Jean-Jacques Legendre

Prediction of the Impact Sensitivity by Neural Networks

A method for optimizing the prediction of impact sensitivity of explosive molecules by neural networks is presented. The database consists of 204 molecules of known sensitivity, containing C, H, N, and O and belonging to several chemical families. Pertinent molecular descriptors are selected by a preliminary evolutionary multiple linear regression treatment, and the effects of the networks topology and the extent of the training are examined and optimized. The predictions are satisfactory with a correlation coefficient R = 0.94 obtained through cross-validation. The neural networks approach proves more accurate than linear methods and more general than all previously used methods.

NEURAL NETWORKS PREDICTION OF PARTITION COEFFICIENTS

Abstract A comparison has been made of the relative ability of neural networks and regression analysis to estimate the octanol—water partition coefficients (log P ) of organic compounds, with a recently proposed model based on geometric and semi-empirical AM1 parameters. The predictive power of the model is estimated by cross-validation. The regression analysis requires the use of a 17-parameter model including higher powers of a 8-parameter model, giving a standard error of prediction (SEP) of 0.371. The neural network approach gives comparable prediction (SEP = 0.379), even with the reduced 8-parameter model. With a 13-parameter model obtained by the addition of five new independent parameters, neural networks are found to give still better results (SEP = 0.300). The variability of the prediction made by neural networks has been related to the leverage of the compounds in the descriptor space of the multilinear model.

Computational Chemistry: A Way To Reach Spectroscopic and Thermodynamic Data for Exotic Compounds

In this paper, we report a general methodology to reach thermochemical properties of molecules in gas phase. We applied these calculations to related complexes of industrial interest, concerning the electrowinning of aluminum. This thermochemical prediction is a demonstration of a quantitative analysis after geometry optimizations and frequencies calculations from density functional computations (or any first-principle techniques). Since this methodology is completely generic, we can study any complexes of interest in order to study the model of the vapor. We have especially investigated the (Na3AlF6, Na2AlF5, NaAlF4, AlF3, NaF, (NaF)2, (NaAlF4)2) system. It has been found that NaAlF4 and (NaF)2 are the major species present in gas phase. Results concerning calcium are also presented, showing that CaAlF5 is in the vapor. Its partial pressure decreases when the one of CaF2 increases and the partial pressure of NaAlF4 (and all minor species containing aluminum) decreases at the same time.

A new method for the structural modelling of disordered compounds—application to molten NaCl

Abstract This paper describes a new method for limiting structural models during modelling procedures on disordered compounds. Instead of using the well known periodic boundary, the data derived from neutron diffraction experiments are modified to take into account a spherical limitation of the model. Compared to cubic periodic modelling, this method reduces the computation time and improves model reliability since it introduces no artificial anisotropy. The method has been tested in the structural modelling of molten NaCl.

Computational and analytical chemistry: Methodology to study chemical reactions between sodium, calcium, and aluminum fluorides in molten cryolite

Reaction constants and composition profiles in molten cryolite have been theoretically investigated. Hartree-Fock and density functional calculations were applied to determine the exact nature (structure and energetic) of complexes existing in molten cryolite. The aim of this work was thus the understanding of chemical processes occurring in the electrowinning of aluminum and was a demonstration of how computational chemistry (based on density functional theory) can help us to determine structures and reaction energies in particularly complex medium such as cryolite. An analytical study, based on mass balance and equilibrium constants has been undertaken. This was performed on molecular liquid entities taking into account the four-, five- and sixfold coordinated aluminum complexes of the AlF3-3NaF melt system. Moreover, the effect of calcium has been studied by substituting two sodium atoms with one calcium atom, thus leading to the CaNaA1F6 system. Two conformers (instead of three for Na) were obtained for this system. They can be described as representing the four- and fivefold coordinated aluminum complexes in molten cryolite. The structurizing effect of calcium was clearly illustrated by the resulting optimized structures, showing that calcium stabilizes the IV and V coordinations of aluminum. By computing reaction constants, we have obtained composition profiles that are presented with those based on experimental data. Comparisons point out that computational chemistry techniques match with experimental results, especially in the case of pure cryolitic melts. For the presence of the fivefold coordinated aluminum complex in cryolite, and the predominance of the fourfold coordinated complex with calcium, it is clear that these computational techniques show us correct trends in predicting the main species in molten media.

