Friday, 23 December 2016
008. Spoilsport Chess
White: M.W. Johnson - BCCA Championship, 1992/93
Issue #102 (April 1989) of the BCCA magazine (before my time as editor) featured an article by Peter Millican on the King's Gambit, Double Muzio. The basic tabiya arises after 1 e4 e5 2 f4 exf4 3 Nf3 g5 4 Bc4 g4 5 0-0 (the Muzio) 5...gxf3 6 Qxf3 Qf6 7 e5 Qxe5 8 Bxf7+ (“doubling up”) 8...Kxf7 9 d4 and if 9...Qxd4+ then 10 Be3 Qf6 11 Bxf4. Yes, White is two pieces down here, but the rest will all soon be in play and attacking, whereas Black's forces are almost all still at home, while the king sits uncomfortably on the f-file. Peter concentrates on this line, assessing it as “objectively equal”, and supplies a fascinating, at times brilliant exposition of the attacking resources at White's disposal.
Offering the article for download on his website, Peter comments: “It's all good fun, though I've since discovered a lot of improvements” and, pertinently, “Some day it would be good to redo the entire article with computer assistance.” Yes, indeed. As I know only too well, undertaking highly complicated, tactical analysis without a chess engine to tidy it all up is fraught with peril. Today's engines can rip everything to shreds in minutes, or even seconds.
For instance, Peter's main game (Millican-Down, BCCA Gambit Tournament 1986/87) continued 11...Ne7 12 Nc3 Nf5 13 Ne4 Qg6 14 g4 Be7 15 Kh1 Nh4 16 Qe3 Kg8. Here he rejected 17 Be5 (Estrin & Glazkov) on account of 17...b6! (Korchnoi, ECO) and played 17 Bh6!! (threatening 18 Nf6+ Bxf6 19 Rxf6 Qxf6 20 Qe8+) which won by force: 17...Qe6 18 Rf2 b6 19 Raf1 Ng6 20 Qd4 Bf6 21 Nxf6+ Kf7 22 Nd5+ Ke8 23 Qxh8+ Nxh8 24 Rf8 mate. Very nice. However, my Houdini software is less impressed, pinpointing Black's 16th move as an outright blunder, and refutes White's play with, ironically, the refutation of 17 Be5, only a move sooner: 16...b6!, when Black keeps the option of both ...Ke8 and ...Kg8 and appears to win in all variations. The emotionless computer gives not one whit about the discovered check.
That doesn't mean it's all over after 12...Nf5. Instead of 13 Ne4(?), White does much better with the natural 13 Nd5(!), while Houdini suggests a startling improvement of its own: 13 Be5!? (the Triple Muzio!), which it rapidly calculates to a draw: 13...Qxe5 14 Qh5+ Ke7 15 Qg5+ Ke8 16 Rxf5 Qe7 17 Re5 Kd8 18 Rae1 Nc6 19 Rxe7 Bxe7 20 Rxe7 Nxe7 21 Nd5 Re8 22 Nxe7 Rxe7 23 Qg8+ Re8 24 Qg5+ and so on. In other words, despite the computer finding flaws, sometimes serious ones, here and there in his 1989 analysis, Peter's intuitive/empirical assessment seems still to stand up. Moreover, in practical play, White has a huge plus score. It's easy to pick holes in variations after the event, especially with multi-processor enhancement.
But before the silicon era was even a dream, there was the first World Champion and first great spoilsport, Wilhelm Steinitz, who subjected his contemporaries' enthusiastic commentary to critical scrutiny and found it wanting. Steinitz looked at the Double Muzio too, but rather than get involved in the shenanigans after 9...Qxd4+, he recommended that Black avoid the whole thing by playing 9...Qf5!. I first discovered this in Stefan Bücker's interesting little book Das neue Königsgambit (Schach bei Franckh 1986), where Stefan wrote:
“Wäre das zweite Figurenopfer wirklich so stark, so hätte man es in der Blütezeit des Muzio-Gambits gewiß nicht übersehen. In Wahrheit hat schon die folgende Partie 1889 den Nachweis für die Inkorrektheit erbracht: Showalter-Taubenhaus, New York 1889. 9...Df5! (“Am besten”, Steinitz) 10 g4 (10 Lxf4 Sf6) 10...Dg6 11 Sc3 Sf6 12 Lxf4 d6 13 Lg3 Kg7 14 Sd5 Sxd5 15 Dxd5 Sc6 16 Dc4 d5 17 Dxd5 Le6 -+ (0:1, 26). Oder 11 Lxf4 Sf6 (11...Le7) 12 Le5 Le7 13 Lxf6 (13 Sc3 d6 14 Lxf6 Lxg4) 13...Lxf6 14 Sc3 (Nugent-Smith, USA 1909) 14...Kg7 15 Sd5 Tf8 16 Kh1 Sc6 -+.” (If you don't read German, Google Translate will give you the gist.)
Peter noted similar lines himself after 10 g4 and therefore, in Millican-Rawlings from the same BCCA tournament, opted for 10 Bxf4 Nf6 11 Qe3, which led to another rout: 11...Bg7 12 Be5 Qe6? (12...Qg6 13 Nc3 Re8 14 Nd5 d6 is more testing) 13 Nc3 d5 14 Rae1 Nbd7 15 Qf3 Nxe5 16 dxe5 Qb6+ 17 Kh1 Rf8 18 Nxd5 Qd4 19 exf6 Bxf6 20 Nxf6 1-0. Houdini evaluates Black as already worse after 15...Nxe5 and losing after 16...Qb6+.
But if you're aiming to be a spoilsport, I think you should do it properly. In such mean spirit, I played 11...Qe4!? in the game below.
What is really annoying about this move is that White can now regain virtually all the sacrificed material – i.e. 12 Qxe4 Nxe4 13 Be5+ Kg8 14 Bxh8 with rook and pawn for the two pieces – but then the anticipated attack has completely vanished. Worse still, after 14...d5 15 Nc3 Nxc3 16 bxc3 Nd7, the two pieces will soon be two bishops, which means a horrible endgame for White. Preserving the h8-bishop from exchange, on the other hand, would cost a pawn; e.g. 15 Be5 Nc6 16 Bxc7 Nxd4 17 Nc3 Nxc3 18 bxc3 Ne2+ 19 Kh1 Nxc3 and White is pretty much just losing.
Understandably, Maurice (who was British Correspondence Champion at the time) wasn't too keen on any of that, so he retreated the queen again. But then Black has a superior version of Millican-Rawlings, and I won quite quickly (albeit without the reassurance of a powerful engine predicting the result in advance). That Houdini of course suggests improvements on my play, too, hardly matters since Black's play doesn't need improving. It's White who has to improve – and there the computer has nothing to say.