Determining usefulness of machine learning in materials discovery using simulated research landscapes.

When existing experimental data are combined with machine learning (ML) to predict the performance of new materials, the data acquisition bias determines ML usefulness and the prediction accuracy. In this context, the following two conditions are highly common: (i) constructing new unbiased data sets is too expensive and the global knowledge effectively does not change by performing a limited number of novel measurements; (ii) the performance of the material depends on a limited number of physical parameters, much smaller than the range of variables that can be changed, albeit such parameters are unknown or not measurable. To determine the usefulness of ML under these conditions, we introduce the concept of simulated research landscapes, which describe how datasets of arbitrary complexity evolve over time. Simulated research landscapes allow us to use different discovery strategies to compare standard materials exploration with ML-guided explorations, i.e. we can measure quantitatively the benefit of using a specific ML model. We show that there is a window of opportunity to obtain a significant benefit from ML-guided strategies. The adoption of ML can take place too soon (not enough information to find patterns) or too late (dense datasets only allow for negligible ML benefit), and the adoption of ML can even slow down the discovery process in some cases. We offer a qualitative guide on when ML can accelerate the discovery of new best-performing materials in a field under specific conditions. The answer in each case depends on factors like data dimensionality, corrugation and data collection strategy. We consider how these factors may affect the ML prediction capabilities and discuss some general trends.

Prominent out-of-plane diffraction in helium scattering from a methyl-terminated Si(111) surface.

Due to their electrochemical and oxidative stability, organic-terminated semiconductor surfaces are well suited to applications in, for example, photoelectrodes and electrochemical cells, which explains the lively interest in their detailed characterization. Helium atom scattering (HAS) is a useful tool to carry out such characterization. Here, we have simulated HAS in He/CH3-Si(111) based on density functional theory (DFT) potential energy surfaces (PESs) and multi-configuration time-dependent Hartree (MCTDH) dynamics. Our analysis of HAS shows that most diffraction taking place in this system corresponds to high-order out-of-plane peaks. This is a general trend that does not depend on the specific features of the simulations, such as the inclusion or not of the van der Waals long-range effects. This is the first and only He-surface system for which such huge out-of-plane diffraction has been described. This striking theoretical finding should encourage new experimental developments to confirm this previously unreported effect.

Experimental and theoretical study of rotationally inelastic diffraction of H2(D2) from methyl-terminated Si(111).

Fundamental details concerning the interaction between H2 and CH3-Si(111) have been elucidated by the combination of diffractive scattering experiments and electronic structure and scattering calculations. Rotationally inelastic diffraction (RID) of H2 and D2 from this model hydrocarbon-decorated semiconductor interface has been confirmed for the first time via both time-of-flight and diffraction measurements, with modest j = 0 → 2 RID intensities for H2 compared to the strong RID features observed for D2 over a large range of kinematic scattering conditions along two high-symmetry azimuthal directions. The Debye-Waller model was applied to the thermal attenuation of diffraction peaks, allowing for precise determination of the RID probabilities by accounting for incoherent motion of the CH3-Si(111) surface atoms. The probabilities of rotationally inelastic diffraction of H2 and D2 have been quantitatively evaluated as a function of beam energy and scattering angle, and have been compared with complementary electronic structure and scattering calculations to provide insight into the interaction potential between H2 (D2) and hence the surface charge density distribution. Specifically, a six-dimensional potential energy surface (PES), describing the electronic structure of the H2(D2)/CH3-Si(111) system, has been computed based on interpolation of density functional theory energies. Quantum and classical dynamics simulations have allowed for an assessment of the accuracy of the PES, and subsequently for identification of the features of the PES that serve as classical turning points. A close scrutiny of the PES reveals the highly anisotropic character of the interaction potential at these turning points. This combination of experiment and theory provides new and important details about the interaction of H2 with a hybrid organic-semiconductor interface, which can be used to further investigate energy flow in technologically relevant systems.

Limitations of machine learning models when predicting compounds with completely new chemistries: possible improvements applied to the discovery of new non-fullerene acceptors.

We try to determine if machine learning (ML) methods, applied to the discovery of new materials on the basis of existing data sets, have the power to predict completely new classes of compounds (extrapolating) or perform well only when interpolating between known materials. We introduce the leave-one-group-out cross-validation, in which the ML model is trained to explicitly perform extrapolations of unseen chemical families. This approach can be used across materials science and chemistry problems to improve the added value of ML predictions, instead of using extrapolative ML models that were trained with a regular cross-validation. We consider as a case study the problem of the discovery of non-fullerene acceptors because novel classes of acceptors are naturally classified into distinct chemical families. We show that conventional ML methods are not useful in practice when attempting to predict the efficiency of a completely novel class of materials. The approach proposed in this work increases the accuracy of the predictions to enable at least the categorization of materials with a performance above and below the median value.

Data-Driven Analysis of Hole-Transporting Materials for Perovskite Solar Cells Performance.

We have created a dataset of 269 perovskite solar cells, containing information about their perovskite family, cell architecture, and multiple hole-transporting materials features, including fingerprints, additives, and structural and electronic features. We propose a predictive machine learning model that is trained on these data and can be used to screen possible candidate hole-transporting materials. Our approach allows us to predict the performance of perovskite solar cells with reasonable accuracy and is able to successfully identify most of the top-performing and lowest-performing hole-transporting materials in the dataset. We discuss the effect of data biases on the distribution of perovskite families/architectures on the model's accuracy and offer an analysis with a subset of the data to accurately study the effect of the hole-transporting material on the solar cell performance. Finally, we discuss some chemical fragments, like arylamine and aryloxy groups, which present a relatively large positive correlation with the efficiency of the cell, whereas other groups, like thiophene groups, display a negative correlation with power conversion efficiency (PCE).

High-throughput virtual screening for organic electronics: a comparative study of alternative strategies.

We present a review of the field of high-throughput virtual screening for organic electronics materials focusing on the sequence of methodological choices that determine each virtual screening protocol. These choices are present in all high-throughput virtual screenings and addressing them systematically will lead to optimised workflows and improve their applicability. We consider the range of properties that can be computed and illustrate how their accuracy can be determined depending on the quality and size of the experimental datasets. The approaches to generate candidates for virtual screening are also extremely varied and their relative strengths and weaknesses are discussed. The analysis of high-throughput virtual screening is almost never limited to the identification of top candidates and often new patterns and structure-property relations are the most interesting findings of such searches. The review reveals a very dynamic field constantly adapting to match an evolving landscape of applications, methodologies and datasets.

Performance of van der Waals DFT approaches for helium diffraction on metal surfaces.

The ability of the different approaches proposed to date to include the effects of van der Waals (vdW) dispersion forces in density functional theory (DFT) is currently under debate. Here, we used the diffraction of He on a Ru(0 0 0 1) surface as a challenging benchmark system to analyze the suitability of several representative approaches, from the ones correcting the exchange-correlation generalized gradient approximation (GGA) functional, to the ones correcting the DFT energies through pairwise-based methods. To perform our analysis, we have built seven continuous potential energy surfaces (PESs) and carried out quantum dynamics simulations using a multi-configuration time-dependent Hartree method. Our analysis reveals that standard DFT within the PBE-GGA framework, although it overestimates diffraction probabilities, yields the best results in comparison with available experimental measurements. On the other hand, although several of the existing vdW DFT approaches yield physisorption wells in very good agreement with experiment, they all seem to overestimate the long-distance corrugation of the PES, the region probed by He scattering, resulting in a large overestimation of diffraction probabilities.