Journal of Chromatography A, 1250 (2012) 227249

Contents lists available at SciVerse ScienceDirect

Journal of Chromatography A
journal homepage: www.elsevier.com/locate/chroma

Review

Design of preparative-supercritical uid chromatography

Arvind Rajendran
School of Chemical and Biomedical Engineering, Nanyang Technological University, Singapore 637459, Singapore

a r t i c l e i n f o a b s t r a c t

Article history: Preparative supercritical uid chromatography (prep-SFC) is an important separation process in the chro-
Available online 16 May 2012 matographers toolbox. Owing to the unique properties of the mobile phase, which is predominantly CO2 ,
the behavior of SFC is markedly different from high performance liquid chromatography (HPLC). This
Keywords: review article focuses on the scale-up of preparative chromatography. The basics of SFC, with particular
Supercritical uid chromatography focus on highlighting the key differences between SFC and HPLC, are introduced. Then, a framework for
High pressure chromatography
rational design of prep-SFC is proposed. This framework is based on obtaining basic system parameters
Carbon dioxide
from analytical scale equipment, i.e., with very small amount of material, and performing design and
Design and optimization
Isotherm determination
optimization in silico to evaluate process performance and to identify operating conditions for scale-
Preparative separations up. The tools required to obtain the input parameters such as adsorption isotherms are discussed and
the development of the design and optimization framework is elaborated. Examples from the literature
which use this approach for successful scale-up are provided. Finally the design of multi-column SFC
systems is discussed.
2012 Elsevier B.V. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
2. Review of SFC fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
2.1. Overview of SFC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
2.2. Microscopic picture of adsorption at supercritical conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
2.3. Key differences between HPLC and SFC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
2.3.1. Inuence of pressure on retention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
2.3.2. Compressibility effects in SFC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
2.3.3. Large-volume injections in SFC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
2.3.4. Swelling of stationary phase in SFC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
3. Prep-SFC operating principles and the case for optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
3.1. Traditional scale-up approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
3.2. Framework for rational design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
4. Process modeling and simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
4.1. Detailed modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
4.2. Numerical solution of chromatographic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
4.3. Equilibrium theory of chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
5. System characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
5.1. Adsorption isotherms: measurement and characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
5.2. Measurement of multi-component isotherms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
5.2.1. Static methods of measuring adsorption isotherms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
5.2.2. Dynamic methods of measuring adsorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
5.2.3. Estimation of binary/multi-component non-linear isotherms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
5.3. Measurement of pressure-drop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

Tel.: +65 6316 8813; fax: +65 6794 7553.

E-mail address: arvind@ntu.edu.sg

0021-9673/$ see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.chroma.2012.05.037
228 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249

6. Optimization of chromatographic separations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

6.1. Problem denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.1.1. Objective functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.1.2. Decision variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.1.3. Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.2. Optimization techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
6.3. Multi-objective optimization of single column prep-SFC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
7. Design of multicolumn chromatography with supercritical mobile phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
8. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
9. Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247

1. Introduction mobile phase. The advantages of using supercritical uids can be

readily appreciated by considering their physical properties. The
Preparative chromatography (prep-chromatography) is an viscosity of sc-CO2 is in the range of 0.020.1 cP (temperature
important separation method that has had signicant impact range of 3075 C) which is still lower than most commonly used
in several industries including pharmaceutical, food, fragrance, mobile phases in high performance liquid chromatography (HPLC).
agrochemicals, etc. [15]. The particular advantage of prep- This allows for higher mobile phase ow rates with moderate pres-
chromatography arises from the ease of scalability, robustness sure drops. Further, the diffusion coefcients of solutes in sc-CO2
of the equipment and the possibility to separate a diverse set is larger than that in organic solvents and it is possible to obtain
of mixtures, both binary and multicomponent. Further prep- higher efciencies [2832]. These properties make sc-CO2 as a
chromatography is an example of a separation process that nds strong contender for a choice as a mobile phase. The non-toxic and
applications from scales ranging from a few micrograms to tonnes non-ammable nature of CO2 makes it an ideal choice for use in
[6]. The impact of chromatography on enantiomer separation is food and pharmaceutical processes.
well known [1,4]. This has been possible due to simultaneous Over the years, SFC has been developed into a mature technology
developments in the synthesis of chiral stationary phases [711] both for analytical and preparative separations [5,28,30,3336].
and process technologies [2,12]. Several blockbuster drugs contain During the early days of the technology SFC was widely expected
active pharmaceutical ingredients (APIs) that were puried using to replace HPLC. However, in the years that followed, it became
at least one chromatographic step [1316]. In recent years prep- increasingly clear that SFC, while not capable of replacing HPLC has
chromatography is starting to make impact on the separation of an important role in separation sciences. The demand for chiral sep-
biomolecules, e.g., peptides, mono-clonal antibodies, etc. [17]. arations in the pharmaceutical industry, improved understanding
Prep-chromatography is routinely used for the purication from of the fundamentals, availability of reliable instrumentation and the
a few grams to tons/day scale. For small scales (typically up to need to reduce operating costs and the emphasis on green sustain-
10 s of kg/day) single column processes and their variations are able chemistry in the mid-late 1990s have made SFC an attractive
used. For larger scales, complex multi-column processes such as separation alternative.
the simulated moving bed (SMB) and its variants are employed The advantages of SFC is now well-documented and several
[1,4,18]. The choice of a process for preparative scale separations excellent review articles and monographs have been published
depends on several aspects, viz., as productivity, solvent consump- [5,32,33,3537]. There are also interesting case studies that have
tion, complexity of design and certainly the cost of separation. A documented the advantages of SFC over HPLC, in terms of speed,
key shortcoming of chromatography is that the solutes that need cost of operation, etc. [38]. The reader is suggested to refer to
to be separated are often dissolved in a solvent and injected into these works to obtain a background of this technology. The focus
the column. The maximum allowable injection concentration is dic- of this review is on the design aspects of preparative SFC (prep-
tated by the solubility of the solute mixture in the solvent. Owing SFC). The basics of prep-SFC are introduced and the methods of
to the inherent nature of elution chromatography, the products are measuring fundamental properties required for the design are dis-
usually collected at dilute conditions compared to the feed. Hence, cussed. Advancements in numerical techniques and optimization
concentrating the solute from the product requires the evaporation algorithms have provided tools that can have signicant impact on
of signicant quantities of the solvent. This process requires energy the design of chromatographic separations. The case for using these
and increases the operating costs [1922]. While the solvent can techniques in routine separations is made and the steps involved
be reused for a certain number of times, they have to be eventu- therein are discussed. Examples of successful applications are pro-
ally disposed. They are either treated by a suitable waste-treatment vided.
process or incinerated to recover heat [23].
Reducing solvent consumption in order to make chromato-
graphic separations more environmentally friendly and to reduce 2. Review of SFC fundamentals
the cost of separations is the key driver in todays industrial opera-
tions [2325]. Further, increasing cost of solvents, either owing to 2.1. Overview of SFC
the increasing costs of crude oil or owing to supply shortages, e.g.,
the acetonitrile shortage of 2008 have pushed separation scientists The schematic of a typical prep-SFC unit is shown in Fig. 1. The
to look for viable alternatives for organic solvents [26]. system consists of the chromatographic separation and fraction-
Supercritical uid chromatography (SFC) which employs a com- collection modules. The CO2 from the tank is liquied and pumped
pressed gas as a mobile phase has been studied since the 1960s using high pressure pumps to which the modier is added. The
when Klesper reported the separation of porphyrin derivatives mixed stream is then heated to the required temperature. The
using supercritical dichlorodiuromethane as a mobile phase [27]. sample to be separated is injected into the mobile phase and the
While several uids including NH3 , SF6 , etc., have been used, super- column separates the mixture into its constituents. A back pres-
critical CO2 (sc-CO2 ) has been the most popular choice for the sure regulator (BPR) is used to control the pressure in the system.
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249
Chromatography
Pin
High pressure
pump
Injection
Valve
Cooler
Rotary pumps
Exhaust to vent or
to recycle
CO2
Mod
Fig. 1. Schematic of a typical preparative SFC system showing the chromatography and the fraction collection modules.
Downstream of the BPR is the fraction collection system that con-
sists of solenoid valves and cyclones. The mobile phase which
consists of CO2 + modier + solute separates into two phases down-
stream of the BPR and the switching valve directs the stream to an
appropriate cyclone where the gas and the liquid phases are sep-
arated. Typically, the modier and solute precipitate in the liquid
phase while the CO2 stream leaves the cyclone. It is not uncommon
for some amount of CO2 to dissolve in the liquid phase or for the
carryover of the modier and solute with the CO2 . The gaseous CO2
phase can be puried; sent for re-compression and used or vented.
The liquid phase is sent to an evaporation unit where the solute is
separated from the solvent.
Carbon dioxide has been the most commonly used mobile phase
in SFC. In addition to the advantages described above, it has criti-
cal points (Pc = 73 bar, Tc = 31.1 C) that are easily accessible and in
a range where thermally labile substances can be separated. The
main advantage of using supercritical uids stems from their tun-
able properties. For example by adjusting the pressure, it is possible
to signicantly alter the solvent strength of the uid. Hence, it is
possible to operate the chromatographic separation at a high pres-
sure, i.e., as a strong solvent, and reduce the pressure to recover
the product. This is an important advantage as it is possible to
evaporate a signicant portion of the mobile phase by lowering the
pressure and hence avoiding the energy consuming step of heat-
ing. While CO2 possess all these advantages, its main drawback is
its low polarity which makes it a relatively poor solvent for most
commonly encountered solutes. In order to overcome this, polar
modiers (typically lower alcohols) are added as modiers. Fig. 2(a)
shows the variation of mobile phase polarity with the addition of
methanol. It is common to see upto 50% modier compositions in
preparative separations. The addition of the modier also alters
the critical properties. Fig. 2(b) shows the variation of critical pres-
sure and temperature with the addition of methanol. It can be seen
that the critical pressure increases with the addition of methanol
and then falls, while the critical temperature increases with the
addition of methanol. It is worth noting that typical SFC separa-
tions are carried out around room temperature (2030 C) and at
pressures of 100200 bar. Under these conditions, the system is
229
Pout
UV
Back Pressure
Column
Regulator
Cyclone
separator
Fraction 1 Fraction 2 Fraction 3
Fraction collection
inherently sub-critical. Hence in most practical situations the use
of the term supercritical is only indicative of the technique and
does not strictly refer to the state of the uid. Finally, from the
two-component phase diagram shown in Fig. 2(c) it can also be
seen that under typical operating conditions, the uid is present in
the single-phase subcritical region.
2.2. Microscopic picture of adsorption at supercritical conditions
Before discussing the design of prep-SFC, it is illustrative to
understand the microscopic picture of adsorption on the solid
surface at supercritical conditions. In the most simplest systems
encountered in prep-SFC, the mobile phase consists of CO2 , organic
modier and the solute. At high pressures owing to the attractive
potential of the solid surface, all three components adsorb on the
surface. In order to understand the effect of competitive adsorption,
an ideal experiment should be able to provide quantitative infor-
mation about the adsorbed amounts of the individual components
as a function of uid phase pressure, temperature and composition.
However, there are several challenges associated with the measure-
ment of adsorption at high pressures. These challenges and some
examples of high pressure adsorption are described below.
In order to illustrate the challenges, let us consider the adsorp-
tion of a pure uid in a pore of volume Vtot . In the presence of
an adsorbent wall, the density prole of the adsorbed phase is as
shown in Fig. 3. Under these conditions the absolute adsorbent
amount, nabs is dened as

1
nabs = (x, y, z)dV (1)
msorb Vtot
where msorb is the mass of the sorbent, and (x, y, z) is the three-
dimensional density prole. At low pressures, it is possible to
explicitly measure nabs as it can be safely assumed that the adsorbed
phase density is sufciently larger than the bulk density of the uid.
However, when the uid phase pressure is high, this assumption
is no longer true and the density of the uid phase is compara-
ble to that of the adsorbed phase. Under these conditions, the only
230 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249

(a) (b) (c)

Fig. 2. Phase behavior of CO2 methanol system (a) polarity of the mobile phase [140], (b) variation of critical temperature and pressure with methanol composition [141]
and (c) binary phase diagram [142].

