Search

Finding accurate exchange-correlation (XC) functionals remains the defining challenge in density functional theory (DFT). Despite 40 years of active development, the desired chemical accuracy is still elusive with existing functionals. We present a data-driven pathway to learn the XC functionals by utilizing the exact density, XC energy, and XC potential. While the exact densities are obtained from accurate configuration interaction (CI), the exact XC energies and XC potentials are obtained via inverse DFT calculations on the CI densities. We demonstrate how simple neural network (NN) based local density approximation (LDA) and generalized gradient approximation (GGA), trained on just five atoms and two molecules, provide remarkable improvement in total energies, densities, atomization energies, and barrier heights for hundreds of molecules outside the training set. Particularly, the NN-based GGA functional attains similar accuracy as the higher rung SCAN meta-GGA, highlighting the promise of using the XC potential in modeling XC functionals. We expect this approach to pave the way for systematic learning of increasingly accurate and sophisticated XC functionals.

The Ryabinkin-Kohut-Staroverov (RKS) and Kanungo-Zimmerman-Gavini (KZG) methods offer two approaches to find exchange-correlation (XC) potentials from ground state densities. The RKS method utilizes the one- and two-particle reduced density matrices to alleviate any numerical artifacts stemming from a finite basis (e.g., Gaussian- or Slater-type orbitals). The KZG approach relies solely on the density to find the XC potential, by combining a systematically convergent finite-element basis with appropriate asymptotic correction on the target density. The RKS method, being designed for a finite basis, offers computational efficiency. The KZG method, using a complete basis, provides higher accuracy. In this work, we combine both the methods to simultaneously afford accuracy and efficiency. In particular, we use the RKS solution as initial guess to the KZG method to attain a significant $3-11\times$ speedup. This work also presents a direct comparison of the XC potentials from the RKS and the KZG method and their relative accuracy on various weakly and strongly correlated molecules, using their ground state solutions from accurate configuration interaction calculations solved in a Slater orbital basis.

This work introduces Extent of Violation Indices (EVIs), a novel metric for quantifying how well exchange-correlation functionals adhere to local conditions. Applying EVIs to a diverse set of molecules for GGA functionals reveals widespread violations, particularly for semi-empirical functionals. We leverage EVIs to explore potential connections between these violations and errors in chemical properties. While no correlation is observed for atomization energies, a link emerges between EVIs and total energies. Similarly, the analysis of reaction energies suggests weak positive correlations for specific conditions, but definitive conclusions about error cancellation require advancements in both functional accuracy and our understanding of cancellation mechanisms. Overall, this study highlights EVIs as a powerful tool for analyzing functional behavior and adherence to local conditions, paving the way for future research to fully elucidate the impact of violations on energy errors.

Accurate solutions to the electronic Schr\"odinger equation can provide valuable insight for electron interactions within molecular systems, accelerating the molecular design and discovery processes in many different applications. However, the availability of such accurate solutions are limited to small molecular systems due to both the extremely high computational complexity and the challenge of operating and executing these workloads on high-performance compute clusters. This work presents a massively scalable cloud-based quantum chemistry platform by implementing a highly parallelizable quantum chemistry method that provides a polynomial-scaling approximation to full configuration interaction (FCI). Our platform orchestrates more than one million virtual CPUs on the cloud to analyze the bond-breaking behaviour of carbon-fluoride bonds of per- and polyfluoroalkyl substances (PFAS) with near-exact accuracy within the chosen basis set. This is the first quantum chemistry calculation utilizing more than one million virtual CPUs on the cloud and is the most accurate electronic structure computation of PFAS bond breaking to date.

The conventional approaches to the inverse density functional theory problem typically assume non-degeneracy of the Kohn-Sham (KS) eigenvalues, greatly hindering their use in open-shell systems. We present a generalization of the inverse density functional theory problem that can seamlessly admit degenerate KS eigenvalues. Additionally, we allow for fractional occupancy of the Kohn-Sham orbitals to also handle non-interacting ensemble-v-representable densities, as opposed to just non-interacting pure-v-representable densities. We present the exact exchange-correlation (XC) potentials for six open-shell systems -- four atoms (Li, C, N, and O) and two molecules (CN and $\text{CH}_2$) -- using accurate ground-state densities from configuration interaction calculations. We compare these exact XC potentials with model XC potentials obtained using non-local (B3LYP, SCAN0) and local/semi-local (SCAN, PBE, PW92) XC functionals. Although the relative errors in the densities obtained from these DFT functionals are of $\mathcal{O}(10^{-3}-10^{-2})$, the relative errors in the model XC potentials remain substantially large -- $\mathcal{O}(10^{-1}-10^0)$.

