Improving the accuracy of hybrid meta-GGA density functionals by range separation

The Minnesota family of exchange-correlation functionals, which consists of meta generalized gradient approximations (meta-GGAs) and global-hybrid meta-GGAs, has been successful for density functional calculations of molecular structure, properties, and thermochemistry, kinetics, noncovalent interactions, and spectroscopy. Here, we generalize the functional form by using range-separated hybrid meta-GGA exchange. We optimize a functional, called M11, with the new form against a broad database of energetic chemical properties and compare its performance to that of several other functionals, including previous Minnesota functionals. We require the percentage of Hartree-Fock exchange to be 100 at large interelectronic distance, and we find an optimum percentage of 42.8 at short range. M11 has good across-the-board performance and the smallest mean unsigned error over the whole test set of 332 data; it has especially good performance for main-group atomization energies, proton affinities, electron affinities, alkyl bond dissociation energies, barrier heights, noncovalent interaction energies, and charge-transfer electronic excitation.