You are on page 1of 25

Fluid Phase Equilibria 201 (2002) 409433

A new method for the estimation of the normal boiling


point of non-electrolyte organic compounds
Wilfried Cordes
a
, Jrgen Rarey
a,b,
a
DDBST GmbH, Industriestr. 1, 26121 Oldenburg, Germany
b
Industrial Chemistry, Carl von Ossietzky University Oldenburg, Fachbereich 9, Postfach 2503, 26111 Oldenburg, Germany
Received 3 September 2001; accepted 21 February 2002
Abstract
Agroup contribution method for the estimation of the normal boiling point of non-electrolyte organic compounds
was developed using experimental data for approximately 2500 components stored in the Dortmund Data Bank
(DDB). Predictions are based exclusively on the molecular structure of the compound. The results of the newmethod
are compared to currently-used methods and are shown to be far more accurate. Structural groups were dened in a
standardized form and the fragmentation of the molecular structures was performed by an automatic procedure to
eliminate any arbitrary assumptions.
2002 Elsevier Science B.V. All rights reserved.
Keywords: Vapor pressure; Model; Method of calculation; Normal boiling temperature; Group contribution
1. Introduction
Pure component vapor pressures of liquid (and solid) compounds are of great importance for many
practical applications in chemical and biochemical engineering as well as for environmental and safety
problems. Vaporliquid equilibria of mixtures, which are for example employed for the design and
simulation of distillation processes, are usually described on the basis of the pure component vapor
pressures of the components in the mixture and a correction term (activity coefcient), which accounts
for the real behavior of the mixture. If an equation-of-state is used for the calculation of the phase
equilibrium, both pure component vapor pressure and real mixture behavior are described by the fugacity
coefcients of the components in the mixture.
While pure component vapor pressures are in most cases accessible by rather simple experimental
techniques and are available for nearly all components of industrial importance, great effort was invested
into the development of methods for the correlation and prediction of the real behavior of liquid mixtures
(Wilson, NRTL, UNIQUAC, UNIFAC, modied UNIFAC, PSRK, . . . ).

Corresponding author. Tel.: +49-441-798-3846; fax: +49-441-798-3330.


E-mail address: rarey@ddbst.de (J. Rarey).
0378-3812/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0378- 3812( 02) 00050- X
410 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
However, with the wide availability of computers and software for the simulation of chemical processes
and environmental simulations (e.g. compartment models for the estimation of the distribution of chemi-
cals in the environment), today we face a great need for physical properties, especially vapor pressures of
a large number of rather exotic compounds (by-products, trace impurities, additives in resign production,
pesticides, . . . ), which are not easily available from literature or experiment.
Another evolving application for accurate physical property estimation methods is computer aided
molecular design (CAMD), which is focused on generating molecular structures for components with
specic properties (vapor pressure, viscosity, polarity, . . . ). During the optimization process, the computer
will generate a large number of structures, for which experimental data are not available and the program
has to rely on the accuracy of the predictive methods employed.
Due to theoretical considerations, physical property estimation usually starts with the estimation of
the critical point, which is directly linked to the molecular parameters of the intermolecular potential. At
the critical point, physical properties of components are not yet strongly inuenced by the formation of
short-range structures, which tend to signicantly complicate property estimation.
Development of estimation methods for critical data with broad applicability is limited by the fact, that
the critical point is difcult to determine and that most of the more complex molecules are not stable at
the critical temperature. Currently, reliable data for less than 600 components are available in literature.
Normal boiling points on the other hand are available for a large number of compounds and are of
special practical importance, but are inuenced signicantly by short-range structures in the liquid phase,
which are difcult to describe by group contributions. This paper presents a method for the estimation of
normal boiling points by group contribution based on a large number of experimental data without the
need for experimental or estimated critical data.
2. Available methods for the prediction of normal boiling points
A variety of estimation methods for normal boiling points from molecular structure are available. A
broad overview on these methods was given by Prausnitz et al. [1]. Several methods are restricted to
individual classes of substances, others require the knowledge of molecular descriptors usually obtained
from quantum mechanical methods. In this work, we will compare our results with those of Constantinou
and Gani [2], Joback and Reid [3], Stein and Brown [4] and Marrero-Morejon and Pardillo-Fontdevila
[5]. All of these methods have in common, that they only require the knowledge about the molecular
structure and therefore are comparable. As the original paper of Marrero-Morejon and Pardillo-Fontdevila
[5] contains several misprints, in this work, we use the corrected parameters supplied by the authors and
published by Prausnitz et al. [1].
3. Theoretical considerations
From the ClausiusClapeyron equation, the normal boiling point can be calculated via the ratio of
H
vap
and S
vap
:
T
b
=
H
vap
S
vap
(1)
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 411
At the normal boiling point, the total interaction between the molecules in the vapor phase is small
compared to that in the liquid phase and the enthalpy of vaporization can be approximated by the total
intermolecular interaction in the liquid phase.
In a homologous series, in which the individual members of the series differ by one CH
2
group, the
volume of the molecule increases linearly. Because molecules are usually not stiff, but tend to tangle to a
more or less spherical form, their outer surface should increase with approximately n
2/3
CH
2
once a certain
length is reached. In case of short chains, however, the gain in entropy is probably not large enough to
favor a spherical form.
As molecular interaction in organic liquids is dominated by nearest neighbor contacts, it should also be
proportional to the molecular surface. Fig. 1a shows H
vap
(J/(mol K)) at the normal boiling for n-alkanes
as function of molecular weight together with a correlation using the expression:
H
vap
= 4067.5 +1738.8M
0.6248
(2)
The exponent of 0.6248 is close the estimated value of 2/3. In case of small chains (14 CH
2
groups),
H
vap
increases linearly with the number of CH
2
groups as expected (Fig. 1b). For very large molecules,
a mutual contact of the complete outer surface becomes more difcult (increasing free volume) and the
increase of H
vap
is less than estimated.
Fig. 1. (a and b) Enthalpy of vaporization of n-alkanes at the normal boiling point as function of the molecular weight (data
taken from Dortmund Data Bank, 2001).
412 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
For the estimation of the normal boiling temperature also knowledge of the entropy change with
vaporization is required. Following Troutons rule, the molar entropy of vaporization of a non-associated
liquid at standard boiling point is approximately constant, about 8288 J/(mol K). Different approaches are
available for a more sophisticated estimation. Different effects contribute to the entropy of vaporization,
e.g.
(a) Generally, external movement (translation) of the molecule is conned to the molar volume minus the
volume occupied by the molecule itself. For an ideal gas, at 1 atmand 273.15 KV
V
trans
amounts to nearly
22.414 dm
3
/mol. In the liquid phase, the available volume V
L
trans
is usually less than a few percent of
the total liquid volume. The translational part of the entropy of vaporization can be calculated via
S
vap, trans
= Rln

