Thursday, July 20, 2006

Trial and error

Today I had a beautiful example of how a patzer approaches a position.
Have a look at diagram 1.

Diagram 1

White to play.

Since I had no clue, I started with trial and error.
I tried the following moves in my minds eye:
1. bxc5
1. b5
1. Rxd5
1. Qxe5
1. Bxa6
1. e4
1. Rxg6
1. Nb7
1. Nc8
1. Nxf7
1. Nc4
None of these tries triggered something in me.
Until I tried 1. Nf5 exf5
At that moment I saw the double function of e6 (protecting the bishop and defending f5) and all pieces of the puzzle felt together. The knightfork, Qg7#, the exchange sacrifice on d5).
It is clear I used an enormous amount of time.
Maybe the thought process that Mousetrapper is developing would have helped me here.

What is remarkable is that the first move of the solution (1. Rxd5) didn't trigger anything in me. The basic pattern of the combination was only revealed to me by imagining 1. Nf5 exf5, which is actually the SECOND move of the combination.

This indicates that you first have to see (as a pattern, in the minds eye) the end position and then work your way back. Which is called calculation.
I have to be more precise. It was not exactly the END position what was recognized.
It was actually the position of diagram 2

diagram 2

Of course this sort postions are not stored in LTM exactly as this diagram. With every pawn and all. Only the main properties are stored and recognized:
  • The knightfork
  • The mate with Qg7
  • The pin along the g-file
  • The black Queen can't prevent the mate without sacrificing herself
The matter I have to solve is:
How can I make that my LTM releases diagram 2 at the moment I see diagram 1.
Or: how can I store diagram 2 in LTM in a better retrievable way.

When you have a close look at diagram 1, you see that the rating of the problem is 1727.
In other words, all 1727 rated players at CTS will solve this problem in 10 seconds at average. What do I have to do so that I can do the same?

I'm experimenting with different methods. Whithout reaching a conclusion sofar.


  1. It probably took me 20 seconds or so.

    The first thing I noticed was that the knight was en prise, then the queen aiming down the long diagonal and the rook opposite the king on the g file. Something like a sacrifice on g6? No, not enough pieces in the area. I did look at Rxd5 briefly (GM Larry Christiansen says to always consider all captures and checks, all the time) but it didn't seem that exiciting.

    What led me to the solution was a search for knight forks, which I usually do if there's a knight in position, especially an advanced one. Hmm, if the black king were on g7 and the black pawn on e6 were missing, I could fork the king and queen with Nf5+! OK, Rxd5 exd5 Qg7+ Kxg7 Nf5+... no, all I've done is lose a rook. But the idea is promising... aha, I can switch the order of moves and threaten the queen and mate simultaneously.

    I'm about your level, 1572 on CTS right now, although I usually am about 1600 and my high is 1635.

  2. Tempo, this is a good example of multiple patterns in a single position. For me, it is too much to imagine that both white rooks, the knight, both queens, the pinned pawn, and the bishop are all one pattern. However, it is easy to recognize the pinned g-pawn, the white queen's access to the squares around the king, and the capture RxB immediately. What is difficult is to recognize that they all work together!

    Another interesting point is that when I first see this diagram, I notice that the white knight is in danger and so I consider protecting it or moving it. Since I'm looking at moving the knight for this reason, I have some help in seeing the usefulness of Nf5. I wonder how the thought process would change if the knight were on g3 instead. The same moves should work, RxB and Nf5, but Nf5 might not come to mind as quickly since the knight doesn't find itself in danger.

    A minor note about your post, you say "the problem is rated 1727. In other words all 1727 rated players will solve this problem in 10 seconds at average." In fact, 1727 rated players will solve a 1727 rated problem half the time, not all the time. Maybe this is what you meant by "at average."

  3. Well, the average rating change when a 1727 rated player faces a 1727 rated problem is 0. And the rating change is 0 when the problem is solved in 10 seconds. So in a real sense the average result for a 1727 rated player facing a 1727 rated problem is that he solves it in 10 seconds. Sometimes he will solve it faster, sometimes he'll solve it slower, sometimes he won't solve it at all. See

  4. Ok, that's about the problem. But can you guys say something about the issue I'm trying to address in this post?

  5. Well, what I was trying to say is that I didn't see position 2 and work my way backwards; I saw some patterns in the original position and played around with them. I think that your position 2 is a little too weird for me to keep it in my head as a pattern, although maybe that is why I am only 1600 on CTS and 1800 USCF.

    It is true that there are certain "goal position" patterns that I have in my head and then try to figure out how to get to. For example, it may be pretty clear that I'm going to mate with a rook on h8 supported by a bishop along the long diagonal, or at least that it's a main theme in the position. But in this case I think the way to solve it is to recognize the little subpatterns (g6 pawn is pinned, queen is on a nasty diagonal, knight is deep in enemy territory so look for forks) and try to combine them rather than recognize a goal position and try to aim for it.

    I'm be interested to hear from people who are better than us and can solve the position quickly, to see how they did it.

