Extrapolation Practice predictions

But given our discussion above, he should not be. So in this case it is only surprising that he is able to extrapolate the predictions - we think that it is inevitable, since he has found a way to utilize only those wavelengths where nonlinearity is absent. Now what we need are ways to extend this approach to samples more nearly like real ones. And if we can come up with a way to determine the spectral regions where all components are linearly related to their absorbances, the issue of not being able to extrapolate a calibration should go away. Surely it is of scientific as well as practical and commercial interest to understand the reasons we cannot extrapolate calibration models. And then devise ways to circumvent those limitations. 

Predicting the combined effects of a mixture from knowledge of the effects of its components requires a reference model of what to expect for a mixture. Reference models used in mixture extrapolation practice are typically based on pharmacodynamic assumptions on the type of interaction between a chemical and a biological system. 

In addition it should be noted that the predicted extrapolations themselves are biased, even when no alteration in the signal of an addition has occurred. Hence, we discourage the use of the extrapolation mode with the standard additions method and propose interpolating the signals instead. We cannot recommend the combination of extrapolation practices with the use of equations related to interpolation. 

Aside from the fundamentals, the principal compromise to the accuracy of extrapolations and interpolations is the interaction of the model parameters with the database parameters (e.g., tray efficiency and phase eqiiilibria). Compromises in the model development due to the uncertainties in the data base will manifest themselves when the model is used to describe other operating conditions. A model with these interactions may describe the operating conditions upon which it is based but be of little value at operating conditions or equipment constraints different from the foundation. Therefore, it is good practice to test any model predictions against measurements at other operating conditions. 

To summarize, in the field of lyotropic cholesterics formed by helical polymers of known geometry, it is possible to predict qualitatively only the entropic steric contribution to the cholesteric twist The prediction relies on the model of Figure 7.4 and should be compared to the experimental Sq data obtained by extrapolation of the twist to l/T = 0. A practical model for calculating the often dominant dispersion contribution is not yet available. 

The problems particular to accelerated tests are related to the extrapolation process. It was stated earlier that it is essential that extrapolation rules from the test conditions to those of service are known and have been verified. In practice this is only an ideal as extrapolation procedures have not generally been comprehensively validated and almost certainly will not give accurate predictions in all cases. The only choice is to use the best techniques available and apply them with caution. 

Before there can be any extrapolation there must be confidence in the model or rules being used. In practice this often has to involve an element of faith because of lack of validation data, particularly where the rule is empirical. The theory or model should be no more complex than is necessary to fit the data. The accuracy of fit to, for example, a creep curve can often be made more precise by applying ever higher order polynomial expressions, but outside the range of points these functions diverge rapidly to infinity (or minus infinity) leading to predictions that are ridiculous. 

In terms of time, for high temperature steels it is conventional to restrict extrapolation to a time factor of 3, although years of practical experience are allowing this to be extended as far as 10. ISO 2578 [4] recommends no more than a factor of 4, from 5,000 to 20,000 h. A maximum value of 100 (two orders of magnitude) is permitted for pipes with the restriction that the temperature difference is limited to 60 °C for polyolefins and to 25 °C for PVC. General advice is no more than a factor of 10. Figures 9.1 and 9.2 show instances of somewhat extreme prediction far into the future of the human race. 

In summary, practical experience with predicting the hydrolytic degradation of polyethylene terephthalate is an example of the use of Arrhenius extrapolation, a demonstration of the problems encountered when there are changes in the state of the polymer as the temperature is raised, and an example of the large variability in prediction of lifetime due to the logarithmic scale. 

The thermodynamic theory of solutions is complete in the sense that the exact relations among thermodynamic coefficients are all known, the Gibbs-Helmholtz equation for example. However in practice it commonly is necessary to make predictions on the basis of incomplete data, therefore to make extrapolations and other approximations. Reliable approximations depend upon a knowledge of the solution structure. 

Contrary to the practical results reviewed above, statistics from correlation work revealed a serious deficiency in the accuracy with which Phase I Equations 3 and 4 predicted -for the Phase II dataset r for Equation 3 predictions for the 103 compound Phase II data was only 0.45 r for Equation 4 predictions for the Phase II dataset was only 0.44. An analysis of the residuals for the Phase II dataset [Potency(observed)-Potency(predicted by Phase I models)] immediately Identified the source of the problem of the 26 Phase II compounds having DICARB >4, 17 had potency for adult observed more than one log unit better than predicted 15 had egg potency observed more than one log unit better than predicted. As schematically shown in Figure 2B, the parabolic functions for DICARB for the Phase I models underpredict at values of DICARB extrapolated beyond those represented in the Phase I dataset. 

A rough prediction of desorption and recovery can be made by extrapolation from similar compounds however, until we know more about the factors involved, this practice can lead to difficulties as illustrated in Table II. The recovery of naphthalene and biphenyl from charcoal are much lower than one would predict from the recoveries of other aromatic compounds, and this stronger attraction may be related to bond angles or steric effects. The order of recovery, ortho (84%) > meta (81%) > para (76%) methyl biphenyl, supports this theory. One may predict from this information that the recovery of benzene would be very poor, which is not the case. 

There is no widely used detector that has predictable absolute response factors. This is why predicted rather than calibrated response factors are only used when calibration is not practical, such as for unknowns, or when a closely related compound has been calibrated, allowing extrapolation to the desired response factor. 

Up to 1970, it was thought that the practical limit of the periodic table would be reached at about element 108. By extrapolating the experimental data on heavy-element half-lives, we concluded that the half-lives of the longest-lived isotopes of the heavy elements beyond about element 108 would be so short (<10-6 s) due to spontaneous fission decay that we could not produce and study them (Fig. 15.10). However, in the late 1960s and early 1970s, nuclear theorists, using techniques developed by Vilen Stmtinsky and Wladyslaw Swiatecki, predicted... 

---