V
V
trans
V
L
trans

(3)
V
L
trans
should decrease with increasing attractive forces (i.e. increasing enthalpy of vaporization) and
be inuenced by structural effects.
(b) Due to close packing in the liquid phase, a difcult to estimate part of the internal movements can be
conned. This effect is considered to be very small and generally not of further importance. In case of,
e.g. cyclohexane and n-hexane, the difference in the entropy of vaporization is only approximately
0.4% which is nearly solely due to the increase in molar vapor volume due to the higher boiling
temperature of cyclohexane.
(c) In case of association, the number of particles is changed when going from the liquid phase (nearly
complete association) to the vapor phase (partial association).
Instead of separately estimating H
vap
and S
vap
, a correlation expression for T
b
based on group
contributions and other easy to obtain quantities will be sought for as a certain correlation between H
vap
and S
vap
can be expected. If both properties would depend very differently on molecular structure, two
sets of group contributions would be required.
In order to accomplish this, the typical dependence of T
b
on the number of groups in different homol-
ogous series will rst be compared to the results of methods previously published.
Figs. 2 and 3 show the normal boiling points of n-alkanes and n-alkanols as a function of chain length
together with the predicted values from different group contribution methods.
Joback and Reid [3] employed a linear relationship between the sum of group increments and the
boiling point, which is only adequate to describe the experimental data over a small range of normal
boiling temperatures. Because the method is based on a rather limited database of only 438 components,
this deciency may not have become apparent to the authors.
Stein and Brown employed the same linear relationship, but their method is based on a much larger set
of data (4426 components from the Aldrich Chemical Catalog [6] and 6584 different compounds from
the HODOC[7] database, normal boiling temperatures were partly extrapolated fromvapor pressure data
at low pressures using the LeeKesler [8] equation). In this case, large systematic deviations became
apparent and the authors applied a correction to the estimated values:
T
b
(corrected) = T
b
+A +BT
b
+CT
2
b
(4)
As the polynomial parameters A, Band Cand the group contribution parameters were regressed separately,
they do not give the best results possible with this type of mathematical expression.
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 413
Fig. 2. Normal boiling points of n-alkanes as a function of the number of CH
2
groups.
Constantinou and Gani [2] greatly improved the method of Joback and Reid [3] by using a logarithmic
dependence of the boiling point on the sum of the group contributions:
T
b
= 204.359 ln

N
i
C
i
+

M
i
D
i

(5)
with N
i
is the number of rst-order groups of type i, C
i
the group contribution of rst-order group i, M
i
the number of second-order groups of type i, and D
i
is the group contribution of second-order group i.
The second-order groups are only used as a correction to the rst-order estimations for a limited number
of special structures.
Marrero-Morejon and Pardillo-Fontdevila [5] employed the equation:
T
b
= M
0.404

N
i
C
i
+156.00 (6)
where N
i
is the number of rst-order groups of type i, C
i
the group contribution of rst-order group i,
and M is the molecular weight (g/mol).
Fig. 3. Normal boiling points of n-alkanols as a function of the number of CH
2
groups.
414 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
In contrary to the other methods, Marrero-Morejon and Pardillo-Fontdevila use bond contributions
instead of group contributions.
From Figs. 2 and 3, it can be seen that, although much better than the linear assumption of Joback and
Reid, neither the logarithmic dependence nor the term M
0.404
are able to describe the dependence of T
b
on chain length in the two homologous series.
For the correlation of group contribution parameters Constantinou and Gani as well as Marrero-Morejon
and Pardillo-Fontdevila used only a limited set of data (392 and 507 components, respectively).
4. Development of the new method
From the analysis of the different estimation methods, it is obvious that a signicantly better method
could be developed based on an improved expression for the dependence of the normal boiling point
on the sum of group increments and a signicantly larger set of reliable experimental information. In
addition, all parameters should be regressed simultaneously in order to nd the optimum value of the
objective function.
4.1. Group denitions and equations
To describe the normal boiling point for a large number of molecules greatly differing in size the
following expression was employed:
T
b
=

