Thermodynamic models for process and product

design. Status and Future trends

Georgios M. Kontogeorgis
Center for Energy Resources Engineering (CERE) &
KT-consortium
Department of Chemical and Biochemical Engineering
Technical University of Denmark
 Outline - Messages
• Industrial needs as expressed by industry ! Similarities and Differences
- EFCE
- Older studies
- Ralf & Dohrn – Zeck articles – chemical industry’s view

• Which thermodynamic model(s) to choose ?

- Many models ?
- How many really different models – approaches ?
- How different are the various modeling approaches ?
- What is “physics” and what is just “parameter estimation” ?

• Handling real Problems and asking the right questions

- Beyond the opinion of model developers
- Beyond just phase equilibria
- Applications: polymers and pharmaceuticals

• Conclusions and future outlook

 EFCE Working Party on Thermodynamics and
transport properties (http://www.wp-ttp.dk )
• EFCE: Founded: 1953 (AIChE: 1908)
 Organised in 19 Working Parties (WP) and 6 Sections
 ECCE, Excellence Awards http://www.efce.info/
• Objectives of the WP:
– Promote measurements, correlations, predictions, simulations and
educational activities
– Offer a forum to Chemical Engineers for exchange of ideas
– Help identify chemical systems and conditions that are important for
novel chemical processes and for which data and / or correlations are
non-existent or inaccurate
• Creation Frankfurt, 12 November 2008
• President (then): Professor Ioannis Economou (‘Demokritos’, Greece)
• Vice-presidents: Prof.Dr. Ralf Dohrn (Bayer), Prof. G.M. Kontogeorgis (DTU)
• Richard Sass, Secretary (Dechema)
• Task groups:
– Educational (P. Alhstrom) and
– Industrial perspectives (E. Hendriks)
Industrial requirements for thermodynamics and
transport properties
• Project carried out by the Working Party on Thermodynamics
and Transport Properties (http://www.wp-ttp.dk ) of EFCE
• Purpose: (i) Review situation in European industries with
respect to thermodynamics and transport properties, (ii)
Identify limitations in current approaches, (iii) Provide specific
recommendations on R&D work
• Questionnaire sent to 59 companies from 16 countries
(mostly in Europe) but also USA, Japan, Thailand, Brazil and
South Africa – 28 replies (47% response rate)
• Results analyzed and discussed in a forum of industrial (Shell,
Bayer, IFP) and academic colleagues (DTU, UAE, Slovenia, UK)
• Manuscript published in 2010
Project carried out by the Task Group on Industrial Perspectives
under the Working Party on Thermodynamics and Transport
Properties (http://www.wp-ttp.dk ) of EFCE
 What do industries want ?
• Fewer models (single?) which can do more
• Emphasis: electrolytes, polymers, CCS,
biomolecules, ”classical forgotten” systems...
• Wide range of properties and conditions
• Both thermodynamic (equilibrium and other) and
transport properties needed
• More and better experimental data
• Standards and immediate use of models by
industries
• Better education in thermodynamics of end-users
• Classical and advanced models should be
developed equally
 Both UNIFAC and SAFT !
Gmehling, 2009,
J.Chem.Thermodynamics,41:73
From Fu and Sandler (1995)
1
 What can be done next ? – Bridge the gap
between academia and industry
• Education
– Define some minimum knowledge required for Chem Eng, and
propose ways to achieve it
– Promote the employment of young experts (PhD) by chemical
industries
– Propose short courses for professionals to the basics of
molecular thermodynamics & more
• Continue and improve communication
– National WP meetings & co-ordination with EFCE
– European workshop (ProcessNet in Würzburg Feb 2010) &
coming ones (Lyon, 2012, March 19-20)
• More discussion with industry – more/encourage participation of
industrial experts in Conferences on Thermodynamics and
Transport properties (do we have too many conferences ?)
 Articles written by industry [1986-2001]
• Tsonopoulos, C.; Heidman, J. L. High-pressure vapour-liquid equilibria with cubic
equations of state Fluid Phase Equilib. 1986, 29, 391-414. (EXXON)

• Zeck, S. Thermodynamics in process development in the chemical industry -

importance, benefits, current state and future development. Fluid Phase Equilib.
1991, 70, 125-140. (BASF)

• Zeck, S.; Wolf, D. Requirements of thermodynamic data in the chemical industry.

Fluid Phase Equilib. 1993, 82, 27-38. (BASF)

• Mathias, P. M.; Klotz, H. C. Take a Closer Look at Thermodynamic Property Models.

Chem. Eng. Prog. 1994, 90, 67-75. (Air Products)

• Carlson, E. C. Don’t Gamble with Physical Properties for Simulations. Chem. Eng.
Prog. 1996, 92, 35 – 46. (ASPEN)

• Bokis, C. P.; Orbey, H.; Chen, C. C. Properly Model Polymer Processes. Chem. Eng.
Prog. 1999, 95, 39 – 52. (ASPEN)

• Agarwal, R.; Li, Y.-K.; Santollani, O.; Satyro, M. A.; Vieler, A. Uncovering the Realities
of Simulation. Chem. Eng. Prog. 2001, 97, 64 – 72. (Virtual Materials)
 Articles written by industry [2002-2010]
• Dohrn, R.; Pfohl, O. Thermo-physical properties - Industrial directions. Fluid Phase Equilib. 2002, 194-197,
15-29. (Bayer)

• Chen, C. C.; Mathias, P. M. Applied Thermodynamics for Process Modeling. AIChE J. 2002, 48, 194 - 200.
(ASPEN)
• Kolar, P.; Shen, J.-W.; Tsuboi, A.; Ishikawa, T. Solvent section for pharmaceuticals. Fluid Phase Equilib. 2002,
194, 771 – 782. (Mitsubishi)

