(Advances in Chemical Engineering 40) Guy B. Marin (Eds.) - Multiscale Simulation and Design-Elsevier - Academic Press (2011)

Academic Press is an imprint of Elsevier
Linacre House, Jordan Hill, Oxford OX2 8DP, UK

32 Jamestown Road, London NW1 7BY, UK
Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands
225 Wyman Street, Waltham, MA 02451, USA
525 B Street, Suite 1900, San Diego, CA 92101-4495, USA
First edition 2011
Copyright Ó 2011, Elsevier Inc. All rights reserved
No part of this publication may be reproduced, stored in a retrieval system

or transmitted in any form or by any means electronic, mechanical, photocopying,
recording or otherwise without the prior written permission of the publisher
Permissions may be sought directly from Elsevier’s Science & Technology Rights
Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333;
email: permissions@elsevier.com. Alternatively you can submit your request online by
visiting the Elsevier web site at http://www.elsevier.com/locate/permissions, and selecting:
Obtaining permission to use Elsevier material.
Notice
No responsibility is assumed by the publisher for any injury and/or damage to persons
or property as a matter of products liability, negligence or otherwise, or from any use
or operation of any methods, products, instructions or ideas contained in the
material herein
ISBN: 978-0-12-380985-8
ISSN: 0065-2377
For information on all Academic Press publications

visit our website at elsevierdirect.com
Printed and bound in USA

11 12 1 3 14 10 9 8 7 6 5 4 3 2 1
PREFACE
Multi-scale Simulation and Design
The increasing importance of multi-scale computation in engineering,

stimulated both by the dramatic development of computer technology
and the better understanding of multi-scale structures, has incited me to
edit this issue on multi-scale simulation and design or so-called “virtual
process engineering”. The title “Multiscale Simulation and Design” indi-
cates the progress achieved by the chemical engineering community since
2005, the year in which volume 35 of this serial with as title “Multiscale
Analysis” was published. The intention of the present volume is to eluci-
date the bottlenecks and to identify the correct directions for the coming
years from the process and product engineering point of view. Both fun-
damental and practical contributions are provided from academia and
industry.
Bridging the gap between micro- and macro-scale is the central theme
of the first contribution. The authors show how a so-called Energy-
Minimization Multi-Scale (EMMS) model allows to do this for circulating
fluid beds. This variational type of Computational Fluid Dynamics (CFD)
modeling allows for the resolution of meso-scale structures, that is, those
accounting for the particle interactions, and enables almost grid-indepen-
dent solution of the gas–solids two-phase flow.
The second contribution spans an even larger range of length and times
scales. Two benchmark examples illustrate the design approach: polymer
electrolyte fuel cells and hard disk drive (HDD) systems. In the current
HDDs, the read/write head flies about 6.5 nm above the surface via the air
bearing design. Multi-scale modeling tools include quantum mechanical
(i.e., density functional theory (DFT)), atomistic (i.e., Monte Carlo (MC) and
molecular dynamics (MD)), mesoscopic (i.e., dissipative particle dynamics
(DPD) and lattice Boltzmann method (LBM)), and macroscopic (i.e., LBM,
computational fluid mechanics, and system optimization) levels.
An old classic but now particularly relevant process, carbon dioxide
(CO2) capture with alkanolamines, is used in the most industrially oriented
of the contributions to illustrate that integrated, high fidelity, multi-scale
ix
x Preface
models from molecular level to site-wide chemical complex or enterprise

level can be deployed in online environments to deliver benefits and
insight. While this contribution does not include molecular simulations
or CFD, it does deal with multi-scale models that integrate layers of
heterogeneous models across significant differences of physical scale
of the chemical supply chain, from single and multiphase systems, to
process equipments and units, to plants, complexes, and enterprises. It is
clearly shown that equation oriented (EO) simultaneous solution tech-
niques make it possible to solve highly complex models in industrial
environments efficiently and very robustly.
The last contribution is the most fundamental one. A large number of
new functional materials have unique meso-scale structures originating
from processes over a wide range of length and time scales. A “classical”
approach combined with DFT can describe molecular conformation dis-
tribution of polymer chains at interface regions. The theoretical predic-
tions are in agreement with experimental observations. It was thanks to an
international cooperation supported by the 111 project “Chemical
Reaction Engineering Science and Technology” that significant progress
was made alone these lines in the recent years. How to predict the exact
macroscopic properties of polymer blends or block copolymers with
meso-phase separation structures from pure component properties
remains a big challenge however.
Guy B. Marin
March 8, 2011
Advances in
CHEMICAL ENGINEERING
MULTISCALE SIMULATION AND DESIGN
VOLUME 40
ADVANCES IN
Editor-in-Chief
GUY B. MARIN
Department of Chemical Engineering,
Ghent University,
Ghent, Belgium
Editorial Board
DAVID H. WEST
Research and Development
The Dow Chemical Company
Freeport, Texas, U.S.A.
JINGHAI LI
Institute of Process Engineering
Chinese Academy of Sciences
Beijing, P.R. China
SHANKAR NARASIMHAN
Department of Chemical Engineering
Indian Institute of Technology
Chennai, India
Advances in
MULTISCALE SIMULATION AND DESIGN
VOLUME 40
Edited by
GUY B. MARIN
Amsterdam * Boston * Heidelberg * London * New York * Oxford

Paris * San Diego * San Francisco * Singapore * Sydney * Tokyo
Academic Press is an imprint of Elsevier
CONTRIBUTORS
Lorenz T. Biegler, Department of Chemical Engineering, Carnegie Mellon

University, Pittsburgh, PA 15213, USA
Chau-Chyun Chen, Aspen Technology, Inc., Burlington, MA 01803, USA
Xueqian Chen, State Key Laboratory of Chemical Engineering and Department of

Chemistry, East China University of Science and Technology, Shanghai 200237,
China
Pil Seung Chung, Department of Chemical Engineering, Carnegie Mellon

University, Pittsburgh, PA 15213, USA
Wei Ge, The EMMS Group, State Key Laboratory of Multi-Scale Complex
Systems, Institute of Process Engineering, Chinese Academy of Sciences, Beijing
100190, P.R. China
Boyd Gochenour, Aspen Technology, Inc., Burlington, MA 01803, USA
Ying Hu, State Key Laboratory of Chemical Engineering and Department of

Chemistry, East China University of Science and Technology, Shanghai
200237, China
Yongmin Huang, State Key Laboratory of Chemical Engineering and

Department of Chemistry, East China University of Science and Technology,
Shanghai 200237, China
Myung S. Jhon, Department of Chemical Engineering, Carnegie Mellon

University, Pittsburgh, PA 15213, USA; School of Advanced Materials Science
and Engineering, Sungkyunkwan University, Suwon 440-746, Korea
Jinghai Li, The EMMS Group, State Key Laboratory of Multi-Scale Complex
100190, P.R. China
Advances in Heterocyclic Chemistry, Volume 102 Ó 2011 Elsevier Inc.
ISSN 0065-2725, DOI All rights reserved
vii
viii Contributors
Honglai Liu, State Key Laboratory of Chemical Engineering and Department of

China
Milo D. Meixell, Aspen Technology, Inc., Burlington, MA 01803, USA
Wei Wang, The EMMS Group, State Key Laboratory of Multi-Scale Complex
100190, P.R. China
Xingqin Xiao, State Key Laboratory of Chemical Engineering and Department of

China
Ning Yang, The EMMS Group, State Key Laboratory of Multi-Scale Complex
100190, P.R. China
CHAP TER 1
Meso-Scale Modeling—The Key
to Multi-Scale CFD Simulation
Wei Wang*, Wei Ge, Ning Yang and Jinghai Li*
Contents 1. Meso-Scale Structure—A Common Challenge for 2

Chemical Engineering
1.1 Multi-scale characteristics of chemical reactors 2
1.2 Micro-, meso-, and macro-scales 4
1.3 Meso-scale structure—spatiotemporal features 5
1.4 Critical effect of meso-scale structure 8
2. Multi-Scale CFD—Solutions for High Predictability and 10
Scalability
2.1 Single-scale approach—its limitation on predictability 10
2.2 Direct numerical simulations—its limitation on 10
scalability
2.3 Multi-scale CFD approach—a compromise of 12
predictability and scalability
3. Meso-Scale Modeling—The Key to Multi-Scale 24
Approaches
3.1 Energy-minimization multi-scale (EMMS) model—a 24
meso-scale model
3.2 Coupling of EMMS and CFD 26
3.3 Application of EMMS to mass/heat transfer and 35
reactions
3.4 Extension of EMMS modeling to gas–liquid flow 40
4. Industrial Applications 43
4.1 Fluid catalytic cracking 43
4.2 CFB boiler 46
The EMMS Group, State Key Laboratory of Multi-Scale Complex Systems, Institute of Process
Engineering, Chinese Academy of Sciences, Beijing 100190, P.R. China
*Corresponding author, Email address: jhli@home.ipe.ac.cn; wangwei@home.ipe.ac.cn
Advances in Chemical Engineering, Volume 40 # 2011 Elsevier Inc.
ISSN 0065-2377, DOI 10.1016/B978-0-12-387036-0.00005-0 All rights reserved
1
2 Wei Wang et al.
5. Summary 51
Nomenclature 52
Greek letters 53
Subscripts 53
Acknowledgments 54
References 54
Abstract Meso-scale structure is critical to characterize complex systems in

chemical engineering. Conventional two-fluid model (TFM) without
meso-scale modeling has proved to be inadequate for describing
gas–solid flow systems featuring multi-scale heterogeneity. In this
review, it is demonstrated that, based on the energy-minimization
multi-scale (EMMS) model, the multi-scale computational fluid
dynamics (CFD) for gas–solid systems has reasonably accounted
the effects of meso-scale structures and hence upgraded both
computational efficiency and accuracy significantly. This approach
has succeeded in predicting the circulating solids flux, revealing the
mechanisms of the choking phenomena and resolving the disputes
in transport phenomena of gas-fluidized beds. It has also been
applied to a wide range of industrial processes including fluid
catalytic cracking (FCC), coal combustion, and so on. In all, the
multi-scale CFD with EMMS modeling is intrinsically multi-scaled,
free from the requirement of clear scale separation, and it can be
expected to be an emerging paradigm for the simulation of multi-
phase flows and reactors.
1. MESO-SCALE STRUCTURE—A COMMON CHALLENGE

FOR CHEMICAL ENGINEERING
Chemical engineering encompasses a broad spectrum of scales with

regard to time and space, or briefly, it is multi-scaled. Enumerating from
small (fast) to big (slow), this spectrum is highlighted with, for example,
molecule, molecular assembly, particle, particle cluster, reactor, and pro-
cess/plant up to the environment, and may be further grouped into three
levels, that is, material, reactor and system, as shown in Figure 1. Each
level has its own multi-scale features, in forms of a variety of structures.
Obviously, how to characterize these multi-scale structures is a common
challenge to all three levels, and hence to the whole domain of chemical
engineering. In this article, as a specific example to meet the challenge, we
will focus on the multi-scale characterization at the reactor level.
1.1 Multi-scale characteristics of chemical reactors
Understanding of a multiphase chemical reactor involves chemical
(catalysis) kinetics, hydrodynamics and heat/mass transfers at scales
Meso-Scale Modeling—The Key to Multi-Scale CFD Simulation 3
[(Figure_1)TD$IG]
Figure 1 Multi-scale characteristics of chemical engineering and its multilevel

classification.
(Adapted from Li et al., 2009.)
ranging from molecules to the reactor (Dudukovic, 2009; Li and Kwauk,

2003). Below the (catalyst) particle scale, quantum and molecular
dynamics coupled with surface chemistry help us understand the
assembly of molecules and (catalysis) reaction mechanism, leading to
detailed chemical kinetics. On the scales ranging from catalyst particles
and particle clusters, the interactions, within or between fluid vortices
and particle aggregates, particle–fluid mixing, as well as heterogeneous
heat/mass transfer and reactions, are required to obtain the closure
laws for computational fluid dynamics (CFD) modeling. On the reactor
scale, global measurement and control can be performed, for optimal
operation and design, for which the detailed CFD modeling may shed
light on the underlying mechanisms.
The scales involved in such a reactor should be defined in a relative
manner. For a chemist, the molecule is at the start and catalyst (particle)
at the end of the scales. To reveal the reaction mechanism over a catalyst
particle, a sequence of network of ‘‘elementary reactions’’ will be
needed. Accordingly, on the basis of, for example, the molecular colli-
sion theory (Turns, 2000), the ‘‘global reaction’’ can be derived in terms
of global rate coefficient and reaction order. Here, the resultant reaction
mechanism is termed ‘‘global’’ by chemists, because the use of it for a
specific problem is normally a ‘‘black box’’ approach, without knowing
exactly the underlying networks or structures of chemical routes from
reactants to products. On the other hand, for a chemical reaction engi-
neer, the catalyst (particle) is often the start and the reactor is the end.
The reaction free of inner-particle and outer-particle diffusions, that is,
4 Wei Wang et al.
without mass transfer, is normally termed as the ‘‘intrinsic reaction’’

(Levenspiel, 1999). On the basis of the ‘‘intrinsic,’’ the ‘‘overall’’ reaction
behavior can be evaluated on the reactor scale by including the effect of
flow and mixing within the reactor. In all, a chemist and a chemical
engineer may study the same phenomena on certain specific scale (say,
the particle scale), but owing to its relative stand, this scale may be
viewed as the ‘‘intrinsic’’ or the ‘‘global.’’ That is the reason why we
need to define the scales in a relative manner, namely, the micro-, meso-
and macro-scales. The ‘‘micro’’ in a field may be the ‘‘macro’’ in another
field, and vice versa.
1.2 Micro-, meso-, and macro-scales

Normally, the scale with respect to the ‘‘elementary’’ or ‘‘intrinsic’’ end
of one research domain (or level) can be termed the micro-scale, below
which the behavior are assumed, or given as input. The scale with
respect to the ‘‘global’’ end of the research domain can be termed the
macro-scale, above which the overall performance can be measured or
adjusted. In between, the wide span of scales between the micro- and
the macro-scales can be termed the meso-scale, which is characterized
by heterogeneous structures with respect to time and space. It is not
surprising, then, that the meso-scale is the critical and also the most
informative scale to understand the whole range of scales, and it is the
bridge between the micro-scale nature and the macro-scale appearance
(Li and Ge, 2007). That is also the reason why we do not distinguish the
usage between ‘‘structure’’ and ‘‘meso-scale structure’’ in the following
discussions.
Following the above definition of scales, we can see, for a fluidized
bed reactor, the single particle stands for the micro-scale, on which the
fluid–particle interactions have been thoroughly investigated and sum-
marized into well-accepted laws, for example, the standard drag coef-
ficient for momentum transfer, the Ranz relation for heat/mass transfer
(Ranz, 1952), and so on. To the opposite end, the reactor represents the
macro-scale, over which we may measure and control the temperature,
pressure, and gas flow rate, to achieve optimal conversion of products
with as least as possible power consumption and pollutant emissions.
In between, the meso-scale is characterized with a variety of heteroge-
neous structures such as bubbles or clusters, and so on, to which the
mirco-scale particles react nonlinearly. Within the meso-scale struc-
tures, the behavior of individual particles is quite different from that
of isolated, single particles, with local asymmetry and accelerations that
may result in additional factors that cannot be accounted for by simple
averaging of single-particle behaviors. That is right the reason why we
need the scale-up of a reactor. To some extent, the scale-up is to grasp
the effects of the meso-scale structures. To this end, in-depth under-

standing of the meso-scale structure is critical.
1.3 Meso-scale structure—spatiotemporal features

The meso-scale structure is likely dynamic and hereby hard to charac-
terize. By comparison, for static structure, as is the case in a fixed bed,
though there is a wide variety of morphology for the packed state, it is
still possible to enumerate its variation and then analyze its statistical
effects through subparticle level simulations (Dixon et al., 2006; Van der
Hoef et al., 2006) or experiments (Ergun, 1952; Ranz, 1952). The micro-
scale difference of static structures can be smoothed up with increase of
samples, and then the structural effects are comparatively easier to
grasp. For dynamic structure, however, as is the case in a fluidized
bed, the degrees of freedom increases significantly with continuous
evolution of shapeless bubbles/clusters (Kunii and Levenspiel, 1991),
and then, it is hard to statistically analyze the structural effects by
enumerating samples.
1.3.1 Time-averaged characterization

To characterize the meso-scale structure from a series of fluctuating/
random signals, the mean (or expectation) values in terms of time
averaging or probability-weighted averaging are mostly used. On the
basis of that, the fluctuation, the variance or its square root, the standard
deviation, and higher order moments can be derived (Pope, 2000). For a
fluidized bed, for instance, it was reported that there exists a bimodal
probability distribution over the entire range of solids fraction, one apex
corresponding to the dense ‘‘cluster’’ phase and the other to the dilute
‘‘broth’’ phase (Li and Kwauk, 1994), respectively. Lin et al. (2001) found
that the dense phase (clusters) has a Gaussian solids fraction distribu-
tion and the dilute phase (broth or void) has a log-normal solids fraction
distribution, and they recommended that the mean solids fraction of the
cluster minus three times its standard deviation as the criterion to
distinguish clusters from the broth. Another approach to identify a
cluster is to set the time-averaged solids fraction plus n-times its stan-
dard deviation as the threshold value for clusters. However, how to
define the value of n over the fluctuating signals remains an art rather
than a science. For example, Soong et al. (1994) and Liu et al. (2005)
suggested n = 3.0, Sharma et al. (2000) suggested n = 2.0, while
Manyele et al. (2002) suggested n = 1.0–1.4 according to their respective
sensitivity analysis. Besides this uncertainty in definitions, measure-
ment ambiguity owing to different configuration of probes will also
lead to different descriptions of structures, as recognized by Reh and
6 Wei Wang et al.
Li (1991). All of these, however, suggest that reducing the meso-scale

structures into two-phase description is a reasonable simplification.
Following the above two-phase description, we can further define
the velocities with respect to the dense ‘‘clusters’’ and dilute ‘‘broth.’’
The slip velocity in dense clusters is normally much lower than that
in the dilute ‘‘broth,’’ which results in dense clusters falling down along
the wall while upflowing gas in broth tearing off particles out of the
clusters (Li et al., 1993). Seeing from the micro-scale using direct numer-
ical simulations (DNS) approaches, Ma et al. (2006) simulated a periodic
suspension of 1024 solid particles and found that heterogeneity is
observable even in such tiny system in forms of particle clusters, formed
progressively from uniform suspension. This was demonstrated to be
results of the compromise between the tendency of the solids to main-
tain low gravitational potential and that of the gas phase to maintain
low energy consumption for suspending and transport when flowing
through the solids. According to the heterogeneity index proposed in
that work, a characteristic scale can be determined where the heteroge-
neity is most evident. The heterogeneity is also found in the velocity of
the solids. Although Gaussian distribution is still valid for the velocity
component in each direction, the variance in the gravitational (vertical)
direction is notably higher than in the horizontal direction. That is, the
velocity distribution is anisotropic. Equal partitioning of the particle
kinetic energy, as in the case of gas molecules, is not reached. The same
thing happens to the drag force distribution. The distributions of the drag
force components are Gaussian or nearly Gaussian, but the solids in the
dilute phase suffer far larger drag forces than those in the center of the
dense phase. These are all clear evidence that the solid phase is in non-
equilibrium state, which must be taken into account in CFB simulations.
The same mean velocity of clusters may have different moving
tendency. If the net force exerted on a cluster is positive against gravity,
its solids fraction decreases and the cluster diffuses or fragments; if the
net force is negative, the cluster is forming or concentrating (Liu et al.,
2005). Such an imbalance between the dense and dilute phases requires
more degrees of freedom to account for the dynamic factors.
1.3.2 Dynamic characterization

The acceleration is a direct measure to the dynamic factors. However,
there are few reports, if not to say none, about that for meso-scale
structures. In a recent attempt, Meng et al. (2009) made use of the
multiple sensors of an X-ray computerized tomography (CT) to mea-
sure the cluster accelerations. Instead of the conventional use of CT for
cross-sectionally scanning the solids distribution, they erected the X-ray
fan-beam and the sensors to follow the vertical movement of clusters
[(Figure_2)TD$IG]
Figure 2 X-ray measurement of cluster acceleration (Meng et al., 2009).

(a) Identification of cluster with mean + 1.4s on two correlated series of X-ray
signals at time t and t + t ; (b) the cluster velocity series calculated by dividing the
displacement with time step; (c) the probability density function of cluster
acceleration. (Air-glass beads, Ug = 3.57 m/s, Gs = 65.13 kg/(m2 s), X-ray measurement
platform was 1 m above the secondary distributors of a 10.5 m high CFB riser.)
along the riser wall of a CFB. As shown in Figure 2, by cross-correlation

analysis of a time series of the detected signals of solids fraction of
clusters (Figure 2a), the cluster velocity series can be determined by
dividing the displacement with the time step (Figure 2b), and likewise,
the cluster accelerations can be calculated by dividing the cluster veloc-
ity with the time step (Figure 2c). It was found that the clusters near the
wall are under force balance in average, in the sense that the measured
accelerations are in normal distribution with mean value of about zero.
The nonzero accelerations closely relate to the deforming, aggregating
and fragmenting of clusters, and this is the basic difference between a
dynamic cluster and a moving porous medium. More experiments are
under way and can be expected to help grasp the dynamic nature of the
meso-scale clusters.
8 Wei Wang et al.
Besides the time-averaged and the dynamic characterizations, the

correlations between different scales add more complexities to the
meso-scale structure. For example, the intensive exchange between meso-
scale clusters and dispersed particles will reduce the cross-correlation
coefficient and then make it hard to discern a cluster from broth.
Such complexity requires more efforts in exploring the dominant
mechanisms underlying the correlations, which will be discussed and
exemplified by multi-scale CFD in later sections.
1.4 Critical effect of meso-scale structure

The meso-scale structure is a common challenge for chemical engi-
neering, owing not only to its difficulty to grasp but also to its critical
effects on the flow, heat/mass transfer, and reaction behavior. For a
gas-fluidized bed, for example, the single-particle behavior has been
thoroughly investigated. There are common senses as to the inter-
phase momentum or heat/mass transfer, and so on (Crowe, 2006).
When the meso-scale structure is involved, however, great disputes
ensue even on how to define the drag force and the added mass
force at those scales; the relative importance of these different forces
(say, the drag force and the added mass force) may be reversed
(De Wilde, 2005), and the resultant variation may be up to several
orders of magnitude (Li and Kwauk, 2001; Wang et al., 2010b).
Normally, these interphase forces play the decisive role in affecting
the reactor behavior. Therefore, how to quantify these forces, espe-
cially the drag force for fluidized beds, is of overwhelming impor-
tance (Agrawal et al., 2001; Li and Kwauk, 1994, 2001).
Figure 3 schematically shows the striking difference in literature, on
flow, mass transfer and reactions in terms of drag coefficient, Sherwood
number and reaction rate coefficient, respectively. For gas–solids riser
flow, different drag coefficients may cause variance up to three orders
of magnitude. Their general trends with increasing solids concentration
are also different. For mass transfer in a CFB, even higher difference up
to five orders of magnitude has been reported (Breault, 2006). For the
reaction rate of, for example, char combustion, the reported values
differ greatly by also several orders of magnitude due to the complex
properties and structures of char (Basu and Fraser, 1991; Nikss et al.,
2003), while it is hard to quantify such differences and no general
correlation is available. How to explain and quantify these structure-
induced differences in flow, transfers and reaction remain challenges
and also critical questions to chemical engineering research. In what
follows we will try to tackle these challenging issues with our tentative
answers by introducing multi-scale CFD approach.
[(Figure_3)TD$IG]
Figure 3 Literature discrepancies in drag coefficient, in mass transfer for CFB due to
meso-scale structures and in char reaction rate coefficient. For the drag coefficients,
curves are adapted from Wang et al. (2010); for the mass transfer, curves are adapted
from Dong et al. (2008a); for the coal reaction, different symbols refer to different
coal data.
10 Wei Wang et al.
2. MULTI-SCALE CFD—SOLUTIONS FOR HIGH

PREDICTABILITY AND SCALABILITY
To capture the meso-scale structure and/or to predict its effects, various

modeling approaches have been proposed. The spatiotemporal resolu-
tion of these approaches grows with the development of the computer
capacity, including the single-scale approaches, direct numerical simu-
lations, and multi-scale approaches.
2.1 Single-scale approach—its limitation on predictability

The simplest approach is the classic chemical engineering models which
include, for example, the plug flow model, the continuously stirred tank
reactor (CSTR) model and their hybrid models (Levenspiel, 1999), and so
on. The plug-flow model is a typical single-scale approach, in which the
multi-scale flow is reduced into a pipe flow with a characteristic velocity
and time scale, every flow elements passing through the pipe uniformly
without any structure. Such a model is an extreme simplification to the
real flow. To approximate the real flow, one may assemble a series of
plug flow models and CSTR models in various networking topologies to
fit the overall output in terms of the residence time distribution (RTD).
However, such agreement cannot be counted on too much as the classic
chemical engineering model in itself is a black box without spatial res-
olution of what happens inside and beneath.
2.2 Direct numerical simulations—its limitation on scalability

Computational fluid dynamics enables us to investigate the time-
dependent behavior of what happens inside a reactor with spatial res-
olution from the micro to the reactor scale. That is to say, CFD in itself
allows a multi-scale description of chemical reactors. To this end, for
single-phase flow, the space resolution of the CFD model should go
down to the scales of the smallest dissipative eddies (Kolmogorov
scales) (Pope, 2000), which is inversely proportional to Re3/4 and of
the orders of magnitude of microns to millimeters for typical reactors.
On such scales, the Navier–Stokes (NS) equations can be expected to
apply directly to predict the hydrodynamics of well-defined system,
resolving all the meso-scale structures. That is the merit of the so-called
DNS.
For multiphase flow that is normally encountered in fluidized bed
reactors, there are two kinds of definitions of the micro-scale: first, it is
the scale with respect to the smaller one between Kolmogorov eddies
and particles; second, it is the scale with respect to the smallest space
required for two-phase continuum. If the first definition is adopted, the
space resolution of the governing equations should be smaller than the

microlength scale, where the governing equations may have a variety of
choices, including NS equations, lattice Boltzmann method (LBM)
(Wolf-Gladrow, 2000), and smoothed particle hydrodynamics (SPH)
(Monaghan, 2005) and particle motion may be described by, for exam-
ple, discrete element method or discrete particle method. In this way,
each particle can be tracked individually with the fluid–particle inter-
actions described in classic Newton mechanics and the particle–particle
interaction with, for example, rigid body mechanics. Here the para-
meters such as the fluid viscosity, restitution coefficient and the elastic-
ity modulus are needed as input from lower scale theory or experimen-
tal measurements. For example, the kinetic theory of gas can be applied
to determine the gas phase viscosity for DNS.
If the second definition is adopted instead, as is the case of two-fluid
model (TFM), the so-called fine-grid TFM will take the place of DNS, by
which the subgrid closure relations may be obtained for the coarse-grid
TFM simulation. That is, coarse-grid TFM will behave as the large eddy
simulations (LES) or Reynolds-averaged Navier–Stokes (RANS) for
two-phase flows. Obviously, the critical issue of this kind of DNS
heavily relies on the accuracy of the basic governing equations for the
continua. It should be noted, the terms of ‘‘fine-grid’’ and the ‘‘coarse-
grid’’ that will appear repeatedly in this article are defined in a relative
manner, as follows: the ‘‘fine-grid’’ means that the grid is smaller
than the micro-scale and there is no need of subgrid modeling any-
more, while the ‘‘coarse-grid’’ means that the grid is larger than the
micro-scale; there are subgrid structures and subgrid modeling must be
included to guarantee the accuracy.
It has been widely recognized that the computational demand of any
type of the above DNS is tremendous. In the literature, DNS of gas–solid
suspension was performed over a domain that is comparable in size with
a computational cell and only thousands of particles are tracked. That
scale is far less than that involved in industrial reactors, which is nor-
mally of the order of meters, with amount of particles (e.g., FCC particle,
10–100 m) in the range of 1012–1015. As to temporal evolution, the time
step of the gas–solid DNS simulations is limited by both the Kolmogorov
time scale that is inversely proportional to Re1/2 and the particle colli-
sion time scale. The typical time step was reported less than microse-
conds (Ma et al., 2006) and the simulated physical time can hardly reach
the order of magnitude for real processes that may be in minutes or
hours. Though recent progress in GPU-based technology harnesses the
speedup of parallel computing by one to two orders of magnitude (Chen
et al., 2009; Xiong et al., 2010), it is still formidable, at least in foreseeable
future, to apply DNS to predict hydrodynamics of a reactor.
12 Wei Wang et al.
2.3 Multi-scale CFD approach—a compromise of predictability

and scalability
To meet the industrial demand for both large-scale computation and
good predictability, the reasonable way out is not to simulate from the
beginning of the micro-scale, but to use coarse-grid simulation with
meso-scale modeling for the effects of structure. This kind of approach
can be termed the ‘‘multi-scale CFD.’’ It is entitled ‘‘multi-scale,’’ not
because the problem it solves is multi-scale, but because its meso-scale
model contains multi-scale structure parameters.
According to the choice of meso-scale models, we may divide the
multi-scale CFD into two branches: the ‘‘correlative’’ and the ‘‘varia-
tional,’’ as named in Li and Kwauk (2003) and will be discussed in the
following sections.
2.3.1 Correlative
The ‘‘correlative’’ multi-scale CFD, here, refers to CFD with meso-scale
models derived from DNS, which is the way that we normally follow
when modeling turbulent single-phase flows. That is, to start from the
Navier–Stokes equations and perform DNS to provide the closure rela-
tions of eddy viscosity for LES, and thereon, to obtain the larger scale
stress for RANS simulations (Pope, 2000). There are a lot of reports
about this correlative multi-scale CFD for single-phase turbulent flows.
Normally, clear scale separation should first be distinguished for the
correlative approach, since the finer scale simulation need clear speci-
fication of its boundary. In this regard, the correlative multi-scale CFD
may be viewed as a ‘‘multilevel’’ approach, in the sense that each span
of modeled scales is at comparatively independent level and the finer
level output is interlinked with the coarser level input in succession.
Following the same methodology, one may start from the basic
governing equations for gas–solid two-phase flows and performs in
sequence the DNS, LES, and RANS for fluidization simulations. As
discussed above, depending on the space resolution of different DNS
approaches, one can further find two paradigms for that practice. The
first may be referred to Agrawal et al. (2001) and Igci et al. (2008), where
the effective drag coefficients for coarse-grid simulation were derived
from the fine-grid TFM simulation results over periodic domains.
Another paradigm starts from subparticle simulations; the effective
drag coefficient can be then obtained from lattice Boltzmann simula-
tions of gas flow through a fixed bed of particles (Van der Hoef et al.,
2006) or SPH simulations of a gas–solid suspension (Ma et al., 2006).
These subparticle simulations also allow in-depth understanding of
the applicability of their closure laws for higher scale simulations. For
example, Ma et al. (2009) found that the typical ‘‘drafting–kissing–
tumbling’’ process describing the instability of sedimentation of few

particles remains effective in shaping the heterogeneity in gas–solid
systems, but it was repeated on groups of particles rather than individ-
ual particles, which eventually leads to the formation of clusters, the
typical meso-scale structure in gas–solid suspensions. It is interesting to
note that, for the whole suspension, the balance between particle grav-
ity and the drag force on the particles are established almost immedi-
ately during this process, and vary very little thereafter, though indi-
vidual particles may experience acceleration and deceleration
constantly. However, the slip velocity between the gas and solid phases
increases in a fluctuating manner for a much longer time before reach-
ing a statistically plateau value, this is driven by, or in other words,
reflected by, the structural changes in the destabilizing process.
Clustering has apparently reduced the intensity of interphase friction,
which, according to the EMMS model, is both favorable for the solid
particles to reach minimum local voidage and the gas phase to produce
least dissipation. Xiong et al. (2010) found more distinct heterogeneous
structures in gas–solid suspensions with more solid particles. With the
help of massive parallel computation and recently by GPU computing,
they were able to simulate an area up to several square centimeters,
laden with 75-m spherical solid particles at a volume concentration of
about 15%, and the gas–solid density ratio above 1000. A preliminary
study on the scale-dependence of the heterogeneity was then possible.
As shown in Figure 4, they simulated up to 30,240 solid particles, and
found that local fluctuations, in both particle velocity and drag force,
increase with the size of the simulated domain, whereas the overall
temporal fluctuation of these quantities is still ablating with the increase
of domain size, which means the heterogeneity is indeed more signif-
icant in larger systems. However, an asymptotic scale-dependence is
evident in different quantities, such as the steady-state average slip
velocity, as wells as its fluctuation in time. That is, the rates of increase
slow down gradually as the domain size increase, and finally a plateau
of these quantities are reached. It is also in agreement with earlier
studies using TFMs (Agrawal et al., 2001; Ge et al., 2008; Wang, 2008)
and theoretical projection of the EMMS model. Together with the mesh-
dependence, these findings validate the use of (nearly) scale-indepen-
dent models for describing the properties of the heterogeneity on meso-
scale, such as the drag law and solid-phase stress law, at least in a
certain range of scales.
Following the above approach, the correlative multi-scale CFD
reduces the computation cost by transforming information over a range
of scales into meso-scale models. With these works, it seems feasible to
establish a numerical experiment facility to consider thoroughly the
14 Wei Wang et al.
[(Figure_4)TD$IG]
Figure 4 Snapshot from a dynamic DNS simulation of two-dimensional gas–solid

system with 30,240 solid particles (Xiong et al., 2010). (The right figure shows the
distribution of solid particles and to the left the gas velocity field is added; |v| in color
spectrum denotes the gas velocity magnitude.)
meso-scale effects of gas–solid systems, which is otherwise a formidable

task for both theoretical studies and experimental measurements. We
may expect that further development of DNS methods for gas–solid
system and the high performance computing technologies will allow
new breakthrough to realize it in future (Chen et al., 2009). On the other
hand, however, severe challenges have to be solved first to meet this
forthcoming possibility.
The first challenge lies in how to sufficiently sweep all the possible
states of flow structures over the wide range of scales. As discussed
above, the meso-scale structure is characterized with dynamic varia-

tion. Its effects are related inherently with accelerations and random
evolution of structures. The normal practice by accounting for these
effects with void fraction has proved to be insufficient (Wang et al.,
2010b). Therefore, how to introduce new variables for the dynamic
structure will be a challenge to both modeling and computation.
Another challenge to the correlative multi-scale CFD is how to incor-
porate the macro-scale influence into the meso-scale modeling. It is well
known that there could be backscatter contributions for the transfer of
energy in turbulent flows (Pope, 2000). In a fluidized bed, there are also
evidence that the meso-scale clusters are affected by both the macro-scale
operating conditions and the micro-scale interstitial flows around parti-
cles (Grace, 1996; Harris et al., 2002; Wang and Li, 2007). However, the
normal practice by performing a finer scale simulation over a small
periodic domain or a static bed of particles is only related with micro-
scale constraint. So, it will be a challenge for correlative methods as to how
to introduce macro-scale effects such as the geometric limitation and
operating conditions of a reactor. Such a challenge in itself reflects the
bilateral coupling/bridging difficulties in meso-scale modeling for most
of the multi-scale problems. Certain trans-scale criterion might be needed,
or, at least helpful, to untangle this challenge in correlative methods, and
this is right to the point that the variational methods will focus on.
2.3.2 Variational
The ‘‘variational’’ type of multi-scale CFD, here, refers to CFD with
meso-scale models featuring variational stability conditions. This
approach can be exemplified by the coupling of the EMMS model
(Li and Kwauk, 1994) and TFM, where the EMMS/matrix model
(Wang and Li, 2007) at the subgrid level is applied to calculate a struc-
ture-dependent drag force.
The variational stability condition describes the ‘‘compromise’’ among
various dominant mechanisms. In mathematics, the ‘‘compromise’’ can
be expressed as a multiobjective optimization (Li and Kwauk, 2003). Not
like the ‘‘correlative’’ multi-scale CFD where separate ranges of scales
can be distinguished over which the finer supplies for the coarser, the
‘‘variational’’ obtains correlations through the ‘‘compromise’’ among the
dominant mechanisms that are coupled over the investigated range of
scales. As a consequence, the scale separation is not a necessary condition
to the ‘‘variational.’’ In this regard, the ‘‘variational’’ multi-scale CFD is
inherently multi-scaled.
More details about the variational type of multi-scale CFD will be
addressed in following sections. As to the major challenges it confronts,
the first lies in how to distinguish the dominant mechanisms over such
16 Wei Wang et al.
broad range of scales for different systems. For a thermodynamic equi-

librium system, we know that the maximum entropy criterion deter-
mines its final state; for linear nonequilibrium systems, minimum
entropy production rate governs their behavior (Prigogine, 1967); while
for the nonlinear nonequilibrium system that is widely encountered in
chemical engineering, no single, universal criterion has been discovered
yet (Gage et al., 1966). That implies we have to search specific stability
conditions for different systems (Li and Kwauk, 2003). Recent work has
found some clues to establish a general strategy to distinguish the
‘‘dominant mechanisms’’ by analyzing the ‘‘compromise’’ between
them (Li and Kwauk, 2003; Li et al., 2004). This strategy has been
extended to six other systems besides the gas–solid fluidization, cover-
ing single-phase flow, gas–liquid flow, granular flow and emulsions,
and so on. More efforts are needed to generalize this strategy for wider
range of applications (Ge et al., 2007).
Another challenge to the ‘‘variational’’ multi-scale CFD lies in the
computing scheme, in the sense that the ‘‘dominant mechanisms’’ as
well as their ‘‘compromise’’ in terms of certain stability conditions may
relate with scales different from those of CFD computation. A clear
example for that situation can be found in gas–solid fluidization (Li
et al., 2004; Zhang et al., 2005), where locally the two dominant mechan-
isms for particles (e = min) and gas (Wst = min) can be realized alter-
nately with respect to time and space, with the term for characterizing
stability condition fluctuating intensively, while their compromise
leads to the stability condition (Nst = min) at the meso-scale. When
CFD computation is performed at a scale smaller than that, how to
incorporate the larger scale stability condition into the CFD description
of hydrodynamics will be a hard topic.
In all, there are different approaches for realizing multi-scale CFD,
each with distinctive characteristics. Table 1 summarizes the characteris-
tics and the challenges of these two kinds of multi-scale CFD approaches.
To manifest their respective utilization and features, in what follows we
will compare them with examples on gas-fluidized bed simulations.
2.3.3 Applications of the multi-scale CFD