N
i
C
i
n
a
+b
+c (7)
where a, b and c are the adjustable parameters, N
i
the number of groups of type i, C
i
the group contribution
of group i (K), and n is the number of atoms in the molecule (except hydrogen).
While this expression gives a good description of the dependence of T
b
on molecular size, it carries the
additional advantage, that via the number of atoms in the molecule, an additional and readily available
quantity more or less independent from the sum of the increments is introduced. The number of atoms or
the molecular weight have been successfully employed by numerous other authors in the past (e.g. Joback
and Reid [3], Marrero-Morejon and Pardillo-Fontdevila [5]) to improve the results of group contribution
methods for various properties.
Correlation of data for 2550 components lead to the following values for the constants:
a = 0.6713, b = 1.4442, c = 59.344 (K)
Results for the new model for n-alkanes and n-alkanols are given in Figs. 4 and 5.
Denition of the structural groups was approached by an iterative procedure. Starting off from the
groups used in the method by Joback and Reid, the adjustable parameters a, b and c as well as all the
group contributions were tted to the experimental data. The results were then carefully analyzed and
the group denitions were revised in order to take into account special effects, which otherwise led to
large deviations. This had to be repeated several times. In the second step, all the molecules, which could
not be fragmented into the structural groups available were reviewed and new groups were dened for
all cases in which enough experimental information was available for the correlation of reliable group
parameters.
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 415
Fig. 4. Normal boiling points of n-alkanes as a function of the number of CH
2
groups together with the results for the proposed
method.
In this procedure, each redenition of molecular groups required the re-fragmentation of about 2550
components. This was made possible by employing an automatic fragmentation algorithm (AutoInkr)
and a data bank with the molecular structures of all components (ChemDB) (Cordes et al. [9]).
In order to perform group fragmentation consistent with the chemical knowledge usually required for
this task, not only the structure of the groups themselves but also their chemical neighborhood had to be
included in the group denition.
Fig. 5. Normal boiling points of n-alkanols as a function of the number of CH
2
groups together with the results for the proposed
method.
416 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
The list of groups, explanations as well as examples for molecules containing these groups are given
in Table 1. The group contributions, the number of components used for regressing these values and the
mean absolute deviation in temperature for these components are given in Table 2.
In the approach used for the development of the new method specication of the chemical neighbor-
hood of a structural group plays a very important part. The correct specication of this neighborhood is
especially critical, because in our approach the fragmentation of molecules is performed by an automatic
algorithm rather than a chemist with his understanding of the electronic structure.
During the development of the method, it became apparent that:
there is no need to distinguish between carbon or silicon as a neighbor atom;
veryelectronegative (N, O, FandCl) or aromatic neighbors oftensignicantlyinuence the contribution
of a structural group;
it is usually of great importance whether a group is part of a chain, ring or aromatic system.
In addition, strong steric or mesomeric effects signicantly worsen the results of the new method. This
is discussed in detail later on.
In a number of cases, for a computer software, it is not as obvious as for a chemist which group to use.
As there are only two avors of the CH
2
group (ring and chain), one could use the chain CH
2
group
(identication number (ID) 4, priority (PR) 82) for the phenylic carbon in ethylbenzene. In this case, the
C(c) (a)-group (phenylic carbon, ID 8, PR 79) would be the correct choice as the neighboring aromatic
carbon is of greater importance than the two hydrogen atoms. To solve this problem, the different groups
were sorted by priority. In case of the automatic algorithm structural groups are matched in the order of
their respective priority. In this way, it is guaranteed that always the group with the lower priority number
(higher priority) is used to estimate the normal boiling point. The priority numbers are given in Tables 1
and 3 in order to allow the reader to correctly apply or program the method.
As was also observed by other authors, the short chain alkanols do not t into the group contribution
scheme. For this reason, special groups are used for short-chain (ID 36) and long-chain (ID 34 and 35)
alkanols. In some cases a larger group could be but should not be constructed of smaller groups (e.g. the
carbamat-group, the CCOCOC group, . . . ). Some of these groups are given at the end of Table 1
as excluded groups with a group ID of 1. Whenever one of these groups is found in a molecule, the
method cannot be applied as no group parameters are available for excluded groups.
4.2. Second-order corrections
In addition to the group increments given in Table 1, special second-order corrections were applied.
A list of these special corrections together with examples is given in Table 3. Table 4 contains the
second-order contributions, the number of components used for regressing these values and the mean
absolute deviation in temperature for these components.
4.3. Database
Normal boiling temperature data for approximately 2800 components are available in the Dortmund
Data Bank (DDB), out of which 2550 components are constructed of the structural groups proposed in
this work. These data were entered since work on the data bank started in 1973 and were extensively
used, e.g. for the calculation of phase equilibrium data. This is the reason why this set of data may be
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 417
Table 1
Group denitions (ID, identication number; PR, priority)
Group Description Name ID/PR Occurs
Periodic group 17
Fluorine
F F connected to C or Si F(C,Si) 19/59 2-Fluoropropane,
trimethyluorosilane
F connected to a C or Si
already substituted with
one F or Cl and one other
atom
F(C([F,Cl]))a 22/56 1-Chloro-1,2,2,2-tetra-
uoroethane[R124],
diuoromethylsilane
F connected to C or Si
already substituted with at
least one F or Cl and two
other atoms
F(C([F,Cl]))b 21/54 Trichlorouoromethane
[R11],
2,2,3,3-tetrauoropropionic
acid
F connected to C or Si
already substituted with
two F or Cl
F(C([F,Cl]
2
)) 23/55 1,1,1-Triuorotoluene,
2,2,2-triuoroethanol,
triuoroacetic acid
F connected to an
aromatic carbon
F(C(a)) 24/58 Fluorobenzene,
4-uoroaniline
F on a C=C (vinyluoride) CF=C 20/57 Vinyl uoride,
triuoroethene,
peruoropropylene
Chlorine
Cl Cl connected to C or Si
not already substituted
with F or Cl
Cl(C,Si) 25/47 Butyl chloride,
2-chloroethanol,
chloroacetic acid
Cl connected to C or Si
already substituted with
one F or Cl
Cl((C,Si)([F,Cl])) 26/46 Dichloromethane,
dichloroacetic acid,
dichlorosilane
Cl connected to C or Si
already substituted with at
least two F or Cl
Cl((C,Si)([F,Cl]
2
)) 27/44 Ethyl trichloroacetate,
trichloroacetonitrile
Cl connected to an
aromatic C
Cl(C(a)) 28/48 Chlorobenzene
Cl on a C=C
(vinylchloride)
CCl=C 29/45 Vinyl chloride
COCl COCl connected to C
(acid chloride)
COCl 77/13 Acetyl chloride,
phenylacetic acid chloride
Bromine
Br Br connected to a
non-aromatic C or Si
Br(C/Si(na)) 30/41 Ethyl bromide,
bromoacetone
Br Br connected to an
aromatic C
Br(C(a)) 31/42 Bromobenzene
Iodine
I I connected to C or Si I(C,Si) 32/39 Ethyl iodide 2-iodotoluene
418 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
Table 1 (Continued)
Group Description Name ID/PR Occurs
Periodic group 16
Oxygen
OH OH for aliphatic chains with
one or two C (even if connected
to aromatic fragments, used in
particular for methanol and
ethanol and their derivatives)
OH short chain 36/64 Ethanol, benzyl alcohol
OH connected to C which has
four non-hydrogen neighbors
(tertiary alcohols)
OH tert 33/63 Tert-butanol, diacetone
alcohol
OH connected to C or Si
substituted with one C or Si in an
at least three C or Si containing
chain (primary alcohols)
HO((C,Si)H
2
(C,Si)
(C,Si))
35/60 1-Nonanol,
tetrahydrofurfuryl alcohol,
ethylene cyanohydrin
OH connected to an aromatic C
(phenols)
OH (Ca) 37/62 Phenol, methyl salicylate
OH connected to a C or Si
substituted with two C or Si in a
at least three C or Si containing
chain (secondary alcohols)
HO((C,Si)
2
H(C,Si)
(C,Si))
34/61 2-Butanol, cycloheptanol
O O connected to two neighbors
which are each either C or Si
(ethers)
(C,Si)O(C,Si) 38/66 Diethyl ether, 1,4-dioxane
O in an aromatic ring with
aromatic C as neighbors
(C(a))O(a)(C(a)) 65/65 Furan, furfural
CHO CHO connected to C
(aldehydes)
CHO(C) 52/29 Acetaldehyde, benzaldehyde
C=O CO connected to two C
(ketones)
O=C (C)
2
51/30 Acetone, methyl
cyclopropyl ketone
O=C(O)
2
Non-cyclic carbonate O=C(O)
2
79/9 Dimethyl carbonate
COOH COOH connected to C COOH(C) 44/18 Acetic acid
COO HCOO connected to C (formic
acid ester)
HCOO(C) 46/21 Ethyl formate, phenyl
formate
COO connected to two C
(ester)
(C)COO(C) 45/19 Ethyl acetate, vinyl acetate
COO in a ring, C is connected
to C (lactone)
C(c)OO 47/20 -Caprolactone,
crotonolactone
(OC
2
) (OC
2
) (epoxide) (OC
2
) 39/27 Propylene oxide
COOCO Anhydride connected to two C COOCO 76/6 Acetic anhydride, phthalic
anhydride
Sulphur
SS SS (disulde)
connected to two C
(C)SS(C) 55/28 Dimethyldisulde,
1,2-dicyclopentyl-1,2-disulde
SH SH connected to C
(thiols)
SH(C) 53/49 1-Propanethiol
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 419
Table 1 (Continued)
Group Description Name ID/PR Occurs
S S connected to two C (C)S(C) 54/50 Methyl ethyl sulde
S in an aromatic ring S(a) 56/51 Thiazole, thiophene
SO
2
Non-cyclic sulfone
connected to two C
(sulfones)
(C)SO
2
(C) 82/12 Sulfolane, divinylsulfone
SCN SCN (thiocyanate) connected to C SCN(C) 81/14 Allyl isothiocyanate
Periodic group 15
Nitrogen
NH
2
NH
2
connected to either
C or Si
NH
2
(C,Si) 40/68 Hexylamine, ethylenediamine
NH
2
connected to an
aromatic C
NH
2
(Ca) 41/67 Aniline, benzidine
NH NH connected to two
neighbors which are each
either C or Si (secondary
amines)
(C,Si)NH(C,Si) 42/71 Diethylamine, morpholine
N N connected to three
neighbors which are each
either C or Si (tertiary
amines)
(C,Si)
2
N(C,Si) 43/72 N,N-dimethylaniline,
nicotine
=N aromatic =N in a
ve-membered ring
=N(a)(r5) 66/70 Piperidine, thiazole
=N aromatic =N in a
six-membered ring
=N(a)(r6) 67/69 Pyridine, nicotine
C

