It would be nice to get good scans of covers, frontpieces, copyright pages, and even prefaces...

(Which I think would qualify as "fair use").

Here's an example with Kmoch's picture and signature:

White moves and draws:

click for larger view

The "instructive composition" is by W.E. Rudolf from La Strategie 1912 and not by Hans Kmoch as I first thought. It was clearly off-topic for the Fischer vs Petrosian, 1971 game so I've posted the relevant links here even though, since it wasn't composed by Hans Kmoch, some may consider it off-topic for this page also. Here are the links in case anyone is interested in what preceded this post.

Fischer vs Petrosian, 1971 (kibitz #162)

Fischer vs Petrosian, 1971 (kibitz #165)

Fischer vs Petrosian, 1971 (kibitz #166)

Fischer vs Petrosian, 1971 (kibitz #167)

Fischer vs Petrosian, 1971 (kibitz #168)

As far as the d=35 barrier, after reading <and understanding> the Komodo documentation, there is a formula that allows you to estimate the amount of time that you can run an MCTS-based analysis as a function of the amount memory you allocate to the MTCS table. And, if the MTCS table becomes full, the analysis just stops.

Sure enough, the number of nodes that it evaluated in all 3 MCTS-based analyses were about the same, so the MCTS table just got full. So, if I want the analysis to run to a greater search depth, I just have to make the MCTS table bigger. I just don't know how much bigger I can make it without the program starting to swap its code and data to disk and slow down to a crawl so I'll have to experiment. This might take some time.

Also mildly interesting and perhaps not surprising (but surprising to me), the analysis with 1 core required about 4X longer to reach d=35 than it took for the analyses using 4 cores. It was surprising to me because these multi-threaded algorithms do not typically scale linearly; in my computer an analysis using a "regular" engine with alpha-beta pruning (ABP) instead of MTCS takes roughly 2X as long with 1 core to reach the same search depth as the same analysis using 4 cores. Again, the Komodo 12 documentation alludes to this, saying that "we believe Komodo MCTS gains more than most normal engines from using multiple threads". I'm sure that they would be very interested in data that indicates that it benefits linearly rather than by a factor of SQRT(2).

Which then leads me to run more analyses to try to find out if:

(a) Komodo MCTS is deterministic just like Komodo "normal" when using only 1 core.

Thanks. I suspected a super-duper box might toil because it is rather long - it has to get to the 50 moves marker to make a draw and there are quite a few pieces left on the board.

A reasonable human player can see within seconds it's a draw but the box has to calculate.

No harm done. It's a freak position as far as a computer goes.

Thanks - in this case humans:1

So I started the analysis in the position after 1.Ba4+ Kxa4 2.b3+ Kb5 3.c4+ Kc6 4.d5+ Kd7

click for larger view

To my surprise, delight, and puzzlement, Stockfish selected 5.e6+ as its 3rd best move with an evaluation of [-15.27] at d=27 and 9 seconds of calculation and selected it as its best move with a [-15.28] evaluation at d=31 after 35 seconds of calculation, doable under classic time control conditions. After d=35 and a little bit over 8 minutes of calculations it evaluated 5.e6+ as the only move to avoid mate.

I'm puzzled why it did that. After 5.e5+ dxe8 6.f5 looking at its material deficit alone and using the classic piece values (Q=9, R=5, B=3, N=2, P=1) the evaluation would have been [-13.00] (down a bishop and 2 rooks) while after 5.exf6 exf6 6.Bxf6 its evaluation would have been [-9.00] (down two rooks but up a pawn). So, for some reason, it evaluated Black's positional disadvantage after 5.e5+ much less than after 5.exf6. I still have no idea why but that's a mystery for another day (and for another person!)

And, like I said, I hardly have a super-duper box. I did some more calculations and I projected that, in my computer, it would take Stockfish 9 about 1,250 hours or 52 days for it to reach d=57 and determine that the position is a draw. But after it evaluated all moves except 5.e5+ as leading to a forced mate, it began to reach additional search depths quicker. And after it passed d=60 I realized that I had made a foolish mistake. It's not a 50 <ply> draw rule, it's a 50 <move> draw rule. So instead of needing to reach d=57 it would need to reach d=114, and I projected that it would take about 850 hours on my machine for it to do that. No thanks, I don't have that much patience to wait to see if its evaluation of 5.e5+ would really be [0.00] at around d=114. So I pulled the plug at d=75 after about 15 hours of calculation. Maybe someone like <RandomVisitor> with his powerful hardware might be interested in seeing if he can get Stockfish 9 to run long enough to give 5.e5+ a [0.00] evaluation.

The other thing I did establish after another analysis run with Komodo MCTS is that, unlike "classical" chess engines which use handcrafted evaluation functions, and minimax-based move selection, combined with alpha-beta pruning, running Komodo MCTS with threads=1 is also non-deterministic. Which in retrospect should have been obvious since MCTS involves running game simulations (playouts) using a random number generator so it would have been <very> surprising if two successive analyses with Komodo MCTS of the same position would yield the same results. Still, it's good to have confirmation even though 2 analysis runs is hardly a statistical sample.

<Stehen sich zwei Einzelbauern im Hinkerabstand gegenüber, so fällt die Rückständigkeit auf den mit der längeren Frontspanne.>

I try a translation: If two isolated pawns stand opposite each other at a limp distance, the backwardness falls on the one with the longer front span.

It's a great book! Kmoch has a way of writing that gets the points across. The only weird thing is he tried to coin a lot of phrases, but it doesn't matter much. The examples he chooses for diagrams are great too. It's one of the first chess books I acquired.

But now we live in a new age! I think we can learn a lot more about chess from a few videos where the masters, or even IMs, GMs, or the world champ himself (!) provide live commentary as they play. That provides the kind of insight into high-level chess thinking that books struggle to get across.

I love <Kingscrusher's> channel because of the way he explains the games. I think Kramnik said that although he can't prove it, he somehow feels it's important, valuable, maybe even necessary for improvement, to be very familiar with the whole history of chess and study games from the best players of each respective era.

With that in mind, <Kingscrusher's> playlist on The Evolution of Chess Theory is perfect! He goes way back to the times of Greco and then Staunton, on to Morphy and then Steinitz, moving forward to the Lasker era and so forth.

I went through the different eras once and tried to emulate the style of play typical for each era in my blitz games. I would play thousands of games just trying to play like Morphy by focusing on quick development and increasing piece activity at all times, with an eye out of tactics. (Supplemented with tactics puzzles).

Then I got into the Steinitz era and that was so interesting, watching him change from a typical 1800s romantic style to the new position style. I then switched my style of play in blitz to Steinitz and resisted my old urges to sac and attack speculatively, focusing on the center, the bishops, and solid play.

My blitz rating went from about 1850-1950 to 2295 (the other day).

Hooper & Whyld, in their Oxford Companion, christen both 4.Qc2 d5 and 4.f3 as Kmoch variations in the Nimzo. I think of the former the name has been used regularly, though probably not anymore in the last decades, and really it's the Noa V. The reason for 4.f3 may be what H&W give: "Some of these appeared in the supplement to the Handbuch (i.e. Bilguer), written by Kmoch and published in 1930."