-
J. of Supercritical Fluids 55 (2010) 421437
Contents lists available at ScienceDirect
The Journal of Supercritical Fluids
journa l homepage: www.e lsev ier .com/ locate /supf lu
Equations of state: From the ideas of van der Wa
Georgiosa Center for Ene ng, Teb The Petroleum Emira
a r t i c l
Article history:Received 21 JuReceived in reAccepted 16 O
Keywords:Phase equilibrvan der WaalsCubic equationExcess Gibbs
eMixing rulesAsymmetric mixturesAssociating uidsSAFTCPANRHB
ltedidelof va
n thisxcessWe illed. Mlatedon of
developed using the so-called EoS/G mixing rules and we
illustrate with the same methodology howthesemixing rules should
best be used for size-asymmetric systems. Despite the signicant
capabilities ofcubic equations of state, their limitations lie
especially in the description of complex phase behavior,
e.g.liquidliquid equilibria for highly polar and/or hydrogen
bonding containing molecules. In these cases,advanced equations of
state based on statistical mechanics that incorporate ideas from
perturbation (e.g.SAFT and CPA), chemical (e.g. APACT) and lattice
(e.g. NRHB) theories are preferred. Some of the mostpromising
approaches are briey outlined here. Their capabilities and
limitations/challenges, pointing
1. Introdu
Equationdynamic mover extensame timebe calculatof state,
liteclaiming toideal liquidvan der Wation, speedwhy suchindustrial
aallel with tfor the liquoped and m
CorresponE-mail add
ieconomou@p
0896-8446/$ doi:10.1016/j.out to future research needs are also
discussed. 2010 Elsevier B.V. All rights reserved.
ction
s of state (EoS) represent the cornerstone of thermo-odels. They
can be used to represent phase equilibriasive temperature and
pressure ranges, while at theother properties like thermal and
volumetric ones caned. Since the advent of the van der Waals
equationrally hundreds such models have been published, allbe used
with different degree of success for non-
s and gases. The elegance of cubic EoS, following theals
developments, combined with ease of implementa-and accurate results
in many practical cases explainmodels dominated and to some extent
still do pplications, especially in the oil & gas sector. In
par-he development of EoS, theories explicitly designedid phase
(excess Gibbs energy models) were devel-uch knowledge was obtained.
These activity coefcient
ding author. Tel.: +45 45252859.resses: [email protected] (G.M.
Kontogeorgis),i.ac.ae (I.G. Economou).
models are limited to low pressure applications and an equa-tion
of state should be used for describing deviations from theideal gas
law in the vapor phase. As more demanding applicationsappeared
combinedwith need for accurate design of industrial pro-cesses, it
became apparent that cubic EoS did have limitations,especially for
describing highly polar and/or hydrogen bondingcompounds.
Consequently, higher order EoS rooted to statisticalmechanics
became increasingly important and widely used bychemical engineers
over the last 20 years. The increased com-plexity of these models,
due to a stronger physical basis, resultin a higher accuracy over
cubic EoS for highly non-ideal mix-tures.
This review is divided into two parts. In the rst part we
illus-trate how far we can go with cubic equations of state and
howmuch we can learn on their capabilities and limitations by
taking acloser look at the excess Gibbs energy and the activity
coefcientexpressions which can be derived from them. The very
importantrole of mixing and combining rules will be illustrated. In
the sec-ond part, we turn our interest on some of the recently
developedadvanced EoS, especially those based on statistical
mechanics forassociating uids. We will show some examples
illustrating theirsignicant capabilities, but we will also outline
some limitationswhich indicate areas where future research is
required.
see front matter 2010 Elsevier B.V. All rights
reserved.supu.2010.10.023M. Kontogeorgisa,, Ioannis G.
Economoub
rgy Resources Engineering (CERE), Department of Chemical and
Biochemical EngineeriInstitute, Department of Chemical Engineering,
PO Box 2533, Abu Dhabi, United Arab
e i n f o
ne 2010vised form 16 October 2010ctober 2010
ia
s of statenergy equations
a b s t r a c t
The ideas of van der Waals have resuand PengRobinson (PR) which
are wthought that the range of applicabilityrelatively similar in
size. We employ iequations of state by looking at the ederived
fromthese equationsof state.to the mixing and combining rules
ussize-asymmetric systems are more rethe functionality of the cubic
equati
Eals to association theories
chnical University of Denmark, Building 229, DK-2800 Lyngby,
Denmarktes
to cubic equations of state like SoaveRedlichKwong (SRK)y used
in the petroleum and chemical industries. It is oftenn der
Waals-type models is limited to mixtures of compoundswork an
approach for investigating the various terms of cubicGibbs energy
and activity coefcient expressions which areustrate that the
results of cubic equationsof state are sensitiveoreover, it is
shown that previously reported deciencies forto the van der Waals
one uid mixing rules used rather thanstate itself. Improved models
for polar systems have been
-
422 G.M. Kontogeorgis, I.G. Economou / J. of Supercritical
Fluids 55 (2010) 421437
2. Cubic equations of state
2.1. Physical concepts in cubic equations of state
The vanadvent of a
P = RTV b
In somewyears sinceare still muical industrreviewed inpractical
puabbreviated
P = RTV b
and the Pen
P = RTV b
SRK anderature retmolecular dattractive tand PR conforces and
calso combinchoice usedder Waals othe classic
a =i=1
j=1
b =i=1
j=1
aij =
aiaj
bij =bi + b
2
There arthem is thacess is largin detail bywhy these sical.
First othe basic varelated to thetc.) of thepound paraacentric
facthree-paramtication ofThe vdW1ffor the secofrom statistthe
cross-etheory, as itric mean copotential (
It is alsofor mixturewill be call
they cannot represent polymer solutions. This is another
miscon-ception, as will be illustrated in Section 2.4. We will see
that theperformance of modern cubic EoS like SRK and PR may not be
asmuch dependent on their precise functional form but relies
much
he cmpotandons oande usped e
cesse equ
ougng cu1978fciedyn
n
lnii
(6) pom asingressthe te exc:(
i
+ PRT
(i
+(
R
+ PRT
(i
+(
R
ablder Waals (vdW) equation of state [1] marked thenew family of
models which are since called cubic EoS:
a
V2(1)
hat different forms than the vdW EoS and almost 130van der Waals
proposed his equation of state, cubic EoSch used today especially
in the petroleum and chem-ies. There are numerous such cubic EoS
and many areliterature [26], but the most widely used ones for
rposes are the SoaveRedlichKwong from 1972, oftenas SRK [7]:
a(T)V(V + b) (2)
gRobinson (PR) one from 1976 [8]:
a(T)V(V + b) + b(V b) (3)
PR as well as many other cubic EoS presented in lit-ain vdWs
repulsive term, despite its mismatch withynamics data for
hard-sphere uids, while differenterms are used. Most cubic EoS
including vdW, SRKtain two parameters (a, b) reecting
intermolecularo-volume effects. These EoS require mixing (and
oftening) rules for application to mixtures and one popularalso by
van der Waals and van Laar is the so-called vanne-uid (vdW1f)
mixing rules, Eq. (4), together with
al combining rules shown in Eq. (5):
xixjaij
xixjbij(4)
(1 kij)j (1 lij)
(5)
e several misconceptions regarding cubic EoS. One oft they are
completely empirical models whose suc-ely attributed to
cancellation of errors. As discussedKontogeorgis and Folas [3]
there are many reasons
imple models should not be considered entirely empir-f all, they
account for free-volumes (Vf =Vb) and forn der Waals forces and the
energy parameter can beephysical properties (polarizabilities,
dipolemoments,intermolecular potential. Moreover, if the pure
com-meters are estimated from the critical properties (andtor),
cubic EoS are representations of the two- or theeter corresponding
states principle. There is also jus-the classical mixing and
combining rules of cubic EoS.mixing rules (Eq. (4)) satisfy the
quadratic mixing rulend virial coefcient (B=
i
jxixjBij), which is derivedical mechanics. The geometric mean
combining rule fornergy parameter, Eq. (5), has its origin on the
Londoncan be shown that for the dispersion forces a geomet-mbining
rule is derived for the cross intermolecularij =
ij).
often stated that cubic EoS cannot be satisfactorily useds with
compounds differing signicantly in size, whiched size-asymmetric
mixtures in this discussion. Thus,
upon tpure coundersequatienergyto makdevelonext.
2.2. Exanalyz
Althstudyilate asity coethermo
gE
RT= l
lni =
Eq.sion frrules. Uthe exp
For2.1, th
vdW
gE
RT=
SRK:
gE
RT=
PR:
gE
RT=hoice of mixing and combining rules and the way theund
parameters are estimated. One way to analyze andthe capabilities
and limitations of (cubic and other)f state is to look at the
expressions for the excess Gibbsactivity coefcient which are
derived from them ande of the knowledge we have for models and
theoriesxplicitly for the liquidphase. This approach is
discussed
Gibbs energy and activity coefcients: a method toations of
state
h the use of excess Gibbs energy/activity coefcients forbic EoS
started with the HuronVidal mixing rules as[9], calculating the
excess Gibbs energy, gE, and activ-
nts, , from cubic EoS is straightforward using
classicalamics:
i
xi lni =
i
xi lni (6)
(7)
ermits the derivation of the excess Gibbs energy expres-n EoS
without any knowledge of mixing and combiningEq. (7) and a choice
for themixing and combining rules,ion for the activity coefcient is
derived.hree cubic EoS (vdW, SRK and PR) presented in Sectioness
Gibbs energy equations are:
xi ln(
Vi biV b
))+(
1RT
(i
xiaiVi
aV
))(
V
i
xiVi
)(8a)
xi ln(
Vi biV b
))
1T
[i
xiaibi
ln(
Vi + biVi
) a
bln(
V + bV
)])(
V
i
xiVi
)(8b)
xi ln(
Vi biV b
))
1
T2
2
[i
xiaibi
ln
(Vi + (1 +
2)bi
Vi + (1
2)bi
)
n
(V + (1 +
2)b
V + (1
2)b
)])+ P
RT
(V
i
xiVi
)(8c)
-
G.M. Kontogeorgis, I.G. Economou / J. of Supercritical Fluids 55
(2010) 421437 423
With Eqs. (8a)(8c) as starting point, it is tempting to
divide,somewhat arbitrarily, gE into three parts. From left to
right, we canidentify a so-called combinatorial-free volume term
(accountingfor size or entropic effects), a residual term
(accounting for ener-getic differeterminologcientmodelLaar,
Florypolymers [2widely used
van Laar
gE
RT= 1
RT
(
FloryHuggzero interac
gE
RT=
i
xi
with the vo
i =xiVijxjV
and is the[10].
