Tadah! a Swiss army knife for developing and deployment of machine learning interatomic potentials

The Tadah! code provides a versatile platform for developing and optimizing Machine Learning Interatomic Potentials (MLIPs). By integrating composite descriptors, it allows for a nuanced representation of system interactions, customized with unique cutoff functions and interaction distances. Tadah! supports Bayesian Linear Regression (BLR) and Kernel Ridge Regression (KRR) to enhance model accuracy and uncertainty management. A key feature is its hyperparameter optimization cycle, iteratively refining model architecture to improve transferability. This approach incorporates performance constraints, aligning predictions with experimental and theoretical data. Tadah! provides an interface for LAMMPS, enabling the deployment of MLIPs in molecular dynamics simulations. It is designed for broad accessibility, supporting parallel computations on desktop and HPC systems. Tadah! leverages a modular C++ codebase, utilizing both compile-time and runtime polymorphism for flexibility and efficiency. Neural network support and predefined bonding schemes are potential future developments, and Tadah! remains open to community-driven feature expansion. Comprehensive documentation and command-line tools further streamline the development and application of MLIPs. Program title: Tadah! CPC Library link to program files: https://doi.org/10.17632/vy6y3tjdr3.1 Developer's repository link: https://git.ecdf.ed.ac.uk/tadah Licensing provisions: GPLv3 Programming language: C++ Supplementary material: Installation instructions and usage examples are available in the Tadah! online documentation at https://tadah.readthedocs.io Nature of problem: Atomistic modelling, particularly molecular dynamics, is among the most popular techniques used in physics and chemistry research. Accurate and efficient methods are required to generate forces for such simulations. Quantum mechanical calculation of the electronic structure is the “gold standard” here, but is restricted to relatively small systems. Hence, interatomic potentials have a role in allowing large scale simulations, provided they have adequate accuracy. Over the past two decades, the paradigm has shifted from developing interatomic potentials using physics-informed functional forms to generating machine learning interatomic potentials (MLIPs) with more flexible mathematical forms, albeit lacking physical meaning. MLIPs typically lack physical insights in trained models, requiring comprehensive datasets from methods like density functional theory during model parametrization. While training on energies and derivatives may ensure good interpolation within the training data, it frequently lacks transferability, failing environments unlike those in the training data. At the same time the ability to fit a model to experimental data increasingly seems to be lost. In essence, the MLIP framework includes a feature vector (descriptor/fingerprint) and a regression method (e.g., kernel ridge regression or neural networks). Currently, there are numerous frameworks, but there is no consensus on a superior approach. Competing factors such as accuracy, transferability, efficiency, and dataset requirements influence user decisions. Because of the fragmentation of the methodologies, users frequently need to compile multiple software tools, which then often prove to be not end-user-ready. A niche exists for a more flexible framework within which users can mix and match descriptor and regression methods. This would allow MLIP developers to experiment, test, and rapidly deploy models to production, using the methods best suited to their system, application and desired outputs. It is desirable to provide additional tools for fitting and verification of models. Furthermore, deployment in molecular dynamics (MD) settings, requires effort to integrate the MLIP with third-party MD software. Plugins for standard MD codes should be available alongside MLIP fitting codes. Solution method: Tadah! is a unified program designed for the development of advanced machine learning interatomic potentials using various descriptors, cutoffs, and regression methods. It also allows rapid deployment of these through the MD LAMMPS plugin [1]. It supports mono- and multi-species systems as well as advanced (custom) fitting procedures. The software provides unique pathways to embed prior domain knowledge into a model and offers a range of additional tools for dataset conversion, manipulation, and plotting. Tadah! allows users to incorporate their physical insight by combining multiple descriptors into a single composite descriptor. Each constituent descriptor can be further customized by specifying a unique cutoff function and distance. Additionally, users can define specific chemical species pairs that each constituent descriptor will target, enabling tailored interactions within the model. Tadah! also provides users with a nested fitting procedure that allows the optimization of model hyperparameters with a custom-defined loss function. In principle, the loss function can incorporate any quantity obtainable from MD simulation and be compared against experimental values for scoring. The Tadah! toolkit provides a command-line interface (CLI) binary as well as a C++ API for advanced use. The majority of features are available via the CLI, while the API allows developers and users to rapidly implement and test new descriptors and various regression strategies. It supports seamless integration of new types of descriptors, with an API that enables their addition. We follow the philosophy: “Implement once, use it everywhere across Tadah! and LAMMPS.” The code is open source, and we aim to drive development based on user feedback. We encourage users to request new features and provide feedback. We are also open to individual contributions and collaborations. Additional comments including restrictions and unusual features: Tadah! provides two end-user packages: Tadah!MLIP for the development of MLIPs and Tadah!LAMMPS for deployment in molecular dynamics settings. The Tadah!MLIP software can be built as a desktop version using OpenMP. For users working with large datasets or complex descriptors, a massively parallel version of Tadah!MLIP is available with MPI, designed for HPC architectures. [1] A.P. Thompson et al., LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales, Comput. Phys. Comm. 271 (2022) 10817.