by IM Larry Kaufman (reprinted with Larry's permission)

(first published in Chess Life May 2005, p.22)

Main Chess Page

Articles Page

Nearly every beginner soon hears that doubled pawns are bad, and some even interpret this to mean that a piece should not be recaptured if it leads to doubled pawns! On the other hand, many grandmasters often seem to play as if the doubling of pawns is a desirable goal. It is the goal of this article to clarify this contradiction and explore the different types of doubled pawns.

Long-time readers of this magazine may recall that in 1999, I ran an article in which the average values of the pieces under differing circumstances were explored using database statistics. I have again made some use of this same method for this article, although software limitations only allow me to confirm a few of the statements made here. Again, I am using a database (now of about 600,000) games in which both players are of at least FIDE Master standard (2300+), and I require that whatever situation is being tested to remain on the board for three full moves. Results are analyzed by performance rating minus actual rating rather than by raw scores, and are expressed as a fraction of a pawn. All opening statistics are based on the “PowerBook 2005” database of about a million games.

The first statement I can confirm is that doubled pawns are indeed on average undesirable, by about 1/8 of a pawn. However this statistic needs to be broken down to be useful. Doubled pawns themselves are really more serious than this generally, but when your pawns are doubled you automatically get an extra half-open file for your rooks (or queen, but the queen can also use diagonals). So it follows logically that the net cost of doubled pawns is much greater in the absence of major pieces. The database shows that with all rooks present, doubled pawns only “cost” about 1/16 of a pawn on average. With one rook each, the cost rises to ¼ pawn, and with no rooks present to 3/8 of a pawn. With the queens present, the cost is again only 1/16 of a pawn; without them it’s a quarter pawn. So the lessons are clear; beware of doubled pawns when major pieces have been exchanged, and beware of exchanging major pieces when you are the one with the doubled pawns.

The next point about doubled pawns is that they arise only by captures. Since players generally capture towards the center when given a choice, the doubling of pawns tends on average to increase central control. However the database shows that the difference in value between pawns on different files is quite small, with the exception of the edge pawns, which are on average worth about 0.85 pawns. This means that a capture by an edge pawn that produces doubled pawns is on average a slightly favorable transaction. With all the major pieces still present, it is a clearly favorable one, and with a rook still on the newly opened file it is generally highly desirable. This explains why there are so many openings in which each player hopes to induce the opponent to exchange pieces on the b3, b6, g3, or g6 squares. But note that if most of the major pieces have been exchanged, even the doubling of pawns by an edge pawn capture is generally undesirable.

Why are doubled pawns bad? There are two main reasons. First of all they are in general a bit weak, because they cannot defend each other and because the front one cannot usually be defended from behind by a rook or queen. This is most apparent in the case of doubled isolated pawns, which are generally considered pretty bad, especially if they are on a half-open file where they are subject to attack by an enemy rook or queen. In general, doubled isolated pawns on a half-open file are worth only slightly more on average than one healthy pawn. Thus, in the Open Sicilian whenever Black can play the sequence …Rxc3 bxc3 Nxe4 he should usually do so, since one pawn plus the severely damaged pawn structure is worth at least the Exchange (par value 1 ¾ pawns).

However, when the doubled isolated pawns are on a closed file, the damage is much less serious, and such damage may be acceptable in return for the bishop pair or similar compensation. Consider the following amazing opening, played last year between two very strong grandmasters, Nigel Short and P. Harikrishna: 1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 Bc5 5. 0-0 d6 6. Nc3 Bg4 7. Na4! Nd7 8. Nxc5 Nxc5 9. Be3 Qf6?!? 10. Bxc6+ bxc6 11. Nxc5 dxc5

.

