and eliminating them can result in a significant leap in strength.

05 March 2021

Training quote of the day #35: Max Euwe

In the middlegame the pieces seldom exert their full powers, since they are obstructed by other pieces, both hostile and friendly. The aim must always be to give every piece its greatest possible radius of action; and this is really the hub on which turns all positional and combinational play.

From The Middlegame, Book One: Static Features by Dr. Max Euwe and Hans Kramer  

27 February 2021

"The Attacking Set-Up That Always Wins" - article

In the context of studying the Stonewall, it was interesting to see the article by GM Gregory Serper "The Attacking Set-Up That Always Wins" on Chess.com. Despite the clickbait title, it's actually a useful conceptual breakdown of some opening and middlegame ideas using the structure, focusing primarily on White's prospects with it (which he terms the "Pillsbury Attack"). Some highlights and observations:

  • The article intro examines the core setup (Ne5, Bd3 with pawns on d4/e3/f4) using a classic Pillsbury game (above image) from a Queen's Gambit opening. The main difference between it and a "pure" Stonewall Attack is the location of the Bg5. Other ideas seen (Qf3-h3 transfer, sacrificing the light-squared bishop, exploitation of the half-open f-file, etc.) are standard Stonewall ones.
  • Even more unusual paths to the formation for White are possible, including via 1.e4, as in the exciting games Bronstein-Zamikhovsky (see move 12 onwards) and Tal-Leonov (move 13 onwards), both from Caro-Kann (!) variations. In these the e-pawn is missing, but the other characteristics remain.
  • Black is given a nod with the citation of one classic attacking Dutch Stonewall game (Maroczy-Tartakower) that features a thematic rook sacrifice for White's h2 pawn, an idea (in reverse) which is also seen in one of the White example games (Steinitz-Mongredien).
The article is well worth the read for anyone looking at how to succeed by employing the Stonewall formation to attack. It also shows how this approach is not necessarily the result of formulaic "system" opening play, rather the idea is to reach specific position-types which can be very effective if their advantages are understood.

23 February 2021

Chess Computing Resources (2021)

The recent release of the Stockfish 13 engine is a good peg point for an updated post outlining available chess computing resources for improving players. This is intended to be a streamlined look at key resources grouped around two main functions - Database/Analysis and Gameplay/Training - along with some observations on how to effectively use these computer tools. It is not intended to be a comprehensive review of everything that is available on the internet. Rather, it reflects the products/sites with which I have some familiarity, and have found more (or sometimes less) useful. For a deeper dive into specific software functionality, although the references are somewhat dated, see Chess Computing Resources for 2015.

I. Database/Analysis packages: the core training functions

Based on the recommendations of top-level trainers and my own personal experience, the two most important practices for progressing in chess strength are analyzing your own games and studying master games. Building your personal databases - I maintain two, one containing my tournament games and one with my training/online games - is fundamental to this. Using computer tools to help analyze and compare your games to master-level ones then becomes a powerful and efficient training practice. 

To maintain full control over your data and for ease of use, having your own software package often makes the most sense. This will consist of three parts: 1) a database program GUI (graphical user interface); 2) databases (your own and at least one general reference one); and 3) chess engine(s) for assisting analytic evaluation. Below we'll look at what's available in each of these categories. Mixing and matching is very possible, so it is worth carefully considering what you actually need for your own training purposes, especially if your budget is limited.

Database/Analysis GUI programs (free)

Scid vs. PC
A very good free option for database software is Scid vs. PC, which also has a Mac version and a related Android mobile app (Scid on the Go). The above screenshot displays the program loaded with the PGN file of the Ruiz Castillo - So commentary game, with the "Commentary - ChessAdmin" database (linked in the sidebar) loaded into the database tree reference tab. The engine analysis window is displayed at the bottom, showing the top three lines being calculated by the (free) Stockfish 13 engine. The text moves in the PGN window are fully annotatable using a right-click menu. The software is quite powerful and has many database, search, and game editing functions available.

Other top free database GUI programs include: Arena; a version of the original Scid; and ChessX. It costs nothing to check them out and one may work better on your particular computer platform (PC/Mac/Linux).