• Gupta, S.; Olson, J. D. Industrial Needs in Physical Properties. Ind. Eng. Chem. Res. 2003, 42, 6359 – 6374.
(Dow Chemicals)
• de Hemptinne, J. C.; Mougin, P.; Barreau, A.; Ruffine, L.; Tamouza, S.; Inchekel, R. Application to Petroleum
Engineering of Statistical Thermodynamics – Based Equations of State. Oil and gas Science and Technology
– Rev. IFP 2006, 61, 363-386. (IFP)

• Chen, C. C.; Crafts, P. A. Correlation and prediction of drug molecule solubility in mixed solvent systems
with the Nonrandom Two-Liquid Segment Activity Coefficient (NRTL-SAC) model Ind. Eng. Chem. Res.
2006, 45, 4816 – 4824. (ASPEN-Astra Zeneca)

• O’Connell, J. P.; Gani, R.; Mathias, P. M.; Maurer, G.; Olson, J. D.; Crafts, P. A. Thermodynamic property
modeling for chemical process and product engineering: some perspectives. Ind. Eng. Chem. Res. 2009, 48,
4619 – 4637.
(ASPEN – Dow – Astra Zeneca)
• Lin, Y.; ten Kate, A.; Mooijer, M.; Delgado, J.; Fosbøl, P. L.; Thomsen, K. Comparison of activity coefficient
models for electrolyte systems, AIChE J. 2010, 56 (5), 1334-1351.
(Shell – Akzo – SQM)
 Century of EoS & Activity Models and DSM’s role
Equation of States Activity Models
PC-SAFT
g-SAFT COSMO3D Year of
introduction
VTPR
S-PCSAFT COSMOSAC
e-PCSAFT 2000

PSRK, SWP IAWPS COSMO-RS COSMOSPACE 1990

Mathias Koningsveld-
e-NRTL
Huron-Vidal PHCS, SAFT Unifac, LiFAC Kleintjens 1975
Helgeson
Mixing rules Uniquac

Sanchez- ASOG Pitzer/Bromley

Peng-Robinson MBWR NRTL
Lacombe Hansen Robinson/Stokes 1950
Flory-Huggins
Redlich-Kwon BWR Scatchard- Wilson Debeye-Huckel
Hildebrand 1925

Van Laar
vdWaals Virial Meyer
Margules

Cubic Empirical Statistical Predictive Solvent Electrolyte

DSM
Involved Used @ DSM

Slide courtesy of Dr. Gerard Krooshof, DSM

 Cubic EoS are still popular and widely used
models especially in the Oil & Gas Industry
Costa Tsonopoulos & Jack Heidman
(Exxon,1986)
”What about complex models ? Cubic EoS
are simple, reliable and allow for direct
incorporation of critical conditions.
We, in the petroleum industry, continue
to find that such simple EoS are very useful
high-pressure VLE models, and we found as yet no
reason to use complex non-cubic equations of state”

• J. Heidman (IVC-SEP discussion meeting, 1998):

”Cubic EoS are here to stay.
SAFT EoS is referred as the beast!”
 Dangers with process simulators
E.g. Flowsheet for styrene/ethylbenzene separation using
SRK
(R. Dohrn, O. Pfohl, Fluid Phase Equilibria, 2002, 194-197, 15-29)

Comparison of
Three
Process
Simulators

Why this difference

51 kmol/h EB in the design ?
47.8 kmol/h S
EB,
2.9-8.55 kmol/h
The Devil is in the Details
 Simple good models but with some
problems
• Parameters ! How to estimate them ?
• Mixtures ! How to extend the models ?
• Two wrongs – can they make one right ?
• Gibbs: ”The whole is simpler than
RT
the sum of its parts” 16 P=
MD data
V −b
12 C-S
VdW

G = H − TS
E E E Zhs
8

RT a
P= − 2 0

V −b V 0 0.1 0.2
y
0.3 0.4 0.5
 One way to apply cubic EoS to mixtures
The van der Waals one fluid (vdW1f) mixing rules

RT a
The Details !! P= −
V − b V (V + b )
a = ∑∑ xi x j aij b = ∑∑ xi x j bij
i j i j

aij = ai a j (1 − k ij ) B = b − a / RT
bi + b j B = ∑ ∑ xi x j Bij
bij = (1 − lij ) i j
2
 n-butane in alkane solvents at 373 K with the Peng-
Robinson equation of state
Same EoS a = ∑∑ xi x j aij
i j

Same pure
1,2
parameters b = ∑ xi bi
Infinite Dilution Activity Coefficient of

1
exp. data
Modified Unifac
Butane

PR EoS (vdw1f mix. rule)

n
a a
= ∑ xi i
0,8 PR EoS (a/b mix. rule)

b i =1 bi

0,6

0,4
20 22 24 26 28 30 32 34 36
Carbon number of Alkane

Kontogeorgis and Folas, 2010. Thermodynamic models for industrial applications. Wiley
 The van der Waals (vdW) EoS for polymers – why not ?
RT a
P= − 2 vdW1f mixing
V −b V rules

 ϕ fvi ϕ fvi   Vi
ln γ i = ln γ icomb-fv + ln γ ires =  ln +1−  +  (δ i − δ j )2 ϕ 2j 
 xi xi   RT 
x (V − b ) ai
ϕ = i i i
tv
δi =
∑ x j (V j − b j )
i
Vi
j
Flory-Huggins (1941, 1942)

 Φi   Φi  Vi
ln γ i = ln  + 1 −   + (δ i − δ j )2 Φ 2j
 xi   xi  RT
Entropic-FV xiV f i
 ϕi fv   ϕi fv  ϕi =
fv
ln γ i
comb − fv
= ln  + 1 −   ∑x V j fj
 xi   xi  j

Elbro et al., 1990 V f i = Vi − Vw i

LLE with vdW for polymer solutions and blends
(Kontogeorgis et al., 1994; Harismiadis et al., 1994)

