Appendix

C8

Estimation of HSP from Group Contribution Methods

Is used throughout all ranges of values for all three HSP

Is based on correlation with a neural network vs. linear
regression, and
Is embodied in a commercial product12 and not
a scientic paper.

H. COMPARISON OF ESTIMATION
METHODS FOR HSP

Figure C8-9 Hydrogen Bonding Solubility Parameter

The ability of this correlative approach to predict two of

three HSP values is strikingdespecially for the hydrogen
bonding HSP whose values are environment-dependent
(Appendix C5, Section C.2)!
Compare the results of the Y-MB (HY) method to
that of the EOS method (Appendix C7dFigures C7-3
[disperse], C7-4 [hydrogen bonding] and C7-5 [polar]).
The Y-MB approach is statistically superior for both the
disperse HSP (SS/Average of 2.35 vs. 6.38) and hydrogen
bonding HSP (SS/Average of 2.89 vs. 16.87).
However, the ability of the Y-MB approach to predict
the polar HSP is not useful relative to the EOS approach
(SS/Average of 77.14 vs. 1.30). The comparison is plotted
in Figure C8-10.
As opposed to the SP group contribution approach, this
approach:

Uses only one set of parameters for all levels of an HSP

Can describe molecules of one and more carbon atoms

Figure C8-10 Polar Solubility Parameter

642

It isnt that the conventional (Joback or Lyderson) or

the SP approaches are invalid or inferior. Rather the
contents of this appendix are an excellent example of
technology growthdof developers building on previous
technology. The EOS and Y-MB (HY) work were
apparently not started until after the SP work was
published.
Outcomes of using the four calculation schemes are
compared in Table C8-10. Note that the comparison is
awed since it is not based on tests of the same slate of
cleaning solvents13.
The statistic used for comparison is the sum of squares of
the difference between the reference HSP and that
computed by the method under consideration, divided by
the average reference HSP14, and further divided by the
number of values in the dataset.

12

This is the reason for the group contribution constants not being
provided here as with the Stefanis Panayiotou method in Tables C8-4
through C8-6.
13
The conventional approach was used with 53 data points because this
was the number of chemicals that could be considered useful as cleaning
solvents, and for which there were credible values of dipole moment
available. The EOS approach was used with 66 chemicals, but not all
these chemicals were the ones used with the conventional approach.
Although the EOS paper by Panayiotou, noted in Appendix C7, shows
the application of his method to signicantly more than 53 chemicals,
many are hydrocarbons whose polar and hydrogen bonding HSP are
zero and others are chemicals of little interest to those doing cleaning
work. So, this author truncated the list of chemicals considered in
Panayiotous paper. Both group contribution methods were evaluated
with the same dataset of cleaning solvents.
14
The desire to use a common statistic for the comparison in Table
C8-10 of all three HSP was the factor inuencing the choice of that
comparative statistic. There were differently sized datasets, and natural
differences in their ranges of values. It was found that percentage
difference was inadequate to evaluate prediction of polar and hydrogen
bonding HSP. For example, prediction of a value of 0.5 MPa when the
true value is 1 MPa constitutes a 100 percent error, yet is a remarkable
achievement given that either polar or hydrogen bonding HSP can have
values from 0 to above 20. The sum of squares of the differences is
somewhat normalized by dividing by the average value of the considered
HSP and further dividing by the number of values in the dataset.
In the gures noted in Table C8-10, the comparative statistic was the
sum of squares of the differences divided by the average value of the
considered HSP. Since the datasets for each set of gures produced by
a common evaluation scheme were the same, there was no need to
divide by the average number of values in the dataset.