Ryabinkin-Kohut-Staroverov (RKS) theory builds a bridge between wave function theory and density functional theory by using quantities from the former to produce accurate exchange-correlation potentials needed by the latter. In this work, the RKS method is developed and tested alongside Slater atomic orbital basis functions for the first time. To evaluate this approach, Full Configuration Interaction computations in the Slater orbitals are employed to give quality input to RKS method, allowing full correlation to be present along with correct nuclei cusps and asymptotic decay of the wavefunction. The RKS method will be shown to be an efficient algorithm to arrive at exchange correlation potentials without unphysical artifacts in moderately-sized basis sets. Furthermore, enforcement of the nuclear cusp conditions will be shown to be vital for the success of the Slater-basis RKS method. Examples of weakly and strongly correlated molecular systems will demonstrate the main features of Slater RKS., Comment: 10 pages, 9 figures

In this paper, we study a sequential decision-making problem, called Adaptive Sampling for Discovery (ASD). Starting with a large unlabeled dataset, algorithms for ASD adaptively label the points with the goal to maximize the sum of responses. This problem has wide applications to real-world discovery problems, for example drug discovery with the help of machine learning models. ASD algorithms face the well-known exploration-exploitation dilemma. The algorithm needs to choose points that yield information to improve model estimates but it also needs to exploit the model. We rigorously formulate the problem and propose a general information-directed sampling (IDS) algorithm. We provide theoretical guarantees for the performance of IDS in linear, graph and low-rank models. The benefits of IDS are shown in both simulation experiments and real-data experiments for discovering chemical reaction conditions.

Accurate exchange-correlation (XC) potentials for 3-dimensional systems -- via solution of the \emph{inverse} density functional theory (DFT) problem -- are now available to test the quality of DFT approximations. Herein, the \emph{exact} XC potential for six molecules -- hydrogen molecule at three different bond-lengths, lithium hydride, water, and ortho-benzyne -- are computed using accurate ground-state densities from full configuration interaction (CI) calculations. These potentials are then compared to model XC potentials obtained from DFT calculations with commonly used non-local (B3LYP, HSE06, SCAN0, and M08-HX) and local/semi-local (SCAN, PBE, PW92) XC functionals. While relative errors in the ground-state densities from these models are order $\mathcal{O}(10^{-3}-10^{-2})$, much larger errors in the model XC potentials are found, $\mathcal{O}(10^{-1}-10^0)$, in both the $L_2$ norm of the potential as well as its gradients. These errors are exacerbated in strongly correlated situations, such as the stretched $\text{H}_2$ molecule. Among the model XC functionals under consideration, SCAN0 offers the best quantitative and qualitative agreement with the exact XC potential, underlining the significance of satisfying the exact conditions as well as the incorporation of non-local effects in the construction of XC functionals. Overall, this work indicates that tests against the exact XC potential will provide a promising new direction for building more accurate XC functionals for DFT.

Molecular geometry prediction of flexible molecules, or conformer search, is a long-standing challenge in computational chemistry. This task is of great importance for predicting structure-activity relationships for a wide variety of substances ranging from biomolecules to ubiquitous materials. Substantial computational resources are invested in Monte Carlo and Molecular Dynamics methods to generate diverse and representative conformer sets for medium to large molecules, which are yet intractable to chemoinformatic conformer search methods. We present TorsionNet, an efficient sequential conformer search technique based on reinforcement learning under the rigid rotor approximation. The model is trained via curriculum learning, whose theoretical benefit is explored in detail, to maximize a novel metric grounded in thermodynamics called the Gibbs Score. Our experimental results show that TorsionNet outperforms the highest scoring chemoinformatics method by 4x on large branched alkanes, and by several orders of magnitude on the previously unexplored biopolymer lignin, with applications in renewable energy.

The electronically excited states of methylene (CH$_2$), ethylene (C$_2$H$_4$), butadiene (C$_4$H$_6$), hexatriene (C$_6$H$_8$), and ozone (O$_3$) have long proven challenging due to their complex mixtures of static and dynamic correlations. Semistochastic heat-bath configuration interaction (SHCI), which efficiently and systematically approaches the full configuration interaction (FCI) limit, is used to provide close approximations to the FCI energies in these systems. This article presents the largest FCI-level calculation to date -- on hexatriene using a polarized double-zeta basis (ANO-L-pVDZ), which gives rise to a Hilbert space containing more than $10^{38}$ determinants. These calculations give vertical excitation energies of 5.58 and 5.59 eV respectively for the $2^1{\rm A}_{\rm g}$ and $1^1{\rm B}_{\rm u}$ states, showing that they are nearly degenerate. The same excitation energies in butadiene/ANO-L-pVDZ were found to be 6.58 and 6.45 eV. In addition to these benchmarks, our calculations strongly support the presence of a previously hypothesized ring-minimum species of ozone that lies 1.3 eV higher than the open-ring minimum energy structure and is separated from it by a barrier of 1.11 eV.