Is e4 significantly better than d4?

[OneTrickPony]

Leela sucks at tactics too much for this question. It wants to play the Catalan because it thinks Vienna is too good for black but it wants to play a losing line there (taking the pawn on e4 and then c3). Once its search is good enough to understand black can't take two pawns in Vienna then 1.d4 will be back as at least a close contender or maybe even surpassing the Berlin positions it wants to play after 1.e4
If you look at the evals in Bf4 QGD mainlines they are higher than in positions arising from the Berlin.

[Laskos]

I don't think you are ever going to be able to decide this question with a root-level search.

--JOn

That's possible.

[Laskos]

Leela sucks at tactics too much for this question. It wants to play the Catalan because it thinks Vienna is too good for black but it wants to play a losing line there (taking the pawn on e4 and then c3). Once its search is good enough to understand black can't take two pawns in Vienna then 1.d4 will be back as at least a close contender or maybe even surpassing the Berlin positions it wants to play after 1.e4
If you look at the evals in Bf4 QGD mainlines they are higher than in positions arising from the Berlin.

I am not sure tactics unseen to Lc0 enters into this. Then, even if it so, Lc0 is still stronger than humans in tactics, and this empiric outcome might be useful, not true. In fact the truth is that Chess is either a draw or a win/loss, if abandoning empiricism.

[Laskos]

Leela sucks at tactics too much for this question. It wants to play the Catalan because it thinks Vienna is too good for black but it wants to play a losing line there (taking the pawn on e4 and then c3). Once its search is good enough to understand black can't take two pawns in Vienna then 1.d4 will be back as at least a close contender or maybe even surpassing the Berlin positions it wants to play after 1.e4
If you look at the evals in Bf4 QGD mainlines they are higher than in positions arising from the Berlin.

I am not sure tactics unseen to Lc0 enters into this. Then, even if it so, Lc0 is still stronger than humans in tactics, and this empiric outcome might be useful, not true. In fact the truth is that Chess is either a draw or a win/loss, if abandoning empiricism.

Thinking a bit, positionally 50 million SF nodes at the leaves, not bad. How many humans have that OTB? Tactics is an issue, not sure where to place it, humans are worse. Again, more in terms of empiric "usefulness".

[Father]

This is simply fascinating! I would like to know about the following responses to e2e4 if they are a mistake or if they are correct: 1 ... e7e6; 1 ... c7c5; 1 ... c7c6; 1 ... g8f6; 1 ... d7d5; 1 ... a7a6; 1 ... b8c6; 1 ... g7g6; 1 ... b7b6; 1 ... d7d6. 1 ... f7f5. Thanks in advance for this wonderful information, which is presented as in the window to the future.

[Laskos]

This is simply fascinating! I would like to know about the following responses to e2e4 if they are a mistake or if they are correct: 1 ... e7e6; 1 ... c7c5; 1 ... c7c6; 1 ... g8f6; 1 ... d7d5; 1 ... a7a6; 1 ... b8c6; 1 ... g7g6; 1 ... b7b6; 1 ... d7d6. 1 ... f7f5. Thanks in advance for this wonderful information, which is presented as in the window to the future.

Not sure it's fascinating, but it's interesting, your proposals too. Maybe the first thing is to check the result with a strong, weakly related net (at probably very high node count, my net was simply the strongest).

[Ovyron]

My success rate over the past 10 years is better with 1.d4

I'm in the same boat. We're talking about some 50 days/game with 50 days added every 10 moves time control level. My wins with 1.d4 at least double my wins with 1.e4. What seems to be happening is that 1.d4 allows more freedom to black about what to do, while 1.e4 variations seem more forcing. This might seem counter-intuitive but people are better at playing forced moves, when they're not forced, they don't know what to play, so their mistakes accumulate until the position blows on their face.

Also:

Leela sucks at tactics too much for this question. It wants to play the Catalan because it thinks Vienna is too good for black but it wants to play a losing line there (taking the pawn on e4 and then c3).

Shouldn't this put a nail in the coffin for these discussions? If you follow Leela's suggestions in the opening it leads to a losing line, so how can Leela's opinion be used at all to decide what move is best? What I can comment on is that Leela's lines from the opening position are laughable, g3 is very drawish on d4 variations, and Bxc6 is very drawish on e4 variations. Honestly these lines don't seem any better than what Rybka 4 was showing in hardware from 9 years ago...

