Rybka Aquarium: Using IDeA in correspondence chess 13 January 2011
I recently completed an informal experiment that lasted eleven months. During that time I participated in three correspondence chess tournaments. My goal was to evaluate IDeA as an analysis tool for correspondence play and see how it performed in a variety of positions that arise in real games, especially middlegame and endgame positions. I made limited use of various resources normally used by correspondence players, since I wanted the results to depend as much as possible on IDeA itself. Those self-imposed restrictions affected the results, but I think they made the experiment more interesting for users who are considering IDeA for correspondence chess analysis.

My Correspondence Chess Background I never played in correspondence chess tournaments before. Previously, I played one informal game on Rybka forum and participated in a couple of forum team games.

Hardware

My main computer for the analysis was an X5355 @2.66Hz (8 cores). This four-year old processor still turns in a respectable performance, although recent processors are much faster.

Sometimes I also had access to six more cores over an Internet connection, but some of those were running 32-bit software and were quite slow. On a few occasions I used up to six additional cores in one of the tournaments.

Preparing for Opponents

I did not study my opponents’ games or openings before playing them. In this experiment I had no interest in scoring points by playing on a weakness in my opponent’s opening repertoire.

If you play correspondence chess, you should of course study your opponents’ games and use all the resources that are allowed in the tournaments in which you participate.

Opening Preparation

I didn’t do any opening preparation before the games. Instead, I played moves straight out of publicly available opening books, usually Jiri Dufek’s excellent Rybka 4 Aquarium book, but also other openings. Sometimes I used the Opening Tree at chessok.com. In several games I played “harmless,” simple openings, such as the Giuoco Piano, which I played in three games. There is little need for an opening book when playing such openings and I often made the moves without checking the books.

When using an opening book, I usually followed the recommended lines more or less blindly. I didn’t do any type of analysis until late in the opening or when entering rare variations.