N C

N (cyanide)
connected to C
C

N(C) 57/31 Acetonitrile, 2,2

-dicyano
diethyl sulde
CONH CONH
2
(amide) CONH
2
50/22 Acetamide
CONH (monosubstituted amide) CONH 49/25 N-methylformamide,
6-caprolactam
CON (disubstituted
amide)
CON 48/26 N,N-dimethylformamide
(DMF)
OCN OCN connected to C or
Si (cyanate)
OCN 80/23 Butylisocyanate,
hexamethylene
diisocyanate
ONC ONC (oxime) ONC 75/24 Methyl ethyl ketoxime
NO
2
Nitrites (esters of nitrous
acid)
O=NO(C) 74/17 Ethyl nitrite, nitrous acid
methyl ester
NO
2
connected to aliphatic C NO
2
(C) 68/15 1-Nitropropane
NO
2
connected to
aromatic C
NO
2
(C(a)) 69/16 Nitrobenzene
NO
3
Nitrate (esters of nitric acid) NO
3
72/8 N-butylnitrate,
1,2-propanediol dinitrate
Phosphorous
PO(O)
3
Phosphates with three
substituents
PO(O)
3
73/5 Triethyl phosphate,
tris-(2,4-dimethylphenyl)
phosphate
420 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
Table 1 (Continued)
Group Description Name ID/PR Occurs
Arsine
AsCl
2
AsCl
2
connected to C AsCl
2
84/11 Ethylarsenic dichloride
Periodic group 14
Carbon
CH
3
CH
3
not connected to either N,
O, F or Cl
CH
3
(ne) 1/75 Decane
CH
3
connected to either N, O, F
or Cl
CH
3
(e) 2/73 Dimethoxymethane,
methyl butyl ether
CH
3
connected to an aromatic
atom (not necessarily C)
CH
3
(a) 3/74 Toluene, p-methyl-styrene
CH
2
CH
2
in a chain C(c)H
2
4/82 Butane
CH
2
in a ring C(r)H
2
9/83 Cyclopentane
CH CH in a chain C(c)H 5/88 2-Methylpentane
CH in a ring C(r)H 10/87 Methylcyclohexane
C C in a chain C(c) 6/90 Neopentane
C in a chain connected to at
least one aromatic carbon
C(c) (a) 8/79 Ethylbenzene, diphenylmethane
C in a chain connected to at
least one F, Cl, N or O
C(c) (e) 7/78 Ethanol
C in a ring C(r) 11/89 Beta-pinene
C in a ring connected to at
least one aromatic carbon
C(r) (Ca) 14/77 Indene, 2-methyl tetralin
C in a ring connected to at
least one N or O which are not
part of the ring or one Cl or F
C(r) (e,c) 12/80 Cyclopentanol, menthol
C in a ring connected to at
least one N or O which are part
of the ring
C(r) (e,r) 13/81 Morpholine, nicotine
=C(a) Aromatic =CH =C(a)H 15/76 Benzene
Aromatic =C not connected to
either O, N, Cl or F
=C(a) (ne) 16/86 Ethylbenzene, benzaldehyde
Aromatic =C with three
aromatic neighbors
(a)=C(a) 2(a) 18/85 Naphthalene, quinoline
Aromatic =C connected to
either O, N, Cl or F
=C(a) (e) 17/84 Aniline, phenol
C=C H
2
C=C (1-ene) H
2
C(c)=C 61/33 1-Hexene
C=C (both C have at least one
non-H neighbor)
C(c)=C(c) 58/38 2-Heptene, mesityl oxide
Non-cyclic C=C connected to
at least one aromatic C
C(c)=C(c) (C(a)) 59/35 Isosafrole, cinnamic alcohol
Cyclic C=C C(r)=C(r) 62/36 Cyclopentadiene
Non-cyclic C=C
substituted with at least
one F, Cl, N or O
(e)C(c)=C(c) 60/34 Trans-1,2-dichloroethylene,
peruoroisoprene
C