Advantages and drawbacks of the quantum chemistry methodology in predicting the thermochemical data of lanthanide trihalide molecules

Abstract DFT calculations have been carried out on six lanthanide trihalide molecules, using Stuttgart ECPs on the lanthanides and a 6-31G * all-electrons basis set on the halogens. These calculations have been compared to previous theoretical and experimental data. Calculated bond lengths and vibrational frequencies are found to be in good agreement with previous theoretical results. The variations observed between experimental and theoretical results are discussed. Therefore, thermochemical data have been derived from such calculations. In this paper, the advantages and drawbacks of this methodology are discussed in the framework of the selected heavy elements. The different factors which influence the estimated thermochemical data have been compared. While molecular parameters seem to have a weak influence on the thermochemical functions, low frequencies (flat energy surfaces along the out-of-plane distortion) and the electronic partition function have been revealed to be the main factors for the accuracy of the predicted thermochemical data.

Prediction of the extraction energy of an atom from the surface of Fe–Cr alloys using topological descriptors

Abstract A 3D model for simulating the passivation of iron–chromium alloys has been developed previously. This paper describes the attempts to estimate the probabilities of dissolution used in the model from the energy of extraction of Fe or Cr atoms according to their chemical environment. Quantum chemistry calculations have been performed to estimate the energy of extraction of Fe or Cr from the cluster for various topological environments of the atom to be extracted. Using the obtained results, a simple multilinear correlation between the energy of extraction of an atom and its environment has been developed. Such a relation will allow us to do more rapid calculations of the energy of extraction for any topology, which can be included in the simulation. The correlation used topological descriptors describing, in a relevant way, the environment of an atom to be extracted. At first, we tested the descriptors connected to the location of the extracted atom in relation with its neighbors. Then, we took into account the parameters describing only the distribution of the neighbors among them, which leads to the autocorrelation function of the cluster. Finally, we combined these two types of descriptors.

GEOMETRICAL ANALYSIS OF THE VOIDS IN STRUCTURAL MODELS OF MOLTEN SALTS

Abstract A new method for analyzing the shapes of the voids in structural models of liquids is described. Each void is approximated by a cluster of elementary cubes the inertial ellipsoid of which provides information about the dimensionality of the void (sphere, disk or rod). A plot of this data vs the volumes of the voids provides a useful insight on the types of voids in such models which may be connected with various melt properties or may be used to check the reliability of the modeling methods. The analysis of some models of molten salts is given as an example.

Density functional calculations for predicting structures and vibrational frequencies of aluminum chloride-fluoride complexes

Abstract Quantum chemical calculations are used to study AlCly−xFx3−y (y = 5 or 6, x = 0,…,y) species that can occur in aluminum electrorefining melts. These theoretical studies are included in a wider research program concerning the chemical instabilities in the bulk of molten salts during the refinement process. Stabilization energies, equilibrium geometries and vibrational frequencies of the complexes are calculated using the Delley functional methodology described in Ref. [1] (B. Delley, J. Chem. Phys., 92 (1990) 508). These computational simulations, discussed and compared with the experimental results demonstrate that density functional calculations can be reliably used in the study of complexes existing in molten salts. Quantum chemical calculations are accurate tools for theoretically predicting structures, physical and chemical properties and vibrational frequencies of known entities as well as unknown compounds.

Complexes responsible for ionic transport in LiCl‐KCl eutectic melt

DFT calculations using DMol software at the DNP/UHF/BLYP level were firstly performed in order to determine the more stable isolated chloro‐complexes of lithium and/or potassium cations. An Inverse Isotropic Monte Carlo technique was then used for obtaining a 3‐D model of this melt from the pair correlation functions computed by Lantelme and Turq. At last, DFT calculation derived complexes were looked for in the melt model using a lab‐made software. This procedure showed the presence of LiCl2− and K2Cl+ complexes in molten KCl eutectic. This procedure allows a better understanding of the structure of melt and suggests answers to some phenomena that occur in molten salts such as the Chemla effect.