DIR Return Create A Forum - Home
---------------------------------------------------------
knight tours
HTML https://knighttours.createaforum.com
---------------------------------------------------------
*****************************************************
DIR Return to: general Hamiltonian cycles problem
*****************************************************
#Post#: 13--------------------------------------------------
Andrew Chalaturnyk's 2008 program
By: gsgs Date: April 23, 2023, 1:20 am
---------------------------------------------------------
is here :
HTML https://github.com/aachalat/hamilton_cycle
It's from his 2008 thesis
HTML https://github.com/aachalat/hamilton_cycle/raw/master/chalaturnykthesis.pdf
improving a program from William Kocay from 1998, which goes
back
to a program from Frank Rubin, 1975.
"multipath-method"
Apparently not the method with pathstructures and borders
as in the 2015-Pettersson thesis.
-----------------------------
on page 88 he gives some benchmarks for knight graphs.
Apparently using the same decomposition of the 8x8 into
41790 subproblems as on my and Yann Denef's webpage !
But without comment, without calculating all closed 8x8 tours.
However it's ~10 times faster than my program.
-------------------------------
It is quite fast.
But it has a "triangle problem", which may increase the running
time a lot
when all but one neighbors of a (small) cycle are included
but none of the cycle vertices yet.
This can't become Hamiltonian.
Another problem is with forced edges : better replace them by
one new vertex of degree 2 and edges to the endpoints.
It often benefits from random restart, when some node-limit is
exceeded,
this may be due to the triangle-problem.
It also benefits from initial resorting of the vertices
for small border-size.
With this reordering and random restarts after 100000 nodes,
it solves 500 of the 1001 FHCP-challenge-graphs in ca. 5 hours
HTML https://sites.flinders.edu.au/flinders-hamiltonian-cycle-project/fhcp-challenge-set/
*****************************************************