We thought:
A provocative statement !
We were the first !
Influence of the error in the separation factors
in the stages of distillation
(R. Dohrn, O. Pfohl, Fluid Phase Equilibria, 2002, 194-197, 15-29)
100%

alpha=1.05
Error in minimum number of stages 80% alpha=1.1
alpha=1.2

60%

40%

20%

0%

-20%

-40%
-5% -4% -3% -2% -1% 0% 1% 2% 3% 4% 5%
Error in separation factor α=(y1/x1)/(y2/x2)
Challenges for thermodynamic models in the
chemical industry (S. Zeck, 1991; FPE, 70: 125)
Challenges for thermodynamic models in the chemical industry
(S. Zeck, 1991; FPE, 70: 125) – cont’d
 Fair Comparisons Please !!
Peng-Robinson
SAFT
“Most past work focused on extensions
of cubic vdW-type EoS … This has occurred
despite the now famous recommendation of
Henderson. No matter how sophisticated a
mixing rule, the use of van der Waals-type
cubic equations of state force their inherent
limitations on the users. These are the ability
to reasonably predict only the vapour pressure
of a select series of components and
only an approximate modelling of the effect
of liquid density and compressibility.
Van der Waals-type cubic equations are
unable to accurately model other liquid
phase properties, for example, enthalpy and
heat capacity also phase equilibria at high
pressures, particularly the mixture critical locus.”

S.Gupta and J.D. Olson, 2003.

Industrial Needs in Physical Properties
Ind.Eng.Chem.Res., 42: 6359

H.Kahl and S.Enders, 2002,

Phys.Chem.Chem.Phys., 4: 931
 Which model to select ?
C.C.Chen and P.M.Mathias (ASPEN)
AIChE J., 2002, 48: 194
Which model to select ? One ”industrial approach” – use of decision trees

Carlsson, 1996 (ASPEN) P < 10 bar

NRTL, UNIQUAC, WILSON and their
variances, UNIFAC LLE, UNIFAC and
Chem.Eng.Progress, 35 its extensions

Non-electrolyte
P?

Schwartentruber – Renon,
Polar P > 10 bar PR or RKS with WS,
E? PR or RKS with MHV2, PSRK

Electrolyte Electrolyte NRTL

or Pitzer

PL?
Real Peng-Robinson,
Redlich-Kwong-Soave
Lee-Kesler-Plocker
All Nonpolar
R? Chao-Seader, Grayson-
Streed or Braun K-10

Pseudo & Real

P?

Vacuum
Braun K-10 or Ideal
Symbols:

Real or Pseudo- Pressure

PL? Polarity R? components E? Electrolyte P?
 Four (?) types of Equations of State

Virial EoS

(Empirical)
specialized Molecular
multiparameter EoS (non-cubic)
Cubic EoS EoS e.g.
e.g. SRK, PR e.g. BWR SAFT, CPA,
classical/EoS-GE Span-Wagner Lattice
mixing rules (CO2)

Cubic vs. Non-Cubic ? (Semi)empirical vs. Theoretical/molecular ?

 Predictive activity coefficient models

Regular solution theory

 Vi
ln γ i =  (δ i − δ j ) ϕ j 
2 2

 RT 

UNIFAC

ln γ i = ln γ i
comb
+ ln γ i
res
The group contribution concept
 Simple principle: We
assume that
molecules are
divided into groups
 Properties are then
calculated by additive
rules

Gmehling, 2009,
J.Chem.Thermodynamics,41:73
Gmehling, 2009,
J.Chem.Thermodynamics,41:731

www.unifac.org
www.ddbst.de

Only 15% DDB

published!!
 Advanced mixing rules for cubic EoS
• Can we combine the advantages of cubic equations of state
and activity coefficient models ?
• ... And have a model (Equation of State) for both high
pressures and polar fluids ?
• Combination of the above  Cubic EoS with activity
coefficient models in mixing rules (Huron-Vidal, MHV2, Wong-
Sandler, PSRK, LCVM,...)
 True value of EoS/GE models

The Huron-Vidal mixing rule (1979)

ai g E ,∞
a
= ∑ xi − SRK
b i bi ln 2

At infinite pressure the comb/free-volume

(size) term disappears
Preferred Act.Coef.Model:
- NRTL
- Residual term of UNIQUAC
or UNIFAC
 This is what ”better models”
should beat ! a = f(g ) E

a = ∑∑ xi x j aij
i j

b = ∑∑ xi x j bij
i j

Target-1: Complex mixtures also at high pressures

Target-2: Use of the same parameters of existing LC activity coefficient models – predictive models
 UNIFAC is no good for complex molecules
Partition coefficient of Penicillin G over organic
and aqueous phases using UNIFAC and experiments
Stockar and van der Wiele, 1997
Thermodynamics in
Biochemical Engineering

J. Biotechnology, 59: 25

10000 tn/year (2000)

 Very difficult problems (also in oil industry):
multicomponent phase equilibria, especially LLE and
VLE/LLE with one parameter set

Plots kindly provided by Henk Meijer and

Eric Hendriks, Shell (Amsterdam)

Photo: Petrobras
The Physical Picture of Association

Acetic Acid (dimers)

Alcohols, phenols (oligomers)

Water – 3D structure
 New generation of thermodynamic models
include the concept of association
• Different variations
- Chemical theories (Anderko, PHCT,...)
- Lattice-Fluid theories (NRHB,...) Z, A, P
- Perturbation theories (SAFT, CPA,...)
• Most often in form of non-cubic EoS
• Different interactions explictly accounted for:
- Attractive vdW forces (dispersion, polar)
Attractive + Repulsion
- Association (hydrogen bonding) + Chain + Association
- Chain term  polymers
• Not easy development but great potential
- Several pure compound parameters (5-6) Z 3
- Need for experimental data (VP, density)
 SAFT – a theoretical concept
• Additive contributions from various effects
(attraction, repulsion, chain, association)
• Wertheim theory for chain and association
effects
• Segment concept
• Parameters with physical meaning
• Parameters following smooth trends / can be
predicted from independent data

