URI:
   DIR Return Create A Forum - Home
       ---------------------------------------------------------
       knight tours
  HTML https://knighttours.createaforum.com
       ---------------------------------------------------------
       *****************************************************
   DIR Return to: magic tours
       *****************************************************
       #Post#: 5--------------------------------------------------
       8x10 magic knight's tours
       By: gsgs Date: March 4, 2023, 9:16 am
       ---------------------------------------------------------
       The smallest rectangular board with magic knight tours is the
       8x8, which has been
       searched completely already. So the next is 8x10.
       An exhaustive search is probably out of reach, so we collect
       single tours ,
       as they had been doing for 183 years now ifor the 8x8.
       I found one in Jan.2020 by linking a 4x8 and a 6x8 with 2 links.
       Then in Feb.2020 Awani Kumar sent a list of 24 (10 different
       magic)
       by linking two 4x10s with 4 links.
       Then I found 12 in 2020 ? by linking a 4x8 and a 6x8 with 4
       links.
       (counting only geometrically different tours)
       The more links, the more magic tours , the more computation time
       , the more complicated ?!
       [img]
  HTML http://magictour.free.fr/8x10.PNG[/img]
       22+1+10 tours are different, 20 open and 13 closed
       file :
  HTML http://magictour.free.fr/8x10.txt
       I use numbers T(0,..,m*n-1) for the board-cells visited at move
       i.
       always starting at zero.
       It's compatible to my programs and modulo-math.
       But I may change it to A(1..m,1..n) to show the movenumbers
       in the cells starting at 1 9as usual ?!).
       However, one line per tour is better IMO for multiple tours in
       one file
       than rectangles. Easier to check,sort by computer.
       -----------------------
       I found 4 more magic 8x10s now
       [edit:12 , searching 4 of 12 start-end combinations]
       ( the last 4 open ones in the updated picture above)
       by partitioning the board into these
       two knight-friendly symmetric parts and touring them separately.
       That gave 1216424180 knight's tours for each part alone.
       For an estimated 1.2e9^2/40=4e16 tours by combining them,
       0,001% of the estimated total of 10^22 open 8x10 tours.
       Then I picked a matching start and end cell at random and
       stored the 18 sums for each of the 20866920 and 25972378
       solutions.
       I calculated the 18 sums for each, normal for the first part
       and magic constant - sum for the second, 1 byte per sum.
       Then I sorted that 800MB file and searched for matching sums.
       This gave the 4 solutions. The whole process took 3 hours
       and luckily it worked without much debugging.
       If someone wants to repeat it for others of the 24
       start-end-combinations or the other knight-friendly partition,
       I can send the programs but it still needs programming skills,
       all the steps are not fully automised.
       I estimate > 100 magic 8x10 tours with this method.
       This same procedure gives 21 of the 108 magic 8x8 tours = 20%.
       There are 7900012 solutions per part , so combinations give
       8e6*8e6/30/2e16 = 0.015% of all tours but 20% of the magic ones.
       -----------------edit 2023-02-21----------------------
       the calculated chance to get the 14 sums matching with this
       method
       is very low. There are an estimated 50 sums that may occur
       in each line, so 1/50^14 (the last 2 sums are forced)
       or < 1e-16 with 20 million sums checked.
       That I still succeeded with 4,1,3,4 matches for the 4 checked
       start-end combinations is presumably because some "inverse"
       sub-paths are symmetric.
       So I do not expect this method to be useful for other,
       non-symmetric,
       knight-friendly partitions.
       With 4 out of 12 start-end combinations checked my estimate for
       the total number of tours with this method is down to 36.
       I started a complete search for 180^ symmetric magic 8x10s on
       one Intel Atom Z3735 - core.(<2W),
       expecting ~10 solutions. It may take 1 month....I assume Yann
       could do it in <1h
       1600 8x10 double symmetric magic 2-kts (two-knight-tours) and
       their
       generating paths
  HTML http://magictour.free.fr/8xtgb.GIF
       #Post#: 32--------------------------------------------------
       Re: 8x10 magic knight's tours
       By: gsgs Date: January 6, 2024, 6:10 am
       ---------------------------------------------------------
       --------------2024-01-06--------------
       I'm rerunning the search for symmetric magic 8x10s now on a
       Z3735 processor, 1 core ...
       hamp4m.d1 8 10
       this run was interrupted after 53 days in Feb.2023 without
       storing results
       The estimated runtime was 30d
       in Linux I see, that the empty output-file was accessed after 14
       days ?!?
       That should have been a solution
       Or a power-failure and restart, I don't remember
       hamp4m.d1 8 8 takes 17h on the tecra
       #Post#: 34--------------------------------------------------
       Re: 8x10 magic knight's tours
       By: gsgs Date: February 20, 2024, 12:53 pm
       ---------------------------------------------------------
       a simple brute-force program (hamp4r.c12 8 12 x) to search for
       (180"-rotation)symmetric magic 8x10 knight's tours
       found no solution after about 1 day on 20 cores.(x=1..20)
       The program found the 16 symmetric 8x8 tours
       in 20 min on 1 core
       and the first symmetric magic 8x12 (see below)
       after 1 day on 20 cores. (estimated time
       for the total search = 10000 days on one core)
       From the nodecounts I had expected some 8x10
       solutions, so I assume there is some mathematical
       argument that symmetric magic knight's tours
       are only possible when both sides of the rectangles
       are divisible by 4.
