Release notes

Git master branch


Version 21.6.0

Jun 24, 2021: 21.6.0

Version 21.1.0

Jan 18, 2021: 21.1.0

  • Corresponding ASE release: ASE-3.21.0.

  • We now use GPAW’s own (faster) implementation for LDA, PBE, revPBE, RPBE and PW91. For most calculation the speedup is unimportant, but for our test-suites it gives a nice boost. There can be small meV changes compared to the LibXC implementation. If you want to use LibXC then use:

    from gpaw.xc.gga import GGA
    from gpaw.xc.libxc import LibXC
    calc = GPAW(xc=GGA(LibXC('PBE')), ...)
  • New Zero-field splitting module.

  • New Scissors operator for LCAO mode.

  • Nonlinear optical responses can now be calculated in the independent particle approximations. See the Nonlinear optical response of an extended system tutorial for how to use it to compute the second-harmonic generation and shift current spectra.

  • New method for interpolating pseudo density to fine grids: gpaw.utilities.ps2ae.PS2AE.get_pseudo_density() (useful for Bader analysis and other things).

  • Now with contribution from “frozen” core: Isotropic and anisotropic hyperfine coupling paramters.

  • Change in parameters of linear response TDDFT

  • Improved relaxation in the excited states in parallel, see linear response TDDFT

  • We now have a Code coverage report updated every night.

  • Plane-wave mode implementation of hybrid functionals can now be selected via a dict: xc={'name': ..., 'backend': 'pw'}, where then name must be one of EXX, PBE0, HSE03, HSE06 or B3LYP. The EXX fraction and damping parameter can also be given in the dict.

Version 20.10.0

Oct 19, 2020: 20.10.0

Version 20.1.0

Jan 30, 2020: 20.1.0

  • Corresponding ASE release: ASE-3.19.0.

  • Self-consistent calculations with hybrid functionals are now possible in plane-wave mode. You have to parallelize over plane-waves and you must use the Davidson eigensolver with one iteration per scf step:

    from gpaw import GPAW, PW, Davidson
    calc = GPAW(mode=PW(ecut=...),
                parallel={'band': 1, 'kpt': 1},
  • We are now using setuptools instead of distutils. This means that installation with pip works much better.

  • No more gpaw-python. By default, an MPI-enabled Python interpreter is not built (use parallel_python_interpreter=True if you want a gpaw-python). The C-extension file (usually only used for serial calculations) will now be compiled with mpicc and contain what is necessary for both serial and parallel calculations. In order to run GPAW in parallel, you do one of these three:

    $ mpiexec -n 24 gpaw python
    $ gpaw -P 24 python
    $ mpiexec -n 24 python3

    The first two are the recommended ones: The gpaw script will make sure that imports are done in an efficient way.

  • Configuration/customization: The file in the root folder of the Git repository is no longer used. Instead, the first of the following three files that exist will be used:

    1. the file that $GPAW_CONFIG points at

    2. <git-root>/

    3. ~/.gpaw/

    This will be used to configure things (BLAS, FFTW, ScaLAPACK, libxc, libvdwxc, …). If no configuration file is found then you get libraries = ['xc', 'blas'].

  • A Lapack library is no longer needed for compiling GPAW. We are using scipy.linalg from now on.

  • Debug mode is now enabled with:

    $ python3 -d
  • Dry-run mode is now enabled with:

    $ gpaw python --dry-run=N
  • New convergence criterium. Example: convergence={'bands': 'CBM+2.5'} will converge bands up to conduction band minimum plus 2.5 eV.

  • Point-group symmetries now also used for non-periodic systems. Use symmetry={'point_group': False} if you don’t want that.

  • Marzari-Vanderbilt distribution function added.

  • New configuration option: noblas = True. Useful for compiling GPAW without a BLAS library. scipy.linalg.blas and will be used instead.

Version 19.8.1

Aug 8, 2019: 19.8.1


Upgrading from version 1.5.2

Some small changes in the code introduced between version 1.5.2 and 19.8.1 (improved handling of splines) may give rise to small changes in the total energy calculated with version 19.8.1 compared to version 1.5.2. The changes should be in the meV/atom range, but may add up to significant numbers if you are doing calculations for large systems with many atoms.

  • Corresponding ASE release: ASE-3.18.0.