Z = Zattr(disp) + Zrepuls(hs) + Zchain + Zassoc

 Parameters of the SAFT equation

Z = Zattr(disp) + Zrepuls(hs) + Zchain + Zassoc

σ = segment diameter

Chain molecules ε/ k B = dispersion energy

m = segment number

ε HB / k B = ass. energy
Associative molecules
κ HB = ass. volume

ε ij = ε iε j (1 − kij ) Parameters (typically) fitted to

vapor pressure and liquid density data
 SAFT – a family of models
SAFT Reference Citations (cit/year)

Original SAFT Chapman et al., 1990 1113 (41)

Original SAFT (II) Huang and Radosz, 1990 1027 (38)

Soft SAFT Vega and Blass, 1997 194 (10)

SAFT-VR Jackson et al., 1997 624 (31)

CPA Kontogeorgis et al., 1996 348 (17)

PC-SAFT Gross and Sadowski, 2001 1078 (67)
sPC-SAFT Von Solms et al., 2003 118 (8)

Source: Web of Science, October 21, 2016

 Many UNIFAC and far too many SAFTs !
Gmehling, 2009,
J.Chem.Thermodynamics,41:731 Fu and Sandler (1995)

Original SAFT
PC-SAFT
sPC-SAFT
SAFT-VR
Soft SAFT
SAFT-LJ
Original UNIFAC SAFT-Mie
Modified UNIFAC – Lyngby Polar SAFT
Modified UNIFAC – Dortmund Quadrupolar SAFT
LLE UNIFAC iSAFT
KOW UNIFAC GC-SAFT
.... .......
We see no significant differences between
the various association models as long as...
• We do not consider polymers
• We focus on phase behavior
• We consider mixtures of associating and non-associating
compounds
• Association terms of chemical, quasi-chemical and
perturbation theories are similar (Economou and Donohue,
1991, AIChE J, 37: 1875)

• We develop the models in the same way (same number of

adjustable parameters – association schemes - input data,
etc)
• Isn’t that remarkable ? For ”fundamentally different”
models ? (M.L.Michelsen)
 SAFT’s complexity not needed in many applications ?
Brainstorming towards a better model

NTUA (Greece) Researchers

1995: A new equation of state is born (CPA)

Z = Zattr(disp) + Zrepuls(hs) + Zchain +Zassoc

CPA first publication:
RT a Kontogeorgis et al., 1996
P= −
V − b V (V + b ) Cubic EoS is good enough !
The concept of association
Proton donor site
….. 1A Aliphatic Acids
Proton acceptor site
R-COOH

….. 2B Alkanols
R-OH

…..
Primary amines
3B
R-NH2
…..

….. 4C Water, glycols

 LLE of water-hexane with three models:
Good results with single kij (4C scheme)
1.E+0

Water in oil (petroleum interest)

1.E-2

Oil in water
(environmental interest)
mole fraction

1.E-4

1.E-6

x (H2O) x (HC) CPA NRHB PC-SAFT

1.E-8
270 320 370 420 470
Temperature / K

Liang, Maribo-Mogensen, Tsivintzelis, Kontogeorgis, 2016, FPE: 47: 2

 Water-hydrocarbon LLE with CPA

0.1 Water in Benzene rich phase

0.01
mole fraction

1E-3
Benzene in water rich phase
zero
1E-4
Water in Hexane rich phase ε associating + ε aromatic
ε =
Ai B j

1E-5 2

β fitted to
Ai B j
1E-6 Hexane in water rich phase
data
260 280 300 320 340 360 380 400 420 440 460 480 500 520 540
Temperature / K kij from
water-hexane
(homomorph)
Kontogeorgis, Tsivintzelis et al., 2011, Fluid Phase Equilibria, 301: 244
 Methanol/CC6 with CPA and PC-SAFT

Similar results for other methanol-HC VLE/LLE

Gross and Sadowski, 2002
 The system methanol-water-methane
Gas phase methanol content and methanol/water in gas phase
and methane in the liquid phase
T=313 K
0.01 0.1

methane in polar
water in gas
methanol in gas

0.01

mole fraction
mole fraction

0.001

273.15K 0.001

298.15K

313.15K

0.0001
0.0001 0 20 40 60 80 100 120 140
0 20 40 60 80 100 120 140 P / bar
P / bar

Exper. Data: Sinyavskaya et al., 1984, 1985 [published in Russian journals]

Kontogeorgis and Folas, 2010. Wiley
 1
1

0.1

n-hexane mole fraction

Methane mole fraction
Aqueous phase
0.1
0.01
Organic phase Aqueous phase

vapor phase 0.001 Organic phase

0.01 vapor phase

0.0001

0.00001
0.001
6 7 8 9 10
6 7 8 9 10 Pressure / MPa
Pressure / MPa
1 1

0.1
Methanol mole fraction

Aqueous phase Aqueous phase

0.1

Water mole fraction

Organic phase Organic phase

Vapor phase 0.01 vapor phase

0.01
0.001

0.001 0.0001
6 7 8 9 10 6 7 8 9 10
Pressure / MPa Pressure / MPa

Water – methanol – methane – hexane: New data and CPA

(Frost et al., 2015)
 MEG-water-hexane vs. MEG-water-benzene
Exp. Data: Razzouk et al., 2010, J.Chem.Eng.Data, 55: 1468 (first paper in 2004)
Folas, Kontogeorgis, Michelsen, Stenby and Solbraa, 2006, J. Chem. Eng. Data, 51: 977

0.01
HC in the HC phase
1
MEG-Water-Benzene at 298.15 K

0.1 1E-3
MEG-Water-Benzene at 298.15 K

mole fraction
mole fraction

0.01

MEG-Water-Hexane at 303.15 K MEG in the HC phase

1E-4
1E-3
HC in the polar phase
1E-4
1E-5
MEG-Water-Hexane at 313.15 K
1E-5

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
MEG mole fraction in the polar phase MEG mole fraction in the polar phase

