Though molecular beam experiments have revealed deficiencies in the Lorentz-Berthelot combining rules, these rules are still used widely to parametrize effective pair potential models or to calculate the thermodynamic properties of mixtures. Gibbs ensemble Monte Carlo and isothermal isobaric Monte Carlo simulations were used to compute the liquid-vapour phase equilibria and the liquid properties of binary mixtures of rare gases modelled by effective pair potentials. Three sets of simple combining rules were tested in this work: the commonly used Lorentz-Berthelot rules, the Kong rules (Kong, J., 1973, J. chem. Phys., 59, 2464) and the Waldman-Hagler rules (Waldman, M., and Hagler, A. T., 1993, J. comput. Chem., 14, 1077). These three sets of rules do not require any additional parameter. It is shown that: (1) the choice of a set of combining rules has a significant effect on the thermodynamic properties, (2) using the Lorentz-Berthelot rules yields significant deviations from experiment and (3) the Kong rules provide a much better description of the mixture properties both for coexistence curves and liquid properties. We therefore recommend the use of the Kong rules instead of the Lorentz-Berthelot when parametrizing potential models.
The parameters of the anisotropic united atoms potential for linear alkanes proposed by Toxvaerd [S. Toxvaerd, J. Chem. Phys. 107, 5197 (1997)] have been optimized on the basis of selected equilibrium properties (vapor pressures, vaporization enthalpies, and liquid densities) of ethane, n-pentane, and n-dodecane. The optimized parameters for the and groups form a regular sequence with those of methane and the force centers are found between the carbon and hydrogen atoms, as expected. The resulting potential, called AUA4, has been compared with Toxvaerd’s potential (AUA3) by using several molecular simulation methods (Gibbs ensemble Monte Carlo, thermodynamic integration, and molecular dynamics). An investigation performed at temperatures ranging from 140 to 700 K and with various chain lengths up to 20 carbon atoms has shown AUA4 to provide systematic improvements of vapor pressures, vaporization enthalpies, and liquid densities for pure n-alkanes. Significant improvements have been also noticed on the critical temperatures of n-alkanes, estimated from coexistence density curves, and on the equilibrium properties of –n-alkane binary mixtures. Self-diffusion coefficients of n-hexane, however, are slightly improved by the new potential, but still exceed experimental measurements at low temperature. As we have only optimized the intermolecular potential in the present study, it is suggested that further optimization of the intramolecular potentials of the anisotropic united atoms model could allow simultaneous prediction of thermodynamic properties and of transport coefficients, particularly in very dense liquids.
We carry out a critical assessment of machine learning and deep learning models for the classification of skin tumors. Machine learning (ML) algorithms tested in this work include logistic regression, linear discriminant analysis, k-nearest neighbors classifier, decision tree classifier and Gaussian naive Bayes, while deep learning (DL) models employed are either based on a custom Convolutional Neural Network model, or leverage transfer learning via the use of pre-trained models (VGG16, Xception and ResNet50). We find that DL models, with accuracies up to 0.88, all outperform ML models. ML models exhibit accuracies below 0.72, which can be increased to up to 0.75 with ensemble learning. To further assess the performance of DL models, we test them on a larger and more imbalanced dataset. Metrics, such as the F-score and accuracy, indicate that, after fine-tuning, pre-trained models perform extremely well for skin tumor classification. This is most notably the case for VGG16, which exhibits an F-score of 0.88 and an accuracy of 0.88 on the smaller database, and metrics of 0.70 and 0.88, respectively, on the larger database.
25 July 2006 in Journal of the American Chemical Society
We use molecular dynamics simulations to shed light on polymorph selection during the crystallization of the Lennard-Jones fluid. By varying pressure at fixed supercooling, we form large crystallites either of the stable face centered cubic form or of the metastable body centered cubic form and even fine-tune the fractions of stable and metastable polymorphs in the crystallite. We demonstrate that the conditions of crystallization, leading to large bcc crystallites, lie within the occurrence domain of the metastable bcc polymorph. We also find that the predominantly fcc crystallites contain a notable amount of the hexagonal close packed form, due to the cross nucleation of the hcp form on the fcc form. By varying temperature at fixed pressure, we prevent cross nucleation and form pure fcc crystallites. Our results reveal that polymorph selection may take place, and be controlled, during the growth step.
1 November 2006 in Journal of the American Chemical Society