Limited to computing capability, the following analysis confines the
DNS to the fine-grid TFM simulation, which offers meso-scale closures
for the correlative, coarse-grid TFM simulations. For comparison, the
variational type of multi-scale CFD takes the EMMS-based models to
close TFM simulations.
2.3.3.1 Periodic domain simulations. As to the fine-grid TFM,
we followed the scheme proposed by Agrawal et al. (2001) and
first performed simulations over periodic 2D domains, whose
Table 1 Comparison between the correlative and the variational types of multi-scale
CFD
Correlative (e.g., closure for Variational (e.g., coupling

TFM with DNS results) between EMMS and TFM)
Characteristics * Clear scale separation is * Clear scale separation is

necessary (multileveled) not necessary (intrinsically
multi-scaled)
* Computationally * Computationally saving
expensive
Challenges * Exhaustive search of all * Determination of
states of dynamic dominant mechanisms
structures and stability conditions
* Incorporation of macro- * Trans-scale coupling
scale influence into meso- scheme between CFD and
scale modeling meso-scale models
dimensionless size maintains the same value of 200 800 (scaled with
the particle diameter), which is comparable in physical size with a grid
in coarse-grid simulations for fine particles of Geldart group A.
The domain was meshed with uniform, square grids and refined
gradually to investigate the effect of grid size. The commercial
software Fluent1 6.2.16 was used as the solver of TFM. The solid stress
and drag coefficient terms therein need to be closed. For simplicity, the
algebraic form of the granular temperature equation derived from the
kinetic theory of granular flow (KTGF) (Gidaspow, 1994) was taken to
close the solids pressure and solids viscosity, and so on. The correlative
CFD adopts the hybrid drag coefficient combining Wen and Yu’s and
Ergun’s relations (Gidaspow, 1994) (for brevity, Model G), which has
been widely accepted as the standard relation, while the variational CFD
adopts the EMMS-based subgrid model, that is, EMMS/matrix (Lu et al.,
2009; Wang and Li, 2007) (for brevity, Model M). Model G was obtained
from homogeneous fluidization and packed bed systems, so it may be
taken as an extreme example without considering subgrid structures.
Model M depends on the structure that is resolved by EMMS. To
account for the effects of physical properties of the materials used,
three types of particles that belong to groups A, B, and D of Geldart
classification (Geldart, 1973), respectively, are compared as shown in
Table 2 (Lu et al., 2011). At the start of simulations, particles of all
cases were uniformly distributed with an identical solids volume
fraction of 5%. The periodic boundaries were prescribed in both
directions to keep constant solids concentration and the gravity of
18 Wei Wang et al.
Table 2 Physical properties of the fluidized systems simulated
Group A Group B Group D
Particle diameter dp, m 75 300 1,020

Particle density rp, kg/m3 1,500 2,500 4,000
Gas density rg, kg/m3 1.3 1.225 1.225
Gas viscosity g, Pa s 1.8 105
Terminal settling velocity uT, m/s 0.2184 2.18 8.3
Archimedes number Ar 24.9 2,499.8 157,233.3
particles was balanced by the imposed pressure drop along the vertical
direction. The drag force exerted on particles relates with the effective
gravity by b = eg(1 eg)(rp rg)g/us, so that the slip velocity is inversely
proportional to the drag coefficient. More details about the settings
should be referred to Lu et al. (2009, 2011).
For a given periodic domain (or a coarse-grid), the two-phase flow
will reach its quasi-steady state with slip velocity fluctuating around its
time-average value after a period of time. Figure 5 shows the variation
of this time-average, dimensionless slip velocity against grid size for
Geldart group A particles. Snapshots of the solids distribution are inset
to manifest the meso-scale clusters at corresponding grid resolutions.
For the Model G, as shown in Figure 5a, the predicted slip velocity
increases with grid refining and finally approaches an asymptote when
the grid is thinning to the size as small as several particle diameters.
[(Figure_5)TD$IG]
Figure 5 Effect of grid resolution (l) on the time-averaged dimensionless slip

velocity (us/uT). Geldart group A particles are used. The ordinate is scaled with the
terminal velocity of single particles (uT 21.84 cm/s) and the abscissa is scaled with
the particle diameter dp. The domain size is 1.5 6 cm2, comparable to the coarse-grid
used in normal simulations.
(After Wang et al., 2010b.)
During this refining process, the slip velocity roughly doubles, and
accordingly, the effective drag coefficient reduces by 50%. This result
coincides with the findings of Agrawal et al. (2001). The asymptotic slip
velocity implies that there is a threshold for the micro-scale, below
which the DNS of TFM can be used to predict the subgrid drag coeffi-
cient for coarse-grid simulations. It is interesting to note that, when
using Model M, over the investigated range of l/dp from about 1 to
100, the solution only weakly depends on the grid size (or even grid
independent), as shown in Figure 5b.
Figure 6 gives more comparison for coarser particles. For Geldart
group B particles, the slip velocity predicted by the Model G only
depends weakly on the grid size, and its value is close to unity. This
phenomenon is obviously different from what is observed for Geldart A
particles. When the Model M is employed, there exists a similar trend,
but with different values. The slip velocity predicted by using Model M
is about two times that by using Model G. This difference should be
attributed to the subgrid meso-scale modeling in the EMMS/matrix
model. For Geldart group D particles, again, results of both models
remain almost grid independent, and both of their predicted slip veloc-
ities are smaller, and much closer to unity. This gradual reduction of the
dimensionless slip velocity reflects the decaying effect of meso-scale
structures with increase of particle diameter (or Ar). By comparison,
Jin et al. (2010) pointed out that, in isotropic turbulent flow laden with
heavy particles, with increase of Stokes number (St > 1), particles
respond to the eddies with larger time scales relative to the
Kolmogorov eddies, and then, the level of accumulation drops and
[(Figure_6)TD$IG]
Figure 6 Effect of the periodic domain size on the time-averaged dimensionless slip
velocity (us/uT).
20 Wei Wang et al.
particles are more uniformly distributed. More efforts are needed to

clarify the relevant mechanisms.
Besides the obvious value difference mentioned above, it is also
natural to question why these two drag models cause such different
trends. The answer is that there is substantial difference between using
the Model G and the Model M. For Model G, the simulations with grid
size larger than the threshold value, for example, l/dp > 10 for FCC
particles in Figure 5, are coarse-grid without consideration of structures,
and hence inadequate for providing the subgrid closure. For Model M,
however, the model in itself contains subgrid structure modeling. That
is, across the investigated range of grid size, all the simulations can be
viewed as ‘‘coarse-grid’’ with subgrid meso-scale modeling. If Model M
accurately captures the grid-size dependency of the structure (i.e., the
core for a subgrid meso-scale model), all these simulations should pre-
dict the same, accurate drag closure, for the larger domain, irrespective
of what grid resolution they are using. Therefore, the Model M seems
satisfying this premise for accuracy, and hence predicts almost the same
value of slip velocity, showing a grid-independent trend.
As we know, the structure effect varies with domain size—the effec-
tive drag coefficient will decrease with increase of domain size before
reaching a plateau of slip velocity (Agrawal et al., 2001; Wang, 2008).
Therefore, an appropriate meso-scale model should take into account
the domain size (or filter size in Andrews et al., 2005), either explicitly or
implicitly. The typical LES model takes into account this effect by a
length scale that selects the minimum resolved eddy size, D, explicitly
(Ferziger, 1993). In Model M (EMMS/matrix), the grid size is not pre-
sented explicitly. However, in its second step, as will be detailed in
following sections, the slip velocity that varies with grid size was intro-
duced besides commonly used voidage, which means the grid size was
taken into account implicitly, and so was the dynamic factor of the
structure, by allowing variation of slip velocity. That may be one of
the reasons why the Model M predicts an almost grid-independent
solution. More details should be referred to (Wang et al., 2010b) to
understand why it is not sufficient to take into account the meso-scale
structure effects only by using a function of voidage.
To verify this domain-size dependency, the Model M was further
applied in a larger periodic domain (3 12 cm2) for FCC particles.
Figure 7 shows the comparison between the results in Figure 5b and
those of larger domain (3 12 cm2). It is clear that larger domain leads to
higher slip velocity and then lower drag coefficient, which is in agree-
ment with our expectation and the other research results (Agrawal et al.,
2001; Wang, 2008). Again, we can find across the whole range of grid
size that the Model M predicts almost unchanged value of slip velocity,
[(Figure_7)TD$IG]
Figure 7 Effect of the periodic domain size on the time-averaged dimensionless slip
velocity (us/uT) with Model M. Geldart group A particles. The red triangles denote the
cases with respect to the domain size of 3 12 cm2 and the blue squares 1.5 6 cm2
confirming that the Model M accurately captures the mechanism

behind the structure.
It should be noted that, Model G and Model M predict quite different
values of slip velocities, though their resolved structure may look sim-
ilar, as shown by insets of Figure 5. Extending this seeming inconsis-
tency to larger scales, we may expect that, simulations of real reactors
with Model G and Model M may predict different solids flux even with
similar impression of structures. Then, it is natural to question, which
solution of these two models coincides with the reality. To answer this
question, simulations of CFB risers are needed to test which will agree
with experiments.
2.3.3.2 Simulations of risers and validations. Two CFB risers were

selected. One is the IPE riser, with inner diameter of 90 mm and
height of about 105 m and FCC particles fluidized (Li and Kwauk,
1994). The other was constructed by the group of Lothar Reh in ETH
(Herbert et al., 1999), with inner diameter of 411 mm and height of 8.5 m
and glass beads fluidized, whose configuration has been detailed in
literature (Zhang et al., 2008). In simulations, the solids fluxes were not
given, instead, the solids entrained out of the risers were circulated into
the inlets to keep the solids inventories constant.
Figure 8 compares Model M with Model G in terms of their predic-
tions of the axial profiles of voidage under various grid resolutions. For
FCC particles, when using Model G, the solids were distributed uni-
formly across the riser height. It seems that the grid refining has little
22 Wei Wang et al.
[(Figure_8)TD$IG]
Figure 8 Axial profiles of cross-sectionally averaged voidage under different grid

resolutions for IPE and ETH risers.
(Adapted from Lu et al., 2009, 2011.)
effects for this case. Moreover, the predicted profiles deviate from the
experimental data significantly, owing to its overpredicted drag coeffi-
cient. The prediction improves when using Model M, by which the
characteristic S-shaped profile of voidage—a dilute top coexisting with
a dense bottom in the axial direction—was reproduced. The solids flux
was also overpredicted when using Model G, its time-averaged value
was around 170 kg/(m2 s), about 10 times the experimental data of
14.3 kg/(m2 s). Updating to Model M improves very much, with its
predicted solids flux at around 19 kg/(m2 s) and close to reality. In
addition, such improvements have been found also applicable to the
other riser cases (Lu et al., 2009).
For glass beads in the ETH riser, the macro-scale structures seem to
be similar for different approaches, as the predicted voidage profiles
using both drag models agree well with the experimental data.
However, their predicted solids fluxes still differ much from each
other—the Model G predicts values near 400 kg/(m2 s), which is about
three times higher than the measured data (151 kg/(m2 s)); while the
Model M predicts values around 147 kg/(m2 s), which are in good
agreement with the data. The other comparisons in terms of radial
profiles of solids distribution were also found in favor of the Model
M. So, it seems that the evaluation based on the time-averaged axial
profiles of solids distribution is insufficient; the circulating solids flux,
which represents the dynamic characterization of the structures at the
macro-scale, should be additionally examined, to judge whether an
approach correctly captures the two-phase flow behavior. From the
above analysis we can see that, at least for the current cases, the con-
ventional TFM seems inadequate to describe gas–solids riser flows
featuring multi-scale heterogeneity while EMMS-based multi-scale
CFD improves its accuracy greatly.
2.3.3.3 Scope of applications. In literature, there are some other reports

concerning the scope of applying various types of multi-scale CFD. For
example, Benyahia (2010) found that the subgrid drag coefficients
obtained from filtered drag model (Igci et al., 2008) and EMMS
model (Yang et al., 2004) are both needed and useful in large-scale
simulations. The solids used in his riser are also FCC particles, but
heavier (1712 kg/m3). The riser was operated under higher gas velocity
(5.2 m/s) and higher solids flux (489 kg/(m2 s)). Wang et al. (2009)
addressed the applicability of TFM for bubbling fluidized bed by
comparing it with discrete particle model (DPM) results. They
reported that TFM can predict the correct bed expansion even without
drag modifications, provided that a sufficiently fine-grid size and small
time step is used. Based on scale separation analysis, Wang and Li (2007)
have ever addressed that for dense gas–solids flow or bubbling fluidized
beds, the subgrid structure modeling may be not needed (or not that
important), but for comparatively dilute CFB, it is necessary. By and
large, we may say that there are different application scopes for the
correlative and the variational types of multi-scale CFD. In
appearance, these differences are related with the operating conditions
and physical properties of the investigated systems. As to the
underlying mechanism, more efforts are still needed, and a tentative
explanation may start from the scale separation analysis, as follows.
For the gas–solids two-phase flow in a CFB riser, there is no clear
separation between the micro-scale (fine-grid scale of TFM continua)
and the meso-scale (cluster scale), as discussed in Goldhirsch (2003) and
Wang and Li (2007). Thus, the meso-scale structure does not vanish
24 Wei Wang et al.
with grid refining of TFM. That is to say, there are certain subgrid terms
that cannot be solved under the conventional framework of TFM. That
is probably the reason why the fine-grid simulations of risers in Figure 8
fail to predict the circulating solids flux. If the EMMS-based Model M
was used instead, there will be intrinsic structure terms in the conser-
vation equations, in the sense that the conservation applies to both the
dense phase and the dilute phase, respectively. That allows capturing
the intrinsic structures. For bubbling fluidized beds or some other dense
flow cases, the micro-scale and the meso-scale may be separated clearly
as analyzed in Wang and Li (2007). That is to say, the meso-scale
structure is larger than the grid size with its characteristic relaxation
time longer than computational time step. In that case, fine-grid TFM
may be sufficiently precise to capture all the meso-scale structure and
then no modification to the conventional drag coefficient is needed.
In summary, we may expect that the correlative type of multi-scale
CFD can be used for the problems with clear scale separations between
the micro-scale and the meso-scale, while the variational type, provided
with appropriate stability condition, seems free of such limitation. In
what follows we will detail some examples of the variational approach
by introducing its basis of the EMMS model.
3. MESO-SCALE MODELING—THE KEY TO MULTI-SCALE

APPROACHES
3.1 Energy-minimization multi-scale (EMMS) model—a

meso-scale model
The EMMS model was first proposed for the hydrodynamics of con-
current-up particle–fluid two-phase flow. Though it is based on a rather
simplified physical picture of the complex system (Li, 1987; Li and
Kwauk, 1994), it harnesses the most intrinsic complexity in the system,
the meso-scale heterogeneity, and this is why it allows better predic-
tions to the critical phenomena in the system which is obscured in other
seemingly more comprehensive models.
In this model, instead of the uniform and interpenetrating continu-
ous phases of the gas and the solids, a distinct heterogeneous structure
is assumed. The elemental volume in the flow field, which has dis-
played observable heterogeneity, is partitioned into fractions occupied
by the gas-rich, dilute phase (denoted by subscript ‘‘f’’) and the particle-
rich, dense phase (denoted by subscript ‘‘c’’), respectively. Within each
‘‘phase,’’ uniformity is assumed, and the dense ‘‘phase’’ is assumed to
occur as spherical clusters. That is, the dense phase is discrete, sur-
rounded by the continuous dilute phase. In this way, eight variables
is needed to define its steady state, namely, the superficial particle and
fluid velocities of the dense and dilute phases (Upc, Upf and Ugc, Ugf),
the voidages in each phase (ec and ef), the volume fraction of the dense
phase (f), and the cluster diameter of the dense phase (dc).
To facilitate the discussion in the rest part of this article, we revisit
the formulation of the EMMS model, while interested readers are
referred to Li and Kwauk (1994) and Li et al. (2005) for more details.
1. Dilute-phase momentum balance: all effective particle weight in
unit dilute-phase volume is balanced by the fluid drag:
3 1 ef
CDf rg U 2sf ð1 ef Þðrp rg Þg ¼ 0; ð1Þ
4 dp
where Usf = Ugf Upff/(1 f) is defined as the dilute-phase super-

ficial slip velocity.
2. Dense-phase momentum balance: effective particle weight in unit
dense-phase volume is partially supported by the dense-phase
fluid flow, and the rest is supported by the bypassing dilute phase
fluid flow,
3 1 ec 3 f
CDc rg U 2sc f þ CDi rg U 2si fð1 ec Þðrp rg Þg ¼ 0: ð2Þ
4 dp 4 dc
3. Pressure balance between dense and dilute phases: the dense
phase pressure drop is balanced by that of the dilute phase plus
the ‘‘interphase,’’
1 ef f 1 1 ec
CDf rg U 2sf þ CDi rU 2si CDc rg U 2sc ¼ 0: ð3Þ
dp 1f dc dp
4. Continuity of the fluid phase:

U g ¼ fU gc þ ð1 fÞU gf : ð4Þ
5. Continuity of the solids:

U p ¼ fU pc þ ð1 fÞU pf : ð5Þ
6. Cluster diameter:
dc ððrp rg ÞgU p Þ=rp ð1 emax Þ N st;mf
¼ ; ð6Þ
dp N st N st;mf
26 Wei Wang et al.
where emax denotes the maximum voidage existent for heterogeneous

particle–fluid flow (Matsen, 1982) and the subscript ‘‘mf’’ denotes the
state of minimum fluidization. Nst is the energy consumption for sus-
pending and transporting the solids with respect to unit mass, which
can be calculated from

W st rp rg ef eg
N st ¼ ¼ Ug fð1 fÞU gf g; ð7Þ
ð1 eg Þrp rp 1 eg
where

rp rg U p emf
N st;mf ¼ U mf þ g: ð8Þ
rp 1 emf
These six equations are insufficient to give a closure of the EMMS

model that involves eight variables. The closure is provided by the most
unique part of the EMMS model, that is, the introduction of stability
condition to constraint dynamics equations. It is expressed mathemat-
ically as Nst = min, which expresses the compromise between the ten-
dency of the fluid to choose an upward path through the particle sus-
pension with least resistance, characterized by Wst = min, and the
tendency of the particle to maintain least gravitational potential, char-
acterized by eg = min (Li and Kwauk, 1994).
A direct application of the model and demonstration of its predict-
ability is the prediction of choking point in fast-fluidization. The abrupt
change in particle concentration in the risers of circulating fluidization
beds with the continuous variation of gas or solids flux has long been a
controversial issue. With the EMMS model, it is now apparent that
regime transition is related to this phenomena where the dominant
mechanism in the system shifts from particle–fluid compromising to
fluid dominating (Li and Kwauk, 1994), and mathematically, it is a jump
between two branches of the stable solution (Ge and Li, 2002), as shown
in Figure 9. This ability of the EMMS model has shown its practical
significance in the designing and scaling-up of industrial fluidization
systems as will be discussed in more detail later.
3.2 Coupling of EMMS and CFD

The EMMS model was proposed for the time-mean behavior of fluid-
ized beds on the reactor scale. A more extensive application of the
EMMS model to gas–solid flow is through its coupling with the two-
fluid CFD approaches, which brings about an EMMS-based multi-scale
CFD framework for gas–solid flow. For this purpose, Yang et al. (2003)
introduced an acceleration, a, into the EMMS model to account for the
[(Figure_9)TD$IG]
Figure 9 Variation of Nst with ec and ef for the FCC–air system (Gs = 50 kg/(m2 s)).
Two minimum points in blue correspond to the dilute and the dense flow,
respectively, and their coexistence corresponds to the choking state of fluidization
(Ge and Li, 2002).
force imbalance between the effective gravity of particles and the drag
force, and the local voidage was also introduced as a new known
quantity. In this way, a simplified structure-dependent drag coefficient
was proposed, and it was found being able to greatly improve the TFM
prediction in terms of the axial profiles of voidage as well as the circu-
lating solids flux. Without EMMS correction, the conventional TFM
simulation was found to overpredict the solids flux by almost 10 times,
while the EMMS-based approach correctly predicts the solids flux and
the axially S-shaped profiles of voidage (Yang et al., 2004).
Later on, based on the in-depth understanding of the applicability of
the stability condition (Li et al., 2004; Zhang et al., 2005), the EMMS
model was extended, to describe the meso-scale structures at the sub-
grid level (Wang and Li, 2007). In more detail, first, as mentioned above
for the challenges to the variational type of multi-scale CFD, though the
hydrodynamic conservation equations were established on the micro-
scale of continuum, the stability condition of the EMMS model was
found applicable only to the higher meso-scales (Li et al., 2004). Such
a mismatch of scales is the main problem encountered in coupling
EMMS and TFM. To coordinate the hydrodynamics and the stability
condition at different scales, the extended EMMS model (named after
EMMS/matrix) adopts a two-step scheme, as follows:
Step 1:
The first step is to determine the meso-scale parameters in terms of the
diameter and voidage of clusters (dc and ec) with the constraint of the
28 Wei Wang et al.
global stability condition Nst ! min. These two parameters are the time-
mean characterization of the clusters, leaving the fluctuating informa-
tion to the second step in terms of velocities and accelerations. The
relevant momentum conservation equations in the first step are as
follows (strictly speaking, the parameters in Step 1 are cross-sectionally
averaged variables, and should be bracketed with < >. For simplicity,
we neglect these angle brackets):
3 1 ec
F1 ðXÞ ¼ CDc rg jU sc jU sc ð1 eg Þðrp rg Þðac þ gÞ ¼ 0; ð9Þ
4 dp
3 1 ef
F2 ðXÞ ¼ CDf rg jU sf jU sf ð1 ef Þðrp rg Þðaf þ gÞ ¼ 0; ð10Þ
4 dp
3 f
F3 ðXÞ ¼ CDi rg jU si jU si fðrp rg Þðeg ec Þðai þ gÞ ¼ 0; ð11Þ
4 dc
where the pressure drop balance assumption relates ai with the other
two inertial terms by
ð1 fÞ½ð1 eg Þðac þ gÞ ð1 ef Þðaf þ gÞ
ai ¼ g: ð12Þ
fðeg ec Þ
The mass conservation equations of the gas and particles are the same
with the original, as follows:
F4 ðXÞ ¼ U p U pf ð1 fÞ U pc f ¼ 0; ð13Þ
F5 ðXÞ ¼ U g U gf ð1 fÞ U gc f ¼ 0: ð14Þ
The average voidage eg relates ef and ec by

F6 ðXÞ ¼ eg f ec ð1 fÞef ¼ 0: ð15Þ
The definition of cluster diameter dc remains the same with the

original (Equation (6)), which read
dc ððrp rg ÞgU p Þ=ðrp ð1 emax ÞÞ N st;mf
F7 ðXÞ ¼ ¼ 0; ð16Þ
dp N st N st;mf
where the energy consumption for suspending and transporting parti-

cles, Nst = min, as is in the original form.
It should be noted that if the inertial terms are omitted, the above
relations will return to the original form of the EMMS model (Li and
Kwauk, 1994). For specified conditions Ug, Gs, and eg, this set of 10
variables X = (Ugc, Upc, Ugf, Upf, ef, ec, f, dc, ac, af) in the model can be
determined by satisfying seven equations, F(X), under constraint of the
minimization of Nst. These seven equations are nonlinear; their solution
follows a scheme that is similar to that used for the original EMMS
model, whose detail is referred to Wang and Li (2007). With the above
scheme, the variation of dc and ec can be determined as functions of
cross-sectionally averaged voidage <eg>. In practice, at least for
Geldart A particles, Nst ! min requires the maximization of ef and
minimization of af, that is, ef ! emax and af ! g. Such two relations
can be used to reduce the computation. Alternative models for dc and ec
may be expected to improve the EMMS-based models; one of such
efforts can be referred to the work of Wang et al. (2008a).
Step 2:
In the second step, dc and ec are known parameters ready for using.
The EMMS model is coupled with TFM in this step—with input from
CFD results, that is, the gas and solids velocities (ug and up) and voidage
eg in each grid, the remaining variables of EMMS, that is (Ugc, Upc, f) for
the dense phase and (Ugf, Upf) for the dilute phase as well as the inertial
terms associated to each phase, can be determined to supply the subgrid
structure parameters for calculating the drag coefficient. The algorithm
is similar to the step 1, but can be simplified according to the relative
invariance, by reorganizing the conservation equations as functions of
slip velocities (Lu et al., 2009). After simplification, only three unknown
variables (Usc, Usi, ac) are needed to solve and the relevant equations are
as follows:
3ð1 ec Þ
CDc rg jU sc jU sc ¼ ðrp rg Þð1 eg Þðac gÞ; ð17Þ
4dp
3
CDi rg jU si jU si ¼ ðrp rg Þðeg ec Þðai gÞ; ð18Þ
4dc
ef ð1 eg Þ
U si ¼ ðU s fU sc Þ : ð19Þ
ef eg
The EMMS-based drag coefficient was calculated by
ðrp rg Þe2g
b¼ ð1 eg Þðac gÞ: ð20Þ
Us
Here the vector division results in a scalar and the vectors on both
numerator and denominator have the same orientation. The relevant
algorithm should be referred to Lu et al. (2009).
30 Wei Wang et al.
Strictly speaking, the values of dc and ec used in the step 2 should be

calculated as functions of the cross-sectionally averaged voidage hegi, as
the step 1 of the algorithm is performed with global operating condi-
tions. In practice, however, correlating them with local voidage was
found only have trivial effects on the final prediction. So, for simplicity,
we may use local voidage in simulations (Lu et al., 2009).
Following the above scheme, we can calculate a structure-depen-
dent drag coefficient. To compare with the correlations in literature, we
take the homogeneous drag coefficient b0 from Wen and Yu (1966) as
the scale and define the heterogeneity index HD with (HD b/b0). Here
b refers to all the other drag coefficients. As shown to the left of
Figure 10, there are big discrepancies up to several orders of magnitude
on the current correlations. By comparison, the right-hand side of
Figure 10 shows a typical surface of HD calculated with EMMS/matrix
for an air–FCC system as a function of local Reynolds number Rep and
voidage. For visualization, only the vertical slip velocity is used here. It
is obvious that the meso-scale modeling results in significant variation
that covers the range of difference in literature, and therein the voidage
appears to be the dominant factor. In most of the range of voidage, gas,
and solids tend to ‘‘compromise’’ to reach each of their dominance. That
is, the particles tend to aggregate for least gravity potential and the gas
tend to flow around aggregates with least resistance. In this way, the
gas–solids interphase momentum transfer decreases and then the effec-
tive drag coefficient is less than that for homogeneous suspensions (i.e.,
HD < 1.0). At the two ends of voidage spectrum, HD approaches unity
corresponding to the homogeneous states of packed bed and extremely
dilute flow, where two-phase ‘‘compromise’’ gives way to particle
dominance and gas dominance, respectively. Higher slip velocity
[(Figure_0)TD$IG]
Figure 10 A tentative answer to disputes on drag coefficient with a typical surface of

the heterogeneity index for a FCC–air system (rp = 930 kg/m3, dp = 54 m, Ug = 1.52 m/s,
Gs = 14.3 kg/(m2 s), emf = 0.4, emax = 0.9997).
means gas dominance is strengthened and then formation of aggregates

is suppressed, leading to higher values of HD. In practice, for the dilute
flow in large-scale gas-fluidized beds, for example, CFB boilers, the gas
turbulence may have important effect on the two-phase flow distribu-
tion, and thus, the dominant mechanisms accounting for turbulence
may need to be considered (Li et al., 1999). A preliminary attempt to
include this effect can be referred to Lu (2009) for CFB boiler modeling.
Figure 11 shows some simulation results of the above-mentioned
ETH CFB using the EMMS/matrix model (Zhang et al., 2008). The
simulation was performed on 3D, full-loop geometry. To the left is a
snapshot of the solids distribution at the wall. Axial coexistence of a
dilute top and a dense bottom can be clearly seen in the riser, with
clusters falling along the riser wall as well as in the cyclone. A dense
bottom was also formed in the downer side, where the higher pressure
drives solids recycling through the siphon valve to the riser. The next to
the snapshot shows an axial profile of the cross-sectionally averaged
solids volume fraction. The right-hand side shows the radial profiles of
the time-average solids volume fraction and the time-average solids
velocity. In general, the simulation agrees well with the experimental
[(Figure_1)TD$IG]
Figure 11 A typical snapshot of the simulated solids distribution at the wall of the
ETH CFB, along with the axial profile of the cross-sectionally averaged solids volume
fraction and the radial profiles of time-average solids volume fraction and solids
velocity (rp = 1400 kg/m3, dp = 60 m, Ug = 3.5 m/s, H0 = 1.7 m).
(Adapted from Zhang et al., 2008.)
32 Wei Wang et al.
data. The characteristic core-annulus structure in radial direction is also

captured. It should be noted that, for this case, the grid size is around
9 mm in radial direction, which is about 150 times the particle diameter.
That meshing is rather coarse for the conventional TFM simulation
according to what has been obtained in periodic-domain simulations,
as shown in Figure 5. For EMMS-based multi-scale CFD, however, the
grid size is still within the acceptable range. So, we can say that the
EMMS-based multi-scale CFD reduces the load of computing to a large
extent.
Another merit of the EMMS-based multi-scale CFD can be repre-
sented by its ability to predict flow regime diagrams of CFB. In practice,
how to situate a reactor to an appropriate regime remains a critical issue
for both design and operating. Running in an inappropriate flow regime
may cause severe instability. Figure 12 shows calculated flow regime
diagrams in forms of a series of iso-ug, which relates the solid flux Gs
with the total pressure drop of riser at specified superficial gas flow rate
Ug. Some characteristic snapshots of solids distribution are also drawn
as insets. To the left is the diagram for an air–FCC system that is
calculated with the simplified EMMS drag model (Yang et al., 2003),
while to the right is that for an air–HGB (hollow glass beads) system that
is calculated with EMMS/matrix model. At least three regimes can be
distinguished in these two diagrams, that is, the dilute transport, the
dense upflow and in between the ‘‘choking’’ or the ‘‘continuous’’ tran-
sitions. The coexistence of the dense upflow with the dilute transport
marks the choking, which occupies the bell-shaped areas where the iso-
ug levels off with Gs equal to the saturation carrying capacity. The
summit of the bell-shaped area can be named after the ‘‘critical point,’’
above which the continuous nonchoking transitions bridges between
[(Figure_2)TD$IG]
Figure 12 Calculated flow regime diagrams for an air–FCC (dp = 54 m, rp = 930 kg/m3)
system and an air–HGB (hollow glass beads, dp = 75 m, rp = 609 kg/m3) system
(Wang et al., 2008).
the dilute transport and the dense upflow. Quantitative comparison

also manifests good agreement between this simulated regime diagram
and the experimental (Li et al., 2007; Wang et al., 2007).
The flow regime in Figure 12 is an ‘‘apparent flow regime’’
(Wang et al., 2008) in that it is associated with not only hydrodynamics
but also geometric factors. It was found that this apparent flow regime
changes gradually with the riser height (Li and Kwauk, 1994; Wang
et al., 2008), as shown in Figure 13a. Higher riser reduces the inlet/outlet
effects and has a higher ratio of riser height for a flow state to be fully
developed, thus leading to wider area for the choking transition. A short
riser whose whole length is affected by the inlet/outlet effect may be
hard to discern the choking transition. Recent experiments on a cold
model of CFB boiler validate this tendency (Hu et al., 2009).
The flow regime that is purely defined by hydrodynamics was
named after the ‘‘intrinsic flow regime’’ (Wang et al., 2008).
Figure 13b gives the intrinsic flow regime diagram calculated with
the original EMMS model for the air–FCC system. Similar partitions
of the flow regimes can be distinguished over the investigated range of
Gs and the averaged solids volume fraction es0 (es0 = DPimp/(rpgH)).
Clearly, the intrinsic choking transition area is sloping and larger than
that of the apparent one.
Comparing the apparent and the intrinsic diagrams in Figure 13, we
may conclude that, the apparent flow regime and the choking transition
area change with the riser height. Accordingly, the critical point will rise
to a higher position with increase of riser height. The upper limit of this
expansion is the intrinsic flow regime that is purely defined by
[(Figure_3)TD$IG]
Figure 13 The apparent flow regime diagram calculated with EMMS-based

multiscale CFD and the intrinsic flow regime diagram for the air–FCC system (fluid
catalytic cracking particle, dp = 54 m, rp = 930 kg/m3) calculated by using the EMMS
model without CFD. The intrinsic flow regime diagram is independent of the riser
height (Wang et al., 2008).
34 Wei Wang et al.
[(Figure_4)TD$IG]
Figure 14 Riser height decides the variation from apparent to intrinsic flow regime
diagrams. Dark cyan columns represent different riser heights with relevant flow
regime diagram sketched above, and the curve denotes the variation of the critical
point with the final end of intrinsic critical point (Wang et al., 2010b).
hydrodynamics, as depicted in Figure 14. In our opinion, this depen-

dency of the flow regime on the riser height is at least one of the major
reasons that cause disputes in literature (say, e.g., the review articles of
Bi et al., 1993 and Yang, 2004) about understanding the choking phe-
nomena. In practice, different research groups perform choking studies
with different designs of CFB in terms of riser height and the other
geometric factors. As a result, different understanding of the choking
occurs with different quantifications of their apparent flow regime
transitions. It is very hard, if not impossible, to unify all the experimen-
tal findings from different research groups, or to perform experiments
getting rid of all the geometric factors. The solution to this puzzle may
be twofold: one is to add an axis of the riser height (some other geo-
metric factors may also affect) to the flow regime diagram, to allow
understanding the complex interrelations between hydrodynamics and
geometric limitations. We called this diagram the ‘‘operating diagram’’
(Wang et al., 2008). The other is to study these phenomena through
virtual experiments of a series of 3D, full-geometry simulations of
CFB, as the virtual experiments are much easier to control the manifold
factors and also getting more reliable and cheaper with rapid develop-
ment of computing technologies. Our recent attempt has unfolded the
latter approach with fresh insight (Zhang et al., 2008). More efforts are
expected to finally reach the common knowledge of the flow regime
transitions of CFB.
3.3 Application of EMMS to mass/heat transfer and reactions

