Also require that 'leads' of FiniteSystems have a 'parameters'
property.

Specifically: finalized Builder, PrecalculatedLead,
              ModesLead and SelfEnergyLead

This can be used to inspect the parameters on which a system depends,
and can be later used to give more informative error messages.

Adding 'parameters' to ModesLead and SelfEnergyLead is strictly a
backwards-incompatible change, however this is an advanced feature of
Kwant and we don't want to let this block adding 'parameters' to the
system API.

Closes #263.

Anton Akhmerov authored 6 years ago and Joseph Weston committed 6 years ago

Also correctly select random sites without repetitions.
Closes issue #265.

This was committed by mistake during debugging.

The build process uses a recipe from a separate repository, which was not
updated for a while.  Instead we should be calling conda-build directly.
However this may be done separately.

The old 'toolchain' package has been removed from the conda-forge
channel, and the 'gcc' package has no builds for Linux (several
non-authoratative sources say that it is "deprecated"), and it
seems that the toolchain3 package provides the necessary compilers

Also fixup some docstrings for better formatting.

add kwant.qsymm module for working with Qsymm

Closes #256

See merge request kwant/kwant!261

Also update CI to install qsymm on relevant platforms, and update
pytest config to skip kwant.qsymm tests when qsymm is not installed

Tómas authored 6 years ago and Joseph Weston committed 6 years ago

Co-authored-by: Daniel Varjas <dvarjas@gmail.com>

add magnetic gauge fixing for infinite systems and systems with leads

See merge request kwant/kwant!267
Closes #246.

Test that the sign convention and units are correct, and that
we correctly recover current flowing along the boundaries of
a bar system when a magnetic field is applied.

Magnetic fields are now expressed in units of magnetic flux quantum
per unit area.

We still refer to these as "phases" everywhere in the code.
Working with phase factors means that users do not have to
exponentiate anything (and potentially get an incorrect sign).
In addition this is closer to what we need for non-abelian
gauge fields.

This means less repeated documentation.

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

The UI will show the test failures as warnings (which we can
address).

Add first and second derivative calculation to kwant.physics.Bands,
and allow eigenvectors to be optionally returned.

See merge request kwant/kwant!262

The 'out' parameter is only used when you don't want numpy to
allocate a new array for the result, which is not the case here.

Thomas Kloss authored 6 years ago and Joseph Weston committed 6 years ago

np.linalg.eigvalsh returns eigenvalues already in ascending order