Kontogeorgis, Tsivintzelis et al., 2011, Fluid Phase Equilibria, 301: 244

>100 multicomponent systems studied with CPA

• Oil-water-MEG
• Hydrocarbons-water-MEG, methanol (LLE,
VLLE, hydrate curves)
• CO2, H2S – water – HCs – glycols, alcohols
• Mercaptanes-water-methane
• Chemicals (water-DME-ethanol; biodiesel,
lipids & esters-water-alcohols; acetic acid-
water-polars; CO2-DMC-water-methanol,
octanol, benzaldehyde, benzyl alcohol,..; )
Fluid Phase Equilibria, 2011, 311, 83-92
 CPA in industry
ASPEN PlusTM
http://www.aspentech.com/products/aspen-hysys/whats-new/

http://energygroup.ir/2015/08/08/aspen-tech-aspen-one-ver-8-8/

Calsep, KBC, Hydrafact, Tie-Line

http://www.pvtsimnova.com/news/24/105/October-November-2015-Introducing-CPA-in-PVTsim-Nova-2-
0.html

http://www.4-traders.com/KBC-ADVANCED-TECHNOLOGIES-4002067/news/KBC-Advanced-Technologies-and-
University-of-Aveiro-announce-cooperation-for-extension-of-CPA-model-18810128/

http://www.hydrafact.com/index.php?page=software

http://tie-tech.com/products/software/

”Every model is wrong – Some models are useful”

More concerns - reservations with
thermodynamic models
Acetone(inert)/hexane VLE and LLE with sPC-
SAFT
3D

kij from VLE

kij from LLE

Amra Tihic’s PhD Thesis 54

 ”Adjustable parameters” – the curse of
thermodynamicists
How many parameters do you need in
order to fit an elephant ?

http://levenspiel.com/octave/elephant.htm

Friedrich Gauss:
Give me four parameters, and I will draw an elephant for you;
with five I will have him raise and lower his trunk and his tail.
 Example of the importance of adjustable parameters!
Solid-Gas Equilibria for CO2 – benzoic acid with the
Peng-Robinson Equation of State
P=
RT
−
a
α= ∑∑ x x α b = ∑∑ xi x j bij
V − b V (V + b ) + b(V − b )
i j ij
i j i j

aij = ai a j (1 − k ij ) kij = 0
bi + b j lij = 0 kij ≠ 0
bij = (1 − lij ) 3%
2 lij ≠ 0

kij ≠ 0
lij = 0
T = 35 oC
11%

Coutsikos et al., 2003, J. Super. Fluids, 25: 197

 The power of adjustable parameters
TEG-water with one kij 440

1.2 0.85 bar

420
332.6 K

Temperature / K
0.9

400
activity coefficient

297.6 K
0.6

380
Exp. data at 0.85 bar
0.3 Exp. Data 297.6 K set-1
Exp. Data 332.6 K set-2
set-1 set-3
set-2 360
set-3 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0.98 0.99 1.00
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Water mole fraction
Water mole fraction

1.0
0
0.9

0.8
-500
0.7

0.6
-1000
γ∞water

0.5

0.4 HE / J mol-1
-1500
0.3

0.2
Exp. data
set-1 Exp. Data
-2000
0.1 set-2 Set 1
set-3 Set 2
0.0 Set 3
300 320 340 360 380 -2500
0.0 0.2 0.4 0.6 0.8 1.0
Temperature / K
mole fraction of water

Arya, Maribo-Mogensen, Tsivintzelis, Kontogeorgis, 2014, IECR, 53: 11766

 The power of adjustable parameters
TEG-water with kij(T) kij(T) & cross association
1.2

1.2
332.6 K
332.6 K
0.9
0.9

activity coefficient
activity coefficient

297.6 K
297.6 K 0.6
0.6

Exp. Data 297.6 K

Exp. Data 297.6 K 0.3 Exp. Data 332.6 K
0.3 Exp. Data 332.6 K set-1
set-1 set-2
set-2 set-3
set-3 set-4
set-4 0.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.2 0.4 0.6 0.8 1.0
Water mole fraction Water mole fraction

1.0
1.0
0.9
0.9
0.8
0.8
0.7
0.7
0.6
0.6
γ∞water

0.5
γ∞water

0.5
0.4
0.4
0.3 Exp. data
0.2
set-1 0.3 Exp. data
set-2 set-1
0.1 set-3 0.2 set-2
set-4 set-3
0.0 0.1
300 320 340 360 380
set-4
0.0
Temperature / K 300 320 340 360 380
Temperature / K

Arya, Maribo-Mogensen, Tsivintzelis, Kontogeorgis, 2014, IECR, 53: 11766

 The role of parameter estimation – the
case of methanol (for PC-SAFT)

Set Tr-range % AAD in Psat % AAD in density

Gross – Sadowski, 0.39-1.00 4.5 0.4
2002
Von Solms et al., 0.39-0.98 1.0 0.5
2003
Ane S. Avlund 0.42-0.95 0.9 0.1

All % AAD reported at Tr-range 0.42-0.95 only !

A. Avlund, 2011 (PhD thesis)

 Methanol/hexane with PC-SAFT

prediction correlation

Von Solms and ASA better prediction of VLE and esp. LLE (no LLE prediction with GS)

GS clearly superior to Von Solms/ASA for LLE correlation (VLE – all similar)
Von Solms/ASA need negative kij values

Similar conclusions for methanol/heptane Luck or used in parameter estimation ?

 Advanced models at a price !