Training methodology: the above layout reflects my core setup for game analysis functions, except that I would also have open a general reference database and my opening repertoire database. Stepping through a game from the beginning then allows you to examine in detail the key decision points, in a computer-assisted manner, while exploring and recording different options. In the opening phase, the databases are the primary reference points for comparison and further investigation. While move statistics (% winning, frequency played) can be helpful guides as to what works best in a position, it is also important to look at related master games in a more in-depth fashion, especially to see what is going on in the early middlegame. Once that point is reached, the engine becomes your primary assistant, especially in highlighting alternative candidate moves that you may not have otherwise considered; this is why I have it set to display the top three lines. Of course along with the benefits, it is useful to keep in mind the pitfalls of computer analysis.

Database/Analysis GUI programs (commercial)

ChessBase 15
Above is displayed the same game in the ChessBase 15 program. The configuration is similar, showing the board, game text, and engine window, in this case running the (commercial) Komodo Dragon engine. One significant addition is the lower right "Let's Check" window which is proprietary, showing the top three engine evaluation lines from the ChessBase cloud database for the position. This is a useful way of automatically tapping into others' analytic efforts, using a variety of different engines. (You automatically contribute your own engine's evaluation by accessing the function, as well.) The feature is included with the latest ChessBase versions, which also have a large suite of advanced analytic database functions, as previously documented. These can be very useful for study and preparation purposes, but are not required for the core practice of game analysis using a reference database, analytic engine, and annotation tools.

Other commercial database GUIs include: Hiarcs Chess Explorer, Chess Assistant 21 and Aquarium 2021. These are all powerful products, but the reality is that ChessBase is considered the commercial standard, including for its proprietary database file format (CBV). Which means that if you want access to a variety of large databases in CBV format, you will need a ChessBase program.

Databases

In contrast with database GUI software, it's difficult to find comprehensive, good quality free games databases, which makes having some type of commercial product more necessary, at least in terms of a general reference database. That said, there are still multiple options available and searching the internet for public domain PGN databases that are currently available is always an option. I would advise being very careful about any advertised "free" file sources, however, to avoid downloading malware onto your system.
  • TWIC (The Week in Chess) since 1994 has been publishing international tournament games on its website and has weekly game files freely downloadable in both PGN and CBV format. According to the site, a PayPal donation will get you the full collection in CBV format.
  • Chess Games Links is a comprehensive site for finding downloadable and online games databases.
  • PGN Mentor is often cited as a games source and has a number of databases broken down by opening, player and event.
  • The standard commercial ChessBase products (sold by many retailers) are its Mega Database and Big Database programs, released as comprehensive general reference databases for international OTB (over-the-board) tournament and match games, with yearly updates. The primary difference is the professionally annotated games contained in the Mega version. Both use the proprietary ChessBase format to facilitate generating reports by player, event, etc. and to take advantage of the advanced search and analytic capabilities of the ChessBase GUI software. The company also publishes a separate Correspondence Database.
Chess engines

In contemporary times, free chess engines are among the world's top performers, so there is no reason not to have one (or more) in your analysis toolkit. A bit of internet background research - for example this non-partisan blog post on Chess.com - will also reveal which engines are essentially clones of publicly available ones. You can then make the decision to go for the free download, or instead choose a slightly tweaked, perhaps even weaker commercial version of an engine.
  • Stockfish is the gold standard chess engine that is freely available. Many internet sites incorporate a version of it as a live analysis engine.
  • Lc0 (Leela Chess Zero) is the main open source neural network based chess engine, also freely available. Stockfish and other top engines now incorporate NNUE (efficiently updatable neural network) evaluations, so the neural network distinction is less important than it used to be.
  • Komodo is a commercial engine that has been developed since 2007 by GM Larry Kaufman and his team. It typically has some extra/unique features and settings included, which are described on the site. The latest (Dragon) version has also incorporated NNUE technology. Previous Komodo versions are available for free download on the site.
It used to be that chess engines had significantly different strengths and demonstrably varied in their evaluation functions; for example, Fritz was notorious for its materialism. Now that the top engines are well over 3000 Elo in strength, there are still differences between their evaluations and approaches, but the practical significance of this is relatively minor, especially at the amateur level. Nevertheless, I believe that the "feel" of an engine's move choices is at least psychologically important for an improving player, so which engine you select as your primary assistant is still a personal decision. Historically I've been able to understand and relate better to Komodo's top 3 choices - the standard number of candidate moves I have displayed during the analysis process - so that has been my go-to engine for a while now.

