Website founded by
Milan Velimirović
in 2006

5:31 UTC
ISC 2021



Remember me

Forgot your
Click here!
to create your account if you don't already have one.

Rating lists


MatPlus.Net Forum General Chess960 problems?
You can only view this page!
Page: [Previous] [Next] 1 2 3
(1) Posted by Geir Sune Tallaksen Østmoe [Monday, Apr 16, 2018 20:32]

Chess960 problems?

Does anyone know if there has been any Chess960 problems?

Obviously, in most positions, there is no difference between Chess960 and regular chess, but retro analysis would be different in Chess960 because of the increased number of legal positions, and the castling rules are different. You can even have situations where it is clear that at most one side is allowed to castle, but it is not clear which side.

Simple example:

(= 4+4 )


1.Rf1? is met by 1...0-0! So White plays 1.0-0! instead, proving that Black is not allowed to castle.
(Read Only)pid=16476
(2) Posted by Joost de Heer [Tuesday, Apr 17, 2018 06:35]

You can search for K='Chess 960' on PDB
(Read Only)pid=16477
(3) Posted by Hauke Reddmann [Tuesday, Apr 17, 2018 09:54]

Ha, good that you reminded me, I just composed one and would have
forgotten it. Since it's below par in construction, I can as well
UL it here as an original.

(= 12+5 )

1.OOO? Sa3! (2.Sc3+ Ka1), Theme H of the White Line Combinations
1.OO! Zugzwang, Theme Shitty Excuse of a Key :-)

(Read Only)pid=16480
(4) Posted by Per Olin [Tuesday, Apr 17, 2018 10:18]; edited by Per Olin [18-04-18]

The search in PDB gave 21 problems of different types. My personal view is that the best area for Chess960 composing is proof games. In a standard proof game the solver has a given initial position, which helps the solving. If we have e.g. wKd1 and wDe1 in the diagram position, in a standard proof game it gives away part of the solution: the pieces have swapped places. In a Chess960 proof game this does not have to be the case. This, I think, is much of the spirit of chess problems; leave as much as possible for the solver to find out.

The twomover starting this thread is an elegant way of showing the specialities of Chess960 castling. If the pawn e3 (needed for try and solution!) is removed and added are black bishop b8 and black pawns b7, b6, c7, then Black has a last move. This must be a Chess960 position, therefore could even be thought that the mention Chess960 below the diagram can be left out! Leave as much as possible for the solver to find out (it might also be said that this would be a joke problem). There are different legalities: standard legality and Chess960 legality. With a black pawn on b7, the wSc6 is to be moved to e.g. d5 and the white king can not be on b1, but well on c1.

One type probably not yet seen is asking for the Chess960 initial position for a certain diagram position. The original below should not be very difficult to solve; perhaps somebody can make a more complicated one. Solving by trial and error is possible, but there are some conclusions that can be made to make the solving easier. Solution tomorrow Wednesday April 18.

Per Olin
(= 16+16 )

Position after 18.0 moves.
Chess960 initial position?


Both parties have made 8 pawn moves and 10 moves with all the other pieces, i.e. on both sides two of the officers have made two moves or one has made three moves (castling counts as a king's move). Finding out which pieces have moved once and which pieces have moved more than once helps in determining the initial positions of all pieces.

- In the diagram position there are three knights on dark squares and one on light squares; this means that a knight has moved twice (or three knights moved twice or same knight three times).
- The wLg2 and bLc7 come from the same file and consequently one of these bishops (or both of them) has (have) made two moves.
- If there were no castling(s) made, then the kings would need five moves, which taking into account the above, would be too many. At least one castling has taken place and thus at least one of the kings has made two moves.
- There are two possibilities for the fourth and last piece to have moved twice: a) the bishops wLf2 and bLd7 have started on e1/e8 and then bTe7 has moved twice or b) the bishops wLf2 and bLd7 have not started on e1/e8 and one of them has moved twice. In the latter alternative, e.g. initial positions DTLSKLST and DLSSTKLT, White can reach his diagram position of the officers in 10 moves, but Black can not. Alternative b) is impossible and consequently the fourth piece to have moved twice is the rook on e7.

Summing up: the pieces to move twice are 1) a knight 2) wLg2 or bLc7 3) one of the kings and 4) bTe7.

Conclusions: - wLf2 and bLd7 start on e1/e8 - the uniting file for two knights is the c-file with knights starting on c1/c8 - the previous means that either wSb2 or bSg7 has made two moves by starting on d- or e-file; as the e-file is occupied by bishops, then the knights start on d1/d8 - the black pieces making two moves are bTe7 and bSg7; the white pieces making two moves are wKc2 and wLg2; one of the moves by wK is c-side castling - the previous excludes bLc7 from making two moves meaning that it has started on b8 or d8 and from this follows that wLg2 has started either on b1 or d1; this can not be d1 as it is reserved for a wS and therefore there is a wL on b1 - with bishops on b1/b8 the queens must start on a1/a8 - the black king moves from its initial position g8 to h7 - a black rook is on h8 and, as bTf7 moves only once, the other rook is on f8.

