Simulation of Screw Dislocations in SmB₆ Topological Insulator Using DFT and Active Learning

Recent research has indicated the potential of dislocations to host exotic quantum states in topological insulators with applications in quantum computing, quantum sensing, acoustic and microwave metamaterials. In topological insulators, dislocations with specific orientations can induce conducting gapless states and enable electron transport through their cores that is unaffected by disorder and has perfectly locked spin and negligible backscattering. Advanced material implementation can be aided by atomistic/quantum simulations that predict the electronic structure in the crystal, to accompany, design and reduce cost and time-demanding experiments. However, acceleration of technologies relying on quantum materials is limited by the computational cost of quantum mechanical simulations and their inefficiency to describe large atomic structures like dislocations at the microscale. The reduction of simulation complexity from first principles to molecular statics to achieve computational efficiency in large systems involves development of interatomic potentials (IPs) suited to simulate the dislocation core in the topological insulator.

This study investigates screw dislocations in SmB₆, a strong topological insulator that exhibits the Kondo effect, with a perfectly bulk insulating nature. Interatomic potentials fit to describe perfectly ordered SmB₆ as well as the system including dislocations do not yet exist. In this work, a data-driven active learning algorithm is developed for the training of a Gaussian Approximation Potential (GAP) to model the interatomic potential from DFT information, in a perfectly ordered system and a system including a screw dislocation. The interatomic potential is trained iteratively in the Quantum mechanics and Interatomic Potentials (QUIP) software until its convergence as derived from the GAP quantification of uncertainty. The new IP is employed by a molecular statics simulation in LAMMPS to describe the structure of the dislocation core and to ensure the core is relaxed under the trained IP; otherwise additional training is done on representative high uncertainty atom structures based on DFT data. The initial dislocation in SmB₆ is generated using the linear elasticity-based Peierls-Nabarro solution for a screw dislocation with given Burgers vector, at a circular slab with a fixed boundary condition. DFT simulations are performed in Quantum Espresso, employing the Perdew-Burke-Ernzerhof (PBE) functional, using the Generalized Gradient Approximation (GGA) for the calculations as well as the pseudopotentials. The molecular statics simulation that uses the trained GAP and yields the relaxed dislocation core structure will predict the elastic constants, lattice parameter, and mechanical properties of the structure.

This study constitutes an exploration of strategy on atomistic simulations for dislocations in the exotic SmB₆ system, where limited computational experience exists, as well as a demonstration of utility in coupling traditional simulation approaches with active learning. The results will aid in the investigation of electronic properties of SmB₆ as a topological insulator and determine the applicability of its microstructural features in the manufacturing of quantum devices.