Machine Learning Force Fields.

In recent years, the use of machine learning (ML) in computational chemistry has enabled numerous advances previously out of reach due to the computational complexity of traditional electronic-structure methods. One of the most promising applications is the construction of ML-based force fields (FFs), with the aim to narrow the gap between the accuracy of ab initio methods and the efficiency of classical FFs. The key idea is to learn the statistical relation between chemical structure and potential energy without relying on a preconceived notion of fixed chemical bonds or knowledge about the relevant interactions. Such universal ML approximations are in principle only limited by the quality and quantity of the reference data used to train them. This review gives an overview of applications of ML-FFs and the chemical insights that can be obtained from them. The core concepts underlying ML-FFs are described in detail, and a step-by-step guide for constructing and testing them from scratch is given. The text concludes with a discussion of the challenges that remain to be overcome by the next generation of ML-FFs.

Improving molecular force fields across configurational space by combining supervised and unsupervised machine learning.

The training set of atomic configurations is key to the performance of any Machine Learning Force Field (MLFF) and, as such, the training set selection determines the applicability of the MLFF model for predictive molecular simulations. However, most atomistic reference datasets are inhomogeneously distributed across configurational space (CS), and thus, choosing the training set randomly or according to the probability distribution of the data leads to models whose accuracy is mainly defined by the most common close-to-equilibrium configurations in the reference data. In this work, we combine unsupervised and supervised ML methods to bypass the inherent bias of the data for common configurations, effectively widening the applicability range of the MLFF to the fullest capabilities of the dataset. To achieve this goal, we first cluster the CS into subregions similar in terms of geometry and energetics. We iteratively test a given MLFF performance on each subregion and fill the training set of the model with the representatives of the most inaccurate parts of the CS. The proposed approach has been applied to a set of small organic molecules and alanine tetrapeptide, demonstrating an up to twofold decrease in the root mean squared errors for force predictions on non-equilibrium geometries of these molecules. Furthermore, our ML models demonstrate superior stability over the default training approaches, allowing reliable study of processes involving highly out-of-equilibrium molecular configurations. These results hold for both kernel-based methods (sGDML and GAP/SOAP models) and deep neural networks (SchNet model).

Challenges for machine learning force fields in reproducing potential energy surfaces of flexible molecules.

Dynamics of flexible molecules are often determined by an interplay between local chemical bond fluctuations and conformational changes driven by long-range electrostatics and van der Waals interactions. This interplay between interactions yields complex potential-energy surfaces (PESs) with multiple minima and transition paths between them. In this work, we assess the performance of the state-of-the-art Machine Learning (ML) models, namely, sGDML, SchNet, Gaussian Approximation Potentials/Smooth Overlap of Atomic Positions (GAPs/SOAPs), and Behler-Parrinello neural networks, for reproducing such PESs, while using limited amounts of reference data. As a benchmark, we use the cis to trans thermal relaxation in an azobenzene molecule, where at least three different transition mechanisms should be considered. Although GAP/SOAP, SchNet, and sGDML models can globally achieve a chemical accuracy of 1 kcal mol-1 with fewer than 1000 training points, predictions greatly depend on the ML method used and on the local region of the PES being sampled. Within a given ML method, large differences can be found between predictions of close-to-equilibrium and transition regions, as well as for different transition mechanisms. We identify key challenges that the ML models face mainly due to the intrinsic limitations of commonly used atom-based descriptors. All in all, our results suggest switching from learning the entire PES within a single model to using multiple local models with optimized descriptors, training sets, and architectures for different parts of the complex PES.

Molecular force fields with gradient-domain machine learning: Construction and application to dynamics of small molecules with coupled cluster forces.

We present the construction of molecular force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of molecular conformations extracted from ab initio molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic forces for these conformations can be computed with high-level wavefunction-based approaches, such as the "gold standard" coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.

Perturbed path integrals in imaginary time: Efficiently modeling nuclear quantum effects in molecules and materials.