measurable quantity, no matter what experimental technique is
used, is the excess adsorbed amount [3942]. The excess adsorbed
amount, nex , is dened as:
 the bulk uid is close to its critical point. Hence, it can be specu-
ex 1
n = [(x, y, z) b ]dV
msorb Vtot
The above equation can also be written as
ex
n = nabs b V a
where Va is the volume of the adsorbed phase. From this equation it
is clear that the difculty in measuring nabs arises from the inabil-
ity to accurately measure Va . Further, it is also clear that at low
bulk densities nabs  b Va and hence the measured quantity, nex , is
indeed equal to the absolute amount, nabs .
Fig. 4 shows the excess adsorption isotherms of CO2 on 13X
zeolite and silica gel [43,44]. Owing to the denition of the excess
adsorbed amount, the isotherms have a distinct shape. At low pres-
sures, nex increases with b , reaches a maximum and decrease with
further increase in density. Further, at a given density, it can also
be seen that the adsorbed amount increases with decreasing tem-
perature. The difference between the shapes of excess isotherms
for silica gel and 13X zeolite is very evident. 13X zeolite, a microp-
orous solid, has a pore size of 12 A (approximately 3 layers of CO2
molecule) while silica gel, a mesoporous sorbent, has a pore size of
60 A (approximately 15 layers of CO2 molecule). For microporous
solids, e.g., 13X zeolites, the pores are saturated even at low bulk
densities and there is no further capacity for adsorption at higher
bulk densities. Hence, at higher density the rst term in Eq. (3)
does not change much and the decrease in nex is due to the sec-
ond term. Hence, nex falls nearly linearly with increasing b . For
meso- and macroporous solids, since the pores can accommodate
multiple layers of the sorbate molecules, the excess increases more
pore wall
b b
0 0
Fig. 3. The concept of (a) excess and (b) absolute adsorption for a uid in a pore.
The shaded area shows the (a) absolute amount adsorbed and (b) excess amount
adsorbed.
smoothly and shows a smooth fall. From the gures, it is worth
noting that the adsorption of CO2 on the solids is non-negligible
and can indeed lead to multi-layer adsorption particularly when
(2) lated that CO2 competes with other components that are present in
the uid phase, a situation that is encountered in SFC. The possible
effects of multilayer adsorption on SFC performance was recently
reported [45].
(3) The experimental results can be further understood with the
help of molecular simulations. Fig. 5 shows lattice-gas model pre-
dictions of excess adsorption at two temperatures T/Tc = 1.2 and
1.0001, i.e., both far from and very close to the critical tempera-
ture, respectively. The effect of the pore-size of the adsorbent is
also shown by considering micro-, meso- and macro-porous sor-
bents. In each plot, the contribution of the rst layer, from rst
three layers and the total excess adsorbed amount are shown. Note
that even at temperatures far from the critical value multi-layer
adsorption is seen. It is also worth noting that the shapes of the
excess isotherms are characteristic of pore sizes. For the case of
micropores, the isotherm rises sharply and falls almost linearly
with density. However, for meso- and macroporous sorbents, the
decrease is rather smooth. This is consistent with experimental
results shown in Fig. 4.
At temperatures close to critical values, i.e., at T/Tc = 1.0001,
the excess isotherm shows interesting behavior. For microporous
solids, since the width of the pore is small, i.e., only a few adsor-
bate layers thick, there is no possibility for contributions beyond
the rst few layers. However, for mesoporous systems at both very
low and very high densities, only a few layers contribute to the
excess adsorbed amount, while at densities near the critical value,
contributions beyond the rst three layers are seen. This effect is
more pronounced in macroporous solids, where layers that are far
away from the walls contribute to the excess adsorbed amount
when the critical density is approached. Fischer and de Gennes
have shown that the excess adsorbed amount on a one dimensional
adsorbent indeed approaches innity as the critical density and
temperature are approached [46]. This phenomenon called criti-
cal adsorption has been the focus of several works in the literature
[4749].
Traditionally, absolute amounts are used to describe adsorption
equilibria in gas and liquid chromatography. From Eq. (3) it can
be seen that in order to convert the excess amounts into absolute
amounts, the volume of the adsorbed phase should be known. For
microporous sorbents this can be approximated from the slope of
the descending part of the excess isotherm [50]. For meso- and
macroporous sorbents this is not straightforward and it is indeed
preferable to model the process using excess quantities [51].
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249 231

8
T/Tc
1.025
1.053
1.104
6
1.150
T/Tc 1.209
7 1.269

n [mmol/g]
1.00246
1.0133
1.366
6
1.532
1.06 4
n [mmol/g]

5 1.17

ex
1.282
4
ex

3 2
2

1
(a) (b)
0
0.5 1 1.5 2
0.0 0.5 1.0 1.5 2.0
b
/
c b
/
c

Fig. 4. Experimentally measured excess adsorption isotherms of CO2 on (a) 13X zeolites and (b) silica gel [44].

The addition of a second component to the uid phase has
interesting consequences. In addition to affecting the uid phase
properties, as discussed earlier, it also competes for adsorption
sites on the solid surface. Some pioneering work in this direc-
tion to study modier adsorption in SFC were performed in the
1990s by the groups of Parcher and Smith using chromatographic
methods [5255]. These studies revealed that the modier does
indeed competitively adsorb with CO2 . Recent work in the area
of binary, high-pressure adsorption is focussed on the study of
the competitive adsorption of methane with CO2 in the context
of CO2 sequestration with the possibility of recovering methane
from unmineable coal seams [5659]. After the initial work there
has been a dearth of such studies related to SFC. Although, per-
forming binary or multi-component adsorption measurements can
be laborious they are expected to offer a wealth of information
concerning the complex interactions between sorbate and sor-
bent at high pressures. From the perspective of process design, SFC
experiments and models do not account for the complex physics at
the uidsolid interface and account for the competitive adsorption
of CO2 and the modier only in an empirical manner.
2.3. Key differences between HPLC and SFC
While the general method of operation of SFC is similar to that
of HPLC, there are some key differences that need to be understood
to make the transition from HPLC to SFC. The main differences are
caused primarily due to the compressibility of the mobile phase.
We discuss below some issues that are specic to SFC and highlight
them using examples form the literature.
2.3.1. Inuence of pressure on retention
The compressibility of the mobile phase plays an important role
in determining the retention characteristics of solutes in SFC. This
is an aspect that clearly differentiates SFC from HPLC and deserves

T/Tc lat 1lat lat

13

2 2 2

1.5 1.5 1.5

1.2 1 1 1

0.5 0.5 0.5

0.25 0.5 0.75 1 0.25 0.5 0.75 1 0.25 0.5 0.75 1

6 6 6
5 5 5
4 4 4
1.0001 3 3 3
2 2 2
1 1 1
0.25 0.5 0.75 1 0.25 0.5 0.75 1 0.25 0.5 0.75 1

Micropore Mesopore Macropore

(3 layers) (20 layers) (200 layers)
Fig. 5. Lattice-gas predictions of excess adsorption at supercritical, T/Tc = 1.2, and near-critical, T/Tc = 1.001, temperatures for different pore sizes [44]. lat denotes the total
excess inside the pores, whereas 1lat and 13
lat
represent the contributions to the excess of the rst and the rst three uid layers, respectively, adjacent to the sorbent walls.
In the lattice gas model  = 1 corresponds to the condensed phase density while  = 0.5 corresponds to the critical density of the uid phase.
232 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249

(a) cm=2.4% 15 (b)

15 cm=4.7%
cm=7.1%
cm=9.5% 10

10
H2

H2
5

cm=2.4%
5 cm=4.7%
cm=7.1%
cm=9.5%

120 140 160 180 200 220 700 750 800 850 900 950 1000
Pressure [bar] Density [g/L]

Fig. 6. Experimental (symbols) and tted (lines) values of the Henry constant of 1-phenyl-1-propanol on Chiralcel-OD as a function of (a) pressure and (b) density [60]. The
mobile phase consisted of CO2 + methanol. The closed symbols and continuous lines represent measurements at 30 C while the open symbols and the dotted lines represent
measurements at 40 C.

attention. At a given temperature, the partitioning of the solute
is inuenced by the solvent strength of the mobile phase, which
in-turn is a function of the pressure. In general, higher the pres-
sure, higher the solvent strength of the mobile phase and hence
weaker the retention. At dilute solute concentrations, the adsorp-
tion isotherm can be written as
ni = Hi ci
where ci and ni represent the uid phase and the corresponding
equilibrium loading on the solid phase respectively and Hi is the
Henry constant. The Henry constant can be measured experimen-
tally by injecting a dilute pulse and determining its retention time
tR,i . The retention time is related to the Henry constant as:
 
L 1
tR,i = 1+ Hi
v
where L, v and refer to the column length, the interstitial velocity
and the column voidage respectively. Fig. 6(a) shows the Henry con-
stants of 1-phenyl-1-propanol on Chiralcel-OD as a function of uid
phase pressure at two different temperatures [60]. These experi-
ments were performed at conditions where the density gradient
across the column was less than 3%. The plot shows that at a xed
temperature, the Henry constants decreases with increasing pres-
sure. Depending on the pressure, the plot also shows that the Henry
constant can increase or decrease with increase in temperature.
This may be counter-intuitive and opposite to the trends observed
in HPLC and gas chromatography (GC), where the Henry constant
strictly decreases with increase in temperature. This aspect can be
resolved when the Henry constants are plotted as a function of
density as shown in Fig. 6(b). When the density is considered as
the primary intensive variable, the expected trend of the Henry
constant is obtained, i.e., at a given density the Henry constant
decreases with increasing temperature. The reason for this behav-
ior is clear as the primary intensive variable that controls the
properties of a supercritical uid is the density and not the pres-
sure. Similar effects have been reported for the retention of alkanes
in SFC [61]. In addition to pressure, the composition of the modier
has signicant effect on the retention. The addition of the modi-
er has two effects. As described earlier, by increasing the polarity
of the mobile phase its solvent strength is increased. Further, the
modier can competitively adsorb on the stationary phase. Both
these will cause a reduction in the Henry constant with increasing
modier composition as seen in Fig. 6. These results indicate that
the density, rather than pressure, is the primary intensive variable.
However, the understanding of SFC is hindered by the simple fact
that most commercial SFC units measure pressure and temperature
and there is typically no direct measurement of density. While part
of the reason could be that the costs of adding a probe can be high,
the addition of even a low-cost, approximate density measurement
(4)
device will enable users to understand SFC in a better way.
Different empirical models have been used to describe the effect
of density and the modier composition in the Henry constant.
van Wasen and Schneider suggested the following relationship to
describe the effect of density on the Henry constant [61]
 bi
0
Hi = Hi0 (6)
(5) 
where Hi0 is the Henry constant at a reference density, 0 , and bi
is an experimentally regressed parameter. Later Lochmuller and
Mink suggested the following empirical relationship to describe the
dependence of the Henry constant on the modier composition, cm
[62]:
1
= ai cm + di (7)
Hi
where ai and di are experimentally regressed empirical parameters.
Rajendran et al., combined the two effects to suggest an empirical
correlation containing four parameters, that describes the relation-
ship between the Henry constant and the uid phase density and
modier composition [63]:
 pi cm +qi
1 0
Hi = (8)
ai cm + di 
The above equation was successfully used to describe the Henry
constants of the enantiomers of 1-phenyl-1-propanol on Chiralcel-
OD [63] and urbiprofen on Chiralpak-ADH [64] columns.
2.3.2. Compressibility effects in SFC
The compressibility of the uid phase plays an important role
in determining SFC column dynamics. An important advantage of
using SFC stems from the fact that the viscosity of the uid is
lower compared to liquids. This immediately opens up the pos-
sibility of operating the chromatographic unit at higher ow rates
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249 233

3
BP =130bar
2 BP =150bar BP = 130 bar
65C BP =180bar 2
BP = 150 bar
BP =210bar BP = 180 bar
1000 1000 BP = 210 bar
Retention time [s]

8
8

Retention time [s]

7 6
6
5 4
4

3 2

2
100
8
6
100
8 4
7
6 (a) (b)
5 2
2 3 4 5 6 7 8 2 3 4 5 6 7 8
3 4 5 6 7 8 2 3 4 5 6 7 8
0.001 0.01 0.1
0.01 0.1
Mass flow rate [g/s] Mass flow rate [g/s]

Fig. 7. Experimentally measured (symbols) and calculated (lines) retention times of phenanthrene on LiChrospher RP-18 at 65 C with (a) pure CO2 as the mobile phase and
(b) CO2 + 7% ethanol as the mobile phase [68,69]. In both gures the closed symbols represent runs where the density difference between column inlet and outlet was less
than 3%. The dashed line in (a) represents the expected trend at 130 bar, if the mobile phase were to have been incompressible.

