(747c) Computer-Aided Reaction Solvent Design Integrated with New Reaction Mechanism Model

Authors: 
Liu, Q., Dalian University of Technology
Zhang, L., Dalian University of Technology
Liu, L., Dalian University of Technology
Du, J., Dalian University of Technology
Meng, Q., Dalian University of Technology

With the development of process industry, solvents are widely used in chemical manufacturing processes, such as extraction, absorption, crystallization and reaction. When involved in liquid homogeneous-phase kinetic reactions, solvents can have significant impacts on reaction rate and selectivity to facilitate the existing synthetic routes or even develop new routes [1]. Thus, the choice of reaction solvents is particularly important to product yield and economic benefit. However, it is impossible to perform thousands of extensive and costly kinetic experiments for screening reaction solvents based on trial and error method or qualitative chemical knowledge. Therefore, efficient theoretical methods should be developed to help the selection of reaction solvents. A very promising avenue of research in this direction is the development of Computer-Aided Molecular Design (CAMD) technique, which has been successfully adopted for the optimal design of solvents in the last three decades [2]. It offers a systematic methodology that numerous of chemical species can be assembled by a predefined set of functional groups, and then Quantitative Structure-Property Relationship (QSPR) method, such as Group Contribution (GC) method, was constructed to formulate the molecular group structures with macroscopic solvent properties for establishing mathematical optimization models. Although CAMD had contributed a lot to the solvent design for separations, the work in reaction systems was still immature comparatively. On the one hand, pure Quantum Mechanics (QM) calculations integrated with CAMD have been performed for the reaction rate constant estimations without any experimental rate constant. However, large quantitative deviations often occurred due to the defect of Conventional Transition State Theory (CTST) and Continuum Solvation Model (CSMs). On the other hand, some empirical models, functioned as surrogate models, have been proposed to correlate some experimental reaction rate constants with solvent properties [3] or QM-based descriptors [4], but the model accuracy and universality can hardly be guaranteed simultaneously for the reaction solvent design. Therefore, reliable reaction kinetic models are urgently in need on the basis of reaction mechanisms.

In this paper, an optimization-based framework is developed for the reaction solvent design. A new reaction mechanism model has been established using the hybrid method of CTST derivation, knowledge-based selection and model evaluation, in which rigorously thermodynamic deduction was first performed to formulate the primary reaction mechanism model consisting of the infinite dilution activity coefficients and solvent molar volumes. The state-of-the-art COSMO-SAC (COnduct-like Screening MOdel-Segment Activity Coefficient) model is employed for the calculation of activity coefficients efficiently. Then, according to the solvatochromic equation [3], knowledge-based selection method was used to add the rate-related descriptors to our primary equation for model correction. Afterwards, optimal linear regression equation with the largest adjusted R square was identified through model evaluation, in which the significances of all model descriptors were tested again to overcome the overfitting problem. This hybrid method was demonstrated by two case studies, namely Diels-Alder and Menschutkin reaction, and impressive consistency of the results was found out that the infinite dilution activity coefficients, hydrogen-bond donor, hydrogen-bond acceptor and solvent surface tension are the most influential descriptors to the reaction rate constant. The multiple linear regression results of determination coefficient (R2) and Average Absolute Percent Error (AAPE) between experimental logk and predictive logkare illustrated in Table 1, where the results of Zhou and Folić's methods are also exhibited for comparisons. As we can see from Table 1, our method shows higher R2 and lower AAPE, which indicates that the reaction mechanism model proposed in this paper exhibits appealing characteristics with wide universality and high accuracy simultaneously.

Table 1. Regression results of different methods for comparisons

Method

Diels-Alder reaction

Menschutkin reaction

R2

AAPE

R2

AAPE

Our method

0.962

1.83%

0.911

8.07%

Zhou's method [4]

0.923

2.57%

0.292

20.22%

Folić's method [3]

-

-

0.654

18.75%

Later GC method was integrated for the prediction of above descriptors. CAMD technique was incorporated with the reaction mechanism model for the optimal reaction solvent design. With an Mixed-Integer Non-Linear Programming (MINLP) problem formulated, decomposition-based solution algorithm was employed to reduce the difficulty in dealing with the nonlinear COSMO-SAC equations. Ultimately, the promising solvents with double bond and carboxyl groups were identified for Diels-Alder reaction, while nitryl, azyl and nitrile groups for Menschutkin reaction. Experimental verification will be further carried out in our future research.

References:

[1] Struebing, H; Ganase, Z; Karamertzanis, P. G; Siougkrou, E; Haycock, P; Piccione, P. M; Armstrong, A; Galindo, A; Adjiman, C. S. Computer-aided molecular design of solvents for accelerated reaction kinetics. Nature Chemistry. 2013, 5(11), 952-957.

[2] Zhang, L.; Babi, D. K.; Gani, R. New Vistas in Chemical Product and Process Design. Annual Review of Chemical and Biomolecular Engineering. 2016, 7(1), 557-582.

[3] Folić M; Adjiman C. S; Pistikopoulos E. N. Computer-Aided Solvent Design for Reactions: Maximizing Product Formation. Industrial & Engineering Chemistry Research. 2008, 47(15), 5190-5202.

[4] Zhou T; Lyu, Z; Qi Z; Sundmacher, K. Robust design of optimal solvents for chemical reactions—A combined experimental and computational strategy. Chemical Engineering Science, 2015, 137(2), 613-625.