Multi-Fidelity Transfer Learning for Quantum Chemical Data Using A Robust Density Functional Tight Binding Baseline

This container includes datasets of QM7x, Tungsten, and molecules with CCSD(T) level of theory. 1. QM7x (partially) 150k molecules were randomly chosen as the training, validation and test data for this project from [1] - qm7x_DFTB.xyz: includes DFT and DFTB properties labeled as "dft_energy/dft_forces" and "dftb_energy/dftb_energy", respectively. - qm7x_Foundation.xyz: includes properties calculated with MACE-OFF23 model, labeled as "MACE_energy/MACE_forces" 2. Tungsten The dataset and DFT properties were obtained from [2] - tungsten_DFTB.xyz: includes DFT and DFTB properties, labeled as "dft_energy/dft_forces" and "dftb_energy/dftb_forces," respectively. - tungsten_Foundation.xyz: includes properties calculated with the MACE-MP-0 model, labeled as "MACE_energy/MACE_forces." 3. CCSDT The dataset and CCSD(T) properties were obtained from [3]. Details on data processing and training can be found in the manuscript [4]. [1] J. Hoja, L. Medrano Sandonas, B. Ernst, A. Vazquez-Mayagoitia, R. DiStasio Jr., and A. Tkatchenko. QM7-x, a comprehensive dataset of quantum-mechanical properties spanning the chemical space of small organic molecules. Sci. Data, 8(1):43, 2021. [2] W. Szlachta, A. Bartók, and G. Csányi. Accuracy and transferability of gaussian approximation potential models for tungsten. Phys. Rev. B, 90(10):104108, 2014. [3] S. Chmiela, V. Vassilev-Galindo, O. Unke, A. Kabylda, H. Sauceda, A. Tkatchenko, and K. Müller. Accurate global machine learning force fields for molecules with hundreds of atoms, 2022. [4] M. Cui, K. Reuter, J. Margraf. Multi-Fidelity Transfer Learning for Quantum Chemical Data Using A Robust Density Functional Tight Binding Baseline. ChemRxiv. 2024; doi:10.26434/chemrxiv-2024-9734b