ELECTRONIC T-NOTES


CHESSBASE USA'S WEEKLY ON-LINE NEWSLETTER


FOR THE WEEK OF FEBRUARY 18, 2001


LEARNING A NEW OPENING -- PART 3

by Steve Lopez

Last week we looked at the initial steps in learning a specific opening variation (input the line by hand, annotate it, and save it into a database). But we're not yet finished with this starting variation; there's a lot more we can do with it.

You'll recall that we looked at the following variation of the Queen's Gambit Accepted:

1.d4 d5 2.c4 dxc4 3.Nf3 Nf6 4.e3 e6 5.Bxc4 c5 6.0-0 a6 7.a4 Nc6 8.Qe2 cxd4 9.Rd1 Be7 10.exd4 0-0 11.Nc3 Nd5

We've annotated the moves to get an idea of why they're played, so that we know how we got to this position. Now we'll consider what happens next as we move into the middlegame (so that it doesn't become the muddlegame).

The easiest thing to do is just fire up a chessplaying engine and have it analyze the position after 11.Nd5. In ChessBase 8, you just hit the F3 key and choose the engine you want (see ETN November 12, 2000). In Fritz6 you do the same thing but, after loading the engine, you'll need to hit ALT-F2 to start Infinite Analysis mode.

Either way, the engine will start chewing on the position and will continue to do so until you stop it. Once it's done, you can paste the analysis into the game (using CB8) by right-clicking in the engine analysis pane and selecting "Copy all to notation"; this will drop the analysis into the game as a set of replayable variations. In Fritz6, you can right-click in the analysis window and select "Clip analysis". This won't copy the analysis directly into the notation as in ChessBase 8, but it will send it to the Windows clipboard and you can then open an annotation window in Fritz (by hitting CTRL-A) and drop in the analysis as text by hitting CTRL-V. Either way, be sure to use "Replace game" if you want to save the analysis.

I let Fritz6 chew on the position after 11...Nd5, displaying the three best lines of play, and here's what it came up with:

1) 12.Ne5 Ncb4 13.Qg4 Bh4 14.a5 b6 15.axb6 Nxb6 16.b3 N6d5 0.25/14
2) 12.Qe4 0.09/13
3) 12.Bb3 Ncb4 13.Nxd5 Nxd5 14.Bc2 b6 15.a5 bxa5 16.Qe4 g6 0.06/13

We see that in the top line, Fritz thinks that White is better by a quarter-pawn. The caveat here is that you can't take computer analysis at face value. Computer programs often misanalyze closed positions or ones in which long-term planning is required. Our primary interest here is in getting a gauge of whether or not a position is sound. In this case, the position's not blocked up, so I think the analysis is pretty trustworthy. White appears to be all right, but Black's not at a huge disadvantage, so neither player needs to worry here.

A more elaborate type of computer analysis is called "Deep Position Analysis", which will generate a whole tree of move possibilities from a given position. With the position set up on the chessboard in Fritz (in this case, we're looking at 11...Nd5), go to the Tools menu, select "Analysis", and then select "Deep Position Analysis" from the submenu. This will bring up the following dialogue:


You'll use this dialogue to set your parameters for the Deep Position analysis. The "Time" and "Depth" settings are mutually exclusive; you can't set both. In the graphic, you'll see that I set the time to 45 seconds per move. You'll need to play with this setting to see what works best on your computer.

The "plus (root)" field lets you set additional time (in seconds or moves) for the computer to use in analyzing the initial position (in this case, the position after 11...Nd5). Some folks think that the initial candidate move selection is the most important part of the process and deserves more time than the consideration of later moves. This field lets this extra time be added to consideration of the initial candidate moves.

Skipping down to the bottom of the dialogue, "Branching" lets you determine whether you want to see move alternates for White, Black, or both. "Length of variations" helps determines the overall size of the tree. In my case, I set the variations to be nine plies (half-moves) deep -- in other words, 4.5 moves. "Evaluation window" is given in 100th/pawn increments. In the case of my example, if an alternate move is worse than the best move by 30/100ths of a pawn or greater, Fritz won't bother investigating that line any further.