  • Important bug fixed: reading of some old gpw-files did not work.

Version 19.8.0

Aug 1, 2019: 19.8.0

  • Corresponding ASE release: ASE-3.18.0.

  • The "You have a weird unit cell" and "Real space grid not compatible with symmetry operation" errors are now gone. GPAW now handles these cases by choosing the number of real-space grid-points in a more clever way.

  • The angular part of the PAW correction to the ALDA kernel is now calculated analytically by expanding the correction in spherical harmonics.

  • Berry phases can now be calculated. See the Berry phase calculations tutorial for how to use it to calculate spontaneous polarization, Born effective charges and other physical properties.

  • How to do Ehrenfest dynamics (TDDFT/MD) has now been documented.

  • Non self-consistent hybrid functional calculations can now be continued if they run out of time. See gpaw.xc.exx.EXX.calculate().

  • When using a convergence criteria on the accuracy of the forces (see Accuracy of the self-consistency cycle), the foceces will only be calculated when the other convergence criteria (energy, eigenstates and density) are fulfilled. This can save a bit of time.

  • Experimental support for JTH PAW-datasets.

  • Fast C implementation of bond-length constraints and associated hidden constraints for water models. This allows efficient explicit solvent QMMM calculations for GPAW up to tens of thousands of solvent molecules with watermodels such as SPC, TIPnP etc. See gpaw/utilities/ and gpaw/test/ for examples.

  • New “metallic boundary conditions” have been added to the for PoissonSolver. This enables simulating charged 2D systems without counter charges. See: gpaw/test/poisson/

  • Removed unnecessary application of H-operator in davidson algorithm making it a bit faster.

Version 1.5.2

May 8, 2019: 1.5.2

  • Corresponding ASE release: ASE-3.17.0.

  • Important bugfix release:

    There was a bug which was triggered when combining ScaLAPACK, LCAO and k-points in GPAW 1.5.0/1.5.1 from January. The projections were calculated incorrectly which affected the SCF loop.

    If you use ScaLAPACK+LCAO+kpoints and see the line “Atomic Correction: distributed and sparse using scipy” in the output, then please rerun after updating.

Version 1.5.1

Jan 23, 2019: 1.5.1

  • Corresponding ASE release: ASE-3.17.0.

  • Small bug fixes related to latest versions of Python, Numpy and Libxc.

Version 1.5.0

Jan 11, 2019: 1.5.0

  • Corresponding ASE release: ASE-3.17.0.

  • Last release to support Python 2.7.

  • The default finite-difference stencils used for gradients in GGA and MGGA calculations have been changed.

    • The range of the stencil has been increased from 1 to 2 thereby decreasing the error from \(O(h^2)\) to \(O(h^4)\) (where \(h\) is the grid spacing). Use xc={'name': 'PBE', 'stencil': 1} to get the old, less accurate, stencil.

    • The stencils are now symmetric also for non-orthorhombic unit cells. Before, the stencils would only have weight on the nighboring grid-points in the 6 directions along the lattice vectors. Now, grid-points along all nearest neighbor directions can have a weight in the stencils. This allows for creating stencils that have all the crystal symmetries.

  • PW-mode calculations can now be parallelized over plane-wave coefficients.

  • The PW-mode code is now much faster. The “hot spots” have been moved from Python to C-code.

  • Wavefunctions are now updated when the atomic positions change by default, improving the initial wavefunctions across geometry steps. Corresponds to GPAW(experimental={'reuse_wfs_method': 'paw'}). To get the old behaviour, set the option to 'keep' instead. The option is disabled for TDDFT/Ehrenfest.

  • Add interface to Elpa eigensolver for LCAO mode. Using Elpa is strongly recommended for large calculations. Use:

         parallel={'sl_auto': True, 'use_elpa': True})

    See also documentation on the parallel keyword.

  • Default eigensolver is now Davidson(niter=2).

  • Default number of bands is now \(1.2 \times N_{\text{occ}} + 4\), where \(N_{\text{occ}}\) is the number of occupied bands.

  • Solvated jellium method has been implemented, see the documentation.

  • Added FastPoissonSolver which is faster and works well for any cell. This replaces the old Poisson solver as default Poisson solver.

  • Range separated functionals (RSF) and improved virtual orbitals, the latter from Hartree-Fock theory.