Wilsonin a form scontributio
gE
RT= x1 ln
x1 l
x2 l
UNIQUA
gE
RT= x1 ln
x1q
x2q
where si
tions calculusing van d[16] or UNI
NRTL [13
gE
RT= x1x2
[Gij = exp(
In all theters in thdifferencesdata.
EntropicUNIQUAC) [
gE
RT=
i
xi
Vw is thevanderWaals volume, taken fromBondi [16]orUNIFAC[17]
tables.
As can be seen in Eqs. (9a)(9h), the van Laar and NRTL mod-els
have no combinatorial-free volume term; they are essentially
etics shoatoral towithtingaboquatombiy coeare Ee) terf
weic-FV/matthaned las mixented
e vann the
an bing th
comcan
i
xi
xiVjx
lumexp
the sij = lij
ln
(
Laarer We (E=bi aivedGibb
1RT
(
(biRT
solt of ve mo
xibi
jxjb
r, Hptionnces) and an excess volume term. This division and
thisy is inspired by the one used in explicit activity coef-s,
derived for the liquid (condensed) phases like the vanHuggins,
local composition models and Entropic-FV for,3]. The expressions
for some of the most known andexplicit activity coefcient models
are given below:
:i
xiaibi
ab
)(9a)
ins and regular solution theory (for binary mixtures,tion
parameters):
lnixi
+ VRT
ij(i j)2 (9b)
lume fraction dened as:
j(9c)
solubility parameter as dened byHildebrand and Scott
[3,11] equation (shown here for binary mixtures anduitable for
identifying the combinatorial and residualns):
1x1
+ x2 ln2x2
n(1 + 2 exp
(21
RT
))n(2 + 1 exp
(12
RT
)) (9d)
C [12] (for binary mixtures):
s1x1
+ x2 lns2x2
+ Z2
(x1q1 ln
1s1
+ x2q2 ln2s2
)
1 ln(1 + 2 exp
(U21
RT
))
2 ln(2 + 1 exp
(U12
RT
))(9e)
i are, respectively, the segment and surface area frac-ated in a
similar way to volume fractions in Eq. (9c) buter Waals volume and
surface areas (taken from BondiFAC [17] tables) instead of
volumes.] (for binary mixtures):
21G21x1 + x2G21
+ 12G12x2 + x1G12
](9f)
aijij) ij =gij gjj
RT(9g)
ree models (Wilson, NRTL and UNIQUAC), the param-e exponential
terms denote the energy parameterand they are typically estimated
from experimental
-FV (the residual term could be taken from UNIFAC or14,15]:
lnf v
i
xi+ g
E,res
RT=
i
xi ln
(Vi Vw,iV Vw
)+ g
E,res
RT(9h)
energmodelcombinidenticis usedinterac
Thecubic ebasic cactivit(compvolumterm
oEntroplogicalratherobtainvariouis pres
2.3. Thsolutio
It cassuming andenergy
gE
RT=
f vi
=
(The voThe
underwith k
lni =
Van(van dthe timuidsVihe derexcess
gE
RT=
lni =
Thecontexthan th
i = Late
assummodels, as is the regular solution theory. The otherwn in
Eqs. (9a)(9h) as well as UNIFAC [17] have both aial and a residual
term. UNIFACs combinatorial term isthat of UNIQUAC, whereas a
group-based residual termtwo (ormore) group-based interaction
parameters pergroups.ve analysis sheds some light on the
capabilities ofions of state. Even the simplest cubic EoS contain
thenatorial-free volumeand residual contributions like
thefcientmodels explicitly developed for the liquid phaseqs. (8)
and (9)). Moreover, the size (combinatorial-freem of cubic EoS is
functionally similar to the equivalentll-established polymer
models, like FloryHuggins and. The above analysis is based,
however, on phenomeno-hematical similarities and may be more of
indicativeof conclusive character. Additional evidence will be
ter by looking at the activity coefcient values usinging rules.
First the special case of the van der Waals EoS.
der Waals EoS against the van Laar and the regularories
e shown for the vdW EoS that starting from Eq. (8a),at the
excess volume is zero and that the vdW1f mix-bining rules apply
(Eqs. (4) and (5)), the excess Gibbs
be equivalently written as:
lnf v
i
xi+ V
RTij(i j)2 (10a)
fi
jVfjVfi = Vi bi i =
ai
Vi(10b)
e fraction is dened in Eq. (9c))ression for the activity
coefcient from the vdW EoSame assumptions (vdW1f mixing and
combining rules=0) is:
f vi
xi
)+ 1
(f v
i
xi
)+ Vi
RT(i j)22j (11)
s starting point was the equation of state of his teacheraals)
and the same mixing and combining rules used atqs. (4) and (5)).
Van Laar further assumed that for liq-nd the excess volume is zero.
Under these assumptions,a liquid phase activity coefcient model,
which for thes energy and the activity coefcient is expressed
as:
i
xiaibi
ab
)= b
RTij(i j)2 (12a)
(i j)22j)
(12b)
ubility parameters and volume fractions are, in thean Laars
model, dened using the co-volumes ratherlar volumes:
ji =
ai
bi(12c)
ildebrand and Scott [10] maintained some of the basics of
regular solutions (gE =hE, VE = SE =0). Their regu-
-
424 G.M. Kontogeorgis, I.G. Economou / J. of Supercritical
Fluids 55 (2010) 421437
lar solution theory is shown in Eqs. (13a)(13c) for
binarymixtures.In absence of any interaction parameters, the
regular solution the-ory is identical to the van Laar model with
the co-volume beingreplaced by the molar volumes and using the
experimental solu-bility param
gE
RT= V
RT
lni =(
ViRT
Now thethe usual w
=
Hv
It is, thution theorievdW, theybe used fotions. In
cocombinatorbe used forwith the regdened in E
gE
RT=
i
xi
lni = ln(
All the etion are bas(Eqs. (4) anvan der Waand (14b).an
activitytures seenas FloryHuimportant fcommon, arcombinatora
free-volumand an enerregular solucubic EoSbeand, if not,EoS (or
both
2.4. Mixing
A moreexaminingEoS using aconsider ththe
activity(15a)(15c)mixing rule
SRK:
lni = ln[
i
PR:
lni = ln[Vi biV b
]+ 1
[Vi biV b
]+ i
2
2ln
[Vi + (1 +
2)bi
Vi + (1
2)bi
]
2
a
Tb + 2
a
RT
[willite dixing=
etering e:
= ln
+
= ln
+
+
wh
nctio)turecanbof thmpaEoS trentutioexp
natorr ananatorcom1
ue toe re
residEoS aetric
bic Ee deanuneters rather than those obtained from the vdW
EoS:
ij(i j)2 (13a)
(i j)22j)
(13b)
volume fractions are dened not via Eq. (12c) but inay via Eq.
(9c) and the solubility parameter is given as:
ap RTV
(13c)
s, clear that both the van Laar and the regular solu-s follow
the spirit of the vdW EoS. However, unlikeare purely energetic
models and alone they cannotr size-asymmetric mixtures including
polymer solu-mbination with the FloryHuggins or the
Entropic-FVial/free-volume terms, see Eqs. (9b) and (14), they
canpolymers [2,3,18,19]. The Entropic-FV model combinedular
solution theory canbewritten as (volume fractionsq. (9c)):
lnf v
i
xi+ V
RTij(i j)2 (14a)
f vi
xi
)+ 1
(f v
i
xi
)+ Vi
RT(i j)22j (14b)
quations presented for activity coefcients in this sec-ed on the
vdW1fmixing and classical combining rulesd (5)). Under these
assumptions, we can compare theals EoS in the form of Eqs. (10) and
(11) with Eqs. (14a)We conclude that the vdW EoS also when written
ascoefcient model contains many of the essential fea-in successful
and well-known polymer models suchggins, regular solution theory
and Entropic-FV. Theseeatures, which vdW EoS and polymer models
have ine an explicit representation of the size-asymmetry via
aial-free volume term which has the same functionality,e denition
similar to the one used in polymer models
gy term which is mathematically similar to that of thetion
theory. Thus, the following question is raised: canused for
size-asymmetricmixtures includingpolymersis this attributed to the
size or the energy term of the)?
and combining rules in the van der Waals spirit
detailed analysis of the cubic EoS can be obtained bythe
expression for the activity coefcient derived fromspecic set of
mixing and combining rules. At rst wee classical vdW1f mixing
rules, Eq. (4), and using themcoefcient equationswith SRK and PR
are shown in Eqs.. Eqs. (15a) and (15b) can be further simplied if
a linearis used for the co-volume parameter, b=
ixibi
Vi biV b
]+ 1
[Vi biV b
]+ i ln
[Vi + bi
Vi
]
ln[V + b
V
]+ (bVi Vbi)
V(V + b) (15a)
where
=bR
bi =
i = b
Weat innuid meter, bparamfollow
SRK
ln1
PR:
ln1
(The fuscripts
Mixwhichbutionand cocubicof diffecontribenergycombiIn
theicombi(16) (lnb2))) done. Thgetic (the PRasymm
(i) Cutivth i2ln
[V + (1 +
2)b
V + (1
2)b
]+ (bVi Vbi)
V(V + b) + b(V b) (15b)
j
xjbij ai = a + 2
j
xjaij
aia
+ 1 bib
] (15c)
simplify these expressions by considering the equationsilution
for binary mixtures using the van der Waals onerules, the linear
mixing rule for the co-volume param-
ixibi, and the geometric mean rule for the cross-energy(with kij
=0; aij =
aiaj). Under these assumptions, the
quations can be derived:
(V1 b1V2 b2
)+ 1
(V1 b1V2 b2
)+ a1
b1RTln
(1 + (b1/V1))(1 + (b2/V2))
a2b2RT
(V1b2 V2b1V22 + b2V2
)+ b1
RT
[a2
b2
a1
b1
]2ln(1 + b2
V2
)(16a)
(V1 b1V2 b2
)+ 1
(V1 b1V2 b2
)a1
2
2b1RTln
f (b1, V1)f (b2, V2)
+ a2b2RT
(V1b2 V2b1
V22 + 2V2b2 b22
)
b1
2
2RT
[a2
b2
a1
b1
]2ln f (b2, V2)
ere f (bi, Vi) =[
Vi + (1 +
2)biVi + (1
2)bi
](16b)
n f(b2, V2) is the same as f(b1, V1), with change of sub-
s containing alkanes are nearly athermal solutionseused for
testing the combinatorial-freevolumecontri-ermodynamic models. It
is, thus, of interest to computere the innite dilution activity
coefcient values fromo experimental data for mixtures containing
alkanessize as well as examining the values of the different
ns. Unlike what was the case with the excess Gibbsressions, it
may be difcult to identify which are theial/free-volume and
residual parts in Eqs. (15) and (16).lysis from 1998, Kontogeorgis
et al. [20] considered theial-free volume to be the rst three terms
of Eqs. (15) orbFV, = ln((V1 b1)/(V2 b2)) + 1 ((V1 b1)/(V2
similarities with polymer models like the Entropic-FV
maining terms of Eqs. (15) or (16) constituted the ener-ual)
term in that analysis. Kontogeorgis et al. [20] usednd the most
important conclusions of their analysis formixtures of alkanes
are:
oS with vdW1f mixing rules and kij =0 result to posi-viations
from Raoults law (activity coefcients higherity), in contradiction
to theexperimentaldata (negative
-
G.M. Kontogeorgis, I.G. Economou / J. of Supercritical Fluids 55
(2010) 421437 425
0,5
0,6
0,7
0,8
0,9
1
1,1
1,2
1,3
14 16 18 20 22 24 26 28 30 32 34 36
Carbon number of Alkane
Infi
nit
e D
ilu
tio
n A
cti
vit
y C
oe
ffic
ien
t o
f H
ep
tan
e exp. data
Modified UnifacPR EoS (vdw1f mix. rule)
PR EoS (a/b mix. rule)
Fig. 1. Activit293373K asequation of staand (5)) and ththe
modied U
deviaticientssee also
(ii) The comunity (much htive vavalidwmean fco-volu(Berthe
(iii) It is iminteracto corrresiduacient foresultsto
veryseriesobtaine
At a rssmooth kij-that cubicalthough ththe modelsand
activitycombinatorafter.
