A dataset of high-efficiency atomic orbital data for first-principles calculations of large atomic systems

<p indent="0mm">Numerical Atomic Orbitals (NAOs) are a highly efficient basis set for describing electron wave functions in atomic structures, particularly suitable for electronic structure simulations of large-scale atomic systems (thousands of atoms or more). Due to the complexity of constructing and testing such data, the research community still lacks data products that provide both comprehensive quality control and broad applicability. To address this, this work systematically constructed a NAO dataset covering 68 elements, supporting various pseudopotentials, cutoff parameters (radius/energy), and basis set size configurations. During the construction of the NAO dataset, a hierarchical optimization strategy was adopted: low-precision orbitals with fewer basis functions were optimized first, followed by a gradual progression to high-precision orbitals with more basis functions, thereby meeting the requirements of first-principles calculations at different levels of efficiency and accuracy. Moreover, to demonstrate the reliability of the data and provide a basis for user selection, this dataset includes comprehensive quality assessment data, such as single-crystal equations of state (EOS) fitting data and bond length-energy data for molecular systems. Assessment results show that the mean energy difference between calculations using this NAO dataset and standard plane-wave basis sets for single crystals is minimal, reaching as low as 0.66 meV/atom. This dataset has been adopted by the ABACUS first-principles computational software, significantly enhancing the efficiency and accuracy of complex system simulations. It has broadt application value in fields such as condensed matter physics, materials science, and chemical simulations.