C HC

C (1-ine) HC

C 64/32 1-Heptyne
C

C C

C 63/37 2-Octyne
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 421
Table 1 (Continued)
Group Description Name ID/PR Occurs
Silicon
Si Si Si 70/53 Butylsilane
Si Si connected to at least
one O, F or Cl
Si (e) 71/52 Trichlorosilane,
hexamethyl disiloxane
Germanium
Ge Ge connected to four
carbons
(C)
2
Ge (C)
2
86/43 Tetramethylgermane
GeCl
3
GeCl
3
connected to
carbons
GeCl
3
85/7 Fluorodimethylsilyl-
(trichlorogermanyl)methane
Stannium
Sn Sn connected to four
carbons
(C)
2
Sn (C)
2
83/40 Tetramethylstannane
Periodic group 13
Bor
B(O)
3
Non-cyclic boric acid ester B(O)
3
78/10 Triethyl borate
Excluded groups
OCON Carbamat OCON 1/1 N-propylcarbamate
CO(NH
2
)
2
Urea and derivatives CO(NH
2
)
2
1/2 Urea
CCOCOC Neighboring keto groups CCOCOC 1/3 2,3-Butanedione
Imidazol Imidazol Imidazol 1/4 Imidazol
Abbreviations: e, very electronegative neighbors (N, O, F, Cl); ne, not very electronegative neighbors (not N, O, F, Cl); a, aromatic
atom or neighbor; c, atom or neighbor is part of a chain; r, atom or neighbor is part of a ring.
regarded as very reliable. In addition, molecular structures for approximately 16000 components were
stored in form of connection tables, so that the molecules may easily be fragmented into groups by the
automatic procedure described before. The components in the DDB comprise a set of substances for
which experimental thermophysical data are available in literature and which are therefore of importance
to science or industry. The frequency of occurrence of the individual structural groups and second-order
corrections in a database of 2550 components is given in Tables 2 and 4, respectively.
4.4. Regression of the group constants
For the simultaneous regression of the model parameters, a special algorithm was developed consisting
of an inner and outer regression loop. In the outer loop, the coefcients a, b and c are optimized in order
to minimize the mean absolute deviation in normal boiling temperature times the square of the standard
deviation with the help of a non-linear regression using a simplex algorithm. Using this, rather uncommon
objective function decreases the number of large deviations while leading to a slightly increased mean
deviation. The inner loop incorporates a multilinear least squares regression of the equations:

