Search

We present high-quality reference data for two fundamentally important groups of molecular properties related to a compound's utility as a lithium battery electrolyte. The first one is energy changes associated with charge excitations of molecules, namely ionization potential and electron affinity. They were estimated for 7000 randomly chosen molecules with up to 9 non-hydrogen atoms C, N, O, and F (QM9 dataset) using DH-HF, DF-HF-CABS, PNO-LMP2-F12, and PNO-LCCSD(T)-F12 methods as implemented in Molpro software with aug-cc-pVTZ basis set; additionally, we provide the corresponding atomization energies at these levels of theory, as well as CPU time and disk space used during the calculations. The second one is solvation energies for 39 different solvents, which we estimate for 18361 molecules connected to battery design (Electrolyte Genome Project dataset), 309463 randomly chosen molecules with up to 17 non-hydrogen atoms C, N, O, S, and halogens (GDB17 dataset), as well as 88418 amons of ZINC database of commercially available compounds and 37772 amons of GDB17. For these calculations we used the COnductor-like Screening MOdel for Real Solvents (COSMO-RS) method; we additionally provide estimates of gas-phase atomization energies, as well as information about conformers considered during the COSMO-RS calculations, namely coordinates, energies, and dipole moments.

Modern machine learning (ML) models of chemical and materials systems with billions of parameters require vast training datasets and considerable computational efforts. Lightweight kernel or decision tree based methods, however, can be rapidly trained, leading to a considerably lower carbon footprint. We introduce generalized convolutional many-body distribution functionals (cMBDF) as highly compute and data efficient atomic representations for accurate kernels that excel in low-data regimes. Generalizing the MBDF framework, cMBDF encodes local chemical environments in a compact fashion using translationally and rotationally invariant functionals of smooth atom centered Gaussian electron density proxy distributions weighted by interaction potentials. The functional values can be efficiently evaluated by expressing them in terms of convolutions which are calculated via fast Fourier transforms and stored on pre-defined grids. In the generalized form each atomic environment is described using a set of functionals uniformly defined by three integers; many-body, derivative, weighting orders. Irrespective of size/composition, cMBDF atomic vectors remain compact and constant in size for a fixed choice of these orders controlling the structural and compositional resolution. While being up to two orders of magnitude more compact than other popular representations, cMBDF is shown to be more accurate for the learning of various quantum properties such as energies, dipole moments, homo-lumo gaps, heat-capacity, polarizability, optimal exact-exchange admixtures and basis-set scaling factors. Applicability for organic and inorganic chemistry is tested as represented by the QM7b, QM9 and VQM24 data sets. Due to its compactness, model training and testing times are reduced from 23 hours to 8 minutes, implying a corresponding reduction in carbon footprint.

We introduce the alchemical harmonic approximation (AHA) of the absolute electronic energy for charge-neutral iso-electronic diatomics at fixed interatomic distance $d_0$. To account for variations in distance, we combine AHA with this Ansatz for the electronic binding potential, $E(d)=(E_{u}-E_s) \left(\frac{E_c-E_s}{E_u-E_s}\right)^{\sqrt{d/d_0}}+E_s$, where $E_u,E_c,E_s$ correspond to the energies of united atom, calibration at $d_0$, and sum of infinitely separated atoms, respectively. Our model covers the entire two-dimensional electronic potential energy surface spanned by distance and difference in nuclear charge from which only one single point (with elements of nuclear charge $Z_1,Z_2$ and distance $d_0$) is drawn to calibrate $E_c$. Using reference data from pbe0/cc-pVDZ, we present numerical evidence for the electronic ground-state of all neutral diatomics with 8, 10, 12, 14 electrons. We assess the validity of our model by comparison to legacy interatomic potentials (Harmonic oscillator, Lennard-Jones, and Morse) within the most relevant range of binding (0.7 - 2.5 A), and find comparable accuracy if restricted to single diatomics, and significantly better predictive power when extrapolating to the entire iso-electronic series. We also investigated $\Delta$-learning of the electronic absolute energy using our model as baseline. This baseline model results in a systematic improvement, effectively reducing training data needs for reaching chemical accuracy by up to an order of magnitude from $\sim$1000 to $\sim$100. By contrast, using AHA+Morse as a baseline hardly leads to any improvement, and sometimes even deteriorates the predictive power. Inferring the energy of unseen CO converges to a prediction error of $\sim$0.1 Ha in direct learning, and $\sim$0.04 Ha with our baseline.

Accurate quantum mechanics based predictions of property trends are so important for materials design and discovery that even inexpensive approximate methods are valuable. We use the Alchemical Integral Transform (AIT) to study multi-electron atoms, and to gain a better understanding of the approximately quadratic behavior of energy differences between iso-electronic atoms in their nuclear charges. Based on this, we arrive at the following simple analytical estimate of energy differences between any two iso-electronic atoms, $\Delta E \approx -(1 + 2\gamma \sqrt{N_e-1})\,\Delta Z\, \bar{Z}$. Here, $\gamma \approx 0.3766 \pm 0.0020$ Ha corresponds to an empirical constant, and $N_e$, $\Delta Z$, and $\bar{Z}$ respectively to electron number, and nuclear charge difference and average. We compare the formula's predictive accuracy using experimental numbers and non-relativistic, numerical results obtained via DFT (pbe0) for the entire periodic table up to Radon. A detailed discussion of the atomic Helium-series is included.