The sought Chess960 initial position is DLSSLTKT
(Read Only)pid=16481
(5) Posted by Valery Liskovets [Wednesday, Apr 18, 2018 12:02]

A fine idea, Geir Sune! Definitely promising in significantly simpler implementing various familiar retrostrategic ideas (and maybe in implementing new ones?). Even your example is much simpler than the simplest known classical chess problem with incompatible w and b castlings by Lapierre (P0001918; 14 pieces). But the former can be reduced further till a miniature. Here is one, even aristocratic:

(= 4+3 )
#2 Chess960

1.Rxg4? O-O-O! [Rd8]
1.O-O! [Kg1,Rf1] O-O-O??
(1...Rxb7(Se5)/Sf6 2.Rf8#/Rc1#.)
(Read Only)pid=16489
(6) Posted by Joaquim Crusats [Wednesday, Apr 18, 2018 20:02]

Several logical Chess960 retractor problems have been published.

One of them, based on "castling dictation" is:

Andrey Frolkin
Joaquim Crusats
Phenix 2014
(= 12+13 )
Chess960, Proca retractor -3 & #1

Both balances are closed:
White: 12 + bxc + cxb + dxc + hxg = 16
Black: 13 + axb>b8 + exf + gxh =16

Main plan: -1.Se5-d7? threatens -2.Sg6-e5 & 1.Rxh8#. Black can only hope that uncastling might help.
-1…0-0 (Ke8, Rf8) -2.Sc4-e5 & 1.Sd6#
-1…0-0 (Ke8, Rg8)? – no mate, but the retraction is illegal. It means that in the initial position White had his king on e1 and one of his rooks on g1. In view of the wPb2 and d2, there is no place for a white dark-squared bishop.
but -1…0-0 (Kf8, Rg8)! refutes: -2.Sc4-e5? ~ -3.Sd6-c4 & 1.Sd7#, but -2…Bc8-b7!

Solution: White has to deprive Black of the latter castling, by means of a foreplan:
-1.0-0 (Kf1, Rg1)? is useless for White (the counterpart of the refutation).
-1.0-0 (Ke1, Rg1)?? is both useless and illegal (see above – no place for the white dark-squared bishop
in the initial position).
-1.0-0 (Ke1, Rf1)! and the main plan works: -1…any (uncastling is still illegal) -2.Se5-d7! (-3.Sg6-e5
& 1.Rxh8#) -2…0-0 (Ke8, Rf8) the refutation is now illegal -3.Sc4-e5 & 1.Sd6#

Note that other castlings by White are illegal because in the starting position h1 was occupied by a white
bishop (bPg7, bBh8).
(Read Only)pid=16491
(7) Posted by Per Olin [Thursday, Apr 19, 2018 12:44]

The standard of this thread will now be lowered, therfore I don't put my name on the top of the diagram.

(= 1+7 )

h#2 1/3 (two and one third) Chess960

b) bKc3->f3

The chess rules recommend concerning Chess960 castling that, 'when castling on a physical board with a human player, it is recommended that the king be moved outside the playing surface next to his final position, the rook then be moved from its starting position to its final position, and then the king be placed on his final square'. A real chess problem does not have a human player, but this joke problem has.

Solution: The castling consists of three parts, all representing 1/3 of a halfmove, i.e. two of these parts count for one third.
a) 0. - Td1, Kc1 1.Kb3 Kb1 2.f3 Txd3#
b) 0. - Tf1, Kg1 1.Ke3 Kg2 2.Sd2 Te1#
(Read Only)pid=16500
(8) Posted by Andrew Buchanan [Friday, Apr 20, 2018 08:30]; edited by Andrew Buchanan [18-04-23]

(= 6+2 )


Being me, of course I had to explore how the Dead Position rule engages with Chess960 (although some aficionados prefer Chess959, I read). Clearly, we can use DP to prove particular start squares for R&K. For example, a regular problem like P1281139 in PDB could be directly stipulated as Chess960. But actually, we can go further. In P1281139, the squares between wK & wR must all be empty, hence the castling move must be checkmate itself, to avoid a cook if wK simply retreats to d1. With Chess960, wR can directly block wK's retreat. This allows longer forward play, and lighter material. A direct flip from fake castling position to real castling position is also shown here.

Solution: White cannot deliver mate this move and cannot give Black a move, unless White is able to castle (kingside). So the position is dead except for possible castling. Black's prior move was 0. ... a3-a2 and this was forced. So if the current position is dead, so was the prior one and the game would have already ended. Therefore we can deduce unconditionally that wKc1 & wRd1 began on their current squares *and* have never moved.

