In order to compile this program, you will need to have GSL and GLPK installed. This version was tested with GSL version 1.15 and GLPK version 4.55. You may also need to make sure that the #include statements in the .cpp and .h files are correct for your GLPK installation.
The program call for the TJ program is:
$ ./tj_3d iconfig.dat parameters.txt output-
Change "3d" to the dimension that you'd like to use. While the program was written with a lot of flexibility available at runtime, the vectors and matrices used are hard-coded according to dimension due to their ubiquity, so you need a different executable for each dimension. Fortunately, most people will only need two different version at most!
-
The first argument is the initial configuration that will be packed using TJ.
-
The second argument tells TJ where to read parameter values such as the influence sphere radius, compression and translation limits, etc.
-
The third argument tells TJ where to write the final configuration as well as any intermediate configurations (see the print_every parameter). IMPORTANT: It will automatically append a ".dat" at the end of the name you supply. This is because intermediate configurations may be written as so many iterations; these are named "output_xx.dat", where "xx" is the LP iteration after which the packing was written.
The TJ algorithm is designed to generate packings in any dimension. The way that the files are written is as follows:
[dim]
[number of spheres]
[sphere diameter]
[Lattice (dim rows)]
[sphere centers (in global coordinates)]
Note that the lattice is written so that its ROWS span the fundamental cell. The sphere centers' locations are written in GLOBAL coordinates, meaning that these are their positions, INDEPENDENT OF THE FUNDAMENTAL CELL. Their LOCAL coordinates (i.e. in terms of the fundamental cell's lattice matrix) may be obtained by multiplying the global coordinates by the lattice matrix's inverse, and vice versa.
One potential issue that may arise while running TJ is that GLPK will print messages saying:
"Warning: numerical instability (primal simplex, phase I)"
or
"Warning: numerical instability (primal simplex, phase II)"
The cause of this is normally that the primal basis has become infeasible due to roundoff errors. GLPK will attempt to fix the problem and continue solving, but in the event that this doesn't work, try lowering "feasible_tol" in the parameters file. The included parameters file has some more suggestions regarding the proper selection of parameter values.
Another alternative is to implement a more advanced LP solver. While GLPK is a decent free option, better options exist that are free to those who are affiliated with an academic institution. My personal favorite is Gurobi. The base class lp_class has been designed so that other solvers can inherit from it and be implemented seamlessly. It should be straightforward to follow the example set by the GLPK implementation for your solver of choice.