Synthetic Parameter Analysis Of Open-framework Aluminophosphates Based On Dimensionality Reduction

Since the first successful synthesis of the aluminophosphate molecular sieve AlPO4-n byWilson et.al in1982, open-framework aluminophosphates (AlPOs) have become an importantfamily of porous crystal materials. Because of the diversity of pore structure and elementscomposition, open-framework aluminophosphate materials have presented potentialapplication value in catalysis, adsorption and separation. However, the synthesis of this kindof materials is affected by many parameters, such as reaction material, gel composition,solvent, template agent, crystallization temperature and crystallization time etc. Thecrystallization mechanism is so complicated that makes the rational synthesis verychallenging.In order to understand the formation mechanism of the open-frameworkaluminophosphate materials, the State Key Laboratory of Inorganic Synthesis and PreparativeChemistry of Jilin University established an open-framework aluminophosphate synthesisdatabase. Based on the open-framework aluminophosphate synthesis database, the work inthis paper concentrates on analyzing and studying the synthetic parameters for rationalsynthesis of open-framework aluminophosphates by dimensionality reduction technology.From the class and class hierarchy points of view, we propose three methods based on featureextraction and feature selection technologies for analyzing the effect of synthetic parameterson the final products, and studying the relationships between the synthetic parameters andfinal products. The research work can provide directive opinions for the rational synthesisexperiments which may improve the success rate of the synthesis. The primary contributionsof this paper are presented as following aspects:1. Feature extraction could find the valuable information which hinds in thehigh-dimensional data space. In order to probe the effect of the synthetic parameters on therational synthesis of the final products, we propose a feature extraction algorithm based on2,1-norm (FEA-2,1) by using2,1-norm on the projection matrix of the classical featureextraction methods. Due to applying2,1-norm on the projection matrix, the projection matrixcould achieve row sparsity which increases the interpretability of feature extraction resultsand reaches the goal of analyzing the features of data from the feature extraction point of view.We apply FEA-2,1in the open-framework aluminophosphate synthesis database, and analyzethe importance degree of various synthetic parameters for rational synthesis of(12,6)-ring-containing,(10,6)-ring-containing,(8,6)-ring-containing and8-ring-containingopen-framework aluminophosphates.2. There are serious correlation relationships between the synthetic parameters in theopen-framework aluminophosphate synthesis database. Based on maximum weight andminimum redundancy criterion, Maximum Weighted and Minimum Redundancy (MWMR)feature selection algorithm could select a subset in which the importance of features ismaximizing and the correlation among features is minimizing. For the purpose of consideringthe importance of the features and the correlation between the selected featuressimultaneously, we apply MWMR in the open-framework aluminophosphate synthesis database, and make deep analysis of the relationship between the synthetic parameters and therational synthesis of (12,6)-ring-containing,(10,6)-ring-containing,(8,6)-ring-containing and8-ring-containing open-framework aluminophosphates.3. In the open-framework aluminophosphate synthesis database,(12,6)-ring-containingand (8,6)-ring-containing open-framework aluminophosphates can be divided into severalsubclasses due to different topology structures. We propose a Hierarchical Feature SelectionModel (HFSM) and a Subclass Discriminant Analysis Score method for feature selection ofsubclass problem. We apply HFSM and SDAS in studying the relationship between thesynthetic parameters and the rational synthesis of (12,6)-ring-containing and(8,6)-ring-containing open-framework aluminophosphate subclasses. In accordance with theexperiment results, we make an in-deep analysis of the synthetic parameters for rationalsynthesis of AFI, ATO, ATS, FAU, AEN, AWO, ERI, and CHA.