It's worth highlighting the use of the word "assistant" for an engine's role in the training process. They can't be called coaches, since they do not provide explanations for their recommendations. The assistance provided to your thinking process by their showing different possibilities in a position can be very valuable, however. Engines will also help hone your evaluation skills, since as part of the analytic process you will need to articulate to yourself why a position is +/- for a particular side. Finally, your own judgment cannot be replaced and sometimes - perhaps even relatively often, in quiet positions - engine "scores" for lines can be misleading or simply not helpful, depending on how practical they would be to play in real life.

II. Gameplay/Training resources

Given the broad capabilities of modern chess software, the lines have been permanently blurred between chess database and chess-playing programs. This is in part because having some sort of database access is now considered integral to playing chess. So depending on what you want, you may not feel the need to have a different application for your playing and other training needs.
  • Scid vs. PC is a good example of this, as it comes with an engine (Phalanx) designed to be played at a user-adjustable level (strong master - beginner) and has basic time and other settings, including specifying which opening complex the computer should play and playing out a game from a particular position.
  • ChessBase deliberately separates the playing function into its Fritz GUI line of software, which has been sold with various primary engines, although ChessBase software owners have the option of playing out a particular position against an online version of Fritz at adjustable levels.
The experience of playing a computer opponent is much more personalized than the database analysis functions, as "fun" and "challenge" and other aesthetic considerations will enter into it. People also have very different goals and interests in playing, everything ranging from repeated blitz games to learn new openings, to slow games that are intended to simulate tournament conditions. Personally, I find it difficult to care about a training game against a computer, especially when there is no evident personality or "real-life" quality conveyed as part of the experience. This is why my preferred practice is to play a weekly game against a human training partner.

That said, here are some additional computer resources designed to provide higher-quality playing experiences.

Shredder 13
  • Shredder Chess is now best known for the game-playing and training features in its Shredder 13 line of commercial products, which are available for all platforms. The interface is intuitive and it offers a variety of customizable options, as well as a more human-like opponent at different strength levels. It has built-in training features for openings and endings, as well as tactics puzzles. Naturally it has an analytic and PGN database capability as well.
  • Lucas Chess is software geared towards playing and training, with special emphasis on offering a number of choices deliberately aimed at beginners, including some very low-rated engine opponents and more basic training options to develop chess skills.
  • It's a shame the old Chessmaster series died out, as especially the last edition (XI or The Art of Learning) was packed with tutorials for all levels and used an expansive range of named opponents with different chess "personalities" that made for more interesting training games. You could for example use the opponents as a "ladder" challenge or specific ones (aggressive attackers, strong defenders) to work on specific aspects of your game. The software is still floating around, but installing it on a modern system is unlikely to work well.
  • The Chess.com bots are a recent implementation of the same idea of varying chess "personalities" and strengths, in the context of an online opponent. From my perspective, they have a good range, but lack some customizable training options, for example a time control above 30 0 (why not 45 45 or 60 10 or just let you input one yourself, up to a certain range?) This makes them unsuitable for tournament-like practice games, since anything that goes into the endgame will result in a time crunch. There is also a "no time" option for casual play, but that doesn't address the issue of wanting an opponent set up for a slow game.
In terms of chess training resources, there are a lot of online sites that offer a wide variety of services. Gone are the days when chess bloggers had to rely on things like CT-ART software. Internet search, which will turn up things like YouTube videos and such, may well be your best bet for researching specific interests or applications in depth, especially free content. Nonetheless I'll close this post by highlighting some of the major commercial chess training sites, which normally include some free content as part of their offerings.
  • Chess.com has online database services, including an openings explorer, various chess puzzle challenges, and a series of its own interactive training videos on all aspects of the game. There is also a large number of recorded instructional videos from professional players hosted on the site, many behind the paywall.
  • Chessable offers a range of training courses on openings, endgames, strategy and tactics. Again, some content is free.
  • 365chess.com is primarily an online database service, but includes a number of training features for its supporters, focusing on openings and puzzles.
  • Chess Tempo is an in-depth site focusing on tactics, openings and endgames training, as well as offering online database services.
  • Forward Chess by this point has accumulated an extensive library of interactive e-books from major publishers for its app, which is best designed for tablets. If you are already using a database GUI program for your analytic work on a computer, I would recommend getting e-books (with accompanying databases) in whatever format you have your other databases, so you can use the data more easily. You can obtain free samples and excerpts of products to "try before you buy" within the app.