  6. Diagram 1 took me 12 seconds. 1. Nf5 exf5; 1. Bh5 (slow), 1. Rxg6 (nothing), 1. Rxd5! exd5 2. Nf5! After working through Polgar's mates several times through, I tend to see mates and mate threats quicker than other types of tactical problems.

  7. Dan, maybe we are saying the same thing about the rating? The average result for a 1727 against a 1727 is 0 points. Half the time they will solve it faster than 10 seconds, half the time slower. I can see that what I wrote initially did not convey this as I left out the time, thanks for the correction/clarification.

    Tempo, sorry to only address the problem of the position and not the problem you brought up in your post. I suppose if I could answer your question of how to train ourselves to use our patterns together, I would be much higher rated!

    On a more useful note, I think CTS does help with this. Once you are able to solve the problems that rely on a direct tactical motif, your rating will go up and you will begin to practice on problems with multiple interacting motifs. I think your approach to resolve carefully and repeatedly the CTS problems you fail will work. Practice makes perfect, as they say.

    Also, 1727 is roughly 130 points higher than your CTS rating, so one might expect that these problems will come much easier to you in time. You have experienced a kind of magical accumulative effect before, it can happen again!

  8. dirk, i am at lunch for second from work and have to run now. but i just loved this post. thank you. keep em coming!

    very smart stuff...

    i couldnt do this problem and i am 1511 at CTS. i never saw Nf5, then again, im in a hurry here, but not in ten seconds, no!

    i did eventually see the loose B at d5, whereby IF the Q took the N, then white could construct something whereby he could attack with e4... and only got as far, at a glance as sensing the key to the solution was there. but ten seconds? no. that is why i am 1511. its amazing to me to know some not only see this stuff, but near instantly.

    or blindfolded. or blindfolded simuls...

  9. 40 seconds for me. I will say that I looked at about half of the moves you listed before looking at the pin on the g-file. Maybe pins and absolute pins are so easy to spot you might consider that to be a deliberate calculation in the first few seconds.

    I would see obvious 2 approach options. One, ignore mistakes, keep moving on, and eventually all parts of your game will be stronger through massive repetition, or two, attempt to understand all mistakes, collect them, and review them.

    I like collecting positions of my worst mistakes, but I am lazy about reviewing them. I think that addressing them will be a life-long challenge. If I followed through this it probably would help me significantly. I suppose you could argue it's training to be cool, not gamma-bursting, when the position (like a sharp opening) or a similar occurred :-).

  10. Trans,
    thx for the cheering!

    I would see obvious 2 approach options. One, ignore mistakes, keep moving on, and eventually all parts of your game will be stronger through massive repetition, or two, attempt to understand all mistakes, collect them, and review them.

    I don't appreciate a slow solution different from an error. As CTS does. Both indicate that you didn't use LTM to solve it. The error you have to understand first of course, but after that it is equal to a slow solution. What is the best method to store it in LTM? I looked in the history of the problem and saw that I didn't solve it either a month ago. What do I have to do to solve it within 10 seconds the next month? I'm not lazy, I'm prepared to do everything. But what is the most efficient way to do this? CTS isn't efficient enough. If you are very determined you can squeeze a few ratingpoints from it, but it's actually not doable if you want some sort of life next to chess. Besides that, I hate the idea to work in a non efficient way.

    I'm experimenting with doing every problem 3-4 times in a row (inspired by how to learn a motor skill), with conscious staring at the main elements etc..
    Maybe there are readers of my blog who are willing to help and do experiments at CTS too.

  11. I counted target features (how I do it? see my blog!): g7 (4), Knight a6 (4) and Queen e7 (3) for attack and Knight d6 (3) for defense. Because two of the major targets could be further attacked by the knight, this could be a hint for the move search. It is also important to see that the key square f5 is only protected by one single pawn. Therefore: Rxd5 exd5 removal of the guard! And now the target feature counts are 5 for the Queen and 8 for the mate in one. Anyway, the important point is that we must know the target before we attack it.

  12. Perhaps I am vague about this. Isn't immediately seeing the absolute pin on the g-file using LTM, or at least partially LTM?

  13. King,
    yes, that's true. Do you sense a flaw in my reasoning?

  14. This comment has been removed by a blog administrator.

  15. Not in your reasoning, exactly, but I guess I disagree with your framework. I think long term memory in chess may seem like simplistic pattern recognition, but I think it includes vital associations, parallels between positions, memories about what looked similar and the strong and weak elements that might be shared, old decisions and rewards and punishments that came with them, etc. I believe that with each memorized pattern there is the "rolled up" reasoning (and calculation) behind it, so even though you might recognize something in under 3 seconds, it typically takes longer to formulate words to express it completely as you unroll it. That's my take on it.

    >>"Since I had no clue, I started with trial and error."

    What seems striking to me is that you did NOT post that you "saw the absolute pin but didn't see the connection", or you "thought it actually wasn't part the answer." Even from a pure pattern recognition point of view, that element alone really stand out, especially if nothing else in the position seems like a recognizable pattern.

    Lastly I believe calculation adds to these patterns when learning them. It seems obvious. I can think of plenty of endgame positions where even if I remember one right move, I might have trouble if my opponent plays perfectly because I didn't formulate something well enough to unroll it.