Meso-scale structure significantly affects the mass transfer but receives
less research attention. The first reason for this ignorance lies in that the
mass transfer is coupled with the momentum transfer, making its
dependency on the structure more complicate. Available few efforts
show that the determination of the mass transfer rate, either experimen-
tal or theoretical, is difficult, especially with variation of dynamic struc-
ture (Wang, 2002; Wang et al., 2005; Yu and Jin, 1994). The second reason
may relate with the underestimation of the relative role of the mass
transfer on the overall reaction rate. For a CFB reactor, fine particles
with large specific area react to the surrounding gas under high slip
velocity. It is normally expected in such a case, the fluid–particle mass
transfer rate is very high and therefore the overall reaction rate is
controlled by the other lower rate steps such as the intrinsic reaction
rate. For example, if the ozone decomposition on a FCC particle was
modeled, its reaction coefficient kr was reported of the order of magni-
tude of 10 s1, while the overall mass transfer coefficient kpap was of the
order of 105 s1 (Dong et al., 2008a; Ouyang et al., 1995), then the differ-
ence of four orders of magnitude can be expected using such micro-
scale analysis and that (Da = kr/kpap 1) means the mass transfer
modeling is negligible. However, the meso-scale structure may reduce
the overall mass transfer rate, in the same way as it affects the drag
coefficient, making Da 1. The last but not the least reason lies in that
the intrinsic reaction rate for the most of heterogeneous systems is hard
to measure. For example, the combustion of a char particle may relate
with manifold factors such as devolatilization, attrition and inner dif-
fusion inside the ash layers, and so on, while each of these processes is
very hard to present a generalized model. For petroleum refinery pro-
cesses as another example, the complex network may involve hundreds
or even thousands of elementary reactions that are very hard to deter-
mine. As a result, we are short of the intrinsic reaction rate data, and this
situation directly hinders the application of mass transfer modeling.
The meso-scale structure promotes gas bypassing clusters, and thus,
decreases the effective interphase mass/heat transfer rate. Following
the EMMS model for resolving the structure, a multi-scale mass transfer
model, EMMS/mass (Dong et al., 2008a, 2008b), was proposed. With the
two-phase structures calculated from the EMMS/matrix model, the gas
concentration in the dense phase and in the dilute phase were distin-
guished at the subgrid level. The mass conservation equations for these
36 Wei Wang et al.
two subgrid continua were then derived in a way similar to the con-
ventional equations for mixture concentrations, as follows:
Mass conservation for the gas mixture:
@
ðf e r Þ þ r ðfk ek rg ugk Þ Sk G k ¼ 0: ð21Þ
@t k k g
Mass conservation of component A in the gas mixture:
@
ðf e r Y Þþr ðfk ek rg YAk ugk fk ek rg Dm rYAk ÞSk G Ak ¼ 0; ð22Þ
@t k k g Ak
where subscript k denotes the phase k (k = c, dense phase; k = f, dilute
phase); the volume fraction fk is f or (1 f) for the dense phase or dilute
phase; YAk denotes the mass fraction of A in the phase k, and it relates
with the averaged concentration by
X
e g rg YA ¼ ðfk ek rg YAk Þ: ð23Þ
k¼c;f
Dm denotes the molecular diffusion coefficient; G k denotes the inter-

phase mass exchange rate between the dense and the dilute phases and
G c = G f, which can be directly calculated with EMMS/matrix model
parameters if the reaction source term, Sk, is negligible compared to the
bulk gas conservation. For vaporization of A, the source term reads
Sk ¼ fk kk ð1 ek Þap rg ðYA;sat YAk Þ; ð24Þ
where kk denotes the mass transfer coefficient between gas and particles
in homogeneous suspension of phase k, and various empirical relations
can be used in this regard, in which most of them take the form of Ranz
relation (Ranz, 1952). The meso-scale mass exchange rate of A, G Ak, can
be approximated by two parts of contributions: one is from a stable part,
just like a virtual ‘‘big particle’’; the other is from a dynamic part via
cluster-interface renewal or transformation. The stable part can be cal-
culated with the classic surface renewal theory (Bird et al., 1960) or the
above mentioned Ranz empirical relations; the dynamic part can be
approximated by G k YAy if the interphase exchange is dominated by
convection, where the subscript y denotes the phase of carrier gas. More
details about the model closures are referred to Dong et al. (2008a).
If the reaction has little effects on the flow and the convection dom-
inates the mass transfer, then Equations (21)–(24) can be reduced into
algebraic expression of subgrid concentration of transferred compo-
nent, as follows:
eg ðugf ug Þ
YAc ¼ YA ð25Þ
f ec ðugf ugc Þ
eg ðug ugc Þ
YAf ¼ YA ð26Þ
ð1 fÞef ðugf ugc Þ
Such a simplified form is easier than the partial differential equa-

tions to analyze the effect of structures on the mass transfer. More
details about its analysis are referred to Zhang (2010). The EMMS/mass
model has been used to evaluate the overall mass transfer coefficient in
a CFB riser (Dong et al., 2008a). For a one-dimensional, steady state and
fully developed riser, the above equations can be simplified by ignoring
all the terms related with @/@t, @/@r. If the cluster variation along the
riser height is further ignored, then the two-phase structure parameters
(fk, ek, ugk) can be calculated from the original EMMS model instead of
the EMMS/matrix model. The final solution of the steady-state version
of EMMS/mass model was plotted in Figure 15b with magenta circles
for a system with naphthalene and air, having Schmidt number, Sc, at
around 2.5. For comparison, Figure 15a is the experimental data in the
literature drawn in conventional way, where the overall Sherwood
number Shovr relates with the superficial Reynolds number
Re0(=Ugdprg/g). Owing to the lack of description of structure (it should
be noted that, sometimes, wrong interpretation models for processing
the experimental data are also the causes, as discussed in Kunii and
[(Figure_5)TD$IG]
Figure 15 A tentative answer to the disputes on mass transfer (adapted from

Dong et al., 2008a): comparison of overall Sherwood number between this work and
the literature data. (a) Conventional Sh–Re curve of (1) Subbarao and Gambhir
(2002); (2) Kettenring and Manderfield (1950); (3) Resnick and White (1949); (4)
Venderbosch et al. (1999); (5) Gunn (1978); (6) Van der Ham et al. (1991); (7) Dry et al.
(1987); (8) Dry and White (1992). (b) EMMS/mass predictions of Shovr as a function of
Re0 and e and its comparison with experimental data.
38 Wei Wang et al.
Levenspiel, 1991), for one specific Re0, the data are diffused over a wide
range of Shovr. If the data are redrawn with consideration of structure,
as shown in Figure 15b, we can see that the diffuse of data are due to
different voidage, and the EMMS/mass model predictions agree well
with these two sets of data. That can be viewed as our tentative answer
or solution to the disputes over mass transfer drawn in Figure 3.
In all, Reynolds number is insufficient to correlate the overall
Sherwood number in a CFB. This is the reason why the conventional
correlations of mass transfer coefficient diffuse over several orders of
magnitudes (Breault, 2006). Introducing structures, in terms of voidage
for this case, improves very much the prediction. The abrupt change of
Sherwood number at Reynolds number around 50–100 and voidage
around 0.85–1.0 corresponds to the jump change of choking. It is inter-
esting to note that the Sherwood number for classical fluidized beds and
fixed beds also displays abrupt change around this range of Re0 (Kunii
and Levenspiel, 1991). Thus, one may expect that certain common
mechanisms are underlying these phenomena. More efforts are needed
to unify the mass transfer theory with progress in meso-scale modeling.
The full version of the EMMS/mass mass transfer model has also
been coupled with CFD to show its improvement over conventional
approaches for reactive multiphase flow simulations (Dong et al.,
2008b). Figure 16 shows the necessity of this meso-scale modeling
through partial and full replacement of the conventional CFD
approaches when simulating an ozone decomposition experiment
(Ouyang et al., 1995). The conventional CFD without any meso-scale
modeling gives rather poor results in that both the interphase momen-
tum transfer and the mass transfer rate were overpredicted and then
ozone were decomposed too fast (Figure 16c). When the drag coefficient
was corrected with a meso-scale model (here, EMMS/matrix) but the
mass transfer model remains unchanged, the prediction was improved,
more ozone being released from the top outlet instead of being totally
consumed in the riser (Figure 16b). If the meso-scale modeling for mass
transfer was also introduced, the prediction improved further, showing
the best agreement with experimental data in terms of ozone concen-
tration along the radial direction at two elevations (Figure 16a). In all,
accurate prediction of reaction behavior greatly depends on the meso-
scale structure modeling, both on mass transfer and momentum trans-
fer. Accurate prediction can be viewed as our tentative answer to the
question of reactive modeling issued in Figure 3.
With additional assumption of analogy between mass and heat
transfer, which is valid for low mass transfer rate processes, similar
approach has ever been proposed to model the particle–fluid heat
transfer (Hou and Li, 2010). The overall heat transfer, however, may
[(Figure_6)TD$IG]
Figure 16 A tentative answer to the question on reactive modeling: snapshots of

dimensionless ozone concentration at time of No. 30 s and related time-averaged
radial profiles at different heights (experiment: Ouyang et al., 1995; Ug = 3.8 m/s,
Gs = 106 kg/m2 s, kr = 57.21 m3(O3)/m3 (catalyst)s). (a) EMMS/matrix for flow and
EMMS/mass for mass transfer; (b) only flow structure is considered through a EMMS/
matrix drag coefficient, mass transfer model is the conventional; (c) conventional CFD
model for both flow and mass transfer without structural consideration (color
spectrum is in log scale) (Dong et al., 2008b).
be mixed with particle–particle heat conduction in dense fluidized beds

and tends to be very rapid. In contrast, the effects of meso-scale struc-
ture on the bed-to-wall heat transfer are often of engineering interest
and have been investigated extensively. In a CFB combustor, the intense
convection and renewal of particles in the vicinity of the wall as well as
the particle–wall collisions represents the main contribution to the total
bed-to-wall heat transfer. To account for this meso-scale effect, the
cluster renewal model was proposed by Basu and Fraser (1991). It
was assumed that any parts of the wall are in alternate contact with
the cluster and dispersed particles. Assume d is the fraction of the wall
contacted with clusters, then, the overall heat transfer coefficient h can
be written as
h ¼ hcon þ hrad ¼ dðhc;con þ hc;rad Þ þ ð1 dÞðhf;con þ hf;rad Þ; ð27Þ
where subscripts ‘‘con’’ and ‘‘rad’’ denote convection and radiation

contributions, respectively. The critical parameter therein, that is, the
cluster fraction d at the wall, can be calculated with EMMS/matrix
subgrid model. This is what we have adopted in simulating an indus-
trial CFB boiler (Wang and Li, 2010; Zhang, 2010), as will be detailed in
industrial applications in Section 4.
40 Wei Wang et al.
3.4 Extension of EMMS modeling to gas–liquid flow

Albeit originally proposed for gas–solid fluidization, the concepts of
structure resolution and compromise between dominant mechanisms
embodied in the EMMS model can be generalized into the so-called
variational multi-scale methodology (Li and Kwauk, 2003) and
extended to other complex systems (Ge et al., 2007). One typical example
out of these extensions is the Dual-Bubble-Size (DBS) model for gas–
liquid two-phase flow in bubble columns (Yang et al., 2007, 2010).
Analogous to the EMMS model, the gas–liquid systems in bubble
columns can be resolved physically into small bubbles, large bubbles,
and liquid phase. Note that the significance of this ‘‘structure’’ resolu-
tion lies not only in the classification of different phase structure but in
the differentiation of different dominant mechanisms that can be math-
ematically expressed as various extremum tendencies. When larger
bubbles break into smaller daughter bubbles, the energy dissipated
for bubble surface oscillation and deformation (Nsurf) decreases. This
process is governed by Nsurf ! min, and hence the global two-phase
flow field in this case is taken over by the liquid phase. On the other
hand, when smaller bubbles coalesce into larger bubbles, the energy
dissipation through liquid turbulence (Nturb) decreases. The process is
dominated by Nturb ! min to favor the formation of large bubbles and
the global two-phase flow field in this case is therefore dominated by
gas phase. However, bubbles may break up at one location and coalesce
at another, and the dynamic balance between the two dominant
mechanisms is usually well established in practical industrial systems,
which can be expressed with a stability condition Nsurf + Nturb ! min
(Ge et al., 2007; Yang et al., 2010; Zhao, 2006).
From the multi-scale point of view, the total energy dissipation NT
can be grouped into three portions, namely, Nsurf, Nturb, and Nbreak. The
last portion is generated from bubble breakage and finally dissipated in
the process of bubble coalescence. While Nsurf and Nturb are considered
to be directly dissipated on micro-scale, Nbreak is counted as a kind of
meso-scale energy dissipation. Therefore, the stability condition can be
either expressed with the minimization of micro-scale energy dissipa-
tion Nsurf + Nturb ! min or conveyed as the maximization of meso-scale
energy dissipation Nbreak ! max.
With such an understanding on system complexity in mind, the
DBS model is composed of two simple force balance equations,
respectively, for small or large bubble classes, and one mass conserva-
tion equation as well as the stability condition serving as a variational
criterion and a closure for conservative equations. For a given operating
condition of the global system, six structure parameters for small and
large bubble classes (their respective diameters dS, dL, volume fraction
fS, fL, and superficial gas velocities UgS, UgL) can be obtained by solving
the nonlinear optimization problem. This implies that these structure
parameters not only obey the mass and momentum conservative equa-
tions but are governed by the stability condition reflecting the compro-
mise between different dominant mechanisms.
The DBS model calculation on structure parameters and total gas
holdup shows that the system structure evolves gradually with the
increase of global superficial gas velocities, and then a jump change
of gas holdup occurs, which coincidently corresponds to the transition
from homogeneous and transition regimes to fully developed hetero-
geneous regime found in experiments (Camarasa et al., 1999; Ruthiya
et al., 2005; Zahradnik et al., 1997), as shown in Figure 17a. The predic-
tion reflects some intrinsic evolution of the system structure whereas
[(Figure_7)TD$IG]
Figure 17 Physical explanation of regime transition in bubble columns (Yang et al.,

2007, 2010).
42 Wei Wang et al.
experiments in reality have to be influenced by extrinsic factors like

aeration non uniformity generated from different distributors, liquid
properties, and column geometries. A physical understanding on such
regime transition can be obtained by scrutinizing Figure 17b and c. The
global minimum of micro-scale energy dissipation falls into the right
potential trough for 0.128 m/s of superficial gas velocity. However, it
jumps into the left potential trough when Ug increases to 0.129 m/s.
This jump change actually leads to the jump change of the diameter of
small bubble class (see the dash lines. Note that the subscript S and L
here only have symbolic significance and dL stands for the small bubble
class since the solutions are fully symmetrical to the cross-section plane
of the cube) and hence other structure variables, implying that it is the
stability condition which drives the evolution of system structure and
finally leads to the regime transition.
Recognizing the role of stability condition in reflecting the compro-
mise between dominant mechanisms and driving the evolution of
multi-scale structure as an additional constraint besides mass and
momentum conservative equations, direct coupling of the stability con-
dition with CFD simulation should supply an ideal framework to the-
oretically simulate the gas–liquid flow. As a first approximation, we
recently propose a simplified method to realize this coupling, just like
that for modeling gas–solid flow in fluidization. The CFD simulation
indicates that DBS-based model could obtain quite reasonable predic-
tion with experiments for radial gas holdup, total gas holdup, and two-
phase flow field as well as the regime transition in bubble columns
without the need of adjusting any model parameters. Although this
work need further validation and verification, the simulation has
shown the great potential and advantage in modeling the complicated
multi-scale structure and achieving a more intrinsic understanding of
gas–liquid two-phase flow. For details, the interested readers are
referred to our previous and upcoming publications (Chen et al.,
2009a, 2009b; Yang et al., 2007, 2010, 2011).
In fact, extremum tendencies expressing the dominant mechanisms
in systems like turbulent pipe flow (Li et al., 1999), gas–liquid–solid flow
(Liu et al., 2001), granular flow, emulsions, foam drainages, and multi-
phase micro-/nanoflows also follow similar scenarios of compromising
as in gas–solid and gas–liquid systems (Ge et al., 2007), and therefore,
stability conditions established on this basis also lead to reasonable
descriptions of the meso-scale structures in these systems. We believe
that such an EMMS-based methodology accords with the structure of
the problems being solved, and hence realize the similarity of the struc-
tures between the physical model and the problems. That is the funda-
mental reason why the EMMS-based multi-scale CFD improves the
conventional simulation methods for the corresponding systems. The

effectiveness of the EMMS-based multi-scale CFD has been also indi-
cated recently in the literature (Benyahia, 2010; Chalermsinsuwan et al.,
2009; Hartge et al., 2009; Hou and Li, 2010; Jiradilok et al., 2006;
Nikolopoulos et al., 2010a, 2010b; Qi et al., 2007; Wang et al., 2010a;
Yang et al., 2009), and in the mean while, industrial applications of it
help us continuously elaborate the multi-scale CFD approach.
4. INDUSTRIAL APPLICATIONS
In this section we will present some examples of using the EMMS-based

multi-scale CFD to solve industrial problems, including fluid catalytic
cracking (FCC) and CFB combustion.
4.1 Fluid catalytic cracking

Fluid catalytic cracking has been widely applied in petroleum refinery to
convert crude oil into a variety of light products, including about 45% of
gasoline worldwide (Chen, 2006). Direct products from FCC process
normally contain too much olefin that is harmful to environments. To
reduce the olefin content for cleaner fuels, the Research Institute of
Petroleum Processing (RIPP), SINOPEC has developed a novel FCC
process for maximizing iso-paraffins (MIP) (Xu et al., 2001). Besides cat-
alyst, this new FCC process has a different design of reactor. The classical
FCC reactor was altered by inserting an enlarged section in the middle of
the riser, to favor the olefins transformation into iso-paraffins and aro-
matics through alkene isomerization and hydrogen transfer reactions.
Changing the mature FCC to novel design in commercial plants is
challenging. Traditional method from lab test to industry step by step is
rather time-consuming and costly, while numerical simulation could
help boost this procedure provided with reliable model of hydrody-
namics. In view of that, RIPP and Institute of Process Engineering (IPE)
of Chinese Academy of Sciences (CAS) reached an agreement in 2001 to
collaborate on hydrodynamics prediction by using EMMS model. At
the time of that first cooperation, the best computing resource available
to IPE was not powerful enough to allow a transient CFD simulation of
the whole riser. Therefore, a steady-state EMMS model was used with
extension to account for the axial distribution. The simulation was
performed to predict the catalyst particles distribution. The results
submitted to SINOPEC finally have helped the designer figure out
the key parameters for the first set of novel FCC process installed in
Gaoqiao, Shanghai in 2002, with capability of annual process of 1.4
million ton of crude oil.
44 Wei Wang et al.
Along with the successful running of this first demonstration plant,

the MIP processes spread all over China through modifying existent
FCC processes. Unlike building a reactor from scratch, modification is
constrained by the old design. That results in various geometric factors
in different MIP reactors, and hence, more complicated flow behaviors
have been found during operation. To help implement the modifica-
tion, it is necessary to determine the detailed flow behavior inside. As a
result, a second collaboration between IPE and RIPP was started in 2005,
to evaluate the flow behavior under various designs of inlet, outlet, and
distributor as well as solids flux of catalyst particles. To this end, the
simplified version of the EMMS-based multi-scale CFD has been used
to predict the multiphase flow behavior (Lu et al., 2007).
First, for validation, a series of experiments were performed on a
cold-model rig of MIP reactor. Figure 18 shows snapshots of simulated
solids distribution in the riser and the axial profile of cross-sectionally
averaged voidage against experimental data. The clustering phenome-
non that was found in experiments can be captured, and the predicted
profile was in good agreement with the experimental data (Lu et al.,
2007). By comparison, CFD simulation without drag correction pre-
dicted a rather dilute riser without obvious clustering, which is contrary
to experimental findings. On that basis, a series of 2D simulation of an
[(Figure_8)TD$IG]
Figure 18 (a) Instantaneous plot of solids volume fraction in a laboratory-scale cold

model of MIP reactor. (b) Sectioned snapshots of solids volume fraction in the first
and second reaction zones. (c) Time-averaged axial profiles of voidage (—EMMS/CFD:
2D result with EMMS-based drag correlation; –D–CFD: result with hybrid drag
coefficient of Wen and Yu model and Ergun equation) (Lu et al., 2007).
industrial MIP riser was performed. About 18,000 unstructured grids

were generated to model the enlarged section of the riser as well as its
inlet and outlet tube. It normally takes about 1 month for one typical
case on a computer with one AMD246 CPU. A typical snapshot of the
predicted solids distribution in the reactor is shown in Figure 19.
The most important information generated from this second collab-
oration is that the ‘‘choking’’ behavior was found in the MIP riser. As
shown in Figure 20, we found the choking may occur in the second
reaction zone under certain set of operating conditions. The under-
standing of that is critical to the design and operation as the industry
needs that information to avoid the choking instability. However, as
discussed in the above sections, the choking mechanism and even its
phenomenon have caused hot disputes in the past decades. One reason
is that we are still short of physics-based explanation. Another reason
is that the choking was found dependent heavily on the riser height
and geometric limitations, while it is quite difficult, if not impossible,
for experimental researchers to carry out studies covering all those
geometric factors and to unify the findings obtained from different
[(Figure_9)TD$IG]
Figure 19 An industrial MIP reactor and the relevant simulation results of solids
distribution.
46 Wei Wang et al.
[(Figure_0)TD$IG]
Figure 20 Predicted flow regime diagram of the industrial MIP reactor, with solids flux
as a function of the imposed total pressure drop at fixed gas flow rate. The snapshots
of voidage profile refer to the transition, from left to right, the dilute transport,
choking transition in between with different solids inventory, to the dense fluidization
(Lu et al., 2007).
research groups. By comparison, CFD simulations with reliable meso-

scale modeling can be expected to help troubleshoot and tackle such
hard problems readily.
The collaboration is still going on. The full-loop, 3D simulations of
MIP reactors are being performed to help further scale-up. To some
extent, the multi-scale CFD is beginning to take the place of virtual
experiment for solving industrial problems, and it is emerging as a
paradigm beneficial to both industry and academia.
4.2 CFB boiler

CFB with internal combustion can be traced back to the invention of
aluminum hydroxide calcination technology by Lurgi (Reh, 1995). A
typical CFB combustor mainly consists of a highly expanded fluidized
bed furnace (riser), with solids externally circulated, through gas–solid
cyclones or separators, standpipes, aerated siphons/valves, and, in
some cases, external heat exchangers. CFB combustors have been
applied widely together with gradual scale-up, from the first coal com-
bustor of 84 MWth at Vereinigte Aluminium Werke AG (VAW) in
L€unen to atmospheric utility boilers of 300 MWe installed in recent
years and up to 800 MWe ultrasupercritical CFB boilers in project.
Scale-up and optimal design of CFB boilers require comprehensive

knowledge on gas–solids two-phase flow and mixing, interphase heat/
mass transfer, reaction, bed-to-wall heat extraction, and separation effi-
ciency. These issues are strongly related with specific designs of coal
particles and complex geometries under wide range of operating con-
ditions, while measurements under high temperature over large-scale
facilities are very hard to carry out. In this situation, CFD simulation is a
good complement or alternative to facilitate design and operation. In
literature, most of CFD simulations reported are limited to the furnace
chamber and usually with 2D simplifications. However, as we have
shown in simulating the ETH cold model CFB (Zhang et al., 2008), to
better understand the manifold factors affecting CFB boilers, we need
detailed information concerning, for example, the dynamic mixing of
gas and solid fuels in each direction, the effects of various coal-feed
inlets, solid-return valves, air-injection nozzles, and the pressure distri-
bution around the whole loop, and so on. All these entail 3D, full-loop
CFD simulations with reasonable models.
As discussed in the above sections, EMMS-based multi-scale CFD
allows using very coarse-grid that is larger than in conventional CFD
methods by 1–2 orders of magnitude without losing accuracy. That
means it even enables reducing the computing load by 3–6 orders of
magnitude. Such a merit is extremely useful to simulate a large CFB
boiler, or else, a reasonable computation with billions of numerical
grids may last for years and is unaffordable.
In what follows we will give an example of using the EMMS-based
multi-scale CFD to study the complex hydrodynamics and reactions in
an industrial 150 MWe CFB boiler, which was designed by Harbin
Boiler Co. Ltd. and installed in Guangdong, China. The hydrodynamics
part has been reported (Wang and Li, 2010; Zhang et al., 2010) and the
reactive part is referred to more recent work of Zhang (2010).
As shown in Figure 21, the boiler consists of a furnace
(15.32 7.22 m2 in cross section and 36.5 m in height), two adiabatic
cyclones, diplegs, and U-type loop-seal valves. The diameter of the
two cyclones is 8.08 m; each dipleg is connected via two loop seals to
the furnace. The primary air was assumed uniformly blown into the
bottom of the furnace, and 26 secondary air and 2 slag-cooler inlets were
all meshed according to the real design with uniform inflow. The gas
flow rates at different inlets were set according to the designed values.
An atmospheric pressure boundary was prescribed at the cyclone out-
lets. To save the computation cost, monodisperse particle with average
diameter of 0.2 mm and density of 2000 kg/m3 was assumed based on
empirical data. At the walls, the nonslip boundary condition was used
for the gas phase and a partial-slip boundary condition was used for the
48 Wei Wang et al.
[(Figure_1)TD$IG]
solid phase. For hydrodynamic simulations, the flow is assumed unaf-

fected by reactions and the whole loop in Figure 21 was modeled under
the design temperature of 917 C. As the reaction was not explicitly
introduced, to keep a constant solids inventory of the whole boiler,
the solids entrained out of the cyclone outlets were returned to the
furnace via the coal feed inlets. Fluent 6.3.26 was used as the solver
with the drag coefficient modified by EMMS. As to the meshes, the
boiler was divided into several blocks, where the connections between
air inlets and the furnace were meshed with polyhedron, and the others
were meshed with hexahedron, all with size scale of 0.1 m. In such a
way, the total mesh number amounts to about 500,000. The turnaround
time of such a simulation was about several weeks on a cluster of
2.6 GHz computers. More detailed description is referred to
Zhang et al. (2010).
Figure 22 shows a snapshot of the solids distribution at the walls of
the whole boiler. Below the secondary air inlets, clearly a dense bottom
was formed. Above that, the dilute top region was predicted with
various forms of clusters, most of which flow down along the wall as
shown by the vector slice at the side wall. At the loop-seal valves, dense
bottom regions were formed with bubbles. The solids captured by the
cyclone were also in forms of certain kind of dynamic aggregates, falling
down spirally along the wall. Unfortunately there is no data we can use
to verify such complex phenomena. Obviously more efforts are needed
to measure the flow behavior in such a hot facility.
A qualitative verification of this hot-model hydrodynamics relates
with a seesaw phenomenon. For this phenomenon, Grace et al. have
[(Figure_2)TD$IG]
Figure 22 Snapshot of the solids distribution at all the walls together with a slice of
the solids velocity vectors at the side wall.
addressed that, ‘‘when two-phase suspensions are conveyed through

identical parallel flow paths, the flow distribution can be significantly
nonuniform in practice,’’ and there is also maldistribution of gas–solids
flow through identical parallel cyclones in their experiments (Grace
et al., 2007; Masnadi et al., 2010). Figure 23 shows the simulated solid
fluxes at the inlets of the two cyclones. During the monitored period of
time, the averaged solid fluxes manifest no obvious difference as these
two cyclones were designed and operated identically. However, their
instant fluxes show the seesaw phenomenon—the maximum and min-
imum values of the fluxes alternates as to time, one minimum corre-
sponding to another maximum. This can also be validated by the alter-
nate appearance of dense aggregates near the cyclone inlets, as
indicated by the circles in Figure 24. Such predicted phenomenon
agrees well with the experimental results of Grace’s group (Grace
et al., 2007; Masnadi et al., 2010).
The reactive modeling was only performed for the furnace chamber
limited to computing cost. Still a monodisperse solid phase was
50 Wei Wang et al.
[(Figure_3)TD$IG]
Figure 23 The seesaw phenomenon of area-averaged solid fluxes at the cyclone

inlets (Zhang et al., 2010), in accordance with the experimental findings in identical
parallel cyclones (Masnadi et al., 2010).
[(Figure_4)TD$IG]
Figure 24 The seesaw phenomenon by instantaneous solids distribution in the boiler

at simulation time of (a) No. 28.9 s and (b): No. 38.7 s. The red circles indicate high
solids volume fraction on the top wall of the furnace (Zhang et al., 2010).
assumed, consisting of ash and carbon. The gas mixture was assumed to
contain only N2, O2, and CO2 to reduce the computation, while the heat
effect of the other components such as the volatile and the moisture
were considered only in the energy conservation equations. The
EMMS/mass model was used to reframe the scalar conservation equa-
tions for mass fractions of carbon, O2, and CO2. The EMMS/matrix
model was used to modify the drag coefficient. The bed-to-wall heat
transfer has great effects on the convergence of code, for simplicity,
constant temperatures at the walls were assumed. More detailed
description should be referred to Zhang (2010) and Wang and Li (2010).
In all, the coal combustion is a very complex process and is far from
being understood comprehensively. As a tentative solution to it, the
EMMS-based multi-scale CFD allows using coarse-grid to predict its
gas–solid flows without losing accuracy, and the multi-scale behavior
of mass/heat transfer can be also integrated into this framework natu-
rally. The current example has shown us the possibility and advantages
of using this multi-scale CFD in dealing with large-scale CFB reactors
with complex coupling of flow, transfers and reactions, though still
preliminarily. More applications of such an approach can be expected
to bring us a new paradigm for reactive multiphase flow simulations.
5. SUMMARY
Meso-scale structure is the bridge between micro-scale nature and
macro-scale performance, and hence it is critical to characterize com-
plex systems in chemical engineering. Without meso-scale modeling,
the conventional TFM fails to describe the intrinsic structural effects
and characteristic behavior of fluidized beds. In contrast, based on
EMMS modeling, the multi-scale CFD approach features intrinsic res-
olution of meso-scale structures and enables almost grid-independent
solution of the gas–solid two-phase flows. Such a variational type of
multi-scale CFD has been used to simulate various CFB reactors, includ-
ing industrial applications in FCC, CFB boiler, and so on. It has proved
to improve both computational efficiency and accuracy significantly, in
the sense that it allows using much coarser grid without losing accu-
racy, succeeds in predicting the circulating solids flux, revealing the
mechanisms of the choking phenomena, resolving the disputes in trans-
port phenomena of gas-fluidized beds, and so on. All these suggest a
breakthrough in CFD simulation can be achieved by resolving the
structure of the physical model in accordance with that of the problems
being solved. With that methodology in mind, we are expected to
52 Wei Wang et al.
unfold a new paradigm for the simulation of multiphase flows and

reactors.
NOMENCLATURE
a the inert term or acceleration of particles, m/s2

Ar Archimedes number
CD effective drag coefficient for a particle
dc cluster diameter, m
dL bubble diameter of large bubbles, m
dp particle diameter, m
dS bubble diameter of small bubbles, m
Dm molecular diffusion, m2/s
f volume fraction of clusters
fL volume fraction of large bubbles
fS volume fraction of small bubbles
g vector of gravity acceleration, m/s2
Gs solids flux, kg/(m2 s)
h heat transfer coefficient, W/(m2 K)
H riser height, m
H0 initial bed height, m
HD heterogeneity index
k mass transfer coefficient between gas and particle, m/s
kp mass transfer coefficient between gas and particle, m/s
kr ozone decomposition rate, 1/s
Nbreak rate of energy consumption due to bubble breakage and
coalescence per unit mass, m2/s3
Nst mass-specific energy consumption for suspending and
transporting particles, W/kg
Nsurf rate of energy dissipation due to bubble oscillation per unit
mass, m2/s3
Nturb rate of energy dissipation in turbulent liquid phase per unit
mass, m2/s3
Re Reynolds number
Rep local superficial Reynolds number (rgdpUs/g)
Re0 global superficial Reynolds number (rgdpUg/g)
Sc Schmidt number (g/rgDm)
Sh Sherwood number (kdp/Dm)
u real velocity, m/s
U superficial velocity, m/s
Wst volume-specific energy consumption for suspending and

transporting particles, W/m3
Y mass fraction of gas species
Greek letters
ap outer surface area per volume of particles, m2/m3
b drag coefficient with structure in a control volume, kg/(m3 s)
b0 drag coefficient without structure in a control volume, kg/(m3 s)
e time-averaged voidage
eg voidage
emax maximum voidage for particle aggregation
emf minimum fluidization voidage
es solids concentration
es0 averaged solids concentration
viscosity, Pa s
r density, kg/m3
G interphase mass exchange rate, kg/(m3 s)
Subscripts
A component A
c dense phase
f dilute phase
g gas phase
gc gas phase in the dense phase
gf gas phase in the dilute phase
i meso-scale interphase
imp imposed pressure across the riser
L large bubbles
mf minimum fluidization
ovr averaging over bed height
p particle
pc solid phase in the dense phase
pf solid phase in the dilute phase
s slip velocity
S small bubbles
sat saturation
sc slip in the dense phase
sf slip in the dilute phase
si slip at the meso-scale interphase
54 Wei Wang et al.
T terminal velocity
y carrier gas phase
(Bold characters are for vectors or tensors.)
ACKNOWLEDGMENTS
The authors acknowledge the financial supports provided by the
National Natural Sciences Foundation of China (NSFC) under the
Grant No. 20821092, Ministry of Science and Technology (MOST) of
China under the Grant No. 2008BAF33B01, and Chinese Academy of
Sciences (CAS) under the Grant No. KGCX2-YW-222.
REFERENCES
Agrawal, K., Loezos, P. N., Syamlal, M. and Sundaresan, S., J. Fluid Mech. 445, 151–185
(2001).
Andrews, A. T., Loezos, P. N. and Sundaresan, S., Ind. Eng. Chem. Res. 44(16), 6022–6037
(2005).
Andrews, M. J. and O’Rourke, P. J., Int. J. Multiphase Flow 22(2), 379–402 (1996).
Basu, P. and Fraser, S. A., ‘‘Circulating Fluidized Bed Boilers: Design and Operations’’.
Butterworth-Heinemann, Boston (1991).
Benyahia, S., Ind. Eng. Chem. Res. 49(11), 5122–5131 (2010).
Bi, H. T., Grace, J. R. and Zhu, J., Int. J. Multiphase Flow 19(6), 1077–1092 (1993).
Bird, R. B., Stewart, E. and Lightfoot, E. N., ‘‘Transport Phenomena’’. Wiley, New York
(1960).
Breault, R. W., Powder Technol. 163(1–2), 9–17 (2006).
Camarasa, E., Vial, C., Poncin, S., Wild, G., Midoux, N. and Bouillard, J., Chem. Eng. Proc.
38, 329–344 (1999).
Chalermsinsuwan, B., Piumsomboon, P. and Gidaspow, D., Chem. Eng. Sci. 64, 1212–1222
(2009).
Chen, Y., Powder Technol. 163, 2–8 (2006).
Chen, F. and Ge, W., Particuology 8(4), 332–342 (2010).
Chen, F., Ge, W., Guo, L., He, X., Li, B., Li, J., Particuology 7, 332–335 (2009).
Chen, F., Ge, W., Wang, L. and Li, J., Microfluidics Nanofludics 5(5), 639–653 (2008).
Chen, J., Yang, N., Ge, W. and Li, J., Ind. Eng. Chem. Res. 48, 290–301 (2009).
Chen, J., Yang, N., Ge, W. and Li, J., Ind. Eng. Chem. Res. 48, 8172–8179 (2009).
Crowe, C. T., ‘‘Multiphase Flow Handbook’’. CRC Press, Boca Raton (2006).
De Wilde, J., Phys. Fluids 17, 113304 (2005).
Dixon, A. G., Nijemeisland, M. and Hugh Stitt, E., Packed tubular reactor modeling and
catalyst design using computational fluid dynamics, in ‘‘Advances in Chemical
Engineering’’, Vol. 31, pp. 307–389, Elsevier, Amsterdam (2006).
Dong, W., Wang, W. and Li, J., Chem. Eng. Sci. 63, 2798–2810 (2008).
Dong, W., Wang, W. and Li, J., Chem. Eng. Sci. 63, 2811–2823 (2008).
Dry, R. J., Christensen, I. N. and White, C. C., Powder Technol. 52, 243 (1987).
Dry, R. J. and White, C. C., Powder Technol. 70, 277 (1992).
Dudukovic, M. P., Science 325, 698–701 (2009).

