On-the-Fly Active Learning of Interatomic Potentials for Large-Scale Atomistic Simulations.

The on-the-fly generation of machine-learning force fields by active-learning schemes attracts a great deal of attention in the community of atomistic simulations. The algorithms allow the machine to self-learn an interatomic potential and construct machine-learned models on the fly during simulations. State-of-the-art query strategies allow the machine to judge whether new structures are out of the training data set or not. Only when the machine judges the necessity of updating the data set with the new structures are first-principles calculations carried out. Otherwise, the yet available machine-learned model is used to update the atomic positions. In this manner, most of the first-principles calculations are bypassed during training, and overall, simulations are accelerated by several orders of magnitude while retaining almost first-principles accuracy. In this Perspective, after describing essential components of the active-learning algorithms, we demonstrate the power of the schemes by presenting recent applications.

Li-ion diffusion in Li intercalated graphite C6Li and C12Li probed by µ+SR.

In order to study a diffusive behavior of Li+ in Li intercalated graphites, we have measured muon spin relaxation (µ+SR) spectra for C6Li and C12Li synthesized with an electrochemical reaction between Li and graphite in a Li-ion battery. For both compounds, it was found that Li+ ions start to diffuse above 230 K and the diffusive behavior obeys a thermal activation process. The activation energy (Ea) for C6Li is obtained as 270(5) meV, while Ea = 170(20) meV for C12Li. Assuming a jump diffusion of Li+ in the Li layer of C6Li and C12Li, a self-diffusion coefficient DLi at 310 K was estimated as 7.6(3) × 10-11 (cm2 s-1) in C6Li and 14.6(4) × 10-11 (cm2 s-1) in C12Li.

Formation of novel transition metal hydride complexes with ninefold hydrogen coordination.

Ninefold coordination of hydrogen is very rare, and has been observed in two different hydride complexes comprising rhenium and technetium. Herein, based on a theoretical/experimental approach, we present evidence for the formation of ninefold H- coordination hydride complexes of molybdenum ([MoH9]3-), tungsten ([WH9]3-), niobium ([NbH9]4-) and tantalum ([TaH9]4-) in novel complex transition-metal hydrides, Li5MoH11, Li5WH11, Li6NbH11 and Li6TaH11, respectively. All of the synthesized materials are insulated with band gaps of approximately 4 eV, but contain a sufficient amount of hydrogen to cause the H 1s-derived states to reach the Fermi level. Such hydrogen-rich materials might be of interest for high-critical-temperature superconductivity if the gaps close under compression. Furthermore, the hydride complexes exhibit significant rotational motions associated with anharmonic librations at room temperature, which are often discussed in relation to the translational diffusion of cations in alkali-metal dodecahydro-closo-dodecaborates and strongly point to the emergence of a fast lithium conduction even at room temperature.

True boundary for the formation of homoleptic transition-metal hydride complexes.

Despite many exploratory studies over the past several decades, the presently known transition metals that form homoleptic transition-metal hydride complexes are limited to the Groups 7-12. Here we present evidence for the formation of Mg3 CrH8 , containing the first Group 6 hydride complex [CrH7 ](5-) . Our theoretical calculations reveal that pentagonal-bipyramidal H coordination allows the formation of σ-bonds between H and Cr. The results are strongly supported by neutron diffraction and IR spectroscopic measurements. Given that the Group 3-5 elements favor ionic/metallic bonding with H, along with the current results, the true boundary for the formation of homoleptic transition-metal hydride complexes should be between Group 5 and 6. As the H coordination number generally tends to increase with decreasing atomic number of transition metals, the revised boundary suggests high potential for further discovery of hydrogen-rich materials that are of both technological and fundamental interest.

A novel inorganic solid state ion conductor for rechargeable Mg batteries.

The conduction of the bivalent magnesium cation, Mg(2+), in an inorganic solid state material, Mg(BH4)(NH2), has been demonstrated. This material exhibited a high ionic conductivity of 10(-6) S cm(-1) at 150 °C, and its electrochemical window was estimated to be approximately 3 V using cyclic voltammetry.

Reactive surface area of the Li(x)(Co(1/3)Ni(1/3)Mn(1/3))O2 electrode determined by µ(+)SR and electrochemical measurements.

The self-diffusion coefficient of Li(+) ions (D(Li)) in the positive electrode material Li(x)(Co(1/3)Ni(1/3)Mn(1/3))O2 has been estimated by muon-spin relaxation (µ(+)SR) using powder samples with x = 1-0.49, which were prepared by an electrochemical reaction in a Li-ion battery. Here, since the implanted muons sense a slight change in the internal magnetic field due to Li-diffusion, µ(+)SR provides an intrinsic D(Li) through the temperature dependence of the nuclear field fluctuation rate (ν) [Sugiyama et al., Phys. Rev. Lett., 2009, 103, 147601]. Both D(Li) at 300 K and activation energy (E(a)) were estimated to be ∼2.9 × 10(-12) cm(2) s(-1) and 0.074 eV for the x = 1 sample, ∼11.0 × 10(-12) cm(2) s(-1) and 0.097 eV for x = 0.70, and ∼8.9 × 10(-12) cm(2) s(-1) and 0.062 eV for x = 0.49, assuming that the diffusing Li(+) ions mainly jump from a regular occupied site to a regular vacant site. The estimated D(Li) was smaller by roughly one order of magnitude than those for Li(x)CoO2 in the whole x range measured. Furthermore, by making comparison with D(Li) obtained by electrochemical measurements, the reactive surface area of the Li(x)(Co(1/3)Ni(1/3)Mn(1/3))O2 electrode in a liquid electrolyte was found to strongly depend on x particularly at x > 0.8.

Hydrogen absorption and desorption by the Li-Al-N-H system.

Lithium hexahydridoaluminate Li(3)AlH(6) and lithium amide LiNH(2) with 1:2 molar ratio were mechanically milled, yielding a Li-Al-N-H system. LiNH(2) destabilized Li(3)AlH(6) during the dehydrogenation process of Li(3)AlH(6), because the dehydrogenation starting temperature of the Li-Al-N-H system was lower than that of Li(3)AlH(6). Temperature-programmed desorption scans of the Li-Al-N-H system indicated that a large amount of hydrogen (6.9 wt %) can be released between 370 and 773 K. After initial H(2) desorption, the H(2) absorption and the desorption capacities of the Li-Al-N-H system with a nano-Ni catalyst exhibited 3-4 wt % at 10-0.004 MPa and 473-573 K, while the capacities of the system without the catalyst were 1-2 wt %. The remarkably increased capacity was due to the fact that the kinetics was improved by addition of the nano-Ni catalyst.