Table 1Percentage decoefcients at(with the vdWactivity coefc
Mixture
nC5/nC6nC5/nC20nC5/nC22nC5/nC24nC5/nC28nC5/nC32nC5/nC36
Following the ideas of van Laar and HuronVidal, we rec-ognize
that at the innite pressure limit (when Vb)
thecombinatorial/free-volume termvanishes and the remaining
termscorrespond to the residual (or energetic) part. Thus, for all
threeEoS consideinnite pre
gE,EoS,P
RT=
= RT
(
= 1 (vdW = ln 2 (S = 0.623
SRKrt oatorhic
=bi):
= ln
+
= lnc1
+2
the fith Snd (1parm ofyingactlyility p
es =y coefcients at innite dilution of n-heptane in alkane
solvents ata function of the alkane carbon number using the
PengRobinsonte. Results are shown using vdW1f mixing rules (with
kij =0, Eqs. (4)e a/bmixing rule, Eq. (19). For comparison are
shown the resultswithNIFAC activity coefcient model [47].
ons from Raoults law). Moreover, these activity coef-become even
higher with increasing size asymmetry;Fig. 1 and Table
1.binatorial/free-volume term has values slightly below
close to one), while the residual term reaches valuesigher than
one, which explains the very large posi-
lues for the activity coefcients. These conclusions arehen the
classical combining rules are used (geometricor the cross energy,
arithmetic mean rule for the crossme, Eq. (5)), while the other
combining rules testedlot, Lorentz, Sandler) failed (in the context
of PR).portant to correct the combining rule for b12 (via antion
parameter), which can result at the same timeect values for the
combinatorial/free-volume and thel contributions, thus also for the
total activity coef-r alkane mixtures. This approach results to
good VLEaswell e.g. a single lij (=0.02) togetherwith PR can
resultgood correlation of VLE for the whole ethanealkane
(up to ethaneC44). Equally good results cannot bed using a
single kij value.
t look, these and other results, for example the noncarbon
number trends for gas/alkanes, may indicateEoS cannot represent
size-asymmetric mixtures well,
Forthe pacombin(16b) wand Vi
SRK
ln1
PR:
ln1
wherement w(18a) a
Comlast teridentifEoS exsolub
ln,r1is could be attributed more to the energetic term of. A
closer look [3,21,22] at the excess Gibbs energycoefcient equations
reveals the true identity of the
ial/free-volume (size-related) term, as explained here-
viations between the experimental and calculated pentane
activityinnite dilution in various alkane solvents at 373K using
the PR EoS1f and the a/b mixing rule, Eq. (19)) and with the modied
UNIFACient model [47].
PR (vdW1f, kij =0) PR (/b rule) Modied UNIFAC
6 4 424 1 1028 5 1139 3 753 6 667 10 585 13 3
Consequsize-asymmity coefcie(18a) or (18the followin
a
b=
i
xia
b
b =
i
xib
It can beEqs. (15a) aple a/b mixthe simpleing the resFigs. 14
anrules (Eq. (1size asymmred (vdW, SRK and PR) here, the excess
Gibbs energy atssure can be written as:
gE,res
RT=
(1RT
(i
xiaibi
ab
))
i
xii )
= ab
, vanLaar)RK)(PR)
(17)
and PR at innite dilution, it can be shown thatf Eqs. (16a) and
(16b) which corresponds to theial/free-volume term (i.e. the part
of Eqs. (16a) andh would disappear at innite pressure by setting
V=bcan be expressed as:
combFV1 = ln(
V1 b1V2 b2
)+ 1
(V1 b1V2 b2
)a1
b1RTln
(1 + (b1/V1))(1 + (b2/V2))
+ a2b2RT
(V1b2 V2b1V22 + b2V2
)(18a)
ombFV = ln(
V1 b1V2 b2
)+ 1
(V1 b1V2 b2
)a12b1RT
lnf (b1, V1)f (b2, V2)
+ a2b2RT
(V1b2 V2b1
V22 + 2b2V2 b22
)(18b)
(bi,Vi) function is given in Eq. (16b). This is also in
agree-akomani and Brignole [21] who called these terms Eqs.8b) as
non-residual.
ingwith Eqs. (16a) and (16b), it can be seen that only theEqs.
(16a) and (16b) remains even at innite pressure,as the residual
term (true energetic tem) of the cubicthis last contributionwhich
contains the EoS-generatedarameters (i = (
ai/bi)) e.g. for SRK:
b1RT
[a2
b2
a1
b1
]2ln(1 + b2
V2
)ently, an assessment of the true value of cubic EoS foretric
mixtures can be obtained by computing the activ-nts from Eqs. (18a)
and (18b). It can be shown that Eqs.b) are the total activity
coefcients fromcubic EoSusingg simple mixing rules:
i
i
i
(19)
proved that Eqs. (18a) and (18b) can be derived fromnd (15b) at
innite dilution conditions using the sim-ing rule shown in Eq.
(19). Thus, based on the above,mixing rules of Eq. (19) have the
property of eliminat-idual term of cubic EoS. Typical results are
shown ind Table 1 illustrating that cubic EoS with these mixing9))
predict reliable values of the activity coefcients foretric alkane
mixtures, even with kij =0. The predictions
-
426 G.M. Kontogeorgis, I.G. Economou / J. of Supercritical
Fluids 55 (2010) 421437
10
15
20
25
30
35
40
0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8mole fraction of Hexane
Pre
ss
ure
(b
ar)
exp (liquid phase)exp (vapor phase)PR eos (a/b rule)PR eos
(vdw1f rule)
Fig. 2. Vaporliquid equilibria for hexane/nC16 at 623K using the
PengRobinsonequation of state. Results are shown using the vdW1f
mixing rules (with kij =0, Eqs.(4) and (5)) and the a/b mixing
rule, Eq. (19).
5
15
25
35
45
55
0
Pre
ssu
re (
bar
)
Fig. 3. Pressu373.15K usingvdW1f mixing
are in qualimental valubetter commixing ruleequilibriumhave a
perc
5
15
25
35
45
0
Pre
ssu
re (
bar
)
Fig. 4. PressurPengRobinso(with kij =0, Eq
with the vdW1f rules. Similarly the errors (respectively with
Eq.(19)/vdW1f) are 12% and 20% for C2/C28 and 13% and 33% forC2/C36
(373K). These are typical results and similar deviations
areobtained for other systems as well. All results with both
mixingrules are prto zero.
This invfor cubic Eoattributedmcially that ofunctionalitmixing
rulenext.
2.5. Excessclassical to a
As illustof the vdWtems like aprovided thOne suitabl
vdWexpractHur] andhichparaexp. data
PR eos (a/b mix. rule)
PR eos (vdw1f mix. rule)
nor thegin areno interst by[2426ruleswenergy0,30,20,1
mole fraction of methane
reliquid phase mole fraction plot for methane/nC36 mixture atthe
PengRobinson equation of state. Results are shown using therules
(with kij =0, Eqs. (4) and (5)) and the a/b mixing rule, Eq.
(19).
tative and often quantitative agreement to the experi-es.
Moreover, these predictions are shown to be muchpared to those of
cubic EoS using the classical vdW1fs (i.e. using Eqs. (16a) and
(16b)), even for vaporliquidcalculations. For example, for the
system C1/C36 we
entage deviation in pressure 3% with Eq. (19) and 14%
0,50,40,30,20,1
mole fraction of Ethane
exp (liquid phase)
PR eos (a/b rule)
PR eos (vdw1f rule)
eliquid phase mole fraction plot for ethane/nC28 at 573K using
then equation of state. Results are shown using the vdW1f mixing
ruless. (4) and (5)) and the a/b mixing rule, Eq. (19).
excess Gibband from anreference p(
gE
RT
)EoSP
=
The Hurand in this cthe size terthe EoS van(
gE
RT
)EoS
=
which in th
a
b=
i
xia
b
vdW = 1SRK = lnPR = 1
2
It is thusapproach (model whic(21) e.g. NRrule is empshown
thatto an excellrily phase ephase behationparameshouldbe reinnite
refeavailable inpressure daedictions, i.e. the interaction
parameters are set equal
estigation illustrates that problems previously reportedS in
representing size-asymmetric systems should beore to the use of
specicmixing/combining rules espe-f the energy parameter, rather
than the van der Waalsy of the vdW repulsive term. The studies of
the EoS/GE
s will shed some additional light on this, as explained
Gibbs energy at zero and innite pressures: fromdvanced mixing
rules (EoS/GE)
rated from the analysis in Sections 2.22.4, cubic EoS-type can
be used successfully for size-asymmetric sys-ctivity coefcient
calculations for mixtures of alkanesat mixing rules other than the
vdW1f ones are used.e choice is the simple a/b rule, Eq. (19), but
neither this1f mixing rules with their random-mixing based ori-
ected to work well for polar mixtures, especially whenion
parameters are used. Another approach, pioneeredon and Vidal [9]
and later by Mollerup [23], MichelsenWongSandler [27,28] is the
so-called EoS/GE mixingprovide away to derivemixing rules for,
especially, themeter of cubic EoS. The starting point is equating
thes energy (or the excess Helmholtz energy) from the EoSexplicit
activity coefcient model (*) at some specic
ressure, P:(gE
RT
)model,P
(20)
onVidal approach assumes innite reference pressurease in
accordance to the discussion in Sections 2.22.4,ms
(combinatorial/free-volume and excess volume) ofish and Eq. (20)
reduces to:(
gE
RT
)model,
(
gE
RT
)EoS,res=(
gE
RT
)model,res(21a)
e case of vdW, SRK and PR can be written as:
i
i 1
gE,
2
2ln
(2 +
2
2
2
) (21b)
understood that in combination with the HuronVidalinnite
pressure limit), an explicit activity coefcienth only has a
residual-type term should be used in Eq.TL or the residual term of
UNIQUAC. The linear mixingloyed for the co-volume parameter. Many
studies haveSRK and PR using the HuronVidal mixing rule resultent
correlation model capable of describing satisfacto-quilibria for
binary complexmixtures and for predictingvior of multicomponent
mixtures [3,9,29]. The interac-ters of the explicit excessGibbs
energymodel, e.g.NRTL-estimated togetherwith the cubic EoS since,
due to therence pressure used, existing parameters of the
modelsdatabases like Dechema which are obtained from lowta cannot
be used.
