Bayesian inference of atomistic structure in functional materials

Abstract

Tailoring the functional properties of advanced organic/inorganic heterogeneous devices to their intended technological applications requires knowledge and control of the microscopic structure inside the device. Atomistic quantum mechanical simulation methods deliver accurate energies and properties for individual configurations, however, finding the most favourable configurations remains computationally prohibitive. We propose a 'building block'-based Bayesian Optimization Structure Search (BOSS) approach for addressing extended organic/inorganic interface problems and demonstrate its feasibility in a molecular surface adsorption study. In BOSS, a Bayesian model identifies material energy landscapes in an accelerated fashion from atomistic configurations sampled during active learning. This allowed us to identify several most favorable molecular adsorption configurations for C60 on the (101) surface of TiO2 anatase and clarify the key molecule-surface interactions governing structural assembly. Inferred structures were in good agreement with detailed experimental images of this surface adsorbate, demonstrating good predictive power of BOSS and opening the route towards large-scale surface adsorption studies of molecular aggregates and films.

Data‐Driven Materials Science: Status, Challenges, and Perspectives

“…[223][224][225][226][227][228] restricts their usability for large datasets because the time taken by a brute force inversion scales as n ( ) 3 with the dataset size n. Other models like neural networks can handle larger datasets since they can be trained by using small batches of the training data, and their performance can be monitored during training. At the other end of the spectrum we find powerful tools for small datasets such as, regression with Gaussian processes and Bayesian optimization, [18,257] the extraction of effective materials descriptors with subgroup discovery [258] or compressed sensing as done in, e.g., the least absolute shrinkage and selection operator (LASSO) or the sure independence screening and sparsifying operator (SISSO). [215,216] Also some forms of input are better suited for certain models.…”

Rocketsled: a software library for optimizing high-throughput computational searches

“…A previous study identified stable structures for stage-I and stage-II lithium-graphite intercalation compounds using 4%-6% of the calculations required to explore the entire search space, a combinatorial problem containing more than 16.7 million total possibilities [51]. Others have found adaptive design can accelerate searches-particularly those where exhaustive computation would be problematic-for a variety of applications including novel crystalline interfaces [56], elastic properties [52], high-pressure Mg-silicate phases [47], ultra-low thermal conductivity structures [53], inorganic/organic molecular interfaces [54], layered materials [50], stable carbide/nitrides [57], piezoelectrics [34], Poisson-Schrödinger simulations of LEDs [48], and stable crystal structures [55] for Y 2 Co 17 . In this study, we demonstrated two additional proofs-of-concept searching for superhard materials and photocatalysts.…”

“…The growing toolkit for running high-throughput calculations [39][40][41][42], the potentially immense search space (e.g. often more than 10 10 viable compounds [43]), and the growing number of studies using adaptive design in the materials domain to guide both simulations [34,[44][45][46][47][48][49][50][51][52][53][54][55][56][57][58][59] as well as experiments [44,[60][61][62][63] makes computational materials design an exciting field to explore with Rocketsled. In the two following case studies, we demonstrate the applicability of Rocketsled to adaptive design for materials discovery.…”

Machine learning for the modeling of interfaces in energy storage and conversion materials

“…A very successful example is the Gaussian Approximation Potential (GAP) by Bartók et al that is based on Gaussian process regression [38]. Various other MLP approaches and implementations have been developed in recent years [51][52][53][54][55][56][57][58][59].…”