How to integrate your own solver

You can use your own impurity solver from DCore by wrapping the executable as appropriate. The null solver serves as a template of wrapper code.

null_solver.py

Follow the instructions below.

Copy the template

In directory python/impurity_solvers, copy the template

$ cp null_solver.py your_solver.py

Edit the template

• Change the class name

  class NullSolver(SolverBase):
  

  Let’s say you changed the class name into YourSolver.

• Set the solver name

  def name(self):
      return "null"
  

• Set input to the solver

  # (1) Set configuration for the impurity solver
  # input:
  #   self.beta
  #   self.set_G0_iw
  #   self.u_mat
  #
  # Additionally, the following variables may be used:
  #   self.n_orb
  #   self.n_flavor
  #   self.gf_struct
  #   self.n_tau
  #   self.use_spin_orbit
  
  # (1a) If H0 is necessary:
  # Non-interacting part of the local Hamiltonian including chemical potential
  # Make sure H0 is hermite.
  # Ordering of index in H0 is spin1, spin1, ..., spin2, spin2, ...
  H0 = extract_H0(self._G0_iw, self.block_names)
  
  # (1b) If Delta(iw) and/or Delta(tau) are necessary:
  # Compute the hybridization function from G0:
  #     Delta(iwn_n) = iw_n - H0 - G0^{-1}(iw_n)
  # H0 is extracted from the tail of the Green's function.
  self._Delta_iw = delta(self._G0_iw)
  Delta_tau = make_block_gf(GfImTime, self.gf_struct, self.beta, self.n_tau)
  for name, block in self._Delta_iw:
      Delta_tau[name] << Fourier(self._Delta_iw[name])
  
  # (1c) Set U_{ijkl} for the solver
  # for i, j, k, l in product(range(self.n_flavors), repeat=4):
  #     self.u_mat[i, j, k, l]
  

  Here, you generate all necessary input files to run your program.

• Run the solver

  # (2) Run a working horse
  with open('./output', 'w') as output_f:
      launch_mpi_subprocesses(mpirun_command, [exec_path, 'input.ini'], output_f)
  

  The second argument to the function launch_mpi_subprocesses is the actual command and arguments used for invoking MPI processes from DCore. These options and arguments to the solver must be given as a list (refer to subprocess module in Python).

• Convert output of the solver

  # (3) Copy results into
  #   self._Sigma_iw
  #   self._Gimp_iw
  

  Read output files generated by the solver, and set them into DCore variables. The self-energy and the Green’s function are stored as BlockGf class of TRIQS library. See other wrappers or the TRIQS documentation.

Register your solver

Finally, you register your own solver to DCore. Edit python/impurity_solvers/__init__.py to import your class

from .your_solver import YourSolver

and add it to the dictionary variable solver_classes as

solver_classes = {
    ...
    'your_solver': YourSolver,
}

Then, you can invoke your solver from DCore by

[impurity_solver]
name = your_solver