We cast the relation between the chemical composition of a solid-state material and its superconducting critical temperature (Tc) as a statistical learning problem with reduced complexity. Training of query-aware similarity-based ridge regression models on experimental SuperCon data achieve average Tc prediction errors of ~5 K for unseen out-of-sample materials. Two models were trained with one excluding high pressure data in training ("ambient" model) and a second also including high pressure data ("implicit" model). Subsequent utilization of the approach to scan ~153k materials in the Materials Project enables the ranking of candidates by Tc while accounting for thermodynamic stability and small band gap. The ambient model is used to predict stable top three high-Tc candidate materials that include those with large band gaps of LiCuF4 (316 K), Ag2H12S(NO)4 (316 K), and Na2H6PtO6 (315 K). Filtering these candidates for those with small band gaps correspondingly yields LiCuF4 (316 K), Cu2P2O7 (311 K), and Cu3P2H2O9 (307 K).

We study the applicability of the Hammett-inspired product (HIP) Ansatz to model relative substrate binding within homogenous organometallic catalysis, assigning $\sigma$ and $\rho$ to ligands and metals, respectively. Implementing an additive combination (c) rule for obtaining $\sigma$ constants for any ligand pair combination results in a cHIP model that enhances data efficiency in computational ligand tuning. We show its usefulness (i) as a baseline for $\Delta$-machine learning (ML), and (ii) to identify novel catalyst candidates via volcano plots. After testing the combination rule on Hammett constants previously published in the literature, we have generated numerical evidence for the Suzuki-Miyaura (SM) C-C cross-coupling reaction using two synthetic datasets of metallic catalysts (including (10) and (11)-metals Ni, Pd, Pt, and Cu, Ag, Au as well as 96 ligands such as N-heterocyclic carbenes, phosphines, or pyridines). When used as a baseline, $\Delta$-ML prediction errors of relative binding decrease systematically with training set size and reach chemical accuracy ($\sim$1 kcal/mol) for 20k training instances. Employing the individual ligand constants obtained from cHIP, we report relative substrate binding for a novel dataset consisting of 720 catalysts (not part of training data), of which 145 fall into the most promising range on the volcano plot accounting for oxidative addition, transmetalation, and reductive elimination steps. Multiple Ni-based catalysts, e.g. Aphos-Ni-P($t$-Bu)$_3$, are included among these promising candidates, potentially offering dramatic cost savings in experimental applications.

We introduce the Vector-QM24 (VQM24) dataset which comprehensively covers all possible neutral closed shell small organic and inorganic molecules and their conformers that contain up to five heavy atoms (non-hydrogen) consisting of $p$-block elements C, N, O, F, Si, P, S, Cl, Br. This dataset has been systematically generated by considering all combinatorially possible stoichiometries, and graphs (according to Lewis rules), along with all stable conformers identified by GFN2-xTB. We have used density functional theory ($\omega$B97X-D3/cc-pVDZ) to optimize the geometries of 577k conformational isomers corresponding to 258k constitutional isomers consistent with 5,599 unique stoichiometries. Single point diffusion quantum Monte Carlo (DMC@PBE0(ccECP/cc-pVQZ)) energies are reported for the sub-set of all the energetically lowest conformers (10,793 molecules) with up to 4 heavy atoms. Apart from graphs, geometries, rotational constants, and vibrational normal modes, VQM24 includes internal, atomization, electron-electron repulsion, exchange-correlation, dispersion, vibrational frequency, Gibbs free, enthalpy, ZPV and molecular orbital energies; as well as entropy, and heat capacities. Electronic properties include multipole moments (dipole, quadrupole, octupole, hexadecapole), electrostatic potentials at nuclei (aka "alchemical potential"), Mulliken charges, and Kohn-Sham orbitals. Supervised machine learning (ML) models of atomization energies on the energetically lowest conformers for the 258k constitutional isomers indicate a significantly more challenging benchmark than the previously introduced QM9 dataset with none of our models reaching chemical accuracy. VQM24 represents an accurate and unbiased benchmark dataset ideal for assessing the efficiency, accuracy and transferability of quantum ML models of real systems.

We investigate trends in the data-error scaling behavior of machine learning (ML) models trained on discrete combinatorial spaces that are prone-to-mutation, such as proteins or organic small molecules. We trained and evaluated kernel ridge regression machines using variable amounts of computationally generated training data. Our synthetic datasets comprise i) two na\"ive functions based on many-body theory; ii) binding energy estimates between a protein and a mutagenised peptide; and iii) solvation energies of two 6-heavy atom structural graphs. In contrast to typical data-error scaling, our results showed discontinuous monotonic phase transitions during learning, observed as rapid drops in the test error at particular thresholds of training data. We observed two learning regimes, which we call saturated and asymptotic decay, and found that they are conditioned by the level of complexity (i.e. number of mutations) enclosed in the training set. We show that during training on this class of problems, the predictions were clustered by the ML models employed in the calibration plots. Furthermore, we present an alternative strategy to normalize learning curves (LCs) and the concept of mutant based shuffling. This work has implications for machine learning on mutagenisable discrete spaces such as chemical properties or protein phenotype prediction, and improves basic understanding of concepts in statistical learning theory.