  • New Jupyter notebooks added for teaching DFT and many-body methods. Topics cover: Catalysis, Magnetism in 2D, Machine Learning, Photovoltaics, Batteries and Introduction to Python and Jupyter notebooks.

  • New experimental local k-point refinement feature: gpaw/test/

  • A module and tutorial have been added for calculating electrostatic corrections to DFT total energies for charged systems involving localised defects: Tools for defect calculations.

  • Default for FFTW planning has been changed from ESTIMATE to MEASURE. See

Version 1.4.0

May 29, 2018: 1.4.0

  • Corresponding ASE release: ASE-3.16.0.

  • Improved parallelization of operations with localized functions in PW mode. This solves the current size bottleneck in PW mode.

  • Added QNA XC functional: Quasi-non-local exchange correlation approxmation.

  • Major refactoring of the LCAOTDDFT code and added Kohn–Sham decomposition analysis within LCAOTDDFT, see the documentation.

  • New experimental keyword, GPAW(experimental={...}) to enable features that are still being tested.

  • Experimental support for calculations with non-collinear spins (plane-wave mode only). Use GPAW(experimental={'magmoms': magmoms}), where magmoms is an array of magnetic moment vectors of shape (len(atoms), 3).

  • Number of bands no longer needs to be divisible by band parallelization group size. Number of bands will no longer be automatically adjusted to fit parallelization.

  • Major code refactoring to facilitate work with parallel arrays. See new module: gpaw.matrix.

  • Better reuse of wavefunctions when atoms are displaced. This can improve performance of optimizations and dynamics in FD and PW mode. Use GPAW(experimental={'reuse_wfs_method': name}) where name is 'paw' or 'lcao'. This will move the projections of the wavefunctions upon the PAW projectors or LCAO basis set along with the atoms. The latter is best when used with dzp. This feature has no effect for LCAO mode where the basis functions automatically follow the atoms.

  • Broadcast imports (Python3 only): Master process broadcasts most module files at import time to reduce file system overhead in parallel calculations.

  • Command-line arguments for BLACS/ScaLAPACK have been removed in favour of the parallel keyword. For example instead of running gpaw-python --sl_diagonalize=4,4,64, set the parallelization within the script using GPAW(parallel={'sl_diagonalize': (4, 4, 64)}).

  • When run through the ordinary Python interpreter, GPAW will now only intercept and use command-line options of the form --gpaw key1=value1,key2=value2,... or --gpaw=key1=value1,key2=value2,....

  • gpaw-python now takes Command-line options directly instead of stealing them from sys.argv, passing the remaining ones to the script: Example: gpaw-python --gpaw=debug=True myscript_arguments. See also gpaw-python --help.

  • Two new parameters for specifying the Pulay stress. Directly like this:

    GPAW(mode=PW(ecut, pulay_stress=...), ...)

    or indirectly:

    GPAW(mode=PW(ecut, dedecut=...), ...)

    via the formula \(\sigma_P=(2/3)E_{\text{cut}}dE/dE_{\text{cut}}/V\). Use dedecut='estimate' to use an estimate from the kinetic energy of an isolated atom.

  • New utility function: gpaw.utilities.ibz2bz.ibz2bz().

Version 1.3.0

October 2, 2017: 1.3.0

  • Corresponding ASE release: ASE-3.15.0.

  • Command-line options --dry-run and --debug have been removed. Please use --gpaw dry-run=N and --gpaw debug=True instead (or --gpaw dry-run=N,debug=True for both).

  • The ase.Atoms.get_magnetic_moments() method will no longer be scaled to sum up to the total magnetic moment. Instead, the magnetic moments integrated inside the atomic PAW spheres will be returned.

  • New sbatch sub-command for GPAW’s Command line interface.

  • Support added for ASE’s new band-structure Command line tool:

    $ ase band-structure xxx.gpw -p GKLM
  • Added tetrahedron method for calculation the density response function.

  • Long-range cutoff for qmmm calculations can now be per molecule instead of only per point charge.

  • Python 2.6 no longer supported.

  • There is now a web-page documenting the use of the in development version of GPAW:

  • BSE calculations for spin-polarized systems.

  • Calculation of magnetic anisotropy.

  • Calculation of vectorial magnetic moments inside PAW spheres based on spin-orbit spinors.