The reason why I chose this approach was that it’s already well-known that IDeA is an excellent tool for opening analysis. This is confirmed by opening experts such as Jiri Dufek, Jeroen Noomen, Nick Carlin, and GM Larry Kaufman. However, less is known about IDeA’s performance in the middlegame and endgame stages. Therefore, I was interested in establishing how IDeA performs in positions that arise after the opening. Is it good enough to create winning chances in balanced middlegame positions, maintain or increase the advantage in good positions and hold the draw in inferior positions? You won’t find any groundbreaking opening novelties using this method. The only advantage was that I could blitz through the opening and speed up the game a little. I was perfectly happy with an approximately even position out of the opening, even when playing the white side. Naturally, I do not recommend such a reckless approach to opening play. You risk ending up in an inferior position or a positions where you have to work very hard to create winning chances, if possible. Of all the self-imposed restrictions, this was the only one that had outcome-changing impact. Engines I used Rybka 4 (single core) with 256MB-512MB hash size for the IDeA analysis, running one instance on each available core. Besides Rybka 4, I used a few commercial chess engines, plus Komodo and Stockfish (all running as single core engines; 256MB hash) for infinite analysis. All the analysis was done using single-core engines (mostly multiprocessor engines limited to one core). Tablebases The chess engines weren’t set up to use tablebases (endgame databases), neither Rybka 4 in IDeA nor the engines I used for infinite analysis. Analysis Method I used IDeA as my analysis tool in all the games, with the advantage of having access to Aquarium 4.0.6 beta. In essence, I was using the IDeA analysis methods that I have described in my previous ChessOK Cafe columns, with emphasis on interactive analysis. I moved the root node around as the analysis evolved or if I wanted to check out my own ideas. I added alternatives and extended lines that I found interesting and I used move coloring to exclude moves from analysis or to focus the analysis on selected moves. Infinite analysis is a crucial component of the IDeA analyst’s toolbox. However, in this experiment I made minimal use of infinite analysis. I used it mostly for generating alternatives for IDeA in certain positions. I used eight different engines for more variety in move choices and I also ran Rybka 4 in multi-variation mode in order to generate more alternatives. I added the results to the notation and then I sent the variations to the IDeA project. The maximum length of the infinite analysis sessions was around thirty minutes. Sometimes I ran infinite analysis daily, but at other times I didn’t touch it for a week or more. In completely won positions I often stopped using IDeA and switched to infinite analysis as a quick check before making my move. The purpose of limiting infinite analysis to short sessions with single-core engines was to put the responsibility of finding deep, hard-to-find moves on my IDeA analysis. I was very interested to see if this would cause me to miss some hidden tactics. Since I wasn’t using any secret IDeA analysis methods there isn’t much more to say here. I have described all the methods I used in previous ChessOK Cafe columns. Rate of Play In the first tournament, the players had forty days at the beginning of a game and forty more days after every ten moves. The two other tournaments were faster: ten days, plus one day/move (no leaves allowed). I switched to faster tournaments because the experiment would have taken too long. I got email notifications when my opponents moved (not including the move itself) and my clock started running when I logged into the server to check the move (or twenty-four hours after a move was made). Number of Simultaneous Games The tournaments were round-robins with seven players, so each player was involved in six games. I played a total of eighteen games. I finished my ongoing games before starting the next tournament. A higher rating was required to be eligible for stronger tournaments, so I wanted the full rating increase for the previous tournament to take effect before registering for the next one. Time Usage I used very little time for the games, compared to my opponents. One reason was that I wanted to finish each tournament as quickly as possible, so I could start the next one. Another reason was that I often had an answer to my opponent’s move in the IDeA tree and didn’t need much additional time for analysis. I also used conditional moves that were played automatically if my opponent made the expected moves. Conditional moves can be a little dangerous and you can easily make mistakes if you are not careful. Therefore, I was taking a slight chance by using them extensively, but I managed to avoid mistakes, so in the end they saved me some time. Sometimes moves were exchanged very quickly and I had sessions that were more like freestyle chess than correspondence chess. On the average, my opponents used around thirty-seven days of their reflection time for each of their games. My total time usage for all eighteen games was less. Results I won sixteen games and drew two (+16 =2 -0), scoring 17/18 or 94.4%. The average rating of my opponents was 1943 (tournament entry rating). My (Elo) rating performance was 2386. In the last tournament, I played four players rated over 2100 with an average rating of 2151 and scored 3?