Nuclear quantum effects (NQE), which include both zero-point motion and tunneling, exhibit quite an impressive range of influence over the equilibrium and dynamical properties of molecules and materials. In this work, we extend our recently proposed perturbed path-integral (PPI) approach for modeling NQE in molecular systems [I. Poltavsky and A. Tkatchenko, Chem. Sci. 7, 1368 (2016)], which successfully combines the advantages of thermodynamic perturbation theory with path-integral molecular dynamics (PIMD), in a number of important directions. First, we demonstrate the accuracy, performance, and general applicability of the PPI approach to both molecules and extended (condensed-phase) materials. Second, we derive a series of estimators within the PPI approach to enable calculations of structural properties such as radial distribution functions (RDFs) that exhibit rapid convergence with respect to the number of beads in the PIMD simulation. Finally, we introduce an effective nuclear temperature formalism within the framework of the PPI approach and demonstrate that such effective temperatures can be an extremely useful tool in quantitatively estimating the "quantumness" associated with different degrees of freedom in the system as well as providing a reliable quantitative assessment of the convergence of PIMD simulations. Since the PPI approach only requires the use of standard second-order imaginary-time PIMD simulations, these developments enable one to include a treatment of NQE in equilibrium thermodynamic properties (such as energies, heat capacities, and RDFs) with the accuracy of higher-order methods but at a fraction of the computational cost, thereby enabling first-principles modeling that simultaneously accounts for the quantum mechanical nature of both electrons and nuclei in large-scale molecules and materials.

Quantum tunneling of thermal protons through pristine graphene.

Engineering of atomically thin membranes for hydrogen isotope separation is an actual challenge which has a broad range of applications. Recent experiments [M. Lozada-Hidalgo et al., Science 351, 68 (2016)] unambiguously demonstrate an order-of-magnitude difference in permeabilities of graphene-based membranes to protons and deuterons at ambient conditions, making such materials promising for novel separation technologies. Here we demonstrate that the permeability mechanism in such systems changes from quantum tunneling for protons to quasi-classical transport for heavier isotopes. Quantum nuclear effects exhibit large temperature and mass dependence, modifying the Arrhenius activation energy and Arrhenius prefactor for protons by more than 0.5 eV and by seven orders of magnitude correspondingly. Our findings not only shed light on the separation process for hydrogen isotope ions passing through pristine graphene but also offer new insights for controlling ion transport mechanisms in nanostructured separation membranes by manipulating the shape of the barrier and transport process conditions.

Stability of functionalized platform molecules on Au(111).

Trioxatriangulenium (TOTA) platform molecules were functionalized with methyl, ethyl, ethynyl, propynyl, and hydrogen and sublimated onto Au(111) surfaces. Low-temperature scanning tunneling microscopy data reveal that >99% of ethyl-TOTA and methyl-TOTA remain intact, whereas 60% of H-TOTA and >99% of propynyl-TOTA and ethynyl-TOTA decompose. The observed tendency toward fragmentation on Au(111) is opposite to the sequence of gas-phase stabilities of the molecules. Although Au(111) is the noblest of all metal surfaces, the binding energies of the decomposition products to Au(111) destabilize the functionalized platforms by 2 to 3.9 eV (190-370 kJ/mol) and even render some of them unstable as revealed by density functional theory calculations. Van der Waals forces are important, as they drive the adsorption of the platform molecules.

Thermal and Electronic Fluctuations of Flexible Adsorbed Molecules: Azobenzene on Ag(111).

We investigate the thermal and electronic collective fluctuations that contribute to the finite-temperature adsorption properties of flexible adsorbates on surfaces on the example of the molecular switch azobenzene C_{12}H_{10}N_{2} on the Ag(111) surface. Using first-principles molecular dynamics simulations, we obtain the free energy of adsorption that accurately accounts for entropic contributions, whereas the inclusion of many-body dispersion interactions accounts for the electronic correlations that govern the adsorbate binding. We find the adsorbate properties to be strongly entropy driven, as can be judged by a kinetic molecular desorption prefactor of 10^{24} s^{-1} that largely exceeds previously reported estimates. We relate this effect to sizable fluctuations across structural and electronic observables. A comparison of our calculations to temperature-programed desorption measurements demonstrates that finite-temperature effects play a dominant role for flexible molecules in contact with polarizable surfaces, and that recently developed first-principles methods offer an optimal tool to reveal novel collective behavior in such complex systems.

