v0.6.0
DFTK v0.6.0
Note that this release increases the minor version and contains a number of breaking changes:
-
The
self_consistent_field
andpotential_mixing
SCF functions now check convergence in the density by default. This means thatself_consistent_field(basis; tol=1e-6)
now converges until the change in density between subsequent steps is below1e-6
. Before it used to converge until the change in energy was less thantol
. To recover the old behaviour useself_consistent_field(basis; is_converged=DFTK.ScfConvergenceEnergy(tol))
. (#800) -
The
load_lattice
,load_atoms
andload_positions
functions are now deprecated and will be removed in DFTK 0.6.1. Users are advised to useAtomsIO
for the purpose of loading structures from external files. See https://docs.dftk.org/v0.6/examples/atomsbase/ and https://github.com/mfherbst/AtomsIO.jl for examples. -
Instead of using NLSolve's Anderson acceleration we have switched to a built-in implementation, which we found more stable in practice. (#791)
The NLSolve-been approach has been removed completely. -
The deprecated
G_to_r
andr_to_G
functions for FFTs have now been removed in favour ofifft
,fft
and similar. (#804) -
The deprecated
n_bands
andn_ep_extra
arguments fromself_consistent_field
have now been removed. (#804)
Closed issues:
- Struct for atoms (#118)
- GPU discussion (#350)
- SCF default tolerance should be on the density (#780)
- Run GPU CI only after activation (#803)
Merged pull requests:
- Bugfix for computations of
δψ
and supercells (#775) (@epolack) - Switch default scf solver to builtin Anderson (#791) (@epolack)
- Add developper documentation for GPU computations (#793) (@GVigne)
- Correct PW parameterization comment (#795) (@epolack)
- Add GPU CI via gitlab (#796) (@mfherbst)
- Switch to density tolerance by default (#800) (@mfherbst)
- Make cigpu badge a link to ci pipelines in gitlab (#801) (@carstenbauer)
- Rely on AtomsIO for structure parsing (#802) (@mfherbst)
- Remove deprecated features (#804) (@mfherbst)