-
G.M. Kontogeorgis, I.G. Economou / J. of Supercritical Fluids 55
(2010) 421437 427
About 10 years after Vidal, Michelsen [24,25] showed how
anEoS/GE mixing rule can be derived at zero reference pressure.
Herethe mathematical analysis is a bit more complicated comparedto
the innite pressure limit and neither the combinatorial/free-volume
noran implicitthe energy(
gE
RT
)mode0
where the rthe qe() fuvalues >function is0.9. Thus, eexplicit
excoften simplMHV1andMimation ofequations [
MHV1:
= 1AM
(g
MHV2:
q1
(
i
=(
gE
RT
)
The origmodels [2,3
The MHexpressionsbecause ofderived from
gE,EoS,0P
RT
where the ction of state
Notice tonly by thcombinatorzero referention can befractions
ofet al. [31] anMHV1/MHVVLE for comup to 200bmodel andwsure data.
Nbeen a breaFor mixtureas noUNIFAeters, e.g. Ccubic EoS w
As Kalospressure mduce the ex
model they are combined with. Moreover, the difference from
thereference activity coefcient model increases with increasing
size-asymmetry of the mixture components [33]. Another issue,
moreof theoretical nature, is the inability of EoS/GE models to
repro-
e thoefhisorule
ults wfact. Actuuce ts, inhat tn evcubis, e.g
ws nohis lg rulis prry exs in.
S/GE
es
sawre limal (enre Eore M
ixiequat actice patorthe. This, asmaterials: Thcel o
nd Aferene-asyan eymmf MH
2 inapo
to revani
s, likode
q. (20our EM [3
(gE
RTthe residual term of the EoS vanish. In the general
case,mixing rule is obtained at zero reference pressure
forparameter:
l,+
i
xi ln(
b
bi
)= qe()
xiq
ei (i) (22)
educed energy parameter is dened in Eq. (15c), whilenction
depends on the EoS used and it is dened only forlim. This excludes
mixtures with gases, since the qe()dened for reduced temperature
Tr-values up to aboutven though Eq. (22) ensures full reproduction
of theess Gibbs energy model the cubic EoS is combined with,er or
even explicit mixing rules are used in practice.
TheHV2rulesarebasedona linearandaquadratic approx-
the qe() function and they are given by the following26,31]:
E,
RT+
i
xi ln(
b
bi
))+
i
xii (23a)
xii
)+ q2
(2
i
xi2i
)
+
i
xi ln(
b
bi
)(23b)
inal publications [26,30] and books on thermodynamic] contain
values for q1, q2 and AM for both models.V1/PSRK mixing rules
correspond to the followingof the excess Gibbs energy at zero
pressure. Exactly
the assumptions made, this is the excess Gibbs energya cubic EoS
at approximately zero pressure:
i
xi lnbib
+(
q
RT
(i
xiaibi
ab
))(23c)
onstant q depends on the functional form of the equa-and the
approximation used.
hat MHV1 differs from the HuronVidal mixing rulee (
ixi ln(b/bi)) term which is a FloryHuggins typeial term,
originating from the cubic EoS at approximatecepressure, as canbe
seen inEq. (23c). TheMHV1equa-also derived [23] assuming equal
values for the packingpure compounds and of the mixture. As shown
by Dahld Dahl andMichelsen [26], using either the exact or the2
mixing rules, it is possible to predict high pressureplex mixtures
like acetonewater and ethanolwater
ar employing UNIFAC as the explicit activity coefcientith UNIFAC
parameters solely obtained from lowpres-o additional adjustable
parameters are used. This has
kthrough development in the eld of predictivemodels.s with
gases, e.g. CO2/alkanes or N2/alkanes, however,C parameters are
available for gases, newgroupparam-O2/CH2 and N2/CH2 should be
determined from theith the incorporated EoS/GE mixing rule.piros et
al. [32] showed, the approximate zero referenceodels (MHV1, MHV2,
PSRK, etc.) do not fully repro-cess Gibbs energy and especially the
activity coefcient
duce thvirial crules. Tmixingful resto thecientreprodmid 90cated
tfrom acess ofmodelperformalkanebons. T(mixinthat
thisfactomodelsection
2.6. Eomixtur
Wepressuresidupressupressution (
Whenexplicireferencombine.g. incel outsystembe
estibinatowritesple can[39] athe difing sizofferssize-ascase
oCO2/CHtorily vordershouldmodelthese mfrom Ethese f
LCV
= C1eoretically correct quadratic mixing rule for the
secondcient, which is satised by the simpler vdW1f mixingbservation
led to thedevelopmentof theWongSandlers [27,28], but it has not
been veried that the success-ith the WongSandler mixing rules can
be attributed
they satisfy the mixing rule for the second virial coef-ally,
the WongSandler mixing rules also fail to fullyhe activity
coefcient model incorporated [34]. In thea number of publications
[3538] it has been indi-
he approximate zero reference pressure models sufferen more
serious problem, which is, in light of the suc-c EoS discussed
previously, somewhat surprising. These. MHV1, MHV2, PSRK but also
WongSandler do notell for size-asymmetric mixtures, neither
mixtures ofr mixtures of gases (ethane, CO2, N2) with hydrocar-ed
to the development of various EoS/GE-type modelses) for correcting
this deciency. We believe, however,oblem must be addressed
generally for obtaining a sat-planation of the limitation which can
result to bettera systematic way. This issue is discussed in the
next
mixing rules: the special case of size-asymmetric
in the previous section, Eq. (21), that in the inniteit for
EoS/GE models, we have in reality an equation ofergetic) terms.
This is not the case in the zero referenceS/GE models, where, for
the approximate zero referenceHV1 and PSRK models, a FloryHuggins
type contribu-ln(b/bi)) results from the EoS, see Eqs. (23a) and
(23c).ting the excess Gibbs energy from an EoS to that of anvity
coefcient model for deriving a mixing rule at zeroressure, it would
be useful if this FloryHuggins typeial and that of the explicit
activity coefcient models,case of UNIQUAC/UNIFAC (
ixi ln(r/ri)), would can-
s is especially important in the case of gas-containingthe new
group interaction parameters that need tod should not carry this
difference of the two com-in the parameter estimation. Mollerup
[23] in 1986
e two FloryHuggins combinatorials should in princi-ut. However,
as shown by Kontogeorgis and Vlamoshlers and Gmehling [40], they
dont! And moreoverce of the two combinatorials increases with
increas-mmetry, e.g. for mixtures of alkanes. This observation
xplanation why models likes MHV1 and PSRK fail foretric systems.
As Coniglio et al. [41] showed, for theV1, unique group interaction
parameters, e.g. for theteraction cannot be obtained for
representing satisfac-rliquid equilibria for the whole CO2/alkane
series. Insolve this problem, the double combinatorial effectsh and
this is what in essence is done in various EoS/GE
e LCVM, k-MHV1, GCVM and CHV (although some ofls have been
derived semi-empirically, i.e. not starting)). The equations for
the reduced energy parameter foroS/GE models are shown
below:537]:
, C2
C1
gE,FH
RT
)+
i
xii (24a)
-
428 G.M. Kontogeorgis, I.G. Economou / J. of Supercritical
Fluids 55 (2010) 421437
where
C1 =
AV+ 1
AM
C2 =1 AM
AV = 0.62LCVM usk-MHV1
= 1AM
(g
R
Using thof the param
= C1(
gE
RT
where
C1 =1
AMAM = 0.53
For PR aCHV (Or
= 1C
(gE
R
where C* =Based on
parameterent dependcoefcientparametersFAC they shConiglio
etbinatorials(0.640.72port the hysuccessful ris the elimi
This coninnite diluing rules anand a/b mixmodel usedtorial
stemmactivity coe
As discEoS/vdW1ftures, whilvanishes. Tcombinatorapplied onlWe
saw thacubic EoS sTheHuronfrom EoS, ifactivity coemodels likeis
eliminate
Recentlyon the prinhave been
11
bution
toth
e
activi
tyco
efci
ent.
Mden
otes
the
explici
tex
tern
alac
tivi
tyco
efci
entm
odel
use
din
the
EoS/
GE
mix
ing
rule
.FH
den
otes
the
FHco
mbi
nat
oria
lst
emm
ing
from
the
equat
ion
ofst
ate.
ln
,c
ombF
V1
(ext
ra)is
the
additio
nal
co
mbi
nat
oria
l/free
-vol
um
eco
ntr
ibution
stem
min
gfrom
the
explici
tac
tivi
tyco
efci
entm
odel
inse
rted
via
the
EoS/
GEm
ixin
gru
le.
Mod
elln
,com
bF
V1
ln
,c
ombF
V1
(ext
ra)
ln
,r
es1
vdW
1fm
ixin
gru
les,
Eqs.
(4)an
d(5
)ln
( V 1
b1
V2
b2
) +1( V 1
b1
V2
b2
) +a 1
b1RT
ln(1
+(b
1/V1))
(1+
(b2/V2))
+a 2
b2RT
( V 1b2
V2b1
V2 2
+b2V2
)b1
RT
[ a 2 b 2
a1
b1
] 2 ln(1
+b2
V2
)a/
bm
ixin
gru
les,
Eq.(
19)
ln
( V 1
b1
V2
b2
) +1( V 1
b1
V2
b2
) +a 1
b1RT
ln(1
+(b
1/V1))
(1+
(b2/V2))
+a 2
b2RT
( V 1b2
V2b1
V2 2
+b2V2
)Not
hin
g
Huro
nV
idal
,Eq.