i
N
i,j
C
i,j
= T
b,j
(n
b
j
+c) a (for all components j) (8)
422 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
Table 2
Group contributions, number of components used for regressing these values and mean absolute deviation in temperature for
these components
Group Group Mean Standard Mean Standard Number of
number contribution absolute deviation absolute deviation components
(K) error (%) (%) error (K) (K)
1 188.555 1.71 2.42 7.42 10.69 1468
2 282.015 2.3 3.02 9.67 13.11 199
3 176.705 2.01 2.78 9.63 13.41 153
4 250.119 1.58 2.26 7.24 10.66 956
5 260.938 1.48 2 6.83 10.11 310
6 273.544 2.15 2.83 8.56 11.37 66
7 278.135 2.14 3.42 8.97 14.35 820
8 210.93 1.5 2.12 7.85 11.71 124
9 246.871 2.07 2.77 8.71 11.65 270
10 241.804 1.55 2.11 6.84 9.43 137
11 265.032 1.72 2.15 7.47 9.24 35
12 264.839 1.95 4.01 8.05 14.93 67
13 304.422 2.49 3.26 10.32 13.21 64
14 281.964 1.73 2.22 9.07 11.52 22
15 245.521 2.32 3.4 11.73 17.5 558
16 322.825 2.08 2.99 10.57 15.65 421
17 377.988 2.79 3.94 13.88 19.82 247
18 386.361 1.68 2.28 9.81 13.89 65
19 129.511 1.95 2.72 6.58 9.44 38
20 73.5088 2.92 4.31 6.69 9.29 9
21 65.9125 2.95 5.04 9.58 15.07 132
22 111.411 2.36 3.24 7.12 9.63 21
23 144.464 3.95 5.7 9.38 12.9 3
24 21.348 2.35 2.78 9.04 10.85 25
25 327.158 2.01 2.89 8.58 12.59 108
26 300.288 1.73 2.23 6.66 8.63 47
27 275.233 2.79 3.98 10.37 14.3 55
28 204.105 2.1 2.75 10.2 13.55 63
29 299.52 1.69 2.21 5.87 7.79 34
30 427.56 2.23 2.94 8.45 11.31 65
31 351.895 1.36 1.85 6.51 8.83 21
32 564.102 1.89 2.35 7.71 9.73 28
33 401.033 1.97 2.81 8.51 11.66 43
34 411.08 1.81 2.43 8.31 11.4 73
35 477.583 2.33 3.1 11.04 14.65 68
36 515.544 2.47 3.01 10.6 12.65 65
37 354.061 3.28 4.05 16.86 21.4 63
38 158.793 2.11 3.14 8.91 13.31 283
39 861.138 2.18 2.66 8.88 11.24 15
40 361.207 2.31 2.97 9.44 12.75 51
41 468.458 2.04 2.83 11.29 15.96 31
42 259.446 2.18 2.93 9.21 12.04 58
43 121.99 2.76 3.54 11.66 15.7 41
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 423
Table 2 (Continued)
Group Group Mean Standard Mean Standard Number of
number contribution absolute deviation absolute deviation components
(K) error (%) (%) error (K) (K)
44 1124.49 1.88 2.87 9.67 14.46 63
45 697.228 2.16 3.01 10.41 15.38 203
46 712.76 1.4 2.11 6.03 9.17 19
47 1230.21 4.76 6.46 20.25 27.66 5
48 1058.87 1.94 2.57 9.28 12.12 10
49 1323.88 2.72 3.29 14.71 17.99 8
50 1479.27 1.78 2.28 8.76 11.21 5
51 654.008 3.12 4.96 12.98 20.17 89
52 626.216 3.24 4.6 14.97 21.69 42
53 459.247 1.37 1.71 5.22 6.35 44
54 479.985 1.43 1.9 6.14 8.41 33
55 874.273 0.94 1.29 3.66 5.09 4
56 309.872 2.1 2.82 9.27 12.6 18
57 804.356 3.05 3.87 14.38 18.51 41
58 516.817 1.66 2.27 6.66 9.37 117
59 607.968 2.37 2.62 12.51 13.81 12
60 521.597 1.74 2.33 6.31 8.58 29
61 437.399 1.95 2.71 7.2 9.82 182
62 507.998 2.56 3.48 10.59 14.21 76
63 556.785 1.19 2.01 4.98 8.34 23
64 468.03 1.74 2.24 5.79 7.7 23
65 79.2981 2.81 3.88 12.68 17.51 11
66 430.782 3.47 4.4 15.37 19.55 15
67 282.737 2.64 4.15 12.39 19.82 36
68 907.229 10.26 16.5 40.71 65.72 9
69 775.752 5.59 7.03 28.47 35.49 37
70 294.323 1.42 2.19 4.9 7.16 33
71 219.416 2.98 4.47 10.47 13.55 70
72 964.373 0.93 1.18 3.79 4.85 6
73 1267.28 2.39 3.02 11.84 14.97 6
74 532.35 0.62 0.78 1.91 2.43 7
75 1082.68 1.95 2.51 8.64 10.9 9
76 1431.22 5.47 6.35 26.53 30.48 7
77 837.687 1.41 1.8 5.99 7.74 18
78 594.43 0.96 1.37 4.14 5.72 7
79 911.983 0.88 1.08 3.65 4.34 4
80 671.441 5.5 7.27 24.7 32.57 14
81 1002.53 1.85 2.29 8.21 10.26 5
82 1502.35 2.4 2.95 13.34 16.46 5
83 525.228 0.33 0.45 1.28 1.76 3
84 1173.13 0.43 0.59 1.97 2.7 6
85 1280.52 1
86 301.028 1
424 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
Table 3
Second-order contributions for normal boiling points
Name Description ID Occurs
Para pair(s) Para positioncounted only once and
only if there are no meta or ortho pairs
87 p-Xylene
Meta pair(s) Meta positioncounted only once and
only if there are no para or ortho pairs
88 m-Xylene
Ortho pair(s) Ortho positioncounted only once and
only if there are no meta or para pairs
89 o-Xylene
Five-ring A ve-membered non-aromatic ring 90 Cyclopentane
Three/four-ring A three- or four-membered non-aromatic ring 91 Cyclobutene
One hydrogen Component has one hydrogen 92 Nonauorobutane
No hydrogen Component has no hydrogen 93 Peruoro compounds
The main advantage of this algorithm is, that only three parameters need to be regressed using the slow
non-linear method while the large number of group increments is calculated from the fast multilinear
least squares regression. The objective function in linear regression is always the mean squared deviation
between calculated and experimental value. Partial derivation of the objective function with respect to
the n coefcients (in this case, the n individual group contributions) leads to n linear equations. At the
minimumof the objective function, all partial derivatives have to be zero. This leads to a set of n equations
with n unknowns, which can readily be solved, e.g. by the well known GaussJordan elimination. The
set of linear equations leads to exactly one solution provided all columns and rows in the coefcient
matrix are linear independent, which can be analyzed by singular value decomposition. In this work, the
ratio of experimental data (normal boiling temperatures) to adjustable parameters is about 26.61 and the
resulting matrix is not close to singular.
Once the coefcients are available from the linear regression, the mean absolute deviation in T
b
and the
squared standard deviation are calculated and their product is returned to the outer loop. Initial estimates
are only required for a, b and c, which are regressed in the outer loop. A full regression of 96 parameters
(a, b, c, 87 group contributions and six second-order corrections) to 2550 experimental normal boiling
temperatures, thus, requires only a few minutes on a regular PC. We experienced no problems with
Table 4
Second-order contributions, number of components used for regressing these values and mean absolute deviation in temperature
for these components
Group
number
Group
contribution
Mean
absolute
error (%)
Standard
deviation
(%)
Mean
absolute
error (K)
Standard
deviation
(K)
Number of
components
87 37.5096 2.87 4.29 14.31 21.8 81
88 3.5994 2.47 3.83 11.79 18.49 77
89 44.8024 2.85 3.82 14.53 19.52 77
90 27.0458 2.06 2.85 8.88 12.42 134
91 39.0849 3.18 5.3 11.14 17.74 50
92 131.323 2.34 3.36 7.64 11.69 36
93 167.799 3.48 7.18 11.58 24.73 97
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 425
local minima. To verify the algorithm, the nal parameters were used as starting values for a non-linear
regression of all parameters to the experimental data using the commercial GRG2 algorithm (Frontline
Systems, Inc., Incline Village, NV, USA). No further improvement was achieved.
4.5. Test of the predictive capability
In order to test the predictive capability of the method, experimental normal boiling temperatures for
126 components not in the database used for regression were compared with the predicted values. This
lead to a mean absolute deviation in temperature of 8.9 K (35.3 K for the Joback method, 13.1 K for Stein
and Brown, 14.5 Kfor Constantinou and Gani, and 12.3 Kfor Marrero-Morejon and Pardillo-Fontdevila).
5. Results and discussion
The results for the different group contribution methods were carefully analyzed to detect any possible
weaknesses of the new method compared to the other methods and to identify typical components for
which to expect extraordinary large deviations.
As the most important criterion for the reliability of a model the probability of a prediction failure
(extreme deviation between experimental and estimated value) was chosen. Fig. 6 shows a plot of the
part of the data with a deviation larger than a certain temperature versus the deviation in temperature.
Calculations are based on a common set of 1863 components, for which the normal boiling point can
be estimated by all models except that of Marrero-Morejon and Pardillo-Fontdevila. Including the latter
Fig. 6. Part of the data with deviations larger than a given temperature.
426 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
Table 5
Deviations of the models for different types of hydrocarbons
Components Number of
components
This work Joback and
Reid [3]
Stein and
Brown [4]
Constantinou
and Gani [2]
