introduction
For predicting compound retention time within capillary columns in gas chromatography (GC), it is necessary to define a model for retention factor k(T). In the most common approach, The relationship between retention factor and thermodynamic parameterrs (/_H) or (/S) is considered, as is the temperature program. >In many studies found within the available literature [1731] The estimation of parameters of the retention factor In GC model is performed by way of experimenting with different isothermal conditions, resulting in a linear regression problem that can be solved by least squares.
In other cases, although non-isothermal temperature programming is utilized, the method for solving the non—linear regression is not described in detail, and the optimization method used is not defined, nor is presented any information about convergence and the objective function.
This work, which is Part II of a series, demonstrates performances of different numerical methods in optimizing two different error criteria used for estimating parameters: the sum of square error (SSE) and the maximum error for absolute value (MEAbs).
Optimizations of each error criterion were tested using two types of initialization: one simple, and another one, called specialized which is what is being proposed here, departing from data and information gathered in Part I, which makes use of the format of the surface ofthe error criterion due to relevant thermodynamic parameters.
According to Part I, from knowledge of the surface of the error function, it is possible to place the points in the area with the high- est possibility for the global optimum to be found upon. This result is in line with the heuristic in optimization, that is, that there is a probability of convergence toward the global optimum the closer the value is to the location ofthe optimum. Additionally a reduction in the numberofobjective function evaluations is expected due to the proximity of the initial guess with respect to the optimum point.
The main objective of this series of works (Parts I and II) is to identify an efficient way of initializing parameter estimation: nonetheless, attention is also given to the physically interpretation of the estimated parameters, as well as to how the validation of the model of retention factor in the prediction of retention time, for temperature programs different from those used in estimation. is done.
The experimental data describedin Part l and used for parameter estimation was obtained from retention time measurements for a series of alkanes. Additionally, an equation was determined for the variation of dead time of the mobile phase with the temperature. Furthermore, additional data for validation ofthe method is shown in Section 5.