Ergun, S., Chem. Eng. Prog. 48, 89–94 (1952).
Ferziger, J. H., Subgrid-scale modeling, in ‘‘Large Eddy Simulation of Complex
Engineering and Geophysical Flows’’ (B. Galperin, S. A. Orszag, Eds.), pp. 37–54.
Cambridge University Press, Cambridge (1993).
Gage, D. H., Schiffer, M., Kline, S. J. and Reynolds, W. C., The non-existence of a general
thermodynamic variational principle, in ‘‘Non-Equilibrium Thermodynamics,
Variational Techniques and Stability’’ (R. J. Donnelly, R. Herman, I. Prigogine,
Eds.), p. 286. University of Chicago Press, Chicago (1966).
Ge, W. and Chen, F., et al., ‘‘GPU-Based Multi-Scale Discrete Simulations and Parallel
Computation’’. Science Press, Beijing (2009) (in Chinese).
Ge, W., Chen, F. and Gao, J., et al., Chem. Eng. Sci. 62(13), 3346–3377 (2007).
Ge, W. and Li, J., Pseudo-particle approach to hydrodynamics of gas/solid two-phase
flow, in ‘‘Circulating Fluidized Bed Technology V’’ (J. Li, M. Kwauk, Eds.), pp. 260–
265. Science Press, Beijing (1996).
Ge, W. and Li, J., Chem. Eng. Sci. 57, 3993–4004 (2002).
Ge, W. and Li, J., Chem. Eng. Sci 58(8), 1565–1585 (2003).
Ge, W., Wang, W., Dong, W., Wang, J., Lu, B., Xiong, Q. and Li, J., Multi-scale structure—a
challenge of CFD for CFBs, in ‘‘Proceedings of the 9th International Conference of
Circulating Fluidized Bed’’ (J. Werther, E. U. Hartge, Eds.), May 13–16, Hamburg.
Geldart, D., Powder Technol. 7, 185–195 (1973).
Gidaspow, D., ‘‘Multiphase Flow and Fluidization: Continuum and Kinetic Theory
Descriptions’’. Academic Press, Boston (1994).
Goldhirsch, I., Ann. Rev. Fluid Mech. 35, 267–293 (2003).
Grace, J. R., Influence of riser geometry on particle and fluid dynamics in circulating
fluidized beds risers, in ‘‘Proceedings of the 5th International Conference of
Circulating Fluidized Bed’’ (M. Kwauk, J. Li, Eds.), May 28–31, Beijing.
Grace, J. R., Cui, H. and Elnashaie, S. S., Can. J. Chem. Eng. 85(5), 662–668 (2007).
Gunn, D. J., Int. J. Heat Mass Transfer 21, 467–476 (1978).
Harris, A. T., Davidson, J. F. and Thorpe, R. B., Powder Technol. 127, 128–143 (2002).
Hartge, E. -U., Ratschow, L., Wischnewski, R. and Werther, J., Particuology 7(4), 283–296
(2009).
Herbert, P., Reh, L. and Nicolai, R., AIChE Symp. Ser. 321, 95 (1999) 61–66.
Hou, C. and Ge, W., Chem. Eng. Sci. 62, 6794–6805 (2007).
Hou, B. and Li, H., Chem. Eng. J. 157(2–3), 509–519 (2010).
Hu, N., Zhang, H., Yang, H., Yang, S., Yue, G., Lu, J. and Liu, Q., Powder Technol. 196(1),
8–13 (2009).
Igci, Y., Andrews, A. T., Sundaresan, S., Pannala, S. and O’Brien, T., AIChE J. 54, 1431–1448
(2008).
Jin, G., He, G. and Wang, L., Phys. Fluids 22(5) 055106.
Jiradilok, V., Gidaspow, D., Damronglerd, S., Koves, W. J. and Mostofi, R., Chem. Eng. Sci.
61(17), 5544–5559 (2006).
Kettenring, K. N. and Manderfield, E. L., Chem. Eng. Prog. 46(3), 139–145 (1950).
Kunii, D. and Levenspiel, O., ‘‘Fluidization Engineering, 2nd ed. Butterworth-
Heinemann, Boston (1991).
Levenspiel, O., ‘‘Chemical Reaction Engineering’’, 3rd ed. John Wiley & Sons, New York
(1999).
Li, J., Multi-Scale Modeling and Method of Energy Minimization for Particle–Fluid Two
Phase Flow, Ph.D. thesis (in Chinese), Institute of Chemical Metallurgy, Chinese
Academy of Sciences, Beijing (1987).
56 Wei Wang et al.
Li, J., Chen, A., Yan, Z., Xu, G. and Zhang, X., Particle-fluid contacting in circulating
fluidized beds, in ‘‘Preprint Volume for Circulating Fluidized Beds IV’’ (A. A.
Avidan, Ed.), pp. 49–54. AIChE, Somerset (1993).
Li, J. and Ge, W., Chem. Eng. Sci. 62, 3285–3286 (2007).
Li, J., Ge, W. and Kwauk, M., arXiv:0912.5407v3 (2009).
Li, J., Ge, W., Zhang, J., Gao, S., Wang, W., Yang, N., Sun, Q., Gao, J. (2007). Analytical
multi-scale methodology for fluidization systems—retrospect and prospect. In:
Berruti, F., Bi, X., Pugsley, T. (eds.), The 12th International Conference on
Fluidization, Vancouver, Canada.
Li, J., Ouyang, J., Gao, S., Ge, W., Yang, N. and Song, W., ‘‘Multiscale Simulation of
Particle-Fluid Complex Systems (in Chinese)’’. Science Press, Beijing (2005).
Li, J. and Kwauk, M., ‘‘Particle-Fluid Two-Phase Flow—the Energy-Minimization Multi-
Scale Method’’. Metallurgical Industry Press, Beijing (1994).
Li, J. and Kwauk, M., Ind. Eng. Chem. Res. 40, 4227–4237 (2001).
Li, J. and Kwauk, M., Chem. Eng. Sci. 58, 521–535 (2003).
Li, J., Zhang, J., Ge, W. and Liu, X., Chem. Eng. Sci. 59, 1687–1700 (2004).
Li, J., Zhang, Z., Ge, W., Sun, Q. and Yuan, J., Chem. Eng. Sci. 54(8), 1151–1154 (1999).
Lin, Q., Wei, F. and Jin, Y., Chem. Eng. Sci. 56(6), 2179–2189 (2001).
Liu, X., Gao, S. and Li, J., Chem. Eng. J. 108(3), 193–202 (2005).
Liu, Y., Ge, W., Wang, J., Chen, J., Yang, N., Wang, W. and Li, J. Powder Technol. (2010)
(accepted).
Liu, M., Li, J. and Kwauk, M., Chem. Eng. Sci. 56, 6805–6812 (2001).
Lu, B., EMMS-based Meso-Scale Model and Its Application in Simulating Gas-Solid Two-
Phase Flows, Ph.D. thesis (in Chinese), Institute of Process Engineering, Chinese
Academy of Sciences, Beijing (2009).
Lu, B., Wang, W. and Li, J., et al., Chem. Eng. Sci. 62, 5487–5494 (2007).
Lu, B., Wang, W. and Li, J., Chem. Eng. Sci. 64, 3437–3447 (2009).
Lu, B., Wang, W. and Li, J. Chem. Eng. Sci. (2011) (in revision).
Ma, J., Ge, W., Wang, X., Wang, J. and Li, J., Chem. Eng. Sci. 61, 7096–7106 (2006).
Ma, J., Ge, W., Xiong, Q., Wang, J. and Li, J., Chem. Eng. Sci. 64, 43–51 (2009).
Manyele, S. V., Parssinen, J. H. and Zhu, J. -X., Chem. Eng. J. 88, 151–161 (2002).
Masnadi, M. S., Grace, J. R., Elyasi, S. and Bi, X., Sep. Purif. Technol. 72(1), 48–55 (2010).
Matsen, J. M., Powder Technol. 32, 21–33 (1982).
Meng, F., Wang, W. and Li, J., A method of measuring velocity and acceleration of particle
clusters in two-phase flows. Chinese Patent No. 200910082721.0 (2009).
Monaghan, J. J., Rep. Prog. Phys. 68, 1703–1759 (2005).
Monaghan, J. J. and Kocharyan, A., Comp. Phys. Commun. 87(1–2), 225–235 (1995).
Nikolopoulos, A., Atsonios, K., Nikolopoulos, N., Grammelis, P. and Kakaras, E., Chem.
Eng. Sci. 65(13), 4089–4099 (2010).
Nikolopoulos, A., Papafotiou, D., Nikolopoulos, N., Grammelis, P. and Kakaras, E., Chem.
Eng. Sci. 65(13), 4080–4088 (2010).
Niksa, S., Liu, G. and Hurt, R. H., Prog. Energy Combust. Sci. 29, 425–477 (2003).
Ouyang, S., Li, X. -G. and Potter, O. E., AIChE J. 41(6), 1534–1542 (1995).
Pope, S. B., ‘‘Turbulent Flows’’. Cambridge University Press, Cambridge (2000).
Prigogine, I., ‘‘Introduction to Thermodynamics of Irreversible Processes’’. Wiley
Interscience, New York (1967).
Qi, H., Li, F., Xi, B. and You, C., Chem. Eng. Sci. 62, 1670–1681 (2007).
Ranz, W. E., Chem. Eng. Prog. 48(5), 247–253 (1952).
Reh, L., Fluid dynamics of CFB combustors, in ‘‘Proceedings of the 5th International
Conference on Circulating Fluidized Beds’’ (M. Kwauk, J. Li, Eds.), pp. 1–15.
Beijing (1996).
Reh, L. and Li, J., Measurement of voidage in fluidized beds by optical probe, in
‘‘Circulating Fluidized Beds Technology III’’ (P. Basu, M. Horio, M. Hasatani,
Eds.), p. 105. Pergamon Press, Oxford, UK (1991).
Resnick, W. and White, R. R., AIChE J. 45(6), 377–390 (1949).
Ruthiya, K. C., Chilekar, V. P., Warnier, M. J. F., van der Schaaf, J., Kuster, B. F. M.,
Schouten, J. C. and van Ommen, J. R., AIChE J. 51, 1951–1965 (2005).
Sharma, A. K., Tuzla, K., Matsen, J. and Chen, J. C., Powder Technol. 111, 114–122 (2000).
Soong, C. H., Tuzla, K. and Chen, J. C., Identification of particle clusters in circulating
fluidized bed, A. A. Avidan, Ed.), pp. 615–620. Vol. IV, AIChE, New York (1994).
Subbarao, D. and Gambhir, S., Gas to particle mass transfer in risers, in ‘‘Proceedings of
7th International Circulating Fluidized Beds Conference’’, pp. 97–104, Canadian
Society for Chemical Engineering, Niagara Falls.
Turns, S. R., ‘‘An Introduction to Combustion: Concepts and Applications, 2nd ed.
McGraw-Hill, Boston (2000).
Venderbosch, R. H., Prins, W. and van Swaaij, W. P. M., Can. J. Chem. Eng. 77(2), 262–274
(1999).
Van der Ham, A. G. J., Prins, W. and Van Swaaij, W. P. M., Hydrodynamics and mass
transfer in a regularly packed circulating fluidized bed, in ‘‘Circulating Fluidized
Bed Technology III’’ (P. Basu, M. Horio, M. Hasatani, Eds.), p. 605. Pergamon Press,
Oxford, UK (1991).
Van der Hoef, M. A., Ye, M., van sint Annaland, M., Andrews IV, A. T., Sundaresan, S. and
Kuipers, J. A. M., Multiscale modeling of gas-fluidized beds, in ‘‘Advances in
Chemical Engineering’’, Vol. 31, pp. 65–149. Academic Press, New York (2006).
Wang, L., Multi-Scale Mass Transfer Model and Experimental Validation for
Heterogeneous Gas-Solid Two-Phase Flow, Ph.D. thesis (in Chinese), Chinese
Academy of Science, Beijing (2002).
Wang, J., Chem. Eng. Sci. 63(8), 2294–2298 (2008).
Wang, W. and Li, J., Chem. Eng. Sci. 62, 208–231 (2007).
Wang, W. and Li, J., Modeling of circulating fluidized bed combustion, Chapter 12 (M.
Lackner, F. Winter, A. K. Agarwal, Eds.), pp. 437–472. Vol. 4, Wiley-VCH, Berlin
(2010).
Wang, X., Liao, L., Fan, B., Jiang, F., Xu, X., Wang, S. and Xiao, Y., Fuel Process. Technol. 91
(8), 927–933 (2010a).
Wang, W., Lu, B., Dong, W. and Li, J., Can. J. Chem. Eng. 86, 448–457 (2008).
Wang, W., Lu, B. and Li, J., Chem. Eng. Sci. 62, 814–819 (2007).
Wang, W., Lu, B., Zhang, N., Shi, Z. and Li, J., Int. J. Multiphase Flow 36, 109–118 (2010b).
Wang, J., Van der Hoef, M. A. and Kuipers, J. A. M. Chem. Eng. Sci.64(3), 622–625 (2009).
Wang, L., Yang, N. and Li, J., Chem. Eng. Commun. 192, 1636–1654 (2005).
Wen, C. Y. and Yu, Y. H., Chem. Eng. Symp. Ser. 62(62), 100–111 (1966).
Wolf-Gladrow, D. A., ‘‘Lattice-Gas Cellular Automata and Lattice Boltzmann Models: An
Introduction’’. Springer-Verlag, Berlin (2000).
Xiong, Q., Li, B., Chen, F., Ma, J., Ge, W. and Li, J., Chem. Eng. Sci. 65, 5356–5365 (2010).
Xu, J., Qi, H., Fang, X., Ge., W., Wang, X., Xu, M., Chen, F., He, X. and Li, J. Particuology
(2010b) (accepted).
Xu, Y., Zhang, J. and Rong, J., Petrol. Process. Petrochem. 32(8), 1–5 (2001) (in Chinese).
Yang, W., Ind. Eng. Chem. Res. 43(18), 5496–5506 (2004).
Yang, N., Chen, J., Ge, W. and Li, J., Chem. Eng. Sci. 65, 517–526 (2010).
Yang, N., Chen, J., Zhao, H., Ge, W. and Li, J., Chem. Eng. Sci. 62, 6978–6991 (2007).
Yang, N., Wang, W., Ge, W. and Li, J., Chem. Eng. J. 96, 71–80 (2003).
Yang, N., Wang, W., Ge, W. and Li, J., Ind. Eng. Chem. Res. 43, 5548–5561 (2004).
Yang, N., Wu, Z., Chen, J., Wang, Y. and Li, J. Chem. Eng. Sci. (2011) (in press).
Yang, B., Zhou, X., Yang, X., Chen, C. and Wang, L., AIChE J. 55(8), 2138–2149 (2009).
58 Wei Wang et al.
Yu, Z. Q. and Jin, Y., Heat and mass transfer, Academic Press, New York (1994).
Zahradnik, J., Fialova, M., Ruzicka, M., Drahos, J., Kastanek, F. and Thomas, N. H., Chem.
Eng. Sci. 52, 3811–3826 (1997).
Zhang, N., EMMS-based Meso-Scale Mass Transfer Model and Its Application to
Circulating Fluidized Bed Combustion Simulation, Ph.D. thesis (in Chinese),
Institute of Process Engineering, Chinese Academy of Sciences, Beijing (2010).
Zhang, J., Ge, W. and Li, J., Chem. Eng. Sci. 60(11), 3091–3099 (2005).
Zhang, N., Lu, B., Wang, W. and Li, J., Particuology 6, 529–539 (2008).
Zhang, N., Lu, B., Wang, W. and Li, J., Chem. Eng. J. 162, 821–828 (2010).
Zhang, D. Z. and VanderHeyden, W. B., Int. J. Multiphase Flow 28, 805–822 (2002).
Zhao, H., Multi-scale Modeling of Gas-Liquid (Slurry) Reactors, Ph.D. thesis (in Chinese),
Institute of Process Engineering, Chinese Academy of Sciences, Beijing (2006).
CHAP TER 2
The Holistic Strategy in
Multi-Scale Modeling
Pil Seung Chung1, Myung S. Jhon1,2 and
Lorenz T. Biegler1
Contents 1. Introduction 60
2. Illustration via Benchmark Examples 63
2.1 Renewable energy (PEFC) 63
2.2 Nanotribology (HDDs) 66
3. Structure of Hierarchical Equations 74
3.1 Quantum level 75
3.2 Atomistic/molecular level 76
3.3 Meso-scale/continuum level 81
3.4 Process-scale level 83
4. System Integration 84
4.1 Bridging methodology between different scale levels 85
4.2 Coarse-graining methods 87
5. Technological Applications: PEFC 92
5.1 PEM 92
5.2 Multiphenomena in gas diffusion layer 97
5.3 Device-scale/process-scale level 102
6. Technological Applications: Hard Disk Drive 103
6.1 The coarse-grained, bead-spring model 104
6.2 Simple reactive sphere model 108
6.3 Meso-scale/continuum level 109
1
Department of Chemical Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA
2
School of Advanced Materials Science and Engineering, Sungkyunkwan University,
Suwon 440-746, Korea
59
60 Pil Seung Chung et al.
7. Summary and Conclusions 112

Acknowledgment 114
References 114
Abstract Theoretical tools for nanoscale systems have become highly

advanced over the past decades. Consequently, the demand for
novel methodologies, which enable scale-correlated analysis,
increases in order to describe a hierarchy of multi-scale systems
in various length and time scales. Recent development in the
multi-scale approach is becoming popular as a new multidisci-
plinary analysis paradigm in science and engineering fields. Via
hierarchical integration methodologies of various numerical
techniques, we can expect a mechanism of communication
between different scale phenomena, and predict macroscopic
level behavior as functions of other lower level scale parameters.
In addition, multi-scale approach can provide integrated system
models for decision-making with high accuracy based on first-
principle phenomena. In this paper, we present an overview of
theoretical tools for specific scales as well as holistic strategies
for multi-scale integration by using two benchmark examples:
special topics dealing with renewable energy and nanotribology,
that is, polymer electrolyte fuel cells and hard disk drive systems.
1. INTRODUCTION
As the demand for nanoscale investigation of systematically specified

functional materials increases, novel analytical techniques have been
broadly studied to satisfy this demand by tuning physical structures
and functionalities of these materials. This progress enlightens the
understanding of nanoscale dynamics to control the functionalities of
materials. However, advances in nanoanalysis direct us to the critical
issue of the interrelation to macroscopic scales since most systems
comprise phenomena at different time and length scales, and often
can be described through a hierarchy of scale-specific models.
Therefore, the multi-scale approach becomes popular as a new multi-
disciplinary analysis paradigm in science and engineering fields. Via
multi-scale approach, we expect that the mechanism of information
communication obtained from one scale and passed to another scale,
allows one to predict macroscopic behavior from first-principles infor-
mation in a bottom-up approach. Alternately, device-scale properties
can be obtained with nanoscale resolution using the top-down
approach. The ultimate goal is to integrate models seamlessly at the
The Holistic Strategy in Multi-Scale Modeling 61
different scales. As system size approaches nanoscale, the observation

of phenomena at atomic/molecular scales must be emphasized. In
addition, many of the novel approaches to analyze these phenomena
have adopted an understanding of the relationship between different
scale phenomena, including nanoscopic details where the phenomena
cannot be explained by the experiments or macroscopic models (Bent
et al., 2003; Brown et al., 2009; Gerde and Marder, 2001; Juan et al.,
2009; Moseler et al., 2005; Rubla et al., 2000).
The vision of the theoretical engineering and science field is cur-
rently in multi-scale modeling by coupling computational tools from
ab initio atomistic/electronic scales to continuum scale including pro-
cess and device-level simulation and decision making, and optimizing
across different scales. This leads to reversal of the normal flow of
knowledge across these scales, and to allow design and synthesis of
new forms of matter with tailored properties, driven by large-scale
process requirements (Figure 1). Concepts of multi-scale simulation
enable us to control the macroscopic phenomena by characterizing
nanoscale properties of matter emerging from the complex correlations
of atomic constituents. Models based on physics suitable for particular
time and length scales use specific numerical techniques to solve the
governing equations and answer questions relevant to those scales,
which cover the atomistic, molecular, device, and plant levels. To more
fully couple advances in knowledge across the enormous range of
scales, and enable better understanding and control of emergent prop-
erties of matter that emerge from complex interactions, it is necessary to
develop holistic multi-scale models, which ensure that knowledge gen-
erated at one scale is transmitted to the other. Figure 2 (Grossmann and
Westerberg, 2000; Marquardt et al, 2000) illustrates multi-scale models
and tasks that arise in all applications of process systems engineering.
[(Figure_1)TD$IG]
Figure 1 Schematic description of multi-scale approach and modeling methods.

[(Figure_2)TD$IG]
Figure 2 Time and length scales in multi-scale systems engineering.
Both time and length scales cover over 15 orders of magnitude and
reflect the behavior of electronic/atomistic/molecular/supramolecular
levels all the way to products, process units, and their integration into
the enterprise. Moreover, these tasks are strongly intertwined; basic
components at the lower scales serve as the building blocks for the
scales above them, while the demands at higher scales impose specifi-
cations at smaller levels, particularly with respect to material proper-
ties. Linkage of submodels ranging from atomistic to the enterprise
requires systematic multi-scale, multiphenomena integration, which
is the core driver for the current multi-scale modeling paradigm.
During the past 5 years, over 1000 publications have dealt with
multi-scale subject and resulting in a broad knowledge of integration
methodologies. Some publications provided perspectives on the multi-
scale approaches by presenting decision-making methodologies on the
applications (Baeurle, 2009; Dudukovic, 2009; Engler et al., 2009;
Vlachos, 2005). Among the various multi-scale approaches, which have
been developed, we motivate the specific need for multi-scale modeling
by focusing on methodologies for the systematic approach to these
techniques. In this review paper, we present the holistic integration
strategy in multi-scale modeling with two different benchmark appli-
cations: special topics dealing with renewable energy and nanotribol-
ogy, that is, polymer electrolyte fuel cells (PEFC) and hard disk drive
(HDD) systems.
2. ILLUSTRATION VIA BENCHMARK EXAMPLES
Owing to the complexity involved in multiphenomena at different

time and length scales, the current trends in renewable energy and
nanotribology can be excellent benchmark candidates for evaluation
of multi-scale modeling approaches. To help understanding of hier-
archical systems (Figure 2) of benchmark systems, we will examine
PEFC (renewable energy) and HDD system (nanotribology).
2.1 Renewable energy (PEFC)

As the need for renewable energy systems increase, fuel cells have
attracted considerable recent attention as a possible replacement for
the massive power generation and portable systems. PEFC is one of
the promising alternative energy sources for portable or small unit
devices (i.e., small scale power generators, and transportations). Since
the PEFC consists of nanoscale subsystems including multiphenomena,
an accurate system model is needed to design PEFC with optimal
performance. Therefore, a fundamental understanding in electrochem-
istry, materials, and heat and mass transport phenomena are critical for
developing accurate models, which can satisfy the high performance
and reliability for the fuel cells. The hydrogen PEFC, which directly
convert chemical energy into electric energy, is promising candidate for
future green car technology in parallel with plug-in and hybrid electric
vehicles.
The hydrogen PEFC device comprises a hydrogen source and a gas
compressor supplying air in a stack, which consists of multiple PEFCs
connected in series (Figure 3) (Xu et al., 2006a). An individual PEFC
(Figure 4) comprises two gas channels (GCs), two gas diffusion layers
(GDLs), and two catalyst layers (CLs) each on the anode and cathode
sides, as well as a central polymer electrolyte membrane (PEM). The
GCs are bipolar plates that are hollow chambers for fluid inlet and
outlet. They also serve as a connection between adjacent cells. The
GDLs are porous materials that support the CL, provide uniform dis-
tribution of gases, and act as a medium for electron transport from the
CL (where the electrochemical reactions typically occur with a platinum
(Pt) catalyst) to the external current collectors. The PEM acts as a proton
transport passage from anode to the cathode, and this conductivity of
the PEM is a strong function of its water uptake. Humidified hydrogen
and air are fed to the anode and cathode inlets, respectively. The species
undergo transport processes in both the GCs and the porous GDLs, after
which they reach the anode/cathode CLs. Within the CLs, hydrogen
and oxygen undergo electrochemical reactions, and the water mole-
cules are transported across the PEM from anode to the cathode. The
[(Figure_3)TD$IG]
Figure 3 Process-level model of a PEFC-based power plant.
[(Figure_4)TD$IG]
Figure 4 Components of a single stack hydrogen PEFC.
protons released in the oxidation reaction at the anode are transported

across the PEM, and electrons released reach the cathode via an external
circuit. On reaching the cathode the protons and electrons combine with
the oxygen in the reduction reaction to generate water as the product.
The direct methanol fuel cell (DMFC) is another type of PEFC, which
[(Figure_5)TD$IG]
Figure 5 Chemical structure of Nafion1.
uses a solution of methanol and water as a fuel instead of humidified

hydrogen gas. Compared to hydrogen fuel cells, DMFCs are advanta-
geous for their ease of fuel delivery and storage, lack of humidification
requirement, and reduced design complexity. Due to the absence of
ancillary equipments (i.e., fuel reformer), the DMFC is ideally suited
for portable electronic devices applicable to laptops or mobile phones
(Mench et al., 2004).
A PEM is a semipermeable membrane generally made from iono-
mers and designed to conduct protons while being impermeable to
gases such as oxygen or hydrogen. This is their essential function when
incorporated into a PEM fuel cell: separation of reactants and transport
of protons. One of the most commonly and commercially available PEM
materials is Nafion1 (Figure 5), which is produced by DuPont. While
Nafion1 is an ionomer with a perfluorinated backbone like Teflon,
there are many other structural motifs used to make ionomers for
PEMs. Many use polyaromatic polymers while others use partially
fluorinated polymers. Nafion1 has received a considerable amount of
attention as a proton conductor for PEM fuel cells because of its excel-
lent thermal and mechanical stability. The chemical basis of Nafion1’s
superior conductive properties remains a focus of research. Recent
reports suggest protons on the SO3H+ (sulfonic acid) groups hop from
one acid site to another. The membrane structure allows movement of
cations but the membranes do not conduct anions or electrons. Nafion1
can be manufactured with various cationic conductivities. In order to
operate the membrane must conduct hydrogen ions (protons), but not
electrons as this would in effect ‘‘short circuit’’ of the fuel cell. The
membrane must also not allow gas to pass to the other side of the
cell, a problem known as fuel crossover (Baxter et al., 1999; Dohle et al.,
2000; Ren et al., 2000), and must be resistant to the reducing environment
at the anode as well as the harsh oxidative environment at the cathode.
Although the PEFC is a prime candidate of the alternative energy source
for the small-scale applications, water management is crucial to perfor-
mance, as power output requires optimized water uptake in the
membrane. Water management is a very difficult subject in PEM sys-

tems. A wide variety of solutions for managing water exist including
integration of electro-osmotic pumps. Furthermore, the Pt catalyst on the
membrane is easily poisoned by carbon monoxide (no more than one
part per million is usually acceptable) and the membrane is sensitive to
materials like metal ions, which can be introduced by corrosion of
metallic bipolar plates. The commercial viability of this device is ham-
pered currently due to its high cost, power density, and low durability.
Key issues or objectives that are needed to be achieved to make a
paradigm shift in PEFC technology are (i) novel materials, obtained
through computational chemistry calculations, including PEMs (pos-
sessing high-temperature operability and low cost), and electrocatalysts
for high reduction reaction kinetics, tolerant to carbon monooxide.
Advanced materials need to be discovered for increased durability
(with an order of magnitude higher than current technology) and meet-
ing the requirements of environment, safety and health; (ii) optimal
design parameters for GDLs (porosity and hydrophobicity), CLs (thick-
ness, composition, and particle size distribution) as well as determina-
tion of operating and design conditions (i.e., optimal temperature, pres-
sure, current density, humidity, geometric parameters, flow
characteristics and arrangements, etc.). This entire PEFC design can
be envisioned as an ultralarge integration of submodels and an optimi-
zation problem with the objective function as a combination of maxi-
mizing the performance, and minimizing the cost, while maintaining
durability. As a series of hierarchical governing equations containing a
large set of parameters to be optimized and conditions to be satisfied,
multi-scale, multiphenomena models are bound to play a pivotal role in
achieving the design goals, in concert with experimentation.
2.2 Nanotribology (HDDs)

Tribology, the subject dealing with friction, wear, and lubrication,
becomes important in academic and industrial community. The
improved technology on friction and wear would save developed coun-
tries up to 1.6% of their gross national product. Such technological
considerations have driven humans to understand friction since pre-
historic ages. By 200,000 B.C., Neanderthals had achieved a clear mas-
tery of friction, generating fire by the rubbing of wood on wood and by
the striking of flint stones. Significant developments are also found 5000
years ago in Egypt, where the transportation of large stone statues for
the construction of the pyramids demanded tribological advances in the
form of lubricated wooden sledges (Figure 6a). Modern microscopic
tribology began 500 years ago, when Leonardo da Vinci deduced the
laws governing the motion of a rectangular block sliding over a flat
[(Figure_6)TD$IG]
Figure 6 Schematic descriptions of examples in tribological advances:

(a) transportation of large stone statues in Egypt (5000 years ago) and
(b) nanotribology of dry surfaces (2000).
surface. His notebooks remained unpublished for hundreds of years.

Later, Amontons’s and Coulomb’s classical microscopic friction laws
have far outlived a variety of attempts to explain them on a fundamen-
tal basis in terms of, say, molecular adhesion (attraction between par-
ticles in the opposing surfaces). Nowadays, the surface consists of many
forms of carbons (hydrogenated or nitrogenerated carbon, C60, carbon
nanotube (CNT), and graphene) as illustrated in Figure 6b. Many efforts
have been made to understand atomistic dry surfaces for nanotribology
including superlubricity (Figure 7) (Coffey and Krim, 2006; Dienwiebel
et al., 2004; Guerra et al., 2010; Lee et al., 2009, 2010; Lucas et al., 2009;
Miura et al., 2003; Schirmeisen, 2010). In this paper, we focus on our
effort in HDD dealing with wet, meso-scale surfaces (Figure 8a).
HDDs have been one of the most dominant data storage systems
with high-density recording capacity yet relatively low cost. Thus, the
information technology area widely utilizes HDD as a main storage
system and its application is currently extended to consumer personal
electronics and portable devices. As shown in Figure 8b, HDD is mainly
composed of several parts: magnetic read/write heads and magnetic
disks (platters), data detection electronics and write circuit, mechan-
ical servo and control system, and interface to microprocessor. A
stack of 3–10 disk platters, each containing a layer of magnetic
medium, is attached to a motor spindle, which rotates the stack at
[(Figure_7)TD$IG]
Figure 7 Examples of nanotribology on dry carbon surfaces for atomic force

microscopy (AFM): (a) schematic description of the out-of-plane graphene
deformation with the sliding AFM (Lee et al., 2010), (b) nanotube without tip (left) and
tip–nanotube interaction under 2.5 nN normal force (right) (Lucas et al., 2009), (c)
stick–slip rolling model with a step rotation of a C60 molecule (Miura et al., 2003), and
(d) ballistic sliding of gold nanocluster on graphite (Schirmeisen, 2010).
speeds of 5,000 to over 15,000 revolutions per minute (RPM). The

read/write head is located on the trailing edge of the head, which is
mounted at the end of the actuator arm, via a servo, which can pre-
cisely control the radial position of the read/write heads to access
data on the rotating hard disk platters.
The actual magnetic processes for the recording occur in the head-
disk interface (HDI) as schematically described in Figure 9, consisting of
flying read/write head and the disk (substrate, underlayer, magnetic
layer, carbon-overcoat, and lubricant layer) (Johnson et al., 1996). The
substrate of choice is aluminum because of its low density and low cost.
However, aluminum by itself is quite soft. Therefore, electroless nickel–
phosphorus-plated (Ni–P) aluminum is universally used today to pro-
vide a hard surface for magnetic film structural support as well as the
capability of being polished to a high degree of smoothness for low-
flying recording heads. The chromium (Cr) underlayer serves to help
[(Figure_8)TD$IG]
Figure 8 (a) Nanotribology in HDD and (b) a typical HDD system.
nucleate and grow the microstructure of appropriate magnetic proper-

ties because of an epitaxial match between the Cr planes and the cobalt
(Co)-based magnetic alloys in the thin film media. Although Co by itself
has a very low coercivity, Co-based binary and ternary alloys are suit-
able for the ferromagnetic material in the magnetic layer, since they
provide a high coercivity that can even be tailored by varying the alloy
composition. Cr is a crucial second or third element here as can reduce
corrosion potential and allow for precipitation of other crystalline
phases at grain boundaries or within grains to reduce noise. The plasma
enhanced chemical vapor deposition (PECVD) amorphous carbon-
overcoat is sputtered over the magnetic layer, because it seals the data
[(Figure_9)TD$IG]
Figure 9 The cross-sectional diagram of a typical HDI for a HDD system with
aluminum substrate.
layer from corrosion and chemical surface deformity, protects the data
layer from physical damage during the intermittent contact between the
head and the disk, and minimizes the thermally induced erasure of data
bits due to the heat transfer to the data layer in the event of a head-disk
contact.
The lubricant film is essentially the first line of protection from the
mechanical damage in the event of intermittent contact between head
and disk, and serves to reduce the friction and wear between carbon-
overcoat and the recording head. It brings additional stability to the
HDD by providing the recording heads a smooth transition from a
region of dragging to flying and by adsorbing some of the energy that
is generated by the head-disk contact. An ideal lubricant candidate is
expected to possess chemical inertness to avoid chemical reaction, low
vapor pressure to prevent evaporation loss, low surface tension to allow
its uniform wetting on the overcoat for the near-field recording, high
stability under shear stress to avoid degradation, and good boundary
lubrication properties (Klaus and Bhushan, 1985). It is equally impor-
tant that the lubricant should present appropriate chemical affinity for
the overcoat and can reside on the disk surface over the HDD lifetime
without desorption, spin-off, and thermal degradation (Karis et al.,
2001; Tani and Matsumoto, 2003; Tyndall et al., 1999). On the other hand,
a reasonable lubricant diffusion capability is also expected for the ‘‘self-
healing’’ purpose in the event of head-disk contact, where lubricant
depleted zones may be formed and thus cause a possible head crash.
The current commercialized lubricant for HDD industry is a class of
Table 1 The endgroup structure, number averaged molecular weight (Mn), and
relevant physical properties of Fomblin Z derivatives
Fomblin Endgroup (X) Mn, Kinematic Vapor Surface

Z structure g/mol viscosity pressure, tension at
at 20 C Torr 20 C
(St)
20 C 100 C
Z03 –CF3 4000 0.3 / / 23

Zdol –CF2CH2OH 2000 0.85 2 105 2 105 24
Ztetraol–CF2CH2OCH2 2200 20 5 107 2 104 /
-CHOH-CH2
-OH
Zdol-TX –CF2CH2(OCH2 2100 1.45 2 105 2 103 23
CH2)pOH
AM2001 2400 0.75 1 107 2 105 25
A20H HO– and/or 3000 / / / 22
oligomeric, random linear copolymers named PFPEs with the chemical

structure of
X[(OCF2CF2)p(OCF2)q]OX (p/q ffi 2/3).
Here, X stands for the functional endgroup. Its weight average
molecular weight (Mw) ranges from 2000 to 4000 g/mol. PFPE Z03
has nonfunctional endgroup (X = CF3), while its derivatives, PFPE
Zdol and Ztetraol, have the functional endgroups (X = CF2CH2OH for
Zdol and X = CF2CH2CH(OH)CH2OH for Ztetraol). Here, the hydroxyl
groups in the chain ends determine the functionality of PFPEs. The
detailed physical properties of various PFPEs are provided in Table 1.
The interaction between PFPEs and disk overcoat is another signif-
icant factor to affect the properties of lubricant films. PFPEs with func-
tional endgroups (e.g., Zdol and Ztetraol) perform better than PFPEs
with nonfunctional endgroup (e.g., Z03) for retention and evaporation
at the expense of the surface mobility or replenishment ability.
However, strong endgroup functionality can lead to the layering and
instability (e.g., surface nonuniformity/dewetting) of PFPE films (Karis
[(Figure_0)TD$IG]
Figure 10 The molecular structure of recently reported lubricants: (a) DDPA-S,