Black has allowed tripled isolated pawns! Not surprisingly, he lost. I think he felt that after the obvious 9…Ne6 White would have a free bishop pair advantage (Black apparently needs to improve on move 5 or 6), and judged that his drawing chances would be better with the tripled pawns in an ending, especially since he could at least inflict doubled pawns on White.. Grandmaster J. Rowson called this “a questionable decision” (I would put it more strongly), but to me the remarkable thing is that such a strong player as Harikrishna would even consider this choice. It’s not that he was unaware of how bad tripled isolated pawns are; it’s just that he had such respect for the bishop pair that he (wrongly) judged the tripling to be the lesser evil.

The other defect of doubled pawns is that when they are part of a pawn majority, they cannot generally create a passed pawn unless the opponent allows them to be undoubled. So, although a doubled pawn gets full value for square control, the back pawn tends to go to waste when it comes to making a queen. Such a “crippled majority” should be thought of as roughly like losing about 3/8 of a pawn. As the bishop pair is worth nearly half a pawn, the rule of thumb is that the bishop pair more than offsets a crippled majority. This can be seen in pure form in the Ruy Lopez Exchange variation after 1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Bxc6 dxc6 5. d4?! (5. 0-0 f6 6. d4 offers chances for a slight edge based on White’s lead in development) exd4 6. Qxd4 Qxd4 7. Nxd4 Bd7.

Although development is even and it is White’s move, Black has a slim edge in database statistics, consistent with the above statement.

Another form of compensation for a crippled majority is development. Consider the Scotch Opening, as often played by Gary Kasparov in the 90’s: 1. e4 e5 2. Nf3 Nc6 3. d4 exd4 4. Nxd4 Bc5 5. Nxc6 Qf6 6. Qd2 dxc6. Here Black has one tempo (the bishop developed) plus a better placed queen in return for the crippled majority. But it is White’s move, and database statistics show White retaining a slightly larger than normal advantage out of 559 games, clearly proving that a crippled majority is a serious defect. In the Caro-Kann, Black may concede the healthy vs. crippled majority advantage to White without compensation after 1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Nf6 5. Nxf6+ exf6, but only one player over 2600 (V. Korchnoi) has played this line for Black in master chess in the past twenty years, which shows how grandmasters feel about taking on a crippled majority without compensation.

The next category of “bad” doubled pawns is the backward doubled pawn, which I define as a doubled pawn whose neighbors are at least as advanced as the front twin. If the neighbors are more advanced than the front twin, I call it a “doubly backward doubled pawn,” which is nearly as bad as a doubled isolated pawn. Even the normal backward doubled pawn is a more serious liability than a normal backward pawn, because its backwardness generally cannot be remedied by advance due to the obstruction of the front twin. Both cases can be illustrated by the Nimzo-Indian Saemisch variation: 1. d4 Nf6 2. c4 e6 3. Nc3 Bb4 4. a3 Bxc3+ 5. bxc3 c5. Despite White’s bishop pair, the backward doubled pawn on c3 (plus the wasted tempo a3) give Black plenty of compensation, and White’s statistical advantage here is below par. If White were to play 6. d5?, he would have a “doubly backward doubled pawn” which would clearly favor Black despite White’s bishop pair. As evidence of the validity of the backward doubled pawn concept, I note that White’s advantage in the Nimzo (after 3…Bb4 above) is a sub-par 28 points, whereas in the same position with 1. Nf3 replacing 1. d4 it is a normal 34 points, because the doubled pawns arising after a later …Bxc3 are not backward.

Just days after I wrote the above paragraph, I won a nice miniature as Black at the U.S. Amateur Team East over 2500-rated IM Peter Vavrak which illustrates the above theme well. !. d4 Nf6 2. Nf3 e6 3. c4 b6 4. g3 Bb7 5. Bg2 Be7 6. 0-0 0-0 7. d5 Na6 8. Nc3 Bb4! 9. Nh4 Bxc3 10. bxc3 Nc5

(now it’s a battle of bishop pair vs. doubly backward doubled pawn) 11. Qd4?! d6 12. f4?! Re8! 13. e4 exd5 14. exd5 Qd7 (Black is already much better) 15. f5 Ba6 16. Rb1?! Qa4 17. Rb4 Qxa2 18. g4 Qc2 19. g5 Nfe4 20. f6 Qxc3 and White soon resigned after losing his remaining queenside pawns.