Explainable chemical artificial intelligence from accurate machine learning of real-space chemical descriptors.

Machine-learned computational chemistry has led to a paradoxical situation in which molecular properties can be accurately predicted, but they are difficult to interpret. Explainable AI (XAI) tools can be used to analyze complex models, but they are highly dependent on the AI technique and the origin of the reference data. Alternatively, interpretable real-space tools can be employed directly, but they are often expensive to compute. To address this dilemma between explainability and accuracy, we developed SchNet4AIM, a SchNet-based architecture capable of dealing with local one-body (atomic) and two-body (interatomic) descriptors. The performance of SchNet4AIM is tested by predicting a wide collection of real-space quantities ranging from atomic charges and delocalization indices to pairwise interaction energies. The accuracy and speed of SchNet4AIM breaks the bottleneck that has prevented the use of real-space chemical descriptors in complex systems. We show that the group delocalization indices, arising from our physically rigorous atomistic predictions, provide reliable indicators of supramolecular binding events, thus contributing to the development of Explainable Chemical Artificial Intelligence (XCAI) models.

Force Field Analysis Software and Tools (FFAST): Assessing Machine Learning Force Fields under the Microscope.

As the sophistication of machine learning force fields (MLFF) increases to match the complexity of extended molecules and materials, so does the need for tools to properly analyze and assess the practical performance of MLFFs. To go beyond average error metrics and into a complete picture of a model's applicability and limitations, we developed FFAST (force field analysis software and tools): a cross-platform software package designed to gain detailed insights into a model's performance and limitations, complete with an easy-to-use graphical user interface. The software allows the user to gauge the performance of any molecular force field,─such as popular state-of-the-art MLFF models, ─ on various popular data set types, providing general prediction error overviews, outlier detection mechanisms, atom-projected errors, and more. It has a 3D visualizer to find and picture problematic configurations, atoms, or clusters in a large data set. In this paper, the example of the MACE and NequIP models is used on two data sets of interest [stachyose and docosahexaenoic acid (DHA)]─to illustrate the use cases of the software. With this, it was found that carbons and oxygens involved in or near glycosidic bonds inside the stachyose molecule present increased prediction errors. In addition, prediction errors on DHA rise as the molecule folds, especially for the carboxylic group at the edge of the molecule. We emphasize the need for a systematic assessment of MLFF models for ensuring their successful application to the study of dynamics of molecules and materials.

Efficient interatomic descriptors for accurate machine learning force fields of extended molecules.

Machine learning force fields (MLFFs) are gradually evolving towards enabling molecular dynamics simulations of molecules and materials with ab initio accuracy but at a small fraction of the computational cost. However, several challenges remain to be addressed to enable predictive MLFF simulations of realistic molecules, including: (1) developing efficient descriptors for non-local interatomic interactions, which are essential to capture long-range molecular fluctuations, and (2) reducing the dimensionality of the descriptors to enhance the applicability and interpretability of MLFFs. Here we propose an automatized approach to substantially reduce the number of interatomic descriptor features while preserving the accuracy and increasing the efficiency of MLFFs. To simultaneously address the two stated challenges, we illustrate our approach on the example of the global GDML MLFF. We found that non-local features (atoms separated by as far as 15 Å in studied systems) are crucial to retain the overall accuracy of the MLFF for peptides, DNA base pairs, fatty acids, and supramolecular complexes. Interestingly, the number of required non-local features in the reduced descriptors becomes comparable to the number of local interatomic features (those below 5 Å). These results pave the way to constructing global molecular MLFFs whose cost increases linearly, instead of quadratically, with system size.

