Gygi Research Group

Qbox Release Notes

1.50.1

This release fixes a bug that occasionally caused segmentation faults when using the wf_dyn = JD option.

1.50.0

Implementation of a Jacobi-Davidson algorithm for wavefunction optimization (set wf_dyn JD). Improved stability of the charge mixing algorithm by including Tikhonov regularization. New -atomic_density option for the run command to start iterations with an approximate superposition of atomic densities.

1.49.1

Improved the accuracy of the calculation of Esr.

1.48.0

Added a plot command to generate files in "xyz" or "cube" format. The files generated can be viewed using visualization programs such as VMD.

1.47.0

Implementation of multidimensional Anderson acceleration of charge mixing (see charge_mix_ndim variable). Implementation of a position constraint (see constraint command). Implementation of blocked molecular dynamics (BMD) for geometry optimization (see atoms_dyn variable). Modification of the Fourier interpolation of pseudopotential to guarantee higher accuracy (note: this may result in differences in total energies compared with previous versions. Differences are of the order of 5*10-6 Hartree).

1.45.3

Implementation of the client-server mode (-server option).

1.45.1

The implementation of SDCellStepper.C is modified to allow for compilation using the PGI compiler on Cray-XT4. The algorithm used is unchanged.

1.45.0

New features

  • A new "strain" command applies a deformation to the sample corresponding to a symmetric strain tensor. Maximally Localized Wannier Functions (MLWFs) can now be computed at each step of an MD simulation by setting the wf_diag variable to MLWF or MLWFC.

    • Bug fixes

  • The calculation of the stress tensor in 1.44.0 was incorrect when the unit cell was small compared to the rcps parameter (width of the gaussian pseudocharge distributions). The value of rcps is set to 1.5 (a.u.) in AtomSet.C. This affected calculations in which the unit cell is very small, e.g. containing only one atom. Bug fixed in 1.45.0.

    • 1.44.0

    New features

  • FFT grids have the same size at all k-points. Note: when reading sample files created with versions prior to 1.44.0, the wavefunctions are interpolated. This may introduce a small difference in the total energy of the sample.

    • Bug fixes

  • Corrected inefficient memory allocation in the "kpoint add" command

    • 1.43.2

    New features

  • Change of ecut: Changing the value of the ecut variable causes the wavefunction to be interpolated on the new basis. Useful to 1) compute the ground state by progressively increasing the value of ecut, or 2) save wavefunctions at low resolution to save disk space. Resizing of wavefunctions works for both increasing and decreasing values of ecut.
  • Change number of states: Changing the number of states (through a change of nempty or a change of net_charge) preserves the current wavefunctions. Useful to compute empty states after a calculation including only occupied states.
  • Parallel I/O: Improved implementation using MPI-IO
  • New structure optimization algorithms: SDA: Steepest Descent with Acceleration, implements a steepest descent algorithm with line minimization. CG: Conjugate gradient with line minimization
  • Flexible use of XML Schema documents species.xsd and sample.xsd: Parsing XML documents in the absence of the species.xsd and sample.xsd is now possible, with a message warning that validation could not be carried out.

    • Bug fixes

  • torsion command: sign of the torsion angle corrected.

    • Known bugs

  • The PSDA algorithm for wavefunction optimization only works for the k=0 (Gamma) k-point.

    • MPI operation

    The x86_64 FC8, FC6 and RedHat executables can be used on workstations where no mpich library is installed. The code can be run as
    $ ./qbox
    In that case, the files opened and written by qbox will reside in the directory where the executable is located. When working in a directory where the executable is not present, invoke qbox as
    $ ~/bin/qbox -p4wd .
    The parameter p4wd is needed to tell mpich to set the working directory to the current directory.
    If mpich is installed, qbox can be invoked as e.g.
    $ mpirun -np 2 ./qbox
    When using the mpirun command, the current working directory is automatically recognized by mpich. Note that on workstations having multiple CPUs, a corresponding number of mpi tasks can be used to speed-up the calculation. The above example can be used on a workstation having a dual core processor.