Now we come to doubled pawns which do not fall into any of the obviously bad categories (isolated, backward, or crippled majority). Most books generally do not make further distinctions, and simply say something to the effect that the resultant open file may fully compensate for the doubled pawn, which, while true, is too general a statement to be a good guide. Let’s consider the following two openings, the Ruy Lopez (1. e4 e5 2. Nf3 Nc6 3. Bb5) and the Rossolimo Sicilian (1. e4 c5 2. Nf3 Nc6 3. Bb5). In the former case, Black usually replies 3…a6, inviting White to double Black’s pawns (at the cost of the bishop pair), and White usually declines to capture, even though he would save a tempo and have the option to give Black a crippled majority. But in the Rossolimo, 3…a6 is considered a poor move, even though after White captures the knight he has no way to give Black a crippled majority. Not only that, but in the Rossolimo after almost any normal third move for Black (3…e6, 3…d6, or 3…g6) White very often takes the knight voluntarily in grandmaster play, without either gain of tempo or crippling of a majority! How can we reconcile these facts?

The answer seems to be a principle I have never seen in print! The point is that doubled pawns are worse when the back twin has advanced. One reason is that if the pawns are attacked, the twins need to separate to form a triangle (such as …c7, b6, c5) so that only one pawn needs defense from a piece. If the base pawn has advanced, this necessitates advancing the front twin into the enemy half of the board, which is usually not possible for tactical reasons. So the advance of the rear twin often means that two pawns need piece defense instead of only one. In the Ruy Exchange (or in the Berlin Defense to the Ruy, 3…Nf6), after Black recaptures with the “d” pawn, he later often aims for the setup …c7, b6, c5, and a5, which is rather safe, since the only weak pawn on c7 is easily defended by a bishop on d6 or by the king. But in the Rossolimo, after Bxc6 dxc6 has been played, we often get the above formation but with the c7 pawn on c6, when both the c6 and b6 pawns need piece protection. These considerations aren’t just hypothetical. In the Ruy Exchange, White scores a slightly below par +29 rating points, whereas in the Rossolimo, after the analogous 3…a6?! 4. Bxc6 dxc6 White scores a massive +86! Even after the most popular move 3…g6, after 4. Bxc6 dxc6 White scores an impressive +48 points. The same results are seen in the mirror image openings where White gives up the bishop pair to double the “f” pawns. After 1. d4 Nf6 2. Bg5 d5 3. Bxf6 exf6 White has hardly any edge (+8), because he has little compensation besides the move for the bishop pair. But after 1. d4 f5 2. Bg5 Nf6 3. Bxf6 exf6 White has a huge +63 point edge, because the rear twin has advanced. In general, I would say that if the rear twin has advanced then one usually needs some real compensation for the doubled pawn.

One more consideration is that when pawns are doubled, sometimes it increases the number of pawn islands by one, which is bad. For example, in the Scotch opening, after 1. e4 e5 2. Nf3 Nc6 3. d4 exd4 4. Nxd4 Nf6 5. Nxc6 bxc6 6. e5 Qe7 7. Qe2 Nd5

White retains a normal opening advantage, despite Black’s lead in development and the fact that the doubled pawns don’t fall into any of the above categories. Probably the main reason is that Black now has three pawn islands versus White’s two. This shows up in variations where Black plays a later …d6 and White captures, undoubling the pawns but still leaving White with the advantage of one less pawn island.

When judging the wisdom of allowing doubled pawns, one must also consider the importance of the resultant half-open file. In general, when the kings are castled on the same side, a half-open file on the other side is only a modest plus, and if the half-open file is near your own king it may well be a disadvantage. But if the kings are castled on opposite sides and the half-open file bears down on the enemy king, it’s a big plus and can easily offset even doubled isolated pawns.