Equation of State Parameters – pure Parameters – mixtures

compounds
Ideal gas equation 0 0
Cubic EoS 0 (Tc, Pc, acentric factor) Often one (kij)
CPA/SAFT 3-5 Often one (kij)

H-O-CH2-CH2-O-H H-O-(CH2)2-O-(CH2)2-O-(CH2)2-O-(CH2)2-O-H

1.0E+02 0.0
1.0E+02 0.0
1.0E+01 2.0 Molar liquid density (mol dm-3

Molar liquid density (mol dm )

-3
1.0E+00 1.0
4.0
1.0E+00

Vapor pressure (bar)

Vapor Pressure (bar)

6.0 1.0E-02 2.0

1.0E-01 8.0 1.0E-04 3.0
1.0E-02 10.0
1.0E-06 4.0
1.0E-03 12.0
14.0 1.0E-08 5.0
1.0E-04
16.0 1.0E-10 6.0
1.0E-05 18.0
1.0E-12 7.0
1.0E-06 20.0 200 300 400 500 600 700 800
250 350 450 550 650 750
Temperature (Kelvin) Temperature (Kelvin)

MEG Tetraethylene glycol

 Vapor pressure data are sometimes
hard to find!
• Polymers
• Complex heavy compounds e.g.
pharmaceuticals, biomolecules
Chymotrypsin molecule
• Ionic liquids regions og helix
and sheet are shown
• Electrolytes

Water

Ribonuclease A Atkins and de Paula, 2006. Physical Chemistry for the Life Sciences.
Oxford Univ.Press
 H-bonding of water against recent
molecular simulation data
Liang, Maribo-Mogensen, Tsivintzelis, Kontogeorgis, 2016, FPE: 47: 2

All models MORE Hydrogen Bonding

Than Experimental Data

MS data from Dufal, Lafitte, Haslam, Galindo, Clark, C. Vega, G. Jackson, 2015, Mol. Physics
 Derivative properties
F= Ar/RT  ∂ 2
F  ∂F 
Cvres (T , V , n ) =
− RT 2  2  − 2 RT  
Cv (=
T , V , n ) Cvideal (T ) + Cvres (T , V , n )  ∂ T V ,n  V ,n
∂T

 ∂P   ∂P 
2
C p (=
T , V , n ) C pideal ( T ) + C pres ( T , V , n ) C pres =
Cvres − T     − Rn
 ∂T V ,n  ∂V T ,n

 ∂P   ∂2 F  P  ∂P   ∂2 F  nRT
  = − RT   +   =
− RT  2 −
 ∂T V ,n  ∂T ∂V n T  ∂V T ,n  ∂V T ,n V
2

C p  ∂P 
u = −V ⋅ ⋅ 
2
MW
Cv  ∂V T ,n
 Speed of sound with PC-SAFT, CPA and SRK
for n-octane at 300 K
3000

NIST C p  ∂P 
u = −V 2 ⋅ ⋅  MW
SRK (24.6%)
Cv  ∂V T ,n
2500

CPA (18.6%)

PC-SAFT (12.7%)
Speed of Sound (m/s)

2000

1500

1000

500
0 30 60 90 120 150
Pressure (MPa)

Ackn. Xiaodong Liang (Liang et al., 2012)

 PC-SAFT for n-C6
•1: Putting Speed of Sound into Parameter
Estimation
• 2: Putting Speed of Sound into both Universal
Constants and Parameter Estimation
2400
1400 NIST
NIST
Original (10.68%)
Original (7.04%) Approach 1 (10.32%)
1200
Approach 1 (6.79%) 2000
Approach 2 (0.63%)

Speed of Sound (m/s)

Speed of Sound (m/s)

Approach 2 (0.58%)
1000
1600

800

1200
600

400 800
228 278 328 378 428 0 30 60 90 120 150
Tempearture (K) Pressure (MPa)

Speed of Sound of saturated Speed of Sound of n-Hexane at

Liquid n-Hexane from different 300K from different approaches
approaches

Xiaodong Liang et al., 2012

 Cv res and Cp res for n-heptane
2.5 30
heptane (Tr = 1.10)
heptane (Tr = 1.10)
PC-SAFT
PC-SAFT
SAFT
SAFT
2 CPA
CPA

20
1.5
Cvres/R

Cpres/R
1
10

0.5

0 0
0 1 2 3 0 1 2 3
P/Pc P/Pc

The maximum in Cv (res) cannot be captured by any of the models !

The maximum in Cp (res) can be captured by all models !

De Villiers et al. (2012)

 VLE and Heat of mixing data in
parameter estimation (with CPA)
Methanol- DMC Case A: Only VLE Case B: VLE + HE
(two adjustable in both cases)
2800 313.15 K
428.15 K
411.15 K
10
2400
391.15 K

2000

HE / J mol-1
Pressure / bar

377.15 K 1600

1 1200
337.35 K
800

313.15 K
400
Case A
Case A Case B
Case B 0
0.1 0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Methanol mole fraction
Methanol mole fraction

I. Tsivintzelis, 2013 (unpublished results)

See also Tsivinitzelis et al., 2013. J. Super. Fluids, 84: 155
 … and the importance in ternary
prediction
exp. data
calculated binodal
(Case A, binary parameters from VLE
0.0 calculated binodal
1.0
(Case B, binary parameters from VLE + HE
0.2
0.8
283.15 K

Me
0.4
ter

0.6

tha
Wa

no
l
0.6
0.4

0.8
0.2

1.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0
DMC

I. Tsivintzelis, 2013 (unpublished results)

See also Tsivinitzelis et al., 2013. J. Super. Fluids, 84: 155
 Challenges we have seen
Challenge Status Challenge Status
Organic acids -water VLE at different T Multifunctional Complex association
needs different chemicals e.g. schemes not needed ?
parameters glycolethers,
alkanolamines
Aromatic acids-water VLE,LLE and SLE Polarity, Are they really really
simultaneously is quadrupolar effects needed ?
difficult
Predicting Few monomer Intramolecular Is it really really
parameters fraction data and not association needed ?
very reliable
Lack of P(sat) data
Electrolytes Which e-terms and Derivative Speed of sound not
description of properties possible with CPA at
activity coefficients high pressures – need
and solubility with to go to specific SAFT
same parameters variants
Adsorption CPA+MPTA – does it Interfacial CPA + DGT – does it
work ? properties work ?
 CPA-Huron Vidal for water-acetic acid
1

0.9 Set 1
Set 2
Set 3
relative volatility

0.8
503 K, Freeman et al.

