FIDE vs. US Chess Swiss Pairing Rules

Running Chess Tournaments
1

For those forum readers who are suffering severe bouts of insomnia, I bring good news. I have prepared two papers discussing the difference between US Chess Swiss pairing rules and the FIDE Dutch system of Swiss pairings. The first is a five-page discussion of the major differences between the rules. The second is more complex, weighing in at fifteen pages, and is a more thoroughly analyzed example of a five round tournament based on the example in the FIDE 2014 Arbiter’s Manual.

Both of these are still in review by the FIDE events committee, so these should be considered preliminary drafts. However, I don’t expect anyone to hurt himself badly with these drafts. :slight_smile:

Comments and feedback are welcome.

2

Nice work. Could you perhaps detail the differences between USCF and FIDE in re the 80- and 200-point limits? i.e. I believe such limits do not exist in FIDE pairing rules.

Those limits always seemed artificial to me, though I guess we need a standard cutoff somewhere. They might work well for larger Swisses, but for small club/local events with players rated all over the place I sometimes extended those limits a bit, to equalize colors. (Using 80/200 as a written-in-stone in such events can lead to problems in the late rounds, I found.)

Funny how the same concept works for high-level FIDE-rated Swisses and club tournaments with class players, but works less well in the middle-ground between those extremes. Go figure.

3

That’s correct. Color is all-important in FIDE pairing rules. If it takes a 450 point swap to fix an alternation problem, you do the 450 point swap.

4

Eric,

The FIDE rules don’t use ratings (or rating differences) at all. They just use rank, within the score group.

The rank within the score group is, of course, determined by ratings, but a 1-point difference is the same as 1000 points as far as FIDE is concerned. If switching players 14 and 15 corrects colors on both boards, then the FIDE algorithm will happy to do so, no matter how big the rating difference is.

Bill Smythe

5

Right. Thanks to the gentlemen from Illinois for the info. I thought as much, and when I directed rated events at a weeknight club I often did the same.

We often had 10-14 players in a 4-round Swiss, where lots of players took a half-point bye at some point, with ratings widely spread out from 1200 to 2100 or so. We also had the “clump effect.” i.e. a few 1900-2000 players, then a drop to several more at 1550-1700 then another group at 1300-1400, with one or two stragglers in between the clumps.

I tried to avoid being forced to give anyone 3-1 colors for the event, especially if it meant three Blacks. (and we learned that an odd number of rounds is better for color fairness, but there are four Mondays in most months, and players wanted tournaments contained within a month—after debating other structures with vigor.)

To make sure of that, if I needed to plug in a 1450 player in place of a 1720 player for color equalization, I did it without hesitation. We had small fields, small prizes, players of widely varied ratings, nearly all of whom were sensitive to getting an “unfair” number of Blacks. Even the ones who were easy to deal with on most things would grumble about that.

The 80/200 standard could be a case where what works best for a medium-to-large Swiss, in terms of entries and prizes, was ordained the main rule and cemented in the rulebook.

I found it better to do it the “FIDE” way, designed for tournaments like Gibralter, let’s say: serious chess with strong players and only a few hundred points from top to bottom of the field, apart from a few at the ‘very’ bottom.

Strange symmetry there.

6

In the “basic differences” document, I did write:

Should I have been more explicit here?

7

My other piece of advice, in a small tournament (fewer than about 20 players), would be to make no transpositions at all just to alternate colors. Transpose only to equalize, not to alternate. (This applies mostly to round 3.)

With pairing software using USCF pairings, change the alternation limit from 80 to 0 but keep the equalization limit at 200.

If the colors work too well, you are effectively dividing the players into two camps, those who started with white and those who started with black, and you are making mostly inter-camp pairings. In a small tournament, you may quickly run out of decent inter-camp pairings in the later rounds. If you have some intra-camp pairings (bad colors) in round 3, you’ll find it much easier to make colors work in round 4.

It’s better to have bad colors in round 3 (alternation) than in round 4 (equalization).

Bill Smythe

8

Can I post these links on my 80/20 site?

9

I’m not sure you can in a document this size.

10

Yes, upon further review I think that is right. I focused on the limits for equalization and alternation since that is an issue I dealt with regularly back when I directed small tournaments out in the far country.

Nice work, Ken.

