Appendix

B2

Calculation of MIR through Group Contribution Methods

Professor Carter1 believes that:

For compounds that react with similar mechanisms, there
may be a correlation between MIR and the OH rate
constant, and thus in that way group contribution methods
may estimate MIRE..

B. PREDICTION OF MIR THROUGH

GROUP CONTRIBUTION ANALYSIS
The work described below was rst done in 2009 by Professor
Hiroshi Yamato, using the database of solvents found in the
computer program HSPiPF. His information source was the
more than 1200 solvents in the database of that program,
Professor Carters database of MIR values; and his method of
selection of the group contribution constants was to determine them by optimization with a neural network. Only
a minority of the solvents in his database from which his
constants were determined are of interest for cleaning work.
The general formula for MIR is given in Equation C8-1. In this
case, the constant A is taken to be zero, and Property is MIR.
This author has completed the same analysis using the
smaller database of Appendix A1 (141 cleaning solvents for
which Professor Carter has measured MIR values, the group
structural denitions used by Professor Yamato (Table
B2-2), and Equation C8-1.

X 
Property A f
Gi
C8  1
The method of selection of the group contribution
constants was to use the non-linear optimization routine
found in a spreadsheet. The outcome is the correlation
shown in Figure B2-1. The same correlation is shown in
Figure B2-2 for the range of MIR values of interest to those
seeking lower VOC solventsd0 to 1.

C.

RESULTS

Figure B2-1 Effectiveness of Group Contribution Methods in

Correlation of ALL MIR DATA

One uses Equation C8-1 by: (1) Identifying the groups

in the cleaning solvent molecule which are found in Table
B2-2, (2) counting how many of each are present, (3)
multiplying that number by the Gi value found in Table
B2-2 for that group, and (4) adding all those products
together as the estimate of MIR. The same identication
and count of groups in each solvent was used as was used
by Professor Yamato to estimate all three HSP values.
Table B2-2 is sorted by increasing Gi value. These associated groups are least likely to generate ozone.
If one chooses to synthesize cleaning solvents with those
atomic groups that are listed with a yellow shaded background in Table B2-2, the opportunity to prevent ozone
formation by use of that solvent is maximized.

D.

THE ANOMALY!

Four solvents have MIR values clearly not consistent with

those predicted by this optimization. They are acetone
(MIR calculated 1.416, MIR measured 0.353, off-scale on

The group denitions and their values (Gi) found in this

study are given in Table B2-22. Example solvents which
contain each group are also given.
1
The chemical mechanism generation system of Professor Carter does
this by estimating rate constants and reaction routes to derive
mechanisms that can be used to calculate reactivities as discussed
abovednot by a brute force regression approach.
2
It should be noted that the set of groups listed in Table B2-2 is not the
complete set as developed by Professor Yamato. Table B2-2 lists only
those groups necessary to dene the structures of the cleaning solvents
in Appendix A1, and where MIR values have been published by Professor
Carter. The group numbers (left-hand column) in Table B2-2 do
correspond to those used by Professor Yamato.
There are 141 such solvents, and 61 different groups, so the
statistical integrity of the determination of each constant is certainly not
as high as desired. The group standard deviation of t is 0.61 MIR units.
The correlation coefcient is 0.89.

558

Figure B2-2 Effectiveness of Group Contribution Methods in

Correlation of Data of Low MIR