21 February 2021

Commentary: 2017 FIDE World Cup Tblisi, Round 1 (Ruiz Castillo - So)

I find it most helpful to group together studies of master (commentary) games and opening structures. I've been looking at the Classical Caro-Kann recently (see Nakamura - Liang) so selected for analysis the following game won by GM Wesley So as a useful follow-on. I've more or less randomly accumulated material for my commentary queue, based on reading reports of international events and seeing which games seemed most relevant, interesting and rewarding of more in-depth study.

So's victory and the choices made by both players along the way are well worth examining in detail. The first significant choice is on move 11. I expect White likely went into this variation with the idea of playing 13. a3, as an attempt to get material off the board and transition into an equal ending. However, So afterwards continually finds ways to pose small problems to his opponent and create more imbalances in the position. 

Some of the choices made by both players appear to have been more practical than objective, for example White's decision not to exchange rooks around move 30, but when playing a human in live chess, these are just as important to take into consideration. Other ideas, such as the ones behind So's moves 28 and 37, demonstrate a deep understanding of the dynamics and potential of the resulting positions.

White only makes a significant mistake on move 38, by allowing a tactic to win material, although on the previous move he sets himself up for it. This is a common psychological dynamic, as it is harder to un-commit from an idea (in this case centralizing the king) once it has already been played. By move 41, we therefore reach the point where classical annotators would simply end with "and now it is a matter of technique," but for those of us who could use some work on our endgame proficiency, So's technique is in fact worth following to its conclusion.