(21b
)ln
( V 1
b1
V2
b2
) +1( V 1
b1
V2
b2
) +a 1
b1RT
ln(1
+(b
1/V1))
(1+
(b2/V2))
+a 2
b2RT
( V 1b2
V2b1
V2 2
+b2V2
)
1 AV
ln
( V 2+
b2
V2
) lnM
,com
b1
1 AV
ln
( V 2+
b2
V2
) lnM
,res
1
MHV1,
Eq.(
23a)
ln
( V 1
b1
V2
b2
) +1( V 1
b1
V2
b2
) +a 1
b1RT
ln(1
+(b
1/V1))
(1+
(b2/V2))
+a 2
b2RT
( V 1b2
V2b1
V2 2
+b2V2
)(ln
M
,com
b1
ln
FH 1
)
1 AM
ln
( V 2+
b2
V2
)
1 AM
ln
( V 2+
b2
V2
) lnM
,res
1
k-M
HV1,
Eq.(
24b)
ln
( V 1
b1
V2
b2
) +1( V 1
b1
V2
b2
) +a 1
b1RT
ln(1
+(b
1/V1))
(1+
(b2/V2))
+a 2
b2RT
( V 1b2
V2b1
V2 2
+b2V2
)(ln
M
,com
b1
kln
FH 1
)
1 AM
ln
( V 2+
b2
V2
)
1 AM
ln
( V 2+
b2
V2
) lnM
,res
1
LCVM
,Eq.
(24a
)ln
( V 1
b1
V2
b2
) +1( V 1
b1
V2
b2
) +a 1
b1RT
ln(1
+(b
1/V1))
(1+
(b2/V2))
+a 2
b2RT
( V 1b2
V2b1
V2 2
+b2V2
)([ A
V+
1
AM
] lnM
,com
b1
[ 1
AM
] lnFH 1
) ln
( V 2+
b2
V2
)[ A V
+1
AM
] lnM
,res
1
ln
( V 2+
b2
V2
)CHV,E
q.(2
4d)
ln
( V 1
b1
V2
b2
) +1( V 1
b1
V2
b2
) +a 1
b1RT
ln(1
+(b
1/V1))
(1+
(b2/V2))
+a 2
b2RT
( V 1b2
V2b1
V2 2
+b2V2
)(ln
M
,com
b1
(1
)
ln
FH 1)
1 Cl
n
( V 2+
b2
V2
)
1 Cl
n
( V 2+
b2
V2
) lnM
,res
13, AM = 0.52es =0.36 (C2/C1) = 0.68[42]:
E,
T k g
E,FH
RT
)+
i
xii (24b)
e t-modied PR+original UNIFAC, the optimum valueeter k=0.65 GCVM
(Coniglio et al. [41]):
, C2
C1
gE,FH
RT
)+
i
xii (24c)
C2 =1 AM
, = 0.285nd UNIFAC VLE the ratio proposed is: C2/C1 =0.715bey
and Sandler [43], Zhong and Masuoka [44]):
,
T (1 )g
E,FH
RT
)+
i
xii (24d)
0.6931 and the proposed values for 1 =0.640.7.the above
explanation, it can be understood why the
C2/C1 in LCVM or the k in k-MHV1 must be differ-ing on the
combinatorial term of the explicit activitymodel used. Thus, in the
case of original UNIFAC theseshould be around 0.650.68, while for
modied UNI-ould around 0.3. Similar values have been reported byal.
[41] for both the original and modied UNIFAC com-and by Orbey and
Sandler [43]/Zhong and Masuoka [44]for original UNIFAC). These
similar values further sup-pothesis that the key concept and
explanation of theesults of these EoS/GE models for asymmetric
systemsnation of the double combinatorial effect.clusion can be
further understood by looking at thetion activity coefcients from
SRK using the abovemix-d compare them to HuronVidal, MHV1/PSRK,
vdW1fing rules (M is the explicit external activity coefcientin the
mixing rule, while FH denotes the FH combina-ing from the equation
of state). These innite dilution
fcients are shown in Table 2.ussed previously, the residual term
from cubicdoes not perform satisfactorily for asymmetric mix-
e when using the a/b mixing rule the residual termhus, in the
latter case the cubic EoS has onlyial/free-volume contributions and
can be rigorouslyy to nearly athermal mixtures like mixtures of
alkanes.t the results are satisfactory in this case, indicating
thatatisfactorily capture size effects in athermal
solutions.Vidalmixing rule has only the same combinatorial terman
extra combinatorial term is not used in the externalfcientmodel.
Theapproximate zero referencepressureMHV1 and PSRK have a double
combinatorial whichd in other mixing rules, e.g. LCVM and CHV.,
more approaches which also fully or partially relyciple of
eliminating the double combinatorial term,proposed; these
approaches, e.g. PSRK/Li [45], VTPR Ta
ble
2In
nite
dilution
activi
tyco
efci
ents
calc
ula
ted
from
SRK
using
the
variou
sm
ixin
gru
les.
ln
,c
ombF
Vis
the
com
binat
oria
l/free
-vol
um
eco
ntr
ibution
toth
eac
tivi
tyco
efci
entan
dln
,res
isth
ere
sidual
contr
i
-
G.M. Kontogeorgis, I.G. Economou / J. of Supercritical Fluids 55
(2010) 421437 429
[40] and UMR-PR [46] also result to EoS/GE models suitable
forsize-asymmetric systems. In some of these approaches the
combi-natorial termof the activity coefcient and of the EoS
(FH-term) arebothdroppedout or/anda combining rule other than the
arithmeticmean comb
2.7. Capabi
The discbilities ofsize-asymmbinary andEoS have alblends.
Thenitz and co-term. Mostnal vdW redeveloped fing severalMasuoka
[5sion of the Sused the quan additionand co-voluwith cubic E
In combEoS represeespecially fmixtures. Cequilibria acorrelated
wrections toStill, even wple performthey are comasymmetricgood
perforvolume term
Thus, thtively of the[2,3,60]:
Correlatioplex mixthydrocarbusing tem
Descriptiosame set
Correlatiocible sywaterpe
PredictionLLE and Sdata.
Very combiomolec
To handneeded whplesof thesethe chemicailarities to tworkingat
tnew concepgenerated w
the aforementioned complex types ofmixtures andphasebehavior.In
some sense, we could state that van der Waals-type equationsof
state really can take us far but not further, in the sense there is
aplethora of systemswhere the physical approaches (vdWand the
ouldctorysful aare dl as p
ociat
igher
deles ationequilulatsta
anceal enf theecadtatise calecurest tpartles (
dp
rN ales,tionaolec
oniacoordsum:
) =
rstraturnds. Intas ha
pley tolar i
m thdynfor thble)
) = A isties iasedrtitiowsimining rule is used for the cross
co-volume parameter.
lities and limitations of cubic equations of state
ussion so far has illustrated the tremendous capa-cubic
equations of state, capabilities which includeetric mixtures, even
polymers, as well as polar VLE ofmulticomponent mixtures. Since the
early 90s cubicso been used for modelling of polymer solutions
andrst cubic EoS for polymers was proposed by Praus-workers [48],
based on a modied vdW-type repulsiveother cubic EoS employed to
polymers use the origi-pulsive term. The vdW, SRK and PR EoS have
all beenor polymers and various mixing rules are used includ-EoS/GE
ones [4959]. It is interesting that Zhong and9]haveused
thea/bmixing rule (Eq. (19)) in their exten-RK and PR equations of
state to polymer solutions. Theyadratic mixing rule for the second
virial coefcient asal equation to obtain a mixing rule for both the
energymeparameters. They obtained good results, better thanoS using
the conventional vdW1f mixing rules.
ination with especially the EoS/GE mixing rules, cubicnt an
excellent correlative and even predictive toolor low and high
pressure VLE for a wide range ofertain complex types of phase
behavior, e.g. solidgasnd VLE of complex mixtures like H2Swater can
beell using two adjustable interaction parameters cor-
the classical vdW1f combining rules, Eqs. (4) and (5).ith the
EoS/GE mixing rules, cubic EoS cannot in princi-muchbetter than the
explicit activity coefcientmodelbinedwith. Asmentioned, however,
for athermal size-mixtures satisfactory results are obtained due to
the
mance of the underlying cubic EoS combinatorial/free-, see the
results for the LCVM model for instance.
e most serious problems with cubic EoS, and irrespec-choice of
mixing and combining rules, are expected for
n of vaporliquid and liquidliquid equilibria of com-ures
containing hydrogen bonding compounds withons and similar ones,
e.g. alcohol-alkanes, especiallyperature-independent interaction
parameters.n of VLE and LLE of e.g. methanol/alkanes using the
of interaction parameters.n of liquidliquid equilibria of highly
immis-stems, e.g. wateralkanes, glycolalkanes andruoroalkanes.of
multiphase, multicomponent equilibria, e.g. VLLE,
LE using parameters solely obtained from the binary
plex mixtures such as those containing electrolytes andules.
le such complex mixtures, new advanced models areich go truly
beyond the ideas of van der Waals. Exam-newconcepts
stemfromtheworkofWertheimor froml and lattice theories. Many of
these theories bear sim-he early chemical approach by Dolezalek
[61] who washesametimeasvanderWaals andvanLaar.Using thesets,
families of advanced equations of state have beenhich can describe
better than the vdW-type cubic EoS
like) shsatisfasuccesyears,as wel
3. Ass
3.1. H
Themolecucorrelaphaseare simappliedappearchemicsome othree d
In stem arthe moof inteso themolecu
Q = c
wheremolecuproporable) mHamiltof theas theso that
H(rNpN
ThetempeU depeactions(suchate comthe wamolecuetc.
Frothermowhichensem
A(NVT
whereproperlated bthe pavery febe combined with the chemical
ones for arriving to arepresentation of the phase behavior. Some of
themostdvanced equations of state, developed over the last
30iscussed next together with characteristic
applications,resentation of their capabilities and limitations.
ion theories
order EoS rooted to statistical mechanics
velopment of chemical processes using complexnd the need for
accuratemodels able to provide reliableand prediction of the
thermodynamic properties andibria of the mixtures involved so that
these processesed and optimized, together with signicant advances
intistical mechanics in the 1980s and onwards led to theof higher
order EoS that gradually gained popularity ingineering community.
In this section, we will reviewmost popular higher order EoS
developed in the last
es based on statistical mechanics.tical mechanics, the
properties of a bulk chemical sys-lculated based on the collective
interactions betweenles that make up the system. Almost all of the
systemso chemical engineering follow Boltzmann statistics andition
function (Q) of a system of constant number ofN) in a specic volume
(V) and temperature (T) is [62]:
N drN exp
[H(rNpN)
kBT
](25)
nd pN denote the coordinates and momenta of all NH(rNpN) is the
Hamiltonian of the system and c is ality constant. For a systemofN
identical (indistinguish-ules: c=1/(h3NN ! ) where h is Plancks
constant. Then provides the total energy of the system as a
functioninates and the momenta of the molecules and is givenof the
kinetic energy (K) and the potential energy (U),
i
p2i
(2mi) + U(rN)(26)
term on the right hand side of Eq. (26) provides thee dependence
of the Hamiltonian. The potential energystrongly on the nature
(complexity) of molecular inter-ermolecular potentials range from
primitive potentialsrd sphere, square well, etc.) to potentials of
moder-xity (such as Lennard-Jones, Stockmayer, etc.) and allcomplex
potentials that account for intra- and inter-nteractions, many body
effects (polarizable potentials),
e partition function, one may calculate macroscopicamic
properties using the so-called bridge function,e case of the
constant NVT system (canonical statistical
is:
kBT lnQ (NVT) (27)
the Helmholtz free energy. Additional thermodynamicncluding
pressure, chemical potential, etc. can be calcu-on standard
thermodynamic relations. Unfortunately,n function Q can be
calculated analytically only for aple
systemsandsignicantapproximationsareneeded
-
430 G.M. Kontogeorgis, I.G. Economou / J. of Supercritical
Fluids 55 (2010) 421437
along the way in order that Eq. (27) can lead to engineering
mean-ingful results. The van der Waals and other cubic EoS can be
alsoderived from statistical mechanics by assuming a simple
inter-molecular potential [62].
