Thermodynamics of Anharmonic Systems: Uncoupled Mode Approximations for Molecules

Li, Yi‐Pei; Bell, Alexis T.; Head‐Gordon, Martin

doi:10.1021/acs.jctc.5b01177

Cited by 42 publications

(99 citation statements)

References 58 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…48 As discussed in previous work, these calculated enthalpies are themselves associated with significant errors, primarily due to weaknesses of B3LYP such as the absence of long-range dispersion interaction but also the lack of rotor or conformer corrections in the calculations. [49][50][51][52] We note that it is possible to use a small amount of high-accuracy coupled cluster training data via a transfer learning approach to minimize the influence of DFT errors. Interested readers are referred to the recent work of Grambow et al 53 In this work, we use the QM9 data as is without any attempt to correct its errors in order to investigate the effects of aleatoric uncertainties.…”

Section: Datamentioning

confidence: 99%

Evaluating Scalable Uncertainty Estimation Methods for Deep Learning-Based Molecular Property Prediction

Scalia

Grambow

Pernici

et al. 2020

J. Chem. Inf. Model.

Self Cite

176

198

View full text Add to dashboard Cite

Advances in deep neural network (DNN) based molecular property prediction have recently led to the development of models of remarkable accuracy and generalization ability, with graph convolution neural networks (GCNNs) reporting state-of-the-art performance for this task. However, some challenges remain and one of the most important that needs to be fully addressed concerns uncertainty quantification. DNN performance is affected by the volume and the quality of the training samples. Therefore, establishing when and to what extent a prediction can be considered reliable is just as important as outputting accurate predictions, especially when out-of-domain molecules are targeted. Recently, several methods to 1 arXiv:1910.03127v1 [cs.LG] 7 Oct 2019 account for uncertainty in DNNs have been proposed, most of which are based on approximate Bayesian inference. Among these, only a few scale to the large datasets required in applications. Evaluating and comparing these methods has recently attracted great interest, but results are generally fragmented and absent for molecular property prediction. In this paper, we aim to quantitatively compare scalable techniques for uncertainty estimation in GCNNs. We introduce a set of quantitative criteria to capture different uncertainty aspects, and then use these criteria to compare MC-Dropout, deep ensembles, and bootstrapping, both theoretically in a unified framework that separates aleatoric/epistemic uncertainty and experimentally on the QM9 dataset. Our experiments quantify the performance of the different uncertainty estimation methods and their impact on uncertainty-related error reduction. Our findings indicate that ensembling and bootstrapping consistently outperform MC-Dropout, with different context-specific pros and cons. Our analysis also leads to a better understanding of the role of aleatoric/epistemic uncertainty and highlights the challenge posed by out-of-domain uncertainty.

show abstract

Section: Datamentioning

confidence: 99%

Evaluating Scalable Uncertainty Estimation Methods for Deep Learning-Based Molecular Property Prediction

Scalia

Grambow

Pernici

et al. 2020

J. Chem. Inf. Model.

Self Cite

176

198

View full text Add to dashboard Cite

show abstract

“…However, it has not yet been established whether this approach can achieve nearly quantitative accuracy for H 2 binding in MOFs in general. 104,105 Alternatively, entropic contributions can be computed directly from finite-temperature molecular dynamics (either ab initio or classical methods). One promising approach for using dynamics trajectories to interpolate and extrapolate entropy over a broad range of temperatures is the two-phase thermodynamic method of Lin et al 106 This method has been used to evaluate the entropy of pure CO 2 , as well as adsorbates confined within zeolites, suggesting a similar approach may be adopted to obtain accurate entropies of bound H 2 .…”

Section: Thermodynamic Factors Beyond Binding Energymentioning

confidence: 99%

An assessment of strategies for the development of solid-state adsorbents for vehicular hydrogen storage

Allendorf

Hulvey

Gennett

et al. 2018

Energy Environ. Sci.