and for using smaller particles for stationary phases. The effect of
using high owrates in HPLC is well studied [6567]. In HPLC, at
increased ow rates, the friction between the mobile and the sta-
tionary phases generates heat and if this heat cannot be dissipated
through the column walls fast enough, a radial and axial ther-
mal gradient is formed that can seriously hamper chromatographic
retention and resolution. These effects become more pronounced at
smaller particle sizes and increased mobile phase ow rates. Mobile
phases in HPLC are relatively incompressible and the only effect of
high ow rate operation is that of viscous heating. However, in the
case of SFC, the uid phase is compressible, especially when the
modier composition is low and the temperatures are close to the
critical point. The ow of a compressible uid in a packed column
resulting in non-negligible pressure drop has two important conse-
quences. Firstly, it causes a velocity and density gradient. Secondly,
the expansion of the uid causes additional thermal effects. Both
these effects are unique to SFC and are not seen in both HPLC or GC.
Let us rst discuss the formation of velocity and retention gradi-
ents along the column. The conservation of mass requires that the
mass ow rate, G, to be constant across the column. In other words:
inlet vinlet = outlet voutlet
From the above equation it can be seen that a gradient in density
along the column translates into a increase in the interstitial veloc-
ity from the column inlet to the outlet. Further, from Eq. (6) it can
be readily inferred that owing to a negative gradient in density, the
Henry constant increases along the column. Hence, in the presence
of a pressure gradient, the retention time of the solute is given by:
 L  
1 1
tR,i = 1+ Hi (z)
0
v(z)
It is worth noting that v(z) and H(z) can be evaluated once (z) is
evaluated by solving Darcys law
dP (z)v(z)(z)
= (11)
dz (z)
along with a suitable equation of state. In the above equation is
treated as a tting parameter and  is the mobile phase viscosity.
The effect of pressure drop on retention is described in Fig. 7, where
the measured and calculated retention times of phenanthrene on
Lichrospher RP-18 is shown along with the calculated values. Let
us consider Fig. 7(a), where the mobile phase was pure CO2 and
the data is reported at four different back pressures [68]. It is clear
that of the four pressure levels reported, the mobile phase is most
compressible at 130 bar. For this pressure, the expected trend of
the variation of retention time as a function of the mass ow rate,
if the uid was incompressible, is shown as the dashed line. In
other words, the retention time is expected to vary linearly with
the mass ow rate with the slope being proportional to the Henry
constant. However, from the gure it is clear that at high ow rates
the deviation from the expected behavior is signicant and they can
be calculated only by explicitly modeling the density gradient and
the corresponding variation of retention characteristics, i.e., using
Eq. (10) together with Eq. (11). Further, it can also be seen that
the measured retention times are lower than what is expected for
the incompressible case. This can be rationalized by the fact that
the experiments were performed at a xed back pressure and an
increasing pressure drop meant that the inlet pressure was higher
at larger ow rates. Higher inlet pressures correspond to lower inlet
velocities, and higher Henry constants and in this case the effect of
density on the Henry constant dominates leading to lower reten-
(9) tion times. It can also be seen that the retention times become more
linear with respect to mass ow rate at higher back pressures, i.e., at
conditions where the uid compressibility is lower. Fig. 7(b) shows
similar measurements performed with a mobile phase that con-
tained CO2 + 7% ethanol [69]. The addition of a modier reduces
the uid compressibility and this is nicely reected in the relation-
ship between the retention time and the mass ow rate which are
almost linear.
Now, let us consider the second effect of pressure drop, i.e., the
(10) associated heat effects. When a compressible uid ows through
a packed column at high ow rates the friction between the uid
and the stationary phase causes viscous heating. In addition, the
expansion of the mobile phase causes adiabatic cooling. In prac-
tice, one of the two effects dominate and impact retention and
resolution. Several authors studied this effect both experimentally
[6876] and theoretically [77,78]. Some of the very interesting work
in the recent times comes from the group of Donald Poe. Fig. 8
shows the separation of four alkanes on 3 m Spherisorb C-8 col-
umn using pure CO2 as the mobile phase at two reduced densities
/c = 1 and 1.5 [76]. Fig. 8 (top) shows a chromatogram measured
234 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249

dphplc98,channel1.dp9-75-2,8,3
Response(mV)
Inject

30 C1
A. Air 1.0
25

20 C18

15

10

0 10 20 30 40 50 60 70
Time(seconds)
dphplc98,channel1.dp9-76-1,8,1
Response(mV)
120

100
Inject B. Ins 1.0
80
C18
60
C1
40

20

0 10 20 30 40 50 60 70
Time(seconds)
dphplc98,channel1.dp9-75-1,3,3
Response(mV)
600
C18
500

C. Air 1.5
400

300

200

100 Inject
C1

0
0 10 20 30 40 50 60 70
Time(seconds)

Fig. 8. Effect of thermal conditions on experimentally measured chromatograms of four alkanes on 3 m, Spherisorb C8 column. (Top) Column placed in an air oven and
uid phase reduced density of /c = 1. (Middle) Column insulated and uid phase reduced density of /c = 1, (Bottom) column placed in an air oven and uid phase reduced
density of /c = 1.5. Reprinted from [76], with permission of Elsevier.

at /c = 1 with the column placed in an air oven. In this case the
column wall is kept at a constant temperature, while the middle of
the column is cooler owing to the expansion of the mobile phase.
This situation leads to the formation of a radial temperature gra-
dient and its adverse effects on the efciency can be seen in the
highly distorted peaks. The authors repeated the experiment but
by adding a layer of insulation around the column, in which case
the column approaches the adiabatic limit and the radial tempera-
ture gradient is minimized. As seen in Fig. 8 (middle), this improves
the band proles signicantly. In the nal experiment (Fig. 8 (bot-
tom)), the density of the mobile phase was increased to /c = 1.5,
a condition where the uid compressibility is reduced. Under these
conditions, no signicant distortion is seen even if the column was
not insulated. These experiments clearly demonstrate that thermal
effects play an important role in determining separation efciency.
In the same work the authors measured the exit temperature
which revealed that gradients of upto 15 C were observed clearly
demonstrating that adiabatic cooling was the dominant heat effect.
The measurement of HETP at higher ow rates was studied using
both experiment and theory for the system phenanthrene- LiChro-
spher RP-18 [68]. Fig. 9(a) shows the HETP measured using pure
CO2 as the mobile phase while Fig. 9(b) shows the results when 7%
ethanol was added as a co-solvent. In this study, the variation of the
HETP along the column was accounted for and the apparent HETP
was calculated by accounting for the local variations. It was found
that the calculations showed good match with the experiments at
high pressures when pure CO2 was used as a mobile phase. How-
ever, at 130 bar, i.e., when the uid phase was highly compressible
the results showed signicant deviations from the calculated values
(see inset of Fig. 9(a)). The systematic deviations were speculated to
be caused due to the thermal effects although they were not exper-
imentally conrmed. It is worth noting that such deviations were
not observed when a modier was added (see inset of Fig. 9(b)), i.e.,
when the uid compressibility was lower.
Recently, Kaczmarski et al. have modeled the thermal effects
in SFC and have shown that signicant radial gradients can be
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249 235

80

60
m]
HET P [

40

20

(a) (b)
0
0.00 0.02 0.04 0.06 0.08 0.10
Mass flow rate [g/s]

Fig. 9. Experimentally measured (symbols) and calculated (lines) HETP vaules of phenanthrene on LiChrospher RP-18 at 65 C with (a) pure CO2 as the mobile phase and (b)
CO2 + 7% ethanol as a the mobile phase [68,69]. The insets show the measured and calculated HETPs at 130 bar. In the inset of (a) notice the signicant deviation of the HETP
from the calculated values when pure CO2 is used as a mobile phase. Legend: () 130 bar; () 150 bar; () 180 bar; () 210 bar.

formed at high ow rates and have shown that under traditional
operating conditions, the adiabatic expansion seems to dominate
[77,78]. However, they also show from calculations that there exists
conditions, although far beyond conventional operational regions,
where the viscous heating is negated by the adiabatic expansion.
Recent works by Tarafder and Guiochon have identied regions on
the isopynic plot, as shown in Fig. 10, where serious loss in col-
umn performance can be observed [79]. It is interesting to note
that this region corresponds to operating conditions where the
uid compressibility is high. While this region provides a wealth
of opportunity for fundamental studies, from an operational per-
spective it is advisable to avoid operating the unit in the vicinity.
Pressure drop has important impact on prep-SFC. Firstly, dur-
ing the measurement of adsorption isotherm and mass transfer
parameters, it is important to ensure that the characterization is
performed at low ow rates where the pressure drop is minimal

Fig. 10. Isopynic plot for pure CO2 along with regions where signicant loss of efciency were observed. Reprinted from [79], with permission from Elsevier.
236 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249

Modifier

Modifier
To column To column

Injection Injection
valve valve
CO2

CO2
Mixed-stream injection Modifier-stream injection
Fig. 11. Type of injections used in prep-SFC.

and hence the retention time and adsorption isotherm parameters front does not exit the modier band. This situation causes distorted
can be directly correlated to the specic operating conditions, e.g., elution proles. The issue of large-volume injections will become
pressure and temperature. Secondly, when attempting to predict important when performing isotherm measurement experiments
retention times and resolutions at higher ow rates, these effects and when performing preparative separations. Further, the solu-
should be explicitly considered if accurate estimates are required. bility of the solute in the modier is typically higher than that in
Most practical separations involve addition of a modier which sig-
nicantly reduces the mobile phase compressibility. Under these
situations most of the effects described above do not occur and the 4
description of column dynamics can be simplied.
3
CO2+ mod
2.3.3. Large-volume injections in SFC Time
In the case of HPLC, the solute to be separated is typically dis- 2
solved in the mobile phase for injection. However, in SFC it is not
possible to prepare a batch of a feed by dissolving the solute in the 1
mobile phase, i.e., CO2 + modier. In order to overcome this prac-
Mod CO2+ mod
tical limitation, the solute is usually dissolved in the modier for 0
0.0 0.2 0.4 0.6 0.8 1.0
injection. The typical injection methodology in HPLC, analytical and Length
some prep-SFC devices is shown in Fig. 11(a). This method often
referred to as a mixed-stream injection involves mixing the CO2 (a) Small-volume injection
and the modier upstream of the injection valve and making an 4
injection into this mixed stream. While this injection is routinely
performed in analytical scale, it can have adverse effects on peak 3
CO2+ mod
shapes especially if the injection volume is large [80]. It is worth
Time

noting that for the duration of the injection, the solute is present in 2
the modier plug from the injection loop. Once injected, the solute
and the modier plugs migrate along the column. It is worth not- 1
ing that based on the relative retention characteristics between the
solute and the modier, three situations can arise: Mod CO2+ mod
0
0.0 0.2 0.4 0.6 0.8 1.0
Length
1. Case 1: The solute is strongly retained compared to the modier.
2. Case 2: The solute and modier have identical retention charac- (b) Medium-volume injection
teristics. 4
3. Case 3: The solute is weakly retained compared to the modier.
3
CO2+ mod
Let us just consider the case where Hmod < Hsolute,mod < Hsolute,mix ,
Time