A powerthe intermoperturbatiosummation
U(rN) = U(0
where U(0)(but yet relabation termchoice of thcations neand, to
a latem underwas proposMcQuarriemodels refements resutheories forby
Barker adler,Weekstheories is g
The formEoS was prPerturbed-Henergy is wrepulsive (r
A = Aig + Ar
The repbased on tAlders Tayby Donohuon Stells p[66]. Finallyory
(APACTmodelled bcal theory,reactions thmaydistingrelevant co
A differeintroducednamic pertu(SAFT) usesand hydrogSAFT,
hydroand the assrial balancewell potenttremendounity and wgases and
htures. In adexplicitly fChain-SAFTvariable ranpolar
intera(electrolyte(29) and sonumber of[3,76,77].
A different approach to model thermodynamic properties ofuids is
based on lattice theory. Early attempts in this directioninclude
the models proposed by Flory [78], and Fowler andGuggenheim [79].
Sanchez and Lacombe formulated a lattice
in ttingdel iids a, knondomtiong beon Ve
ram
ll ofaramulatacterlarle).
etersg. Into exasesticaled fretersily exle, thhe paequixtu
curaclly ij is
ratur
pres
herorangessurial inCT aands syue trongutuafor v5(a)n fr
nt agodelte msupraoligoE ofhere
to 2ive ren wis ne[68]ful and elegant approach to simplify the
complexity oflecular potential is based on the perturbation theory.
Inn theory, the intermolecular potential is written as theof two
terms:
)(rN) + U(1)(rN) (28)
rN) refers to the reference system and is the dominant,tively
easy to calculate term, and U(1)(rN) is the pertur-,which ismore
complex and less accurately known. Thee reference system and the
approximations and simpli-eded for the perturbation term are
practically endlessrge extent, depend on the nature of the chemical
sys-investigation. One of the earliest perturbation theoriesed by
Zwanzig in 1954 [63] followed by the work ofand Katz in 1966 for
high-temperature gases [64]. Bothr to relatively simple systems.
Subsequent advance-lted in much more useful and accurate
perturbationreal uids (gases and liquids) such as those proposednd
Henderson in the period 19681972 and by Chan-andAndersen in1971.
Anexcellent descriptionof theseiven by McQuarrie [62].ulation of
the perturbation theory into an engineeringoposed by Donohue and
Prausnitz [65] based on theard-Chain Theory (PHCT). In PHCT, the
Helmholtz free
ritten as the sum of contributions due to ideal gas (ig),ep) and
attractive (att) interactions, so that:
ep + Aatt (29)
ulsive and attractive contributions were calculatedhe
Carnahan-Starling EoS for hard spheres and thelor series expansion,
respectively. PHCT was extendede and co-workers to polar and
polarizable uids basederturbation theory for dipolar and
quadrupolar uids, in the Associated-Perturbed-Anisotropic-Chain
The-), hydrogen bonding between molecules was explicitlyased on
chemical theory [67,68]. According to chemi-hydrogen bonding is
modelled as a series of chemicalat result in the formation of
oligomers. In this way, oneuishbetweenphysical and chemical forces
and theirntribution to the overall model.nt family of EoS based on
perturbation theory wasin 1990. Rooted to Wertheims rst order
thermody-rbation theory, the Statistical Associating Fluid
Theorythe hard sphere as the reference uid while dispersionen
bonding interactions are perturbations [69,70]. Ingen bonding is
modelled using a square well potentialociation term is calculated
based on a simple mate-. For the case of an innitely deep and
narrow squareial, a molecular chain model is obtained. SAFT
receiveds attention by the academic and industrial commu-as used
for a very wide range of uids, from simpleydrocarbons to ionic
liquids and pharmaceutical mix-dition, it has been extended and
generalized to accountor chain formation in the reference uid
(Perturbed) [71], Lennard-Jones reference term (soft-SAFT) [72],ge
square well attractive interactions (SAFT-VR) [73],ctions (PC-Polar
SAFT) [74], chargecharge interactions-SAFT) [75,76], etc.
Additional terms are added in Eq.
the model becomes signicantly more complex. Areviews of the SAFT
equation of state is now available
theoryaccounthe moto liqutheorynon-rainteracbondinbased
3.2. Pa
In anent pbe calca charmolecumolecuparambondinmodelsome cthe
criobtainparambe easexampfrom t
Thenent mthe acis usuacases ktempe
3.3. Re
Higawideand prindust
APAforcesaqueouideal, dand stThe mknownIn Fig.is showexcellesite
mtwo sisionalchainthe LLbilities(by 1that gbetwevalueAPACThe
isothermalisobaric (NPT) statistical ensemble byfor empty sites
(holes) in the lattice [80]. In this way,s able to account for
pressure effects and is applicablend gases. In the most recent
development of the latticewn as non-random hydrogen bonding (NRHB)
theory,
distribution of molecular sites due to intermoleculars is
calculated explicitly [81]. In addition, hydrogentween rst
neighbouring molecular sites is calculatedytsmans statistics.
eter estimation in higher order EoS
the higher order EoS presented above, pure compo-eters have a
clear physical basis and, in principle, caned from molecular
considerations. Parameters includeistic dispersion energy,
molecular segment size, andsize (expressed in terms of spherical
segments perFor the case of associating components,
additionalinclude energy and entropy (or volume) of
hydrogenpractice, these parameters are calculated by tting
theperimental vapor pressure and saturated liquid (and invapor)
densities from low temperature up to close topoint. The hydrogen
bonding parameters can be alsoom spectroscopic data. For a
homologous series, thevary smoothly with molecular weight and so
they cantrapolated or interpolated to missing components. Fore EoS
parameters for polyethylene may be estimatedrameters of
n-alkanes.ations of state are extended to binary and multicompo-res
using appropriate mixing rules. In order to improvey of
calculations, a binary interaction parameter (kij)tted to
experimental phase equilibrium data. In mosta constant, although
for highly non-ideal mixtures ae-dependent kij is used.
entative applications of higher order EoS
rderequationsof statehavebeenappliedsuccessfully toeof chemical
systemsover abroad rangeof temperaturere conditions. Some
representative examples of directterest are presented here.ccounts
explicitly for weak dispersion forces, polarhydrogen bonding and is
an appropriate model forstems. Waterhydrocarbon mixtures are highly
non-o strong hydrogen bonding between water moleculeshydrophobic
nature of the hydrocarbon molecules.
l solubility is very low but it should be accuratelyarious
petrochemical and environmental applications.
, the liquidliquid equilibria (LLE) of watern-hexaneom 315 to
480K. APACT predictions (kij =0.0) are
inreementwithexperimentaldata. The threeassociating-for water
provides more accurate predictions than theodel [68]. The former
model results in three dimen-molecular structureswhile the two
sitemodel providesmers and is more suitable for 1-alcohols. In Fig.
5(b),waterm-xylene mixture is shown. The mutual solu-are
signicantly higher than in the previous mixture
orders of magnitude) due to electrons in m-xyleneise to strong
polar interactions (dipolequadrupole)ater and m-xylene molecules. A
relatively small kijeded here for an accurate correlation of the
data by. Despite its accuracy, APACT never gained signicant
-
G.M. Kontogeorgis, I.G. Economou / J. of Supercritical Fluids 55
(2010) 421437 431
Fig. 5. Mutual solubilities of (a) watern-hexane and (b)
waterm-xylene. Experimental data (points) and APACT calculations
using a 2-site and 3-site model for water (takenfrom [68]).
popularity, probably due to its relative complexity over
otherEoS.
SAFT has been extensively applied to polymer mixtures. InFig. 6,
SAFT correlation of hydrocarbon Henrys law constant inlow density
polyethylene (LDPE) at low pressure is shown. Usinga temperature
independent kij value, the model provides accuratecorrelation over
a wide temperature range [82].
The polymer phase behavior becomes signicantly more com-plex at
high pressure. In polyethylene technology, polymerizationtakes
place at very high pressures, up to a few kbar, followedby a low
pmost cases,is known tolibria signipolyethylenequilibriummixture.
Poas cloud poa monodispcurve) is obdata. SAFT
1
Hen
ry's
con
stan
t (ba
r)
Fig. 6. Weightimental data (n-butane, n-he
othermal pressurecomposition curves for polydisperse PEethylene
mix-23.15K. Experimental data (points) and SAFT/PC-SAFT
calculations (lines).odelled as a 36-component mixture (taken from
[82]).
mixture [81]. Although thismay be a surprising result
givenC-SAFT uses the hard chain reference uid over the harduid used
by SAFT, it is clear that an appropriate param-tion of a model can
result in very accurate results.ressure separation and product
recovery process. Inthe polymer produced is polydisperse.
Polydispersityaffect the high pressure polymersolvent phase
equi-cantly [81]. In Fig. 7, the high pressure polydispersee (PE;
Mn =56,000 and Mw =99,000)ethylene phaseat 423K is shown. PE is
modelled as a 36-componentlydispersity results in increasing
pressure (referred toint curve) at low polymer compositions. If one
assumeserse polymer, then the shadow point curve (dashedtained,
that deviates signicantly from experimental
correlation is more accurate than PC-SAFT correlation
1000
0000
Fig. 7. Isture at 4PE was m
for thisthan Psphereeteriza1
10
100
300250200150100
Temperature (oC)
fraction Henrys constant of hydrocarbons in LDPE at 1 atm.
Exper-points) and SAFT correlation (lines). From top to bottom:
ethylene,xane, n-octane, benzene and toluene (taken from [76]).
Fig. 8. Experimpoint pressureental data (points) and tPC-PSAFT
correlation (lines) for the bubble-of CHF3[bmim+][PF6] at various
temperatures (taken from [83]).
