The idea of MALADA is to automate as many steps of the data generation for MALA workflow as posible. It is not meant to be a universal tool for DFT-MD workflows, and is specifically tailored to this application. In order to create training data for MALA networks, the following steps need to be undertaken:
- Crystal structure determination (What element in what structure do you want to model?)
- Supercell construction (How many atoms will be in your supercell?)
- DFT convergence (Which cutoff energy and k-grid should be used for DFT calculations?)
- DFT-MD performance test (How do you need to parallelize a DFT-MD run so that you can get results in reasonable time?)
- DFT-MD simulations (Calculate a trajectory with your given parameters)
- Snapshot selection (Filter this trajectory)
- LDOS parameter determination (How should the LDOS be discretized? On which grid should it be discretized?)
- DFT simulations, LDOS calculation (Perform a DFT calculation and calculate the LDOS from the final wave functions)
To automate this, MALADA provides a pipeline, that can be controlled using the global parameters.