Atomic basis sets are widely employed within quantum mechanics based simulations of matter. We introduce a machine learning model that adapts the basis set to the local chemical environment of each atom, prior to the start of self consistent field (SCF) calculations. In particular, as a proof of principle and because of their historic popularity, we have studied the Gaussian type orbitals from the Pople basis set, i.e. the STO-3G, 3-21G, 6-31G and 6-31G*. We adapt the basis by scaling the variance of the radial Gaussian functions leading to contraction or expansion of the atomic orbitals.A data set of optimal scaling factors for C, H, O, N and F were obtained by variational minimization of the Hartree-Fock (HF) energy of the smallest 2500 organic molecules from the QM9 database. Kernel ridge regression based machine learning (ML) prediction errors of the change in scaling decay rapidly with training set size, typically reaching less than 1 % for training set size 2000. Overall, we find systematically lower variance, and consequently the larger training efficiencies, when going from hydrogen to carbon to nitrogen to oxygen. Using the scaled basis functions obtained from the ML model, we conducted HF calculations for the subsequent 30'000 molecules in QM9. In comparison to the corresponding default Pople basis set results we observed improved energetics in up to 99 % of all cases. With respect to the larger basis set 6-311G(2df,2pd), atomization energy errors are lowered on average by ~31, 107, 11, and 11 kcal/mol for STO-3G, 3-21G, 6-31G and 6-31G*, respectively -- with negligible computational overhead. We illustrate the high transferability of adaptive basis sets for larger out-of-domain molecules relevant to addiction, diabetes, pain, aging.

The paper by No\'e et al. [F. No\'e, S. Olsson, J. K\"ohler and H. Wu, Science, 365:6457 (2019)] introduced the concept of Boltzmann Generators (BGs), a deep generative model that can produce unbiased independent samples of many-body systems. They can generate equilibrium configurations from different metastable states, compute relative stabilities between different structures of proteins or other organic molecules, and discover new states. In this commentary, we motivate the necessity for a new generation of sampling methods beyond molecular dynamics, explain the methodology, and give our perspective on the future role of BGs., Comment: 8 pages

Exact exchange contributions are known to crucially affect electronic states, which in turn govern covalent bond formation and breaking in chemical species. Empirically averaging the exact exchange admixture over compositional degrees of freedom, hybrid density functional approximations have been widely successful, yet have fallen short to reach high level quantum chemistry accuracy, primarily due to delocalization errors. We propose to `adaptify` hybrid functionals by generating optimal admixture ratios of exact exchange on the fly, i.e. specifically for any chemical compound, using extremely data efficient quantum machine learning models that carry negligible overhead. The adaptive Perdew-Burke-Ernzerhof based hybrid density functional (aPBE0) is shown to yield atomization energies with sufficient accuracy to effectively cure the infamous spin gap problem in open shell systems, such as carbenes. aPBE0 further improves energetics, electron densities, and HOMO-LUMO gaps in organic molecules drawn from the QM9 and QM7b data set. Obtained with aPBE0 in a large basis, we present a revision of the entire QM9 data set (revQM9) with an estimated quality vastly superior to the original containing on average, stronger covalent binding, larger band-gaps, more localized electron densities, and larger dipole-moments. While aPBE0 is applicable in the equilibrium regime, outstanding limitations include covalent bond dissociation when going beyond the Coulson-Fisher point.

We recently introduced the Alchemical Integral Transform (AIT) enabling the prediction of energy differences, and guessed an Ansatz to parametrize space $\pmb{r}$ in some alchemical change $\lambda$. Here, we present a rigorous derivation of AIT's kernel $\mathcal{K}$ and discuss the parametrization $\pmb{r}(\lambda)$ in $n$ dimensions, i.e. necessary conditions, mathematical freedoms and additional constraints when obtaining it. Analytical expressions for changes in energy spectra and densities are given for a number of systems. Examples include homogeneous potentials like the quantum harmonic oscillator, Hydrogen-like atom, and Dirac well, both for one- and multiparticle cases, and a multiparticle system beyond coordinate scaling for harmonic potentials.

We study alchemical atomic energy partitioning as a method to estimate atomisation energies from atomic contributions which are defined in physically rigorous and general ways through use of the uniform electron gas as a joint reference. We analyze quantitatively the relation between atomic energies and their local environment using a dataset of 1325 organic molecules. The atomic energies are transferable across various molecules, enabling the prediction of atomisation energies with a mean absolute error of 20 kcal/mol - comparable to simple statistical estimates but potentially more robust given their grounding in the physics-based decomposition scheme. A comparative analysis with other decomposition methods highlights its sensitivity to electrostatic variations, underlining its potential as representation of the environment as well as in studying processes like diffusion in solids characterized by significant electrostatic shifts.

For many machine learning applications in science, data acquisition, not training, is the bottleneck even when avoiding experiments and relying on computation and simulation. Correspondingly, and in order to reduce cost and carbon footprint, training data efficiency is key. We introduce minimal multilevel machine learning (M3L) which optimizes training data set sizes using a loss function at multiple levels of reference data in order to minimize a combination of prediction error with overall training data acquisition costs (as measured by computational wall-times). Numerical evidence has been obtained for calculated atomization energies and electron affinities of thousands of organic molecules at various levels of theory including HF, MP2, DLPNO-CCSD(T), DFHFCABS, PNOMP2F12, and PNOCCSD(T)F12, and treating tens with basis sets TZ, cc-pVTZ, and AVTZ-F12. Our M3L benchmarks for reaching chemical accuracy in distinct chemical compound sub-spaces indicate substantial computational cost reductions by factors of $\sim$ 1.01, 1.1, 3.8, 13.8 and 25.8 when compared to heuristic sub-optimal multilevel machine learning (M2L) for the data sets QM7b, QM9$^\mathrm{LCCSD(T)}$, EGP, QM9$^\mathrm{CCSD(T)}_\mathrm{AE}$, and QM9$^\mathrm{CCSD(T)}_\mathrm{EA}$, respectively. Furthermore, we use M2L to investigate the performance for 76 density functionals when used within multilevel learning and building on the following levels drawn from the hierarchy of Jacobs Ladder:~LDA, GGA, mGGA, and hybrid functionals. Within M2L and the molecules considered, mGGAs do not provide any noticeable advantage over GGAs. Among the functionals considered and in combination with LDA, the three on average top performing GGA and Hybrid levels for atomization energies on QM9 using M3L correspond respectively to PW91, KT2, B97D, and $\tau$-HCTH, B3LYP$\ast$(VWN5), TPSSH.