  • Added a simple gpaw.occupations.occupation_numbers() function for calculating occupation numbers, fermi-level, magnetic moment, and entropy from eigenvalues and k-point weights.

  • Deprecated calculator-keyword dtype. If you need to force the datatype of the wave functions to be complex, then use something like:

    calc = GPAW(mode=PW(ecut=500, force_complex_dtype=True))
  • Norm-conserving potentials (HGH and SG15) now subtract the Hartree energies of the compensation charges. The total energy of an isolated pseudoatom stripped of all valence electrons will now be zero.

  • HGH and SG15 pseudopotentials are now Fourier-filtered at runtime as appropriate for the given grid spacing. Using them now requires scipy.

  • The gpaw dos sub-command of the Command line interface can now show projected DOS. Also, one can now use linear tetrahedron interpolation for the calculation of the (P)DOS.

  • The gpaw.utilities.ps2ae.PS2AE tool can now also calculate the all-electron electrostatic potential.

Version 1.2.0

Feb 7, 2017: 1.2.0.

  • Corresponding ASE release: ASE-3.13.0.

  • New file-format for gpw-files. Reading of old files should still work. Look inside the new files with:

    $ python3 -m abc.gpw
  • Simple syntax for specifying BZ paths introduced: kpts={'path': 'GXK', 'npoints': 50}.

  • Calculations with fixdensity=True no longer update the Fermi level.

  • The GPAW calculator object has a new band_structure() method that returns an ase.spectrum.band_structure.BandStructure object. This makes it easy to create band-structure plots as shown in section 9 of this awesome Psi-k Scientfic Highlight Of The Month:

  • Dipole-layer corrections for slab calculations can now be done in PW-mode also. See Dipole-layer corrections in GPAW.

  • New get_electrostatic_potential() method.

  • When setting the default PAW-datasets or basis-sets using a dict, we must now use 'default' as the key instead of None:

    >>> calc = GPAW(basis={'default': 'dzp', 'H': 'sz(dzp)'})

    and not:

    >>> calc = GPAW(basis={None: 'dzp', 'H': 'sz(dzp)'})

    (will still work, but you will get a warning).

  • New feature added to the GW code to be used with 2D systems. This lowers the required k-point grid necessary for convergence. See this tutorial Quasi-particle spectrum of two-dimensional materials.

  • It is now possible to carry out GW calculations with eigenvalue self- consistency in G. See this tutorial GW0 calculations.

  • XC objects can now be specified as dictionaries, allowing GGAs and MGGAs with custom stencils: GPAW(xc={'name': 'PBE', 'stencil': 2})

  • Support for spin-polarized vdW-DF functionals (svdW-DF) with libvdwxc.

Version 1.1.0

June 22, 2016: 1.1.0.

  • Corresponding ASE release: ASE-3.11.0.

  • There was a BUG in the recently added spin-orbit module. Should now be fixed.

  • The default Davidson eigensolver can now parallelize over bands.

  • There is a new PAW-dataset file available: gpaw-setup-0.9.20000.tar.gz. It’s identical to the previous one except for one new data-file which is needed for doing vdW-DF calculations with Python 3.

  • Jellium calculations can now be done in plane-wave mode and there is a new background_charge keyword (see the Jellium tutorial).

  • New band structure unfolding tool and tutorial.

  • The get_pseudo_wave_function() method has a new keyword: Use periodic=True to get the periodic part of the wave function.

  • New tool for interpolating the pseudo wave functions to a fine real-space grids and for adding PAW-corrections in order to obtain all-electron wave functions. See this tutorial: Obtaining all-electron wave functions and electrostatic potential.

  • New and improved dataset pages (see Periodic table). Now shows convergence of absolute and relative energies with respect to plane-wave cut-off.

  • Interface to Wannier90.

  • Updated MacOSX installation guide for Homebrew users.

  • topological index

Version 1.0.0

Mar 17, 2016: 1.0.0.

  • Corresponding ASE release: ASE-3.10.0.

  • A BUG related to use of time-reversal symmetry was found in the \(G_0W_0\) code that was introduced in version 0.11. This has been fixed nowplease run your calculations again.

  • New gpaw.external module.