Hence 1. 0-0+ Kb2 2. Sxa2 (zz) Ka3 3. Rf3 (zz) Kb2 4. c4+ Ka3/Ka1/Kb1 5. Bd1/Rf1/Rf1#
(Read Only)pid=16518
(9) Posted by François Labelle [Friday, Apr 20, 2018 20:51]

Per Olin,

It's possible to computer-test the original you posted on Apr 17 with Jacobi v0.4.3. There are two separate problems to solve (using German notation since you like it):

stip dia 18.0 forsyth 8/sdllttsk/bbbbbbbb/8/8/BBBBBBBB/DSKTSLLT/8
cond Chess960

test not dia forsyth 56/DLSSLTKT AddPieces
stip dia 18.0 forsyth 8/sdllttsk/bbbbbbbb/8/8/BBBBBBBB/DSKTSLLT/8
cond Chess960

The first one is solving your problem as a PG. Jacobi takes 9 seconds and then pauses because it found at least 100 solutions, proving that the initial position DLSSLTKT is possible. (Unpausing is pointless as there are too many solutions.)

The second one searches for a solution excluding the initial position DLSSLTKT. Jacobi takes 0.160s and returns no solutions, proving that the problem is C+.

This could help somebody compose a more complicated problem of the same type.
(Read Only)pid=16528
(10) Posted by Siegfried Hornecker [Friday, Apr 20, 2018 23:09]

As the standard position is one of the possible Chess960 positions, the stipulation "Not RSBQKBSR starting position" should be added, I agree.

What does Per Olin think?
(Read Only)pid=16530
(11) Posted by Per Olin [Saturday, Apr 21, 2018 09:58]

Post nr 9: After having tested a couple of Chess960 proof games with Jakobi, I was fairly sure that it will find the initial position fast. Before publishing the problem in post 4 I ran the first mentioned test, which verifies that the intended solution functions. This is by no means a full test as there might be other Chess960 initial positions that can provide a solution. Thanks to the additional second test, newly designed I suppose, the problem is now C+.

With Jakobi we can now test Chess960 proof games (what was the Chess960 initial position? what was the play leading up to the diagram position?) and the shorter question 'what was the Chess960 initial position?', where the play is not asked for.

Thank you, François !

Post nr 10: If we state Chess960, then there are 960 initial positions to chose from including the standard initial position RSBQKBSR. If we add "Not RSBQKBSR starting position", then there are 959 to chose from. Probably both are possible, but I don't see a need to exclude the standard initial position, it is just one quite similar to the others. On the other hand, if we exclude RSBQKBSR, then we help the solver a little, he has one less to investigate. The other way around: in a normal proof game we help the solver a lot, he knows where to start from!
(Read Only)pid=16532
(12) Posted by Andrew Buchanan [Saturday, Apr 21, 2018 11:17]

@Francois: in the test line, why is it optimal to search for 8 units and not e.g. 32 or 6? What’s going on here please?
(Read Only)pid=16533
(13) Posted by Siegfried Hornecker [Saturday, Apr 21, 2018 11:34]

Sorry, Per. I misread the post about the computer testing. I thought I had read that the standard position had over 100 solutions. I retract my post.
(Read Only)pid=16534
(14) Posted by François Labelle [Saturday, Apr 21, 2018 15:21]

Ok, some explanation about the test line. The entire problem is of the type "chained stipulation", so the position of the test line is important. It is the first command (after the language command), so it applies at ply 0. The line

test not dia forsyth dlssltkt/bbbbbbbb/8/8/8/8/BBBBBBBB/DLSSLTKT

would have worked too and maybe is easier to understand.

My original line (searching for 8 units) is just for human convenience. It uses the fact that, in an initial Chess960 position, the pawns are deterministic and black pieces are mirrored. So I removed the pawns and the black pieces from the forsyth and compensated with the "AddPieces" option.

The test line is not specific to Chess960 initial positions, it can be used at any point in any problem to enforce any aim like + or #. In our case, the aim is "dia" and the test result is negated with "not".
(Read Only)pid=16535
(15) Posted by Andrew Buchanan [Monday, Apr 23, 2018 15:59]; edited by Andrew Buchanan [18-04-23]

Some other retro points about Chess960, illustrated by 4 very simple problems. Clickbaitily, let me say: "number 3 may provoke some debate" :-)

(1) In regular chess, one can (almost) never guarantee that an apparent castling is legal. Not so in Chess960:

(= 7+2 )

(2) In regular chess, one cannot lose parity, when returning to homebase. Not so in Chess960:

Retros mailing list 08/12/2012
(= 16+16 )
PG 11.0

(3) In Chess960, choice of game array and castling rights are separate concerns.
What I'm going to call Constitutional moves - castling, or the deployment of a pocket Knight - are permitted unless the historian of the diagram can prove otherwise. Police-court moves are in general forbidden. Such are en passant capture, promotion to Grasshopper, and claiming a draw under the 50-move rule.
Ronald Turnbull
The Problemist July 1994

Now Chess960 game array is not a "move", but still we need to make a conscious decision whether RS can force it. If we want it to operate in a Constitutional way, as other compositions here have implicitly done, then it will play nice with the castling convention. But the alternative works too: if Chess960 game array is a Police-Court matter, then we can end up with compositions like this:

(= 8+12 )

What's the name of that theme where each subset of the set of threats is exhibited by at least one defence? Because it's one of those.
EDIT: It turns out it's "partial combinative separation".

