Journal of Chemical Theory and Computation最新文献

Comprehending water dynamics is crucial in various fields, such as water desalination, ion separation, electrocatalysis, and biochemical processes. While ab initio molecular dynamics (AIMD) accurately portray water's structure, computing its dynamic properties over nanosecond time scales proves cost-prohibitive. This study employs machine learning potentials (MLPs) to accurately determine the dynamic properties of liquid water with ab initio accuracy. Our findings reveal diversity in the calculated diffusion coefficient (D) and viscosity of water (η) across different methodologies. Specifically, while the GGA, meta-GGA, and hybrid functional methods struggle to predict dynamic properties under ambient conditions, methods on the higher level of Jacob's ladder of DFT approximation perform significantly better. Intriguingly, we discovered that both D and η adhere to the established Stokes-Einstein (SE) relation for all of the ab initio water. The diversity observed across different methods can be attributed to distinct structural entropy, affirming the applicability of excess entropy scaling relations across all functionals. The correlation between D and η provides valuable insights for identifying the ideal temperature to accurately replicate the dynamic properties of liquid water. Furthermore, our findings can validate the rationale behind employing artificially high temperatures in the simulation of water via AIMD. These outcomes not only pave the path to designing better functionals for water but also underscore the significance of water's many-body characteristics.

We introduce a method for reducing the number of valence states entering the calculation of screened the Coulomb interaction W in GW calculations. In this way, denoting with N the generic size of a system, the computational cost is brought from the typical O(N⁴) to the more favorable O(N² ln N). The method becomes effective for large model structures. For enhancing the potentialities of our scheme, we combined it with a linear-response GW approach, which can exploit the symmetries of the simulation cell in direct space. We registered quadratic scaling up to more than thousand atoms with an almost 10-fold speed-up with respect to a standard implementation. Our scheme can be extended to any linear response calculation.

A common approach for computing free energy differences among multiple states is to build a perturbation graph connecting the states and compute free energy differences on all edges of the graph. Such perturbation graphs are often designed to have cycles. Because free energy is a function of states, the free energy around any cycle is zero, which we refer to as the cycle consistency condition. Since the cycle consistency condition relates free energy differences on the edges of a cycle, it could be used to improve the accuracy of free energy estimates. Here, we propose a Bayesian method called the coupled Bayesian multistate Bennett acceptance ratio (CBayesMBAR) that can properly couple the calculations of free energy differences on the edges of cycles in a principled way. We apply the CBayesMBAR to compute free energy differences among harmonic oscillators and relative protein-ligand binding free energies. In both cases, the CBayesMBAR provides more accurate results compared to methods that do not consider the cycle consistency condition. Additionally, it outperforms the cycle closure correction method that also uses cycle consistency conditions.

Atom-centered neural network (ANN) potentials have shown high accuracy and computational efficiency in modeling atomic systems. A crucial step in developing reliable ANN potentials is the proper selection of atom-centered symmetry functions (ACSFs), also known as atomic features, to describe atomic environments. Inappropriate selection of ACSFs can lead to poor-quality ANN potentials. Here, we propose a gradient boosting decision tree (GBDT)-based framework for the automatic selection of optimal ACSFs. This framework takes uniformly distributed sets of ACSFs as input and evaluates their relative importance. The ACSFs with high average importance scores are selected and used to train an ANN potential. We applied this method to the Ge system, resulting in an ANN potential with root-mean-square errors (RMSE) of 10.2 meV/atom for energy and 84.8 meV/Å for force predictions, utilizing only 18 ACSFs to achieve a balance between accuracy and computational efficiency. The framework is validated using the grid searching method, demonstrating that ACSFs selected with our framework are in the optimal region. Furthermore, we also compared our method with commonly used feature selection algorithms. The results show that our algorithm outperforms the others in terms of effectiveness and accuracy. This study highlights the significance of the ACSF parameter effect on the ANN performance and presents a promising method for automatic ACSF selection, facilitating the development of machine learning potentials.

