| C1 |

Appendix

Optimization Method for Determination

of Solubility Parameters
Some computer programs are organized to extract numerical meaning from a collection of measurements. Without
programming knowledge, one can do this with a spreadsheet. Actually one can use a spreadsheet to convert
measurements of solubility or compatibility between
solvents to values of solubility parameters. This appendix
provides an explanation of and directions for how to do
that, and in Appendix C2 the method is applied to
measured data.

A. METHODOLOGY OF AN
OPTIMIZATION
Optimization of multiple response processes requires an
answer to two simple questions: Just what is it that is being
optimized, and how? Here:

The ddisperse, dpolar, dhydrogen bonding, and RO

parameters1, are being optimized.
One performs an optimization by satisfying some
mathematical criterion, while varying the above
parameters. The optimization is complete when the
criterion is satised.

There are three requirements for conducting an

optimization:
(1) A denition of correctness.
(2) A mathematical criterion to know when as much

correctness as possible has been achieved.

1
These are the three solubility parameters and the radius of the
solubility sphere each representing all three intermolecular forces. Note
that this optimization scheme, though normally used with Hansen
Solubility Parameters, can be used with any three-dimensional set
of solubility parameters that is based on measurements of solubility. Hoy
Solubility Parameters are calculated from correlations of molecular
parameters and no measurements of solubility are involved.

(3) A method for varying the four parameters, ddisperse,

dpolar, dhydrogen bonding, and RO, so that they can

converge to satisfy that criterion.

A.1 DEFINITION OF CORRECTNESS

Correctness is evaluated by placing experimental solubility
data into the appropriate one of four groups: GOOD2 and
IN and BAD3 and OUT; or GOOD and OUT and
BAD and IN. IN or OUT placement is relative to
a three-dimensional sphere in solubility parameter space (this
is shown in two-dimensional space in Figure 2.42).
While one may certainly do so, it is not necessary to use
two- or three-dimensional plotting software to conduct
a solubility optimization. It is not necessary to do any
plotting (as in Figure 2.42) to identify an optimum. This
can be achieved non-graphically by tabulating the
outcomes and comparing them to the criterion.
Data sets to which this optimization process has been
applied can be found in this book as Table C2-1 in
Appendix C2, Table C3-1 in Appendix C3, and Table C4-1
in Appendix C4. These data are about the miscibility of
a material with one of many solvents or solvent mixtures.
Simply, the purpose of optimization is to locate and size
the solubility sphere so that it includes conditions of
solubility and excludes conditions of non-solubility. This
means that the volume of a right-sized and properly
located solubility sphere will encapsulate all solvents
which are deemed to be compatible and no solvent
deemed to be not compatible.
A GOOD solvent is one which is found to be compatible with
another solvent (or a polymer) to an acceptable degree.
A BAD solvent is one which is found not to be compatible with
another solvent (or a polymer) to an acceptable degree.

2
3

565

Appendix

C1

Optimization Method for Determination of Solubility Parameters

Conventionally, complete correctness is taken to be:

All data tabulated (or plotted) in three-dimensional

solubility parameter space are from either the GOOD
and IN or BAD and OUT groups, and
No data tabulated (or plotted) from the GOOD and
OUT, and BAD and IN groups.

One can quantify the lack of complete correctness by

noting how far in three-dimensional space4 a GOOD
point is plotted OUT of the sphere and how far a BAD
point is plotted IN the sphere.
For simplicity, lack of correctness in solubility data is
referred to as an error. So in Equation C1-1, which
denes an error, RO is the distance5 in three-dimensional
space from the center of the solubility sphere to its
boundary. And RA is the distance from the center of the
solubility sphere to the location of the center of the
location of the ith solvent.


Error i R A  R O i   fFACTORgi
C1  1
The parameter FACTOR is articialdit is used algebraically
to produce zero error values when the solubility data are
correctly tabulated (plotted), and non-zero error when they
are not. Such an implementation is noted in Table C1-1.
There can be as many Errori values as there are solvents in
the experimental protocol; the maximum value of i is the
number of solvents tested in the experimental protocol.