[Event "FIDE World Cup 2017"] [Site "Tbilisi"] [Date "2017.09.03"] [Round "1.1"] [White "Ruiz Castillo, Joshua Daniel"] [Black "So, Wesley"] [Result "0-1"] [ECO "B19"] [WhiteElo "2377"] [BlackElo "2792"] [Annotator "ChessAdmin/Komodo 13.2"] [PlyCount "118"] [EventDate "2017.??.??"] 1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Bf5 5. Ng3 Bg6 6. h4 h6 7. Nf3 Nd7 8. h5 Bh7 9. Bd3 Bxd3 10. Qxd3 e6 11. Bf4 {White's first significant decision in the main line of the Classical Caro-Kann. The alternative, Bd2, prevents Black's next, but has been considered less challenging.} Qa5+ 12. Bd2 {in olden times, the expected result would have been for Black to retreat with ...Qc7 and enter into more standard lines, rendering the Bf4/Bd2 decision essentially meaningless. GMs since the early 2000s have played this position more actively, however.} Bb4 {Black has a remarkable plus score (almost 54 percent) in the database after this move.} 13. a3 {this move is the third most played, but relatively rare. It essentially offers a transition to an equal ending and should therefore be viewed as a drawing attempt for White.} (13. c3 {is standard, after which Black does best to retreat with} Be7) (13. Ne4 $5 { starts off complications if Black does not wish to trade pieces, starting with} Ngf6 14. Nd6+ Ke7) 13... Bxd2+ 14. Qxd2 {at this point, Black scores well in the database both by trading queens or by keeping them on with either ...Qc7 or . ..Qd5.} Qxd2+ {So opts to head straight for the ending, having over 400 Elo points on his opponent and no doubt both greater experience and endgame skills.} 15. Nxd2 Ngf6 16. O-O-O {castling kingside would decentralize the king and leave the h-pawn weakened after the rook transfer, so White castles queenside.} Ke7 {Black feels no need to castle at all. The king is centralized for the coming endgame, covering the d6 square, and the rooks are connected.} 17. Nde4 {re-developing the knight, uncovering the Rd1 and offering another piece trade.} Rad8 {So develops the rook and implicitly declares he will be using it on the d-file, pursuing a central strategy.} 18. Rhe1 Nxe4 {So decides to simplify the task at hand, now that White has shifted his kingside rook over. This means that Black would gain a small initiative in the event of of a knight recapture.} 19. Rxe4 (19. Nxe4 Nf6 {now the h-pawn is threatened.} 20. Nxf6 gxf6 {and after ...Rhg8 Black will have a pleasant, if small, advantage in his rook activity. The doubled f-pawns are not a weakness that White can exploit and in fact help shore up Black's central control (e5 and e6 squares).}) 19... Nf6 {the knight goes to its best square with a tempo on the rook.} 20. Re5 {on the 5th rank the rook guards the h-pawn and is not badly placed, so Black harrasses it again.} Ng4 {also threatening f2.} 21. Re2 { retreating and protecting the f2 pawn at the same time.} Rd5 {now it is Black's turn to influence the 5th rank with a rook, provoking White's next.} 22. c4 Rd7 {the point of Black's maneuver was to make the d-pawn a better target for him. The position remains equal, but is easier to play for Black, who has more obvious targets (White's h- and d-pawns).} 23. f3 {pushing back the advanced knight and removing the f-pawn from its attack. This comes at the cost of the pawn's influence over g3 and e3.} (23. Nf5+ {is the choice of the engines, taking advantage of the pin on the e-pawn to drive the king away from its ideal square.}) 23... Nf6 24. Re5 {reoccupying the 5th rank, now that the rook cannot be driven away by the knight.} Rc8 {the king's rook finally gets into the game, So choosing to line it up against his opponent's king. One potential idea that results is the ...b5 pawn advance.} 25. Nf5+ {now White chooses to drive Black's king back. However, this is not as effective now, since the rook formerly on h8 has found freedom on the other side of the board. } Kf8 26. Ne3 Rcd8 {a simple but effective doubling of rooks against the d-pawn.} 27. Nc2 {with White's pieces more awkwardly placed and restricted, Black by this point has a slightly more concrete edge than simply enjoying an easier position.} b6 {increasing Black's control of the 5th rank and preventing ideas of Ra5, for example. At this point Black can be said to have the initiative as well, since White is running out of useful moves and will have to respond to Black's ideas.} 28. Rde1 {taking advantage of the fact that the knight is anchoring the pawn on d2, also removing the under-protected rook from any potential tactics involving the d-file.} Rd6 {this potentially frees up the d7 square for the knight and provokes White's next move. These types of ideas can be difficult for amateurs to spot. Often we look, in a too-narrow fashion, to always "do something" with the piece being moved. In this case, the rook itself is no better off on d6, but it accomplishes something useful for Black's other pieces.