After a quick look, I don't like 18.Rg1.  After 18...Ne6 19.Qe3 Qg3 the rook is out of place on g1 (just a wasted tempo) and the h4-pawn is hanging.  Although some speculative lines may be possible where the h4-pawn is sacrificed and the open h-file provides compensation, I'm doubtful of this (besides, Black is not obliged to take the h-pawn and the Rg1 will still be out of place).  I will briefly look at lines with Qf2 (instead of Qe3) to make sure I'm not discarding Rg1 too hastily (and will do the same for 18.g3).

Right now 18.Qe3 is looking very likely, as long as no fatal flaws turn up in concrete lines.  I will also at least briefly consider other playable moves to see if there is an important plan or resource that I'm overlooking by focusing too narrowly on the Qe3/Rg1/g3 plan mentioned above.

Moves like g4, Nc5, and Re1 come to mind, although none seem likely to be preferable to Qe3 (sometimes analysis of such moves reveals resources or pitfalls which will be important later, so even unlikely moves can be worth looking at briefly).

Since it was easier to narrow down the top choices than I anticipated, I will likely post the next move late tonight (or maybe in wee hours of Thurs morning depending on your time zone).

Sorry for the longwindedness (is this even a word?).  Maybe with practice I will learn to be more succint...  

Another thought about the 15...Qc7 16.h4 line.

In the only other game where this specific line was previously played, I couldn't understand why White chose 17.Rd2, when there seemed to be several much more purposeful moves available:

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 g6 6. Be3 Bg7 7. f3 O-O 8. Qd2 Nc6 9. O-O-O d5 10. exd5 Nxd5 11. Nxc6 bxc6 12. Bd4 Bxd4 13. Qxd4 Qb6 14. Na4 Qa5 15. b3 Qc7 16. h4 h5 17. Rd2 Rd8 18. c4 e5 19. Qc5 Nb6 20. Rxd8+ Qxd8 21. Nc3 Qf8 22. Ne4 Qxc5 23. Nxc5 Nd7 24. Nxd7 Bxd7 25. Bd3 Kf8 26. Re1 Re8 27. c5 Bf5 28. Ba6 f6 29. Rd1 e4 30. Rd6 exf3 31. gxf3 Re6 32. Rd8+ Re8 1/2-1/2 Daurelle-Koch, corr 2005.

It now occurs to me that the main purpose of of White's 17th move may have been to avoid the fork ...Qf4+, forcing the exchange of queens!  In this game, Daurelle seems have concluded that White's best winning chances required that queens remain on the board.  Looking at the game continuation after 17. Rd2, we see it is Black who eventually pushes for the queen exchange.  I doubt this could be considered a model game (for one thing, neither side seemed to be striving for a win - perhaps it was a tournament situation where White only needed a draw?).  However, since the queen exchange has been a major point of discussion in this thread I thought the apparent role reversal seen in this game (i.e., White avoids/Black seeks the exchange) is quite interesting.

Smyslov_Fan, I know that you advocated for the queen exchange in some lines that we discussed, but my impression was that you saw this not as something desirable for Black but rather as a lesser evil.

Smyslov_Fan wrote on 07/19/06 at 07:10:29:

There wasn't a specific reason for this choice, but this seemed to work out better than a few more commital alternatives (e.g., 18...e5).  I don't recall, but it might also have been a Fritz suggestion.

I didn't like moving the bishop yet since the knight is going to be pushed back to e6 soon (however, bishop moves other than ...Be6 are still plausible), and I didn't like moving the queen.  I've also considered moves like ...Kg7 or ...Kh7 in order to bring a rook over to the h- or g-file, but White can probably aim for the same kind of setup.  18...Re8 and 18...Rb8 were two of the moves I looked at first because they are useful without being commital.

With 17...Nf4 on the board, White's main plan involves Rg1, Qe3, g3, and Bh3 followed by a well-timed g3-g4.  There are transpositional and move order issues to work out, and a few minor alternatives or slight variations of this plan to consider (e.g.: Bg2 instead of Bh3; fixing Black's kingside pawns on light-squares instead of opening files; bringing the knight back into play via c3 or b2 instead of waiting for the opportune moment to play Nc5).