We present an automated data-collection pipeline involving a convolutional neural network and a large language model to extract user-specified tabular data from peer-reviewed literature. The pipeline is applied to 74 reports published between 1957 and 2014 with experimentally-measured oxidation potentials for 592 organic molecules (-0.75 to 3.58 V). After data curation (solvents, reference electrodes, and missed data points), we trained multiple supervised machine learning models reaching prediction errors similar to experimental uncertainty ($\sim$0.2 V). For experimental measurements of identical molecules reported in multiple studies, we identified the most likely value based on out-of-sample machine learning predictions. Using the trained machine learning models, we then estimated oxidation potentials of $\sim$132k small organic molecules from the QM9 data set, with predicted values spanning 0.21 to 3.46 V. Analysis of the QM9 predictions in terms of plausible descriptor-property trends suggests that aliphaticity increases the oxidation potential of an organic molecule on average from $\sim$1.5 V to $\sim$2 V, while an increase in number of heavy atoms lowers it systematically. The pipeline introduced offers significant reductions in human labor otherwise required for conventional manual data collection of experimental results, and exemplifies how to accelerate scientific research through automation.

Optimizing a target function over the space of organic molecules is an important problem appearing in many fields of applied science, but also a very difficult one due to the vast number of possible molecular systems. We propose an Evolutionary Monte Carlo algorithm for solving such problems which is capable of straightforwardly tuning both exploration and exploitation characteristics of an optimization procedure while retaining favourable properties of genetic algorithms. The method, dubbed MOSAiCS (Metropolis Optimization by Sampling Adaptively in Chemical Space), is tested on problems related to optimizing components of battery electrolytes, namely minimizing solvation energy in water or maximizing dipole moment while enforcing a lower bound on the HOMO-LUMO gap; optimization was done over sets of molecular graphs inspired by QM9 and Electrolyte Genome Project (EGP) datasets. MOSAiCS reliably generated molecular candidates with good target quantity values, which were in most cases better than the ones found in QM9 or EGP. While the optimization results presented in this work sometimes required up to $10^{6}$ QM calculations and were thus only feasible thanks to computationally efficient ab initio approximations of properties of interest, we discuss possible strategies for accelerating MOSAiCS using machine learning approaches.

Despite its fundamental importance and widespread use for assessing reaction success in organic chemistry, deducing chemical structures from nuclear magnetic resonance (NMR) measurements has remained largely manual and time consuming. To keep up with the accelerated pace of automated synthesis in self driving laboratory settings, robust computational algorithms are needed to rapidly perform structure elucidations. We analyse the effectiveness of solving the NMR spectra matching task encountered in this inverse structure elucidation problem by systematically constraining the chemical search space, and correspondingly reducing the ambiguity of the matching task. Numerical evidence collected for the twenty most common stoichiometries in the QM9-NMR data base indicate systematic trends of more permissible machine learning prediction errors in constrained search spaces. Results suggest that compounds with multiple heteroatoms are harder to characterize than others. Extending QM9 by $\sim$10 times more constitutional isomers with 3D structures generated by Surge, ETKDG and CREST, we used ML models of chemical shifts trained on the QM9-NMR data to test the spectra matching algorithms. Combining both $^{13}\mathrm{C}$ and $^{1}\mathrm{H}$ shifts in the matching process suggests twice as permissible machine learning prediction errors than for matching based on $^{13}\mathrm{C}$ shifts alone. Performance curves demonstrate that reducing ambiguity and search space can decrease machine learning training data needs by orders of magnitude.

We show that even order contributions to energy differences between any two iso-electronic compounds vanish when using perturbation theory around an averaged electronic reference Hamiltonian. This finding generalizes the previously introduced alchemical chirality concept [von Rudorff, von Lilienfeld, Science Advances 7 (2021)] by lifting the symmetry requirements for transmutating atoms in the iso-electronic reference system. The leading order term corresponds to twice the Hellmann-Feynman derivative evaluated using the electron density of the averaged Hamiltonian. Analogous analysis reveals Mel Levy's formula for relative energies [J. Chem. Phys. 70, 1573 (1979)] to include the first order contribution while overestimating the higher odd order energy contributions by a factor linearly increasing in order. Using density functional theory, we illustrate the predictive power of the leading order term for estimating relative energies among diatomics in the charge-neutral iso-electronic 14 proton series N2, CO, BF, BeNe, LiNa, HeMg, HAl, and the united atom, Si. The framework's potential for the simultaneous exploration of multiple dimensions in chemical space is demonstrated for toluene by evaluating relative energies between all the possible 35 antisymmetric BN doped isomers (dubbed ``alchemical diastereomers''). Based solely on toluene's electron density, necessary to evaluate all the respective Hellmann-Feynman derivatives, mean absolute errors of predicted total potential energy differences between the alchemical diastereomers are on the scale of mHa.