the way in which they describe the experimental data

upon which they are based6.
It has been conventional when determining solubility
parameters to use a metric called the desirability functionA. This metric is based on the idea that, for a product or
process that has multiple quality characteristics, it is undesirable to have one of them outside of some desired limit.
The method allows the coordinates of the center of the
solubility sphere (ddisperse, dpolar, dhydrogen bonding), and
radius (RO) parameter values to be determined which
provide the most desirable response valuesdthe best
description of the solubility data points.
While the desirability function can be used with nonbinary solubility data (Appendices C3 and C4), it is also
conventional to use binary data (Appendix C2) for the
optimization. This means that with an experimental solubility experiment, one assigns the number 1 if the solvent
was compatible with or dissolved the soil or polymer
satisfactorily (to a certain low level of residue or within
a goal time, for example), and the number 0 if there was
not complete compatibility or solution in adequate time.
The general format of the desirability function (DF) is
Equation C1-2, where the individual dfi values are those
which result from testing with each solvent, and i is the
number of dissolving tests. In Table C2-1, for example, i is
41 as there are 41 dissolving tests with different solvents,
which may also be called responses.
DF fdf 1  df 2  df 3  df 4  df 5 .  df i g1=i

A.2 MATHEMATICAL CRITERIA

TO DEFINE AN OPTIMUM
One simply needs a metric to do thatda standard of
performance about the optimization scheme. The
purpose of the metric is to allow recognition that the
optimum has been found; that movement of the coordinates of the center of the solubility sphere (ddisperse,
dpolar, dhydrogen bonding), and adjustment of the radius
(RO) of the solubility sphere parameters wont improve

This is extremely difcult to do visually in a two-dimensional graph.

Thats why the invention of the spreadsheet has made the technology
which Dr. Hansen invented 20 years earlier signicantly more easy for
him and others to practice.
Data points often appear within a projection of a sphere onto two
dimensionsda circledwhen in fact they are outside the circle. The
reason for this is that usually the disperse solubility parameter is not
specied in the two-dimensional plot, and that affects the location of the
data point in three-dimensional space.
Data in Table C2-1 shows that actual data arent as simple to
characterize as are the imaginary data shown in Figure 2.42.
Two-dimensional solubility plots, which ignore the disperse solubility
parameter, allow discrimination among solvents only when the disperse
solubility parameter of the polymer/soil/solvent being tested is essentially
constant among all tested solvents. This is seldom the case.
5
This is the radius of the three-dimensional sphere. If RA < RO, the
solvent is within the sphere; if RA > RO, the solvent is outside the sphere.
RA is calculated from Equation 2.8.

566

C1  2
The individual dfi functions can be any function of the
errors (Equation C1-1) which one wants to eliminate or
minimize through the optimization scheme.
Conventionally, again, a simple relationship with
Equation C1-1 is used to formulate the individual dfi
values. They are equal to the exponential function (exp) of
the error values, as shown in Equation C1-3.



df i h exp Error i  exp jR A  R Oi   fFACTORgi
C1  3
Equation C1-3 is simpler than it looks because when
a solvent is noted as being GOOD and IN or BAD
and OUT, the value of Errori is 0, and exp (0) has the
value of 17. Only solvents noted as being GOOD and
OUT, or BAD and IN have non-zero error values and
hence non-unity dfi values.

In curve-tting by linear regression, the metric is the sum of the

squares of the deviations between the experimental data and values
calculated using the curve-t equation.
7
The reason one doesnt use the error values directly in Equation C2-3 is
that when any single solvent is correctly tabulated/plotted relative to
a soil/polymer/solvent, the error is zerodso the overall error function is
zero. When the overall error function is zero, further optimization is
impossible. Only the fact that exp (0) 1 allows the optimization to be
completed.

Optimization Method for Determination of Solubility Parameters

C1

Appendix

Table C1-1 Denition of Solubility Errors

Outcome of
Solubility
Experiment

Position of Solvent Relative

to Solubility Sphere

GOOD Solvent
GOOD Solvent
BAD Solvent
BAD Solvent

Value of
Error
9

Inside

Outside

(RO  RA)
88

Inside

Outside

(RA  RO)

Combining Equations C1-2 and C1-3, one gets the

operating Equation C1-4 (shown only for three terms). This
is the equation which will be used for the optimization.
DF fexp jR A  R O j  fFACTORg1 g
fexp jR A  R O j  fFACTORg2 g .
fexp jR A  R O j  fFACTORgi g1=i
C1  4
When the optimization has been completed, by the
spreadsheet function described in Footnote 10, DF has the
value 1dor just slightly less as is possible. If any dissolving
test is not correct, its dfi value is <1, and so the overall DF is
also <1.

