Group contribution methods

Group contribution methods (also fragment methods or increment methods ) are a widely used method in technical chemistry for estimating material data.

Procedure

Principle of a group contribution method

Chemical properties that are required in process simulation , for example, are always properties of a substance or a mixture of substances. Since there is an almost infinite and exponentially increasing number of pure substances and mixtures, group contribution methods have been developed that no longer assign substance properties to whole substances, but to fragments.

The effect achieved is that from a few group properties, typically a dozen to a few hundred, the material data for many thousands of substances and their mixtures can be determined.

These fragments (the groups ) are generally the functional groups of a molecule , such as the hydroxyl group (-OH), the amino group (-NH ₂ ) or the carboxy group (-COOH). Often other molecular features are also added as groups, such as ortho / meta / para positions on aromatics , ring sizes and chain lengths.

Basic properties

The property you are looking for is calculated as a function of the sum of the group contributions : ${\ displaystyle X}$ ${\ displaystyle G_ {i}}$

{\ displaystyle X = f \ left (\ sum G_ {i} \ right)}

Some approaches to the estimation of basic substance quantities are estimated directly from the sum of the group contributions. In many cases, however, the sum of the group contributions does not describe the substance size sought, but only a calculation variable with which the substance size sought is correlated. In addition, there is often a correlation with other substance quantities. For example, when calculating the critical temperature , the normal boiling temperature is usually used as a further input parameter: ( correlation equation according to Joback and Reid ) ${\ displaystyle T _ {\ text {c}} = T _ {\ text {b}} \ left [0 {,} 584 + 0 {,} 965 \ sum {G _ {\ text {i}}} - \ left ( \ sum {G _ {\ text {i}}} \ right) ^ {2} \ right] ^ {- 1}}$

Mixed properties

In models that estimate the properties of mixtures, not only the sums of the group contributions are used, but group interaction parameters and are used. ${\ displaystyle G_ {ij}}$ ${\ displaystyle G_ {ji}}$

{\ displaystyle \ gamma = f \ left (\ sum G_ {ij} \ right) + f \ left (\ sum G_ {ji} \ right)}

One property that is typically calculated by group interaction models such as UNIFAC or ASOG is the activity coefficient . ${\ displaystyle \ gamma}$

One negative impact of using group interactions is the massive increase in the parameters required. For example, interaction parameters are already required for 10 groups . Therefore, group interaction models are usually not completely parameterized. ${\ displaystyle 2 \ cdot 45}$

Determination of group contributions

The group contributions are usually adapted directly to experimentally determined material data using multilinear or nonlinear regression . Nonlinear regressions generally represent multimodal optimization problems , i.e. optimization problems with more than one optimum in the solution space under consideration . Evolutionary algorithms (e.g. (nested) evolution strategies, genetic algorithms, etc.) are therefore often used to adapt group interaction parameters , since deterministic optimization algorithms are usually not able to find the global optimum (in the case of regressions: minimum).

As a database of experimentally determined material data, z. B. fact databases such as Beilstein , the Dortmund database or the DIPPR 801 database. Often, experimental measurements are also carried out as a supplement if there are gaps in the considered group interaction matrix or if group contribution methods additionally describe a temperature and / or pressure dependency.

Assessment of the accuracy

The predictive accuracy of a group contribution method is influenced by two factors: the accuracy of the reproduction of the experimental data by the group contribution method and the accuracy of the underlying experimental data.
When testing the quality of a prediction, usually only the difference between the prediction and experimental data is taken into account. It is crucial that the comparison also takes place with external data. For many group contribution methods (such as the Joback method ), only the accuracy of the representation of the experimental data used for method development is given in the relevant publications. The problem here is that the parameters of the method (e.g. the group contributions) were adapted to precisely these data. So they were optimized for exactly this data set. The prediction errors obtained therefore often do not reflect the actual accuracy.
In order to be able to make a reliable statement about the accuracy and reliability of the prediction, an external validation is necessary. For this purpose, part of the available experimental data (the so-called test set ) is usually removed from the database before the method development begins . With the help of the remaining data (the so-called training set ), the method is then developed and the parameters are adjusted. The method is then applied to the substances in the test set and the corresponding error is calculated. Only an error found for a test set should be viewed as the accuracy of a group contribution method (or analogous to a QSPR model). Otherwise the accuracy of the prediction must be considered unknown.

literature

Bruce E. Poling, John M. Prausnitz, John P. O'Connell, " The Properties of Gases and Liquids, " McGraw-Hill Publishing Co., 5th Edition.
Jürgen Gmehling , Bärbel Kolbe, “Thermodynamics”, Georg-Thieme-Verlag, 1988.
Peter Ulbig, group contribution models UNIVAP & EBGCM, Gelsenkirchen, 1996, ISBN 3-920088-70-0 .
Hannes Geyer, Development and investigation of group contribution methods for the prediction of thermodynamic material quantities using Computational Intelligence methods, Shaker, Aachen, 2000, ISBN 3-8265-7000-6 .

Individual evidence

↑ A. Tropscha, P. Gramatica, VJ Gombar: The Importance of Being Earnest: Validation is the Absolute Essential for Successful Application and interpretation of QSPR Models . In: QSAR & Combinatorial Science . 22 , No. 1, pp. 69-77. doi : 10.1002 / qsar.200390007 .

Web links

[1] A. Tropscha, P. Gramatica, VJ Gombar: The Importance of Being Earnest: Validation is the Absolute Essential for Successful Application and interpretation of QSPR Models . In: QSAR & Combinatorial Science . 22 , No. 1, pp. 69-77. doi : 10.1002 / qsar.200390007 .