       40,77,02,61,38,75,58,19,94,33,56,17
       03,62,39,76,01,60,95,34,57,18,93,32
       78,41,70,05,74,37,20,59,24,91,16,55
       63,04,73,42,71,00,35,84,21,54,31,92
       44,79,06,69,36,83,48,23,90,25,52,15
       07,64,43,72,11,68,85,26,53,22,89,30
       80,45,66,09,82,47,12,49,28,87,14,51
       65,08,81,46,67,10,27,86,13,50,29,88
       40,--,02,--,38,--,--,19,--,33,--,17
       03,--,39,--,01,--,--,34,--,18,--,32
       --,41,--,05,--,37,20,--,24,--,16,--
       --,04,--,42,--,00,35,--,21,--,31,--
       44,--,06,--,36,--,--,23,--,25,--,15
       07,--,43,--,11,--,--,26,--,22,--,30
       --,45,--,09,--,47,12,--,28,--,14,--
       --,08,--,46,--,10,27,--,13,--,29,--
  HTML http://magictour.free.fr/8x12-sy.bmp
       that gives the idea to search
       such patterns only ...
       ... but no other such solution was found with complete search,
       1d,12 cores
       a 2nd solution was found after 50 core*days (from 10000
       estimated)
       and it also looks special. 88 of its 96 edges have their
       horizontal reflection included.
  HTML http://magictour.free.fr/8x12-sy.GIF
       --------update 2024-02-29--------------------
       These were just the first 2 found (randomly) out of an estimated
       500.
       I'm now searching for all such tours with 3 visited cells per
       row
       after 24 moves.
       This should finish in some weeks or days.
       ---------------
       from the first 40 such solutions 19 were of that 2nd type above,
       i.e, had only 8 edges whose horizontal reflection was not
       part of the tour.
       Another 8 had only 12 such edges
       ----------2024-03-02-----------------
       solutions are getting rarer, I expect 80 (istead of 500)
       and these include tours and inverse tours, so 40 different ones
       -----------------2024-03-04----------------
       55066 2-symmetric magic 12x12 tours of that "semi-4-symmetric
       A,B,-Sy(A),-Sy(B),Sxy(A),Sxy(B),-Sx(A),-Sx(B)" type
       were found in 2 hours on my laptop  (27319 different ones)
  HTML http://magictour.free.fr/cxc.mtn
       also :
       30 * 16x12s in 5min
       661 * 16x8s in 4sec
       4 * 16x16s in 10min
       2532 * 20x16s in 30min
       0 * 20x20s in 10h
       estimated 2e12 * 16x16s in 1e13sec
       -------------------------------------
       this reminds me to early 2023 when I examined
       4-symmetric 2kts . Now I see, that you can make 1kts
       with "modified" 4-symmetry
       A,B,-Sy(A),-Sy(B),Sxy(A),Sxy(B),-Sx(A),-Sx(B)
       instead of
       A,B,Sxy(A),Sxy(B),-Sx(B),-Sx(A),-Sy(B),-Sy(A)
       ---------2024-03-12----------------------
       in order to create symmetric 4m*4n knight's tours ...
       it's probably easier and more suitable to create
       these special A,B,-Sy(A),-Sy(B)  tours .
       And again it's probably easier to require
       in addition that the cells in A and B are from the
       left half and have no two cells in the same aligned 2x2 box.
       These tours can then be drawn as double king-tours
       e,g. this 12x12 tour :
       [code]
       (4,6)-(6,5)
       o-o-o  o-o-o
       |  |  |  |
       o o o  o o-o
       | |\    | |
       o o o  o o o
       | | |  | | |
       o o o  o o o
       |  \    |  |
       o o o  o o o
       |/| |  |/| |
       o o-o  o o-o
       03-02-01  24-23-22
       |    |  |    |
       04 15 00  25 20-21
       |  |\    |  |
       05 14 16  26 19 35
       |  |  |  |  |  |
       06 13 17  27 18 34
       |  \    |    |
       07 09 12  28 30 33
       |/ |  |  |/ |  |
       08 10-11  29 31-32
       [/code]
       you can attach columns on the left to make it
       12x16,12x20,...
       -----------------
       on the 8x8 we have 272 symmetric magic knight tours
       from 16 symmetry classes and 32 of type AB-a-b from
       2 symmetry classes none of them of the special type
       as in the 12x12 shown above.
       So the cases 8*4n are not included yet
       -------------------------
       number of such specian symmetric magic knight tours :
       12x12:4
       12x16:249
       12x20:19617
       12x24:1279902
       16x12:27
       16x16:17272
       16x20:>4e6
       20x12:104
       20x16:1411614
       20x20:>5e5
       20x24:?
       24x12:1703
       24x16:?1960
       28x12:>6700
       *****************************************************