11

I’ll be happy to have you post them, but I’ll ask you to wait a bit. I’d like to make sure the FIDE events committee has time to look them over, and between the two, there are twenty pages of less than scintillating text. :slight_smile:

12

I’ve made a few minor changes to this document based on feedback so far. I added an explicit statement that there is no FIDE equivalent of the US Chess 80/200 rules. I also added a statement that FIDE pairings are deterministic, while US Chess pairings are non-deterministic. (Technically, this is not a difference in the rules but rather in the way the FIDE pairing regulations are written.)

The link posted in the original post now points to the revised version.

13

First of all, a huge thanks to Ken for posting the two papers.

The FIDE Dutch system has some major advantages over the USCF system. By eliminating the 80- and 200-point rules, and going simply by rating order within the score group, the FIDE system has made pairings more nearly deterministic (or more nearly algorithmic, if you will) than the USCF system.

There was a thread a couple of years ago in which WinTD was forced (unavoidably) to make a 300-point transposition in order to prevent two players from playing each other a second time. But then, having made this transposition, WinTD refused to make an additional 1-point transposition to improve color equalization, presumably because this player had already been transposed beyond the 200-point limit for color equalization.

This is the kind of anomaly that can occur whenever there is an attempt to convert non-algorithmic rules into an algorithmic computer routine. And it’s why I’d like to see USCF move toward something resembling the FIDE Dutch system.

But a few peculiarities remain in the FIDE system. In Ken’s longer paper the following example is presented: Players {1,2,3} are to be paired against players {5,6,7}. But the raw pairings (I dislike the term “natural pairings”) of 1-5, 2-6, 3-7 would produce bad colors on all three boards. So which transposition is better, {5,7,6} or {6,5,7}? Either would reduce the number of bad colors from 3 to 1. FIDE prefers {5,7,6} because it pushes the transposition down to a lower board. But in doing so, this transposition essentially pushes the bad color up to a higher board. The color conflict is now on board 1 instead of board 3.

For a FIDE system that supposedly places greater weight on colors than the USCF system does, it seems a little odd that FIDE pushes transpositions down and bad colors up, rather than the other way around.

Also, the FIDE system (and, for that matter, the USCF system too) divides color due-ness into only four categories:

  1. No color is due because the player has played no games yet, e.g. byes or forfeits in all previous round(s).
  2. The player is due W or B in order to alternate.
  3. The player is due W or B in order to equalize.
  4. The player is very strongly due W or B, e.g. to avoid 3 of the same color in a row or to avoid going out of balance by more than 2.

(Ken has assured us that the distinction between b. and c. will be added to the FIDE version in 2017.)

It seems to me there should be additional levels of due-ness, along the lines of USCF rule 29E4. For example, going into round 5, a player with WBWB should be considered more strongly due white than a player with BWWB.

I would like to see the FIDE rules (and USCF, too) adopt an algorithm that would work as follows (for example):

  1. If there are more players due white than due black, then players due no color would be considered to be due black.
  2. If 1. is not sufficient to create equality between due-whites and due-blacks, then players due white more weakly than others would be considered to be due black.

For example, going into round 2, if there are six players with color histories
W (due black)
B (due white)
x (due no color)
B (due white)
W (due black)
B (due white)
then the player with x (no color history) would be treated as due black.

Or, if going into round 3 we have
WB (due white)
BW (due black)
BW (due black)
WB (due white)
WB (due white)
xB (due white)
then one of the players with WB (perhaps the lowest?) would be treated as due black, rather than the player with xB who is more strongly due white than the others.

Or, if going into round 5 we have
WBWB (due white)
BWBW (due black)
BWBW (due black)
BWWB (due white)
WBWB (due white)
WBWB (due white)
then the player with BWWB would be treated as due black, as he is less strongly due white than the others.

Bill Smythe

14

The deterministic nature of FIDE pairings is the best feature of them by far, in my opinion. No arguments, easy to verify (there’s a publicly available, free, java program to confirm pairings), etc.

It’s an unpopular opinion here, but I feel this is yet another example of how FIDE has better rules than the USCF. Why we don’t save oodles of time (and money) and just adopt FIDE rules instead of maintaining our own cumbersome, outdated, and sometimes contradictory set of rules is beyond me.

Next year, I’m seriously considering getting someone to propose an ADM to do just that. Who’s with me?