where the subscripts mod and mix refer to the Henry constants 2
in the modier and in the mixed phase respectively. The migration
of the pulse under these circumstances is shown in Fig. 12 for very 1
dilute injections and for varying injection volumes [81]. These g-
ures show the trajectories of the three bands, viz., modier (green), Mod CO2+ mod
0
less retained solute (blue) and more retained solute (red) on the 0.0 0.2 0.4 0.6 0.8 1.0
physical plane and the corresponding elution proles that will be Length
measured at the column outlet. These proles are calculated using
(c) Large-volume injection
the equilibrium theory of chromatography under the assumption
of dilute (Henrys law region) condition. For a given injection vol- Fig. 12. Inuence of injection volume on band proles in mixed stream injection.
ume, it can be seen that the tail of the solute travels predominantly The plots show the physical plane and the elution prole calculated using equilib-
in the mixed phase while the front travels for a certain distance in rium theory of chromatography. The trajectories of the modier band (green), less
the pure modier band. Since the retention of the modier band is retained component (blue) and more retained component (red) are shown. The cal-
culated elution prole at the column outlet is plotted alongside the physical plane.
weaker compared to the solutes, the band prole measured at the
Notice the deformation of elution proles as the injection volume is increased. (For
column outlet is broadened compared to the injected pulse. Further interpretation of the references to color in this gure legend, the reader is referred
when the volume of injection is large enough, it can be seen that the to the web version of the article.)
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249
the mixed phase, and hence when the solute band migrates from
the modier plug into the mixed phases the solute could precip-
itate leading to distorted band [80]. Alternate strategies such as
the modier-stream injection (Fig. 11(b)), wherein the injection
is made in the modier stream which is then mixed with the CO2
stream have been proposed [80].
2.3.4. Swelling of stationary phase in SFC
The addition of CO2 to the polymer lowers its glass transi-
tion temperature and transforms it into a rubbery state [82]. This
causes polymer matrices to swell in the presence of CO2 [8386]. In
systems such as CO2 -poly(methyl methacrylate), swelling of upto
25% have been observed. Springston measured the swelling of the
stationary phase in capillary SFC [87]. The authors used a methylsil-
icone stationary phase with pure CO2 and butane as mobile phases.
By measuring the retention of a non-retained marker, helium in
this case, the degree of swelling was measured. Swelling factors of
upto 1% and 4.8% were measured for CO2 and butane respectively.
In another study which investigated the separation of the enan-
tiomers of 1-phenyl-1-propanol,1,3,5-tri-tert-butylbenzene (TTBB)
was used as a non-adsorbing species to measure the bed voidage
of the column [60]. In these experiments a dilute pulse of TTBB
was injected and its retention time was measured. By assuming
that TTBB does not adsorb, the bed voidage, , was calculated. The
measured varied between 0.728 and 0.86. In general the void frac-
tions were larger at lower pressures. It was surmised that this can
be caused due to either the adsorption of TTBB; or the swelling of
the stationary phase. In that study the lowest measured value of ,
measured at the high pressure and modier composition, was used
for further characterization. There are no detailed studies to explore
the issue of stationary phase swelling and it is important to under-
stand this effect. In practice, if the adsorption isotherm parameters
are measured using chromatographic methods, it might be difcult
to discern the effect of swelling and retention unless other inde-
pendent measurements, either of swelling or the adsorption, can
be made.
3. Prep-SFC operating principles and the case for
optimization
The design of a preparative separation involves the following
steps: screening for a suitable stationary and mobile phase combi-
nation; detailed characterization of the separation; determination
of scale-up conditions; and implementation of the separation. The
rst step, i.e., screening, is performed to identify the best combi-
nation of stationary and mobile phases that will lead to a desired
separation. This step is typically carried out in an automated fash-
ion where a series of analytical columns and modiers are chosen
and various combinations are used and an injection is made. Key
properties that are evaluated are the retention characteristics, the
selectivity and resolution. Once reasonable values of selectivity and
resolution are obtained, a stationary phase-mobile phase combina-
tion is chosen for detailed characterization.
3.1. Traditional scale-up approaches
In traditional approaches, the detailed characterization is per-
formed through a series of loading studies. In these studies, a series
of overloaded pulses are injected in to the column and the response
is measured. Two kinds of loading studies are typically performed:
volume and mass overloading. In volume overloading, the con-
centration of the injected sample is kept xed while the injection
volume is varied. In mass overloading, the volume of injection is
xed and the injection concentration is varied. These injections are
performed until the two peaks are no longer base-line resolved. The
237
Fig. 13. Example of a scale-up of single column liquid chromatography based
on experiments at analytical scale. The chromatograms show that a million-fold
scale-up can be performed reliably in chromatography. Reprinted from [88], with
permission from American Chemical Society.
injection volumes and concentrations at which baseline separation
is achieved are noted and the scale-up calculations are performed.
The rule of thumb is to maintain the same injection concen-
tration, stationary phase particle size and the interstitial velocity
both in analytical and preparative units. Identical velocities can be
maintained by ensuring that the following relationship is satised
 2
Qprep dprep
= (12)
Qanalyt danalyt
where Q and d denote the ow rates of the mobile phase and diam-
eter of the columns respectively. Once these are xed, the only
remaining operating parameter is the injection volume, Vinj . This
can be calculated by
 2
Vinj,prep dprep
= (13)
Vinj,analyt danalyt
In most cases, provided the packing of the column is uniform,
a reasonably good rst injection can be made and the operating
can be ne-tuned for large-scale separation by using stacked injec-
tion. Fig. 13 shows such an approach, although from HPLC, where
a million-fold scale-up was performed from a 300 m to a 30 cm
column [88].
3.2. Framework for rational design
The scale-up method described above is simple and rather
straightforward. However, it is very often the case that such a
scale-up strategy is sub-optimal as it does not explore the entire
operating domain, i.e., combination of operation variables, within
which the separation is feasible and hence may not exploit the
fullest potential of the system. Further, in practical situations, the
specic objectives in a given separation may vary. For example,
in shorter campaigns, speed could be the key objective while for
larger campaigns or in the manufacturing stage, minimizing costs,
e.g., reducing solvent consumption, may be the key objective. Very
often these objectives cannot be simultaneously achieved, i.e., the
achievement of one is accompanied by the loss of the other(s).
Hence, it is important to quantify the trade-off between these
objectives in order to evaluate the economic potential and the
versatility of the process. Once these trade-offs are quantied,
depending on the operational requirements, it will be necessary to
determine the operating conditions that will guarantee the desired
outcome. Identifying these trade-offs is essential not only to eval-
uate its potential but also to compare the effectiveness of different
process options, e.g., SFC vs. HPLC, SFC vs. SMB, etc. It should be
238
Fig. 14. Framework for rational design and optimization of prep-SFC separations.
noted that comparisons between processes are meaningful only if
all the processes under consideration are optimized allowing them
to exploit their fullest potential.
We believe that since traditional scale-up approaches do not
sufciently explore the operating domain and do not account for
the complex interactions of the operating variables, they cannot
guarantee optimal operating conditions and hence do not exploit
the full potential of a process. Hence, there is a need to develop a
rational scale-up strategy that can allow the user to quantify the
performance limits of the process and provide a combination of
operating conditions that will guarantee the achievement of certain
objectives. Needless to say that such an approach should not be time
consuming, minimize trial and error and be general in its approach
so that it can be incorporated in routine operations. Further, very
often in practical situations the amount of material available for
characterization is low and no pure compounds are available.
Hence, the framework should be able to handle such situations.
Fig. 14 provides a framework for rational scale-up of SFC.
It consists of three modules: system characterization; in-silico
design; and scale-up. The rst module involves the measurement
of important physical properties, such as adsorption isotherms,
mass transfer kinetics, pressure drop parameters and solubility. All
these characterization can be performed on analytical-scale units,
thereby minimizing the usage of solutes and solvents. The second
module involves the design and optimization of the process on a
computer. This module consists of a process simulator that is capa-
ble of calculating band proles and an optimization algorithm that
can interact with the process simulator to identify operating condi-
tions that optimize certain objective functions. It is also important
that the user is able to incorporate physical and operational
2.0
(a)
R
Concentration [g L ]
-1
1.5
1.0
0.5
0.0
400
Time [s]
Fig. 15. Experimentally measured (symbols) and calculated (lines) elution proles of the enantiomers of urbiprofen on Chiralpak ADH [64]. Pressure: 100 bar, mobile phase
CO2 + 13% (w/w) methanol. (a) Pure component injection and (b) racemic injection.
A. Rajendran / J. Chromatogr. A 1250 (2012) 227249
constraints that should not be violated. The result of the design
and optimization are trade-off plots and operating conditions that
can be translated to a preparative separation (Figs. 15 and 16).
It is worth noting that this approach has been proposed and
investigated in the academic literature [8997]. With advances
in computational power, development of numerical methods that
allow for rapid calculation of band proles and robust optimization
algorithms, these tools can be implemented on a desktop computer
and results can be achieved in a short time (1 day). Further, initia-
tives such as the quality by design (QbD) from the US Food and Drug
Administration (USFDA), encourage practitioners to adopt more
model-based tools into the manufacturing domain [98,99]. In addi-
tion to attaining obvious process objectives such as reducing costs,
environmental impact and time for scale-up, these tools provide
valuable insights and help us develop a deeper understanding of
the complex interactions between the operating parameters and
the process performance. Considering all these factors, we believe
that the time is appropriate for these (academic) concepts to be
incorporated into industrial manufacturing. The next three sections
describe the different steps involved in the framework and shows
examples from the literature. We will particularly focus on the situ-
ation where scale-up needs to be performed with minimal material
that are often available only in the form of mixtures.
4. Process modeling and simulation
4.1. Detailed modeling
The detailed modeling of SFC column dynamics consists of the
following: overall mass balance; component balances for each
2.0
(b)
300 g/L
Concentration [g L ]
-1
1.5
S 1.0
150 g/L
0.5
75 g/L
0.0
600 800 400 600 800
Time [s]
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249 239

8 30

S1

S2
R enantiomer
S enantiomer
25
6
20

ni [g/L]
ci [g/L] 4 15

10
2
5
(a) (b)
0 0
1.0 2.0 3.0 4.0 0 1 2 3 4 5 6 7 8
Dimensionless time, [-] ci [g/L]

Fig. 16. Estimation of isotherms using the binary retention time method [126]. (a) Estimation of retention times of the discontinuties and (b) comparison of isotherm
calculated by BRTM (lines) and inverse method (symbols).

solute, and the modier; adsorption isotherms for all components Under the assumptions described above, the model describing
including CO2 and modier; appropriate mass transfer kinetics; the transport of a solute reduces to:
energy balance; pressure-drop correlations and equations of state. Component mass balance
However, there has been no attempt in the literature to solve such 2
a detailed model. Part of the limitations arise from the measure- ci ci c 1 ni
= Dax,i v i (14)
ment of adsorption isotherms of CO2 , modier and to describe their t z 2 z t
competition with the solutes. Owing to the compressibility of the Solid phase mass balance
mobile phase, describing the thermal effects, as discussed earlier,
dni
is difcult. Finally, the equations are highly non-linear and solving = kf,i (ni ni ) (15)
them can be computationally expensive. In practice, the model is dt
often simplied using the following assumptions: Adsorption isotherm

The adsorption of CO2 and modier is not explicitly modeled: As dis-
cussed earlier, at high pressures, CO2 shows strong adsorption
on solid surfaces. Further there are reports that conrm the com-
petitive adsorption of the modier. These measurements, while
offering a wealth of information about the microscopic picture
of adsorption, are laborious and may not be well suited when
preparative separations are to be designed within a short dura-
tion. Hence, for the purpose of process design, it is quite common
to ignore these effects when measuring the equilibrium and mass
transfer parameters of the solutes. Note that the effects of CO2 and
modier adsorption, in fact, manifest themselves when tting
models to experimental results. Hence, interpreting the thermo-
dynamic meaning of these parameters should be performed with
care [100].
The compressibility of the mobile phase is ignored: Under practi-
cal conditions, especially when the amount of modier in the
mobile phase is high, the condition of negligible compressibil-
ity is approached. Under this assumption, the pressure drop and
energy balance equations can be omitted and the SFC model equa-
tions resemble those used for HPLC. At this juncture it should be
pointed out that if the operating conditions are indeed within the
region where the uid is compressible, the pressure drop and the
energy balance equations should be included together with an
appropriate equation of state.
Simplied description of mass transfer: For SFC separations, it has
become quite common to use stationary phases smaller than
10 m. The use of small particles and the excellent diffusion
properties of the supercritical mobile phase favour rapid mass
transfer. Under these conditions, lumping all mass transfer resis-
tances into an effective mass transfer coefcient can provide a
practical description. It is worth noting that studies that used this
approach have consistently observed that the tted mass transfer
coefcient have values larger than 1 s1 [64,100].
ni = f (ci ) (16)
where Dax,i kf,i are the axial dispersion coefcient and mass transfer
coefcient respectively.
At this point is worth emphasizing that the aim of making the
assumptions is to reduce the complexity and to arrive a set of equa-
tions that can be solved with reasonable computational effort so
that process optimization can be performed. However, if the goal of
the study is to understand the fundamentals the above assumptions
have to be carefully evaluated.
4.2. Numerical solution of chromatographic models
The numerical solution of chromatographic models described
in Eqs. (14)(16) have been an important area of study. The nature
of chromatographic equations is such that upon injection of high
concentrations, that correspond to the non-linear region of the
isotherm, discontinuities in concentration are formed and prop-
agate through the column. Although the nite rate of mass transfer
and dispersion effects smoothen these discontinuities, the numer-
ical technique used to calculate the band proles should be stable
in handling these proles. In SFC, owing to the use of small parti-
cles and a supercritical mobile phase, both of which lead to reduced
band broadening, the sharp discontinuities are less degenerated as
compared to HPLC and hence there is a need to use more robust
numerical techniques.
Several numerical techniques have been described in the lit-
erature [101,102]. These include, nite difference, nite element
and nite volume. In nite difference, the differential terms in
Eq. (14) are replaced by difference terms. Both semi- and fully
discrete schemes have been described in the literature. In semi-
discrete schemes, one of the two coordinates, i.e., space or time,
is discretized and the PDEs are converted to ordinary differen-
tial equations (ODEs) and a suitable solver is used for integration
240 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249
[103105]. In fully discrete schemes, both time and space coordi-
nates are discretized and the resulting set of algebraic equations
are solved to calculate the band proles. While nite difference
schemes are easy to implement, they are inherently dispersive
and offer a lower order of convergence. In nite element schemes,
Eq. (14) is considered in the differential form and the solution to
the equation is assumed, e.g., using a polynomial. Then, the form
of this solution is adjusted, by solving an optimization problem,
such that it satises the differential equation. While nite element
schemes offer high accuracy, they consume signicant computa-
tional resources [106].
Finite volume techniques, which are more recent developments,
are being increasingly used in the solution of conservation laws
[107]. In this technique, Eq. (14) is written in the integral, instead of
the differential, form. The length axis is discretized into nite num-
ber of volumes and the average concentration at the centre of the
volume is calculated. Convection and dispersion give rise to mass
uxes between two adjacent volumes and hence they have to be
calculated at the boundaries. If the concentration associated with
the ux is assumed to be the same as that of the cell volume to the
left, then the technique reduces to a nite difference scheme. How-
ever, if an appropriate interpolation formula is used to calculate the
uxes, higher order convergences can be obtained. The choice of
the interpolation formula is evaluated based on the slope of con-
centration prole at a certain location in the column. This approach
effectively handles sharp fronts and is able to provide higher accu-
racy with lower number of grid points and hence provide higher
accuracy with reduced computational effort [108,109].
4.3. Equilibrium theory of chromatography
When the mass transfer between the uid and solid phase is fast
enough, the condition of local equilibrium can be assumed. Under
such conditions, the main band broadening contribution comes
from axial dispersion. Contributions to axial dispersion arise from
molecular diffusion and eddy diffusion (a function of the stationary
phase particle size). In prep-SFC since small size particles are used,
the contributions can be neglected for preliminary analysis. Under
these conditions the mass balance is given by
ci c 1 ni
= v i (17)
t z t injected into another mobile phase can precipitate owing either to
Although the above equation is highly idealized, it is amenable
to analytical solutions, for a certain class of isotherms, through a
rather elegant theory [110116]. The advantage of using equilib-
rium theory is that it can yield very quick estimates of productivities
and solvent consumption that are achievable when the process
is scaled-up. Results from our laboratory indicate that the band
proles predicted by equilibrium theory are very good representa-
tions of experimentally measured SFC elution proles. SFC certainly
opens up a new area for the deployment of equilibrium theory,
although such examples are rare.
5. System characterization
5.1. Adsorption isotherms: measurement and characterization
The key thermodynamic property that describes retention of the
solute is the adsorption isotherm. The isotherm describes the dis-
tribution of the solute between the mobile and stationary phases.
The partitioning depends on the chemical potential of the com-
ponent in the two phases. In the case of SFC, the mobile phase
consists of the sc-CO2 , modier and the mixture of solutes. At con-
stant temperature, the chemical potential of the solute is a function
of the density, temperature, the composition of the mobile phase
and the concentrations of the solutes. Since the concentration of
the solutes in the uid are typically low enough, the interactions
between them can be ignored. Under this assumption, the chemical
potential in the uid phase is determined by density, temperature
and modier composition. The picture on the stationary phase is
equally complex. The solid has a nite capacity and all the species,
i.e., CO2 , modier and the solutes compete for adsorption. In order
to describe the adsorption of the solute under these conditions, the
quantication of the solubility and the competitive adsorption is
needed. In the following we briey review the methods to mea-
sure adsorption isotherms and solubility. A detailed description
of measuring single-component isotherms is provided elsewhere
[5,117]. We focus only on techniques for the measurement of multi-
component mixtures without the need to obtaining pure solutes.
Note that in order to scan the entire operating conditions, the effect
of operating variables such as temperature, pressure and modier
composition should be studied. The specic region needed to be
explored depends on the system and the availability of material
and time.
5.2. Measurement of multi-component isotherms
Owing to the inherent difculties associated with measuring the
competitive adsorption of all the components, for practical pur-
poses, the adsorption of CO2 and modier are only considered in
an empirical way. Traditionally, methods to measure adsorption
isotherms can be classied as static and dynamic [101,5].
5.2.1. Static methods of measuring adsorption isotherms
Static methods of isotherm measurement refer to techniques in
which no ow is involved. The general procedure is to rst contact
the stationary phase and the mobile phase, add a known quan-
tity of the solute and monitor its concentration. The monitoring
can be performed either by the use destructive sampling, i.e., to
remove a small amount of the uid phase and analyze its contents
externally or by non-destructive sampling by employing an online
detector. Removing a sample from the measuring cell has to be per-
formed with caution. Firstly, care should be taken not to remove
signicant amount of material as this will disturb the uid phase
from achieving equilibrium. Secondly, the injection of the sample
into the analytical device, e.g., HPLC/SFC, has to be performed with
care. The sample which is removed is at high pressures and when
a drop in pressure or due to the reduction of solubility. Avoiding
such a precipitation is important as it might lead to distorted peaks
or in a more serious situation to the incorrect estimation of the
sample amount. These problems can be eliminated by using non-
destructive techniques, e.g., by incorporating an online detector
such as UV within the measurement system and by circulating the
uid using a pump [118]. While this may provide a good method to
obtain equilibrium data, measuring kinetics needs to be performed
with caution as dead volume effects may have signicant impact
on the reliability of the data. Baring the single advantage that it is
possible to estimate isotherm data up to high concentrations with-
out wasting signicant amount of material, static methods are in
general cumbersome and time-consuming.
5.2.2. Dynamic methods of measuring adsorption
Dynamic methods involve the use of standard HPLC and SFC
equipment for the measurement of isotherm parameters. A typical
dynamic measurement involves providing a known disturbance at
the column inlet and measuring the response of the column using
a detector. The different dynamic methods differ either in the type
of disturbance provided or the method in which the response is
analyzed. In all dynamic methods, a ow of CO2 and modier is
established and an injection is made. In the case of HPLC, since the
mobile phase is incompressible, the owrate at the column inlet
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249 241