(b) DDPA-D, and (c) ZTMD for ultrasmall HMS.
et al., 2005; Waltman et al., 2002). This interaction can also be tuned via
modifying the surface functionality of carbon-overcoat by introducing
dopants such as hydrogen and nitrogen into the vacuum chamber
during the sputtering process (Lee et al., 1993; White et al., 1996). To
date, the bonding mechanism between PFPEs and overcoat was inves-
tigated via electron spin resonance (ESR) or time of flight secondary ion
mass spectroscopy (TOF-SIMS), where it is postulated that a hydrogen
atom is transferred from a hydroxyl group in Zdol chain end to a
dangling bond site on the carbon film (Kasai, 2002; Kasai and Spool,
2001; Kasai et al., 1999; Zhu et al., 2003). The ab initio calculation sug-
gested the possible hydrogen bonding interaction (Waltman et al.,
1999a), while annealing can also lead to the esterification
(Waltman et al., 1998). Recently, new types of lubricants, whose chem-
ical structures are shown in Figure 10, have been reported to enhance
the performance and reliability of HDD with ultrasmall head-medium
spacing (HMS). Researchers introduced new functional group, dipro-
pylamine, in the Demnum chain end. Both monofunctional (DDPA-S)
and difunctional (DDPA-D) were synthesized and characterized
(Sakane et al., 2006). It was found that DDPA-S can significantly reduce
the head-disk adhesive interaction at near-contact operation as shown
in Figure 10, which is believed to be a promising lubricant for ultrasmall
HMS. With the additional functional groups in the center of PFPE chain,
Ztetraol multidentate (ZTMD) can form additional anchors on the over-
coat surface and be suitable for HMS less than 5 nm (Guo et al., 2006;
Marchon et al., 2006).
[(Figure_1)TD$IG]
Figure 11 A schematic of (a) IBM 3370 and (b) negative pressure heads.
In the current HDDs, the read/write head flies about 6.5 nm above
the surface via the air bearing design. Earlier models of head featured
two straight and flat rails having a taper at the front, known as positive
pressure head, for example, IBM 3370 head (Figure 11a). Air is com-
pressed in the taper region of the head, creating an air bearing that
supports the head above the surface of the disk. Negative or subambient
pressure heads (Figure 11b) are introduced to reduce the normal load,
while maintaining the stiffness of positive pressure designs. The aero-
dynamic operation of negative pressure heads can be explained from
consideration of positive and negative pressure region. The side rails of
the head generate a positive pressure, which tend to separate the head
from the disk. The main recessed region confined by a cross rail and side
rails produces a suction force that attracts the head to the disk. Negative
pressure heads generally exhibit favorable characteristics of low normal
load, high air bearing stiffness, and better altitude performance.
Naturally, one might come up with a way to reduce fly height by
permitting the head to slide over the disk surface (so-called contact
recording) much like the head for a tape drive. To use this zero-spacing
approach, there are still many technological hurdles to be resolved,
such as protecting the head and the disk from heat generation, high
friction, and wear. Instead of achieving full contact, liquid bearing
technology (Lemke et al., 1994), where a viscoelastic liquid is fully
flooded in the HDI instead of compressible air, has been proposed.
In HDI, the head flies above the disk surface with the fluid mechan-
ics of air bearing generated by the rotation of disks at high speeds.
Contact between the head and the disk occurs during the cycles of the
operation start/stop and intermittently during disk operation due to
the fluctuation of the fly height. Since the fly height has been reduced to
6.5 nm, the head can even interact with the disk surface via van der
[(Figure_2)TD$IG]
Figure 12 A magnetic head slider flying over a disk surface (slider located on the
position 1.25 in. from the disk center rotating in 7600 rpm) compared with an aircraft
flying in 560 mile/h over ground with a close physical spacing.
Waals interaction, which induces the growth of roughness of lubricant

surface (Pit et al., 2001). Figure 12 helps understanding a degree of a
close physical spacing of HDI via geometric comparison of an aircraft
flying over ground although nanoscale phenomena cannot be
explained by scaling up to macroscale. As coercivity and areal density
of magnetic layer define the HDD capacity, HDI determines the stability
and durability, which cause the tribological issues due to the mechan-
ical contact. Since magnetic recording is a nanoscale near-field process,
it is essential to maintain a stable interface in the proximity recording for
the long-term reliability of HDDs. In addition, to increase the areal
density, each component in HMS must be reduced in dimension, which
places significant constraints on both head and disk parameters.
Therefore, current HDD demands intensive research and development
on the system stability and reliability, which can be enhanced by multi-
scale approaches on the lubricant molecules, carbon-overcoat on the
magnetic layer, and read/write head (via shape optimization to obtain
reliable and constant flying).
3. STRUCTURE OF HIERARCHICAL EQUATIONS
During the past few decades, various theoretical models have been
developed to explain the physical properties and to find key parameters
for the prediction of the system behaviors. Recent technological trends
focus toward integration of subsystem models in various scales, which
entails examining the nanophysical properties, subsystem size, and
scale-specified numerical analysis methods on system level perfor-
mance. Multi-scale modeling components including quantum mechan-
ical (i.e., density functional theory (DFT) and ab initio simulation), atom-
istic/molecular (i.e., Monte Carlo (MC) and molecular dynamics (MD)),
mesoscopic (i.e., dissipative particle dynamics (DPD) and lattice
Boltzmann method (LBM)), and macroscopic (i.e., LBM, computational
fluid mechanics, and system optimization) descriptions have also

gained tremendous attention. Here, we will explain the general
modeling methods and the structure of hierarchical equations, which
are broadly utilized in multi-scale approaches. System specified
modification of each modeling method will be discussed in the next
chapter.
3.1 Quantum level

Theoretical models at the quantum level are important ingredients in
the multi-scale modeling strategy, since the techniques do not require
empirical knowledge for effective calculation. At this level of scale, the
molecular system is described by utilizing various ab initio quantum
mechanical calculations, which are currently available to solve many
body wave-function problems represented by the Schr€ odinger equation
(Jensen, 1999). The DFT is an alternative method to reduce the compu-
tational load by using energy functionals depending on the density of
the particles (Parr and Yang, 1989). In this calculation, the ground-state
electronic energy is determined by the electron density representing the
correspondence between the electron density and the energy
(Hohenberg and Kohn, 1964). Although the quantum level calculation
promises accuracy since the model is based on the fundamental and
nonempirical rules, the calculation demands a huge computational cost;
as a consequence the technique cannot be utilized for the upper level of
the scale represented as massive molecular systems, which are more
than approximately 1000 atoms. Therefore, quantum level models are
generally utilized with a combination of molecular-level theories. For
instance, Leconte et al. (2010) investigated the electronic and transport
properties of ozone-treated graphene by using multi-scale ab initio with
a real space order-N transport computational methodology based on a
reparameterized tight-binding Hamiltonian. Many researchers have
pursued a multi-scale computational approach to the theoretically
inspired optimization of the electroactive properties of organic and
hybrid materials (Dalton et al., 2007; Kim et al., 2008; Olbricht et al.,
2008; Pereverzev et al., 2008; Sullivan et al., 2007). Most recently, this
approach has evolved into a correlated quantum/statistical mechanical
approach based on improvements to real-time, time-dependent density
functional theory (RTTDDFT) and pseudo-atomistic Monte Carlo/
molecular dynamics (PAMCMD) calculations (Dalton, 2009). While
the initial target for this effort was the transformative improvement of
organic electrooptic materials, theoretical approach is also relevant to
developing improved organic electronic, light emitting, photorefrac-
tive, and most particularly organic photovoltaic materials. Although
DFT presented an advantage in the balance of computational cost and
accuracy in quantum mechanics calculations, the method cannot

accurately handle the ubiquitous dispersion interactions (London
dispersion forces), which is difficult to describe most of theoretical
models including ab initio calculation (Johnson and DiLabio, 2006;
Kohn et al., 1998; Meijer and Sprik, 1996). Empirical atom–atom
based correction terms of C6/R6-type are commonly used to deter-
mine the parameters (LeSar, 1984; Meijer and Sprik, 1996). Recently,
DFT for dispersion interaction has been investigated including
parameterized functionals and dispersion-correcting potentials
(Johnson et al., 2009).
At this level, we note that we can obtain first-principles (no
adjustable parameters) prediction of properties such as electrooptic
activity in complex organic and hybrid materials. Organic and
hybrid materials have been also integrated within silicon electronic
and photonic device structures, and it enables theoretical analysis of
the performance of these novel device architectures, which involve
enormous enhancement of optical and electric fields. This work
provides a proof of concept that detailed quantum mechanical meth-
ods can be used to inform macroscale experimental systems, pro-
ducing new materials with dramatically improved properties. In the
field of PEFC and HDI, quantum chemical modeling based on atom-
istic simulation was introduced extensively to investigate the per-
fluoro polymeric systems (i.e., PEM and PFPE) and the interactions
of their functional bodies (Goddard et al., 2006; Waltman et al., 1999a,
1999b).
3.2 Atomistic/molecular level

Classical molecular simulation methods such as MC and MD represent
atomistic/molecular-level modeling, which discards the electronic
degrees of freedom while utilizing parameters transferred from quan-
tum level simulation as force field information. A molecule in the sim-
ulation is composed of beads representing atoms, where the interac-
tions are described by classical potential functions. Each bead has a
dispersive pair-wise interaction as described by the Lennard–Jones
(LJ) potential, ULJ(rij):
" # 12 6
s s
U LJ ðrij Þ ¼ 4eij ; ð1Þ
rij rij
where s is the diameter of beads, eij corresponds to the well depth of LJ

interaction, rij denotes the distance between two beads i and j. To model
the electrostatic effects, the nonbonding interactions were modified by
the combination of van der Waals interaction and the electrostatic

Coulombic interaction:
" 6 #
s 12 s qi qj e2 1
U LJC ðrij Þ ¼ 4eij þ ð2Þ
rij rij 4pe0 rij
where, e0 is the dielectric constant of vacuum, e represents elementary

electric charge, and qi and qj are the charge parameters for Coulombic
interaction. Chemical bond between adjacent beads is commonly repre-
sented by harmonic potential energy:
U r ¼ ð1=2ÞKr ðr ro Þ2 and U u ¼ ð1=2ÞKu ðu u0 Þ2 ð3Þ
where Ur and Uu are the stretching and bending potentials, respec-

tively, Kr is the stretching force constant, Ku is the bending force con-
stant, ro is the equilibrium bond length, and uo is the equilibrium angle.
The stretching interactions take place between two bonded atoms,
while the bending accounts for the bending of the angle formed by
two adjacent bonds. Alternatively, anharmonic finitely extensible non-
linear elastic (FENE) springs connect the adjacent beads separated by a
distance of rib:
8 " 2 #
>
< 1 2 rib
kR0 ln 1 rib < R0
U FENE ðrib Þ ¼ 2 R : ð4Þ
>
:
0
1 otherwise
Here, rib denotes the interbead distance (i.e., the bond length between
two adjacent beads), k is the spring constant that quantifies the rigidity
of the bond, and R0 is the maximum extensibility of the spring. The form
of torsional potential parameters, describing four bonded atoms, is
X
N 1
U f ¼ kf An cosn f ð5Þ
n¼0
where Uf is the torsional potential, kf is a constant, An are the torsional

potential coefficients, and f is the torsional angle. To incorporate clas-
sical potential functions with quantum mechanical calculations, eigen-
value analysis is developed to calculate stretching and bending para-
meters from the ab initio Hessian matrix for the harmonic potential form,
and the torsional potential parameters are calculated by generating the
ab initio torsional energy profiles via a series of constrained geometry
optimization.
[(Figure_3)TD$IG]
Figure 13 A trial movement is generated in a cube with a dimension of rmax centered

by the selected bead. Red line represent skeletal bond.
MC, which is a time-independent molecular simulation, utilizes the

stochastic motion of the bead to measure the static properties or to
equilibrate the system before measuring time-dependent dynamic
properties using MD due to its advantages in computational cost. A
single bead is selected randomly for the MC trial movement during each
attempt. The displacement vector is set randomly in a cube centered by
the selected bead as shown in Figure 13 (r ¼ xi þ yj þ zk with rmax / 2
< x, y, z < rmax / 2). Here, r denotes the position vector of beads, i, j, and
k are unit vectors in x, y, and z directions, and rmax is the maximum
displacement allowable in the trial movement. For each trial movement,
the total potential energy difference before and after the trail movement
is calculated from the system potential energies as follows:
U total U inter þ U intra ð6Þ
where Uinter and Uintra represent intermolecular (e.g., LJ and Coulombic

interactions) and intramolecular (e.g., stretching, bending, and torsion)
energies, respectively. The acceptance of a trial move is determined by
the probability distribution law based on the detailed balance to keep
the system in equilibrium. Metropolis acceptance probability W deter-
mines whether the trial movement will be accepted or not (Frenkel and
Smith, 2000; Metropolis et al., 1953):
W ¼ min½expðU total =kB TÞ; 1: ð7Þ
Here, kB and T are the Boltzmann constant and absolute

temperature, respectively. When W exceeds a random number ranging

between 0 and 1, the trial movement is accepted, and the position of
bead is updated. Otherwise, the bead is placed back to the original
position. The trial movement of beads will be repeated for enormous
number of time steps, which depends on the system size, until the
potential energy of the system approaches the statistical constant, indi-
cating the equilibrium state. MC provides enormous advantages in
calculation cost by comparing MD, which utilizes the dynamic equa-
tions, for novel algorithms and ensembles. These have been intensively
investigated during past few decades for molecular-level analysis of
complex phenomena including phase equilibria and massively
entangled polymeric systems (Theodorou, 2010).
The molecular motion in MD simulation is deterministic by solving a
Hamiltonian system (Allen and Tildesley, 1996). For the precise
description of the polymeric systems, Langevin dynamics (Grest,
1996) were employed, where the force acting on the ith bead in the
ath molecule can be calculated by the following equation:
d2 rai @U drai
m 2
¼ z þ f ai ðtÞ: ð8Þ
dt @rai dt
Here, m and rai are the mass and position vector of beads, respec-
tively. z is the friction tensor, which is assumed to be isotropic for
simplicity in our simulation, that
is, z = G I, where I is the unit dyad
and G = 0.5t 1 t ¼ s ðm=eÞ0:5 (Grest, 1996). Further, f ai is the
Brownian random force, which obeys the Gaussian white noise, and
is generated according to the fluctuation–dissipation theorem:
0 0
hf ai ðtÞf bj ðt Þi ¼ 2kB Tdab I dij dðt t Þ; ð9Þ
where the angular bracket denotes an ensemble average. dab and dij are
Kronecker deltas, and d(t – t0 ) is a Dirac’s delta function. T and quantify
the magnitude of Brownian force. Langevin equation is a phenomeno-
logical stochastic differential equation of motion describing time evo-
lution of a subset of the degrees of freedom for slowly relaxing (mac-
roscopic) variables while the rapidly relaxing (microscopic) variables,
which result in the stochastic nature in the equation. Langevin equa-
tions can be systematically derived via standard Mori and Zwanzig
projection operator method (Hijón et al., 2010) by projecting out fast
degrees of freedom (irrelevant variables contained in fluctuating ran-
dom force, f). The dynamics of relevant variables are described by
potential energy calculated from quantum mechanics and information
containing irrelevant variables through f or equivalently, z, which are
related by fluctuation–dissipation theorem shown in Equation (9).
Using EMD simulations, transport properties can be obtained via the

Green–Kubo linear response theory and the corresponding time corre-
lation functions. However, one of the main concerns for EMD method is
that the fluctuations naturally occurring in the equilibrated system are
fairly small. As a result, the signal-to-noise ratio (SNR) is very unfavor-
able at large time range, where the time correlation functions exhibit
long tails giving a significant contribution to the integrals defining the
corresponding transport properties. However, if we can introduce
much larger fluctuations artificially, the SNR of the measured response
may be improved dramatically. Therefore, nonequilibrium molecular
dynamics (NEMD) is introduced to study the nonequilibrium dynamics
such as rheological properties. Historically, the first fictitious force
method proposed for simulating viscous flow was the Doll’s tensor
method (Hoover et al., 1980) as shown in Equation (10), which can be
derived from the Doll’s tensor Hamiltonian.
drai 1
¼ pai þ rai ru
dt m
ð10Þ
dpai @U
¼ ru pai :
dt @rai
y
Here, pai is the bead momentum vector and uðrai ; tÞ ¼ ig rai is the
linear streaming velocity profile, where g @ux =@y is the shear strain
rate. Doll’s method has now been replaced by the SLLOD algorithm
(Evans and Morriss, 1984), where the Cartesian components that couple
to the strain rate tensor are transposed (Equation (11)).
drai 1
¼ pai þ rai ru
dt m
ð11Þ
dpai @U
¼ pai ru:
dt @rai
y
With uðrai ; tÞ ¼ ig rai , Equations (11) is equivalent to
2
d rai @U dg y
m ¼ þ im r : ð12Þ
dt2 @rai dt ai
Both the Doll’s and SLLOD algorithms are correct in the limit of
zero-shear rate. However, for finite shear rates, the SLLOD equations
are exact but Doll’s tensor algorithm begins to yield incorrect results at
quadratic order in the strain rate, since the former method has suc-
ceeded in transforming the boundary condition expressed in the form
of the local distribution function into the form of a smooth mechanical
force, which appears as a mechanical perturbation in the equation of
motion (Equation (12)) (Evans and Morriss, 1990). To thermostat the
SLLOD algorithm, the thermostat Gaussian multiplier c is introduced:

drai 1 1 y
¼ pai þ rai ru ¼ pai þ ig rai
dt m m
ð13Þ
dpai @U @U y
¼ pai ru cpai ¼ ig pai cpai :
dt @rai @rai
In the
P isokinetic
P version of the algorithm, where the total kinetic
energy a i p2ai =2m is held constant, c is given by
X X @U y
XX
c¼ pai g pxai pai = p2ai : ð14Þ
a i
@r ai a i
To bridge the time-scale gap between microscopic and macroscopic

scales and accurately capture dynamic phenomena on the coarse-
grained level, systematic time-scale-bridging molecular dynamics was
recently introduced by using an alternative MC–MD iteration scheme,
which also shows higher calculation efficiency than standard NEMD
(Ilg et al., 2009).
3.3 Meso-scale/continuum level

LBM was introduced as an alternative meso-scale/continuum-level
modeling tool, which has the advantages in capturing clear physics in
the system with the complex geometry and nanoscale physics. Since
LBM covers broad range of the system scale and is based on the particle
assumption, the method is considered as a multi-scale method from
meso-scale to continuum scale including buffer region simulation to
substitute computational fluid dynamics as well as a promising candi-
date for hierarchical integration with atomistic/molecular-level mod-
els. Physical phenomena in air bearings/viscoelastic liquid bearings in
the HDI system and heat transfer phenomena have been modeled via
LBM (Ghai et al., 2005, 2006a, 2006b; Kim et al., 2005a, 2005b). Especially,
due to the convenience of complex geometry manipulation, LBM is
suitable for modeling flow in porous media, which can be utilized for
GDL simulation in PEFC.
LBM has emerged as a promising numerical tool for simulating fluid
flows and thermal management with complex physics (Chen and
Doolen, 1998). The numerous advantages including clear physical pic-
tures, an inherently transient nature, multi-scale simulation capabilities,
and fully parallel algorithms make LBM to be an attractive candidate as
a multi-scale simulation tool. Kim et al. (2005a, 2005b) have developed a
novel LBM by adopting a spatially dependent relaxation time model to
predict the nanoscale air bearing performance. Unlike conventional
numerical methods, which discretize the macroscopic equations, LBM
constructs simplified kinetic models incorporating the essential physics

of microscopic processes so that the macroscopic properties obey the
desired equations. The two-dimensional, lattice Boltzmann kinetic
equation (LBKE) with Bhatnagar–Gross–Krook (BGK) approximation
can be written as (Mei et al., 2000)
1
f i ðx þ ci t; t þ tÞ ¼ f i ðx; tÞ ½f i ðx; tÞ f eq
i ðx; tÞ
t ð15Þ
for i ¼ 0; 1; . . . ; N
where t is the single relaxation time, which controls the rate of

approach to equilibrium; fi(x, t) is the discrete one particle distribution
function, which is the probability of finding a particle with the velocity
ci at (x, t); t is the time step; N is the number of discrete particle velocities
in each node, which is chosen to be 9 for D2Q9 (i.e., two-dimension and
9 directions of streaming) model used in this simulation; f eq i ðx; tÞ is the
discrete equilibrium distribution function given as

e v vv : ðei ei c2 IÞ
f eq
i ðx; tÞ ¼ w i r 1 þ þ
c2 2c4
8
> 4=9; i ¼ 0 ð16Þ
<
with wi ¼ 1=9; i ¼ 1; . . . ; 4
>
:
1=36; i ¼ 5; . . . ; 8
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where v is the fluid velocity, c s=t ¼ 3kB T=m is the lattice speed, Ds
[(Figure_4)TD$IG]
Figure 14 A lattice node of D2Q9 model (2D lattice with 9 directions of streaming)
is the lattice spacing, kB is the Boltzmann constant, T is the fluid tem-

perature, and m is the particle mass. The 9 discrete velocities in D2Q9
model (Figure 14) are given by
8
< ð0; 0Þ; i ¼ 0
ci ¼ ð 1; 0Þc; ð0; 1Þc; i ¼ 1; . . . ; 4 ð17Þ
:
ð 1; 1Þc; i ¼ 5; . . . ; 8:
The density r and v are calculated by

X 1 X
rðx; tÞ ¼ fi and vðx; tÞ ¼ cf ð18Þ
i
rðx; tÞ i i i
In order to integrate polymeric systems via bridging among meso-

scale/continuum scales, LBM and atomic/molecular scales, MD, and
macroscopic scales, intermediate level models may be required. These
maintain the nature of polymeric system with or without functional
groups and provide coarse-grained levels that can be integrated to
LBM. In Section 4.2, we provide a simple reactive sphere (SRS) model,
which is one of the intermediate levels of system integration for reactive
functional endgroups.
3.4 Process-scale level

Process-scale models represent the behavior of reaction, separation and
mass, heat, and momentum transfer at the process flowsheet level, or
for a network of process flowsheets. Whether based on first-principles
or empirical relations, the model equations for these systems typically
consist of conservation laws (based on mass, heat, and momentum),
physical and chemical equilibrium among species and phases, and
additional constitutive equations that describe the rates of chemical
transformation or transport of mass and energy. These process models
are often represented by a collection of individual unit models (the so-
called unit operations) that usually correspond to major pieces of pro-
cess equipment, which, in turn, are captured by device-level models.
These unit models are assembled within a process flowsheet that
describes the interaction of equipment either for steady state or
dynamic behavior. As a result, models can be described by algebraic
or differential equations. As illustrated in Figure 3 for a PEFC-base
power plant, steady-state process flowsheets are usually described by
lumped parameter models described by algebraic equations. Similarly,
dynamic process flowsheets are described by lumped parameter mod-
els comprising differential-algebraic equations. Models that deal with
spatially distributed models are frequently considered at the device
level, with partial differential equations that model fluid flow, heat and
mass transfer and reactions. These are usually considered too expensive
to incorporate within an overall process model. Process flowsheeting
models, embodied in commercial process simulators such as ASPEN
Plus1, HYSYS1, or PRO-II1, have become the accepted standard to
describe and evaluate process engineering systems, for petroleum refin-
eries, chemical processes, and power plants. Not only do they provide
quantitative information on the flow of material and energy throughout
a process network but they are used extensively to assess and analyze
the current state of the process, improve the operation of existing pro-
cesses, and guide and validate the design of new processes. As a result
of their application in all of these engineering tasks, these process
models have also become the medium of communication for crucial
technical material shared among networks of project teams that partic-
ipate in any engineering operation. On the other hand, simplifications
used in process models (e.g., equilibrium-based, lumped parameter,
and other short-cut models) often lead to inaccuracies and performance
limitations.
Consequently, these models often require model tuning through
semiempirical correlations and data integration. Such tasks are time-
consuming and problem specific as they often require information from
additional experiments and pilot plant trials, with missing information
leading to start-up and operational risks. The incorporation of more
accurate multi-scale phenomena (at device-, meso- and even molecular
scales) captured by reduced-order models (ROMs) will overcome these
limitations (Lang et al., 2009, 2011).
4. SYSTEM INTEGRATION
As introduced in the previous sections, multi-scale modeling can be

realized via effective hierarchical analysis and simulation strategies
correlating various models in different scales, where each level
addresses specific physical phenomena (Gorban et al., 2006; Gubbins
and Moore, 2010; Maginn and Elliot, 2010). The classical concept of the
multi-scale modeling method is straightforward integration among the
various scale levels, which utilizes the simultaneous description of all-
scale subsystems. This method provides real-time multi-scale observa-
tion yet the computational time length depends on the slowest calcula-
tion. To overcome the disadvantage in the computational cost while
retaining multi-scale advantages, the hierarchical multi-scale model,
which utilizes parameters in lower level of scale to obtain new para-
meters for the upper level degree of freedom, has been investigated
(Theodorou, 2005; Ulherr and Theodorou, 1998). The hierarchical multi-

scale model allows each level of calculation performed independently,
and the input parameters can be estimated from other scale level mod-
els, which discard unnecessary calculations during the bridging proce-
dure between the different scales (Delle Site and Kremer, 2005; Delle
Site et al., 2002, 2004; Doi, 2003; Glotzer and Paul, 2002). Recently, the
focus of multi-scale modeling strategy is on developing bridging
methodology connecting quantum–atomistic/molecular–meso-scale–
continuum levels (Broughton et al., 1999; Csanyi et al., 2004; Delgado-
Buscalioni and Coveney, 2003; Faller, 2004; Flekkoy et al., 2000;
Hadjiconstantinou, 1999; Laio et al., 2002; Li et al., 1998; Neri et al.,
2005; O’Connell and Thompson, 1995; Rafii-Tabar et al., 1998;
Smirnova et al., 1999; Villa et al., 2004).
4.1 Bridging methodology between different scale levels

The bridging procedure finds reduced-order parameters for upper level
scale models. As shown in Figure 15, ROMs are introduced to capture
the predictive behavior of the lower scale model and provide the links to
capturing behavioral information from all of the lower scales, while
[(Figure_5)TD$IG]
Figure 15 Linking models at various scales using ROMs and deriving lower scale
specifications through an inverse optimization formulation. The ROM included at each
scale is a reduced representation of the model at the scale below that could range from
a set of parameters such as, for example, elementary rate constants to complex models
derived from proper orthogonal decomposition and perhaps even to the full lower
scale model. This is symbolized by coloring the ROM box with the same color as that of
the box representing the adjacent lower scale model.
allowing the integrated formulation to be tractable. The major role of

multi-scale ROMs is that they allow feasible realizations of complex
domain models (consistency) and capture accurate complex model
behavior over a wide range of the decision space (performance). With
the development of ROMs at each level, modeling and optimization
formulations at a given level will capture performance and feasibility at
neighboring levels. By allowing this communication between levels,
accurate and efficient decision making can be made. Also, note that
the bidirectional flow between levels in Figure 15 easily allows us to
develop and generalize ideas of reverse engineering and inverse
problems.
The coupling of models at different scales is a challenging subject in
the integration procedure, yet is most important for the accuracy of the
multi-scale models. Especially the computational cost associated with
large-scale calculations often precludes their integration over time and
length scales, while rigorous models are now widely applied at all
modeling scales. As an essential tool to overcome this barrier, ROM
plays a critical role to link detailed phenomena at all modeling scales.
While ROMs are widely applied over the entire modeling spectrum,
their development is usually done as a one-time activity at an ad hoc
level. In the development of a multi-scale modeling and optimization
framework, it is required to develop a systematic approach for ROM
development and integration. This will lead to the development of a
single framework that promises much more detailed predictions of
system-wide dynamics with high accuracy and interdependencies for
large-scale decision making.
Depending on the particular length and time scale and the applica-
tion domain, ROMs take a variety of different forms. Often, physics-
based analytic models are derived from simplified, limiting behaviors
of transport, reaction and equilibrium phenomena, and conservation
laws. At the process engineering level, these lead to a rich and widely
applied model library, which must nevertheless be extended to newer
technologies in reaction and separation. Examples of these include
macroscopic models for rate laws, vapor liquid equilibrium and ther-
modynamic properties used at the level of process networks. In addi-
tion, process models are often derived that consist of semiempirical
functional forms fitted with data including reactor models and process
models derived from data-derived correlations. At the device level,
which is dominated by models in continuum mechanics, ROMs take a
variety of forms, ranging from reduced-order solutions involving
proper orthogonal decomposition (POD), variable resolutions models
on meshes with varying degrees of refinement, variable-fidelity physics
models such as inviscid, irrotational, and incompressible flow for
Navier–Stokes equations. Lastly, at atomistic and molecular levels, the

task is to infer thermodynamic and kinetic properties and constants that
are later used in macroscopic physics-based models. Often these calcu-
lations are used directly to regress to physics-based ROMs, such as with
cluster expansion and kinetic MC modeling, or through coarse-grained
models described below. Moreover, for complex device or molecular
models, such as time consuming continuum mechanics and probabilis-
tic models with little or no structural information, general regression-
based model are derived that apply data-driven regression approaches
(e.g., PCA, compressed sensing, neural nets, and wavelets). At all of
these levels, ROMs have approximation errors that prevent their direct
integration for multi-scale decision making. Nevertheless, uncertainty
bounds can be developed along with these models, as described in
Oberkampf et al. (2002) and Frenklach et al. (2002, 2004). These bounds
lead to confidence regions in the parameter space of the ROM, which
can be propagated through the multi-scale model hierarchy for the
verification and validation task.
The overall approach introduces challenges for numerical methods,
error estimates and linking with physics-based models. Moreover, with
the availability of ROMs across the spectrum of relevant problem scales,
one can develop multi-scale optimization formulations that act as the
‘‘glue’’ toward their integration, leading to the entire model. The fol-
lowing sections detail these formulations and discuss enabling algorith-
mic and computational technologies required to realize this integrated
multi-scale framework.
4.2 Coarse-graining methods

4.2.1 Quantum—atomistic/molecular levels
From the equilibrium geometries in the quantum mechanical models,
the atomistic intramolecular force field parameters, which are potential
parameters in atomistic MD simulation, can be obtained by using the
method of Seminario (1996). With the combination of atomistic MD
simulation, quantum mechanical properties of the materials have pro-
vided the advances in the study of the fundamental phenomena in
nanoscale. However, the processes at the device or system level occur
above the microscale, where the quantum level as well as atomistic level
cannot cover. Therefore, the development of a coarse-graining proce-
dure, which simplifies the detailed structure of molecules while keep-
ing the fundamental phenomena from quantum/atomistic level mod-
els, becomes one of the critical issues in atomistic/molecular-level
modeling. Izvekov and Voth (2005a, 2005b) introduced the force-match-
ing method for coarse-graining, which determines a coarse-graining
potential from atomistic information through a variational minimiza-

tion procedure, and the method has been applied to the study of various
material systems (Iuchi et al., 2007; Izvekov et al., 2005, 2006a, 2006b; Shi
et al., 2006; Wang et al., 2006a; Zhou et al., 2007). Reverse Monte Carlo
(RMC) techniques were introduced to simulate a particle system to
produce the correct radial distribution function without the explicit
need for a potential (McGreevy, 1995; McGreevy and Pusztai, 1988;
Soper, 1996). The method was further extended to DNA studies by
Lyubartsev and Laaksonen (1995). Reith et al. (2003) successfully devel-
oped coarse-grained modeling for the polymer by using iterative
Boltzmann inversion. Further, the coarse-graining framework was
expanded with a general statistical mechanical theory by introducing
the many-body potential of mean force, which enables coarse-graining
model consistent with atomistic models (Noid et al., 2008a, 2008b; Wang
et al., 2006a). The coarse-grained sites are constructed by clusters of
atoms where the mass of the coarse-grained site is the sum of the
included atoms. The new coarse-grained bond lengths are calculated
from the distance between the centers of mass of the clusters of atoms as
shown in Figure 16.
The method aims to pass the distribution of structural parameters
from the atomistic structure to the coarse-grained. The procedure
involves performing an atomistic MD simulation and calculating the
[(Figure_6)TD$IG]
Figure 16 Mapping of atomistic clusters onto coarse-grained sites. RCM represents

the bond between coarse-grained sites, where the sites are the centers of mass of the
atomistic clusters.
equilibrium probability distributions of the corresponding coarse-

grained bond lengths, bond angles, and torsional angles. A
Boltzmann distribution is fit to the distribution determined from the
MD simulation by varying the parameters of the bonded potential:
PðrÞ / expðU r ðrÞ=kB TÞ ð19Þ
where P is the probability distribution of bond lengths, Ur is the bond

potential as defined in Equation (3), kB is the Boltzmann constant, and T
is the absolute temperature. A similar approach is used for the bond
angles and torsions. In order to correct for any error in the parameter
calculation due to fitting a Boltzmann distribution to a condensed phase
simulation, a coarse-grained MD simulation is run by using the param-
eter from the fit, and a new probability distribution is calculated. An
updated Boltzmann distribution is fit to the existing distribution, gen-
erating another set of potential parameters. Parameters determined
from both the atomistic and the coarse-grained simulations are used
to calculate the final coarse-grained potential parameters as shown in
Equations (20) and (21):
KIr ¼ ðKIIr Þ2 =KIII

r ð20Þ
rIo ¼ 2rIIo rIII

o ð21Þ
where I denotes the final coarse-grained parameter, II denotes a param-

eter determined from the atomistic simulation, and III denotes a param-
eter determined from the coarse-grained simulation. A similar
approach can be applied to the bond angle and torsional potentials.
4.2.2 Molecular—meso-scale levels

Complementary hierarchical models have been developed very
recently (Pearson et al., 2009) that use MD and finite-element methods
to examine nanoscale friction. However, for most of the systems, a
buffer simulation level between molecular and meso-scale has rarely
been investigated for the system containing complicated dynamics.
However, this is a critical and challenging issue for the complete
multi-scale simulation up to the continuum/device levels, although
LBM reduces the scale difference as a meso-continuum scale simulation
tool. For the HDD, it has been found that functional PFPE anchored on
the surface or the coiled conformation of oligomer shows polarity due to
the high interaction of functional groups, which forms coupling
between functional PFPEs. A SRS model, which is highly coarse-
grained molecular model, was invented to simplify polymer molecules
[(Figure_7)TD$IG]
Figure 17 A sketch of the rigid units of an oligomeric PFPE molecule: (a) the flexible
bonds with freely jointed beads and springs for coarse-grained bead-spring model and
(b) SRS model with polarity (red arrow).
by considering the nature of coil conformation of polymer and polarity

due to the functional groups. The SRS model describes the polymer
chain with simple sphere with spins representing polarity (Figure 17).
SRS was originally developed by Ma et al. (1999a) to simulate the
spreading profiles of single component PFPE films over the carbon-
overcoat via the MC. In the integration procedure of multi-scale model-
ing, the static properties (the spin orientation and radius of gyration of
SRS, see Figure 18) from coarse-grained model can be estimated as SRS
parameters. This implies that the SRS model can be systematically
derived from ab initio via sequential coarse graining. SRS can be used
as fundamental building block for constructing an LBM description for
a lubricant/viscoelastic bearing.
[(Figure_8)TD$IG]
Figure 18 Generalized SRS model with orientational distribution g(u) of molecules in

(a) top and (b) bottom layers.
4.2.3 Meso-scale—continuum levels

For the integration of the multi-scale and multiphenomenological sim-
ulation, the strategy is to describe the entire system using LBM with
parameters provided by lower level models. For the system with static
geometry in equilibrium (e.g., PEFC), the geometric parameters can be
accurately estimated via molecular-level simulation transferable to the
continuum scale simulation. Wang et al. successfully simulated the
continuum/device-level PEFC with direct numerical simulation tech-
niques incorporating to a mesoscopic pore-scale description
(Mukherjee and Wang, 2006, 2007; Wang et al., 2006b, 2006c, 2007),
and Jain et al. (2010) expanded a continuum model to the molecular
and system optimization scales via a middle-out approach to examine
the effect of subsystems on the fuel cell performance. Systems with
more complicated dynamics of subsystems (e.g., HDI) demand specific
intermediate level of coarse-graining steps such as SRS. To devise LBM
elucidating the behavior of the lubricant film, SRS, which is constructed
using simplified information from MD via reactive coarse-grained par-
ticles, needs to be provided.
Since Boltzmann transport equation (BTE), which is derived to
LBKE, is particle assumption-based theory, an SRS model can be imple-
mented to BTE as follows:
@t f ji þ v@s f ji þ F@v f ji ¼ ðf j;0 j

i f i Þ=t ð22Þ
where s, n, i, 0, t , and F are space, velocity, discrete directions, equilib-

rium state, relaxation time, and an external force, respectively.
Superscript j denotes the orientation spin, which is statistically calcu-
lated from coarse-grained bead-spring model. The Ising model in fer-
romagnetism, the SRS model in lubricant theory, and a lattice gas are
analogous mathematically (Jhon and Choi, 2001; Lee and Yang, 1952).
At a lattice site, the Ising and SRS models simplify its spin as upward (")
or downward (#). Lattice gas considers that each lattice site is vacant (0)
or occupied (1). Therefore, it is possible to construct a corresponding
LBM model that can describe the mechanism drawn by spin interac-
tions. The SRS model represents an oversimplified molecular picture
but contains the essence of the molecule/surface interactions for
describing molecular functionality (Karis et al., 2005). In the model,
the spreading properties of lubricants are explained via spin character-
istics, that is, S = 1 (occupied sites) or S = 0 (vacant sites). In case of
spreading of lubricants with polar endgroups, Sz (the projection of S in
the direction perpendicular to the substrate, z) is used to identify the
orientation of a polar endgroup (Sz = 1, where a positive value denotes
an upward-pointing endgroup) (Jhon and Choi, 2001). Only four
different interactions, molecule/molecule, molecule/surface, end-

group/endgroup, and endgroup/surface are considered for a simpli-
fied modeling. Thus, the Hamiltonian of this system may be written as
JX X Si;j;k
H ¼ Si;;j;k Si 1;j 1;k 1 A
2 i;j;k i;j;k k3
X1 X ð23Þ
K ðSzi;j;k þ 1ÞðSzi;j;kþ1 1Þ W Szi;j;k
i;j;k
4 i;j;k¼1
where k is a counting index normal to the surface. The first two terms in
the right-hand side represent interactions for nonreactive molecules
(molecule/molecule interaction and molecule/surface interaction),
implying that J is the nearest neighbor coupling constant, and A is
related to the Hamaker constant originating from van der Waals inter-
actions. The third and fourth terms represent characteristics of end-
group (descriptive for polar endgroup molecules only), where K and
W represent the interaction between endgroups and the interaction
between endgroups and the surface, respectively. Therefore, by setting
K = W = 0, the Hamiltonian for nonreactive molecule systems is recov-
ered for nonfunctional polymeric systems.
Using the Hamiltonian, we can obtain attractive or repulsive forces
that play a role of external forces in Equation (22). A spin analogy/
lattice gas model will be developed that can describe the oversimplified
molecular structure, while still capturing the essence of the molecule/
surface interaction. The relaxation time in SRS–LBM will contain shear
rate and other nanoscopic information.
5. TECHNOLOGICAL APPLICATIONS: PEFC
This section deals with multi-scale models for the PEFC and consists of
three subsections, 4.1, 4.2, and 4.3, that relate to molecular-level models,
bridging models between scales and device/process level models,
respectively. The objectives of these subsections are to survey the devel-
opment and application of these models.
5.1 PEM
The PEM generally consists of polytetrafluoroethylene chains with
hydrophilic perfluorosulfonate side groups. Water molecules within
the system agglomerate in the vicinity of hydrophilic groups (i.e., sul-
fonic acid groups) and form hydrophilic clusters. A network of these
clusters forms passages for proton conduction within PEM, which is
critical to PEFC performance. Hence a detailed relationship between

