Multiscale computations and artificial intelligent models of electrochemical performance in Li‐ion battery materials

Abstract

Li-ion battery (LIB) is widely used as one of renewable energy resources for powerful electronics and electric vehicles. The main challenge in developing next-generation LIB is to further improve the energy density, rate capability, and cycling stability of electrode and electrolyte materials. With the rapid development of computational science, the material design has changed from the traditional trial-and-error approach to integrated database-based computation. Multiscale computational methods and machine learning (ML) not only speed up the development of new materials, but also provide insight into the intrinsic relationship between the microscopic composition and macroscopic performance of materials. In this review, we revealed the correlation of electrochemical activity of LIB materials with the response of energy to variable parameters such as charge-transfer number and Li-ion diffusion coordinates.Based on the structure-property relationship, we reviewed multiscale calculation and ML methods for electrochemical properties in LIB materials. A comparison for various applied methods was outlined to provide a methodselecting reference in LIB material design. It is expected that a deep combination of multiscale computations, experimental data, ML should be more powerful methods for discovering new LIB materials.

Structural Design Principle of Rocksalt Oxides for Li-Excess Cathode Materials

“…The electronic energy convergence criterion and atomic forces were set as 10 –5 eV and 0.02 eV Å –1 , respectively. The redox mechanism upon delithiation was calculated by the Bader charge analysis method − and the magnetic moment from VASP. The supercell program was utilized to sample Li-vacancy arrangements in delithiated structures.…”

Chemical transformations and transport phenomena at interfaces

“…144 Like the air-water interface, this must be achieved through the coupling of various theoretical methods ranging from quantum mechanics, molecular dynamics, to analytical continuum models and machine learning, obviously dependent on the practical tradeoff between computational cost and accuracy. 144,145 But more importantly, this choice should be based on the desired physical observations that one aims to explain or predict, that is, electronic structure, vibrational, or adsorption spectroscopy observables naturally fall into the realm of DFT and Schrödinger's framework; atomic structures and energetics can be obtained using DFT and AIMD, reactive force field and MD, and/or ML energies and ML-driven forces; and concentration profiles that can be predicted through mean-field microkinetics, stochastic kinetic Monte Carlo, analysis of radial distribution functions, and/or analytical continuum models. A basic introduction to the individual simulation methods such as DFT, MD, ML, kinetic models, and continuum models, in the context of electrochemistry, can be found here for students or novices, 146 and thus these basics will not be discussed here.…”

“…Taking the electrochemical solid/liquid interface as an example, the region of interest has length scale of angstroms, nanometers, to micrometers for the Stern layer, diffusion layer, and bulk liquid, respectively, with timescales ranging from femtoseconds for charge transfer to hours when considering cell operation conditions 144 . Like the air–water interface, this must be achieved through the coupling of various theoretical methods ranging from quantum mechanics, molecular dynamics, to analytical continuum models and machine learning, obviously dependent on the practical tradeoff between computational cost and accuracy 144,145 …”

Halide solid-state electrolytes for all-solid-state batteries: structural design, synthesis, environmental stability, interface optimization and challenges

“…64 If traditional trial-and-error approaches are used to screen them one by one, it can't meet the urgent demand for the development of new materials and energy technologies. Luckily, with the rapid development of material informatics, numerous advanced technologies such as articial intelligence, 65,66 machine learning, 67,68 high-throughput screening, 40,69 ab initio molecular dynamics (AIMD), 70,71 density functional theory (DFT), 72 and rst-principles calculations, [73][74][75] are used to guide component screening, structural design, and ion diffusion prediction of crystalline inorganic SSEs.…”