A.3 METHOD OF VARIATION OF

PARAMETERS
The mathematical technique of linear programming is used to
solve multi-variable optimization problems such as identifying the best (when DF 1) location of the solubility
sphere (ddisperse, dpolar, dhydrogen bonding), and radius (RO)
parameter based on a set of solubility experiments.
Developed during World War II to plan expenditures and
returns in order to reduce costs to the army and increase
losses to the enemy, linear programming found postwar
applications in such diverse situations as renery optimization, food blending, inventory management, portfolio
and nance management, resource allocation for human
and machine resources, and design of scientic experiments.
While its mathematical derivation and explanation is
beyond the scope of this book, its power is easily harnessed
by readers. A set of sophisticated linear (and non-linear)
programming algorithms is a part of current spreadsheet
software products8.

Chapter 2, Footnote 69 has a more detailed description.

10

11

Value of
FACTOR

Comments

Correct Placement

False Positive

Correct Placement

False Negative

The word linear in linear programming basically

means no surprises (Box C1.1).

Box C1.1
This optimization technology will nd a global optimum as long
as the relationships among the four parameters and DF are linear.
Mathematically, the word linear does not mean straight line. It
does not mean that there cant be curvature in the relationships
between the four parameters underlying DF. Linear means that
the relationships are continuous, proportional (although the
proportionality constant may change), and additive.
In fact, these conditions of linearity do not have to be strictly
followed with the powerful optimization routines available in
modern spreadsheets. One can apply constraints (limitations
placed on trial values of the four parameters). One can specify the
degree of precision of the answer and the maximum number of
iterations to be made. And one can also specify which of the
linear or non-linear goal-seeking algorithms is to be used.
However, absence of linearity in the relationships between the
four parameters and DF can mean that there are multiple optima.
And only one of which will have the value of DF closest to 1.
Different starting points (parameter values) can produce
different optima, without noticing that there are multiple optima.
This author has experienced this outcome with some sets of
experimental data. Thats why he uses the same initial guesses
for every optimization run. If multiple optima are found, the only
valid one in terms of selection of solubility parameters is one in
which DF equals exactly 1. If an optimum is found by the
optimization routine where DF has any other value than 1, choose
another set of initial guesses. If there are no optima found where
DF has the value of 1, choose the optimum which has the DF value
closest to 1.

9
If a solvent is correctly positioned relative to the solubility sphere, it
matters not what are the relative values of RA and ROdthe error is zero
(and FACTOR 0).
10
The units of all errors are the units of solubility
parametersdconventionally MPa1/2 in this book.
11
By the conventions in place in Table C1-1 for differences in solubility
distance, the values of the desirability function (DF) will be between
0 and 1, and cant exceed 1. Should the opposite conventions be
chosen, DF will be above 1, or exactly 1. In either case, the outcome of
the optimization will be unaffected.

567

Appendix

C1

Optimization Method for Determination of Solubility Parameters

The algorithms require initial or trial guesses about

values of the parameters to be found in an optimization.
From these initial values supplied by the user, the linear
programming algorithms apply small and randomly chosen
levels of change, in each of the six possible directions, to each
of the four parameters. The algorithms then note which
outcomes reduce or increase the DF valuedas calculated
from Equation C1-4. In other words, the proposed direction
of an optimum from the initial values is learned.
The process, repeated many many times, is called goalseeking. Subsequent trial values of all four parameters are
proposed and evaluated by the algorithms based on
previously calculated results and a denition of the goal of
the optimization.
When no improvement (increase of DF toward 1) is
possible, or an integer value of 1 for DF is reached, the
optimization algorithms cease calculation and report their
outcome. Either an optimum was produced, with DF 1,

568

or no feasible solution found, or progress toward solution changes too slowly.

In this book, the dened optimization goal is an
integer value of 1 for DF where there are no experimental
data noted as being GOOD and OUT or BAD and
IN. When DF has the value of 1, all solvent data are
GOOD and IN or BAD and OUT relative to the
sphere whose center coordinates for the soil/polymer/
solvent are ddisperse, dpolar, dhydrogen bonding, and whose
radius is RO.
This author has had good success using initial guesses
of ddisperse, dpolar, dhydrogen bonding, and RO of 15, 5, 5, and
5 MPa1/2 respectively.

Endnotes
A. Harrington EC, Jr. The Desirability Function. Industrial Quality Control.
1965;21(10):494e498.