/4. That’s a small number of games, but for comparison the performance rating against them was 2486. I have little information about the rating level on the server where I played. I did a quick check on a few established, 2400+ players who also played on ICCF. It indicated that their rating was similar on both servers (I only compared a few players, so this may be inaccurate). I was surprised by how strong some of the lower-rated players were compared to the higher-rated players. One reason is probably that they are underrated because of a low initial rating and the slow climb up the rating ladder. Perhaps the use of chess engines also plays a role here: players can upgrade their hardware, etc. Avoiding Draws Looking back, one thing that stands out is the emphasis I put on avoiding draws. I always tried to find ways to keep the game going, even when the position out of the opening wasn’t particularly promising. I feel that IDeA is very useful for this purpose. This strategy paid off well; however, my primitive method of playing the opening worked against it. The main problem with the method I used for choosing opening moves is that it sometimes leads to completely even positions where it’s impossible to create winning chances. My two draws were a direct result of that. As an example, I chose a (formerly?) popular variation in engine games leading to the following position Many readers will likely recognize that this is a position from the Sicilian Najdorf variation. I had the black pieces and played 18…Rxa2. It’s probably the best move, but after that White can keep the balance with 19.Kxa2, 19.Nc1, and even 19.Qxb4. Black has no winning chances in this apparently complex position against a good correspondence player, although the draw wasn’t agreed until move forty-three. In other cases I came out of the opening in completely even positions, which required plenty of patience (and a bit of luck) to maneuver into something more promising I had black and looked at some alternatives in this position, but wasn’t optimistic that I would find a way to improve the position. I decided to go for the pawn sacrifice with 15…e4 16.dxe4 d4. White should be able to hold this, but he gets a passive position if he isn’t careful and his extra pawn is of little help. I thought that White was more likely to go wrong here and that’s what happened. This is an example where a strong correspondence player would easily have held the draw. Since I wanted to win as many games as possible, I sometimes avoided the “best” move if I felt it would lead to increased drawing chances for my opponent. Here is one example where I had a difficult decision to make as black The most natural continuation in this position is 22…Nxb6 23.axb6 Qxb6, but after that massive piece exchanges will take place; e.g., 24.Rb3 Rxc3 25.Qxc3 e4 26.Qd4 Qxd4 27.Nxd4 exd3, etc. I felt that White would have drawing chances in the resulting simplified position; therefore, I chose 22…Qd6. This keeps all the pieces on the board with a complicated position. I don’t think you will find a chess engine that would choose this move, as 22…Nxb6 is just too tempting for them. I found that IDeA is often good at distinguishing between drawing lines and winning/losing lines. Here is one such position. It comes from a variation that could have occurred in one of my games Which move draws? It is White’s move and according to IDeA he has only one move that allows him to draw. Which move is that and (more importantly!) why does it draw? The IDeA analysis tree for this position looks like this I have covered up the moves, but if you look at the scores from the Master tree, you will see how clearly IDeA differentiates between the drawing move and the losing moves. If you have several chess engines, try to analyze this position in two-variation mode. Do they find a drawing variation and do they see that all other moves lose, or at least lead to a very difficult game for White? Do all the engines agree on the best move? I would like to hear from readers who analyze this position and I’ll publish their analysis along with my answers next month. What if? I am sure that many readers will have questions after reading this column. I have anticipated some of these; however, nothing can be stated definitively, so take my answers with a grain of salt. How would it have affected my performance if I had used fewer cores? If I had half the number of cores, I would have needed twice as much time to do the same analysis. Since I had so much reflection time accumulated in all my games this would probably not have been a problem. It would simply have taken me longer to complete the games. Further reduction in the number of cores would have affected my play in the end. The effect of having fewer cores is similar to what happens if you have more active games. Up to a point it has little effect, but beyond it the analysis will start to suffer. Would using more cores have improved my performance? In my case, the two draws were unavoidable given how I played the opening. Simply adding more cores would not have changed the outcome of those games. In general, IDeA improves with more powerful hardware, as you can either analyze more positions or analyze each position to a deeper extent in the same amount of time. How would stronger opposition have affected my performance? I believe that my performance rating might have improved somewhat if I had played higher-rated opponents. However, at some point the improvement would stop and further increase in the rating of my opponents would not have much effect. This could have happened even after a modest increase in my opponent’s rating, but it’s impossible to say without actually playing more games against stronger opponents. Conclusions My experience has convinced me that IDeA is a powerful analysis tool for correspondence chess. My opponents never managed to refute the moves found with the help of IDeA. In the beginning I was wondering if I would run into types of positions where IDeA underperformed and failed to find some critical moves. That never happened and none of the moves made by my opponents came as a surprise to me (i.e., I had them in my analysis tree). Although none of my opponents was rated over 2200, the results indicate that IDeA could be used successfully against higher-rated players. At this time I am not sure if I will continue my correspondence chess “career,” but if I do, I will start meeting players rated above 2200. There are still many interesting questions that I want answered regarding the use of IDeA in correspondence chess against strong players. One thing is certain, though – it will be a relief to be able to play at full strength without any self-imposed restrictions. As a byproduct of having used IDeA instead of only infinite analysis in these tournaments, I now have extensive analysis on several openings (mostly late opening stages). One of the advantages of IDeA is that all this analysis is conveniently stored and I can continue the analysis where I left off. I am aware that many correspondence players are using IDeA, but there is little information available about their analysis methods and results. I hope this column will encourage more players to give it a try, although I recommend a gradual transition from older analysis methods