# Changelog
All notable changes to this project will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [unreleased]

### Removed

+ Support of python < 3.9

## [1.4.0] - 2023-12-16
+ Additionally allow using strings or sympy expressions in `PointGroupElement`
  for the rotation part [gitlab issue
  #7](https://gitlab.kwant-project.org/qt/qsymm/-/issues/7).
+ Avoid a numerical instability in symmetry finder [gitlab issue
  #49](https://gitlab.kwant-project.org/qt/qsymm/-/issues/49).
+ Fix compatibility with latest sympy v1.12.

## [1.3.0] - 2020-11-01

### Added

+ integration test: run the tests for Kwant's qsymm module against the current version
  of qsymm.

### Fixed

+ Avoided using deprecated functionality in latest scipy, numpy, and sympy
+ Fix syntax for python 3.9

### Removed

+ Support of python < 3.7, numpy < 1.16, scipy < 1.1, sympy < 1.3
  (similar to the upcoming Kwant release)

## [1.2.6] - 2019-11-12

### Fixed
- Downgraded scipy version requirement to 0.19, to maintain
  compatibility with Kwant
- Correctly set model format when reshaping models with sparse values

## [1.2.5] - 2019-11-11
### Added
- Added basic reference documentation
- Added citation instructions in CITING.md
- Added a tutorial with downloadable Python files and Jupyter notebooks

### Fixed
- Allow 'display_family' to work when IPython is not installed
- Bug when multiplying a model by a sympy expression (undefined variable)
- Bug when multiplying BlochCoeff with sympy expression (sympy multiplication was
  being used, when BlochCoeff multiplication should have been used)
- Remove copying of Sympy symbols (the hash values for the original and copy is
  different in some cases (probably a bug in Sympy)), which broke comparison
  between Models.
- Remove 'real' assumption from Sympy symbols in 'hamiltonian_from_family'; this
  makes all Sympy symbols in Qsymm consistently have no assumptions.
- Minor clarifications to various docstrings

### Removed
- Removed the notebooks from the source repository

## [1.2.4] - 2019-10-17
### Fixed
- Set correct content-type on long description metadata

## [1.2.3] - 2019-10-17
### Fixed
- Add long description to package metadata

## [1.2.2] - 2019-10-17
### Fixed
- Adding zero to a model is now idempotent

## [1.2.1] - 2019-09-02
### Added
- Add CHANGELOG.md

### Fixed
+ Restore compatibility with Python 3.5

## [1.2.0] - 2019-08-30
### Added
- Factory functions for discrete symmetries: `time_reversal`, `particle_hole`, `chiral`, `inversion`, `rotation`, `mirror`.
- Better representation of `PointGroupElement`s and `ContinuousGroupGenerator`, using `_repr_pretty_` and `_repr_latex_`. Remove `print_PG_elements` and update example notebooks.
- Implemented new functionality in `Model`:
 + Implement `__matmul__` (`@`).
 + Support of sparse matrices and `LinearOperator` as values.
 + Consistent support of scalar valued `Model`s.
 + Add `keep` to only keep track of certain symbolic coefficients.
 + More options and more transparent initialization, allow string keys which are automatically symmpified by default.
 + Several new utility functions, such as `trace`, `reshape`, `allclose` etc.

### Changed
- Slight change to the internal representation of `PointGroupElements`, allow mixing integer tinyarray with either sympy matrix or floating point tinyarray in the rotation part, but do not allow mixing the latter two representations. This removes the need to have two different representations for simple operators. Optimize multiplication by caching.
- Changes in the API of `Model`:
 - Add `format` attribute to keep track of the type of data entries (i.e scalar, dense or sparse array).
 - Change the behaviour of `*` for `Model` objects, matrix multiplication only works with `@` from now on. This breaks backward compatibility, a few uses in the qsymm code base were fixed.
 - Stop supporting a sequence of ints as `momenta`, instead string keys (e.g. `'k_x'`) or `sympy` expressions can be used. Certain other, rarely used ways to initialize `Model` don't work anymore, some tests were changed accordingly.
 - Stop rounding and removing small entries from `Model` objects. This means that testing that the `Model` is empty is not a good test anymore to see if it is approximately zero. Use `Model.allclose` instead.
- Optimizations in `Model`:
 + Remove unnecessary deep copying, which was slow.
 + Optimize the implementation of arithmetic operations to minimize number of loops and function calls.
 + Fast initialization by making restructuring optional when initialized with a dict.
 + Clean up the code of `BlochModel` to utilize improvements in `Model`.
 + Update symmetry finder to work with sparse models.

### Deprecated
- Deprecate initializing empty `Model` without providing `shape` and `format`.

## [1.1.0] - 2018-12-05
Many of these changes were made in anticipation of integrating `qsymm` with `kwant`,
allowing conversion between the Hamiltonian formats used.

### Added
+ `BlochModel` as a subclass of `Model` in order to store Bloch Hamiltonians
  with floating point hopping vectors.
+ `bravais_point_group` to find the point group of a Bravais lattice using the
  translation vectors only. This is intended for generating candidades for `symmetries`.

### Changed
+ Change the way equality of `Model`s is tested, the current implementation treats
  tolerances more consistently, this fixes some bugs in symmetry finding.
+ Allow using floating point rotation matrices in `PointGroupElement`.

### Fixed
+ Several bugs and internal code restructuring.