image

Welcome to MALA!

What is MALA?

MALA is a software package for building ML models that replace density functional theory (DFT) calculations. DFT is one of the most widely used methods for simulating materials at a quantum level and predicting their properties, employed by researchers worldwide.

Although DFT is more efficient compared to other quantum mechanical simulation techniques, it still exhibits cubic scaling with the number of atoms and temperature. This limitation poses significant computational challenges for conducting large-scale simulations.

Recent research activities focus on harnessing the potential of ML and DFT. MALA leads these efforts by facilitating the construction of ML-DFT models, offering comprehensive access to the electronic structure and observables, including the total free energy of the system. Moreover, MALA ensures scalability with system size and various parameters, such as temperature.

How does MALA work?

MALA operates by predicting the so called local density of states (LDOS). The LDOS is an efficient and versatile representation of the electronic structure of a material. Based on the LDOS, a range of useful observables can be calculated.

MALA calculates a grid-based representation for any given atomic structure (so called descriptors). These descriptors are subsequently fed into a neural network, an efficient ML algorithm, to predict the LDOS. The prediction is performed individually at each point of the real space grid, facilitating straightforward scalability and parallelization.

Subsequently, MALA offers a variety of post-processing features to compute desired observables. Similar to a DFT calculation, but with enhanced speed. For further details on the workflow, refer to the first MALA publication.

Where to start?

You can effortlessly set up MALA on your system by following the installation instructions. To get acquainted with MALA, explore the basic usage guide, along with the example files in the repository. Additionally, MALA offers various options for advanced use, allowing you to enhance training and inference speed, scale to large systems, and visualize results.

If you use MALA in your research, please make sure to cite accordingly.

Who is behind MALA?

MALA is jointly developed by the Sandia National Laboratories (Scientific supervisor: Sivasankaran Rajamanickam, code maintenance: Jon Vogel) and the Center for Advanced Systems Understanding, an institute of the Helmholtz-Zentrum Dresden-Rossendorf (Scientific supervisor: Attila Cangi, code maintenance: Lenz Fiedler).

MALA publications

MALA has been employed in various publications, showcasing its versatility and effectiveness.

Contents