(4) Other familiar tropes from regular chess are still very possible:

(= 15+15 )
PG 8.5

(Read Only)pid=16537
(16) Posted by Siegfried Hornecker [Monday, Apr 23, 2018 18:32]; edited by Siegfried Hornecker [18-04-23]

Hi, Andrew.

What's the name of that theme where each subset of the set of threats is exhibited by at least one defence? Because it's one of those.

Anti-Fleck? Although to me this looks more like a normal Fleck, as one defense is left always. Maybe a complete Fleck if each subset is left intact also. Is that what "exhibited" means?

As for number 3, I don't see what the issue is, i.e. why it might provoke discussion? Aren't the rules in the Codex clear, as are those you cited from the Problemist? The black pawn position even proves that the black king must come from b8, with a black rook on a8, so White provably has king and rook in their initial position. In that case, castling rights are usually given.
(Read Only)pid=16538
(17) Posted by Joost de Heer [Monday, Apr 23, 2018 19:07]; edited by Joost de Heer [18-04-23]

I'm probably missing the joke, but What was white's last move in 1)?
And the theme for 3) is combinative separation.
(Read Only)pid=16539
(18) Posted by Andrew Buchanan [Monday, Apr 23, 2018 19:15]

Hi Siegfried,

The idea is that if the threats are A+B+C then there are defences which result in mates A, B, C, A+B, A+C, B+C, A+B+C. It would be nice to get the exact name, maybe it's something to do with Fleck. When I google Anti-Fleck I mainly get stain repellent :-)

Anyway, the little #2 is consistent with either the Police-Court or the Constitutional view of Chess960. The controversial aspect is that three other problems posted earlier here in the thread are inconsistent with the Police-Court view! No disrespect to the esteemed composers, but it is possibly a bit cheap if free White castling in Chess960 can blow up the possibility of later Black castling, if the kings are not on the same file, with no retro logic required.

In general, castling must be Constitutional because otherwise there is (almost) no way for it to happen. It can (almost) never be proved legal because any game can be prefixed with a little dance of rook and knights that specifically disables one castling right.

En passant is much more amenable to retro logic, either to prove it definitely legal or definitely illegal. We do need a convention that says what happens if we can't prove it either way, and rightly the decision is that in this case you can't e.p. E.p. is a cool effect and you need to work for it.

So which of these scenarios is uncertainty in the Chess960 game array closer to? There's lots of scope for retro logic to prove various aspects of the start position. So that suggests maybe this should be a Police-Court item: we need to work to prove the starting squares of R&K before we deserve to castle. After we have confirmed the starting squares, we have a Constitutional right to castle.

Either way, my point is we must make a conscious choice, Constitutional or Police-Court, for the Chess960 game array convention.
(Read Only)pid=16540
(19) Posted by Andrew Buchanan [Monday, Apr 23, 2018 19:25]; edited by Andrew Buchanan [18-04-23]

Hi Joost,

Thanks for combinative separation! There's an article by Ian Shanahan:, which reveals that my problem is "partial" not "total" c.s., because some of the variations merely repeat some of the mate-combinations. Very reasonable: I thought the definition might include some such criterion.

Problem (1) just used the official joke, Codex Article 15: "If the first move does not lie with the conventional party, this should either be indicated in the stipulation or deducible from retroanalysis." In the case of a helpmate, Black loses their first move. However, there is no deception, nor clean retro try, nor reason to make the problem retro in nature, so maybe I should stop telling people a joke they've heard a thousand times before and just set the problem to be h#3.5?

EDIT: I have updated my earlier post on both these points - thanks again, Joost.
(Read Only)pid=16541
(20) Posted by Joost de Heer [Monday, Apr 23, 2018 20:07]

Codex Article 15: "If the first move does not lie with the conventional party, this should either be indicated in the stipulation or deducible from retroanalysis."

For (1), h#4 would mean white starts and black delivers mate on his first move, since the conventional party (black) doesn't have the first move. So it should be h#3.5.
(Read Only)pid=16542

Page: [Previous] [Next] 1 2 3

MatPlus.Net Forum General Chess960 problems?