} 29. c5 Rd5 {So was evidently fine with allowing a trade of rooks. From a practical standpoint, a player with stronger technique usually finds it easier to win a single rather than double rook ending. This move further pressures the h5 pawn, so White is forced to deal with that threat over the next couple of moves.} (29... R6d7 {also looks good, maintaining the tension on the d-file. The square d5 is now also available for the knight.}) 30. cxb6 {White chooses not to exchange rooks and by doing so has to otherwise reinforce or remove his pawns on the 5th rank.} (30. Rxd5 { is what the engines advise.} Nxd5 {here Black's knight is unassailable on d5 and clearly superior to its counterpart, so White may not have liked that board picture when calculating.}) (30. g4 {played immediately lets Black gain positionally, after} R5d7 {followed by ... Nd5 and ...Ke7, centralizing the knight and king.}) 30... axb6 31. g4 R5d6 $15 {clearing d5 for the knight and keeping the rook on the 6th rank to defend the pawns from potential attack.} 32. b4 {gaining space on the queenside and helping blockade c5.} Nd5 {at this point, simply comparing the Black and White knights' scope for action shows why the engines show a noticeable (if not yet decisive) edge for Black.} 33. Kb2 {advancing the king towards the action and clearing the first rank for potential rook moves.} Nf4 {the better knight begins to make its presence felt, forcing White to respond to the fork threat on d3.} 34. R1e3 f6 {the engines prefer the idea of ...b5 here. However, squeezing White's rook has a certain appeal to it and So uses this to open the d-file.} 35. R5e4 e5 {now White has no choice.} 36. dxe5 Nd3+ 37. Kc3 {this intuitive-looking move, centralizing the king rather than moving it to the edge, in fact offers So a way to further imbalance the position.} (37. Ka2 {is the engines' choice.}) 37... fxe5 { similar to White's previous setup with his d-pawn, now Black's knight supports a central pawn against the pressure of two rooks. The difference is that Black's knight is much better placed than White's was, and Black's rooks are more dangerous. As we will see shortly, there is also a tactical point to the e-pawn's newly-won control of d4.} 38. a4 $2 {under pressure, White goes wrong. With this move, he evidently thought that the situation in the center was stable.} (38. Kb3 {and now Black no longer has the idea of a Nf2-d1 maneuver with check.}) 38... Nf2 $1 {now White loses material, since the Black e-pawn prevents the Re4 from escaping the knight's attack and covering the d1 square with Rd4.} 39. Rxe5 Nd1+ 40. Kc4 Nxe3+ 41. Rxe3 $17 {White has a pawn for the exchange, but with R+N versus a strong two rooks, this is not enough.} Rd5 { first shutting down potential White counterplay involving a queenside pawn advance. White's king is much more active than Black's, so allowing White to get a queenside passed pawn would be dangerous.} 42. Rc3 {lining up on the Black c-pawn.} Kf7 43. Ne3 {White has done a good job of reactivating the knight, but So can temporarily ignore it to resolve the situation on the queenside, while his Rd5 is still dominant.} b5+ {Black uses White's advanced king placement against him, forcing a pawn exchange.} 44. axb5 (44. Kb3 bxa4+ 45. Kxa4 Rd3 $17) 44... cxb5+ 45. Kb3 Rd3 {Black presses his advantage, keeping up maximum pressure on White's pieces.} 46. Nf5 Kf6 {the dominating placement of Black's pair of rooks means that he should avoid exchanging one, until he can get something concrete for it.} 47. Kc2 Rd2+ {the next sequences forces the White king away and keeps Black's rooks dominant.} 48. Kc1 Rd1+ 49. Kc2 R8d2+ 50. Kb3 Rd3 {White would now lose the f-pawn with a rook exchange. Note how Black has no interest in allowing any White attempts at counterplay.} 51. Ne3 Rb1+ {forcing the next sequence.} 52. Kc2 Rxc3+ 53. Kxc3 Ke5 {although the material balance is the same, Black effectively dominates the center (and White's knight) with his king. White's weak pawns are now more vulnerable as well.} 54. Nf5 Rf1 {Black is not afraid to trade pawns in this manner, since the rook will clean up on the kingside before White can do anything useful.} 55. Nxg7 Rxf3+ 56. Kd2 {note how White's king was effectively "boxed out" of the center and is essentially out of the action.} Kf4 {going after the g-pawn next. Black only needs to keep one pawn on the board, as White will then not be able to overcome Black's K+R combo, who can keep the knight and king away from it.} 57. Nf5 (57. Ne6+ {does not help, as after} Kxg4 58. Nd4 Rf7 59. Nxb5 Rb7 {with ...Rxb4 to follow, along with the fall of the h5 pawn.}) 57... Kxg4 58. Nxh6+ {a desperate attempt to eliminate Black's pawns, but now the knight will be trapped.} Kxh5 59. Ng8 Rf7 0-1