-Matt Phelps

15

“You might very well think that; I couldn’t possibly comment.” – Francis Urquhart, House of Cards

16

Has anyone tried using the FIDE pairing rules for, say, round two of one of the really big sections at Nationals? Trying to apply the “deterministic” rules to a 250 player score group with teammate and state pairing preferences might take quite a while. (125! is a bit bigger than 10^200).

17

Actually, I think it would be rather straightforward. There are simple optimizations that can be made to the programming to avoid trying stupidly useless transpositions that will not reach the needed goal.

In the course of pairing a score bracket, the arbiter determines P1, the number of pairings that must be made in the score bracket (if possible) and X1, the minimum number of pairings that will not satisfy both players’ color preference. (That’s simply based on more players needing white or black, and not being able to satisfy everybody.) Then, we know that transpositions to improve color allocation have to be made on the bottommost boards. It should not be a difficult matter to count up how many color mismatches there are from the bottom, see how far up in the S2 set the transpositions have to go, and come up with an intelligent starting point. In fact, by not having to pay attention to rating differences, I would think the process would be simpler than under US Chess rules, where the director would keep looking for better transpositions involving smaller rating differences. In the FIDE Dutch system, as soon as you have found pairings that satisfy all the “quality of pairings” criteria, you are done with that score group (subject to backtracking, of course, if a lower score bracket can not be paired).

(As an aside, there is a really easy way to come up with an interesting example of how the backtracking works. Pair a hypothetical sixteen player event where black always wins. In the third round, the four players in the two point score group will have an absolute color preference for white, while the four in the zero point score group will have an absolute color preference for black. There is no choice but to downfloat the four players in the two point score bracket to be paired against the four players in the one point score bracket due black, and then to downfloat all four players in the remainder score bracket (who are due black) to be paired against the players in the zero point score bracket.)

Similarly, although the FIDE Dutch system pairing algorithm appears to be iterative (the funky loop termination in C.10, “Lowering requirements”), I’m not persuaded this is necessary. In the worst case, one ends up examining every possible transposition and interchange (which should really only be necessary in small score brackets), measuring the quality of the pairings produced as an 6-tuple (number of color allocation problems, number of color equalization problems in even rounds, number of downfloaters who had a downfloat the previous round, number of downfloaters who had a downfloat two rounds ago, number of upfloaters who had an upfloat the previous round, number of upfloaters who had an upfloat two rounds ago). For a perfect pairing, that 6-tuple will be (X1, Z1, 0, 0, 0, 0). As one examines transpositions and interchanges in turn, one should be able to apply a lexicographical ordering to the six-tuple to determine which pairing is the best that has been seen so far. The actual quality of the pairings is determined by that 6-tuple, and the value of the 6-tuple does not depend on the state of the constraints in C.10. So, one pass over the possible transpositions and interchanges should be sufficient to determine how it is necessary to lower the requirements as described in C.10 to produce pairings. The only part of this analysis that I’m not quite sure of is how it interacts with any need to backtrack.

(While it may not seem believable, I simplified the discussion in the previous paragraph a wee bit by ignoring C.10.f and C.10.g, but I would think the same logic applies.)

I will stipulate that the teammate and state pairing restrictions do add complexity to the problem. But it seems it might be a reasonable test to take one of those sections, prepare a TRF file corresponding to the results of the first round, add the “no pair” constraints, and see how JaVaFo does producing pairings for the second round.

18

There is nothing stopping organizers from using FIDE rules for their tournaments already. It’s just a matter of announcing their use. I don’t agree that FIDE (Dutch) pairing rules are better than the US Chess system. As I see it, the Dutch system puts too much emphasis on color vs. rating differences.

19

Zero chance Matt’s hypothetical ADM passes. 0.00001% chance it gets as far as being referred to rules committee. Trying to figure out how to mesh FIDE rules with the remaining USCF rules not replaced by FIDE rules seems like a monumental task.

20

I believe I have more sympathy for Mr. Phelps’s position than most Delegates, as I have gone on record more than once favoring some FIDE rules in cases where they conflict with USCF rules.

That said, I suspect any ADM even approximating the one he proposes will be DOA. I further suspect Mr. Mulford has overestimated the likelihood of referral by a factor roughly equivalent to the second Skewes number.