Machine Learning Force Fields: Recent Advances and Remaining Challenges.

In chemistry and physics, machine learning (ML) methods promise transformative impacts by advancing modeling and improving our understanding of complex molecules and materials. Each ML method comprises a mathematically well-defined procedure, and an increasingly larger number of easy-to-use ML packages for modeling atomistic systems are becoming available. In this Perspective, we discuss the general aspects of ML techniques in the context of creating ML force fields. We describe common features of ML modeling and quantum-mechanical approximations, so-called global and local ML models, and the physical differences behind these two classes of approaches. Finally, we describe the recent developments and emerging directions in the field of ML-driven molecular modeling. This Perspective aims to inspire interdisciplinary collaborations crossing the borders between physical chemistry, chemical physics, computer science, and data science.

Accurate Description of Nuclear Quantum Effects with High-Order Perturbed Path Integrals (HOPPI).

Imaginary time path-integral (PI) simulations that account for nuclear quantum effects (NQE) beyond the harmonic approximation are increasingly employed together with modern electronic-structure calculations. Existing PI methods are applicable to molecules, liquids, and solids; however, the computational cost of such simulations increases dramatically with decreasing temperature. To address this challenge, here, we propose to combine high-order PI factorization with perturbation theory (PT). Already for conventional second-order PI simulations, the PT ansatz increases the accuracy 2-fold compared to fourth-order schemes with the same settings. In turn, applying PT to high-order path integrals (HOPI) further improves the efficiency of simulations for molecular and condensed matter systems especially at low temperatures. We present results for bulk liquid water, the aspirin molecule, and the CH5+ molecule. Perturbed HOPI simulations remain both efficient and accurate down to 20 K and provide a convenient method to estimate the convergence of quantum-mechanical observables.

Machine learning of accurate energy-conserving molecular force fields.

Using conservation of energy-a fundamental property of closed classical and quantum mechanical systems-we develop an efficient gradient-domain machine learning (GDML) approach to construct accurate molecular force fields using a restricted number of samples from ab initio molecular dynamics (AIMD) trajectories. The GDML implementation is able to reproduce global potential energy surfaces of intermediate-sized molecules with an accuracy of 0.3 kcal mol-1 for energies and 1 kcal mol-1 Å̊-1 for atomic forces using only 1000 conformational geometries for training. We demonstrate this accuracy for AIMD trajectories of molecules, including benzene, toluene, naphthalene, ethanol, uracil, and aspirin. The challenge of constructing conservative force fields is accomplished in our work by learning in a Hilbert space of vector-valued functions that obey the law of energy conservation. The GDML approach enables quantitative molecular dynamics simulations for molecules at a fraction of cost of explicit AIMD calculations, thereby allowing the construction of efficient force fields with the accuracy and transferability of high-level ab initio methods.

Modeling quantum nuclei with perturbed path integral molecular dynamics.

The quantum nature of nuclear motions plays a vital role in the structure, stability, and thermodynamics of molecules and materials. The standard approach to model nuclear quantum fluctuations in chemical and biological systems is to use path-integral molecular dynamics. Unfortunately, conventional path-integral simulations can have an exceedingly large computational cost due to the need to employ an excessive number of coupled classical subsystems (beads) for quantitative accuracy. Here, we combine perturbation theory with the Feynman-Kac imaginary-time path integral approach to quantum mechanics and derive an improved non-empirical partition function and estimators to calculate converged quantum observables. Our perturbed path-integral (PPI) method requires the same ingredients as the conventional approach, but increases the accuracy and efficiency of path integral simulations by an order of magnitude. Results are presented for the thermodynamics of fundamental model systems, an empirical water model containing 256 water molecules within periodic boundary conditions, and ab initio simulations of nitrogen and benzene molecules. For all of these examples, PPI simulations with 4 to 8 classical beads recover the nuclear quantum contribution to the total energy and heat capacity at room temperature within a 3% accuracy, paving the way toward seamless modeling of nuclear quantum effects in realistic molecules and materials.