triqs_ana_cont_interface

A TRIQS front-end for ana_cont. Give it a Gf or BlockGf on an imaginary mesh, get the continued object back on a real-frequency mesh, with the physics conventions applied once and the diagnostics needed to judge the result attached to the output.

Install

Needs TRIQS 4.0 and ana_cont importable by the same python interpreter. ana_cont is not on PyPI, so install it from source first:

git clone https://github.com/josefkaufmann/ana_cont && pip install ./ana_cont

Then build against TRIQS, as with any TRIQS application:

source path_to_triqs/share/triqs/triqsvars.sh
git clone https://github.com/harrisonlabollita/ana_cont_interface ana_cont_interface.src
mkdir ana_cont_interface.build && cd ana_cont_interface.build
cmake ../ana_cont_interface.src
make
make test
make install

Without -DCMAKE_INSTALL_PREFIX=... the package installs into TRIQS_ROOT and is importable straight away. For a custom prefix, source the generated share/triqs_ana_cont_interface/triqs_ana_cont_interfacevars.sh afterwards.

Usage

Build a problem record, then solve it. The record is data — inspect it, edit one element's model or error bar, solve it repeatedly with different settings.

from triqs_ana_cont_interface import gf_problem, solve, linear_grid, validate

prob = gf_problem(g_iw, grid=linear_grid(-10, 10, 501), error=1e-4, n_iw=60)
res  = solve(prob, preblur=0.5)

res.g_w                  # Gf / BlockGf on MeshReFreq, complex retarded
res.a_w                  # the spectral matrix A = (i/2pi)(g_w - g_w^dag)
print(validate(res))     # per-element table of fit quality and spectral weight

Self-energies go through sigma_problem, which handles Sigma_inf and the model norm:

from triqs_ana_cont_interface import sigma_problem, tangent_grid

res = solve(sigma_problem(sigma_iw, grid=tangent_grid(20, 501), error=err, n_iw=400))
res.sigma_w

error is required. ana_cont has no error estimator, and a defaulted error bar produces a plausible-looking but meaningless spectrum.

Documentation

• doc/usage.md — imaginary time, self-energies, matrix-valued input, degenerate blocks, the preblur scan, solver options
• doc/validation.md — backtransform, check_moments, validate
• doc/conventions.md — what is applied internally, warnings, HDF5, limitations
• doc/testing.md — running the tests

Status

Fermionic MeshImFreq and MeshImTime, Green's functions and self-energies, diagonal and matrix-valued. No bosonic kernels and no Pade yet. Verified against real SrVO3 CTHYB data: the quasiparticle weight from the continued Sigma(w) agrees with the Matsubara estimate to under 1%.

License

GPLv3, the same as TRIQS — see LICENSE.txt and COPYING.txt. The continuation itself is done by ana_cont, which is MIT-licensed and is not redistributed here.