PEM structure, water uptake, and proton conduction is necessary for
synthesis of novel membrane materials, which overcome the limitations
of state-of-the-art PEMs. Ab initio models of PEM have been studied to
explain the first-principle dynamics of proton conduction mechanism
in hydrated PEM (Choe et al., 2010; Habenicht et al., 2010). Quantum
mechanical calculations are incorporated to atomistic MD simulations
by providing accurate potential energy functions as inputs to MD simu-
lations. The atomistic MD exhibits the correspondence to the experi-
mental data as well as provides PEM structural information and proton
transfer mechanisms in PEM (Jinnouchi and Okazaki, 2003).
Komarov et al. (2010) utilized particle-based and field-based simulation
techniques (i.e., integration of atomistic MD and dynamic DFT) to
investigate the processes of self-organization in the systems of sulfo-
nated poly(ether ether ketone)s in the presence of water. Goddard et al.
(2006) used similar overlapping simulation methodologies for the PEM
as well as CL and successfully applied a reactive force field from quan-
tum mechanical calculations to larger scale MD simulations retaining
the accuracy. The molecular-level model can be also applied for differ-
ent electrocatalyst materials to determine atomistic or molecular
mechanisms for electrochemical reactions and degradation of electrode.
Franco and Gerard (2007, 2008) and Franco et al. (2009) analyzed deg-
radation mechanisms of cathode CL in PEFC by using multi-scale mech-
anistic models.
Atomistic MD models can be extended to the coarse-grained level
introduced in the previous section, which is determined by the dimen-
sion of the backbone chain and branch. For the precise description of
water molecular behavior, simple point charge (SPC) model was
adopted (Krishnan et al., 2001), which can be used to simulate complex
composition systems and quantitatively express vibrational spectra of
water molecules in vapor, liquid, and solid states. The six-parameter
(DOH, a, b, Lgu, Lgg , and Luu) SPC potential used for the water molecules
is shown in Equation (24):
U 123 ¼ DOH ½ð1 exp½a r12 Þ2 þ ð1 exp½a r13 Þ2

" #
Luu ðr23 Þ2
þ Lrr r12 r13 þ Lgu ðr12 þ r13 Þr23 þ ð24Þ
2
expfb½ðr12 Þ2 þ ðr13 Þ2 g with rij ¼ rij re;ij ;
where re,ij is the equilibrium distance between the ith and jth atoms. The
subscripts 1, 2, and 3 correspond to the oxygen and the two hydrogen
atoms, respectively. The water uptake in the electrolyte was examined
[(Figure_9)TD$IG]
Figure 19 (a) Density of electrolyte (circle: experiment by Gierke et al. (1981), square:
calculation by Jinnouchi and Okazaki (2003), diamond: our model) and (b) water
uptake dependence of radius of gyration of Nafion1 molecules.
from 1 to 14.3 H2O/SO3. Note that the electrolyte in the state of water
uptake at 14.3 H2O/SO3 is almost identical to the immersed proton
form in this simulation.
One of the most important phenomena related to the electrolyte
structure is water uptake variation in the PEM. The electrolyte swells
upon water uptake, and the density of the electrolyte decreases in a
humidified atmosphere. Swelling deformation has been considered as
forming transport paths for the cluster of water molecules and protons.
We first calculated the density of the PEM system for various amounts
of water uptake. The squares and diamonds in Figure 19a indicate the
experimental and simulated density for the dry and hydrated proton
form of electrolyte (Gierke et al., 1981; Jinnouchi and Okazaki, 2003). As
water content increases, the PEM density decreases in the simulations
as well as experiments, and predicted values are in good agreement
with experimental values, indicating that swelling phenomena of the
electrolyte can be captured through this simulation. This swelling phe-
nomena of Nafion1 molecules is also illustrated via Figure 19b, which
shows that the radius of gyration among the simulation species
increases when the number of water molecules in the system increases.
This indicates that as the area occupied by water molecules around the
end of the side-chain increases, the Nafion1 molecules tend to stretch-
out geometrically. Haubold et al. (2001) found that side-chain unfolding
of the polymer can be observed when the water content increases. In this
simulation, we observed that increase in clustered regions of water not
only affects the side-chain conformations but also the conformation of
the backbone chain, making the polymer backbone chain stretched.
[(Figure_0)TD$IG]
Figure 20 Radial distribution function: (a) H+–H2O, (b) H+–S, (c) H+–CF2.
The evidence of water–water cluster generation in the wet electro-

lyte has been provided through several experimental techniques such
as small-angle X-ray scattering, transmission electron microscopy, and
Fourier transform infrared spectroscopy (FTIR) (Falk, 1980; Porat et al.,
1995; Roche et al., 1981). It has been reported that polar particles (water
and protons) cohere in the electrolyte. We also confirmed such phe-
nomena in this simulation via radial distribution functions (RDFs)
among the species (Figure 20). The sharp peak in RDFs between polar
species indicates that almost all the protons and the water molecules
always exist within the clustered region. However, we did not observe
significant peaks between the polar and nonpolar particles. These
results indicate that polar particles cohered in our simulations as well.
Figure 21 illustrates the two-dimensional snapshot of water density
distribution within the simulation cell for various amounts of water
content (H2O/SO3). It is clearly observed that the high water density
region increases as the water content increases. This high water density
region corresponds to the clusters observed in previous experimental
studies. The increase of the high water density region implies an
increase of the number of the transport paths for polar particles. The
alignment of these clusters is disorderly, and highly tortuous transport
pathways are constructed in our electrolyte model.
[(Figure_1)TD$IG]
Figure 21 Water density distributions in a cross section of the system.

[(Figure_2)TD$IG]
Figure 22 (a) Water content dependence of mean square displacement (MSD) of

protons and (b) MSD of protons and water molecules (H2O/SO3: 1).
Researchers have shown that protons and water molecules move

together through the clusters constructed in electrolytes. Although sev-
eral theoretical methods using macroscopic modeling have been pro-
posed based on the information of the electrolyte structure, there is no
model that estimates transport properties from the subcontinuum
modeling viewpoint. From Figure 22a, we observe that the MSD of
protons increases as the water content increases. In other words, the
diffusivity of the protons (proportional to the gradient of MSD)
increases as the water content increases. Thus, the diffusion of protons
in the PEM depends on the water content. This phenomenon can be
explained in terms of the electrolyte structure. As discussed before, the
water molecules and ions cohere with each other and construct hydro-
philic clusters in a humidified electrolyte. These clusters join with each
other and form transport pathways for the protons and water mole-
cules. Evidence that the protons and water molecules move through this
cluster region has been obtained from the RDF. Hence, we can conclude
that the cluster region is the transport path for protons and water
molecules. The shape parameters of the clustered region, such as the
cluster size and clusters connectivity, can strongly affect the transport
properties in the electrolyte. The low diffusivity in the electrolyte is due
to the tortuous shape of the cluster region. The ions and water molecules
in an electrolyte solution can move freely in all directions, but particles
in the PEM cannot, due to the tortuous hydrophobic wall. Figure 22b
illustrates the comparison between diffusivity of water molecules and
protons. We observe that the diffusion coefficient of the protons is
smaller than that of the water molecule, and the activation energy of
protons is higher than that of water molecule. This indicates that the
resistance to transport of protons is higher than the water molecules.
5.2 Multiphenomena in gas diffusion layer

An entire PEFC model has been successfully investigated via contin-
uum mechanics (Mukherjee and Wang, 2006, 2007; Rama et al., 2010;
Wang et al., 2006b, 2006c, 2007), which does not contain detailed molec-
ular structural information or simulate complex multiphase, multiphy-
sical phenomena. In addition, conventional continuum mechanics has
limitations to simulate the complex two-phase gas–liquid flow phe-
nomena within the porous GDL. To overcome this issue, LBMs can be
introduced. This method is specifically useful for simulating multi-
phase flows as well as hybridizing molecular with continuum level
theories (Hao and Cheng, 2010; Sukop et al., 2008; Yu and Fan, 2010).
Since the LBM covers from meso-scale to continuum levels, it is straight-
forward to combine GDL (meso-scale/continuum level) and GC (con-
tinuum level).
For the porous media flow simulation, LBM can be applied on the
complex geometry of porous media due to its advantages on handling
such geometry. However, in spite of this convenience, the calculation
cost will be drastically increased as the system size increases up to the
device level approximately in millimeter scale and above. To develop
larger scale buffer simulation, representative elementary volume (REV)
method was invented, where single lattice represents the volume of
porous media. The fuel flowing in the porous electrode can be described
by a continuity equation and the Brinkman–Forchheimer–extended
Darcy equations (generalized momentum equation) (Nithiarasu et al.,
1997):
r v¼0 ð25Þ

@v v 1
þ ðv rÞ ¼ rðfPÞ þ vr2 v þ F ð26Þ
@t f r
where f is the porosity, P is the cell total pressure, and v is the kinematic
viscosity. F represents total body force due to the presence of the porous
medium and other external force fields, expressed by
fv fFf
F¼ v pffiffiffiffi jvjv þ fG ð27Þ
K K
where G is the body force induced by an external force, and Ff and K

are, respectively, the geometric function and permeability, which can be
estimated from Ergun’s experimental results and expressed by (Ergun,
1952; Vafai, 1984)
1:75 f3 d2p
Ff ¼ pffiffiffiffiffiffiffiffiffiffiffiffi and K¼ ð28Þ
150f3 150ð1 fÞ2
where dp is the effective average diameter of the solid in the porous

electrode.
Generally, the BTE with single relaxation time approximation can be
written as Equation (15). In order to simulate flow in the porous media,
we consider the drag effect of the medium and present the LBM equa-
tion by the following form of the statistical average:
1 eq
f i ðx þ ci t; t þ tÞ ¼ f i ðx; tÞ ½f i ðx; tÞ f i ðx; tÞ þ Fi t ð29Þ
t
eq
where f i ðx; tÞ and f i ðx; tÞ are volume-averaged distribution function
and equilibrium distribution function at REV scale, respectively (from
now on, the overbars will be omitted for the sake of convenience). Fi is
the force term for ith particle of fluid. According to Cancelliere et al.
(1990), Fi is chosen as

1 ei F ðei vÞðei FÞ v F
Fi ¼ vi r 1 þ ð30Þ
2t c2 fc4 fc2
The density and velocity of the fluid are defined by

X 1 X t
rðx; tÞ ¼ fi and vðx; tÞ ¼ ci f i þ F ð31Þ
i
rðx; tÞ i 2
The macroscopic equations for fluid flowing in porous media may

be recovered by Taylor expansion and Chapman–Enskog expansion,
which become
@r
þ r ðrvÞ ¼ 0 ð32Þ
@t

@ðrvÞ rvv
þr ¼ rP þ r ½rvðrv þ vrÞ þ F ð33Þ
@t f
We see that above equations recover Equations (25) and (26) for
r = constant. Note that as f = 1, Equation (26) or (33) is reduced to the
standard LBKE for the fluid flows in the absence of porous media.
Xu et al. (2006b) have successfully introduced Brinkman–
Forchheimer–extended Darcy equation in order to solve the perfor-
mance of molten carbonate fuel cell. As a verification of REV method,
Poiseuille flow profiles in the porous media modifying LBM with
[(Figure_3)TD$IG]
Figure 23 (a) Comparison of normalized velocity profile (Re = 0.1); solid line:
analytical solution, symbol: LBM; blue: Da = 104, red: Da = 103, f = 0.1 and (b)
comparison of REV (blue) and standard (red) methods: Re = 0.1, Da = 104, f = 0.1.
Brinkman–Forchheimer–extended Darcy equation were simulated. The

porosity is set to be 0.1, Reynolds number (Re) changes from 0.01 to 100,
and the Darcy number (Da) changes from 106 to 102. The lattice used is
an 80 80 square mesh, and the relaxation time is set to be 0.8. Periodic
boundary conditions are applied to the entrance and the exit. The
velocity field is initialized to be zero at each lattice node with a constant
density r = 1.0, and the distribution function is set to be its equilibrium
at t = 0. In Figure 23a, the numerical results of the REV LBM are com-
pared with the finite difference results, which were solved by Guo and
Zhao (2002). Excellent agreement is observed between REV LBM and
the finite-difference solutions, which confirms the validity of the REV
LBM for the continuum scale exhibiting that velocity profiles tend to be
the same when the finer porous media is used with same porosity. REV
LBM is also compared to the standard LBM with manipulated porous
media geometry having same parameter conditions as a verification in
lower scale level. Figure 23b shows that the standard LBM with porous
geometry shows the behavior similar to the results from REV LBM.
Since the flows in PEFC have multicomponent/multiphase character-
istics, which result in clogging effects, the details of porous media
geometry cannot be discarded for an accurate GDL model. Therefore,
multi-scale integration in GDL can be obtained via combining REV LBM
and LBM with the porous media geometry.
LBKE can be further modified to capture the multicomponent/mul-
tiphase phenomena in GDL as well as GC. This modification also allows
modeling the hydrophilic effect such as the bubble transport phenom-
enon. Lu and Wang visualized the bubble phenomenon in situ using the
hydrophilic carbon cloth since it has more regularly distributed pores

than the hydrophobic one. Therefore, the hydrophilic diffusion layer is
more preferred to remove bubbles. In order to simulate a multicompo-
nent, multiphase flow in the fuel cell, we need to solve the same number
of LBKEs as the number of components. The additional effects from the
multiphase flow, such as the surface tension, the fluid–solid force, and
the buoyancy force can be treated as external forces for the momentum
equation. Therefore, modified LBM equation can be expressed as
X s
rs ðx; tÞvs ðx; tÞ ¼ f i ðxi ; tÞci þ Fstotal ðx; tÞ ð34Þ
i
as the general momentum equation for the multicomponent, multi-

phase flow. In Equation (34), Fstotal ðx; tÞ represents the total external
force parameter (momentum per volume) of the component s (species)
contributed by the surface tension, the fluid–solid interaction force, and
the buoyancy force
Fstotal ðx; tÞ ¼ Fssurface tension ðx; tÞ þ Fsfluidsolid ðx; tÞ þ Fsbuoyancy ðx; tÞ: ð35Þ
In order to model the surface tension force in the multicomponent

fluid, an interaction potential f(x, x0 ) was defined. This potential can be
represented by
0 0
0 0
fðx; x Þ ¼ Gss cs ðxÞcs ðx Þ: ð36Þ
where cs (x) is a function0 of the mass density of the species s at the

position vector x; and Gss is set to be the interaction strength between
the 0species s and the other species s 0 . For simplicity, we assume that
Gss only accounts for the nearest-neighbor interactions. The rate of
change of the momentum per volume for the multicomponent fluid
then becomes
dFssurface tension s
X 0 X 0
ðx; tÞ ¼ C ðx; tÞ Gss C s ðx þ ei t; tÞei : ð37Þ
dt
s0 i
Hence, the force parameter contributed by the surface tension can be

approximated by the following equation:
s
X 0 X 0
Fssurface tension ðx; tÞ ¼ t s ½C ðx; tÞ Gss C s ðx þ ei t; tÞei ð38Þ
0 i
s
where t s is the collision time for the species s .

The interaction force at the fluid–solid interaction can be expressed
by
[(Figure_4)TD$IG]
Figure 24 Multi-scale/holistic interpretation of physical system.
[(Figure_5)TD$IG]
Figure 25 Multi-scale modeling of HDI: (a) ab initio (atomic), (b) MC/MD (molecular),
(c) coarse-graining procedure (molecular/meso-scale), (d) diffusion characteristics and
surface topography, and (e) LBM (meso-scale) and integration HDI (system design).
X
Fsfluidsolid ðx; tÞ ¼ rs Gsi sðx þ ei tÞei ð39Þ
i
where Gs is the fluid–solid interaction potential parameter; s is a

function of the position of the particle, s = 0 when the particle is in
the fluid, and s = 1 when the particle is at the fluid/solid interface.
The angle between the fluid and the wall, due to the hydrophilicity,
can be controlled by adjusting the fluid–solid interaction potential
parameter Gs . The angle increases with larger Gs . When the contact
angle increases, the wall is less hydrophilic. The momentum con-
tributed by the buoyancy force can be expressed by the following
equation:
X
Fsbuoyancy ðx; tÞ ¼ g rs ðx þ ei t; tÞei ð40Þ
i
where g is the gravitational constant.
5.3 Device-scale/process-scale level

A more physically accurate representation includes space–time
dependencies on physical parameters via conservation laws as well
as electrochemistry based on irreversible thermodynamics. As shown
in Figure 4, the model for PEFC system consists of an integrated
assembly of several interacting physical components, each compris-
ing of multidimensional, multiphysical transport and electrochemical
reaction processes. There are seven chambers in the model, the GCs,
the GDLs, and the CLs both on anode and the cathode sides, and a
central PEM region. This device-level model is based on multiphase
continuum mechanics coupled with species, and energy conservation
along with electrochemistry. The equations descriptive for various
subcomponents are integrated and incorporated into the optimization
framework. Although this approach primarily originated from con-
tinuum mechanics, one can obtain spatial dependencies and temporal
resolution of physical parameters through an optimization scheme,
which contains design criteria one level deeper. This multi-scale
approach was applied to obtain an optimal CL design with spatial
variation of the CL layer and platinum loading (Jain et al., 2008).
Further work demonstrated the sensitivity of this design for refining
these PEFC models (Jain et al., 2010).
At the process level, efficient flowsheet optimization strategies
based on lumped parameter models are now widely used in practice
(Biegler et al., 1997). At this scale, the PEFC is embedded within a power
plant flowsheet model, as shown in Figure 3. The process comprises
three subsystems (1) the fuel processing, (2) the PEFC stack, and (3) the
postcombustion (Xu et al., 2006a). The design questions at this level are
addressed through optimization of network connections of these sub-
systems, the heat exchanger network, as well as equipment design of
individual components (PEFC, reactors, compressors) within each sub-
system. As considered in Xu et al., the flowsheet is designed to maxi-
mize power for a given feed and an equilibrium model is used for the
PEFC. While this model neither handle the spatial characteristics of
device-level PEFC models nor the design of the membrane material,
the process-scale model can be replaced by a ROM to handle these
features. By applying the ROM-based optimization strategy developed
in Lang et al. (2009, 2011), information from device- and molecular-scale
models can be used to construct the ROM in order to include decisions
that relate to material and device-level performance.
6. TECHNOLOGICAL APPLICATIONS: HARD DISK DRIVE
As emphasized in the previous section, it is critical for the multi-scale

integration to develop mathematically simple (with few parameters)
yet physically realistic models with nanoscopic information. For the
description of the multi-scale framework of HDI, we provide a bot-
tom-up approach as given below to sketch multi-scale modeling as
described in Figures 24 and 25 (Jhon et al., 2011).
Perform atomistic simulations to obtain force field parameters to be
used in step 2 (Figures 24a and 25a). In this step, ab initio methods and/
or DFT are utilized to calculate the intramolecular force field para-
meters (stretching, bending, and torsional) from the Hessian matrix
(Smith et al., 2011). The potential energy among PFPE molecules and
PFPE–carbon surface interactions are also calculated as a function of the
endgroup structure (e.g., Zdol and Ztetraol) via parameter estimation
for the given potential energy functional form. This information is used
for MD simulation in step 2.
Employ classical MD to calculate mesoscopic properties of molecularly
thin lubricant films from the atomistic input obtained from step 1
(Figures 24b and 25b). During the past decade, atomistic/coarse-
grained MD models for PFPE systems based on Langevin equations
have been investigated (Chung et al., 2009; Guo et al., 2006; Ma et al.,
1999a, 1999c; Phillips and Jhon, 2002). Step 1 will provide additional
input to these models by delivering first-principle information as well
as numerical accuracy.
Further employment of the coarse-graining process to represent
PFPE molecules with few parameter models (having the least
parameters yet preserving essence of physics) such as size, shape, and

orientation of endgroups (simplest example is SRS (Figures 24c and 25c)
where the shape is spherical and orientation is spin-like quantized
form).
Develop an LBM scheme with generalized SRS model to accurately
describe the dynamics of PFPE systems. The model is based on the
mathematically simple yet physically realistic LBM models capturing
the bottom level (atomistic) information. This novel formulation is based
on our system for electron–phonon coupling with two states (Ghai et al.,
2005), which is analogous to spin system description for endgroups.
Couple LBM to be developed in step 4 with LBM schemes descrip-
tive for air bearing and thermal phenomena (Figures 24e and 25e).
From the mesoscopic model described in steps 2 and/or 3, calculate
physical properties including spreading, surface energy, diffusion pro-
cesses, and compare the simulation results with experimental data. This
can be done for pure and nanoblended PFPE systems (Figures 24d and
25d).
Design the advanced HDD from our multi-scale simulation tools
described in steps 4 and 5. Integration of HDI simulation stated in
steps 1–6, can be easily achieved via a slight extension of our current
understanding, although at the current stage we are not employing
systematic parameter estimation techniques yet by introducing
sophisticated optimization tools. Furthermore, the techniques like
reduced-order methods may be useful in reducing the degrees of
freedom systematically. This methodology will be useful in our future
HDD development.
6.1 The coarse-grained, bead-spring model

Molecular simulations via the coarse-grained, bead-spring model were
carried out to examine the nanostructure of PFPE lubricant films,
including the anisotropic radius of gyration and endbead density pro-
file (Izumisawa and Jhon, 2002a, 2002b; Jhon et al., 2003). Guo et al. (2004,
2005) pioneered MD simulation to investigate both static and dynamic
properties of PFPE films to explain the underlying physics in the exper-
imental findings of PFPE films as well as develop a powerful numerical
tool to understand the properties of PFPE films at molecular level. The
atomic or united atom models are mainly adopted for the simulation of
simple particles or short chains, while the coarse-grained, bead-spring
model is suitable to simulate the polymeric system with a larger length
scale. With the advantage of reduced computational cost, the coarse-
grained, bead-spring model has been widely adopted to simulate
the behaviors of polymeric systems (Milchev and Binder, 1996;
Milchev et al., 1993; Sheng et al., 1994). Since PFPE and its derivatives
[(Figure_6)TD$IG]
Figure 26 (a) A molecular model of an oligomeric PFPE Zdol molecule with p = q = 10

and Mn = 1974 g/mol and (b) simplification of the rigid units of an oligomeric PFPE
molecule and the flexible bonds with freely jointed beads and springs.
have relatively high molecular weights, Jhon et al. adopted the coarse-
grained, bead-spring model thereby simplifying the detailed atomistic
information while preserving the essence of the internal molecular
structure to examine static and dynamic properties of PFPEs (Guo
et al., 2003; Izumisawa and Jhon, 2002a, 2002b). In the coarse-grained,
bead-spring model, PFPEs are characterized by a sequence of freely
joined rigid beads as shown in Figure 26.
In the HDI system, the root mean square (RMS) roughness of the
carbon-overcoat surface, which corresponds to the solid surface, has
been reduced to 2.3 (Fung et al., 2000). Therefore, it is reasonable to
assume a perfectly flat and structureless wall on the bottom of the
simulation cell for the system with coarse-grained, bead-spring model.
This assumption has been justified via the spreading profiles of PFPE
Zdol on the silica surface (molecularly smooth) (O’Connor et al., 1996)
similar to those on the carbon surface (molecularly rough) (Ma et al.,
1999b, 1999c, 1999d). The van der Waals interaction between bead and
the surface UVDW(z) is then calculated by
9
1 s 3 s 3
U VDW ðzÞ ¼ ew : ð41Þ
2 z 2 z
Here, z is the distance from the surface and ew is the potential depth
of the bead and surface interaction. Due to the repulsive contribution in
Equation (41), beads are prevented from moving very close toward the
solid surface. Figure 27 illustrates a schematic of the coarse-grained,
bead-spring PFPEs with the flat surface assumption. The functional
[(Figure_7)TD$IG]
Figure 27 A schematic of coarse-grained, bead-spring model with potential energies

(PFPE).
beads in PFPE interact much differently from LJ or dispersive type

interactions; however the mechanism has not yet been clarified,
which could be hydrogen bonding, hydrogen transfer, or esterifica-
tion. To maintain the generality and demonstrate potential energy
characteristics of a short-range interaction without losing the
essence of the chain-end related problem, such as that found in
functional PFPE films, short-range exponential (EXP) decay func-
tions in addition to dispersive interaction (i.e., ULJ and UVDW) are
used between functional beads, UEXP1(ree), as well as between the
functional bead and the surface, UEXP2(z) as illustrated by Equations
(42) and (43). Specifically, for the interaction between endbeads,
Ueb–eb = ULJ + UEXP1; and for the interaction between endbeads and
surface, Ueb-surface = UVDW + UEXP2. Due to the strong short-range
interaction in the decay functions, functional endgroups prefer to
couple with the solid surface rather than backbone chains, as long as
they are close enough to the bottom surface. Here,
r r
U EXP1 ðree Þ ¼ eb exp
p ee c
; ð42Þ
d
and
zz
U EXP2 ðzÞ ¼ ew exp
p c
: ð43Þ
d
where, ree denotes the distance between endbeads and d is the
characteristic decay length for the short-range interaction and eb
p
and ew are the potential depths at ree = rc and z = zc. A larger value
p
of eb indicates a stronger endgroup functionality.

p
Endbead density profiles for functional and nonfunctional PFPEs

were examined by Guo et al. (2003). Figure 28a shows the endbead
density profiles of single component PFPEs as a function of the bead
and surface interaction strength ew in Equation (41) and endgroup
functionality eb and ew in Equations (42) and (43). For nonfunctional
p p
[(Figure_8)TD$IG]
Figure 28 (a) The endbead density profiles of PFPE nanofilms as a function of ew,
epb ; and epw with Np = 10 and T = 1.0e/kB by Guo et al. and (b) a schematic of the
‘‘layering’’ structure for functional PFPE nanofilms.
PFPEs eb ¼ ew ¼ 0 , the endbeads are uniformly distributed through

p p
the entire film thickness (z). Slightly intensive adsorption on the surface
was also observed for a stronger surface attraction (ew = 4e), which
shows the surface interaction strength is negligible for the film confor-
mation of nonfunctional PFPEs as well as backbone of functional PFPEs.
For functional PFPE eb ¼ ew ¼ 2e , the endbead density is maximized
p p
near the bottom surface with s = 1. Oscillation of the endbead density

profile occurs for functional PFPEs, originating from the endbead–end-
bead couplings. We also notice that the distance between two neigh-
boring peaks in Figure 28a is about 5.5s , which corresponds to the
thickness of two layers (Guo et al., 2003). Therefore, this result verifies
that the coarse-grained, bead-spring model qualitatively describes the
endgroup coupling between functional PFPE molecules as well as the
layering structure in the functional PFPE film. This is ideally described
in Figure 28b, where the functional PFPEs tend to anchor with end-
groups onto the surface in the first layer and form the subsequent layers
with coupled endgroups.
[(Figure_9)TD$IG]
Figure 29 Spreading profile of SRS models with (a) nonfunctional and (b) functional
PFPEs by Ma et al.
6.2 Simple reactive sphere model

SRS with spins representing the endgroup orientation was first pro-
posed by Ma et al. (1999a) to simulate the spreading profiles of single
component PFPE films over the carbon-overcoat via the MC method
(Figure 29). This work was later extended to simulate the spreading
profiles of single component PFPE films over hydrogenated and nitro-
genated carbon-overcoats (Phillips et al., 2001; Vinay et al., 2000) as well
as their surface characteristics (Phillips and Jhon, 2002). However, the
SRS model does not capture the detailed structure of PFPE molecules.
Therefore, in order to utilize SRS to understand the geometry of sub-
monolayer lubricant films, as well as investigate the molecular confor-
mation and the resultant static and dynamic properties of PFPE films, it
is necessary to incorporate intramolecular degrees of freedom from
high-resolution scale models such as coarse-grained, bead-spring or
atomistic models.
Due to the van der Waals interaction between backbone bead and
carbon-overcoated surface, PFPE has an oblate rather than spherical
conformation. So the current SRS model should be modified by allow-
ing position dependent deformations. To analyze molecular structures
in the nanofilm, the molecular conformation of coarse-grained, bead-
spring PFPEs on the surface is defined with the radius of gyration (Rg),
including the parallel (Rjj) and perpendicular (R?) components
(Guo et al., 2003), corresponding to the radius of gyration in the direc-
tion parallel and perpendicular to the bottom surface (Figure 30):
1 X
Np
R2jj ¼ ½ðxi xg Þ2 þ ðyi yg Þ2 ;
2N p i¼1
ð44Þ
1 X
Np
R2 ¼ ðzi zg Þ2 with R2g ¼ R2 þ 2R2jj
N p i¼1
Here, Np is the total number of beads in each PFPE molecule, (xg, yg, zg)
are the coordinates of the PFPE molecular center of mass, and (xi, yi, zi)
denote the bead coordinates. The perpendicular size of PFPE
[(Figure_0)TD$IG]
Figure 30 A schematic of the parallel and perpendicular radius of gyration of PFPE

molecule on the surface.
especially controls the thickness of monolayer film. The radius of

gyration can be considered as a SRS configuration parameter, which
identifies the form of SRS model on the surface (e.g., oblate, spherical,
or prolate conformation).
6.3 Meso-scale/continuum level

The air bearing of the HDI operates under rarefied gas flow, which is the
high Knudsen number (Kn) flow regime. The conventional Navier–
Stokes equations are not eligible to describe the flow inside the air
bearing because the continuum hypothesis is no longer valid in the high
Kn regime. Therefore, the modified Reynolds equation (MRE) has been
constructed to solve high Kn flow of the air bearing (Fukui and Kaneko,
1987; Kang et al., 1999). Accurate velocity slip modeling on the wall is
crucial for the high Kn flows and several slip models have been devel-
oped that incorporate the molecular rarefaction effect to describe the
slip flow via two parameters, Kn and surface accommodation coeffi-
cient. Although the MRE is widely used for the slider design, it is not
well suited to HDI integrated modeling, because it does not have flex-
ibility and capability for integration of the multi-scale system including
the air-bearing, lubricant film, and nanoscale heat transfer in media.
Recently, Myong et al. (2005) developed the boundary model based on
the theory of adsorption phenomena pioneered by Langmuir.
According to Langmuir’s theory, gas molecules do not reflect directly,
but rather reside on the surface for a brief period of time due to the
intermolecular forces between the gas molecules and the surface atoms.
After some lag in time these molecules may reflect from the surface.
This time lag causes macroscopic velocity slip. From this physical con-
text, the fraction of surface covered at equilibrium for monatomic and
diatomic gases is written respectively as
bp
a¼ ð45Þ
1 þ bp
pffiffiffiffiffiffi
bp
a¼ pffiffiffiffiffiffi : ð46Þ
1 þ bp
It was shown that the parameter a is function of pressure p. b is the

reaction constant for surface–gas molecules interaction and is defined
as

Al =Kn De
b¼ l exp ; ð47Þ
KB T w KB T w
where A, KB, De, and ll are the mean area of a site, the Boltzmann
constant, a potential energy parameter, and the molecular mean free
path, respectively. The potential energy of heat adsorption De varies
with the type of gas and the nature of the wall material. a is a function of
the wall temperature Tw, and plays a role similar to the slip coefficient of
the Maxwell model slip model. The slip velocity, us of the Langmuir slip
model can be calculated by
us ¼ auw þ ð1 aÞug ; ð48Þ
In LBM, the slip velocity on the wall can be defined by applying

streaming and bounce-back process. Typical bounce-back and slip con-
dition on the wall in LBM can be expressed as
0 1 0 1 0 1
f7 f5 r 0 s
@ f A ¼ k @ f A; k ¼ @ 0 r þ s 0 A; ð49Þ
4 2
f8 f6 s 0 r
where two parameters, r and s represent the fraction of the distribution

function to be bounced-back and slipped forward, respectively
(Sbragaglia and Succi, 2005), and they should be chosen such that r +
s = 1. The slip velocity on the wall can be defined by summation of the
distribution in consideration with the direction. The difference of two
functions along the wall f 1 f 3 within the summation can be denoted
as the moving wall momentum ðruw Þ. Thus
X
r us ¼ s ci f þ rruwi : ð50Þ
Equation (50) physically has the same meaning as the Langmuir slip
model when r is chosen as a. For the 3D case, the 5 5 matrix k can be
easily built up by using r and s in the same manner as in 2D.
Kim et al. (2007) examined the normalized velocity profiles at vari-
ous values of Kn. In Figure 31, the computation result using the
Langmuir slip model exactly matches those of the Maxwell and
Beskok’s slip model (Beskok, 2001) at Kn < 0.1. For the cases of
Kn = 0.1 and 1 velocity profiles are also presented in Figure 31 and the
results are compared with those obtained from the first-order slip
model flow, Hsia and Domoto (1983), and the Maxwell models. The
result using the Langmuir slip model exactly matched the one from the
first-order slip flow but showed a slight deviation with the result using
the Maxwell slip model, which may be caused by the fact that the
Maxwell slip model can be considered to be a subset of the Langmuir
slip model in the microchannel flow analysis at Kn < 0.1 only. 2D flow
induced by a moving plate (Couette flow) with Kn = 0.1 and 1 was also
[(Figure_1)TD$IG]
Figure 31 Normalized velocity distribution of Poiseuille flow at various Kn.
[(Figure_2)TD$IG]
Figure 32 Normalized velocity distribution of Couette flow at various Kn.
simulated to verify the applicability of the slip model to air bearing

simulation. The velocity comparisons with the first-order slip flow are
shown in Figure 32. In this figure, we can observe that the present model
gives satisfactory agreement with the prediction of the first-order slip
velocity. Cavity flow at various values of Kn was solved since the
complex geometry of cavity underneath the slider plays a critical role
in determining the stability of the slider under high speed operation of
the HDI system. Figure 33 shows streamlines of cavity flow. The center
[(Figure_3)TD$IG]
Figure 33 Streamlines of cavity flow at various Knudsen numbers.
of primary vortex moves down as Kn increases and the flow pattern was
well resolved in Figure 33.
In fact, for the transition flow regime (0.1 < Kn < 10) and free molec-
ular regime (Kn > 10), it is extremely difficult to accurately predict the
flow physics using the LBM because LBM’s finite discrete velocities
cannot simulate the real molecule motion at high Kn. As a remedy to this
problem, virtual wall collision (VWC) method has been developed
(Toschi and Succi, 2005) and it was reported that LBM with VWC can
solve the flow regime (Kn < 30). This is capable of simulating the nano-
scale air bearing flow using LBM. By incorporating the SRS model
introduced in Section 4.2, the entire HDI can be described by LBM
providing the framework for a tribological study of the disk surface,
lubricant, and read/write head. Optimization strategies can be devised
by macrosopic observables from the integrated HDI model with respect
to the fundamental molecular architectures, and will allow for efficient
evaluation of the decision variables enabling inverse optimization for-
mulation (Figure 15).
7. SUMMARY AND CONCLUSIONS