Marrero-Morejon
and Pardillo-
Fontdevila [5]
n-Alkanes 27 6.5 55.7 12.1 18.7 13.6
Alkanes (non-cyclic) 166 6.5 26.7 15.2 8.9 7.7
Alkanes (cyclic) 65 7.1 8.9 8.6 6.2 17.3
Aromatic hydrocarbons 112 9.4 24.3 10.7 15.7 8.3 (70)
Alkenes (HC) 167 7.4 9.4 7.7 6.0 17.8 (123)
Alkenes (cyclic C=C) (HC) 38 9.4 8.1 8.2 6.7 16.0 (24)
Alkines (HC) 33 5.4 13.1 12.1 13.5 3.8 (31)
Hydrocarbons 568 7.4 17.7 10.8 9.5 11.6 (457)
HC denotes that the components taken into account only contain the elements C and H. The numbers in brackets for the method
of Marrero-Morejon and Pardillo-Fontdevila specify the number of components to which the method was applicable.
method would have further reduced the number of components in this test signicantly. It can be seen,
that in case of the new method, only 8% of the data are estimated with a deviation larger than 20 K.
The other models range between 17% (Stein and Braun) and 36% (Constantinou and Gani). Only 14
components (0.75%) exhibited a deviation larger than 50 K. The other models range between 3.6 and
9.7%. The reasons for these failures will be discussed later. First, the behavior of the different models is
compared for different classes of compounds.
5.1. Hydrocarbons
A comparison of the different models is given in Table 5. While the new model yields small deviations
(less than 10 K) for all classes of components, in some cases the method of Constantinou and Gani
gives slightly better results. This is due to the larger number of parameters. In this work, only four
different types of alkene groups are used compared to six groups plus three second-order groups in the
ConstantinouGani method. The use of more than twice the number of parameters results in only minor
improvements. The method of Marrero-Morejon and Pardillo-Fontdevila gives, especially good results
for alkines. In case of aromatic hydrocarbons, the range of applicability is signicantly smaller (70 instead
of 112 components), which may be a reason for the good results.
Especially, large deviations (T
b,exp
T
b,calc
) were found for the following components:
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 427
5.2. Alkanols
Table 6 gives a comparison of the different models for various types of alkanols. For all 343 components,
the newmodel gives the lowest mean deviation. In case of n-alkanols and primary and secondary alkanols
the method of Stein and Brown gives slightly better results. This is due to the greater exibility of the
polynomial linearization. As shown in Fig. 7, up to about C
7
the increase of normal boiling temperature
with molecular weight is over-predicted by the new model. Especially, high deviations are found for
alkandiols, whereby geminal diols show no special behavior. If the correction function
T
b,corr
= T
b,calc
+39.3 5.26n
C
(n
C
, number of carbon atoms) (9)
is applied, the mean deviation reduces to 5.2 K (not taking into account ethanediol).
Table 6
Deviations of the models for different types of alkanols
Components Number of
components
This work Joback and
Reid [3]
Stein and
Brown [4]
Constantinou
and Gani [2]
Marrero-Morejon
and Pardillo-
Fontdevila [5]
n-Alkanols 21 10.6 29.8 7.9 11.3 17.5
Aliphatic alkanols 110 7.3 23.6 12.1 10.3 15.7 (103)
Primary alkanols 36 8.5 20.8 7.1 10.7 15.1
Secondary alkanols 35 6.0 27.7 5.3 10.7 14.7
Tertiary alkanols 21 6.4 37.9 7.0 18.1 19.1 (14)
Alkandiols 15 15.6 (5.2) 28.9 23.3 23.9 14.3
Alkanols (HC) 343 7.5 22.5 12.5 10.6 9.9 (301)
HC denotes that the components taken into account only contain the elements C and H in addition to one or more OH-groups.
The numbers in brackets for the method of Marrero-Morejon and Pardillo-Fontdevila specify the number of components to
which the method was applicable.
428 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
Fig. 7. Calculated vs. experimental normal boiling points of n-alkanols.
5.3. Other oxygenated compounds
A comparison of the different models is given in Table 7. Deviations for the new model are amongst
the lowest achieved by the different models. The method of Marrero-Morejon and Pardillo-Fontdevila
gives exceptionally good results for aldehydes.
Table 7
Deviations of the models for different types of oxygenated hydrocarbons (except alkanols)
Components Number of
components
This work Joback and
Reid [3]
Stein and
Brown [4]
Constantinou
and Gani [2]
Marrero-Morejon
and Pardillo-
Fontdevila [5]
Ethers (HC) 85 4.7 10.7 6.7 11.1 5.0 (82)
Epoxides (HC) 9 8.6 19.9 16.2 30.9 6.31 (7)
Aldehydes (HC) 17 9.9 9.8 6.3 7.8 3.0 (16)
Ketones (HC) 43 6.9 13.3 7.4 8.8 7.0 (38)
Non-cyclic carbonates (HC) 4 3.7 25.1 49.7 3.5 5.5 (4)
Carboxylic acids (HC) 32 5.7 30.2 6.0 16.4 9.3
Esters (not CHO
2
)(HC) 120 9.2 34.2 9.0 9.6 13.4 (101)
HC denotes that the components taken into account only contain the elements C and H in addition to one or more groups
containing oxygen. The numbers in brackets for the method of Marrero-Morejon and Pardillo-Fontdevila specify the number of
components to which the method was applicable.
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 429
Table 8
Deviations of the models for different types of halogenated hydrocarbons
Components Number of
components
This work Joback and
Reid [3]
Stein and
Brown [4]
Constantinou
and Gani [2]
Marrero-Morejon
and Pardillo-
Fontdevila [5]
Fluorinated saturated hydrocarbons 52 5.4 16.9 64.3 25.6 10.1 (51)
Fluorinated hydrocarbons 74 6.7 16.8 50.3 26.6 9.6 (71)
Chlorinated saturated hydrocarbons 55 8.5 27.6 12.6 9.3 13.9 (51)
Chlorinated hydrocarbons 88 7.7 23.0 12.5 10.1 13.5 (80)
Brominated saturated hydrocarbons 31 7.6 16.9 10.3 8.0 14.9 (22)
Brominated hydrocarbons 47 8.2 14.6 10.1 8.4 14.6 (33)
Iodinated saturated hydrocarbons 17 5.7 14.7 8.8 8.1 13.4 (14)
Iodinated hydrocarbons 20 5.9 13.7 8.1 7.2 11.7 (17)
The numbers in brackets for the method of Marrero-Morejon and Pardillo-Fontdevila specify the number of components to
which the method was applicable.
5.4. Halogenated hydrocarbons
A comparison of the different models is given in Table 8. In all cases, the proposed model gives the
lowest deviation, which is probably due to a stronger differentiation in the structural groups and the large
set of data used in the regression.
5.5. Isomer differentiation
The method developed here was mainly focused on covering a broad range of components also, e.g.
including germanium, silicon, tin and phosphor and very different in size. For non-cyclic saturated
hydrocarbons, only four different groups are used and a correct differentiation between alkane isomers
cannot be expected. For a list of 18 different octane-isomers, an average error of 5.1 K was obtained,
3.3 K for the method of Constantinou and Gani and 2.7 K for Marrero-Morejon and Pardillo-Fontdevila.
The latter two methods use 9, resp. 10 parameters for non-cyclic saturated hydrocarbons.
5.6. Compounds for which large deviations are observed
Although the group contribution approach for the calculation of normal boiling temperatures generally
leads to good results, the assumption of simple additivity is not always valid.
In cases, where a strong inductive effect inuences a mesomeric system, the molecule has a rather
high dipole moment or polarizability. In these cases, the estimated normal boiling temperature is sig-
nicantly too low. Typical examples are: 4-hydroxybenzaldehyde (T
b,est
= 583.20 K, T
b
= 523.78 K);
4-dimethylaminobenzaldehyde (T
b,est
= 529.86 K, T
b
= 588. K); 4-nitroaniline (T
b,est
= 551.53 K,
T
b
= 607. K); 1,4-dihydroxybenzene (T
b,est
= 512.35 K, T
b
= 558.5 K); and resorcinol (T
b,est
= 554.55 K,
T
b
= 506.16 K).
Generally, the inuence of a larger number of strong I- or M-effects on mesomeric systems may lead
to high deviations as in 2,4,6-trinitrotoluene (T
b,est
= 624.60 K, T
b
= 573.0 K), and 2,4,6-trinitrophenol
(T
b,est
= 654.12 K, T
b
= 597.0 K).
430 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
Table 9
Estimation of the normal boiling temperature (K) of ethylbenzene
Component: ethylbenzene; number of atoms: 8
Group Atoms Frequency Contribution Total
1 (CH
3
(ne)) 1 1 188.555 188.555
8 ( C(c) (a)) 2 1 210.930 210.93
15 (=C(a)H) 48 5 245.521 1227.605
16 (=C(a) (ne)) 3 1 322.825 322.825
Sum 1949.915
T
b
=
1949.915
8
0.6713
+1.4442
+59.344 = 414.98 (remark : atom2 is matched by the group C(c) (a) as the group C(c)H
2