0.7

0.6

0.5

0.4
0 0.2 0.4 0.6 0.8 1
mole fraction of acetic acid (liquid)
Breil et al. (2011) Ind.Eng.Chem.Res., 50(9): 5795-5805
 CPA-Huron Vidal for water-acetic acid
1

0.9

0.8
relative volatility

0.7

0.6
Set 1
0.5 Set 2
Set 3
0.4 293 K, Lazeeva et al.

0.3
0 0.2 0.4 0.6 0.8 1
mole fraction of acetic acid (liquid)

Breil et al. (2011) Ind.Eng.Chem.Res., 50(9): 5795-5805

Product design
 Thermodynamics in product design
• Poymer thermodynamics
• Solvent choice for polymers/paints
• Solvent choice for pharmaceuticals
Choice of elastomeric glues in thermopane Controlled drug release
windows – permeability of gases and water
 Molecules in PC-SAFT

d2

m2 m1
hard spheres
chain formation
d1
 Molecules in PC-SAFT

ε2
association
d2 κ AB , ε AB

m2
hard spheres m1
chain formation
dispersion ε1 d1
 Z = Zattr(disp) + Zrepuls(hs) + Zchain + Zassoc PC-SAFT

ε2
association
d2 κ AB , ε AB

m2
hard spheres m1
chain formation
dispersion ε1 d1
 Simplified PC-SAFT (CERE)
1
 ∑ xi mi d 3
 3
 i i

d = 
 ∑ xi mi 
 i 

association
d

hard spheres
chain formation
dispersion d
 Computer-Aided Product Design with
Advanced Thermodynamic Models
First Phase of the project (2002-2005):
The development of sPC-SAFT and applications to polymers (binary, ternary)

Second Phase of the project (2005-2008):

The group-contribution sPC-SAFT
Associating mixtures (water, alcohols, glycols)
Comparisons with the NRHB EoS
Pharmaceuticals

Funding: Danish Research Council for Technology and Production

 Scatter plots using GC PC-SAFT
Parameters for 500 non-associating compounds
15 45 FOG and 26 SOG
R2 = 0.9752

12
GC-estimated m

4.5
6
R2 = 0.9892
4

GC-estimated σ
3

0 3.5
0 3 6 9 12 15
DIPPR-estimated m
3

2.5
2.5 3 3.5 4 4.5
mmolecule = ∑ (ni mi ) FOG + ∑ (n j m j ) SOG DIPPR-estimated σ 450
i j R2 = 0.9152
400

GC-estimated ε/k
mσ 3 molecule = ∑ (ni miσ i ) FOG + ∑ (n j m j σ j ) SOG
3 3 350

i j 300

(mε / k ) molecule = ∑ (ni mi ei / k ) FOG + ∑ (n j m j ε j / k ) SOG 250

i j 200

150
150 250 350 450
DIPPR-estim ated ε/k
Tihic et al., 2008
 VLE of complex binary mixtures

CO2 + methyl oleate CO2 + methyl palmitate

Tihic et al., 2009

Example: Poly(iso-propyl methacrylate) (PIPMA)

FOG (i ) Occurrences ni mi miσ i3 mi ε i / k

CH3 3 1.93309 102.5086 388.1598
CH2 1 0.384329 24.33981 102.3238
CH 1 0.043834 13.95391 68.20842
C 1 -0.49208 2.325415 -10.983
COO 1 1.43911 32.51328 351.1344
∑ ni ⋅ mi = ∑ ni ⋅ miσ i3 = ∑ ni ⋅ mi ε i / k =
3.30828 175.64107 898.8434

SOG ( j ) Occurrences nj mj m jσ 3j m jε j / k
(CH3)2CH 1 0.01626 0.28087 -9.83615
∑nj ⋅mj = ∑ n j ⋅ m jσ =3
j ∑ n j ⋅ m jε j / k =
0.01626 0.28087 -9.83615

m / Mw = (3.30828 + 0.01626) / 128.17 = 0.02594

σ = 3 ( mσ 3 ) / m = 3 (175.922 / 3.3245 = 3.7543 Å

ε / k = (mε / k ) / m = 889.0073 / 3.3245 = 267.41 K

Tihic et al., 2008

 LLE (UCST) with a single
interaction parameter
Polyisobutylene-Diisobutylketone
340

330 6000000
285000
320
22700
310

300
T (K)

kij=0.0053
290

280

270

260 Shultz et al. J.Amer.Chem.Soc. 74, 4760 (1952)

250
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
weight fraction polymer

Von Solms et al., 2004

 LLE of polymer systems - PS

With acetone With

methylcyclohexane
500 600
450
Temperature [K]

500

Temperature [K]
400 Mw : 48 000
350 kij=0.00
kij= -0.0085 400 kij = 0.0065
300
250
300
200
150
200
0 0.1 0.2 0.3 0.4 0 0.05 0.1 0.15 0.2
PS weight fraction PS weight fraction

Tihic et al., 2008

 GC SAFT is a popular recent trend
• There are GC versions of original (CK) SAFT,
PC-SAFT and SAFT-VR
• Two widely applied GC SAFT methodologies:
- (Polar) PC-SAFT by the IFP group (Tamouza et
al., 2005, 2006; Thi et al., 2005, 2006)
- SAFT-gamma (a GC version of SAFT-VR) by the
Imperial group by Lymberiadis et al., 2007,
2008)
• All these approaches are under continuous
developments
 Solvent choice for pharmaceuticals
”30% of Mitsubishi’s thermodynamic group is related to solvent selection”
(Kolar et al., 2002) O
O
O
O O
HN CH3 CH3
HN CH3
HN CH3

