> king + rock + pawn

Do you mean rock + paper + scissors?

Even with rock it is krppkrp.

BTW, endings with pawns are much more dificult to generate because of all promotions should be considered, i.e. adequite tablebases should be generated.

Thanks for the math!

Think about economy.

As I understand you need 64 bits (2bytes *32bits) for every position. But you need not store draw positions. Send positions with mate in 1 to the first file, with mate in 2 - to the second and so on. You need not store draw positions. So you will need 2 times less space (even more for some endings).

With modern 64-bit computers you may save one position even in 1 byte. So 2 times less space for 64-bit computers.

Terrific!!!

1TB is for one ending. It is $100. So it will be $100 000 for all endings. Now I understand your problems with disk space. Looks like we need 10 years more to have 7-piece endings available for own analysis. Sorry, I was very naive at my first posts.

Incredible!! Do you have a link to what is known (longest mates, longest conversion) for various 7 piece endgames?

I have checked the position suggested by Mark & Yakov.
And I have found out that there also exists mate in 545 moves, but DTC in my line is shorter than in Mark's (514 and 517). So both sides are right.

Well, I decided to check other endings of Yakov&Mark. This night it was krrnkrr. The work was not so hard as previous time.
(2.5 hours for krrnkrr ending plus about hour for minor endings that were not generated before).

My program confirmed that in Yakov&Mark position

mate is given in 296 moves.

But it offered another way to mate with DTC=285 instead of Yakov&Mark DTC=290. That is logical. I don't search the largest DTC.

As well a new position with longer mate was found with mate in 304 (DTC=288)!



This is logical too. 6 moves to mate after the conversion for the Yakov&Mark position is evidently too short. For longest DTM Black should try to transform to the postion with longest 6-men mate (may be with slightly less 7-men DTC). For my position White needs 17 moves to mate after the conversion.
It is funny that final mate is in the center of the board.

PGN files are attached (krrnkrr_dtm296.pgn, krrnkrr_dtm304.pgn). Let me know if visualization on the board is essential, so I will attach it too.

Next endings will be
1) KQBN-KQB
1k6/1b5q/N7/8/8/1Q6/8/B1K5 b - -
Yakov&Mark claim 330 moves DTC

2) KQBN-KQN
Nqn3k1/8/8/Q7/8/8/8/1K2B3 w - -
Yakov&Mark claim 317 moves DTC

Then I will need a pause to solve the problems with disk space. If the disk space problem will be solved I start systematic generation.

Generation of kqbnkqb took more than 8 hours.

Mate in 345

The results coincide with Yakov and Mark. My program has offered another position, but agreed that Yakov&Mark's leads to the longest mate too. I am not sure how many moves until checkmate there are in their line.

kqbnkqn endings demanded more research than I expected.


The line offered by Yakov&Mark is mate in 350 by DTC. (see kqbnkqn_Yakov.pgn attached)

I was able to get only mate in 335 moves by DTM (see kqbnkqn_dtm335b.pgn attached).

I didn't like the fact that DTC line is longer, so I decided to consider lines more carefully.

First of all I made modifications to allow generation of DTC tables. This was quite easy. Then I ran generation of DTC kqbnkqn tablebase. It was more than 2 times faster than DTM (possibly with optimizations even more speed up may be achieved).

It was quite pretty to see that my DTC-generation program exactly confirmed every move in Yakov&Mark's line, i.e. every move in line was optimal by DTC.

So I checked positions in DTC line with DTM table. And I found a lot of positions in this ending where the best move by DTC is not optimal by DTM.

For example in this position in Yakov's DTC line (before 313.Qd3)

DTC gives "9 half-moves to conversion" and then mate in 33 moves (65 halfmoves) by 6-men tablebases. Totally 74 halfmoves.
DTM gives "15 half-moves to mate" (all 7 pieces remain on the board). It is very short mate and may be checked by Rybka even without 7-men tablebases.

(See visualization kqbnkqn.aqs)

So both tables look ok. And I think that due a lot of mate posibilities this ending is very good for finding diferences between DTM and DTC tables.

For now I've finished comparison of DTC and DTM long mates. After number of tests I am more or less sure that the program works Ok.

As result interesting comparison of different endings with DTM and DTC metrics was received. For some endings lines are very similar, for some endings they are different.

For now I will take a break before October when I am going to start more systematic generation.

so great. this geniality with the computer. this helps us. now i think it should by modificated the fifty moves low. it shows it is possible to win the other positions longer tnan in fifty moves no pawn moves and no piece changes not only without position with two knight vs. one pawn, which is special, in 100 moves. it will be interesting ni chess lows. maybe in world norm using new temp of playing (for example in master of world championship) is ,or will be, used computers to detect way to win and modify conventional laws. i enjoy it, future.