Empirical parametrization underpins many scientific methodologies including certain quantum-chemistry protocols [e.g., density functional theory (DFT), machine-learning (ML) models]. In some cases, the fitting requires a large amount of data, necessitating the use of data obtained using low-cost, and thus low-quality, means. Here we examine the effect of using low-quality data on the resulting method in the context of DFT methods. We use multiple G2/97 data sets of different qualities to fit the DFT-type methods. Encouragingly, this fitting can tolerate a relatively large proportion of low-quality fitting data, which may be attributed to the physical foundations of the DFT models and the use of a modest number of parameters. Further examination using "ML-quality" data shows that adding a large amount of low-quality data to a small number of high-quality ones may not offer tangible benefits. On the other hand, when the high-quality data is limited in scope, diversification by a modest amount of low-quality data improves the performance. Quantitatively, for parametrizing DFT (and perhaps also quantum-chemistry ML models), caution should be taken when more than 50% of the fitting set contains questionable data, and that the average error of the full set is more than 20 kJ mol^-1. One may also follow the recently proposed transferability principles to ensure diversity in the fitting set.

Boron-doped graphene nanoribbons are promising platforms for developing organic materials with magnetic properties. Boron dopants can be used to create localized magnetic states in nanoribbons with tunable interactions. Controlling the coherence times of these magnetic states is the very first step in designing materials for quantum computation or information storage. In this work, we address the connection between the relaxation time and the position of the dopants for a series of boron-doped graphene nanofragments. We combine Redfield theory and ab initio calculations of magnetic properties to unveil the mechanism that governs spin relaxation in solution. We demonstrate that relaxation times can be in the order of 1 ms for the selected graphene nanofragments. A detailed analysis of the relaxation mechanism reveals that the spin decoherence is fundamentally driven by fluctuations of the spin-orbit coupling, and the hyperfine interaction facilitated by the thermal motion of the graphene nanofragments. The close connection between relaxation time, hyperfine interaction and the spin-orbit coupling offers the perspective of designing attractive materials with long-lived spin states.

Estimating absolute binding free energies from molecular simulations is a key step in computer-aided drug design pipelines, but the agreement between computational results and experiments is still very inconsistent. Both the accuracy of the computational model and the quality of the statistical sampling contribute to this discrepancy, yet disentangling the two remains a challenge. In this study, we present an automated protocol based on OneOPES, an enhanced sampling method that exploits replica exchange and can accelerate several collective variables to address the sampling problem. We apply this protocol to 37 host-guest systems. The simplicity of setting up the simulations and producing well-converged binding free energy estimates without the need to optimize simulation parameters provides a reliable solution to the sampling problem. This, in turn, allows for a systematic force field comparison and ranking according to the correlation between simulations and experiments, which can inform the selection of an appropriate model. The protocol can be readily adapted to test more force field combinations and study more complex protein-ligand systems, where the choice of an appropriate physical model is often based on heuristic considerations rather than systematic optimization.

Doped conjugated polymers have a variety of potential applications in thermoelectric and other electronic devices, but the nature of their electronic structure is still not well understood. In this work, we use time-dependent density functional theory (TD-DFT) calculations along with natural transition orbital (NTO) analysis to understand electronic structures of both p-type (e.g., poly(3-hexylthiophene-2,5-diyl), P3HT) and n-type (e.g., poly{[N,N'-bis(2-octyldodecyl)-naphthalene-1,4,5,8-bis(dicarboximide)-2,6-diyl]-alt-5,5'-(2,2'-bithiophene)}, N2200) conjugated polymers that are both p-doped and n-doped. Of course, the electronic transitions of doped conjugated polymers are multiconfigurational in nature, but it is still useful to have a one-electron energy level diagram with which to interpret their spectroscopy and other electronic behaviors. Based on the NTOs associated with the TD-DFT transitions, we find that the "best" one-electron orbital-based energy level diagram for doped conjugated polymers such as P3HT is the so-called traditional band picture. We also find that the situation is more complicated for donor-acceptor-type polymers like N2200, where the use of different exchange-correlation functionals leads to different predicted optical transitions that have significantly less one-electron character. For some functionals, we still find that the "best" one-electron energy level diagram agrees with the traditional picture, but for others, there is no obvious route to reducing the multiconfigurational transitions to a one-electron energy level diagram. We also see that the presence of both electron-rich and electron-poor subunits on N2200 breaks the symmetry between n- and p-doping, because different types of polarons reside on different subunits leading to different degrees of charge delocalization. This effect is exaggerated by the presence of dopant counterions, which interact differently with n- and p-polarons. Despite these complications, we argue that the traditional band picture suffices if one wishes to employ a simple one-electron picture to explain the spectroscopy of n- and p-doped conjugated polymers.