5.0.0 STEMsalabim 5.0.0 releaseRelease 5.0.0
February 28th, 2019
simulation.skip_simulationare deprecated now. The groups
adf/center_of_massnow have a dimension for energy loss. It is usually
1unless plasmon scattering feature is used.
- Speed improvements by increasing the grid sizes to match efficient FFT sizes. Note, that this may result
in a higher simulation grid density than specified in
- Alternative parallelization scheme, see :ref:
parallelization-scheme. When appropriate, different MPI procs now calculate different frozen phonon configurations / defoci in parallel. This reduces the required amount of communication between the processors.
- Automatic calculation of
center of massof the CBEDs for all ADF points. The COMs are calculated when
adf.enabled = trueand stored in the NC file next to
adf/center_of_mass. Unit is mrad.
- New executables
ssb-run. The former prepares an input NC file from which the latter can run the simulation. This has multiple advantages. See :ref:
simulation-structurefor more information.
- Single plasmon scattering.
Other changes ^^^^^^^^^^^^^
- Ability to disable thermal displacements via
frozen_phonon.enable = falseparameter.
- Fixed a serious bug with the integrated defocus averaging.
- Input XYZ files can now contain more than one space or TAB character for column separation.
- Removed Doxygen documentation and doc string comments.
- Default FFTW planning is now
FFTW_MEASURE. This improves startup times of the simulation slightly.
- Changed the chunking of the
cbed/cbed_intensitiesvariables for faster write speed.
AMBER/slice_coordinatesvariable to the output file, that contains the
zcoordinate of the upper boundary of each slice in nm.
- Removed HTTP reporting and CURL dependency.
- Significant code refactoring and some minor bugs fixed.
- Improved documentation.
- Speed improvements by increasing the grid sizes to match efficient FFT sizes. Note, that this may result in a higher simulation grid density than specified in
March 9th, 2018
I'm releasing this version as 4.0.0, but neither the input nor output files changed. The parameter
precisionhas become deprecated and there is a parameter
tmp-dir. Please see the documentation.
- Removed option for double precision. When requested, this may be re-introduced, but it slowed down compilation times and made the code significantly more complicated. The multislice algorithm with all its approximations, including the scattering factor parametrization, is not precise enough to make the difference between single and double precision significant.
- Improved the Wave class, so that some important parts can now be vectorized by the compiler.
- Introduced some more caches, so that performance could greatly be improved. STEMsalabim should now be about twice as fast as before.
- Results of the MPI processors are now written to temporary files and merged after each configuration is finished.
This removes many MPI calls which tended to slow down the simulation. See the
- Moved the
Scatteringclasses to their own (isolated) library
libatomic. This is easier to maintain.
- Simplified MPI communication by getting rid of serialization of C++ objects into char arrays. This is too error-prone anyway.
- Added compatibility with the Intel parallel studio (Compilers, MKL for FFTs, Intel MPI). Tested with Intel 17 only.
- Some minor fixes and improvements.
- Reworked input/output file format.
- Reworked CBED storing. Blank areas due to bandwidth limiting are now removed.
- Changes to the configuration, mainly to defocus series.
- Compression can be switched on and off via config file now.
- Prepared the project for adding a Python API in the future.
- Added tapering to smoothen the atomic potential at the edges as explained in
I. Lobato, et al, Ultramicroscopy 168, 17 (2016) <https://www.sciencedirect.com/science/article/pii/S030439911630081X>_.
- Added analysis scripts for Python and MATLAB to the Si 001 example.
1.0b The first official beta version to the stable 1.0