OH OH
O

CH3

Acetanilide Paracetamol Phenacetin Methyl-paraben

O
O
O O
OH OH
OH

H3C CH3 CH3

H3C O
CH3 CH3

Ibuprofen Naproxen Ketoprofen

Solvent screening for pharmaceuticals – an
application from Mitsubishi Chemicals

Why is RST
no good here ?

x 2ideal
x 2real =
γ2

Kolar et al., 2002

 V2L  2  V2L   V2L 2 

ln x 2 ≈  − δ 1 +  2δ 2 δ 1 +  ln x 2 −
id
δ 2 
 RT   RT   RT 
The NRHB – an association model for pharmaceuticals
Tsivintzelis, Economou and Kontogeorgis AIChE J., 55, 2009, 756-770
Tsivintzelis, Economou and Kontogeorgis J. Phys. Chem. B, 113, 2009, 6446-6458

–OH….OH– R-OH Alkanols

>NH….NH< N-methyl-aniline

Paracetamol  Estimate the parameters for specific interactions

from simple molecules
 A ”group” contribution type approach for
pharmaceuticals
Pharmaceuticals’ NRHB: A group table developed for association parameters

HB interactions Ehb (J mol-1) Shb (J mol-1 K-1) Adopted from

–OH….OH– -24000 -27.5 Alkanols
–OH….O=C< -16868 -13.25 Acetone-Methanol
–OH…. –O– -10943 -13.75 Ethanol- Diethylether
–OH…. –O–C=O– -12000 -13.75 Propylacetate – 1-Propanol
>NH….O=C< -9176 -6.94 Acetanilide
>NH….–O– -10900 -6.90 Phenacetin
>NH….NH< -8500 -10.25 N-methyl-aniline
>NH….OH– -18100 -16.0 Paracetamol
HOH….HOH -16100 -14.7 Water
HOH….O=C< -13412. -7.35 Water-Cyclohexanone
–COOH….COOH– -16380 -20.4 Propionic acid

HB Interactions
EiH + E Hj S iH
1/ 3
+ S Hj
1/ 3

>NH…. –O–C=O – E H
ij = S H
=( )3
2 ij
2
HOH….–O–
HOH…. >NH Two types of cross-association
HOH…. –OH
EiH SiH
–OH…. –COOH
Eij =
H
S =
H
ij
>COOH….>C=O 2 2
Solubility of P aracetam ol in various solvents w ith N RHB

0.08

0.06 Paracetamol - Ethanol

Solute mole fraction

0.04

Paracetamol - Acetone
0.02
Paracetamol - Acetonitrile

0.00
265 270 275 280 285 290 295 300 305
Temperature / K
 Solubility of naproxen in various solvents with NRHB

Naproxen - Octanol
0.08
Naproxen - Acetone 0.01
Solute mole fraction

Solute mole fraction

0.06
1E-3 Naproxen - Propylene glycol

0.04 Naproxen - Chloroform 1E-4

Naproxen - Ethanol
Naproxen - water
0.02 1E-5
Naproxen - Methanol

0.00 1E-6
290 300 310 320 330 290 295 300 305 310 315
Temperature / K Temperature / K

0.01
293.15 K

1E-3
Solute mole fraction

1E-4
Water-ethanol mixture
1E-5

1E-6
0 20 40 60 80 100
Weight fraction of ethanol in solvent
 Solubility of ketoprofen with NRHB

O O
OH

CH3
Water-acetone mixture
0.1
-1
10
Ketoprofen - acetone
0.01
10-2

Solute mole fraction

Solute mole fraction

1E-3
10-3
283 K
-4 1E-4
10

Ketoprofen - water
10-5 1E-5

10-6
1E-6
280 290 300 310 320 330 0 10 20 30 70 80 90 100

Temperature / K Weight fraction of water in solvent

 NRHB vs. COSMO-RS

0
COSMO-RS
NRHB
Log10 (Xpredicted)

-2

-4

-4 -2 0
Log10(Xexperimental)
 Many association models (EoS) recently
developed for pharmaceuticals
• CPA (Queimada, Macedo)
- F.L.Mota, A.J. Queimada, S.P.Pinho, E.A. Macedo, 2011. Solubility of drug-like
molecules in pure organic solvents with the CPA EoS. Fluid Phase Equilibria .
• PC-SAFT (Sadowski, Economou)
- J. Cassens, F. Ruether, K. Leonhard, G. Sadowski, 2011. Solubility calculations of
pharmaceutical compounds. A priori parameter estimation using quantum
chemistry. Fluid Phase Equilibria.
- F. Ruether, G.Sadowski, 2009. Modeling the solubility of pharmaceuticals in pure
solvents and solevnt mixtures fir drug process design. J. Pharmaceutical Sciences.
- T.Spyriouni, X.Krokidis, I.G. Economou, 2010. Thermodynamics of pharmaceuticals:
prediction of solubility in pure and mixed solvents with PC-SAFT. Fluid Phase
Equilibria.
• NRHB – Tsivintzelis and Kontogeorgis

There are also ”practical” pharmaceutical models like ASPEN’s NRTL-SAC

Mathias, 2005, FPE, 228-229: 49; O’Connell et al., 2005, IECR, 218:. 4619
 Conclusions
• There are fewer ”really different and useful” models
– concepts than thought by many e.g. GC, RST,
cubics, LC, association
• For most models, ”details” in parameter estimation
are more important than model functionality e.g.
• Almost all association models perform similarly e.g.
• CPA, PC-SAFT and NRHB have all been developed for
polymers and pharmaceuticals
• When looking at many ”details” things can becove
(very) difficult e.g. properties other than phase
equilibria incl. derivative properties, electrolytes,
adsorption, …