This silences a deprecation warning on scipy < 1.2 and numpy > 1.15.

add automatic Peierls phase calculation

Closes #19

See merge request kwant/kwant!219

Co-authored-by: Pablo Piskunow <pablo.perez.piskunow@gmail.com>
Co-authored-by: Daniel Varjas <dvarjas@gmail.com>

use the shift-inverse method for finding sparse eigenvalues
in the finite system tutorial.

See merge request kwant/kwant!265

This avoids "sparse efficiency" warnings.

add testing environment specification as Dockerfiles and conda environments

See merge request kwant/kwant!260

Sometimes the testing jobs fail because the build artifacts
are removed.

Only builds testing images when the relevant dockerfile/conda envs
have changed.

+ factor kernel (jackson or lorentz etc.) into separate functions, and
  expose these as part of the API
+ factor out the random vector factory, and allow to specify finite
  collections of vectors (e.g. for sampling over some subdomain)
+ implement KPM expansion of correlators, and add a factory function
  for calculating conductivity.

Closes #138

See merge request kwant/kwant!218

allow computations even if a discrete symmetry is broken

Closes #236 and #242

See merge request !257

Previously if there were more site families than colours in the
matplotlib cycle, then plotting would raise an exception. Now
we cycle through the colours indefinitely.

Closes #257

This was fixed on stable, but due to interim changes and a badly
solved merge conflict, the fix was not applied to master.

This has a lower chance to give an impression that the substitution
happens in-place.

Closes gitlab issue #237

The 'mpl_available' flag was relegated to '_plotter' in
the development version, but the check in the test was still
referring to 'plotter'. The change 'plotter → _plotter' was
made on the development version, so this fix does not need to
be backported to the stable branch.

Currently we test against 5 environments: Ubuntu LTS, Debian stable,
minimum required dependency versions from conda forge, minimum
required dependencies without extras (sympy and matplotlib), and latest
dependency versions from conda forge.

Dániel Varjas authored 7 years ago and Joseph Weston committed 6 years ago

Improve the logic of 'cvp', now it provably works in all cases.
Add 'reduced' option to 'linalg.lll.voronoi', this eliminates
superfluous Voronoi vectors that do not have a corresponding face
of the Voronoi cell, or this face would be smaller than 'rtol'.

Simplify 'lattice.Polyatomic.neighbors' by providing the option
to group equidistant vectors 'group_by_length' in 'linalg.lll.cvp'.

See !165 for details.

Closes #163.