  • The gradients of the cavity and the dielectric in the continuum solvent model are now calculated analytically for the case of the effective potential method. This improves the accuracy of the forces in solution compared to the gradient calculated by finite differences. The solvation energies are expected to change slightly within the accuracy of the model.

  • New \(f_{\text{xc}}\) kernels for correlation energy calculations. See this updated tutorial.

  • Correlation energies within the range-separated RPA. See this tutorial.

  • Experimental interface to the libvdwxc library for efficient van der Waals density functionals.

  • It’s now possible to use Davidson and CG eigensolvers for MGGA calculations.

  • The functional name “M06L” is now deprecated. Use “M06-L” from now on.

Version 0.11.0

July 22, 2015: 0.11.0.

  • Corresponding ASE release: ASE-3.9.1.

  • When searching for basis sets, the setup name if any is now prepended automatically to the basis name. Thus if setups='<setupname>' and basis='<basisname>', GPAW will search for <symbol>.<setupname>.<basisname>.basis.

  • Time-propagation TDDFT with LCAO.

  • Improved distribution and load balance when calculating atomic XC corrections, and in LCAO when calculating atomic corrections to the Hamiltonian and overlap.

  • Norm-conserving SG15 pseudopotentials and parser for several dialects of the UPF format.

  • Non-selfconsistent spin-orbit coupling have been added. See tutorial for examples of band structure calculations with spin-orbit coupling.

  • Text output from ground-state calculations now list the symmetries found and the k-points used. Eigenvalues and occupation numbers are now also printed for systems with k-points.

  • GW, RPA correlation energy, and response function calculation has been rewritten to take advantage of symmetry and fast matrix-matrix multiplication (BLAS).

  • New symmetry keyword. Replaces usesymm.

  • Use non-symmorphic symmetries: combining fractional translations with rotations, reflections and inversion. Use symmetry={'symmorphic': False} to turn this feature on.

  • New forces keyword in convergence. Can be used to calculate forces to a given precision.

  • Fixed bug in printing work functions for calculations with a dipole-correction

  • A continuum solvent model was added.

  • A orbital-free DFT with PAW transformation is available.

  • GPAW can now perform Classical electrodynamics simulations using the quasistatic finite-difference time-domain (QSFDTD) method.

  • BEEF-vdW, mBEEF and mBEEF-vdW functionals added.

  • Support for Python 3.

Version 0.10.0

Apr 8, 2014: 0.10.0.

  • Corresponding ASE release: ASE-3.8.1

  • Default eigensolver is now the Davidson solver.

  • Default density mixer parameters have been changed for calculations with periodic boundary conditions. Parameters for that case: Mixer(0.05, 5, 50) (or MixerSum(0.05, 5, 50) for spin-paired calculations). Old parameters: 0.1, 3, 50.

  • Default is now occupations=FermiDirac(0.1) if a calculation is periodic in at least one direction, and FermiDirac(0.0) otherwise (before it was 0.1 eV for anything with k-points, and 0 otherwise).

  • Calculations with a plane-wave basis set are now officially supported.

  • One-shot GW calculations with full frequency integration or plasmon-pole approximation.

  • Beyond RPA-correlation: using renormalized LDA and PBE.

  • Bethe-Salpeter Equation - Theory.

  • Improved RMM-DIIS eigensolver.

  • Support for new libxc 2.0.1. libxc must now be built separately from GPAW.

  • MGGA calculations can be done in plane-wave mode.

  • Calculation of the stress tensor has been implemented for plane-wave based calculation (except MGGA).

  • MGGA: number of neighbor grid points to use for FD stencil for wave function gradient changed from 1 to 3.

  • New setups: Y, Sb, Xe, Hf, Re, Hg, Tl, Rn

  • Non self-consistent calculations with screened hybrid functionals (HSE03 and HSE06) can be done in plane-wave mode.

  • Modified setups:


    Most of the new semicore setups currently require eigensolver dav, cg eigensolvers or rmm-diis eigensolver with a couple of iterations.

    • improved eggbox: N, O, K, S, Ca, Sc, Zn, Sr, Zr, Cd, In, Sn, Pb, Bi

    • semicore states included: Na, Mg, V, Mn, Ni, Nb, Mo, Ru (seems to solve the Ru problem gpaw/test/big/Ru001/), Rh, Pd, Ag, Ta, W, Os, Ir, Pt

    • semicore states removed: Te

    • elements removed: La (energetics was wrong: errors ~1eV per unit cell for PBE formation energy of La2O3 wrt. PBE benchmark results)


    For some of the setups one has now a choice of different number of valence electrons, e.g.:

    setups={'Ag': '11'}

    See PAW datasets or pseudopotentials and list the contents of GPAW_SETUP_PATH for available setups.

