3

I have only recently started correspondence chess and I have read that engines are allowed in correspondence chess. I definitely do know that the world champions of correspondence chess have said that using engines to a 100% degree is a bad idea. My question is this: What is the most efficient way to use engines in this format of chess? Any experiences would be helpful!

7
  • 10
    Word of caution: engines are allowed in the ICCF (International Correspondence Chess Federation). Engines are by default not allowed in "correspondence chess" on sites like lichess.org and chess.com. Read the rules of the organisation you are playing with before starting.
    – Remellion
    Jan 22, 2020 at 5:58
  • @Remellion Excellent comment. Jan 22, 2020 at 10:15
  • What is an example of an organization that allows engines? I skimmed the linked article but didn't find it. Jan 22, 2020 at 12:21
  • @Remellion quote and source for that ICCF rule. There is no way to tell or police computer use and they have been used ever since Hans Berliner won the world championship using some at IBM. Jan 22, 2020 at 15:38
  • 3
    I found ICCF rules here iccfwebfiles.blob.core.windows.net/rules/2020/… "In ICCF event games, players must decide their own moves. Players are allowed to consult prior to those decisions with any publicly available source of information including chess engines (computer programs), books, DVDs, game archive databases, endgame tablebases, etc." that etc on the end make the rule a bit unclear, but it clearly includes engines. Jan 22, 2020 at 19:57

3 Answers 3

7

Obviously, as long as they are permitted within the rules of the organization, here is how I would use a computer.

First, realize that different programs are better at different things, and I will divide them into two categories. I heard GMs Peter Svidler and Jan Gustafsson discuss this during the Firouzja-Carlsen game just yesterday.

The first program is for opening play, and that is where Leela shines. I have not had much experience with this program, but Jan mentioned that it can, apparently, blunder at times. They both agreed that Stockfish is still the king of analyzing in the middlegame. I watch a lot of relays of top tournaments, and I follow along with Stockfish, and it is currently the best program out there for the middlegame, and the best part is that it is free.

The next thing is that in most circumstances, I only look at the top two lines (in ChessBase, you can click the little plus/minus sign in the engine box to add/remove the number of lines). I would probably increase this to the top 5 lines for correspondence since there may be a line that fits in better with what you are trying to do, so you can, at least, know to explore it.

Next, in the opening, it helps if you have a good understanding of the opening pawn structure, and the typical plans that are played. As you get into the opening, look at a bunch of GM games from a database to see what they have done before you decide on your moves. Have Stockfish going while looking at them...you never know what you might find if it pops up an unusual eval.

Lastly, and the stronger you are, the easier this is, and the better you will be at it, but you want to play "advanced chess". "Advanced chess" is term for using both the computer and your knowledge to pick the best move. In its infancy, Anand was considered the best at this, and there were a few tournaments where these top players were allowed to use a computer. Here is an article. As part of this cyborg play, and this is the key, you want to force in certain series of the computer's best moves or even some of your own human moves, so, in essence, the program only starts analyzing from that point.

For example, if you and I are playing and we are on move 28, and you just let your computer analyze from there, it may analyze out to move 38. Not bad, but if I am able to force in 7 likely moves based on my 40 years' of experience and the computer's evaluation along the way, and only then let it think uninterrupted, I start at move 35, and might get to move 45.

If you are evaluating a position based on the end result at move 38, but I am evaluating it based on move 45, 7 moves later; I am much more likely to be successful, and win the game.

That is how I would use one if I played correspondence chess. Frankly, I am too impatient so I was never into correspondence in the old days of postcards, but the advent of computers really turned it into more machine versus machine, so that really killed any chance I would get into this in the future; but to each his own.

3
  • 5
    I haven't played this way, but it seems like these days whoever knows how to tune their engine and has the most / correct kind of compute power would win. I can't find the exact quote right now but I recall that Magnus said about computer play that they play like they don't understand anything and then they win. If computers win by playing weird moves then how am I to judge these moves? Jan 22, 2020 at 12:32
  • I would be interested in a match between a grandmaster with a computer vs just the computer with the same specs. I'm not convinced the grandmaster actually adds any strength in the long run.
    – orlp
    Jan 22, 2020 at 14:50
  • I've talked to a few correspondence players on the leela discord who say it's pretty obvious when someone is just letting an engine play for them. Turns out the engines are fallible in ways that are sometimes easier to see than to fix. Jan 22, 2020 at 22:15
4

This is by no means complete, but here's a few things.

  1. Use at least 2 engines. More different is generally better. Lc0 and SF would be an obvious pair.
  2. Especially with leela, look at the point of view of the engine. Dig into disagreements.
  3. Realize most of your games will be draws
  4. Make sure you have tuned your engines for long time control. For leela, that typically means using a bigger net since you will have plenty of time to get nodes.
1
  • 1
    I contribute with this idea: Look for positions in which several good moves looks almost the same to an engine. Its like a flat landscape. Or on the contrary, look for very complex positions that the horizon is far deep that your opponent will burnt his machine, before finishing on time his analisis.
    – djnavas
    Jan 23, 2020 at 10:10
2