A holistic strategy in hierarchical modeling, which enables the commu-
nication between physical phenomena in different length and time
scales and provides understanding of the systematic properties using
nanoscale parameters has been presented in this paper via two bench-
mark systems, that is, HDD and PEFC. By illustrating representative
modeling methods on each level of scale, physical phenomena in each
subsystem have been examined and the possibilities of integrating

different scale methods were provided. As detailed in the benchmark
systems separately, bridging methodologies enabling a link to the sub-
systems organically were discussed as a key to build the successful
multi-scale model. Bridging the molecular to the continuum scales
can be resolved via LBM in a broad range of scales from meso-scale to
continuum levels. Molecular-level models with quantum level para-
meters can be incorporated to LBM via coarse-graining procedure,
which simplify detailed molecular structures while transferring essen-
tial physics to upper scale level. The parameters in highly coarse-
grained SRS model include nanoscopic information stochastically
obtained from sequential coarse graining procedures including ab initio
and bead-spring model.
As reviewed in this paper, multi-scale modeling currently is an
active area with many multi-scale integration methods recently pub-
lished. Although these advances in modeling techniques and multi-
scale approaches for each application bring profound understanding
of complex systems and provide multidisciplinary impact in multiple
science and engineering fields, critical challenges are yet to be over-
come in obtaining feasible methodologies. Current state of the bridg-
ing methods for the specific systems must be improved to establish a
sound theoretical structure of hierarchical multi-scale integration for
feasible methods. In particular, the supplemental integration meth-
odologies between molecular and continuum scales are a major hur-
dle to be resolved first. In addition to the integration of time and
length scales differences, the multiphenomenological integration in
an identical space–time phase (e.g., heat and mass transfer and reac-
tion kinetics) should be incorporated. For instance, the thermal effects
on the nanoscale systems can be described more accurately by LBM
due to its treatment of broad length and time scales. Electrons as well
as phonons play a vital role in the energy transport in the nanoscale
systems, therefore a thermal behavior of both electrons and phonons
needs to be simultaneously considered to predict the transient sub-
continuum thermal transport. LBM can also successfully accommo-
date for such complicated systematic problems since the complexity
of the collision term in the BTE can be significantly reduced by using
the single relaxation time approximation (Ghai et al., 2005, 2006a,
2006b).
The demand for the research will cover the development of the novel
algorithms utilizing parallel computation methods. The development
of a hierarchical multi-scale paradigm will consolidate theoretical anal-
ysis and will lead to large-scale decision-making criteria of the process
level design based on the first-principle dynamics.
ACKNOWLEDGMENT
One of the authors (MSJ) was supported by Korea Science &
Engineering Foundation through the WCU Project.
REFERENCES
Allen, M. P. and Tildesley, D. J., ‘‘Computer Simulation of Liquids’’. Oxford, Clarendon
(1996).
Baeurle, S. A., J. Math. Chem. 46, 363 (2009).
Baxter, S. F., Battaglia, V. S. and White, R. E., J. Electrochem. Soc. 146, 437 (1999).
Bent, J., Hutchings, L. R., Richards, R. W., Gough, T., Sparez, R., Coates, P. D., Grillo, I.,
Harlen, O. G., Read, D. J., Graham, R. S., Likhtman, A. E., Groves, D. J., Nicholson,
T. M. and McLeish, T. C. B., Science 301, 1691 (2003).
Beskok, A., Num. Heat Trans.: Fundam. 40, 451 (2001).
Biegler, L. T., Grossman, I. E. and Westerberg, A. W., ‘‘Systematic Methods of Chemical
Process Design’’. Prentice-Hall, Upper Saddle River, NJ (1997).
Broughton, J. Q., Abraham, F. F., Bernstein, N. and Kaxiras, E., Phys. Rev. B 60, 2391 (1999).
Brown, A. E. X., Litvinov, R. I., Discher, D. E., Purohit, P. K. and Weisel, J. W., Science 325,
741 (2009).
Cancelliere, A., Chang, C., Foti, E., Rothman, D. H. and Succi, S., Phys. Fluids A 2, 2085
(1990).
Chen, S. and Doolen, G. D., Annu. Rev. Fluid Mech. 30, 329 (1998).
Choe, Y. K., Tsuchida, E., Ikeshoji, T., Ohira, A. and Kidena, K. J., Phys. Chem. B 114, 2411
(2010).
Chung, P. S., Park, H. and Jhon, M. S., IEEE Trans. Magn. 45, 3644 (2009).
Coffey, T. and Krim, J., Phys. Rev. Lett. 96, 186104 (2006).
Csanyi, G., Albaret, T., Payne, M. C. and De Vita, A., Phys. Rev. Lett. 93, 75503 (2004).
Dalton, L. R., Thin Solid Films 518, 428 (2009).
Dalton, L. R., Sullivan, P. A., Bale, D. H. and Olbricht, B. C., Solid State Electron. 51, 1263 (2007).
Delgado-Buscalioni, R. and Coveney, P. V., Phys. Rev. E 67, 46704 (2003).
Delle Site, L., Abrams, C. F., Alavi, A. and Kremer, K., Phys. Rev. Lett. 89, 56103 (2002).
Delle Site, L. and Kremer, K., J. Q. Chem. Int. 101, 733 (2005).
Delle Site, L., Leon, S. and Kremer, K. J., Am. Chem. Soc. 126, 2944 (2004).
Dienwiebel, M., Verhoeven, G. S., Pradeep, N., Frenken, J. W. M., Heimberg, J. A. and
Zandbergen, H. W., Phys. Rev. Lett. 92(12), 126101 (2004).
Dohle, H., Divisek, J. and Jung, R. J., Power Sources 86, 469 (2000).
Doi, M., Macromol. Symp. 195, 101 (2003).
Dudukovic, M. P., Science 325, 698 (2009).
Engler, A. J., Humbert, P. O., Wehrle-Haller, B. and Weaver, V. M., Science 324, 208 (2009).
Ergun, S., Chem. Eng. Prog. 48, 89 (1952).
Evans, D. J. and Morriss, G. P., Phys. Rev. A 30, 1528 (1984).
Evans, D. J. and Morriss, G. P., ‘‘Statistical Mechanics of Nonequilibrium Liquids’’.
Academic Press, London (1990).
Falk, M., Can. J. Chem. 58, 1495 (1980).
Faller, R., Polymer 45, 3869 (2004).
Flekkoy, E. G., Wagner, G. and Feder, J., Europhys. Lett. 52, 271 (2000).
Franco, A. A. and Gerard, M. J., Electrochem. Soc. 154, B712 (2007).
Franco, A. A. and Gerard, M. J., Electrochem. Soc. 155, B367 (2008).
Franco, A. A., Passot, S., Fugier, P., Anglade, C., Billy, E., Guetaz, L., Guillet, N., Vito, E. D.
and Mailley, S. J., Electrochem. Soc. 156, B410 (2009).
Frenkel, D. and Smit, B., ‘‘Understanding Molecular Simulation’’. Academic Press, New
York (2000).
Frenklach, M., Packard, A., and Seiler, P. (2002). American Control Conference,
Anchorage, May 8–10, pp. 4135–4140.
Frenklach, M., Packard, A., Seiler, P. and Feeley, R., Int. J. Chem. Kinet. 36, 57 (2004).
Fukui, S. and Kaneko, R., JSME Int. J. 30, 1660 (1987).
Fung, M. K., Lai, K. H., Lai, H. L., Chan, C. Y., Wong, N. B., Bello, I., Lee, C. S. and Lee, S. T.,
Diamond Relat. Mater. 9, 815 (2000).
Gerde, E. G. and Marder, M., Nature 413, 286 (2001).
Ghai, S. S., Chung, P. S., Kim, W. T., Amon, C. H. and Jhon, M. S., IEEE Trans. Magn. 42,
2474 (2006).
Ghai, S. S., Kim, W. T., Escobar, R. A., Amon, C. H. and Jhon, M. S., J. Appl. Phys. 97, 10P703
(2005).
Ghai, S. S., Kim, W. T., Amon, C. H. and Jhon, M. S., J. Appl. Phys. 99, 08F906 (2006).
Gierke, T. D., Munn, G. E. and Wilson, F. C., J. Polym. Sci. 19(Part A-2), 1687 (1981).
Glotzer, S. C. and Paul, W., Annu. Rev. Mater. Res. 32, 401 (2002).
Goddard III, W., Merinov, B., Van Duin, A., Jacon, T., Blanco, M., Molinero, V., Jang, S. S.
and Jang, Y. H., Mol. Simulat. 32, 251 (2006).
Gorban, A. N., Kazantzis, N., Kevrekidis, I. G., Ottinger, H. C. and Theodoropoulos, K.,
‘‘Model Reduction and Coarse-Graining Approaches for Multiscale Phenomena’’.
Springer, New York (2006).
Grest, G. S. J., Chem. Phys. 105, 5532 (1996).
Grossmann, I. E. and Westerberg, A. W., AIChE J. 46, 1700 (2000).
Gubbins, K. E. and Moore, J. D., Ind. Eng. Chem. Res. 49, 3026 (2010).
Guerra, R., Tartaglino, U., Vanossi, A. and Tosatti, E., Nat. Mater. 9, 634 (2010).
Guo, Q., Chen, H. G., Smith, B. C. and Jhon, M. S., J. Appl. Phys. 97, 10P301 (2005).
Guo, Q., Chung, P. S., Chen, H. G. and Jhon, M. S., J. Appl. Phys. 99, 08N105 (2006).
Guo, Q., Izumisawa, S., Jhon, M. S. and Hsia, Y. T., IEEE Trans. Magn. 40, 3177 (2004).
Guo, Q., Izumisawa, S., Phillips, D. M. and Jhon, M. S., J. Appl. Phys. 93, 8707 (2003).
Guo, Z. L. and Zhao, T. S., Phys. Rev. E 66, 036304 (2002).
Habenicht, B. F., Paddison, S. J. and Tuckerman, M. E., J. Mater. Chem. 20, 6342 (2010).
Hadjiconstantinou, N. G., Phys. Rev. E 59, 2475 (1999).
Hao, L. and Cheng, P J., Power Sources 195, 3870 (2010).
Haubold, H. G., Vad, T., Jungbluth, H. and Hiller, P., Electrochim. Acta 46, 1559 (2001).
Hijón, C., Español, P., Vanden-Eijnden, E. and Delgado-Buscalioni, R., Faraday Discuss.
144, 285 (2010).
Hohenberg, P. and Kohn, W., Phys. Rev. 136, B864 (1964).
Hoover, W. G., Evans, D. J., Hickman, R. B., Ladd, A. J. C., Ashurst, W. T. and Moran, B.,
Phys. Rev. A 22, 1690 (1980).
Hsia, Y. T. and Domoto, G. A., ASME. J. Lub. Tech. 105, 120 (1983).
Ilg, P., Öttinger, H. C. and Kr€oger, M., Phys. Rev. E 79, 011802 (2009).
Iuchi, S., Izvekov, S. and Voth, G. A., J. Chem. Phys. 126, 124505 (2007).
Izumisawa, S. and Jhon, M. S., J. Chem. Phys. 117, 3972 (2002).
Izumisawa, S. and Jhon, M. S., Tribol. Lett. 12, 75 (2002).
Izvekov, S., Violi, A. and Voth, G. A., J. Phys. Chem. B 109, 17019 (2005).
Izvekov, S. and Voth, G. A., J. Phys. Chem. B 109, 2469 (2005).
Izvekov, S. and Voth, G. A., J. Phys. Chem. 123, 134105 (2005).
Izvekov, S. and Voth, G. A., J. Chem. Phys. 125, 151101 (2006).
Izvekov, S. and Voth, G. A., J. Chem. Theory Comput. 2, 637 (2006).
Jain, P., Biegler, L. T. and Jhon, M. S., J. Electrochem. Soc. 157, B1222 (2010).
Jain, P., Jhon, M. S. and Biegler, L. T., Electrochem. Solid. St. 11, B193 (2008).
Jensen, F., ‘‘Introduction to Computational Chemistry’’. Wiley, Chichester (1999).
Jhon, M. S. and Choi, H. J., J. Ind. Eng. Chem. 7, 263 (2001).
Jhon, M. S., Izumisawa, S., Guo, Q., Phillips, D. M. and Hsia, Y. T., IEEE Trans. Magn. 39,
754 (2003).
Jhon, M. S., Smith, R., Vemuri, S. H., Chung, P. S., Kim, D. and Biegler, L. T., IEEE Trans.
Magn. 47, 87 (2011).
Jinnouchi, R. and Okazaki, K. J., Electrochem Soc. 150, E66 (2003).
Johnson, E. R. and DiLabio, G. A., Phys. Lett. 419, 333 (2006).
Johnson, E. R., Mackie, L. D. and DiLabio, G. A., J. Phys. Org. Chem. 22, 1127 (2009).
Johnson, K. E., Mate, C. M., Merz, J. A., White, R. L. and Wu, A. W., IBM J. Res. Dev. 40(5),
511 (1996).
Juan, M. L., Plain, J., Bachelot, R., Royer, P., Gray, S. K. and Wiederrecht, G. P., ACS Nano 3,
1573 (2009).
Kang, S. -C., Crone, R. M. and Jhon, M. S., J. Appl. Phys. 85, 5594 (1999).
Karis, T. E., Kim, W. T. and Jhon, M. S., Tribol. Lett. 18, 27 (2005).
Karis, T. E., Marchon, B., Flores, V. and Scarpulla, M., Tribol. Lett. 11, 151 (2001).
Kasai, P. H., Tribol. Lett. 13, 155 (2002).
Kasai, P. H. and Spool, A. M., IEEE Trans. Magn. 37, 929 (2001).
Kasai, P. H., Wass, A. and Yen, B. K., J. Inf. Stor. Proc. Syst. 1, 245 (1999).
Kim, W. T., Ghai, S. S., Zhou, Y., Staroselsky, I., Chen, H. D. and Jhon, M. S., IEEE Trans.
Magn. 41(10), 3016 (2005).
Kim, W. T., Jhon, M. S., Zhou, Y., Staroselsky, I. and Chen, H. J., Appl. Phys. 97(10), 10P304
(2005).
Kim, H. M., Kim, D., Kim, W. T., Chung, P. S. and Jhon, M. S., IEEE Trans. Magn. 43, 2244
(2007).
Kim, T. -D., Lao, J., Cheng, Y. -J., Shi, Z., Hau, S., Jang, S. -H., Zhou, X. -H., Tian, Y.,
Polishak, B., Huang, S., Ma, H., Dalton, L. R. and Jen, A. K. -Y. J., Phys. Chem. C 112
(21), 8091 (2008).
Klaus, E. and Bhushan, B., Tribol. Mech. Magn. Storage Syst. SP-19, 7 (1985).
Kohn, W., Meir, Y. and Makarov, D. E., Phys. Rev. Lett. 80, 4153 (1998).
Komarov, P. V., Veselov, I. N., Chu, P. P., Khalatur, P. G. and Khokhlov, A. R., Chem. Phys.
Lett. 487, 291 (2010).
Krishnan, M., Verma, A. and Balasubramanian, S., Proc. Indian Acad. Sci. (Chem. Sci.) 113,
579 (2001).
Laio, A., VandeVondele, J. and R€ othlisberger, U. J., Chem. Phys. 116, 6941 (2002).
Lang, Y.-D., Biegler, L. T., and Zitney, S. E. (2011). Comput. Chem. Eng., doi:10.1016/
j.compchemeng.2011.01.018.
Lang, Y-D., Malacina, A., Biegler, L. T., Munteanu, S., Madsen, J. I. and Zitney, S. E., Energ.
Fuel. 23, 1695 (2009).
Leconte, N., Moser, J., Ordejon, P., Tao, H., Lherbier, A., Bachtold, A., Alsina, F.,
Sotomayor Torres, C. M., Charlier, J. -C. and Roche, S., ACS Nano 4, 4033 (2010).
Lee, H., Lee, N., Seo, Y., Eom, J. and Lee, S., Nanotechnology 20, 325701 (2009).
Lee, C., Li, Q., Kalb, W., Liu, X. -Z., Berger, H., Carpick, R. W. and Hone, J., Science 328, 76
(2010).
Lee, J. K., Smallen, M., Enguero, J., Lee, H. J. and Chao, A., IEEE Trans. Magn. 29, 276 (1993).
Lee, T. D. and Yang, C. N., Phys. Rev. 87, 410 (1952).
Lemke, J. U., French, W. W., and McHargue, W. B. (1994). U.S. Patent 5317463.
LeSar, R. J., Phys. Chem 88, 4272 (1984).
Li, J., Liao, D. and Yip, S., Phys. Rev. E 57, 7259 (1998).
Lucas, M., Zhang, X., Palaci, I., Klinke, C., Tosatti, E. and Riedo, E., Nat. Mater. 8, 876 (2009).
Lyubartsev, A. P. and Laaksonen, A., Phys. Rev. E 52, 3730 (1995).
Ma, X., Bauer, C. L., Jhon, M. S., Gui, J. and Marchon, B., Phys. Rev. E 60, 5795 (1999).
Ma, X., Gui, J., Grannen, K., Smoliar, L., Marchon, B., Jhon, M. S. and Bauer, C. L., Tribol.
Lett. 6, 9 (1999).
Ma, X., Gui, J., Smoliar, L., Grannen, K., Marchon, B., Bauer, C. L. and Jhon, M. S., Phys.
Rev. E 59(1), 722 (1999).
Ma, X., Gui, J., Smoliar, L., Grannen, K., Marchon, B., Jhon, M. S. and Bauer, C. L., J. Chem.
Phys. 110, 3129 (1999).
Maginn, E. J. and Elliot, J. R., Ind. Eng. Chem. Res. 49, 3059 (2010).
Marchon, B., Guo, X. C., Karis, T., Deng, H., Dai, Q., Burns, J. and Waltman, R. J., IEEE
Trans. Magn. 42(10), 2504 (2006).
Marquardt, W., Von Wedel, L. and Bayer, B., AIChE Sym. Ser. 96(323), 192 (2000).
McGreevy, R. L., Nucl. Instrum. Meth. A 1, 354 (1995).
McGreevy, R. L. and Pusztai, L., Mol. Simul. 1, 359 (1988).
Mei, R., Shyy, W., Yu, D. and Luo, L. J., Comp. Phys. 161, 680 (2000).
Meijer, E. J. and Sprik, M. J., Chem. Phys. 105, 8684 (1996).
Mench, M. M., Wang, C. Y. and Thynell, S. J., Electrochem. Soc. 151, A144 (2004).
Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., Teller, A. H. and Teller, E. J., Chem.
Phys. 21, 1087 (1953).
Milchev, A. and Binder, K., Macromolecules 29, 343 (1996).
Milchev, A., Paul, W. and Binder, K. J., Chem. Phys. 99, 4786 (1993).
Miura, K., Kamiya, S. and Sakai, N., Phys. Rev. Lett. 90(5), 055509-1 (2003).
Moseler, M., Gumbsch, P., Casiraghi, C., Ferrari, A. C. and Robertson, J., Science 309, 1545
(2005).
Mukherjee, P. P. and Wang, C. Y., J. Electrochem. Soc. 153, A840 (2006).
Mukherjee, P. P. and Wang, C. Y., J. Electrochem. Soc. 154, B1121 (2007).
Myong, R. S., Reese, J. M., Barber, R. W. and Emerson, D. R., Phys. Fluids 17, 087105 (2005).
Neri, M., Anselmi, C., Cascella, M., Maritan, A. and Carloni, P., Phys. Rev. Lett. 95, 218102
(2005).
Nithiarasu, P., Seetharamu, K. N. and Sundararajan, T., Int. J. Heat Mass Trans. 40, 3955
(1997).
Noid, W. G., Chu, J. W., Ayton, G. S., Krishna, V., Izvekov, S., Voth, G. A., Das, A. and
Andersen, H. C., J. Chem. Phys. 128, 244114 (2008).
Noid, W. G., Liu, P., Wang, Y., Chu, J. W., Ayton, G. S., Izvekov, S., Andersen, H. C. and
Voth, G. A., J. Chem. Phys. 128, 244115 (2008).
Oberkampf, W. L., Trucano, T. G., and Hirsch, C. (2002). Proc. Foundations 02, Johns
Hopkins University, Laurel, MD, October.
O’Connell, S. T. and Thompson, P. A., Phys. Rev. E 52, R5792 (1995).
O’Connor, T. M., Back, Y. R., Jhon, M. S., Min, B. G. and Karis, T. E., J. Appl. Phys. 79, 5788
(1996).
Olbricht, B. C., Sullivan, P. A., Wen, G. -A., Misty, A., Davies, J. A., Ewy, T. R., Eichinger, B.
E., Robinson, B. H., Reid, P. J. and Dalton, L. R., J. Phys. Chem. C 112(21), 7983 (2008).
Parr, R. G. and Yang, W., ‘‘Density-Functional Theory of Atoms and Molecules’’. Oxford
University, New York (1989).
Pearson, J. D., Gao, G., Zikry, M. A. and Harrison, J. A., Comput. Mater. Sci. 47, 1 (2009).
Pereverzev, Y. V., Gunnerson, K. N., Prezhdo, O. V., Sullivan, P. A., Liao, Y., Olbricht, B.
C., Akelaitis, A. J. P., Jen, A. K. -Y. and Dalton, L. R., J. Phys. Chem. C 112, 4355 (2008).
Phillips, D. M. and Jhon, M. S., J. Appl. Phys. 91, 7577 (2002).
Phillips, D. M., Khair, A. S. and Jhon, M. S., IEEE Trans. Magn. 37, 1866 (2001).
Pit, R., Marchon, B., Meeks, S. and Velidandla, V., Tribol. Lett. 10, 133 (2001).
Porat, Z., Fryer, J. R., Huxham, M. and Rubinstein, I. J., Phys. Chem. 99, 4667 (1995).
Rafii-Tabar, H., Hua, L. and Cross, M. J., Phys. Condens. Matter 10, 2375 (1998).
Rama, P., Liu, Y., Chen, R., Ostadi, H., Jiang, K., Gao, Y., Zhang, X., Fisher, R. and Jeschke,
M., Energy Fuel 24, 3130 (2010).
Reith, D., Putz, M. and M€ uller-Plathe, F., J. Comput. Chem. 24, 1624 (2003).
Ren, X. M., Springer, T. E. and Zawodzinski, T. A., J. Electrochem. Soc. 147, 466 (2000).
Roche, E. J., Pineri, M., Duplessix, R. and Levelut, A. M., J. Polym. Sci. Polym. Phys. Ed. 19, 1
(1981).
Rubla, T. D., Zblb, H. M., Khralshl, T. A., Wirth, B. D., Victoria, M. and Caturla, M. J.,
Nature 406, 871 (2000).
Sakane, Y., Wakabayashi, A., Li, L. and Kasai, P. H., IEEE Trans. Magn. 42(10), 2501 (2006).
Sbragaglia, M. and Succi, S., Phys. Fluids 17, 093602 (2005).
Schirmeisen, A., Nat. Mater. 9, 615 (2010).
Seminario, J. M., Int. J. Q. Chem. 60, 1271 (1996).
Sheng, Y. J., Panagiotopoulos, A. Z., Kumar, S. K. and Szleifer, I., Macromolecules 27, 400 (1994).
Shi, Q., Izvekov, S. and Voth, G. A., J. Phys. Chem. B 110, 15045 (2006).
Smirnova, J. A., Zhigilei, L. V. and Garrison, B. J., Comput. Phys. Commun. 118, 11 (1999).
Smith, R., Chung, P. S., Steckel, J. A., Jhon, M. S. and Biegler, L. T., J. Appl. Phys. 109, 07B728
(2011).
Soper, A. K., Chem. Phys. 202, 295 (1996).
Sukop, M. C., Huang, H., Lin, C. L., Deo, M. D., Oh, K. and Miller, J. D., Phys. Rev. E 77,
026710 (2008).
Sullivan, P. A., Rommel, H., Liao, Y., Olbricht, B. C., Akelaitis, A. J. P., Firestone, K. A.,
Kang, J. -W., Luo, J., Choi, D. H., Eichinger, B. E., Reid, P. J., Chen, A., Jen, A. K. Y.,
Robinson, B. H. and Dalton, L. R., J. Am. Chem. Soc. 129, 7523 (2007).
Tani, H. and Matsumoto, H., Tribol. Int. 36, 397 (2003).
Theodorou, D. N., Comput. Phys. Commun. 169, 82 (2005).
Theodorou, D. N., Ind. Eng. Chem. Res. 49, 3047 (2010).
Toschi, F. and Succi, S., Europhys. Lett. 69, 549 (2005).
Tyndall, G. W., Karis, T. E. and Jhon, M. S., Tribol. Trans. 42, 463 (1999).
Ulherr, A. and Theodorou, D. N., Curr. Opin. Solid State Mater. Sci. 3, 544 (1998).
Vafai, K., J. Fluid Mech. 147, 233 (1984).
Villa, E., Balaeff, A., Mahadevan, L. and Schulten, K., Multiscale Model. Simul. 2, 527 (2004).
Vinay, S. J., Phillips, D. M., Lee, Y. S., Schroeder, C. M., Ma, X. D., Kim, M. C. and Jhon, M.
S., J. Appl. Phys. 87, 6164 (2000).
Vlachos, D. G., Adv. Chem. Eng. 30, 1 (2005).
Waltman, R. J., Khurshudov, A. and Tyndall, G. W., Tribol. Lett. 12, 163 (2002).
Waltman, R. J., Pocker, D. and Tyndall, G. W., Tribol. Lett. 4, 267 (1998).
Waltman, R. J., Tyndall, G. W. and Pacansky, J., Langmuir 15, 6470 (1999).
Waltman, R. J., Tyndall, G. W., Pacansky, J. and Berry, R. J., Tribol. Lett. 7, 91 (1999).
Wang, Y. T., Izvekov, S., Yan, T. Y. and Voth, G. A., J. Phys. Chem. B 110, 3564 (2006).
Wang, G., Mukherjee, P. P. and Wang, C. Y., Electrochim. Acta 51, 3139 (2006).
White, R. T., Bhatia, S. S., Friedenberg, M. C., Meeks, S. W., and Mate, C. M. (1996).
Tribology of Contact/Near Contact Recording for Ultra High Density Magnetic
Storage, ASME/STLE Tribology Conference, San Francisco, CA.
Xu, C., Biegler, L. T. and Jhon, M. S., AIChE J. 52, 2496 (2006).
Xu, Y. S., Liu, Y., Zu, X. Z. and Huang, G. X., J. Electrochem. Soc. 153, A607 (2006).
Yu, Z. and Fan, L. S., Phys. Rev. E 82, 046708 (2010).
Zhou, J., Thorpe, I. F., Izvekov, S. and Voth, G. A., Biophys. J. 92, 4289 (2007).
Zhu, L., Zhang, J., Liew, T. and Ye, K. D., J. Vac. Sci. Technol. A 21, 1087 (2003).
CHAP TER 3
Industrial Applications of
Plant-Wide Equation-Oriented
Process Modeling—2010
Milo D. Meixell Jr., Boyd Gochenour and
Chau-Chyun Chen

2. Sequential Modular Solution Techniques Versus 123
Simultaneous Solution Techniques
3. Modeling objectives 124
4. Model variables 124
5. Simulation 125
6. Parameter Estimation and Reconciliation 126
7. Optimization 127
8. Summary of Simulate/Optimize/Parameter/Reconcile 128
Cases
9. Model Scope 130
10. Model Fidelity 130
11. Embedded Solution Strategies 131
12. Process Economics 132
13. Offline and Online Usage 132
14. Model Maintenance 133
15. Examples of Industrial Applications 134
16. CO2 Capture with Aqueous Alkanolamine Solution 136
17. Modeling CO2 and Absorbent Physical Properties and 137
Chemical Reactions
18. Modeling CO2 Capture Process 140
19. Modeling CO2 Capture Process and the Plant 143
Aspen Technology, Inc., Burlington, MA 01803, USA

119
120 Milo D. Meixell Jr. et al.
20. Modeling Site-Wide Chemical Complex 148

21. Summary 150
Acknowledgment 150
References 151
Abstract Mathematical models of industrial processes have long been

used to better design and operate facilities. Traditional applica-
tions of such models have been limited in scope and usage due to
the rapid rise in complexity, execution time, and difficulties
encountered as model size increases to encompass sufficient
fidelity and scope to address overall economic impact to busi-
ness units. Simultaneous solution techniques, referred to as equa-
tion-oriented (EO) modeling, as opposed to sequential modular
(SM) approaches, have addressed many of the issues that previ-
ously limited model fidelity and scope across multiple scales
from delivering actionable, value-adding results in a frequent,
day-to-day, or even hour-to-hour time frame. Models of alkano-
lamines-based carbon dioxide (CO2) capture facilities exemplify
the ability of EO modeling to include the best high fidelity, multi-
scale, mechanistic models along with sufficient scope to optimize
operations, allowing economic trade-offs among plant through-
put and solution regeneration costs, in the context of the larger
process that the CO2 capture system serves. These models span
scales from physical and chemical properties of CO2 and absor-
bent molecules, process equipment or units, process plants, to
site complexes. A steady-state flowsheet model of the CO2
absorption and solution regeneration system is illustrated and
discussed, both in a parameter estimation mode, elucidating sys-
tem performance from observed plant data, and in an optimiza-
tion mode, honoring operating constraints, reflecting control
system configuration, while maximizing operating profit. The
model is also able to help identify and quantify debottlenecking
alternatives. The topics of model robustness, accuracy, and exe-
cution speed are covered as well. This application illustrates that
integrated, high fidelity, multiscale models from molecular level
to site-wide complex can be deployed in nonideal online envir-
onments to deliver benefits and insight that cannot be elucidated
with simpler, less rigorous, more empirical models.
1. INTRODUCTION
Mechanistics-based simulation models of industrial refining and

chemical processes have long been used to better design and operate
facilities Evans, 2009. However, most traditional applications of such
Industrial Applications of Plant-Wide Equation-Oriented Process Modeling—2010 121
simulation models have been limited in scope and usage due to the
rapid increase in model complexity, execution time, and solution
difficulties encountered as model size increases to encompass suffi-
cient scope to address overall economic impact to business units.
Linear programming (LP) models are used effectively for large-scope
problems. However, LP models include limited fidelity and address
planning issues rather than improvement of operating conditions,
detailed design, or revamping. Simultaneous solution techniques,
referred to as equation-oriented (EO) modeling, as opposed to
sequential modular (SM) approaches, have addressed many of the
simulation issues that previously limited model scope and fidelity of
mechanistics-based simulation models Chen and Stadtherr, 1985;
Kisala et al., 1987; Alkaya et al., 2001. Furthermore, significant
advances have been made in understanding and modeling detailed
mechanistic phenomena that when widely and frequently applied to
monitor and improve operations can yield considerable additional
revenue from existing assets Cutler and Perry, 1983.
After decades of industrial practice and continued refinement,
delivery of model-based, actionable, value-adding operational or
design improvement results in an efficient manner offline, or on a
frequent, day-to-day, or even hour-to-hour time frame in an online
environment is now very achievable and in fact, widely practiced in
the industry. Applications range widely including petroleum refineries
Liporace et al., 2009; Camolesi et al., 2008; Mudt et al., 1995, petrochem-
ical plants Cutler and Perry, 1983; Rejowski et al., 2009; Paules and
Meixell, 1994; Fatora and Ayala, 1992; Fatora et al., 1992a; Fatora et al.,
1992; Houk et al., 1992; Kelly et al., 1991 and chemical plants Mercang€ oz
and Doyle, 2008; Lowery et al., 1993; Meixell and Tsang, 1988. This
article presents the state of the art in the industrial applications of
plant-wide EO process modeling. We discuss the importance of model-
ing objectives and the significance of specifying model variables. We
present steady-state simulation models of different mode: simulation
models, optimization models, models for parameter estimation, and
models for reconciliation. We highlight various aspects of modeling
in industrial practice: model scope, model fidelity, embedded solution
strategies, process economics, offline and online usage, and model
maintenance. Also discussed are the current industrial applications
including critical success factors. While a major driving force for
developing simultaneous solution approaches has been to meet the
needs for real-time optimization (RTO) applications, including exe-
cution speed, robustness, and ease of posing the problem to represent
the present process situation, we discuss the much more broad usage
and benefits of these methods. Lastly, we present the details of a

particularly relevant process, carbon dioxide (CO2) capture with alka-
nolamines, to illustrate that integrated, high fidelity, multiscale mod-
els from molecular level to site-wide chemical complex or enterprise
level can be deployed in online environments to deliver benefits and
insight.
Biegler Biegler, 2009 discussed methods for integrating ‘‘process
models over multiple scales, ranging from open-form, declarative mod-
els that arise in real-time process optimization to ‘black-box’ models
that characterize molecular simulations.’’ While this article does not
include molecular simulations or computational fluid dynamics
(CFD), it does deal with multiscale models that integrate layers of
heterogeneous models across significant differences of physical scale
of chemical supply chain National Research Council, 2003 from single
and multiphase systems, to process equipment and units, to plants,
complexes, and enterprises. Figure 1 shows the types of equations
solved in each scale. Such integrated, high fidelity, multiscale process
models are essential to gain better process understanding and to
achieve optimal design of new plants, revamping of older ones, and
operations of existing facilities.
[(Figure_1)TD$IG]
Figure 1 Integrated, high fidelity, multiscale process modeling of chemical supply

chain.
2. SEQUENTIAL MODULAR SOLUTION TECHNIQUES

VERSUS SIMULTANEOUS SOLUTION TECHNIQUES
Models become more complex as their fidelity and scope increase. As

models become more complex, the methods used to solve the model
equations become more important. SM solution techniques are ade-
quate for solving models of modest complexity, while models of signif-
icant complexity are solved efficiently using simultaneous solution
methods Chen and Stadtherr, 1985; Kisala et al., 1987, typically using
sparse matrix methods. These simultaneous solution methods pose the
equations of the model in ‘‘0 = f(x)’’ or ‘‘Residual(i) = f(x)’’ format where
the ‘‘0’’ is the desired value of the equations’ residuals at solution.
Models posed in this manner are referred to as ‘‘equation-oriented’’
or ‘‘EO’’ models. Figure 2 illustrates the relationship between model
complexity and modeling difficulty with SM solution method and EO
modeling method.
Care should be taken when embarking on a modeling effort to
choose the modeling environment and method, which assure that the
model is solved efficiently when its complete scope and level of fidelity
are reached.
[(Figure_2)TD$IG]
Figure 2 Sequential modular (SM) and equation-oriented (EO) modeling difficulty

versus complexity.
3. MODELING OBJECTIVES
It is important to understand the objective or objectives of a modeling

undertaking. Some models can be quite useful even if simple, while
others require high fidelity and rigorous mechanistic underpinnings to
meet their objectives. Some models are not required to be predictive
(e.g., models used only for monitoring performance). Others must be
predictive to deliver their intended benefits (e.g., models used for opti-
mization). A well-planned modeling undertaking can be designed to
meet the immediate requirements while making it easier to extend the
model features to meet future requirements.
Some models require dynamic relationships to compute conditions
changing in time, while others employ very detailed transport phenom-
ena, such as CFD models. Most, if not all, models have steady-state
foundations onto which accumulation terms are added to capture
dynamic effects, or onto which transport properties are imposed to
investigate detailed flow, composition, temperature, or other property
effects in detailed, nonhomogeneous distributions. Since most models
require good steady-state models as foundations, the following discus-
sions will focus on steady-state modeling employing rigorous, nonlin-
ear, mechanistic models. Models representing important mechanisms
occurring at relatively small scales of composition (which is related to
small length scales) capture many causes and effects across wider
ranges of operating conditions than models that attempt to bypass the
underlying mechanistic steps and pose the model empirically and only
deal with larger scales of physical dimension. For example, ionic con-
centrations in liquids are important at concentration levels far below the
concentrations of the species from which they dissociate. Failure to
model at these small concentration levels diminishes the accuracy of
the macroscopic performance of the equipment models in which these
ionic solutions are employed. Similarly, reactions that mechanistically
occur homogeneously via free radicals or heterogeneously via adsorbed
surface species also are affected profoundly by species concentrations
even as low as 1010 mol fraction levels.
4. MODEL VARIABLES
A significant part of developing a model used for other than determin-

ing static sets of heat and material balances (which are sufficient for
some model objectives, such as providing the basis for new plant
design) is specifying which variables are independent and which are
dependent. Far more variables are dependent variables than are inde-
pendent in essentially all models. For simulation and optimization
cases, there is close to a one-to-one correspondence between actual in-

the-plant independent variables and model independent variables.
Independent variables within a plant are typically boundary conditions
(e.g., feed compositions), equipment parameters (such as heat
exchanger areas, number of trays in a distillation column, etc.), or con-
ditions that can be set because ‘‘final control elements’’ (typically
valves) exist to allow controllers to manipulate flows to meet flow set
points (flow indicators and controllers, FICs), temperature set points
(temperature indicators and controllers, TICs), pressure set points
(pressure indicators and controllers, PICs), and analyzer (composition)
set points (analyzer indicators and controllers, AICs). There are many
measured dependent or ‘‘controlled’’ variables in a plant, such as flow
indicators (FIs), temperature indicators (TIs), pressure indicators (PIs),
and analyzer indicators (AIs).
It is important to have the correct set of variables specified as inde-
pendent and dependent to meet the modeling objectives. For monitoring
objectives observed conditions, including the aforementioned indepen-
dent variables (FICs, TICs, etc.) and many of the ‘‘normally’’ (for simu-
lation and optimization cases) dependent variables (FIs, TIs, etc.) are
specified as independent, while numerous equipment performance
parameters are specified as dependent. These equipment perfor-
mance parameters include heat exchanger heat transfer coefficients,
heterogeneous catalyst ‘‘activities’’ (representing the relative number
of active sites), distillation column efficiencies, and similar para-
meters for compressors, gas and steam turbines, resistance-to-flow
parameters (indicated by pressure drops), as well as many others.
These equipment performance parameters are independent in simu-
lation and optimization model executions.
Models used in a predictive mode (simulation and optimization)
should be validated by comparing their ‘‘steady-state gains’’ (many of
which are nonlinear across their operating range) to those of the plants
they are mimicking, or to reasonable gains experienced in similar equip-
ment configurations. A steady-state gain is the change in a dependent
variable for a unit change in an independent variable.
The aforementioned specifications reflect the control system in a
plant (existing or under development). The importance of having a
model appropriately specified cannot be overstated.
5. SIMULATION
Models used for simulation studies must be predictive. Traditional

