[Diff since v0.4.7](https://github.com/JuliaMolSim/DFTK.jl/compare/v0.4.7...v0.5.0)
Note that this release increases the minor version and contains a number of breaking changes:
- The `atoms` field in the `Model` data structure, which used to contain pairs mapping from an element to a list of associated positions, is now split up into a plain `atoms` vector (containing a flat and repeated list of all elements) and a `positions` vector (containing the list of respective positions). I.e. whereas before one would have, for example:
```julia
Ga = ElementCoulomb(:Ga)
As = ElementCoulomb(:As)
atoms = [Ga => [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], ], As => [[0.7, 0.8, 0.9], ], ]
```
we now have
```julia
Ga = ElementCoulomb(:Ga)
As = ElementCoulomb(:As)
atoms = [Ga, Ga, As]
positions = [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], [0.7, 0.8, 0.9], ]
```
This change is carried forward to **all interfaces** of DFTK. This means that all `Model` constructors as well as the `model_atomic`, `model_DFT`, `model_LDA`, `model_PBE`, `model_SCAN` functions change interface. E.g. constructing an LDA model for this fictitious gallium arsenide system could now be done as such:
```julia
Ga = ElementCoulomb(:Ga)
As = ElementCoulomb(:As)
atoms = [Ga, Ga, As]
positions = [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], [0.7, 0.8, 0.9], ]
lattice = 5.6 * I(3)
model = model_LDA(lattice, atoms, positions)
```
All examples and tests have been updated and can be used to draw examples for the new interface. For convenience and to simplify amending downstream code the old `model_LDA` etc. function have not been completely removed for now, but only marked as deprecated. However, removal will occur at the next patch release. (#626)
- The `compute_forces` and `compute_forces_cart` functions now also return arrays of `Vec3` (Static vector of 3 dimensions). These vectors are the forces on a particular atom in the same order as the `model.positions` and `model.atoms` arrays. (#626)
- The `Model(lattice; atoms, positions, kwargs...)` constructor has been dropped in favour of `Model(lattice, atoms, positions; kwargs...)`, i.e. `atoms` and `positions` are now positional arguments. This was done to make the constructor resemble more the structure of the `model_LDA`, `model_PBE`, etc. functions. (#629).
- The default LDA model constructed by `model_LDA` is now `[:lda_x, :lda_c_pw]` instead of `[:lda_xc_teter93]`. To return to the old behaviour replace `model_LDA(lattice, atoms, positions; kwargs...)`
by `model_DFT(lattice, atoms, positions, :lda_xc_teter93; kwargs...)`. (#629)
- The default `kshift` in the `PlaneWaveBasis` constructor is now always `(0, 0, 0)`. It used to be `1/2` in one direction in case the `kgrid` was even in this direction. Manually specifying the desired `kshift` returns the old behaviour. (#630)
- The `load_psp(element; kwargs...)` function is discontinued and has been removed. Instead use a combination of `list_psp(element; kwargs...)` with `load_psp(identifier)` or the `attach_psp(system::AbstractSystem)` function. (#558)
- The deprecated `compute_stresses(scfres)` function has now been removed in favour of `compute_stresses_cart(scfres)`. (#628)
- The deprecated `PowerNonlinearity(C, α)` constructor has now been removed in favour of `LocalNonlinearity(ρ -> C * ρ^α)`. (#628)
**Closed issues:**
- Autodiff (#107)
- Figure out notations for the symmetry operations (#266)
- Possibility to "unfold" the k grid (#382)
- Make another round of profiling (#425)
- Improve density computation (#456)
- Employing DFTK results for Machine-Learning (#595)
- Refactor SymOp to be primarily (W,w) rather than (S,tau) (#618)
- Remove ksymops (#621)
**Merged pull requests:**
- Integrate with AtomsBase (#558) (@mfherbst)
- Adding framework for pairwise interactions (#602) (@epolack)
- Random optimization (#610) (@antoine-levitt)
- Fix performance issue in block array (#611) (@antoine-levitt)
- Small cleanup in LOBPCG (#612) (@mfherbst)
- Remove coordinate_cart field from Kpoint (#613) (@niklasschmitz)
- Small optimizations (#615) (@antoine-levitt)
- Refactoring diagonalize_all_kblocks (#617) (@antoine-levitt)
- Symop refactoring (again!) (#619) (@antoine-levitt)
- Fixes for julia 1.8 (#620) (@mfherbst)
- Excise ksymops (#622) (@antoine-levitt)
- Excise symops, part 2 (#624) (@antoine-levitt)
- Refactor atoms field in Model (#626) (@mfherbst)
- Remove deprecated features (#628) (@mfherbst)
- Make atoms and positions positional arguments of the Model (#629) (@mfherbst)
- Change default kshift to zero (#630) (@mfherbst)
- Remove kpoint fallback (#631) (@mfherbst)
- Remove final mention of ksymops (#632) (@mfherbst)