  • new dzp basis set generated for all the new setups, see

Version 0.9.0

Mar 7, 2012: 0.9.0.

  • Corresponding ASE release: ase-3.6

  • Convergence criteria for eigenstates changed: The missing volume per grid-point factor is now included and the units are now eV**2. The new default value is 4.0e-8 eV**2 which is equivalent to the old default for a grid spacing of 0.2 Å.

  • GPAW should now work also with NumPy 1.6.

  • Much improved Command line interface now based on the new tool in ASE.

Version 0.8.0

May 25, 2011: 0.8.0.

  • Corresponding ASE release: ase-3.5.1

  • Energy convergence criterion changed from 1 meV/atom to 0.5 meV/electron. This was changed in order to allow having no atoms like for jellium calculations.

  • Linear dielectric response of an extended system (RPA and ALDA kernels) can now be calculated.

  • RPA correlation energy.

  • Non-selfconsistent calculations with k-points for hybrid functionals.

  • Methfessel-Paxton distribution added.

  • Text output now shows the distance between planes of grid-points as this is what will be close to the grid-spacing parameter h also for non-orthorhombic cells.

  • Exchange-correlation code restructured. Naming convention for explicitely specifying libxc functionals has changed: Exchange-Correlation functional.

  • New PAW setups for Rb, Ti, Ba, La, Sr, K, Sc, Ca, Zr and Cs.

Version 0.7.2

Aug 13, 2010: 0.7.2.

  • Corresponding ASE release: ase-3.4.1

  • For version 0.7, the default Poisson solver was changed to PoissonSolver(nn=3). Now, also the Poisson solver’s default value for nn has been changed from 'M' to 3.

Version 0.7

Apr 23, 2010: 0.7.

  • Corresponding ASE release: ase-3.4.0

  • Better and much more efficient handling of non-orthorhombic unit cells. It may actually work now!

  • Much better use of ScaLAPACK and BLACS. All large matrices can now be distributed.

  • New test coverage pages for all files.

  • New default value for Poisson solver stencil: PoissonSolver(nn=3).

  • Much improved MPI module (MPI communicators).

  • Self-consistent Meta GGA.

  • New PAW setup tar-file now contains revPBE setups and also dzp basis functions.

  • New $HOME/.gpaw/ configuration file.

  • License is now GPLv3+.

  • New HDF IO-format.

  • Advanced GPAW Test System Introduced.

Version 0.6

Oct 9, 2009: 0.6.

  • Corresponding ASE release: ase-3.2.0

  • Much improved default parameters.

  • Using higher order finite-difference stencil for kinetic energy.

  • Many many other improvements like: better parallelization, fewer bugs and smaller memory footprint.

Version 0.5

Apr 1, 2009: 0.5.

Version 0.4

Nov 13, 2008: 0.4.

  • Corresponding ASE release: ase-3.0.0

  • Now using ASE-3 and numpy.

  • TPSS non self-consistent implementation.

  • LCAO mode.

  • VdW-functional now coded in C.

  • Added atomic orbital basis generation scripts.

  • Added an Overlap object, and moved apply_overlap and apply_hamiltonian from Kpoint to Overlap and Hamiltonian classes.

  • Wannier code much improved.

  • Experimental LDA+U code added.

  • Now using libxc.

  • Many more setups.

  • Delta scf calculations.

  • Using localized functions will now no longer use MPI group communicators and blocking calls to MPI_Reduce and MPI_Bcast. Instead non-blocking sends/receives/waits are used. This will reduce synchronization time for large parallel calculations.

  • More work on LB94.

  • Using LCAO code forinitial guess for grid calculations.

  • TDDFT.

  • Moved documentation to Sphinx.

  • Improved metric for Pulay mixing.

  • Porting and optimization for BlueGene/P.

  • Experimental Hartwigsen-Goedecker-Hutter pseudopotentials added.

  • Transport calculations with LCAO.

Version 0.3

Dec 19, 2007: 0.3.