Data-hungry machine learning methods have become a new standard to efficiently navigate chemical compound space for molecular and materials design and discovery. Due to the severe scarcity and cost of high-quality experimental or synthetic simulated training data, however, data-acquisition costs can be considerable. Relying on reasonably accurate approximate baseline labels with low computational complexity represents one of the most effective strategies to curb data-needs, e.g.~through $\Delta$-, transfer-, or multi-fidelity learning. We introduce baseline labels in the form of a generic coarse-graining Hammett Inspired Product (HIP) {\em Ansatz} which generalizes the empirical Hammett equation towards arbitrary systems and properties. Numerical evidence for the applicability of HIP includes solvation free energies of molecules, formation energies of quaternary elpasolite crystals, carbon adsorption energies on heterogeneous catalytic surfaces, HOMO-LUMO gaps of metallorganic complexes, and activation energies for S$_\text{N}$2 reactions. After calibration on the same training sets, HIP yields an effective baseline for improved $\Delta$-quantum machine learning models with superior data-efficiency when compared to previously introduced specialised domain-specific models.

Density functional theory (DFT) plays a pivotal role for the chemical and materials science due to its relatively high predictive power, applicability, versatility and computational efficiency. We review recent progress in machine learning model developments which has relied heavily on density functional theory for synthetic data generation and for the design of model architectures. The general relevance of these developments is placed in some broader context for the chemical and materials sciences. Resulting in DFT based machine learning models with high efficiency, accuracy, scalability, and transferability (EAST), recent progress indicates probable ways for the routine use of successful experimental planning software within self-driving laboratories.

The feature vector mapping used to represent chemical systems is a key factor governing the superior data-efficiency of kernel based quantum machine learning (QML) models applicable throughout chemical compound space. Unfortunately, the most accurate representations require a high dimensional feature mapping, thereby imposing a considerable computational burden on model training and use. We introduce compact yet accurate, linear scaling QML representations based on atomic Gaussian many-body distribution functionals (MBDF), and their derivatives. Weighted density functions (DF) of MBDF values are used as global representations which are constant in size, i.e.~invariant with respect to the number of atoms. We report predictive performance and training data efficiency that is competitive with state of the art for two diverse datasets of organic molecules, QM9 and QMugs. Generalization capability has been investigated for atomization energies, HOMO-LUMO eigenvalues and gap, internal energies at 0 K, zero point vibrational energies, dipole moment norm, static isotropic polarizability, and heat capacity as encoded in QM9. MBDF based QM9 performance lowers the optimal Pareto front spanned between sampling and training cost to compute node minutes,~effectively sampling chemical compound space with chemical accuracy at a sampling rate of $\sim 48$ molecules per core second.

Recent advances in experimental methodology enabled studies of the quantum-state and conformational dependence of chemical reactions under precisely controlled conditions in the gas phase. Here, we generated samples of selected gauche and s-trans 2,3-dibromobutadiene (DBB) by electrostatic deflection in a molecular beam and studied their reaction with Coulomb crystals of laser-cooled $\mathrm{Ca^{+}}$ ions in an ion trap. The rate coefficients for the total reaction were found to strongly depend on both the conformation of DBB and the electronic state of $\mathrm{Ca^{+}}$. In the $\mathrm{(4p)~^{2}P_{1/2}}$ and $\mathrm{(3d)~^{2}D_{3/2}}$ excited states of $\mathrm{Ca^{+}}$, the reaction is capture-limited and faster for the gauche conformer due to long-range ion-dipole interactions. In the $\mathrm{(4s)~^{2}S_{1/2}}$ ground state of $\mathrm{Ca^{+}}$, the reaction rate for s-trans DBB still conforms with the capture limit, while that for gauche DBB is strongly suppressed. The experimental observations were analysed with the help of adiabatic capture theory, ab-initio calculations and reactive molecular dynamics simulations on a machine-learned full-dimensional potential energy surface of the system. The theory yields near-quantitative agreement for s-trans-DBB, but overestimates the reactivity of the gauche-conformer compared to the experiment. The present study points to the important role of molecular geometry even in strongly reactive exothermic systems and illustrates striking differences in the reactivity of individual conformers in gas-phase ion-molecule reactions.

Large machine learning models with improved predictions have become widely available in the chemical sciences. Unfortunately, these models do not protect the privacy necessary within commercial settings, prohibiting the use of potentially extremely valuable data by others. Encrypting the prediction process can solve this problem by double-blind model evaluation and prohibits the extraction of training or query data. However, contemporary ML models based on fully homomorphic encryption or federated learning are either too expensive for practical use or have to trade higher speed for weaker security. We have implemented secure and computationally feasible encrypted machine learning models using oblivious transfer enabling and secure predictions of molecular quantum properties across chemical compound space. However, we find that encrypted predictions using kernel ridge regression models are a million times more expensive than without encryption. This demonstrates a dire need for a compact machine learning model architecture, including molecular representation and kernel matrix size, that minimizes model evaluation costs.