appears later in the priority list).


Table 10
Estimation of the normal boiling temperature (K) of 1,1,2,2,-tetraphenylethane
Component: 1,1,2,2,-tetraphenylethane; number of atoms: 26
Group Atoms Frequency Contribution Total
8 ( C(c) (a)) 1, 2 2 210.930 421.860
15 (=C(a)H) 48, 1014, 1620, 2226 20 245.521 4910.420
16 (=C(a) (ne)) 3, 9, 15, 21 4 322.825 1291.300
Sum 6623.580
T
b
=
6623.580
26
0.6713
+1.4442
+59.344 = 699.05.
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 431
Table 11
Estimation of the normal boiling temperature (K) of N-methyl-pyrrolidon
Component: N-methyl-pyrrolidon; number of atoms: 7
Group Atoms Frequency Contribution Total
2 (CH
3
(e)) 7 1 282.015 282.015
9 (C(r)H
2
) 3, 4 2 246.871 493.742
13 ( C(r) (e,r)) 2 1 304.422 304.422
48 (CON ) 1, 5, 6 1 1058.87 1058.87
90 (ve-ring) 15 1 27.0458 27.0458
Sum 2112.0032
T
b
=
2112.0032
7
0.6713
+1.4442
+59.344 = 470.508.
A phenyl-group can be sterically hindered, so that the electron system in hardly available for inter-
molecular interaction as in the case of 1,1,2,2-tetraphenylethane (T
b,est
= 699.07 K, T
b
= 633.0 K).
Generally, results for molecules showing an extraordinary number of functional groups should be used
only with great care. This is true for all group contribution methods.
In some cases, the pure component can rather be regarded as a mixture of tautomers. A typical ex-
ample is found in case of -alkoxy-ketones: 2-butoxy-3-hexanone (T
b,est
= 421.73 K, T
b
= 494.44 K);
2-butoxy-3-pentanone (T
b,est
= 409.06 K, T
b
= 475.94 K); and 2-butoxy-3-butanone (T
b,est
= 400.62 K,
T
b
= 456.22 K).
These examples show, that the relatively few cases of high deviations can usually be contributed to
signicant sterical or electronic effects. Yet, these are not easy to integrate into a group contribution
method, where the only required input for a property estimation should be the molecular structure.
6. Examples
To illustrate the application of the proposed method, detailed procedure for the estimation of T
b
is given
for ethylbenzene, tetraphenylethane and N-methyl pyrrolidon in Tables 911.
7. Conclusion
A group contribution method for the estimation of the normal boiling point of non-electrolyte organic
compounds was developed. The predictions are based exclusively on the molecular structure of the
compound. The results of the newmethod are in most cases far more accurate and never signicantly worse
432 W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433
than previous methods. Structural groups were dened in a standardized formand the fragmentation of the
molecular structures was performed by an automatic procedure to eliminate any arbitrary assumptions.
Second-order corrections were limited to those cases, in which larger structures or structural ef-
fects could not be dened as structural groups (e.g. o-, m- and p-positions at rings, three-, four- and
ve-membered rings and molecules with only one or two hydrogen atoms).
Structural groups were usually dened including the neighboring atoms, thus, implementing knowledge
about the electronic structure in the respective context.
Currently, the database for normal boiling points is extended. In case where only low pressure data
are available, these are used for extrapolation. Once, a signicant amount of new data is available, the
method will be updated and further extended.
In addition, work has started to use the same approach for the prediction of other pure component
properties.
List of symbols
a, b, c adjustable parameters
A, B, C adjustable parameters
B second virial coefcient
H enthalpy
k Boltzmann constant
M molar mass
r distance from center of molecule
R gas constant
S entropy
T absolute temperature
V volume
Greek letters
density
(r) function for the intermolecular potential
Subscripts
b boiling
est estimated
free available for translation
vap vaporization
Superscripts
L liquid
V vapor
References
[1] J.M. Prausnitz, R.N. Lichtenthaler, E.G. de Azevedo, Molecular Thermodynamics of Fluid Phase Equilibria, 3rd Edition,
Prentice-Hall, Upper Saddle River, 1999.
W. Cordes, J. Rarey / Fluid Phase Equilibria 201 (2002) 409433 433
[2] L.R. Constantinou, R. Gani, AIChE J. 40 (10) (1994) 16971710.
[3] K.G. Joback, R.C. Reid, Chem. Eng. Commun. 57 (1987) 233243.
[4] S.E. Stein, R.L. Brown, J. Chem. Inf. Comput. Sci. 34 (1994) 581587.
[5] J. Marrero-Morejon, E. Pardillo-Fontdevila, AIChE J. 45 (3) (1999) 615621.
[6] Aldrich Handbook of Fine Chemicals, Aldrich Chemical Co., Milwaukee, WI, 1990.
[7] J. Graselli (Ed.), Handbook of Data of Organic Compounds (HODOC), CRC Press, Boca Raton, FL, 1990.
[8] B.I. Lee, M.G. Kesler, AIChE J. 21 (1975) 510527.
[9] W. Cordes, J. Rarey, F. Delique, J. Gmehling, in: D. Ziessow (Ed.), Software Development in Chemistry, in: Proceedings of
Computer in der Chemie, Vol. 7, Springer, Berlin, 1993.

You might also like