As I dig a bit deeper into the variations I'm finding more instances where positions which I thought looked safe for Black just dissolve a few moves later -so it looks like Black's game might be a bit more difficult to navigate that it seemed a few days ago.

The sample line gave a few days ago (http://www.chesspub.com/cgi-bin/yabb2/YaBB.pl?num=1151879306/60#71) is still representative of what I consider to be the main line: 18.Qe3 [18.Rg1 could be subtly different, but probably tranposes] 18...Rb8 19.Rg1 probably followed by 20.g3 Ne6 21.Bh3 (I think this set up will be possible against most 18th and 19th move choices by Black) 

One emerging feature of some of these lines that I didn't give much consideration to earlier: exchanging white's light-squared bishop for Black's knight on e6 seems to add considerable force to the g3-g4 advance.  After this exchange: in a position with white rooks on the h- and g-files and the queen on e3, the sequence g4, ...hxg4, then h5 followed by hxg6 can be pretty devastating.

Setting up this type of kingside breakthrough takes time, and there aren't enough forcing/tempo-gaining moves to ensure that White can always aim for this.  In most lines, best play (or my best estimation of this) for both sides leads to an ending where White has a tangible advantage (e.g., sometimes pawn-up; other times three connected pawns vs. three isolated pawns) but it is not always clear whether the advantage will be enough to win.

Candidate moves are 18.Qe3, 18.Rg1, and 18.g3.  I will make a choice in a day or two as soon I have had time to work out whether there is an optimal move-order.

Edit:  I don't like the look of 18.g3 because after 18...Ne6 the queen is attacked and the g3-pawn is hanging.  So probably 18.Qe3 or 18.Rg1 first and g3 later.

I worry about Black's position after 18.g3 Ne6 19.Qe3 because the natural looking 19...c5 and 20...Nd4 could get swatted by 21.Nxc5.  If black doesn't have the d4 square, I don't know how he will successfully contest the d-file and the dark squares.  This was my rationale for the rather poor choice of 17...Qf4.  As bad as that was, this doesn't seem any better.  In fact, Black's pieces seem to get in each other's way more in this line.   

Scholar, 

I'm sure you'll prove me wrong, and I look forward to learning from you on how you handle this difficult position.

OstapBender wrote on 07/17/06 at 16:46:39:

Actually it's not very much.  Just that after 17...e5 18.Qc5 I thought that Black should continue with 18...Nb6 (instead of 18...Nf4) buying time to get the bishop out and connect the rooks.  Then we have 19.Nc3 (this specific move is not forced, but White can't really tolerate 19...Nxa4 so Black still gets the tempo to develop the bishop against alternatives) 19...Be6. 

Now 20.Rd6 (my main objection to 17...e5 in other lines) is well met by 21...Rad8 (actually, I think Black is better here; 20.Rd6 is dubious).  But Black still has to contend with 20.Qd6.  Like I said, it isn't much.

I have't looked at the position after 20.Qd6, but I guess with queens coming off the board (e5 hangs, so there's not much choice) Black's knight doesn't serve much of a purpose on b6 (i.e., the ...a5-a4 advance lacks punch).

So far I still like 17...Nf4 best.  17...Nb4 sounds interesting, though.  White might be provoked to weaken the queenside to drive back the knight - but the knight winds up on a6 when it could be very time-consuming to bring it back into play.  I doubt this will prove to be better than ...Nf4, but specific lines need to be worked out before this can be ascertained.

Scholar, 

I think you're right.  I underestimated 18.Qc5 Nf4 19.Qd6 and White will control the d-file.  Even worse for Black, I think White will win a pawn, grab the 7th rank, or both.  

If White has control of the d-file after the queens are traded, then White will have achieved something special.  If Black can block up the d-file and then contest it himself, then I think Black won't have minded trading queens. 

PS: OSTAP:  I'm missing the diagrams for this page, and it's almost full already!