This is a discussion of the actual program files for the static code. The discussion of the format of the input and output files are given elsewhere.

The files in the source code distribution of static are:

• makefile, which is discussed in the Installation section.
• The parameter files P1, P2 and P3, which are also discussed in the Installation section.
• static.f is the main program. It organizes the calculation, opens input (SKIN) and output (SKOUT, SKENG) files, and passes information to the other processors, as needed. Note that only static.f and bande.f make explicit calls to MPI. All other functions and subroutines either run independently on each processor, or only on the master processor, which is assumed to have MPI rank 0. static.f calls the following subroutines:
  • In scalable parallel mode, static calls several routines in the Message Passing Interface (MPI) Library. In serial mode, we define mpifake.h, a fake MPI header file, which tells the code that there is only one processor available, and mpifake.f, a fake MPI library source. This subterfuge is designed to increase the portability of the code with little penalty in execution speed for serial computers. In one case, in the subroutine bande.f, using fake MPI calls would cause a large amount of needless copying. In this case we have written the code to skip the MPI calls when only one processor is being used. Note that this trick also saves time if, for some reason, you want to run an MPI compiled program on only one processor.
  • input.f reads the Slater-Koster parameter file. The current version of the file reads files in the format listed in the Tight-Binding Parameters Database. This routine may be modified to select other versions of the input parameters. input.f runs only on the master processor.
  • input1.f reads the structural information for each lattice in the input (SKIN) file. This includes the primitive vectors of the lattice, the number of atoms, the number of electrons, and the internal parameters of the crystal. This is discussed in detail in the examples in the Usage section.
    Starting in Version 1.05, you may define an arbitrary set of Bravais vectors use a predefined lattice. To give the user this choice, input1.f calls
    • setlat.f, which allows the user to select from a preselected set of lattice types or to select an arbitrary lattice. After the lattice is selected, setlat also allows you to construct an arbitrary or pre-selected strain, e.g., to calculate energy versus strain and hence the elastic constants of the crystal.
  • kptin.f reads the information which determines the k-point mesh in the problem. This may be a list of k-points, a set of instructions to build the k-point mesh, or a file in which the k-point information may be obtained. Construction of a k-point mesh is discussed in the examples in the Usage section. This routine and all of its subprocesses run only on the master processor. It may be feasible to run the k-point generation program over many processors, but it is probably not time efficient. kptin.f calls the following subroutines:
    • rdspgp.f, which reads the space group file. Space group files for many lattices can be found in the Crystal Lattice Structures database. The proper space group file is required only if the program is to generate the k-points. If pre-generated k-points are used, one can use this null space-group file. rdspgp.f calls matmlt.f, which performs some matrix multiplications.
    • mapk.f, which converts from lattice to Cartesian coordinates, and visa versa
    • kpgen.f, which actually actually generates the k-point mesh, if needed. This routine calls the routines gdot.f, which performs inner products with a metric, hsort.f, which sorts the k-points, and also mapk.f.
  • search2.f determines which atoms are close enough to be connected by the Slater-Koster hopping and overlap integrals. This determines the number of Slater-Koster pairs used in the calculation. The distance, direction, and atom types involved in each pair will be passed via the /pairs/ common block. Note that it may be profitable to convert search2.f to a scalable program. At the moment it runs on each processor, in order to minimize the amount of message passing required between the nodes.
  • bande.f performs the actual Slater-Koster computation of the energy and, optionally, the pressure or electronic structure. bande.f runs in parallel, assigning k-points to each processor and collecting the results. This parallelization scheme works best for a problem which has a large number of k-points an a relatively few number of processors. See the discussion in diag.f for a possible alternative scheme for making the code scalable. bande.f calls:
    • setpar.f, which uses the information from search2.f and the tight-binding parameter file to construct the Slater-Koster on-site, Hamiltonian, and overlap matrix elements. Prior to version 1.11 this routine was part of setup.f If you have an alternative method for determining Slater-Koster parameters you can place it into the code by modifying setpar.f.
    • setparv.f is used to set up the presure derivatives of the Slater-Koster parameters found in setpar.f. Prior to version 1.11 this routine was part of setvol.f.
    • setham.f, which uses the information from setpar.f to determine the Hamiltonian and overlap matrices at a given k-point. This code was formerly part of setup.f. setham.f calls
      • rotate.f, which transforms the Slater-Koster matrix elements into the appropriate Hamiltonian and overlap matrix elements by rotating from the standard molecular basis set onto the actually Cartesian direction, and applies Bloch's theorem to incorporate the k-point mesh.
    • diag.f, which calls the LAPACK routine zhpev to solve the generalized Hermitian eigenvalue problem, or zhpgv to solve the generalized Hermitian eigenvalue problem with eigenvectors. For large diagonalization problems, especially with a large number of available processors and only a few k-points, it is likely that better scalability will result by replacing these subroutines with the corresponding ones from the ScaLAPACK distribution.
      • The relevant LAPACK and BLAS routines called by diag.f and perturb.f (below) are collected in the file lapack.f. If your computer installation has an optimized version of LAPACK and/or BLAS, it is undoubtably better to link in this version rather than use the generic versions which are listed here. Another alternative, the open-source ATLAS (Automatically Tuned Linear Algebra Software) package. The SGI Origin makefile discussed in the Installation section shows how this can be accomplished on one system.
    • setparv.f, which, if the pressure is to be calculated, uses the results of diag.f and setparv.f to set up the perturbation matrices required to determine the Hellmann-Feynman pressure for this calculation. The execution of setparv.f is very similar to setham.f. Like setham.f, setparv.f also calls rotate.f to populate the Hamiltonian and overlap matrix.
    • perturb.f, which, if the pressure is to be calculated, determines the perturbation energies for the Hellmann-Feynman pressure, using the LAPACK routine zhpev to split degenerate eigenvalues. The routine is similar in operation to diag.f.
  • Finally, the tweaks.f routine analyzes the contents of the TWEAKS file, if it exists, and signals the main program as to which diagnostic information is the be printed.

static Home Page   Introduction   Installation   List of Files   Usage   Input Files   Output Files   Trouble Shooting   Appendix

Current URL: http://esd.cos.gmu.edu/tb/static/list_of_files.html