I feel a bit like I'm giving away my secrets, but here we go

  1. Learn how to get the most performance out of your computer setup. For example, you can run analysis 1) in the browser on chess.com or lichess, 2) using a downloaded version of stockfish connected to your GUI, or 3) compile an optimized build of stockfish from source code for your specific hardware. Those are in order of increasing complexity and you gain effective ELO with each step, with diminishing returns as you increase complexity. At the least I'd recommend staying up-to-date with the latest releases.
  2. Don't use engine parameter defaults. For example, things like the number of threads and the hashtable size need to be increased for longer analysis and for maximal search speed. On my primary analysis machine, I have 3 threads for lc0 (one per GPU + one extra) and 21 threads for stockfish, with a 16GB hashtable for stockfish. The total number of threads you use should be the number of cores your system reports. In my example above, I have 12 cores with hyperthreading and the OS reports 24 cores.
  3. Use syzygy tablebases. All the 6-man positions are a few hundred GB IIRC. I'd also suggest a handful of the most popular 7-man positions and store those locally too. Store these on a fast SSD. Learn how to configure your engine to use these, and how to verify that they're working. Having these available will sometimes drastically change the evaluation of late middlegame positions because the engine will have an absolutely true evaluation of some endgame positions as opposed to using its heuristics to evaluate. This will prevent going into lines that look good from the engine's perspective but are tablebase draws, and could lead you toward positions that are favorable but which the engine has a hard time seeing the benefit of.
  4. Track the current engine evaluation of every game you're playing, and spend more time on games where the evaluation is in your favor. The eval tells you how hard to work analyzing variations. For example, if I see +1.3 for me on move 22, I'm going to try to look at dozens of lines into the late middlegame or early endgame if I can to see if I can find lines where the eval holds, or better yet, increases, as I go deeper. The hard truth is that often you'll see the +1.3 advantage just evaporate as you play on, so it's critical to explore and decide if you think you can find lines where your perceived advantage does not vanish.
  5. The corollary to the above is basically don't spend any time on positions that are 0.00 eval. I'd say don't offer or accept a draw in these positions, but choose your move from the 0.00 options and make it after double and triple checking that the position on the board and your move match what you analyzed. If your opponent wants a draw, they have to not mess up and get to a tablebase position, 50 move rule position, or find a threefold repetition and can then automatically claim a draw. I've gotten one or two wins in drawn 0.00 positions where my opponent made a mistake. And frankly, all wins in chess are a mistake on your opponents part because its all probably a theoretical draw from move one. So you might as well play in any 0.00 position because the starting position is probably theoretically 0.00 and no one has qualms about starting a game. If you play move one because you think you have a chance at a win, then you should think you have a chance at a win in any 0.00 position.
  6. Dive into the details of disagreements between engines. If lc0 and stockfish see a position drastically differently, play out the lines and see which engine changes its mind first as you go deeper and deeper. You'll sometimes find that one engine or the other misses a critical move and only sees it when you get closer or in some cases after you've played the move. Use this to your advantage: once you've resolved the discrepancies by going deeper, you have a better true evaluation of the position, while your opponent may not have as good of a picture, and as they play on thinking they are okay according to their engine, you will surprise them with a line their engine missed.
  7. Use MultiPV. This is one no one talks about. Engines will give you their top N lines and sort them by evaluation. So in a position where there are several good moves, I'll start with many lines and let the engine crank for a while. I'll mentally cluster the top moves and reduce the PV to just consider them. For example, if I do a PV=5 analysis to depth 30 and see evals of {0.47, 0.45, 0.42, 0.22, 0.16}, I'll notice that the top three cluster together, then the next moves down come after a larger gap in eval. So then I'll reduce the PV to three, and let the computer crank to more depth on just those three to see if it will clearly separate those three and if one is clearly better than the others. If not, I'll explore all three in depth and decide which line I like.
  8. Using the computer just for record keeping of your analysis is huge benefit to your game. I write extensive comments in my games using my GUI, including my impressions of the evaluation, how many alternate lines I think I should consider, and if/when I think my opponent missed something or played an inaccuracy.
  9. Analyze past games for opening ideas. All the ICCF PGN files are available and can be made into one big database, which can then be viewed in "Show Statistics" mode in the Scid software (for example, there's lots of ways to do this). This shows me immediately what's been played, how relatively popular different opening moves are, and lets me dig into opening lines in depth. If a line is popular but statistically bad for one side, I'll often delve deeper to see if there is an unpopular sideline that is more favorable. For example, there are lines in the Najdorf Poisoned Pawn that are statistically overall bad for black, but if black plays the second most popular move at the right moments, he is fine and will maintain equality. I also keep a separate file of just my completed games (and my current ongoing games). I'll look to my completed games for guidance, e.g., if a move looks statistically bad in the opening database but I've drawn with it many times, I'm more confident that its okay to play (but I'll always check variations to make sure I won't be unhappily surprised).
  10. Use your opening database and engine to find the perfect moment to uncork a novelty. If you see a move has not been played in a previous position and its top or close to the top of engine recommendations, you can consider going for it and bringing your opponent into previously uncharted territory. In my opinion, the earlier you deviate from the database the better, because less about the structure of the position is cemented.
2
  • 1
    Very nice writeup! Some small things I would add: in 0.00 positions I think the character of the position matters. In a complicated confusing position you still need to be cautious. However, in a clearly defined settled position indeed you do not need to spend as much time. And, for 10., you can also use the database of your opponents past games to figure out if their repertoire has any weaknesses that you want to probe.
    – koedem
    Oct 11, 2023 at 9:19
  • Those are very good points, thank you! Looking at my opponents openings using the database and planning around their typical play is something I do and I agree wholeheartedly that that is a great way to use a computer when playing on ICCF, and it is one I neglected to include in my answer.
    – rajb245
    Oct 11, 2023 at 15:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.