So in conclusion, most doubled pawns (other than those arising from a capture by a rook’s pawn) are a liability, but to widely varying degrees. If they don’t fall into any of the above bad categories, and if the resultant half-open file can be occupied by a rook, then and only then can it be said that no other compensation is required for the doubling.

Larry's correspondence is in Black; mine is in Green.

Larry,

Another great article. Is this one available on the web?

No.

If not, I offer to host it, too.

That would be okay with me, though I think we should wait a few weeks to make sure that USCF would not lose any magazine sales because of it.

I have a suggestion for a further study on this subject. It has long been my contention that if a doubled pawn in the opening gives to the ‘doubled pawn’ player the only semi-open file(s), then it is more beneficial for that side than if there are, say, other open files on the board that the rooks and queen can use. For example, in the Exchange Ruy Lopez, Black’s temporary use of the semi-open d-file is more beneficial to him until other lines are open where rooks can contend. Or in the four knights game if Black captures Bc5xBe3 and white plays fxe3, then the f-file is a big advantage for White until Black can open up another file, say the d-file. This seems true, but can you verify it by your database method?

I could easily check this by testing whether the harm of doubled pawns varies based on the number of pawns on the board. This would match what you ask for assuming that you don't distinguish between other open files and other semi-open files. However there is one big problem. Doubled pawns tend to arise due to BxN. The side with doubled pawns tends to have the bishop pair (or bishop vs. knight), which benefits from opening the game. So the conclusion would probably be the opposite of what you expect, but for the wrong reason. I could avoid this problem by specifying that the number of bishops must be the same on both sides. Let me know which way you think I should do the test.

Regards,

Dan Heisman

Regards, Larry

---

Larry,

Do you have an electronic version you can e-mail me to host? Just wait until next month’s CL is out, and if you can send it then, I will host it.

As for the BxN B-pair issue for the “only semi-open files”, of course you are correct. So we would either have to adjust for the B-pair or just check it for situations where the material is otherwise identical. I assume the latter would be more accurate since it eliminates the side issue of how effective the B-pair is with 8 pawns initially on the board, so I would lean toward that (your suggestion where the bishops number the same is essentially this), assuming it is equally easy for you to do.

Thanks for offering to do this – I am pretty sure I am on to something, because it would be almost the only way to explain why players with the only semi-open file get such good positions where otherwise the doubled pawn would offset any extra central control. Yet this same advantage is vanishing if there are also open files (which also open lines for the opponent to attack any vulnerable doubled pawn).

Thanks again,

Dan H

-----

OK, I ran the test. Subject to some caveats, the results confirmed your theory. I required identical pieces on each side (no material advantage or B/N imbalance), six ply continuous presence, 2300 min. rating. Expressed as a fraction of a pawn, the "cost" of doubled pawns can be seen below.

Number of pawns per side               Cost

8                                     .19

7                                     .11

6                                     .25

5                                     .20

4                                     .29

3                                     .33

2                                     .47

So, although there is some bouncing around, the cost does rise rather significantly with less pawns on the board. Of course this could partly be due to other issues, such as the fact that with less pawns on the board, there are also usually less major pieces remaining. I could test for that by specifying that each side must have queen and only one rook (for example); perhaps I'll do that when I have more time. The results with many pawns on the board are heavily influenced by theoretical lines, whereas with 5 or less pawns the positions are usually not ones that have occurred in other games.

The results suggest that the "true" cost of a doubled pawn, if we could separate out the benefit of the semi-open file, is rather large, perhaps 0.4 pawns (averaging the last two entries). They also suggest that the average cost, even after allowing for the benefit of the semi-open file, is more than the 1/8 pawn I mentioned in my article, perhaps closer to 1/4 pawn. Apparently, normalizing for the number of pawns makes the effect more pronounced.