The branching factor settings will control the size of the tree more than any other factor. My setting of "3" for "Branching in 1st move" means that Fritz will consider and display (as long as the "evaluation window" criterion is met) three possible 12th moves for White. At any point in those three variations, Fritz may also show up to two alternatives (controlled by "Branching in 2nd move"). And, going still deeper, Fritz may even branch off from those subvariations, providing up to two alternatives at any point in those (controlled by the "Branching in 3rd move" value). This, combined with "Length of variations", gives you total control over the size of the tree.

At this point, it would probably be beneficial to examine Fritz' output for the tree after 11...Nd5 before we proceed further:

12.Ne5 [12.Be3 Ncb4 (12...Bb4 13.Rac1 (13.Qd3 Be7 14.Ne5 Ncb4 15.Qe4 f6 16.Bd3 Nxd3= 0.12/12 ) 13...Nce7 14.Qc2 b6 15.Bg5 Bxc3 16.bxc3 Bb7+/= 0.34/12 ; 13.Ne5 (13.Rac1 b6 14.Ne5 Bb7 15.Qg4 Nxe3 16.fxe3 Bg5= 0.09/12 ) 13...Nxe3 14.fxe3 Bd6 15.Nd3 Nxd3 16.Bxd3 b6= 0.09/13 ; 12.Qe4 Nf6 (12...Ncb4 13.Ne5 (13.Re1 Nf6 14.Qe2 b6 15.Be3 Bb7 16.Ne5 Nfd5= -0.09/12 ) 13...a5 14.Qg4 Kh8 15.Bd2 f6 16.Nf3 Bd7+/= 0.28/11 ; 13.Qf4 (13.Qh4 Qc7 14.Bg5 h6 15.Be3 Rd8 16.Rac1 Bd7= 0.00/11 ) 13...Nb4 14.Qg3 Bd6 15.Qh4 Qc7 16.b3 Nfd5= 0.00/11 ] 12...Ncb4 [12...Bb4 13.Qd3 (13.Qf3 f6 14.Nxc6 bxc6 15.Bd2 a5 16.Rac1 Rb8+/= 0.37/12 ) 13...Nce7 14.Bb3 f6 15.Nc4 Bd7 16.Bd2 Qc7+/= 0.47/11 ] 13.Qg4 [13.Qe4 Qd6 14.Qg4 Kh8 15.Qh5 Kg8 16.Nxd5 exd5+/= 0.31/12 ] 13...Kh8 14.Qh5 Qe8 15.Bxd5 Nxd5 16.Qf3 b6+/= 0.37/12

Now I'll be the first one to admit that this is practically unreadable in text format, but it makes perfect sense when viewed in replayable .CBH format in Fritz. Please allow me to do some editing and some descriptive exposition and we'll work our way through some of the analysis.

The first thing Fritz did was analyze the position after 11...Nd5. It was set for a 45 second analysis, and didn't take much longer than that to come up with three possibilities for White:

12.Ne5
12.Be3
12.Qe4

Looking at these one at a time makes things a bit easier. For 12.Ne5, we see:

12.Ne5 Ncb4
[12...Bb4 13.Qd3 (13.Qf3 f6 14.Nxc6 bxc6 15.Bd2 a5 16.Rac1 Rb8+/= 0.37/12 ) 13...Nce7 14.Bb3 f6 15.Nc4 Bd7 16.Bd2 Qc7+/= 0.47/11 ]

13.Qg4
[13.Qe4 Qd6 14.Qg4 Kh8 15.Qh5 Kg8 16.Nxd5 exd5+/= 0.31/12 ]

13...Kh8 14.Qh5 Qe8 15.Bxd5 Nxd5 16.Qf3 b6+/= 0.37/12

Isn't that a whole lot more workable? Going back to the "branching factors" in the dialogue, we see there were three moves in the first branch (White's 12th move). At a couple of key points in the 12.Ne5 branch, we got a second branch (at Black's 12th and White's 13th branches). And in the 12...Bb4 secondary branch, we got a third level branch at White's 13th move. And the nice thing is that White's fairly safe here -- every line ends with a slight advantage for White.