simulation models are used to design new facilities and to explore
alternative conditions and configurations of existing facilities. Models
can be built with basic functionality, such as material balances, and can
be improved by adding sufficient detail to meet their intended objec-
tives. Material balances and heat balances can be developed or config-
ured with certainty. Thermodynamic equilibrium relationships, while
well known, may just supply limits (constraints) beyond which sepa-
ration and chemical reactions are not feasible. Rate-based relationships
(heat, mass, and momentum transfer) are far more uncertain than ther-
modynamic state functions and demand more detailed modeling.
When simulating an existing facility, a model’s predictive performance
can be greatly improved by using measurements to update model
parameters before simulation. And it is best to validate a model by
evaluating its performance parameters at different operating conditions
and understanding the sensitivity of simulation results, especially eco-
nomic ones, to parameter values.
6. PARAMETER ESTIMATION AND RECONCILIATION
Models used for monitoring equipment or process performance do not

necessarily need to be predictive, and therefore are considerably less
expensive to develop. These kinds of models are typically used solely in
parameter estimation or reconciliation modes to track key process indi-
cators (KPIs) or equipment performance. Parameter estimation can be
done with a model that is ‘‘square,’’ that is, has an equal number of
equations and unknowns, whereas reconciliation estimates the same
parameters but has degrees of freedom (DOFs) that allow the reconcil-
iation to distribute errors among redundant measurements and uncer-
tain values that are otherwise assumed to be constant, while minimizing
a classic least-squares objective function, or a more sophisticated gross
error detection (GED) objective function Özyurt1 and Pike, 2004; Tjoa
and Biegler, 1991. The DOFs are typically offsets or multipliers between
measured and model values and the aforementioned uncertain ‘‘other-
wise constant’’ variables. These latter DOFs can be variables such as
equipment performance ones that have no single measurements that
directly reflect their values.
While some reconciliation models only have material balance rela-
tionships, more meaningful reconciliation results are obtained with mod-
els that include material balances, heat balances, equilibrium constraints
(both in the separation and reaction domains), rate relationships (heat
transfer, mass transfer, momentum transfer, and kinetics), as well as
equipment-specific relationships. In other words, one should include
more than just material balance constraints when reconciling a model.
Heat balance, kinetics, transport relationships—if needed for the
predictive capability of the model—need to be parameterized in the

reconciliation. This is not replacing measurements, but using available
measurements for reconciliation to determine optimal model parameters.
7. OPTIMIZATION
A model used for optimization must be predictive, and must have

several other features to be an effective tool. Today’s optimizers have
the model and the solution ‘‘engine’’ segregated. Engineers focus on
posing the appropriate problems, and are relieved to a great extent of
how the optimization problem is solved. Some previous generations of
models used for optimization have had the model and optimization
algorithms intertwined and were extremely difficult to modify and
maintain with changing plant configurations. Today’s optimizers
almost exclusively use gradient search techniques, and therefore must
have continuous functions and first derivatives (the Jacobian) to be
efficient and robust. Optimizers that use second derivative information
(such as Successive Quadratic Programming (SQP)4) require good esti-
mates or direct calculation of the second derivatives (the Hessian) of the
equations and the objective function. Optimizers must be able to honor
limits of operating conditions so that the solutions are meaningful.
These limits are typically imposed as upper and lower limits on both
dependent variables and independent DOFs, and if not judiciously
imposed can lead to mathematically infeasible problems. The efficiency
of solution (number of successive iterations and therefore execution
times) is greatly affected by the optimization algorithm and how limits
are imposed. Many optimization methods are degraded and even
caused to fail by imposing too many unnecessary limits since the path
to the solution is affected. Most efficient optimization methods use
infeasible path methods, and only require feasibility at the solution.
The model equations themselves limit solution values to be between
reasonable limits if posed well. For example, there is no need to limit
mole fractions to be between zero and one when the model equations
are well posed.
DOFs in an optimization case are independent variables that have
final control elements, such as valves. These DOFs include FICs, PICs,
TICs, and AICs. Objective functions typically are ones that maximize
P P P
operating profit ( Product values Feed costs Utility costs). The
aforementioned objective function can serve well in both sold out and
market limited economic environments. The different economic envi-
ronment can be imposed by simply bringing a product flow limit into
play (for the market limited situation) or out of play (for the sold out
situation). In sold out conditions more physical limits (maximum allow-

able temperatures, pressures, maximum valve openings, product qual-
ities, etc.) become active than in the market-limited environment.
8. SUMMARY OF SIMULATE/OPTIMIZE/PARAMETER/
RECONCILE CASES
Most modelers are more familiar with ‘‘Simulate’’ cases that generate
results with naturally independent variables kept constant and natu-
rally dependent variables being calculated. For models with all relation-
ships (equations) that are universally true, such as material balances
and heat balances, Simulate cases are many times sufficient. When
equipment-specific performance relationships are integrated into the
aforementioned type of models, for new plants Simulate cases may still
be sufficient since equipment performance is estimated using best prac-
tices, and no actual performance results are available to enhance the
estimated performance parameters. For existing facilities, actual plant
(including equipment) performance results are available, and can be
used as feedback to greatly enhance a model’s predictive accuracy.
Feedback can be incorporated by solving ‘‘Parameter’’ or ‘‘Reconcile’’
cases where many variables that were dependent in the aforementioned
Simulate cases are now known, and therefore held constant (become
independent) while other variables that were independent (equipment
performance factors) are now dependent (calculated).
Both the Parameter and Reconcile cases determine (calculate) the
same set of parameters. However, these cases do not get the same
values for each parameter. A Parameter case has an equal number of
unknowns and equations, therefore is considered ‘‘square’’ in mathe-
matical jargon. In the Parameter case, there is no objective function that
drives or affects the solution. There are typically the same measure-
ments, and typically many redundant measurements in both the
Parameter and Reconcile case. In the Parameter case we determine,
by engineering analysis beforehand (before commissioning an online
system for instance) by looking at numerous data sets, which measure-
ments are most reliable (consistent and accurate). We ‘‘believe’’ these,
that is, we force the model and measurements to be exactly the same at
the solution. Some of these measurements may have final control ele-
ments (valves) associated with them and others do not. The former are
of FIC, TIC, PIC, AIC type whereas the latter are of FI, TI, PI, AI type.
How is any model value forced to be exactly equal to the measured
value? The ‘‘offset’’ between plant and model value is forced to be zero.
For normally independent variables such as plant feed rate, tower
pressures, and so on, that is straightforward, and no ‘‘measurement

model’’ containing a plant model and offset is really required (but there
are many reasons it is a best practice to use a measurement model for all
measurements). For otherwise dependent variables (FI, PI, etc.) in a
Parameter case, we declare equipment performance parameters to be
dependent. This allows these kinds of measurements to be believed and
‘‘square up’’ the problem, keeping it well posed. For a measurement
that is redundant and cannot be used to update an equipment perfor-
mance parameter (since another measurement may already be updating
that equipment parameter) its offset (plant minus model) is assigned to
be dependent, again squaring up the problem. In this process, judg-
ments have been made on where to allocate the model-measurement
mismatch (‘‘error’’).
A Reconcile case is configured starting from the Parameter case
described above. An objective function is added, either a classic least-
squares formulation or a more sophisticated GED approach where the
contribution to the objective function does not continue to grow for
larger and larger model-measurement offsets, but if the offset is large
enough, its contribution to the objective function actually decreases.
This GED approach is an automated, during-solution equivalent of
‘‘throwing out’’ bad data that is so inconsistent that its confidence limits
are reduced (standard deviation increased). The specifications of the
offsets of the measurements we believe absolutely in the Parameter case
are changed from Constant (independent) to Reconcile, which declares
them to be DOFs in the Reconcile case. These offsets, as well as the ones
associated with the redundant measurements which are already depen-
dent are included in the Reconcile case objective function, along with
their 95% confidence limits (1/s )2, where s is the standard deviation of
the measurement. At the solution to this problem offsets that were
constant and zero in the Parameter case will be nonzero; consequently
error will be distributed differently than in the Parameter case. So the
equipment parameters will also be somewhat different. Since the
Reconcile case is nonsquare it can honor constraints (upper and lower
limits on any set of variables). Unless slack variables are introduced
limits cannot be honored in the Parameter case since there are no DOFs;
all the equations must be satisfied with a given set of ‘‘knowns.’’
Reconcile cases can assure that many undesirable solutions are elimi-
nated. Furthermore, when reconciling there is no need to reconcile the
offsets of every variable that had a constant offset in the Parameter case.
Judicious choices can be made on which variables to reconcile. Well-
designed software includes a convenient set of specification types that
allow users to assign appropriate specifications during model develop-
ment (or ‘‘on the fly,’’ driven by data) and then to switch from
Parameter to Reconcile to Simulate to Optimize cases without changing

specifications. The great majority of variables are of ‘‘Calculated’’ spec-
ification. Far fewer are ‘‘Constant,’’ and even fewer are DOFs (i.e., of
‘‘Reconcile’’ and ‘‘Optimize’’ specification). The specifications assigned
to variables, their values, their upper and lower limits, along with the
Case type (Parameter, Reconcile, Simulate, or Optimize) and an objec-
tive function determine the problem that is being posed.
9. MODEL SCOPE
The scope of a model is typically chosen so that it is sufficient to envelop
the feeds, product, and utilities to be able to have a meaningful eco-
nomic objective function. Sometimes it is preferable to include
upstream or downstream facilities even in a simple form to be able to
use actual feed, product, and utility costs rather than transfer prices.
Smaller scope models, even of individual unit operations, are of
course useful in design and monitoring modes. However, larger scope
models allow far greater opportunities for meaningful economic anal-
ysis and optimization, either of new or of existing plants.
10. MODEL FIDELITY
Typically it is best to develop a model that has just sufficient fidelity

to meet the modeling objectives. Care should be taken and informed
decisions made to assure that future needs can be met with minimal
cost, so at times some preinvestment can be made in a model by
including higher fidelity than absolutely required for immediate
objectives.
Many modelers may feel that it is necessary to employ quite sim-
plified models for some applications, for instance for models
deployed online. Today’s high-speed servers and efficient modeling
software allow very detailed and high fidelity models to be used
online where execution speed and robustness are important. A model
that is too simple may cause difficulties when deployed to optimize
existing facilities. The optimization benefits are often ‘‘small differ-
ences between large numbers’’ since the improvement in operating
profit will likely be a small fraction of the operating profit itself.
Models of sufficient fidelity and accuracy, often across multiple
scales, must be deployed to deliver benefits. Models of insufficient
fidelity and accuracy may leave a significant fraction of potential
benefits uncaptured or worse yet can compute benefits in the simple
model that result in decreased operating profit in the real plant.

Typically, the more mechanistic the model the easier it is to understand
the causes and effects between operating conditions and operating
profit. Troubleshooting is typically more difficult when simplified mod-
els are employed. Empirical models have their place even in otherwise
high fidelity models. An additional consideration when determining
the appropriate modeling approach is maintainability. For example,
correlation-based models can be very effective, but may be invalidated
by changes to the process.
Sensitivity analysis can be done to understand the impacts of each of
the overall model’s individual models and equations to the objective
function and to the computed optimal operating conditions.
11. EMBEDDED SOLUTION STRATEGIES
Models can have the characteristic of different types and sizes of equa-
tion sets relative to a general set of algebraic equations. Some common
example situations include physical property models and models con-
taining differential equations. In posing the mathematical problem to be
solved, a completely simultaneous solution approach can be used or a
‘‘mixed mode’’ that combines specialized solution techniques within
the overall EO approach.
There are advantages and disadvantages to using an embedded
solution strategy. Two examples of advantages are (1) resolve issues
around multiple solutions to the equation set and (2) more efficient
solution by exploiting model structure. First, in physical property mod-
els employing a cubic equation of state, it is possible to have multiple
solutions to the cubic equation. This can present robustness problems if
solved simultaneously with all other equations. A dedicated algorithm
to solve the cubic equation with logic to select the proper root is a
preferred approach. Second, differential equation sets can present
highly structured models that can be exploited by dedicated solvers
for more efficient solution. The efficiency can result in part by reduction
of the number of variables and equations presented to the general EO
solution.
Requirements to use embedded solution techniques include giving
accurate function and first derivative evaluations for Newton-based
methods. Embedded convergence needs to be tight to ensure the results
are accurate and precise. However, this can yield longer solution times
and make a simultaneous solution approach preferred. Embedded
solutions provide a way of reducing the model variables exposed to
the overall solution. The modeler should ensure the eliminated
variables are not needed for calculation of constraints, objective func-

tion, of model sensitivities.
12. PROCESS ECONOMICS
Accurate representations of the process economics with market and

emission constraints are just as important as process modeling. The
pricing along with market and emission constraints provide the eco-
nomic basis and reflect changing business conditions over time. This
can allow the optimization of the model to produce novel operating
conditions. These economic inputs may be generated with planning and
scheduling models that have a different, usually broader model scope.
In this way, business aspects across multiple sites or across a business
unit can be reflected at the unit optimization level.
The model scope greatly effects the requirements for process eco-
nomics. The boundaries define the required input and modeling asso-
ciated with pricing. By selecting different model scope or model
method, the pricing can be made easier or near impossible. For example,
streams and products that are bought or sold externally to the business
unit are far easier to price than streams and products that are internal to
the business unit or company. Also, it would result in less information
to validate the model if one tries to price a stream without measure-
ments. Since the optimization algorithm is making relative trade-offs to
optimize the objective function, the pricing or costing should be on a
marginal basis.
13. OFFLINE AND ONLINE USAGE

Models used in an offline environment have less stringent requirements
since the user is responsible for the validity of the input and output,
whereas in online environments models and their surrounding support
software must handle validity automatically. Additionally, models
used online typically must be more robust to solve across sometimes
rather wide ranges of operating conditions. Commissioning an online
model application can take as much as one-third more effort compared
to developing a model for offline usage.
Today’s online model applications can sense conditions that indi-
cate what equipment is in operation or out of service, and how equip-
ment is configured (‘‘lined up’’). ‘‘Presolve’’ logic, based on measured
data, automatically turns equipment models on and off and configures
equipment model interconnections and values correctly. Posing the
correct problem for the moment in an online system is an important part

of its application’s requirements.
14. MODEL MAINTENANCE
Maintenance of large size, high fidelity online applications needs to be

considered in the same way maintenance of any other model-based
operations improvement or process control application is handled.
First of all, commissioning cannot be considered complete until the
system is robust under normal operating conditions and during changes
in operating conditions and/or plant configuration that will occur even
if only relatively infrequently. It is necessary to run two cases every
optimization cycle. The first case uses observed conditions to update
equipment and process performance parameters to make the model
solution best match actual operation. The second case uses the eluci-
dated equipment performance factors along with all other model equa-
tions to better predict how dependent conditions will change with
manipulation of independent DOFs. In the first Parameter or
Reconcile case, observed conditions are used to automatically ‘‘pose the
right problem,’’ a very important task that exists in online model appli-
cations but is not required when doing offline modeling. All equipment
that can come on or off line, and all reconfiguration of piping lineup
have to be anticipated so that proper action can be taken to use available
data to pose the right problem. This set of tasks of posing the right
problem is at times referred to as run time initialization or RTI. An
integral part of the model, RTI logic is executed in scripts after online
data is fetched and before the Parameter or Reconcile case is executed.
Quite complex yet robust RTI logic is employed in online systems. For
example, in an olefins plant there are typically 10 to even 20 steam
crackers in the plant, some of which are typically out of service for
decoking. Also, each steam cracker online may be lined up to use dif-
ferent feedstocks. RTI logic uses observed data to pose the right problem
for all the possible combinations of these on/off situations and feedstock
configurations. Far more project execution time is spent on handling
abnormal operating conditions than for normal operating conditions. A
Reconcile case can be posed to improve model robustness since imper-
fect data can be better processed while maintaining reasonable results.
Reconciliation can help where decisions cannot be made from raw data
alone, but only when this data is used to solve the model equations. For
example, raw temperature and flow measurements may indicate that a
heat exchanger has a ‘‘temperature cross’’ that violates the Second Law
of Thermodynamics. The solution can be forced to honor Second Law
constraints by lower bounding temperature crosses to be slightly larger

than zero, and measured flows and temperatures can be reconciled
slightly from their observed values to allow these constraints to be
honored. Weighting of measurement confidence limits during reconcil-
iation should reflect normal instrument error. Once a model-based appli-
cation is commissioned with robust performance, maintenance will be
limited to infrequent changes in plant equipment, instrumentation, and
piping lineup. Most maintenance can be expected to occur after a major
plant turnaround. EO modeling, because of its simultaneous solution
approach makes adding, deleting, or changing a part of a complex model
easier than if SM methods were used, since SM methods require atten-
tion to the order in which they are solved, and employ many nested
convergence loops that have to be modified when changes are made.
15. EXAMPLES OF INDUSTRIAL APPLICATIONS
Since the advent of efficient and robust simulation and optimization

solution ‘‘engines’’ and flowsheeting software packages that allow for
relatively easy configuration of complex models, numerous integrated,
high fidelity, and multiscale process model applications have been
deployed in industrial plants to monitor performance and to determine
and capture improvements in operating profit.
Well over 50 large-scale EO model-based RTO applications have
been deployed for petroleum refining processes. These model applica-
tions have been deployed in petroleum refineries Liporace et al., 2009;
Camolesi et al., 2008; Mudt et al., 1995, both on separation units (crude
atmospheric and vacuum distillation units) and on reactor units
(including fluidized catalytic crackers (FCC), gasoline reformers, and
hydrocrackers).
Petrochemical plants, especially olefins plants that can manufacture
numerous products in different proportions from the same feedstock,
have had probably the greatest success at delivering value from sophis-
ticated online plant-wide models Cutler and Perry, 1983; Rejowski et al.,
2009; Paules and Meixell, 1994; Fatora and Ayala, 1992; Fatora et al.,
1992a; Fatora et al., 1992; Houk et al., 1992; Kelly et al., 1991. Over 50
ethylene RTO applications have been deployed, as well as several
others on nonolefin petrochemical processes.
Chemical plants that have small conversion per pass, large expensive
recycle ratios (typically vapor, requiring recompression), selectivities
and yields that are not close to ideal, and which have parallel reactors,
for instance, also have proven to be good candidates for plant-wide
optimization applications. Well over 30 of these plant-wide applications
are in use today Mercang€ oz and Doyle, 2008; Lowery et al., 1993; Meixell
and Tsang, 1988. Microkinetics-based reactor models Dumesic et al.,
1993, have been deployed in several online, plant-wide optimization
applications (e.g., in an ethylene oxide plant using a published reaction
mechanism Stegelmann et al., 2004) with good success. These reactor
models span the scales from surface chemistry to macroscopic selectivity
improvements, which result in significant reduction of feedstock usage.
The microkinetics approach is basically addressing what has tradition-
ally been done by mathematical manipulation of reaction mechanism
relationships to arrive at closed form reaction rate expressions. By
numerically dealing with more open forms of the same mechanisms,
reaction models can be more easily developed with fewer assumptions,
and can be more easily modified, maintained, and understood.
On the modeling level, there are several elements required to deliver
success and sustain benefits. They include opportunity for improve-
ment, DOFs (independent conditions) manipulated across an effective
range, and trade-offs must exist for optimization modeling applications
to deliver benefits. There are additional critical success factors, beyond
the modeling level, for successful modeling applications in industry.
Summarized in Table 1, these critical success factors include but are not
limited to having a process control system to support online applica-
tion, integration of business processes and operations, long-term focus
Table 1 Critical success factors for successful modeling applications
Model objectives Sufficient functionality and resolution for present

objectives are included. Path for expanding scope is
thought about so ‘‘dead ends’’ are avoided
Modeling Equation-oriented models and large scale solvers for
technology speed, robustness, and maintainability
Model accuracy Rigorous process operations and physical properties
for accuracy
Model consistency Integration with offline modeling
Model constraints Correct limits and bounds
Control system Effective multivariable process control system in place
for online applications
Planning/ Integration of business processes, operations, and
scheduling pricing
Sustained Long-term focus with clearly defined resources to
performance sustain benefits
Skill set Many disciplines are required
Metrics Continual auditing process in place
for sustained performance, skill set of the organization, and continuing

auditing process for performance metrics.
16. CO2 CAPTURE WITH AQUEOUS ALKANOLAMINE

SOLUTION
The model for CO2 capture section of a chemical plant is used as an

example to illustrate integrated, high fidelity, multiscale process mod-
els that are being applied to improve plant operations. CO2 capture by
absorption with aqueous alkanolamines represents an especially rele-
vant modeling study these days because it is considered a key technol-
ogy to reduce CO2 emissions from fossil-fuel fired power plants and
other CO2 emitters to help alleviate global climate change Rochelle,
2009; Zhang et al., 2009. The demonstrated success in deploying this
CO2 capture model online is an important milestone toward future use
of similar models configured and used to help reduce CO2 emissions in
the most economical manner from power plants, chemical plants, and
refinery CO2 emission sources.
Figure 3 shows a comprehensive CO2 capture process model, which
involves thermophysical property and reaction kinetic models for CO2
[(Figure_3)TD$IG]
Figure 3 Integrated, high fidelity, multiscale process modeling of CO2 capture and
storage. Modified based on ‘‘Beyond the Molecular Frontier,’’ NRC report (2003).
and absorption solvents, tray or packed column performance models

for the absorber and the stripper, models for CO2 absorption and sol-
vent regeneration process, models for chemical or power plants inte-
grated with the CO2 capture process, and on a larger scale, network of
chemical or power plants with CO2 capture process, CO2 transportation
pipeline, and CO2 storage reservoirs.
17. MODELING CO2 AND ABSORBENT PHYSICAL

PROPERTIES AND CHEMICAL REACTIONS
Numerous chemical solvents including carbonates, ammonia, and alka-
nolamines have been or are being developed for CO2 capture. The CO2
capture process examined in this study makes use of an aqueous solu-
tion of n-methyldiethanolamine (MDEA), activated with piperazine
(PZ). Figure 4 shows the molecular structure of MDEA and PZ.
MDEA is a tertiary amine. Upon absorbing CO2, MDEA associates with
hydronium ion H3O+ to form a protonated MDEA ion MDEAH+. PZ is
a cyclic amine. It can also associate with H3O+ to form a protonated ion
PZH+, and react with CO2 to form a carbamate ion PZCOO and a
dicarbamate ion PZ(COO)22. In all, for the CO2 capture system with
MDEA/PZ, there are nine ionic reactions that take place in the aqueous
solution. Table 2 shows the nine aqueous phase reactions. Resulting
from the aqueous phase reactions are four cations (H3O+, MDEAH+,
PZH+, HPZH2+), five anions (OH, HCO3, CO32, PZCOO, PZ
(COO)22), and one zwitterion (+HPZCOO).
Table 3 shows the 19 chemical species considered in the model, that
is, 5 inert gases, CO2 to be absorbed, solvent water, MDEA, and PZ
amines, and 10 ionic species. All ionic species, zwitterion included, exist
only in the liquid phase. The vapor phase components are mainly inert,
that is, supercritical components that dissolve sparingly into the liquid.
Among the nine chemical reactions that occur in the aqueous solu-
tion, chemical equilibrium is only attained for six of the reactions. The
[(Figure_4)TD$IG]
Figure 4 Molecular structure of n-methyldiethanolamine (MDEA) and piperazine (PZ).

Table 2 Aqueous phase reactions for CO2 capture with aqueous MDEA/PZ solution
Equilibrium
controlled or
Reaction kinetics controlled Stoichiometry
1 Equilibrium 2H2O $ H3O+ + OH

2 Equilibrium HCO3 + H2O $ CO32 + H3O+
3 Equilibrium MDEAH+ + H2O $ MDEA + H3O+
4 Equilibrium PZH+ + H2O $ PZ + H3O+
5 Equilibrium HPZH2+ + H2O $ PZH+ + H3O+
6 Equilibrium +
HPZCOO+ H2O $ PZCOO + H3O+
7 Kinetic CO2 + OH $ HCO3
8 Kinetic PZ + CO2 + H2O $ PZCOO + H3O+
9 Kinetic PZCOO + CO2 + H2O $ PZ(COO)22 + H3O+
Table 3 Chemical species considered for CO2 capture with aqueous

MDEA/PZ solution
Abbreviation Name Chemical formula
H2 Hydrogen H2
N2 Nitrogen N2
Argon Argon Ar
CH4 Methane CH4
CO Carbon monoxide CO
CO2 Carbon dioxide CO2
H2O Water H2O
MDEA Methyldiethanolamine C5H13NO2
MDEA+ MDEA+ C5H14NO2+
HCO3 HCO3 HCO3
CO32 CO32 CO32
H3O+ H3O+ H3O+
OH OH OH
PZ Piperazine C4H10N2
PZH+ PZH+ C4H11N2+
HPZH2+ HPZH2+ C4H12N22+
HPZCOO HPZCOO C5H10N2O2
PZCOO PZCOO C5H9N2O2
PZ(COO)22 PZ(COO)22 C6H8N2O42
three capture reactions of CO2 with OH, PZ, and PZCOO are kinetics
controlled.
Accurate thermodynamic modeling is essential for meaningful pro-
cess modeling and simulation of the CO2 capture system with aqueous
amine solutions Zhang and Chen, 2011. To accurately describe thermo-
physical properties of the CO2-loaded aqueous amine solutions, one
must explicitly account for the solution chemistry, that is, the chemical
equilibria of the nine ionic reactions summarized in Table 2. In addition,
activity coefficients play a key role in phase equilibrium calculations,
liquid phase chemical equilibrium calculations, and in calculations for
heat of CO2 absorption, liquid heat capacity, liquid enthalpy, pH, and so
on. Rigorous electrolyte thermodynamic models need to be applied as
the electrolyte solutions are known for the highly nonideal liquid phase
behavior resulting from the long-range ion–ion interactions and the
short-range molecule–molecule, molecule–ion, and ion–ion interactions
in the liquid phase. The most common choice is the electrolyte NRTL
(nonrandom two-liquid) activity coefficient model Song and Chen, 2009.
Figure 5 shows comparison of the predicted CO2 partial pressures
for the MDEA–CO2–water system with the experimental data of Jou
[(Figure_5)TD$IG]
Figure 5 Vapor–liquid equilibrium of MDEA–CO2–H2O (MDEA mass fraction at

et al. (1982) and Ermatchkov et al. (2006). The model predictions on

CO2 partial pressure, species concentrations, heat of absorption, heat
capacity, and enthalpy must be validated for the aqueous amine
solutions before the activity coefficient model can be used for process
modeling and simulation.
Apparent rate expressions are available to describe the reaction
kinetics for the forward and reverse reactions of the three CO2 capture
reactions, that is, reactions 7–9 shown in Table 2, Bishnoi and Rochelle,
2002, Further mechanistic enhancements to the rate expressions are
forthcoming through use of zwitterions (i.e., +HPZCOO), first pro-
posed by Caplow (1968) and later reintroduced by Danckwerts (1979).
18. MODELING CO2 CAPTURE PROCESS
Figure 6 shows a simplified CO2 capture process with two major pro-
cess units: absorber and stripper Zhang et al., 2009. A lean amine solvent
(low CO2 loading) is fed into the top of the absorber and is in counter-
current contact with the gas containing CO2. The CO2 is chemically
absorbed by the amine solvent and the treated gas exits the top of
the absorber. The rich (high CO2 loading) amine leaves the bottom of
the absorber and is preheated by a cross heat exchanger before
[(Figure_6)TD$IG]
Figure 6 Typical CO2 capture unit with absorber and stripper.

entering the top of the stripper. At stripper conditions, typically at

higher temperature, the reaction between the amine and CO2 is
reversed, liberating the CO2. A concentrated CO2 stream is obtained
from the top of the stripper. The lean solvent from the stripper is
cooled and goes back to the absorber.
There are two well-established approaches in modeling perfor-
mance of tray or packed columns used for the absorber and the stripper.
They can be modeled either with the simpler traditional equilibrium-
stage approach or more rigorous and higher fidelity rate-based
modeling approach Zhang et al., 2009. The traditional equilibrium-stage
models assume each theoretical stage is composed of well-mixed vapor
phase and liquid phases and these two bulk phases are in phase equi-
librium with each other. Tray ‘‘efficiencies’’ can be introduced to
improve matching with column performance data. Although consid-
ered at times to be an adequate approximation for heat and mass
balance calculations, this equilibrium-stage assumption yields inade-
quate model fidelity in modeling performance of chemical absorption
columns where the contacting phases are far from in equilibrium. In
contrast, rate-based multistage separation models assume that separa-
tion is caused by heat and mass transfer between the contacting phases;
phase equilibrium is achieved only at the vapor–liquid interface, and
the Maxwell–Stefan theory is used to calculate mass transfer rates.
In rate-based multistage separation models, separate balance equa-
tions are written for each distinct phase, and mass and heat transfer
resistances are considered according to the two-film theory with
explicit calculation of interfacial fluxes and film discretization for non-
homogeneous film layer. The film model equations are combined with
relevant diffusion and reaction kinetics and account for the specific
features of electrolyte solution chemistry, electrolyte thermodynamics,
and electroneutrality in the liquid phase.
Specifically the mathematical model for rate-based multistage
separation model consists of the following equations for each stage:
(1) material balances for bulk liquid, bulk vapor, liquid film, and vapor
film, (2) energy balances for bulk liquid, bulk vapor, liquid film, and
vapor film, (3) phase equilibrium at the interface, (4) summations,
(5) mass fluxes for liquid film and vapor film, and (6) heat fluxes for
liquid film and vapor film. Reaction terms are accounted for in the
material balance equations. The Maxwell–Stefan multicomponent mass
transfer equations are used to describe the mass fluxes. A driving force
due to electric potential in each liquid film region is introduced to
satisfy electroneutrality conditions at the boundary of the film region.
When the films are discretized to account for enhanced transfer rates
due to reactions, equations of material balances, energy balances, mass
fluxes, and heat fluxes are formulated and solved for each discretized
film segment.
Figure 7 illustrates the discretized liquid film for CO2 transfer across
the vapor and liquid films. Here Y is gas phase composition, X is liquid
phase composition, T is temperature, I is interface, V is vapor, and L is
liquid. Note that the liquid film is discretized into multiple film seg-
ments to accurately model the nonhomogeneous film layer.
Figure 7 further shows that, as gaseous CO2 moves up the absorber,
phase equilibrium is achieved at the vapor–liquid interface. CO2 then
diffuses through the liquid film while reacting with the amines before it
reaches the bulk liquid. Each reaction is constrained by chemical equi-
librium but does not necessarily reach chemical equilibrium, depend-
ing primarily on the residence time (or liquid film thickness and liquid
holdup for the bulk liquid) and temperature. Certainly kinetic rate
expressions and the kinetic parameters need to be established for the
kinetics-controlled reactions. While concentration-based kinetic rate
expressions are often reported in the literature, activity-based kinetic
rate expressions should be used in order to guarantee model consis-
tency with the chemical equilibrium model for the aqueous phase solu-
tion chemistry.
Success of rate-based multistage separation modeling is ultimately
tied to underlying equipment hydrodynamics performance correla-
tions for tray or packed columns. For example, the thickness of the film
[(Figure_7)TD$IG]
Figure 7 Discretized liquid film for CO2 capture with chemical absorbent.
in each phase is computed as the ratio of the average diffusivity and the
average mass transfer coefficient. In the case of packed columns oper-
ated countercurrent-wise, correlations are required for the mass trans-
fer coefficients, the effective mass transfer area, the pressure drop, and
the flood capacity of the column. Dependable and dimensionally con-
sistent equipment performance correlations validated for modeling of
CO2 capture with aqueous amines have recently been developed for
random and structured packing families Hanley and Chen, 2011. Used
in rate-based calculations for columns, these performance correlations
form part of the underlying equations for accurate modeling of the
absorber and the stripper. They require information on packing or tray
types and geometries along with transport properties such as surface
tension, viscosity, density, diffusivity, and thermal conductivities.
19. MODELING CO2 CAPTURE PROCESS AND THE PLANT
The example CO2 capture process, shown in Figure 8 as an Aspen Plus

EO model representation, is part of an ammonia plant. Designed to
scrub CO2 from ammonia synthesis gas, it includes an absorber and
two solution regeneration columns, one stripping the rich, CO2 laden
solution leaving the absorber to semilean concentration of absorbed
CO2, and the other cleaning the solution even further to lean solution
[(Figure_8)TD$IG]
Figure 8 Aspen Plus EO model for an MDEA/PZ CO2 capture unit.

[(Figure_9)TD$IG]
Figure 9 Aspen Plus EO model for an ammonia plant.
quality. In addition, there are the associated heat exchangers, pumps, a

power recovering expander, and flash drums. Figure 9 shows the hier-
archy-level model representation for the Aspen Plus EO ammonia plant
model. The CO2 capture unit model is shown only as a hierarchy block
in the ammonia plant model. Other blocks in the plant model represent
other major process units such as air compressors, primary reformer,
secondary reformer, water gas shift, methanator, ammonia synthesis
loop, refrigeration loop, and steam system.
The gas being scrubbed of CO2 is ammonia synthesis gas, a mixture
of hydrogen and nitrogen in about a 3:1 ratio with small amounts of
methane, CO, and argon present. The gas entering the absorber has
about 18 mol% CO2 and the gas leaving the absorber ranges in CO2
content from about 2 to 500 ppm by volume, depending on plant rate
and the performance of the solution regeneration equipment, which is
affected significantly by numerous factors, including among others,
cooling water temperature, solution strength, and heat exchanger
fouling.
The gas composition is optimized with DOFs outside the CO2 scrub-
bing system with regard to inert composition (methane and argon) and
hydrogen to nitrogen ratio since the levels of these components affect
downstream (ammonia synthesis) reaction kinetics. Improved kinetics
at lower inert levels are achieved at the expense of using more fuel or
feedstock, since lower inerts can be achieved by firing the primary
reformer harder upstream of the CO2 system, or more purge can be used
downstream from the synthesis loop to lower the inerts level. Similarly,
the optimum nonstoichiometric H2:N2 ratio in the gas is determined
using high fidelity reaction kinetics that account for the rate-limiting
step being the adsorption of nitrogen onto the heterogeneous promoted
magnetite catalyst. Constraints may limit the range of operating condi-
tions allowable, and active constraints profoundly affect the optimiza-
tion results. This situation gives rise to classical optimization trade-offs
that are only quantifiable with a high fidelity plant-wide model.
The separation models for absorber and regenerators were first
developed as equilibrium models without the activating PZ compo-
nent, and then were enhanced to include PZ and were migrated to
rate-based (kinetic and mass transfer rate limited) models. Many of
the modeling configuration and robustness issues were addressed in
the simpler models. These issues included several material balance as
well as numerical stiffness challenges that are not intuitively obvious.
Most of the issues related to model specification and use of industrial
data were solved with the equilibrium-based models. The model of the
illustrated flowsheet, along with the complete upstream and down-
stream equipment models for the entire ammonia plant, was deployed
online before upgrading the separation models. Data consistency and
several very important modeling details would have been more diffi-
cult to identify and fix had the more complex rate-based model been
deployed first.
EO models require that the problem be well specified since the
whole system is solved simultaneously. SM techniques are more for-
giving and allow less well posed problems to be solved since incon-
sistencies may only become apparent when the whole system is solved
together. It became apparent that a makeup stream was necessary for
MDEA since although only tiny amounts of MDEA leave with the
scrubbed synthesis gas and with the recovered CO2, the simultaneous
solution could not close the material balances to normally used very
close tolerances without the makeup stream. The makeup stream has to
be dependent since the solution must seek out the makeup amount to
just offset the miniscule MDEA losses dictated by vapor–liquid equi-
librium and approach to equilibrium. There must be a dependent water
makeup stream for the same reasons. Also, even with these additions
the almost total recycle nature of the circulating solution causes the
simultaneous solution to be difficult to solve robustly at different oper-
ating conditions. A very small purge of the solution, as is done in the
plant by filters removing degradation products, makes the model very
robust. The model has been running for several months in open loop,
cycling from parameter estimation to optimize case modes with very
few failures. Closed loop commissioning will occur in a few months.

CO2 slip from the absorber can be changed by several hundred parts per
million from Parameter case to Parameter case, and the plant rate may
change quite significantly (15–20%) and the model solves without pro-
blems. Prior to addressing data consistency, model specification, and
configuration (makeup MDEA, makeup water) issues the model was
not fit for in-the-plant use. These issues were repaired offline before any
attempt was made to deploy the model in the plant environment.
Significant economic benefits have been identified and will be thor-
oughly investigated prior to closed loop operation.
Execution times for the overall ammonia plant model, of which the
CO2 capture system is a small part, are on the order of 30 s for the
parameter estimation case, and less than a minute for an Optimize case.
The model consists of over 65,000 variables, 60,000 equations, and over
300,000 nonzero Jacobian elements (partial derivates of the equation
residuals with respect to variables). This problem size is moderate for
RTO applications since problems over four times as large have been
deployed on many occasions. