This method has been prototyped by Adrien Sorgniard with contributions
by Michal Nowak.  The prototype has been cleaned-up, debugged, optimized,
and integrated into Kwant by Christoph Groth.

It is now impossible to reduce the number of random vectors or moments.
It is also no longer possible to set the number of energy points used by
the method -- this can be supplied by providing a sequence of energies
when calling the KPM object. We replace the 'increase_energy_resolution'
and 'increase_accuracy' methods by 'add_moments' and 'add_vectors'.
The tests have also been updated to reflect these changes.

num_rand_vecs → num_vectors
num_sampling_points → num_energies
ham → hamiltonian
epsilon → eps

e → expr
discrete_coordinates → discrete_coords
substitutions → subs

The optimized routine now replicates the graph faithfully, exactly
preserving the hopping "directions" and values.

Filling now works directly on the graph dictionary, thus avoiding many
unnecessary intermediate steps.  Since filling is essentially graph
replication, the time gains are considerable.

The optimized fill is also more faithful to the template graph.

The new fill() sets each hoppings only once.

The filling starts now at (one or multiple) sites, not at a domain.
This is more consistent, since the flood fill works on the graph of
sites.

Other changes:

* max_sites is now the maximal number of sites that is still OK
* fill() no longer raises an exception when no sites were added, it
  warns now.

Rafal Skolasinski authored 7 years ago and Joseph Weston committed 7 years ago

Previously, only onsite matrix elements were tested, whereas the
test should have also tested the hopping elements.

Additionally, make sure to test the LLL code accordingly.

Rafal Skolasinski authored 7 years ago and Bas Nijholt committed 7 years ago

Calling "kwant.continuum.discretize('A * k_x', substitutions={'A': 'A + B'})"
ends with performing substitutions twice. Fixed by removing redundant second
substitution. Fixes issue #121.

Previously 'Builder.fill' did not respect the symmetry of the target
Builder. This change also adds a test to check that target system
symmetry is respected.

Fixes #119

Rafal Skolasinski authored 7 years ago and Joseph Weston committed 7 years ago

Current implementation uses ``kwant.physics.Bands``. This example fails
when one substitute ``k_x**2`` with for example ``k_x**4`` due to farther
then nearest neighbours hoppings. Using wraparound provides also easier way
to extend it to multidimensional band structures (except of plotting).

Closes #109.

The Mathjax CDN is shutting down:
https://www.mathjax.org/cdn-shutting-down/

Closes #116.

When the fundamental domains of the constituent symmetries
of a wrapped around system do not coincide, it was previously
possible for wrapping around to fail when 'keep' was provided,
even if wrapping around succeeded with 'keep=None'.

Rafal Skolasinski authored 7 years ago and Joseph Weston committed 7 years ago

Arguments of "discretize", "discretize_symbolic", and "build_discretized"
that does not represent Hamiltonian or discrete coordinates are now keyword-only.
Also "lattice_constant" argument is renamed to "grid_spacing" in order to
distinguish it from real lattice constant of a crystal. Closes 110.

Closes #113.

This allows to provide a helpful error message when someone tries to
import Kwant with Python 2.

Note that I moved the version code into a module of its own:
kwant.version.  The rationale is to separate this code from unrelated
stuff in kwant._common that requires various external libs.  The
kwant.version module should work on its own without any third-party
libraries.

This makes it possible to use colormaps that are white at zero and still
distinguish areas with and without sites.

This function is adapted to plotting band structure in the case
where the translation symmetry vectors are not orthogonal
(e.g. graphene). It can only handle a very limited subset of
systems (2D lattices in 2D realspace), and most of the time
use of 'plotter.spectrum' should be preferred.