Here's how Fritz did the analysis. After 45 seconds or so, it saw three possibilities for White's 12th move that were within the 30/100ths of a pawn range we'd set. It then started analyzing the position after 12.Ne5. It saw two replies for Black that fell within the 0.30 range: 12...Ncb4 and 12...Bb4. It then looked at 12...Bb4 for around 45 seconds and spotted two replies for White: 13.Qd3 and 13.Qf3. It then considered 13.Qf3 for about 45 seconds and decided that Black's best reply was 13...f6. It then looked at this position for about 45 seconds and decided on 14.Nxc6 for White. It then continued to look at positions for about 45 seconds a crack until it got out to 16.Rb8 (taking it to the nine ply variation length we'd set in the dialogue), gave an evaluation of the position, and then jumped back to 13.Qd3 to start analyzing that line out to nine plies. After finishing that line (and evaluating it), it went back to 12...Ncb4 and started analyzing. After about 45 seconds, it came up with two replies for White (13.Qg4 and 13.Qe4) and started down those lines.

The point I'm making here is that Deep Position Analysis is very thorough (much more so than the ALT-F2 Infinite analysis) but also very time consuming. This is a feature that you'll definitely want to have running overnight on your computer, when you're not going to be using it for other tasks. It's also a lot of fun to watch (for a while) as Fritz creeps through the analysis step by step, rolling slowly ahead like an unstoppable juggernaut until the task is completed. (And, while we're talking about completion, make sure you use "Save game" or "Replace game" when the analysis is finished, or you'll be an extremely embittered individual -- Deep Position Analysis is not saved automatically, so be sure to do it yourself).

Let's have a looksee at White's second alternative, 12.Be3:

12.Be3 Ncb4
[12...Bb4 13.Rac1 (13.Qd3 Be7 14.Ne5 Ncb4 15.Qe4 f6 16.Bd3 Nxd3= 0.12/12 ) 13...Nce7 14.Qc2 b6 15.Bg5 Bxc3 16.bxc3 Bb7+/= 0.34/12 ]

13.Ne5
[13.Rac1 b6 14.Ne5 Bb7 15.Qg4 Nxe3 16.fxe3 Bg5= 0.09/12 ]

13...Nxe3 14.fxe3 Bd6 15.Nd3 Nxd3 16.Bxd3 b6= 0.09/13

And, again, we see no danger here for White.

Finally we consider 12.Qe4:

12.Qe4 Nf6
[12...Ncb4 13.Ne5 (13.Re1 Nf6 14.Qe2 b6 15.Be3 Bb7 16.Ne5 Nfd5= -0.09/12 ) 13...a5 14.Qg4 Kh8 15.Bd2 f6 16.Nf3 Bd7+/= 0.28/11]

13.Qf4
[13.Qh4 Qc7 14.Bg5 h6 15.Be3 Rd8 16.Rac1 Bd7= 0.00/11]

13...Nb4 14.Qg3 Bd6 15.Qh4 Qc7 16.b3 Nfd5= 0.00/11

And it's again apparent that White's in no trouble here, though the evaluations are best after 12.Ne5 compared to the two alternatives.

So now we have some idea of the relative safety of the variations we've chosen to study. The task for us now is to play through all of this analysis and take a look at what each player is doing in the early middlegame so that we'll have some idea of what we can expect in our own games. We should also look at the final positions to see if we find them congenial -- in other words, are we comfortable with them? Are they positions we wouldn't mind seeing in our own games?

But, again, remember to take Deep Position Analysis with more than a grain of salt. Computers play like computers and they can't make plans the way humans can. So the analysis you get should be used as a guidepost but shouldn't be taken as some divine oracle of truth (and, as Indiana Jones said, if it's truth you're looking for, the Philosophy class is down the hall). What we're looking for here is advice -- we want a rough roadmap of what to expect in the middlegame. Computer chessplaying programs are pretty good at this, but they're not infallible. So be sure to always think for yourself! Certainly consider what the computer shows you, but consider it with care and always use your own head.

Another way of getting advice and middlegame information (this time from humans) is to use a really killer feature of ChessBase 8. We'll examine that one next week. Until then, have fun!

You can e-mail me with your comments, suggestions, and analysis for Electronic T-Notes.