In the past decade, quantum diffusion Monte Carlo (DMC) has been demonstrated to successfully predict the energetics and properties of a wide range of molecules and solids by numerically solving the electronic many-body Schr\"odinger equation. We show that when coupled with quantum machine learning (QML) based surrogate methods the computational burden can be alleviated such that QMC shows clear potential to undergird the formation of high quality descriptions across chemical space. We discuss three crucial approximations necessary to accomplish this: The fixed node approximation, universal and accurate references for chemical bond dissociation energies, and scalable minimal amons set based QML (AQML) models. Numerical evidence presented includes converged DMC results for over one thousand small organic molecules with up to 5 heavy atoms used as amons, and 50 medium sized organic molecules with 9 heavy atoms to validate the AQML predictions. Numerical evidence collected for $\Delta$-AQML models suggests that already modestly sized QMC training data sets of amons suffice to predict total energies with near chemical accuracy throughout chemical space.

We present an intuitive and general analytical approximation estimating the energy of covalent single and double bonds between participating atoms in terms of their respective nuclear charges with just three parameters, $[{E_\text{AB} \approx a - b Z_\text{A} Z_\text{B} + c (Z_\text{A}^{7/3} + Z_\text{B}^{7/3})}]$. The functional form of our expression models an alchemical atomic energy decomposition between participating atoms A and B. After calibration, reasonably accurate bond energy estimates are obtained for hydrogen-saturated diatomics composed of $p$-block elements coming from the same row $2\le n\le 4$ in the periodic table. Corresponding changes in bond energies due to substitution of atom B by C can be obtained via simple formulas. While being of different functional form and origin, our model is as simple and accurate as Pauling's well-known electronegativity model. Analysis indicates that the model's response in covalent bonding to variation in nuclear charge is near-linear -- which is consistent with Hammett's equation.

Conventional kernel-based machine learning models for ab initio potential energy surfaces, while accurate and convenient in small data regimes, suffer immense computational cost as training set sizes increase. We introduce QML-Lightning, a PyTorch package containing GPU-accelerated approximate kernel models, which reduces the training time by several orders of magnitude, yielding trained models within seconds. QML-Lightning includes a cost-efficient GPU implementation of FCHL19, which together can yield energy and force predictions with competitive accuracy on a microsecond-per-atom timescale. Using modern GPU hardware, we report learning curves of energies and forces as well as timings as numerical evidence for select legacy benchmarks from atomisitic simulation including QM9, MD-17, and 3BPA., Comment: 25 pages, 5 figures

Despite their fundamental importance for science and society at large, experimental design decisions are often plagued by extreme data scarcity which severely hampers the use of modern ready-made machine learning models as they rely heavily on the paradigm, 'the bigger the data the better'. Presenting similarity based machine learning we show how to reduce these data needs such that decision making can be objectively improved in certain problem classes. After introducing similarity machine learning for the harmonic oscillator and the Rosenbrock function, we describe real-world applications to very scarce data scenarios which include quantum mechanics based molecular design and organic synthesis planning. Analysis of the query and training data proximity confirms that only a fraction of data is necessary to converge to competitive performance. Finally, we derive a relationship between the intrinsic dimensionality and volume of feature space, governing the overall model accuracy.

We use energies and forces predicted within response operator based quantum machine learning (OQML) to perform geometry optimization and transition state search calculations with legacy optimizers. For randomly sampled initial coordinates of small organic query molecules we report systematic improvement of equilibrium and transition state geometry output as training set sizes increase. Out-of-sample S$_\mathrm{N}$2 reactant complexes and transition state geometries have been predicted using the LBFGS and the QST2 algorithm with an RMSD of 0.16 and 0.4 \r{A} -- after training on up to 200 reactant complexes relaxations and transition state search trajectories from the QMrxn20 data-set, respectively. For geometry optimizations, we have also considered relaxation paths up to 5'500 constitutional isomers with sum formula C$_7$H$_{10}$O$_2$ from the QM9-database. Using the resulting OQML models with an LBFGS optimizer reproduces the minimum geometry with an RMSD of 0.14~\r{A}. For converged equilibrium and transition state geometries subsequent vibrational normal mode frequency analysis indicates deviation from MP2 reference results by on average 14 and 26\,cm$^{-1}$, respectively. While the numerical cost for OQML predictions is negligible in comparison to DFT or MP2, the number of steps until convergence is typically larger in either case. The success rate for reaching convergence, however, improves systematically with training set size, underscoring OQML's potential for universal applicability.

Equilibrium structures determine material properties and biochemical functions. We propose to machine learn phase-space averages, conventionally obtained by {\em ab initio} or force-field based molecular dynamics (MD) or Monte Carlo simulations. In analogy to \textit(ab initio} molecular dynamics (AIMD), our {\em ab initio} machine learning (AIML) model does not require bond topologies and therefore enables a general machine learning pathway to ensemble properties throughout chemical compound space. We demonstrate AIML for predicting Boltzmann averaged structures after training on hundreds of MD trajectories. AIML output is subsequently used to train machine learning models of free energies of solvation using experimental data, and reaching competitive prediction errors (MAE $\sim$ 0.8 kcal/mol) for out-of-sample molecules -- within milli-seconds. As such, AIML effectively bypasses the need for MD or MC-based phase space sampling, enabling exploration campaigns throughout CCS at a much accelerated pace. We contextualize our findings by comparison to state-of-the-art methods resulting in a Pareto plot for the free energy of solvation predictions in terms of accuracy and time.