-
432 G.M. Kontogeorgis, I.G. Economou / J. of Supercritical
Fluids 55 (2010) 421437
0.08
Naproxen - Acetone
100806040200
0.01
Weight fraction of ethanol in solvent
293.15 K
Fig. 9. Solubil 93.15K (right). Experimental data (points) and
NRHB calculations (lines).Dashed lines c the right plot) and
correlations (temperature independent kij value; solidlines; taken
fr
The expresulted inof the phasetative examAn IL consiwide
tempeical reactiostrong dipouiddensityexperimentfor the
bubinteractionment betweexception o
Thermodceutical indsolvents areon for proceceutical
moencounterepolar and hof thermodRecently,
Tseterizationpharmaceuthe solubilicritical uinaproxen,wmarketed
unaproxen toare in goodroform andfor methanoexcellent co
A mixedorder to conaproxen inmore than3accurately pternary
mixprocess sim
Evaluatidynamic mstraightforwracy of thepredictionpendent
kij
15 binary mixtures was performed [86]. Although differ-were
identied for specic mixtures, the overall deviationsenexperimental
data andmodel predictions and correlationsimilar for the two
models.ides phase equilibria, thermodynamic models that accounttly
for hydrogen bonding can provide reliable prediction ofent of
hydrogen bonding as a function of temperature, pres-d composition.
In Fig. 10, experimental data and predictionsreefor ps
arespecfurt
mpli
T inessfhydroweval in
ial chay b
8
03253203153103053002952900.00
0.02
0.04
0.06
Naproxen - Methanol
Naproxen - Ethanol
Naproxen - Chloroform
Sol
ute
mol
e fr
actio
n
Temperature / K
1E-6
1E-5
1E-4
1E-3
Sol
ute
mo
le fr
act
ion
ity of naproxen in various pure solvents (left) and in
ethanolwater mixture at 2orrespond to predictions (kij =0 in the
left plot and tted to single solvent data inom [84]).
licit inclusion of polar interactions into PC-SAFT hasa model
(PC-PSAFT) that provides accurate descriptionbehavior of strongly
polar uid mixtures. A represen-ple refers to an ionic liquid
(IL)solvent mixture [83].sts of a cation and an anion and is liquid
over a veryrature range. ILs are increasingly used today for
chem-ns and separations. In PC-PSAFT, ILs are modelled aslar
molecular uids. Model parameters are tted to liq-datawhile avery
lowvaporpressure is ensured. In Fig. 8,al data and truncated
PC-PSAFT correlation are shownble pressure of CHF3[bmim+][PF6]. A
single binaryparameter (kij =0.031) is tted to the data. The
agree-en experiments and calculations is very good, with thef high
CHF3 concentration.ynamic models are increasingly used by the
pharma-ustry at the product development stage when variousscreened
for new pharmaceutical molecules and laterss development and
optimization.Most of the pharma-lecules are signicantly more
complex than moleculesd in oil and chemical industry, with multiple
functionalydrogen bonding groups. Accurate parameterizationynamic
models for such molecules is far from trivial.ivintzelis et al.
proposed a methodology for the param-of the NRHB equation of state
for several widely usedticals [84]. NRHB was subsequently used to
calculatety of these pharmaceuticals in various liquid and super-d
solvents and mixed solvents. An example refers tohich is a
commonnonsteroidal anti-inammatorydrug
nder various trade names. In Fig. 9(left), the solubility
ofvarious polar solvents is presented. NRHB predictionsagreement
with experimental data for ethanol, chlo-acetone and in fair
agreement with experimental data
LLE ofencesbetwewere s
Besexplicithe extsure anfrom thsentedmodelrole ofcussed
3.4. Si
SAFbe succuids,cals. HchemicessentEoS m
0.
1.
onl. Using a temperature independent binary
interaction,rrelation is obtained.solvent is often used in
pharmaceutical industry in
ntrol drug solubility. In Fig. 9(right), the solubility ofthe
ethanolwater solvent at 293.15K that varies byorders ofmagnitude
frompurewater to pure ethanol isredicted. No parameter adjustment
was made for theture and so NRHB can be safely used for the
relevantulation [84].on of the relative accuracy of the various
thermo-odels and selection of one over another is not aard task.
Recently, an extensive evaluation of the accu-simplied PC-SAFT
(sPC-SAFT) and of NRHB for the
(kij =0.0) and correlation (using a temperature inde-) of the
VLE of 104 binary mixtures [85] and of the
0.00.0
0.2
0.4
0.6
mo
nom
er fr
act
i
Fig. 10. Propaperatures (takpredictions (liassociating theories
(sPC-SAFT, NRHB and CPA) are pre-ropanoln-heptane in the range 1555
C [87]. All threein very good agreement with spectroscopic data.
Thetroscopic data in parameter estimation in EoS is dis-
her in the next section devoted to the CPA EoS.
ed association theories: the CPA equation of state
its numerous variations and NRHB have been proven toulmodels
formany applications ranging fromnon-polarogen bonding ones up to
polymers and pharmaceuti-er, for many practical applications in the
oil & gas anddustries, a simpler approach which retains some of
thearacteristics of association models but yet rely on cubice
useful. In this direction the Cubic-Plus-Association
Exp. data 15 oC
Exp. data 35 oC
Exp. data 55 oC CPA NRHB sPC-SAFT0.2 0.4 0.6 0.8 1.0
mole fraction of propanol
nol monomer fraction in propanoln-heptane mixture at three
tem-en from [87]). Experimental data (points), NRHB, CPA and
sPC-SAFTnes).
-
G.M. Kontogeorgis, I.G. Economou / J. of Supercritical Fluids 55
(2010) 421437 433
(CPA) EoS has been developed. CPA is a simplied version of
SAFT,which has been shown to be particularly useful for
applicationsin the petroleum and chemical industries. In CPA, the
physicalterm of SAFT is replaced by SRK, PR or another cubic EoS.
TheCPA modelrecent reviepublicationchemicals [from 1996used the
PRtions, the vthe associaneeded forassociatingmatic
hydrCO2water.
While mglycolshydorganic acigases (CO2,alkanolamicompoundshad
additio(waterNaChave beenasphaltenestions mentiof a characculations
al[90,11211WaalsPlat
3.4.1. The pWe pres
tioned invethe limitatiit can be stcapable of
ppredictionshydrocarbodata. Excephighly immcarbons, asThe CPA
Eoequallywelimum of thaccording tfrom 1983zouk et al.needed in
tparameterwater or gparametervation (ind[116], howeing similarkij
from wwaterbenzinteraction
Associat(number ofbeen establperforms bemum schemTheefforts
[monomer f
280 290 300 310 320 330-5
-4
-3
0.01
Exp. data Derawi et al. (2002) Exp. data Razzouk et al.
(2010)
CPA, kij=0.059
Hexane in MEG rich phase
Temperature / K
MEG in HC rich phase
MEGhexane LLE. Experimental data (points) and CPA correlation
(lines,9). The experimental data are from Derawi et al. [128] and
the recent datazzouk et al. [126].
e selection led to conclusions of mostly qualitative
nature,hough satisfactory results are obtained in many cases (see).
Suulatemee is pto ded pic da
y wahe sle accor mbe phaspropumb
ines,harm
280 300 320 340 360 380 400 420 440 460 480-7
-6
-5
-4
-3
01
.1
Exp. data Tsonopoulos et al. (1983) Exp. data Razzouk et al.
(2010)
CPA, k =0.0355
CPA, k =0.0
Water in HC rich phase
Temperature / K
Hexane in water rich phase
Waterhexane LLE. Experimental data (points) and CPA correlation
(lines,55). The experimental data are from Tsonopoulos and Wilson
[127] and theata from Razzouk et al. [126].is extensively described
in literature including severalws [8789] and two books [2,3], while
several recents present specic applications related to oil &
gas and9092]. The CPA version by Kontogeorgis et al. [93]is based
on SRK, while other researchers [94,95] have
or the ESD [96] EoS. In the majority of the applica-dW1f mixing
rules are used (Eqs. (4) and (5)), whiletion term needs no mixing
rules. Combining rules arecross-associating mixtures, e.g. waterMEG
or induced(solvating)mixtures likewaterBTEX (BTEX= thearo-ocarbons;
benzenetolueneethylbenzenexylene) or
ost original publications focused on wateralcohols orrocarbons
VLLE, several recent investigations includeds [9799], uorocarbons
[100], mixtures with acidH2S) [101,102], glycolethers [103], amines
[104] and
nes [105], and complex multifunctional polyphenolic[106,107].
The PR-CPA from Wu and Prausnitz [94]
nal electrostatic terms for applications to
electrolyteslmethane), while recently more electrolyte CPA
EoSdeveloped [108,109]. Finally, an extension of CPA tohas been
reported [110]. In addition to the applica-
oned above, oil applications include the developmentterization
method for CPA [111] and gas hydrate cal-so in presence of
inhibitors like methanol and glycols5]. In the latter case, CPA is
combined with the van derteeuw model.
erformance of CPA in briefent now our conclusions on some of the
aforemen-stigations with CPA illustrating both the successes
andons/challenges of the model at its current state. In brief,ated
that CPA is a successful thermodynamic model,roviding satisfactory
multiphase (VLLE, LLE, SLE, etc.)for mixtures containing water,
alcohols or glycols andns based solely on binary parameters tted to
binarytionally good correlations are obtained for the LLE ofiscible
systems like glycols and water with hydro-shown for two typical
mixtures in Figs. 11 and 12.S can correlate the solubilities in
both liquid phasesl over extensive temperature ranges, except for
themin-e hydrocarbon solubility in water which is observedo the
experimental data of Tsonopoulos and Wilson[127] but not in the
recent measurements of Raz-
[126]. In the typical case, one interaction parameter ishe
calculations with CPA (the kij in the cross-energyin Eq. (5)) but
for solvating mixtures for examplelycols with BTEX compounds we
need an additionalto account for the enhanced interactions due to
sol-uced association) effects. It has been recently shownver, that
the kij can be obtained from a mixture hav-physical interactions as
the solvating system, e.g. theaterhexane or MEGhexane can be used
to modelene and MEGbenzene, respectively, leaving only oneparameter
to t (the cross-association volume).ionmodels like CPA require that
the association schemesites and location) is known. Over the years
it has
ished that, in the CPA model, the two-site (2B) schemetter for
methanol (and other alcohols), while the opti-e for water and
glycols is a four-site (4C) one [2,3].
86,87,117119] inusing spectroscopy results especiallyraction
data towards better parameter estimation and
1E
1E
1E
mo
le fr
act
ion
Fig. 11.kij =0.05from Ra
schemeven tFig. 10ior calc2B
schschemmatureimprovtroscop(mostlcases ttionab[120] fcannot
CPAacetic,large nnolamsmall p
1E
1E
1E
1E
1E
0.