can be safely assumed to be that at the pump head. Hence the total where S is the detector signal, is assumed. An optimization algo-
ow of the mobile phase, Qmob can be calculated as a summation of rithm is invoked to t the parameters in Eq. (19) in order to
the solvent and the co-solvent ow rates. However, in the case of minimize the error dened as
SFC, Qmob refers to the mobile phase ow rate at the column inlet
 2
and has to be calculated carefully. In SFC, the CO2 at the pump head
is kept at low temperatures and is pumped as the liquid. Later, the
modier stream is mixed and the mobile phase is brought to the
required temperature. Owing to the change of temperature and the
addition of the modier, the density of the mobile phase changes,
thereby the effective volumetric ow rate at the column inlet is
obtained by:
CO2 ,pump QCO2 ,pump + mod,pump Qmod,pump
Qmob = (18)
mob,col
where the subscripts pump and col refer to the densities calcu-
lated according to the conditions (P, T, modier composition) at the
pump head and column inlet respectively. It is worth noting that
mob,col can be calculated either by using thermodynamic models
that require appropriate mixing rules or by using experimental data
available in the literature.
5.2.3. Estimation of binary/multi-component non-linear
isotherms
Prior to the measurement of non-linear isotherms, it is always
preferable to measure the bed voidage, and the Henry constants.
The measurement of the bed voidage is performed by measuring
the retention times of a non-adsorbing tracer. In SFC, it is recom-
mended that this experiment be performed at the highest modier
composition possible in order to minimize the possible adsorption
of the tracer. The measurement of the Henry constants is achieved
by injecting a very dilute mixture of the solute in the mobile phase.
Once the retention time is measured, the Henry constant can be
calculated using Eq. (5). It is important to understand the effects
of compressibility that have been discussed earlier, and hence all
measurements should be performed at ow rates where the density
gradient across the column is minimal (<5%).
Most preparative separations involve overloading the column
with fairly high concentrations of the solute. At high concentra-
tions, owing to the limited capacity of the adsorbent, the adsorption
isotherm is non-linear. There are several methods described in the
literature to measure non-linear isotherms [101,117]. Herein we
discuss a few techniques that can be used to measure competitive
adsorption of binary/multicomponent mixtures.
5.2.3.1. Inverse methods. The inverse method is now one of the
most widely used techniques to determine competitive isotherms
from mixtures [119]. In this method, a known concentration and
volume of the solute mixture is made using an injection valve
and the chromatogram is measured. For improved reliability, other
injections either with different injection volumes or concentrations
are made. The rst step is to convert the detector response, typ-
ically recorded as a current/voltage or an equivalent signal into
concentration units. Very often the detector calibration curve, i.e.,
the relationship between the detector signal and the concentration
of the solute in the mobile phase is non-linear and hence needs to
be established. One approach would be continuously inject known
concentrations of the solute into the mobile phase, measure the
response on the plateau and obtain a suitable calibration curve.
However, in situations where sufcient amounts of solute are not
available, this approach may not be feasible. An alternate approach
to circumvent this limitation was developed by Asnin and Guio-
chon [120,121]. In this method the chromatograms obtained for
the isotherm measurements are also used for detector calibration.
As a rst step, a functional form of the calibration curve
ci = f (S) (19)
Nexp loop
ck V loop Qmob ck (S)dt
J= (20)
loop
ck V loop
k=1
In the above equation, the index, k refers to multiple injections that
can be combined to obtain a reliable calibration curve. Using the
best t parameters, Eq. (19) can be used to reconstruct the chro-
matogram in concentration units. Ottiger et al. used the following
calibration equation [100]
b2
c = b1 ln (21)
b2 S
while Chen and Rajendran [64] used the following:
 S

c(S) = b1 ln 1 + b3 S (22)
b2
The variables bi in the above equations are the tted parameters. It
is worth noting that unless the components have identical detec-
tor responses, e.g., UV signals for enantiomers, the above method
would require that each component is at least baseline resolved or
that measurements can be made also at conditions, e.g., by scanning
the signal at multiple wavelengths, where the detector response is
unique for the different components. It is worth emphasizing that
the calibration curve is dependent on factors such as pressure, tem-
perature and modier composition. Hence, if any of the operating
conditions is varied, the curve needs to recalibrated. The inverse
method relies on the fact that when the appropriate isotherm
and mass transfer models along with the correct parameters are
used, it will be possible to calculate the band proles. Hence, in
the inverse method, an isotherm and mass transfer model is rst
assumed. The form of the isotherm is indeed predicted from the
chromatograms themselves. It is worth noting that some of the
isotherm models have the ability to describe competitive effects
in a straightforward manner, e.g., competitive Langmuir, compet-
itive bi-Langmuir, generalized Langmuir etc. In cases where such
extensions are not possible, models which allow combining sin-
gle component isotherms to describe binary adsorption, such as
the ideal adsorbed solution (IAS) theory [122], are invoked. Once
the isotherm function and a framework to describe the competi-
tion is chosen, these information are incorporated into a simulation
model, given in Eqs. (14)(16), to obtain the calculated elution pro-
le. This model is then incorporated within an optimizer whose
role is to determine the value of the parameters in such a way
that the error between the calculated and measured elution pro-
les is minimized. The values of the parameters hence obtained
are considered as those describing the isotherm. The method has
many advantages. Firstly, it can be applied for systems character-
ized by any form of the isotherm. Secondly, multiple experiments
may be combined to calculate the error and hence a set of val-
ues that describe a wide range of concentrations can be obtained.
The main shortcoming of this method arises from the dilution
that the pulse undergoes as it travels through the column. Since
the measured isotherms are strictly valid only until the maxi-
mum concentration measured at the column outlet, it will not be
possible to obtain reliable parameters at high concentrations. On
the other hand, this method is ideally suited when the user has
only a small amount of material for the purpose of characteriza-
tion. Ottiger et al. used the inverse method to obtain equilibrium
isotherm parameters for the enantiomers of 1-phenyl-1-propanol
on Chiralcel-OD [100]. Chen and Rajendran obtained the isotherm
parameters of the enantiomers of urbiprofen on Chiralpak-ADH
[64].
242 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249
5.2.3.2. Perturbation technique. In this technique, the chromato-
graphic column is saturated with a known concentration by
continuously pumping the solute of the solute in the mobile phase
[123]. At a certain time, a small disturbance is introduced at the col-
umn inlet. This can be done either by injecting very small amounts
of the solute or the solvent. Since the uid phase now consists of
N + 2 number of species, where N is the number of solutes in the
mixture, the response of the column should ideally consists of N + 2
peaks. The measured retention time of each peak can then be corre-
lated to the local slope of the adsorption isotherm by the following
expression:

L 1 dni
tR,i = 1+ |c (23)
v dci i
Note that the retention times measured are indicative of the
competition and hence the single component isotherms should
be calculated by taking this into account. The group of Monika
Johannsen and Gerd Brunner has used this technique to measure
the adsorption isotherms of -tocopherol on Nucleosil [124,125].
The main advantage of the perturbation technique is that it
requires no detector calibration and its implementation is straight-
forward. Since the technique ensures that the column is saturated
with a certain concentration of the solute, it is possible to perform
the characterization at very high concentrations as there are no
dilution effect as in the case of the inverse method. The main dis-
advantage of this technique arises from the fact that signicant
amounts of the solute is required for the characterization and may
be a limitation when only small quantities of the material is avail-
able for characterization.
5.2.3.3. Binary retention time method. This method is based on
the principle that the retention time of the discontinuity formed
by injecting a high-concentration pulse can be calculated by
equilibrium theory of chromatography [126]. The type of experi-
ments required for this technique are the same as for the inverse
method. However, the only piece of information extracted from
the chromatograms the retention time of the two shock fronts. The
equilibrium theory for the case of competitive Langmuir isotherms
is used to estimate these retention times for the given injection con-
centration and volume with assumed Langmuir parameters. These
parameters are regressed until the error between the calculated and
experimental retention times is minimized. Rajendran and Chen
applied this technique successfully to estimate the isotherms of 1-
phenyl-1-propanol and showed that they were consistent with the
isotherms measured using the inverse method [64].
This method has some advantages. Firstly, no detector calibra-
tion is required as only retention times are measured. Secondly, the
retention times of the shocks can be calculated using rather simple
analytical equations and hence the time required for paramet-
ric estimation is signicantly shortened compared to solving the
detailed model numerically. The main shortcoming of the method
is that equilibrium is assumed to be described by the competitive
Langmuir isotherm and hence applying this to non-Langmuirian
systems will result in ctitious parameters. Further, if there are
signicant mass transfer effects, the methods can lead to erro-
neous parameters. However, this method can be used to obtain
good initial guess for more rigorous methods such as the inverse
method.
5.3. Measurement of pressure-drop
Operating the unit at pressure drops larger than the recom-
mended limit (typically provided by the column manufacturer)
can lead to damaging the stationary phase and result in undesir-
able band-broadening. Hence, prior to the separation, it is worth
characterizing the pressure drop of columns. Note that pressure
drop along the column is given by Darcys equation, i.e., Eq. (11). The
objective of performing an experiment is to determine the value of
. Note that it is, in principle, sufcient to perform one experiment
at a moderate ow rate to estimate the permeability for a given col-
umn. Once this is obtained the pressure drop at other conditions can
be estimated. Further, these experiments can be performed without
the solute and it would sufce that the experiment is repeated only
when it is suspected that the stationary phase has been signicantly
altered.
6. Optimization of chromatographic separations
6.1. Problem denition
In the previous sections we have described the techniques to
obtain parameters required for numerical simulation of elution
proles. In this section we consider the optimal design of prep-SFC.
Although the discussion here will specically address prep-SFC, the
concepts are generally applicable also to other forms of chromatog-
raphy.
6.1.1. Objective functions
In general an optimization problem is comprised of objective
functions, decision variables and constraints. Objective func-
tions are performance indicators of the process that need to be
maximized or minimized. Based on the number of objectives,
optimization problems can be classied as single-objective and
multi-objective. In single-objective problems, the goal is to nd a
combination of decision variables (operating conditions) that maxi-
mize or minimize a certain objective, e.g., overall cost of the process.
In multi-objective problems, the goal is to maximize or minimize
multiple objectives simultaneously. In practice it is quite common
to expect that a process satises multiple objectives. In some cases
these objective functions may be complimenting each other, i.e.,
the betterment of one is accompanied by the betterment of the
other. Under these situations one can envisage a unique operat-
ing condition at which the multiple objectives are maximized (or
minimized). However, in most practical situations, the objectives
are conicting, i.e., the improvement of one is accompanied by the
deterioration of the other. Hence, the solution obtained is not a
unique point but a combination of non-dominated points. This set,
which is called the Pareto curve, given the best trade-off between
the two (or more) objective functions.
It is indeed possible to combine different objectives, e.g., pro-
ductivity and solvent-consumption, into the cost of separation
[1921]. This would mean that specic costs be assigned to each of
the performance indicator. For example costing related to solvent
consumption could involve the costs of energy required for concen-
trating the product, the amount of make-up solvent used etc. On the
other hand, the costs associated with productivity could involve the
costs associated with use of equipment time, depreciation of the
stationary phase and the costs of manpower etc. When the costs
can be assigned accurately, then it is possible to convert multiple
objectives into a single objective, i.e., the overall cost of purifying a
batch of feed. However, in some situations such as the case when
the time to perform the separation is critical, the only objective
that matters is the productivity. This is a typical situation encoun-
tered when molecules are puried for pre-clinical trials where time
is crucial. On the other hand, e.g., for large-scale manufacturing,
solvent recovery costs tend to dominate. In yet other situations,
e.g., hybrid separations, minimizing the cost of the chromatogra-
phy step need not necessarily lead to minimizing overall process
costs. Hence it is suggested that instead of solving a single-objective
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249
problem, a general multi-objective problem be solved and the user
is given the option to choose appropriate operating conditions.
6.1.2. Decision variables
Decision variables refer to operating parameters that affect the
separation and can be varied by the optimizer to search for opti-
mum operating conditions. In principle, any number of decision
variables can be used. While an increased number of decision
variables expands the operational space, it is accompanied with
a corresponding increase in computational effort. Hence a care-
ful choice has to be made to balance the two. In single-column
prep-SFC, typical decision variables include, pressure, temperature,
modier composition, ow-rate, injection concentration, injection
volume and column length.
6.1.3. Constraints
Constraints can be classied into two: physical and operational.
Physical constraints arise due to limitations associated with equip-
ment. Aspects such as the maximum pressure drop that can be
tolerated by the stationary phase, maximum concentration of the
solute that can exist in the column fall under this category. Oper-
ational constraints may involve the requirement on the product
quality, e.g., the product purity.
6.2. Optimization techniques
The role of the optimizer is to work in unison with the simula-
tor to obtain operating conditions that satisfy the given objectives.
In general, optimization tools can be categorized as gradient-based
and non-gradient based. Gradient-based methods attempt to locate
the minima of an appropriately dened objective function by cal-
culating the local gradient of the m + n-dimensional surface formed
between m decision variables and n objective functions. The opti-
mizer then proceeds in the direction of decreasing slope until the
minima is reached. While the main advantage of these methods
is that they are fast, they suffer from a few shortcomings. These
methods suffer from the possibility of being trapped within the
local minima and hence may not locate the global minima. Fur-
ther, they are sensitive to the initial guess of decision variables.
Unless the user is aware of the nature of the problem and is assured
that the surface has a unique minima, it is advisable to avoid these
methods.
The other class of optimization techniques are non-gradient
based such as genetic algorithms (GA). The scheme of GA is based
on the evolutionary principle of survival of the ttest which is
briey explained here. The optimizer chooses a population of
individuals. Each individual represents an unique combination of
decision variables. The set of decision variables is sent to the pro-
cess simulator, which calculates the elution prole and based on
the cut strategy provided, evaluates the performance indicators and
objective functions, i.e., purity, recovery, productivity and solvent
consumption. The performance indicators are veried against the
constraints to check if any of constraints are violated. If yes, the
objective function is penalized by a value that is proportional to
the degree by which the constraints are violated. The values of the
objective function are then transferred to the optimizer and each
individual is assigned a certain tness value. Individuals with bet-
ter tness values are allowed to reproduce and create the next
generation of individuals. However, at this stage two more impor-
tant operations: crossover, i.e., random mating of two individuals;
and mutation, i.e., random change of the individuals identity is
performed. The crossover and mutation provide the much impor-
tant diversity in the population and help the optimizer to escape
from a local minima. As the number of generation increases the
populations improve in the objective functions until they reach
a point where no further improvement is possible. For single
243
Fig. 17. Pareto curves for single-column and a 6-column SMB process for the separa-
tion of Trgers base enantiomers on microcrystalline cellulose triacetate stationary
phase [95]. PR and DR refer to the productivity and desorbent requirement respec-
tively.
objective problems, we are likely to arrive at the global minima,
while for the case of multi-objective optimization we obtain the
pareto set, which consists of a set of non-dominated solutions, i.e.,
none of the solution is better than the other. Genetic algorithm has
been used to optimize several chemical processes [127]. Starting
from early 2000, this technique has been applied to optimize chro-
matography, particularly multi-column processes [8991,95] and
for the design of ternary separation of biomolecules using solvent-
gradient processes [96,97].
Paredes and Mazzotti considered the multi-objective opti-
mization of the separation of Trgers base enantiomers on
microcrystalline cellulose triacetate stationary phase [95]. The
optimization study compared a single column and a 6-column
simulated moving bed process. The objectives were to maximize
productivity and minimize solvent consumption. The pareto curves
are shown in Fig. 17. Let us rst consider the pareto curve for the
single column process. The curve separates the two-dimensional
objective function space into two regions. Towards the bottom right
of the curve is the unfeasible region. In other words, no combination
of decision variables together with the imposed constraints will
allow the process to provide a performance in this region. Towards
the top left of the pareto is the sub-optimal region. It might be pos-
sible to nd operating conditions that will result in a performance
that will correspond to this region. However, that operation will be
sub-optimal as it will be characterized by objective function val-
ues that are both worse than the point on the pareto curve. Note
that all points on the pareto are non-dominated, i.e., for all of the
points at least one of the objective function values are better than
the other. It is also worth noting that each point on the pareto is
obtained by an unique combination of decision variables and the
optimizer can provide this information. Finally the exact operating
point is chosen according to the process requirements. From the
gure, it is also seen that compared to the single column process,
the pareto for the SMB process is towards the lower right and it
clearly outperforms the simpler process. It is worth highlighting
that both processes were provided the exibility to choose operat-
ing conditions that allow exploit their fullest potential and hence
a comparison of their performances is now justied. In situations
where optimization is not performed, comparing two processes is
subjective as it is possible that two sub-optimal points are being
compared.
244 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249

200 6
Yi =95%, P i =95%
Yi =95%, P i =97% a)

productivity, PR (kg/kg/day)
Yi =100%
5 Yi =95%, P i =99%
180
solvent consumption, S (L/kg rac)

P i =100%
Yi =100%, P i =100%
4
160
3
140
Yi =95% 2
120 P i =99%
Yi =95% 1

100 P i =97%
0
10 12 14 16 18 20 22 24
80 cm (%)
Yi =95%
P i =95% 200

solvent consumption, S (L/kg rac)

60 Yi =95%, P i =95%
180 Yi =95%, P i =97% b)
Yi =95%, P i =99%
40 160 Yi =100%, P i =100%
0 1 2 3 4 5 6 140
productivity, PR (kg/kg/day) 120

Fig. 18. Pareto curves for the SFC enantioseparation of urbiprofen on Chiralpak- 100
ADH. Mobile phase: CO2 + methanol, pressure: 135 bar. 80

60

6.3. Multi-objective optimization of single column prep-SFC 40

10 12 14 16 18 20 22 24
cm (%)
Recently, Chen et al. reported the rst optimization study of
prep-SFC [128]. The authors considered the problem of maximiz- Fig. 19. Plot of modier composition corresponding to the pareto points agains the
ing productivity and minimizing organic solvent consumption. two objective functions (a) productivity and (b) solvent consumption [128].
Detailed estimation of isotherm parameters over a range of pres-
sures and modier compositions were made. The decision variables
include modier composition, injection concentration, injection the value of modier composition for the pareto points shown
volume and ow rate. Physical constraints included maximum in Fig. 18 as a function of the two objective functions. The gure
pressure drop and maximum injection concentration, while oper- shows that most points crowd near cm = 20%. In their study the
ational constraints included purity and recovery. Fig. 18 shows modier composition 20% represented the upper range of the mod-
the pareto curves for different purity and recovery requirements. ier composition for which the characterization was performed.
From the gure it is clear that the paretos move towards the The optimization results indicated that both maximum produc-
bottom right when the purity and recovery requirements are tivity and minimum solvent consumption were both obtained at
relaxed. Further, for a purity and recovery of 95%, a productiv- the maximum modier composition. This result is rather counter-
ity of 5 kg racemate/kg stationary phase/day can be achieved with intuitive as one would expect that increasing modier composition
a solvent consumption of 80 L methanol/kg racemate. It is worth would lead to higher solvent consumption. However, the results
noting that this scale of productivity is comparable to liquid phase indicated that the high modier composition allows the injection
multi-column separations, while the organic solvent consumption of higher concentration, owing to favourable solubility and reduces
is lower than the typical values reported for liquid phase systems. the cycle time for the separation, both of which lead to improved
Very interesting insights about the operation of the SFC pro- performances. Finally, the authors also demonstrated the validity of
cess were obtained from the optimization results. Fig. 19 shows the results through the experimental injection of a pulse using the

5
a) 14 b)
4 12
concentration (g/L)

concentration (g/L)

Fraction 2 Fraction 1 Fraction 2

Fraction 1
10
3
8

2 6

4
1
2

0 0
0 200 400 600 0 100 200 300 400
time (s) time (s)

Fig. 20. Experimental implementation of optimized results (a) with xed ow of CO2 and (b) with ow rate as a decision variable [128]. The symbols represent experiments,
while the lines correspond to calculated optimal elution proles.
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249 245

Fig. 21. Schematic of a four-section, two columns per section, simulated moving bed (SMB) process.

decision variables corresponding to a point on the pareto (circled phase SMB is performed based on ow rate ratios, mj , that are
point in Fig. 18). Fig. 20 shows the experimental and the calcu- dened as:
lated proles and the good match between the two is evident. These Qj t V
studies, along with others in the literature highlight the signicant mj = (24)
V (1 )
benets of using these tools for industrial separation.
where Qj is the volumetric ow rate of the mobile phase in section
j, t* is the switch time. In the context of the SF-SMBs, the ow rate
7. Design of multicolumn chromatography with ratios are dened as
supercritical mobile phases Gj t V4
mj = (25)
V (1 )
So far we have discussed separations using single column SFC.
where Gj are the mass ow rate of the mobile phase in section j and
The single column process has several advantages: its imple-
4 is the average uid phase density in section 4 [133]. Note that
mentation and scale-up is straightforward; its well suited for
the internal ow rates Gj are related to the external ow rates by
the separation of multi-component (more than two) mixtures.
the following node balances:
However, the productivity of the unit is rather limited owing to
the xed-bed conguration. For the purication of large batches, G1 = G4 + GD (26)
multi-column chromatography (MCC) start to offer signicant
advantages in terms of productivity and solvent consumption G2 = G1 GE (27)
[1,4]. G3 = G2 + GF (28)
One of the most popular implementation of the MCC is the sim-
ulated moving bed (SMB) chromatography [129,130]. The main G4 = G3 GR (29)
objective in a SMB is to overcome the xed bed operation of the sin- where GD , GE , GF and GR refer to the mass ow rates of the desor-
gle column chromatography and to implement a conguration in bent, extract, feed and rafnate streams respectively. Under linear
which the stationary and the mobile phases move in countercurrent conditions, the explicit constraints that guarantee pure products
directions. However, since practical difculties exist in physically and complete regeneration of the mobile and stationary phases are
moving the stationary phase, the movement is simulated by switch- given by
ing the feed and product ports. The conguration of a four-section,
two columns per section SMB is shown in Fig. 21. The feed consist- HA 1 m1 (30a)
ing of the two solutes diluted in a solvent (typically the modier in HB 2 < m2 HA 2 (30b)
the supercritical uid SMB, SF-SMB) is introduced between sections
2 and 3. The objective is to operate the process in such a manner HB 3 m3 HA 3 (30c)
that the light component (B) moves in the direction of the mobile
m4 HB 4 (30d)
phase and is collected at the rafnate port located between sec-
tions 3 and 4. The heavy component (A) moves in the direction These constraints, particularly those concerning sections 2 and
of the stationary phase and is collected at the extract port located 3, i.e., Eqs. (30b) and (30c) can be conveniently plotted as shown
between sections 1 and 2. The mobile phase is introduced upstream in Fig. 22. When the isotherm is non-linear, the region of complete
of section 1. The countercurrent movement of the solid is simu- separation is no longer a right-triangle but retains the general form
lated by switching the inlet and outlet ports in the direction of of a skewed triangle [4,132]. It is worth noting that the informa-
the uid ow. It is now clear that sections 2 and 3 perform the tion required to construct the triangle are the adsorption isotherm
important duty of separating the two solutes while sections 1 and parameters and the feed concentration. Operating the unit within
4 are responsible to regenerate the stationary and mobile phases the complete separation region ensures pure products at the extract
respectively. and rafnate port. Further, the orthogonal distance from the operat-
The design of SMBs has been a topic of intense research ing point to the diagonal is proportional to the productivity. Hence,
[1,4,12,18,131,132]. The triangle theory, based on the equilib- operating the unit at the vertex of the triangle, indicated by the
rium theory of chromatography is one of the most well known point (HB,3 , HA,3 ) or (HB,2 , HA,2 ) in Fig. 22 and, yields the maximum
approaches to design the SMB [12,131,132]. The design of liquid productivity.
246 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249

Region of compete separation

4.2
under gradient operation PG

4.0

3.8 IM
m3

n
n io
io at

m3
3.6

at ar
er ep
op e s
ic et
at p
cr m
so co
r i of
3.4

de n
un egio
R

m2
Fig. 22. Complete separation region for a SF-SMB under isocratic and gradient oper-
ation.
In 1996, Clavier and Nicoud proposed the use of supercritical
uids as a mobile phase for the SMB process [134]. They further sug-
gested that it is possible to exploit the dependence of the retention
characteristics on the uid phase density to enhance the produc-
tivity. The idea was to operate the SMB in such a way that the
retention in section 3 is stronger than in section 2, i.e., to oper-
ate under gradient mode. In the case of SFC this can be achieved
by operating section 2 at a higher pressure compared to section
3. This mode of operation termed as pressure gradient mode can
be implemented by introducing a back pressure regulator between
sections 2 and 3. Under these conditions for a general case where
HA,3 > HB,3 > HA,2 > HB,2 , the region of complete separation is given
by the quadrilateral as shown in Fig. 22. Hence, in a pressure gra-
dient mode, the best productivity is obtained by operating the unit
a the vertex of the quadrilateral, i.e., at point (HB,2 , HA,3 ). It is clear
that this point is further away from that obtained by operating the
unit at isocratic conditions. The possibility of productivity enhance-
ment by using a pressure gradient is unique to SF-SMB and has been
demonstrated in the separation of the enantiomers of -tetralol on
Chiralcel-OD [135]. The region of complete separation along with
experimental data is shown in Fig. 23 where it was shown that a
3.2
P E< 97% OR P R< 99%
P E> 97% AND P R> 99%
3.0
3.0 3.2 3.4 3.6 3.8 4.0 4.2
m2
Fig. 23. Region of complete separation for the isocratic mode (IM) and pressure gra-
dient (PG) for the SF-SMB separation of the enantiomers of -tetralol on Chiralcel-OD
[135]. The symbols denote the experimental results on a pilot-scale unit.
pressure gradient operation offers three times more productivity
compared to the isocratic operation.
The application of the triangle theory for the separation of
1-phenyl-1-propanol on Chiralcel-OD was demonstrated by Rajen-
dran et al. [63]. Fig. 24 shows the regions of complete separation
and the effect of switch time on the extract and rafnate purities.
The performance of the SF-SMB is identical to liquid phase SMBs
and all strategies to improve productivities that have been devel-
oped for liquid systems can be, in principle, translated to SF-SMB
systems.
The group of Johannsen has successfully demonstrated the
application of SF-SMB separation for the separation of isomers
of phytol [136], enantiomers of ibuprofen [137], and tocopherols
[125]. In their approach since a Hill isotherms were used to describe
the adsorption equilibria, it was not possible to obtain the regions
of complete separation explicitly. Hence, they adopted an alternate
approach where the region on the m2 m3 plane was discretized
into a grid and complete numerical simulations were performed at

9 100
P E < 98%, P R < 98%

7 8 9 10
8 90
10

7 80
Purity [%]
m3 [g/cc]

8
6 70
7

5 60
Extract
Raffinate
(a) (b)
4 50
4 5 6 7 8 9 4 5 6 7 8
m 2 [g/cc] m 2 [g/cc]

Fig. 24. The SF-SMB enantioseparation of 1-phenyl-1-propanol on Chiralcel-OD. Mobile phase: CO2 + 2.6%(w/w) methanol. (a) Location of experimental runs (symbols) on
the region of complete separation and (b) inuence of m2 on the purities of the extract and rafnate.
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249 247

each grid point. In this manner, the region of complete separation L column length [cm]
was mapped and the operating points were chosen for experimen- mj ow rate ratio in SMB
tal demonstration. They were able to obtain high purities for both n solid phase concentration of solute [g L1 ]
rafnate and extract products. p empirical constant
While pressure is certainly a means of implementing a gradi- P pressure [bar]
ent in retention, a gradient in modier composition can also be q empirical constant
used to the same effect. This principle has been exploited in liquid Q volumetric ow rate [cm3 min1 ]
SMBs [138] but has not been explored for SF-SMB applications. It S detector signal
is worth noting that although the pressure gradient improves the T temprerature [ C]
performance of the process, its implementation can be limited by t* switch time in SMB
physical constraints. According to the operating conditions when tR retention time [s]
the column moves from section 1 to 4 after a switch, it undergoes v interstitial velocity [cm s1 ]
a sudden drop in pressure. This can result in the expansion of the z axial coordinate [cm]
packing or pounding if axial compression columns are used. These
may result in damaging the stationary phase. Hence the trade-off Subscripts and superscripts
of implementing pressure and/or modier gradients have to be 0 reference
evaluated. a adsorbed phase
The optimization of liquid SMB processes have been explored A more retained component
in the literature. It has been shown that optimization approaches abs absolute amount
provide rational means of comparing processes, and in realizing analyt analytical
the true potential of operations. However, rigorous optimization b bulk
techniques have not been applied for SF-SMB design. The SF-SMB B less retained component
process, while having signicant potential to improve productivity c critical
an to reduce organic solvent consumption has not taken-off at the D desorbent stream in SMB
industrial scale. The part of the reason is due to the heavy invest- E extract stream in SMB
ment costs associated with the equipment [139]. However, if there ex excess amount
are needs for large-scale purications, e.g., tons of product, then it F feed stream in SMB
would be worth considering the option of an SF-SMB separation. i component index
mix mixed
8. Conclusions mod modier
prep preparative
Supercritical uid chromatography is now a well accepted tech- R rafnate stream in SMB
nology for preparative separations. SFC is now acknowledged as a sorb sorbent
one of the best examples of green chemistry as it helps reduce
solvent consumption and the associated energy required for con- Greek symbols
centrating solutes. Although the rst use of the technology was column permeability [m mL1 ]
reported about 50 years ago, it is still considered as a bit esoteric  void fraction of column
compared to its more well established cousin, HPLC. The technique  viscosity [Pa s]
went through ups and downs and has seen a resurgence in the last  density [g L1 ]
decade. The cloud of mystery around the technique seems to be
slowly clearing and we now understand its advantages and limi- Acknowledgments
tations. The unusual character of the supercritical uid results in
several interesting and sometimes surprising effects. In this review The author is thankful to Georges Guiochon, University of Ten-
we have highlighted some of the differences between HPLC and nessee, Knoxville for providing a copy of Fig. 10 and for interesting
SFC, which we believe will help practitioners in understanding the discussions on prep SFC on numerous occasions. Marco Mazzotti,
technique. We also emphasized the importance of rational design of Massimo Morbidelli, ETH, Zurich are acknowledged for introducing
prep-SFC systems. With the power of modern computers, it is pos- the author to this technology. The author is particularly thankful to
sible to develop robust design and optimization tools for rational Wenda Chen, Reza Haghpanah, Dai Yun and several undergraduate
scale-up of SFC. We have highlighted the steps needed to perform students at Nanyang Technological University, Singapore for their
such a scale-up. These techniques are also in line with initiatives splendid work on SFC. Partial funding from the GSK-Singapore part-
such as QbD from the USFDA, which encourage pharmaceutical sci- nership for sustainable manufacturing is gratefully acknowledged.
entists to adopt more model-based approaches for process design.
These methods will enable designing efcient processes that can References
guarantee better performance compared to conventional strategies
and help towards greening chromatography. [1] M. Juza, M. Mazzotti, M. Morbidelli, Trends Biotechnol. 18 (2000) 108.
[2] G. Guiochon, J. Chromatogr. A 965 (2002) 129.
[3] H. Schmidt-Traub, Preparative Chromatography of Fine Chemicals and Phar-
9. Notation maceutical Agents, Wiley-VCH, Weinheim, 2005.
[4] A. Rajendran, G. Paredes, M. Mazzotti, J. Chromatogr. A 1216 (2009) 709.
[5] G. Guiochon, A. Tarafder, J. Chromatogr. A 1218 (2011) 1037.
[6] S. Nagamatsu, K. Murazumi, S. Makino, J. Chromatogr. A 832 (1999) 55.
[7] Y. Okamoto, Y. Kaida, J. Chromatogr. A 666 (1994) 403.
a empirical constant
[8] C. Welch, J. Chromatogr. A 666 (1994) 3.
b empirical constant [9] E.R. Francotte, J. Chromatogr. A 906 (2001) 379.
c uid phase concentration of solute [g L1 ] [10] E. Yashima, J. Chromatogr. A 906 (2001) 105.
d column diameter [cm] [11] V. Schurig, J. Chromatogr. A 906 (2001) 275.
[12] D.M. Ruthven, C.B. Ching, Chem. Eng. Sci. 44 (1989) 1011.
G mass ow rate [g s1 ] [13] M. McCoy, Chem. Eng. News 78 (2000) 17.
H Henry constant [14] Anon, Chem. Eng. News 81 (2003) 18.
248 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249