[lkaufman]

Leela sucks at tactics too much for this question. It wants to play the Catalan because it thinks Vienna is too good for black but it wants to play a losing line there (taking the pawn on e4 and then c3). Once its search is good enough to understand black can't take two pawns in Vienna then 1.d4 will be back as at least a close contender or maybe even surpassing the Berlin positions it wants to play after 1.e4
If you look at the evals in Bf4 QGD mainlines they are higher than in positions arising from the Berlin.

This is actually a pretty good summary of the situation. I would like to say I agree totally with Kai, after all my soon-to-be published opening book recommends 1.e4 and I agree that the lines Kai gives are among the best, but after doing some interactive analysis with the same network Kai used, on my 2080 GPU, I have to agree with the above. The main line Berlin as given above up to move 9 gives about 54.5% after a couple million nodes from that position, as does 5.Re1 as well. This does seem to be the best Black can achieve after 1.e4 (top players consider the Marshall about equally good but this network doesn't like it). After 1.d4 if White goes for the Catalan the line given is perhaps the best, and is my personal choice as Black, and White only gets about 53.6%. The Nimzo isn't any better for White. But 3.Nf3 d5 4.Nc3 is probably best, and then as noted the Vienna has a tactical problem, so the Ragozin (4...Bb4) looks best. Then 5.cxd5 exd5 and lately 6.Bf4 has caught on, and is the choice of this network, with a 54.6% score. So I would say that according to this network, with reasonable interactive analysis, 1.e4 and 1.d4 are virtually tied for being the best move. I still believe 1.e4 is the better move, but I can't prove it with the current hardware/software.

[Ovyron]

I still believe 1.e4 is the better move, but I can't prove it with the current hardware/software.

What do you think about the Marshall Counter-Attack?

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 O-O 8. c3 d5

[d]r1bq1rk1/2p1bppp/p1n2n2/1p1pp3/4P3/1BP2N2/PP1P1PPP/RNBQR1K1 w - -

Around here THIS is the line that kills 1.e4. At least this is the thing that tips the balance clearly in favor of 1.d4.

I've been analyzing extensively these variations since 2012. What's funny is that the variations I analyzed back then have held over the years, and still hold up today, no improvement on software had shaken them up, so this is an anomaly (I haven't seen anything like this in any other line.)

My conclusion:

If black defends correctly then this position is 100% drawn.

This is a reason I don't play this line myself, because I want to try to win as black, and black has no winning chances. But can someone find a line where white can win? My next claim is:

Black can play this position correctly with ease.

At 12 +2 time control against the strongest Stockfish derivatives and Leelas in GPUs nobody could touch it.

The position is 0.00

Engines don't understand this, they give +0.50 scores at minimum, you can even play sub-optimally and let them reach 1.00 and arrive at endgames that are trivially drawn.

So I'd put two exclamation marks in all the black moves, and engines that want to give any meaningful scores to the opening position would need to know that this is black's best defense against the Spanish, and that any white attack would need a different white move, and what remains doesn't seem better than 1.d4.

[lkaufman]

I still believe 1.e4 is the better move, but I can't prove it with the current hardware/software.

What do you think about the Marshall Counter-Attack?

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 O-O 8. c3 d5

[d]r1bq1rk1/2p1bppp/p1n2n2/1p1pp3/4P3/1BP2N2/PP1P1PPP/RNBQR1K1 w - -

Around here THIS is the line that kills 1.e4. At least this is the thing that tips the balance clearly in favor of 1.d4.

I've been analyzing extensively these variations since 2012. What's funny is that the variations I analyzed back then have held over the years, and still hold up today, no improvement on software had shaken them up, so this is an anomaly (I haven't seen anything like this in any other line.)

My conclusion:

If black defends correctly then this position is 100% drawn.

This is a reason I don't play this line myself, because I want to try to win as black, and black has no winning chances. But can someone find a line where white can win? My next claim is:

Black can play this position correctly with ease.

At 12 +2 time control against the strongest Stockfish derivatives and Leelas in GPUs nobody could touch it.

The position is 0.00

Engines don't understand this, they give +0.50 scores at minimum, you can even play sub-optimally and let them reach 1.00 and arrive at endgames that are trivially drawn.

So I'd put two exclamation marks in all the black moves, and engines that want to give any meaningful scores to the opening position would need to know that this is black's best defense against the Spanish, and that any white attack would need a different white move, and what remains doesn't seem better than 1.d4.

Indeed, I recommend the Marshall in my new book, so I'm certainly not going to argue with you! White should avoid it by either 8.a4 or by 6.d3, or simply by playing the Italian (I give the latter two options for White in my new book). Almost all top players choose one of these three options. They all give White a small edge, maybe 54% or so. Whether 1.d4 achieves more against best defense is not at all clear.