We show that the energy of a perturbed system can be fully recovered from the unperturbed system's electron density. We derive an alchemical integral transform by parametrizing space in terms of transmutations, the chain rule and integration by parts. Within the radius of convergence, the zeroth order yields the energy expansion at all orders, restricting the textbook statement by Wigner that the $p$-th order wave function derivative is necessary to describe the $(2p+1)$-th energy derivative. Without the need for derivatives of the electron density, this allows to cover entire chemical neighborhoods from just one quantum calculation instead of single systems one-by-one. Numerical evidence presented indicates that predictive accuracy is achieved in the range of mHa for the harmonic oscillator or the Morse potential, and in the range of machine accuracy for hydrogen-like atoms. Considering iso-electronic nuclear charge variations by one proton in all multi-electron atoms from He to Ne, alchemical integral transform based estimates of the relative energy deviate by only few mHa from corresponding Hartree-Fock reference numbers.

We propose to relax geometries throughout chemical compound space (CCS) using alchemical perturbation density functional theory (APDFT). APDFT refers to perturbation theory involving changes in nuclear charges within approximate solutions to Schr\"odinger's equation. We give an analytical formula to calculate the mixed second order energy derivatives with respect to both, nuclear charges and nuclear positions (named "alchemical force"), within the restricted Hartree-Fock case. We have implemented and studied the formula for its use in geometry relaxation of various reference and target molecules. We have also analysed the convergence of the alchemical force perturbation series, as well as basis set effects. Interpolating alchemically predicted energies, forces, and Hessian to a Morse potential yields more accurate geometries and equilibrium energies than when performing a standard Newton Raphson step. Our numerical predictions for small molecules including BF, CO, N2, CH$_4$, NH$_3$, H$_2$O, and HF yield mean absolute errors of of equilibrium energies and bond lengths smaller than 10 mHa and 0.01 Bohr for 4$^\text{th}$ order APDFT predictions, respectively. Our alchemical geometry relaxation still preserves the combinatorial efficiency of APDFT: Based on a single coupled perturbed Hartree Fock derivative for benzene we provide numerical predictions of equilibrium energies and relaxed structures of all the 17 iso-electronic charge-netural BN-doped mutants with averaged absolute deviations of $\sim$27 mHa and $\sim$0.12 Bohr, respectively.

We introduce an electronic structure based representation for quantum machine learning (QML) of electronic properties throughout chemical compound space. The representation is constructed using computationally inexpensive ab initio calculations and explicitly accounts for changes in the electronic structure. We demonstrate the accuracy and flexibility of resulting QML models when applied to property labels such as total potential energy, HOMO and LUMO energies, ionization potential, and electron affinity, using as data sets for training and testing entries from the QM7b, QM7b-T, QM9, and LIBE libraries. For the latter, we also demonstrate the ability of this approach to account for molecular species of different charge and spin multiplicity, resulting in QML models that infer total potential energies based on geometry, charge, and spin as input.

Quantum Machine Learning (QML) models of molecular HOMO-LUMO-gaps often struggle to achieve satisfying data-efficiency as measured by decreasing prediction errors for increasing training set sizes. Partitioning training sets of organic molecules (QM7 and QM9-data-sets) into three classes [systems containing either aromatic rings and carbonyl groups, or single unsaturated bonds, or saturated bonds] prior to training results in independently trained QML models with improved learning rates. The selected QML models of band-gaps (at GW, B3LYP, and ZINDO level of theory) reach mean absolute prediction errors of $\sim$0.1 eV for up to an order of magnitude fewer training molecules than for conventionally trained models. Direct comparison to $\Delta$-QML models of band-gaps suggest that selected QML possesses substantially more data-efficiency. The findings suggest that selected QML, e.g. based on simple classifications prior to training, could help to successfully tackle challenging quantum property screening tasks of large libraries with high fidelity and low computational burden., Comment: 19 pages, 20 figures

We present a systematic study of interaction induced dipole electric properties of all molecular dimers in the S66 set, relying on CCSD(T)-F12b/aug-cc-pVDZ-F12 as reference level of theory. For field strengths up to $\approx$5 GV m$^{-1}$ the interaction induced electric response beyond second order is found to be insignificant. Large interaction dipole moments (i.e. dipole moment changes due to van der Waals binding) are observed in the case of hydrogen bonding oriented along the intermolecular axis, and mostly small interaction dipole moments are found in dimers bonded by $\pi$-stacking or London dispersion. The interaction polarizabilities (i.e. polarizability changes due to van der Waals binding) were generally found to be small but always with a positive-valued principal component approximately aligned with the intermolecular axis, and two other negative-valued components. Energy decompositions according to symmetry adapted perturbation theory (SAPT0/jun-cc-pVDZ) suggest that electrostatics dominates the interaction dipole moment. First-order SAPT0 decompositions into monomer-resolved contributions establish a quantitative link between electric properties of monomers and dimers. Using the aug-cc-pVQZ basis and non-empirical PBE semilocal exchange-correlation kernels, we also assess how density functional approximations in the nonlocal exchange and correlation parts affect the predictive accuracy: While dRPA@PBE0 based predictions are in excellent overall agreement with coupled cluster results, the computationally more affordable LC-$\omega$PBE0-D3 level of theory also yields reliable results with relative errors below 5\%. PBE alone, even when dispersion corrected, produces larger errors in interaction dipole moments and polarizabilities.