[15] D. McCormick, Pharm. Technol. 30 (2006) 54. [76] D.P. Poe, J.J. Schroden, J. Chromatogr. A 1216 (2009) 7915.
[16] S. Abel, M. Juza, in: G. Subramanian (Ed.), Chiral Separation Techniques: [77] K. Kaczmarski, D.P. Poe, G. Guiochon, J. Chromatogr. A 1217 (2010) 6578.
A Practical Approach, 3rd edition, Wiley-VCH, Weinheim, 2007, pp. xxii, [78] K. Kaczmarski, D.P. Poe, G. Guiochon, J. Chromatogr. A 1218 (2011) 6531.
618 p. [79] A. Tarafder, G. Guiochon, J. Chromatogr. A 1218 (2011) 7189.
[17] L. Aumann, M. Morbidelli, Biotechnol. Bioeng. 98 (2007) 1043. [80] T.A. Berger, K. Fogelman, US Patent 6,428,702 (2002).
[18] A. Seidel-Morgenstern, L. Keler, M. Kaspereit, Chem. Eng. Technol. 31 (2008) [81] A. Rajendran, International Symposium, Exhibit and Worshops on Preparative
826. and Process Chromatography PREP, Cambridge, MA, 2011.
[19] A.M. Katti, P. Jagland, Analusis 26 (1998) M38. [82] P.D. Condo, K.P. Johnston, J. Polym. Sci., Part B: Polym. Phys. 32 (1994) 523.
[20] J. Strube, S. Haumreisser, H. Schmidt-Traub, M. Schulte, R. Ditz, Org. Process [83] R.G. Wissinger, M.E. Paulaitis, J. Polym. Sci., Part B: Polym. Phys. 25 (1987)
Res. Dev. 2 (1998) 305. 2497.
[21] A. Jupke, A. Epping, H. Schmidt-Traub, J. Chromatogr. A 944 (2002) 93. [84] U. Fehrenbacher, T. Jakob, T. Berger, W. Knoll, M. Ballauff, Fluid Phase Equilib.
[22] J.R. Bruno, Chim. Oggi-Chem. Today 22 (2004) 32. 200 (2002) 147.
[23] C. Capello, U. Fischer, K. Hungerbuhler, Green Chem. 9 (2007) 927. [85] A. Rajendran, M. Mazzotti, B. Bonavoglia, N. Forrer, G. Storti, M. Morbidelli,
[24] P.T. Anastas, J.C. Warner, Green Chemistry: Theory and Practice, Oxford Uni- Ind. Eng. Chem. Res. 44 (2005) 2549.
versity Press, New York, 1998. [86] B. Bonavoglia, G. Storti, M. Morbidelli, A. Rajendran, M. Mazzotti, J. Polym.
[25] C.J. Welch, N.J. Wu, M. Biba, R. Hartman, T. Brkovic, X.Y. Gong, R. Helmy, Sci., Part B: Polym. Phys. 44 (2006) 1531.
W. Schafer, J. Cuff, Z. Pirzada, L.L. Zhou, Trac-Trend Anal. Chem. 29 (2010) [87] S.R. Springston, P. David, J. Steger, M. Novotny, Anal. Chem. 58 (1986) 997.
667. [88] C.J. Welch, P. Sajonz, G. Spencer, W. Leonard, D. Henderson, W. Schafer, F.
[26] A. Tullo, Chem. Eng. News 86 (2008) 27. Bernardoni, Org. Process Res. Dev. 12 (2008) 674.
[27] E. Klesper, A.H. Corwin, D.A. Turner, J. Org. Chem. 27 (1962) 700. [89] Z. Zhang, K. Hidajat, A.K. Ray, M. Morbidelli, AlChE J. 48 (2002) 2800.
[28] C. Berger, M. Perrut, J. Chromatogr. 505 (1990) 37. [90] Z. Zhang, M. Mazzotti, M. Morbidelli, J. Chromatogr. A 989 (2003) 95.
[29] T.A. Berger, W.H. Wilson, Anal. Chem. 65 (1993) 1451. [91] J.M.M. Araujo, R.C.R. Rodrigues, J.P.B. Mota, J. Chromatogr. A 1132 (2006) 76.
[30] M. Perrut, J. Chromatogr. A 658 (1994) 293. [92] Y. Kawajiri, L.T. Biegler, J. Chromatogr. A 1133 (2006) 226.
[31] T. Clifford, Fundamentals of Supercritical Fluids, Oxford University Press, New [93] Y. Kawajiri, L.T. Biegler, AlChE J. 52 (2006) 1343.
York, 1999. [94] Y. Kawajiri, L.T. Biegler, Ind. Eng. Chem. Res. 45 (2006) 8503.
[32] T.L. Chester, J.D. Pinkston, Anal. Chem. 74 (2002) 2801. [95] G. Paredes, M. Mazzotti, J. Chromatogr. A 1142 (2007) 56.
[33] T.A. Berger, Packed Column SFC, R. Soc. Chem., London, 1995. [96] A. Tarafder, L. Aumann, M. Morbidelli, J. Chromatogr. A 1195 (2008) 67.
[34] G. Teroth, J. Chromatogr. A 906 (2001) 301. [97] A. Tarafder, G. Strohlein, L. Aumann, M. Morbidelli, J. Chromatogr. A 1183
[35] L.T. Taylor, J. Supercrit. Fluids 47 (2009) 566. (2008) 87.
[36] L.T. Taylor, Anal. Chem. 82 (2010) 4925. [98] L.X. Yu, Pharm. Res. 25 (2008) 781.
[37] K. Anton, C. Berger, Supercritical uid chromatography with packed columns: [99] A.S. Rathore, H. Winkle, Nat. Biotechnol. 27 (2009) 26.
Techniques and applications, Chromatogr. Sci. Ser., Marcel Dekker, New York, [100] S. Ottiger, J. Kluge, A. Rajendran, M. Mazzotti, J. Chromatogr. A 1162 (2007)
1998. 74.
[38] C. Welch, R. Majors, W. Leonard Jr., J. DaSilva, M. Biba, J. Albaneze-Walker, D. [101] G. Guiochon, A. Felinger, S.G. Shirazi, A.M. Katti, Fundamentals of Preparative
Henderson, B. Laing, D. Mathre, LCGC North America 23 (2005) 16. and Nonlinear Chromatography, Academic Press, Boston, 2006.
[39] S. Sircar, J. Chem. Soc., Faraday Trans. 1 (81) (1985) 1527. [102] G. Guiochon, B.C. Lin, Modeling for Preparative Chromatography, 1st edition,
[40] S. Sircar, Ind. Eng. Chem. Res. 38 (1999) 3670. Academic Press, Amsterdam; Boston, 2003.
[41] L.D. Gelb, K.E. Gubbins, R. Radhakrishnan, M. Sliwinska-Bartkowiak, Rep. Prog. [103] P. Rouchon, M. Schonauer, P. Valentin, G. Guiochon, Sep. Sci. Technol. 22
Phys. 62 (1999) 1573. (1987) 1793.
[42] S. Sircar, AlChE J. 47 (2001) 1169. [104] M. Czok, G. Guiochon, Anal. Chem. 62 (1990) 189.
[43] O. Di Giovanni, W. Drer, M. Mazzotti, M. Morbidelli, Langmuir 17 (2001) [105] M. Czok, G. Guiochon, Comput. Chem. Eng. 14 (1990) 1435.
4316. [106] Z. Ma, G. Guiochon, Comput. Chem. Eng. 15 (1991) 415.
[44] T. Hocker, A. Rajendran, M. Mazzotti, Langmuir 19 (2003) 1254. [107] R.J. LeVeque, Numerical Methods for Conservation Laws, 2nd edition,
[45] A. Tarafder, G. Guiochon, J. Chromatogr. A 1229 (2012) 249. Birkhuser Verlag, Basel; Boston, 1992.
[46] M.E. Fisher, P. de Gennes, C.R. Acad. Sci. Ser. B 287 (1978) 207. [108] S. Javeed, S. Qamar, A. Seidel-Morgenstern, G. Warnecke, Comput. Chem. Eng.
[47] G.F.S. Blmel, Phys. Rev. Lett. 54 (1985) 447. 35 (2011) 2294.
[48] U.M.B. Marconi, Phys. Rev. A: At. Mol. Opt. Phys. 38 (1988) 6267. [109] B. Medi, M. Amanullah, Ind. Eng. Chem. Res. 50 (2011) 1739.
[49] M. Thommes, G.H. Findenegg, H. Lewandowski, Ber. Bunsenges. Phys. Chem. [110] H.K. Rhee, R. Aris, N.R. Amundson, Phil. Trans. R. Soc. Lond. Ser. A: Phys. Sci.
98 (1994) 477. Eng. 267 (1970) 419.
[50] R. Humayun, D.L. Tomasko, AlChE J. 46 (2000) 2065. [111] F. Helfferich, G. Klein, Multicomponent Chromatography. Theory of Interfer-
[51] S. Sircar, J. Chem. Soc., Faraday Trans. 1 (1985) 81. ence, Marcel Dekker, New York, 1970.
[52] C.R. Yonker, R.D. Smith, J. Chromatogr. 505 (1990) 139. [112] S. Golshan-Shirazi, G. Guiochon, J. Phys. Chem. A 93 (1989) 4143.
[53] J.R. Strubinger, H.C. Song, J.F. Parcher, Anal. Chem. 63 (1991) 98. [113] H.K. Rhee, R. Aris, N.R. Amundson, First-order Partial Differential Equations,
[54] J.R. Strubinger, H.C. Song, J.F. Parcher, Anal. Chem. 63 (1991) 104. Volume I, Dover Publications, Mineola, NY, 2001.
[55] C.R. Yonker, R.D. Smith, J. Chromatogr. 550 (1991) 775. [114] H.K. Rhee, R. Aris, N.R. Amundson, First-order Partial Differential Equations,
[56] M. Sudibandriyo, Z.J. Pan, J.E. Fitzgerald, R.L. Robinson, E.A.M. Gasem, Lang- Volume II, Dover Publications, Mineola, NY, 2001.
muir 19 (2003) 5323. [115] M. Mazzotti, Ind. Eng. Chem. Res. 45 (2006) 5332.
[57] J.E. Fitzgerald, Z. Pan, M. Sudibandriyo, R.L. Robinson, K.A.M. Gasem, S. Reeves, [116] A. Rajendran, M. Mazzotti, Ind. Eng. Chem. Res. 50 (2011) 352.
Fuel 84 (2005) 2351. [117] A. Seidel-Morgenstern, J. Chromatogr. A 1037 (2004) 255.
[58] M. Mazzotti, R. Pini, G. Storti, J. Supercrit. Fluids 47 (2009) 619. [118] B.G. Su, H.B. Xing, Y.S. Han, Y.W. Yang, Q.L. Ren, P.D. Wu, J. Chem. Eng. Data
[59] R. Pini, S. Ottiger, L. Burlini, G. Storti, M. Mazzotti, Int. J. Greenh Gas Con. 4 54 (2009) 2906.
(2010) 90. [119] A. Felinger, A. Cavazzini, G. Guiochon, J. Chromatogr. A 986 (2003) 207.
[60] A. Rajendran, M. Mazzotti, M. Morbidelli, J. Chromatogr. A 1076 (2005) 183. [120] L. Asnin, G. Guiochon, J. Chromatogr. A 1089 (2005) 101.
[61] U. van Wasen, G.M. Schneider, Chromatographia 8 (1975) 274. [121] L. Asnin, G. Guiochon, J. Chromatogr. A 1089 (2005) 105.
[62] C.H. Lochmuller, L.P. Mink, J. Chromatogr. 471 (1989) 357. [122] A.L. Myers, J.M. Prausnitz, AlChE J. 11 (1965) 121.
[63] A. Rajendran, S. Peper, M. Johannsen, M. Mazzotti, M. Morbidelli, G. Brunner, [123] C. Blmel, P. Hugo, A. Seidel-Morgenstern, J. Chromatogr. A 865 (1999) 51.
J. Chromatogr. A 1092 (2005) 55. [124] G. Brunner, M. Johannsen, J. Supercrit. Fluids 38 (2006) 181.
[64] C. Wenda, A. Rajendran, J. Chromatogr. A 1216 (2009) 8750. [125] M. Lubbert, G. Brunner, M. Johannsen, J. Supercrit. Fluids 42 (2007) 180.
[65] K. Kaczmarski, J. Kostka, W. Zapala, G. Guiochon, J. Chromatogr. A 1216 (2009) [126] A. Rajendran, W. Chen, Sep. Purif. Technol. 67 (2009) 344.
6560. [127] V. Bhaskar, S.K. Gupta, A.K. Ray, Rev. Chem. Eng. 16 (2000) 1.
[66] K. Kaczmarski, F. Gritti, J. Kostka, G. Guiochon, J. Chromatogr. A 1216 (2009) [128] W. Chen, R. Hagpanah, A. Rajendran, M. Amanullah, J. Chromatogr. A 1218
6575. (2011) 162.
[67] J. Kostka, F. Gritti, G. Guiochon, K. Kaczmarski, J. Chromatogr. A 1217 (2010) [129] D.B. Broughton, G.G. Gerhold, Continuous sorption process employing
4704. xed bed of sorbent and moving inlets and outlets, 1961, U.S. Patent
[68] A. Rajendran, O. Kruchi, M. Mazzotti, M. Morbidelli, J. Chromatogr. A 1092 2,985,589.
(2005) 149. [130] R.M. Nicoud, G. Fuchs, P. Adam, M. Bailly, E. Kusters, F.D. Antia, R. Reuille, E.
[69] A. Rajendran, T.S. Gilkison, M. Mazzotti, J. Sep. Sci 31 (2008) 1279. Schmid, Chirality 5 (1993) 267.
[70] P.J. Schoenmakers, L.G.M. Uunk, Chromatographia 24 (1987) 51. [131] G. Storti, M. Mazzotti, M. Morbidelli, S. Carra, AlChE J. 39 (1993) 471.
[71] X.W. Lou, H.G. Janssen, H. Snijders, C.A. Cramers, HRC-J High Res. Chrom. 19 [132] M. Mazzotti, G. Storti, M. Morbidelli, J. Chromatogr. A 769 (1997) 3.
(1996) 449. [133] M. Mazzotti, G. Storti, M. Morbidelli, J. Chromatogr. A 786 (1997)
[72] C. Bouigeon, D. Thiebaut, M. Caude, Anal. Chem. 68 (1996) 3622. 309.
[73] D.P. Poe, J. Chromatogr. A 785 (1997) 129. [134] J. Clavier, R. Nicoud, M. Perrut, in: P. von Rohr, C. Trepp (Eds.),
[74] D.P. Poe, P.J. Marquis, T. Tomlinson, J. Dohm, J. He, J. Chromatogr. A 785 (1997) High Pressure Chemical Engineering, Elsevier Science, Amsterdam, 1996,
135. p. 429.
[75] W.S. Xu, D.L. Peterson, J.J. Schroden, D.P. Poe, J. Chromatogr. A 1078 (2005) [135] F. Denet, W. Hauck, R.M. Nicoud, O. Di Giovanni, M. Mazzotti, J.N. Jaubert, M.
162. Morbidelli, Ind. Eng. Chem. Res. 40 (2001) 4603.
 A. Rajendran / J. Chromatogr. A 1250 (2012) 227249 249

[136] A. Depta, T. Giese, M. Johannsen, C. Brunner, J. Chromatogr. A 865 (1999)
175.
[137] S. Peper, M. Lubbert, M. Johannsen, G. Brunner, Sep. Sci. Technol. 37 (2002)
2545.
[138] S. Abel, M. Mazzotti, M. Morbidelli, J. Chromatogr. A 944 (2002) 23.
[139] S. Peper, M. Johannsen, G. Brunner, J. Chromatogr. A 1176 (2007) 246.
[140] V. Roskar, R.A. Dombro, G.A. Prentice, C.R. Westgate, M.A. Mchugh, Fluid Phase
Equilib. 77 (1992) 241.
[141] J.G. Liu, Z.F. Qin, G.F. Wang, X.L. Hou, J.G. Wang, J. Chem. Eng. Data 48 (2003)
1610.
[142] E. Brunner, W. Hultenschmidt, G. Schlichtharle, J. Chem. Thermodyn. 19
(1987) 273.