Self Cite

229

187

View full text Add to dashboard Cite

Nanoporous adsorbents are a diverse category of solid-state materials that hold considerable promise for vehicular hydrogen storage. Although impressive storage capacities have been demonstrated for several materials, particularly at cryogenic temperatures, materials meeting all of the targets established by the U.S. Department of Energy have yet to be identified. In this Perspective, we provide an overview of the major known and proposed strategies for hydrogen adsorbents, with the aim of guiding ongoing research as well as future new storage concepts. The discussion of each strategy includes current relevant literature, strengths and weaknesses, and outstanding challenges that preclude implementation. We consider in particular metal-organic frameworks (MOFs), including surface area/volume tailoring, open metal sites, and the binding of multiple H 2 molecules to a single metal site. Two related classes of porous framework materials, covalent organic frameworks (COFs) and porous aromatic frameworks (PAFs), are also discussed, as are graphene and graphene oxide and doped porous carbons. We additionally introduce criteria for evaluating the merits of a particular materials design strategy. Computation has become an important tool in the discovery of new storage materials, and a brief introduction to the benefits and limitations of computational predictions of H 2 physisorption is therefore presented. Finally, considerations for the synthesis and characterization of hydrogen storage adsorbents are discussed. IntroductionStorage of hydrogen with sufficient gravimetric and volumetric capacity for vehicular use remains a significant obstacle to the widespread adoption of hydrogen fuel cell electric vehicles (FCEVs). Several FCEV models are now commercially available in limited locations around the world, and in these vehicles hydrogen is stored as a gas at room temperature with a fill Table 1 along with the current performance of 700 bar systems. These values pertain to the entire storage system, which includes the mass and volume of hydrogen in addition to the tank and associated balance-ofplant (BOP) components. Notably, it is physically impossible to meet the 2025 and ultimate volumetric capacity target with pressurized gas, as the density of H 2 gas at 700 bar and room temperature is just 40 g L À1 without accounting for the BOP.The search for solid-state H 2 storage materials that can supplant compressed gas systems has been ongoing for at least two decades. The development of a viable storage material Broader contextThe widespread use of hydrogen as a clean, sustainable energy carrier has the potential to provide several significant benefits, including a reduction in oil dependency and emissions, improved energy security and grid resiliency, and substantial economic opportunities across many sectors. Hydrogen-fueled vehicles are already appearing internationally, and one of the critical enabling technologies for increasing their availability is on-board hydrogen storage. Stakeholders in developing a hydrogen in...

show abstract

“…Some promising advances have been made in this area. These include advanced TS configuration sampling via uncoupled mode approximations [165], metadynamics [166], and umbrella sampling [167] approaches, to name just a few. Nonetheless, considerable challenges remain in adequately modeling entropic effects at affordable computational costs.…”

Section: Model Detailsmentioning

confidence: 99%

Impact of long-range electrostatic and dispersive interactions on theoretical predictions of adsorption and catalysis in zeolites

et al. 2018

Self Cite

View full text Add to dashboard Cite

In this paper, we review the importance of long-range zeolite framework interactions in theoretical predictions for a variety of zeolite-catalyzed processes, and we show why such interactions must be determined accurately in order to reproduce experimentally measured adsorption and activation energies. We begin with an overview of the different strategies that have been used to account for long-range coulombic and dispersive interactions of zeolite framework atoms with species adsorbed at an active site. These methods include full periodic-DFT calculations and multi-layer hybrid techniques. Electrostatic interactions are observed to have a more significant impact than dispersive interactions on the geometries of ion-pair transition states and adsorbed species. Stabilization of the TS relative to reactant complexes is also dictated by electrostatic interactions. Dispersion effects are found to significantly stabilize both transition and reactant states for adsorbed species, especially those which have dimensions that provide good fits within the zeolite pore or cavity. We also show that the relevance of particular active site configurations can be missed, if the effects of long-range interactions are neglected. As a case in point, we demonstrate that a site previously considered inactive for ethane dehydrogenation, [GaH 2 ] + may in fact be more active than previously thought, when the impact of long-range interactions on the predicted activation energy is taken into account. Finally, the use of hybrid quantum mechanics/molecular mechanics approaches on extended, finite zeolite clusters has emerged as an accurate, highly cost-effective, and versatile alternative towards overcoming some of the present-day limitations of periodic calculations.

show abstract

Thermodynamics of Anharmonic Systems: Uncoupled Mode Approximations for Molecules

Cited by 42 publications

References 58 publications

Evaluating Scalable Uncertainty Estimation Methods for Deep Learning-Based Molecular Property Prediction

Evaluating Scalable Uncertainty Estimation Methods for Deep Learning-Based Molecular Property Prediction

An assessment of strategies for the development of solid-state adsorbents for vehicular hydrogen storage

Impact of long-range electrostatic and dispersive interactions on theoretical predictions of adsorption and catalysis in zeolites

Contact Info

Product

Resources

About