Diels-Alder cycloadditions are efficient routes for the synthesis of cyclic organic compounds. There has been a long-standing discussion whether these reactions proceed via stepwise or concerted mechanisms. Here, we adopt a new experimental approach to explore the mechanistic details of the model polar cycloaddition of 2,3-dibromo-1,3-butadiene with propene ions by probing its conformational specificities in the entrance channel under single-collision conditions in the gas phase. Combining a conformationally controlled molecular beam with trapped ions, we find that both conformers of the diene, gauche and s-trans, are reactive with capture-limited reaction rates. Aided by quantum-chemical and quantum-capture calculations, this finding is rationalised by a simultaneous competition of concerted and stepwise reaction pathways, revealing an interesting mechanistic borderline case.

The computational prediction of atomistic structure is a long-standing problem in physics, chemistry, materials, and biology. Within conventional force-field or {\em ab initio} calculations, structure is determined through energy minimization, which is either approximate or computationally demanding. Alas, the accuracy-cost trade-off prohibits the generation of synthetic big data records with meaningful energy based conformational search and structure relaxation output. Exploiting implicit correlations among relaxed structures, our kernel ridge regression model, dubbed Graph-To-Structure (G2S), generalizes across chemical compound space, enabling direct predictions of relaxed structures for out-of-sample compounds, and effectively bypassing the energy optimization task. After training on constitutional and compositional isomers (no conformers) G2S infers atomic coordinates relying solely on stoichiometry and bond-network information as input (Our numerical evidence includes closed and open shell molecules, transition states, and solids). For all data considered, G2S learning curves reach mean absolute interatomic distance prediction errors of less than 0.2 {\AA} for less than eight thousand training structures -- on par or better than popular empirical methods. Applicability test of G2S include meaningful structures of molecules for which standard methods require manual intervention, improved initial guesses for subsequent conventional {\em ab initio} based relaxation, and input for structural based representations commonly used in quantum machine learning models, (bridging the gap between graph and structure based models).

To unravel the structures of C12H12O7 isomers, identified as light-absorbing photooxidation products of syringol in atmospheric chamber experiments, we apply a graph-based molecule generator and machine learning workflow. To accomplish this in a bias-free manner, molecular graphs of the entire chemical subspace of C12H12O7 were generated, assuming that the isomers contain two C6-rings; this led to 260 million molecular graphs and 120 million stable structures. Using quantum chemistry excitation energies and oscillator strengths as training data, we predicted these quantities using kernel ridge regression and simulated UV/Vis absorption spectra. Then we determined the probability of the molecules to cause the experimental spectrum within the errors of the different methods. Molecules whose spectra were likely to match the experimental spectrum were clustered according to structural features, resulting in clusters of > 500,000 molecules. While we identified several features that correlate with a high probability to cause the experimental spectrum, no clear composition of necessary features can be given. Thus, the absorption spectrum is not sufficient to uniquely identify one specific isomer structure. If more structural features were known from experimental data, the number of structures could be reduced to a few tens of thousands candidates. We offer a procedure to detect when sufficient fragmentation data has been included to reduce the number of possible molecules. The most efficient strategy to obtain valid candidates is obtained if structural data is applied already at the bias-free molecule generation stage. The systematic enumeration, however, is necessary to avoid mis-identification of molecules, while it guarantees that there are no other molecules that would also fit the spectrum in question.

Free energies govern the behavior of soft and liquid matter, and improving their predictions could have a large impact on the development of drugs, electrolytes or homogeneous catalysts. Unfortunately, it is challenging to devise an accurate description of effects governing solvation such as hydrogen-bonding, van der Waals interactions, or conformational sampling. We present a Free energy Machine Learning (FML) model applicable throughout chemical compound space and based on a representation that employs Boltzmann averages to account for an approximated sampling of configurational space. Using the FreeSolv database, FML's out-of-sample prediction errors of experimental hydration free energies decay systematically with training set size, and experimental uncertainty (0.6 kcal/mol) is reached after training on 490 molecules (80\% of FreeSolv). Corresponding FML model errors are also on par with state-of-the art physics based approaches. To generate the input representation for a new query compound, FML requires approximate and short molecular dynamics runs. We showcase its usefulness through analysis of FML solvation free energies for 116k organic molecules (all force-field compatible molecules in QM9 database) identifying the most and least solvated systems, and rediscovering quasi-linear structure property relationships in terms of simple descriptors such as hydrogen-bond donors, number of NH or OH groups, number of oxygen atoms in hydrocarbons, and number of heavy atoms. FML's accuracy is maximal when the temperature used for the molecular dynamics simulation to generate averaged input representation samples in training is the same as for the query compounds. The sampling time for the representation converges rapidly with respect to the prediction error.

Chemical compound space (CCS), the set of all theoretically conceivable combinations of chemical elements and (meta-)stable geometries that make up matter, is colossal. The first principles based virtual sampling of this space, for example in search of novel molecules or materials which exhibit desirable properties, is therefore prohibitive for all but the smallest sub-sets and simplest properties. We review studies aimed at tackling this challenge using modern machine learning techniques based on (i) synthetic data, typically generated using quantum mechanics based methods, and (ii) model architectures inspired by quantum mechanics. Such Quantum mechanics based Machine Learning (QML) approaches combine the numerical efficiency of statistical surrogate models with an {\em ab initio} view on matter. They rigorously reflect the underlying physics in order to reach universality and transferability across CCS. While state-of-the-art approximations to quantum problems impose severe computational bottlenecks, recent QML based developments indicate the possibility of substantial acceleration without sacrificing the predictive power of quantum mechanics.