0
mo
le fr
actio
n
Fig. 12.kij =0.03recent dch investigations provided additional
(to phase behav-ions) assurance that the use of 4C scheme for water
andfor most alcohols is correct, although a three-site (3B)ossibly
a better choice for methanol. It is perhaps pre-etermine whether
monomer fraction data will result toarameters for association
models. Currently such spec-ta are available for only a few
compounds andmixtures
ter, alcohols and some glycolethers). Moreover, in
somepectroscopic data for monomer fractions are of ques-uracy. For
example, the monomer fraction data of Luckethanol and ethanol
coincide, a surprising trend whichredicted by any of the
association models [87].been also applied to amines, small organic
acids (formic,ionic) and recently also heavy organic acids and aer
of multifunctional compounds (glycolethers, alka-polar aromatic
acids, polyphenolic compounds andaceuticals and) [9799,103107]. The
results are over-
-
434 G.M. Kontogeorgis, I.G. Economou / J. of Supercritical
Fluids 55 (2010) 421437
all satisfactory also for these chemicals although most of
theseapplications have been so far limited to binary mixtures (VLE,
LLEand SLE). An interaction parameter is almost always needed
par-ticularly for aqueous mixtures for which this parameter is
oftennegative, tyan inabilityeffects whi1). Still thetypes of
phature rangescan be reprrange of 250combiningone-site (1Aorganic
acidinvestigatedthe 2B and 32B scheme
The 1A,from the noto the SAFTtional compKontogeorgof heavy
glyLLE, excesswith a 6D sthe two hypler 4C scheOn theothescheme
peralkanolami
Aqueouschallengingcult caseformic acidvdW1f mixcannot desctation
of reAs Muro-SuHuronVidafor this pro
Some cpresented,HuronVidaand glycolcontainingwatermetclearly
supresults arefewer adjus
Using arecently shpolar chemtems for wh
The sucpolar/hydrosical cubic Eabove cancombined wor
limitatiowill hereaftimportant cstudies.
3.4.1.1. Parhydrogen b
established, for others especially multifunctional ones it has
not.The choice may be quite difcult, partly due to lack of
reliablevapor pressure and liquid density data over extensive
temperatureranges. Moreover, especially when ve parameters are
tted, they
t betimizpreFord an
investermd hasglycydro
alcohusedbe dless
. Theced io alkaterrs inompountomplcaseeter,ed btions(all p.
Thethatonene forase,entinn extly aethanVLEmetmostions
. Thecountionractiighlytingehyapprot bd LLeternaleudor acelopms)
maion sshinga m
eringpically around 0.1 to 0.2. This result may indicateto
account fully satisfactorily for the cross-association
ch may be due to the combining rule used (Elliott,
CR-correlation is, in most cases, satisfactory for differentse
behavior (VLE and LLE) and over extensive tempera-. For instance,
watermethanol and waterethanol VLEesented very well with a single
kij over a temperatureC. It is not always possible to determine a
prioriwhich
rule performs best. In terms of association schemes, the)
(resulting to dimerization) is found to be the best fors, while the
4C is the best choice for the alkanolamines(MEA, DEA and MDEA). For
amines and glycolethers,B schemes perform equally good and for
simplicity the
is adopted.2B, 3B and 4C schemes are well-known and
originatemenclature and recommendations made in connectionmodel
byHuang andRadosz [70]. For
certainmultifunc-oundsnewschemesmaybemoreappropriate. Breil andis
[121] showed that many thermodynamic propertiescols like
triethylene glycol can be described well (VLE,enthalpies and innite
dilution activity coefcients)cheme (six equal sites for the four
oxygen atoms anddrogen atoms) although the differences from the
sim-me are not signicant, especially for VLE and LLE alone.r hand,
as shownbyAvlundet al. [122], the four-site (4C)forms as well as
more complex association schemes fornes.mixtures with other
associating compounds are oftento model, as mentioned previously. A
particularly dif-
is represented by the mixtures acetic acidwater andwater. In
these cases, the original form of CPA with theing rules and one (or
even two) interaction parametersribe satisfactorily VLE including
the accurate represen-lative volatilities over extensive
temperature ranges.n et al. [98] showed, a combination of CPA with
thelmixing rules is a successful but empirical way to solve
blem at the cost of additional adjustable parameters.omparisons
of CPA with other models have beene.g. SRK with classical [3,88],
MHV2 [90] andl [3,29]mixing rules for
alcohols/glycolshydrocarbonswaterhydrocarbons and with PC-SAFT for
methanol-mixtures [123] (including multicomponent LLE for
hanolhydrocarbons). Compared to SRK/vdW1f, CPA iserior for
associating mixtures, while overall similarobtained against
SRK/HuronVidal (although CPA usestable parameters) and PC-SAFT for
the systems studied.rather simple characterization method, CPA has
beenown [111] to predict satisfactorily the solubilities oficals
(water, methanol, glycols) in the few reservoir sys-ich such data
are available.cess of association models like CPA for manygen
bonding systems especially as compared to clas-oS described in
Section 2 which has been summarizedbe largely attributed to
Wertheims association termith careful parameterization. Some of the
difculties
ns of the approach have already been presented but weer continue
along these lines and outline some of thehallenging topics which
need to be addressed in future
ameterization-pure compounds. While for a number ofonding
compounds the association scheme has been
may nothe op(vaportainty.requireunderfor demethowater,those
hheavynot beshouldmay be
3.4.1.2enhanpared twith wor estepolar cfor accrule)
ctionedparamprovidpredicibilityalone)[3,91]ticompthe caslatter
crepresover aexpliciand mfor thewater,be, insiderat
3.4.1.3not acinteracan intetreat hassociaacetonThesemay nVLE
anparamadditioThe pstions foA devemodelextensestabliin
suchconsidunique in the sense that several sets are obtained
fromation which can represent pure compound propertiesssures,
liquid densities) within experimental uncer-choosing the best
parameter set additional data ared we nd that LLE data for the
associating compoundtigation and aliphatic hydrocarbons is a
successful wayining the best parameter set. This
parameterizationits merits when such data are indeed available,
e.g. for
ols, small alcohols, glycolethers and alkanolamines. Forgen
bonding compounds like amines, small acids andols which are
miscible with alkanes, this approach can-and the establishment of
the optimum parameter set
etermined based on VLE data on which the
parameterssensitive.
importance of solvation. There is evidence that there
arenteractions resulting to high solubilities (higher com-anes)
inmixtures of aromatic andolenichydrocarbons, glycols,
alkanolamines and acids, mixtures of etherswater or in mixtures
with acid gases (CO2, H2S) and
ounds. But is the approach we have used so far in CPAing for
these enhanced interactions (the modied CR-1etely convincing?
Itworks in practice for the aforemen-s, sometimes at a cost of a
second adjustable interactionbut additional evidence is necessary
and this could bey checking other properties or even
multicomponentwhere there isnoadditional
interactionparameterex-arameters should be determined from binary
systemsstudy with multicomponent systems has illustratedfor the
waterglycolsBTEX the solvation is for mul-t LLE as important as for
binary LLE. This is not alwaysacid
gaswateralcohol/glycolshydrocarbons. In the
accounting for the CO2water solvation is important forg the
water solubility in CO2 for CO2watermethanetensive pressure range.
However, the signicance ofccounting of the solvation for acid gases
with glycolsol is, for multicomponent calculations, small.
Overall,
-based multicomponent systems containing acid gases,hanol or
glycols and hydrocarbons, the solvation couldcases, safely ignored
(even though the solvation con-improve the representation of the
individual binaries).
role of polarity (and weaker forces). The CPA EoS doest
explicitly for polar (or quadrupolar) effects and suchs need to be
accounted for implicitly, e.g. via includingon parameter kij.
Another semi-empirical approach is topolar (non-hydrogen bonding)
compounds as pseudo-ones and such approach is shown to work well
fordrocarbons VLE and sulfolanehydrocarbons LLE [3].oaches (use of
kij or the concept of pseudo-association)e always satisfactory. For
example, acetone/hexaneE cannot be described well using a single
interactionover the whole temperature range. PC-SAFT, withoutpolar
terms, suffers from the same limitation [3,124].-association
approach is also shown to have limita-tone-containingmixtures, e.g.
acetonechloroform [3].ent towards a polar CPA (along the lines of
polar SAFTy be a way to accommodate for these problems. Thishould,
however, be done in a consistent way especiallywhich values for the
dipole moment should be usedodel (vacuum ones or from the liquid
phase) and byboth polarity and association for hydrogen bonding
-
G.M. Kontogeorgis, I.G. Economou / J. of Supercritical Fluids 55
(2010) 421437 435
compounds. A polar CPA model should be further tested for
binaryand multicomponent mixtures containing polar, associating
andinert compounds.
3.4.1.4. Limpounds ormat least frolimitationsgen bondinTEG and
glyaccountingmixtures w
4. Conclus
We haveexcess Gibbfrom cubicing the caphave showeform
similatively a correffects, domasymmetricmers. On ththe use of
thwhich resuterm, whichBest resultstype term imodel, prefaccurate
reuse one of tCHV, VTPR oproblem, wtic for sompressure mdo have
othcoefcient,not satisfacliquidliquimixtures arsize-asymma
problem.address theof multicomespecially tbasic
princiapplicationbilities of thwhich indic
Acknowled
The authuseful discushown in Fthe calculatalso
gratefudiscussions
References
[1] J.D. VanDisserta
[2] G.M. Kontogeorgis, R. Gani, Computer-Aided Property
Estimation for Processand Product Design, Elsevier, 2004.
[3] G.M. Kontogeorgis, G.K. Folas, Thermodynamic Models for
Industrial Appli-cations. From Classical and Advanced Mixing Rules
to Association Theories,John Wiley & Sons, New York, 2010..S.
Webria, A. Orbeate an98.
I. Saniley &. Soavate, C.-Y. Peginee.J. Hurpor-l(1979.
Hild
ons In. Wilsee En713
.S. Abrew exstems. Renoons fo.S. Elbiction074
.M. Koodel fEngin. Bondiley &a. Fre
activ861Thorlelastuilibr. Lindthe
726.M. Koos, Abic eq
hemicA. Sachermeering.M. Koefcions o743
Molleodels.L. Miuatio.L. Miate, FlDahl
nifac-b.S.H. Wres ustrapoginee
.S.H. Wons of.K. Foquid-Lonentuid PhHoldsed
oDahl,uatiouilibr
esearc.S. KaferenC. Vouquid-edustritations of the Wertheim
theory. Finally, there are com-ixtures forwhich theWertheim theory
has limitations,
m a theoretical point of view. One of the signicantis the
presence of intramolecular association (hydro-g within the
molecule) present in e.g. heavy glycols likecolethers. It is of
interest to evaluate the signicance offor the intramolecular
association in modelling of suchith SAFT-type theories. Such work
is in progress [125].
ions
presented a methodological approach based on thes energy and
activity coefcient expressions derivedequations of state (EoS) for
analyzing and understand-abilities and limitations of these
classical models. Wed that cubic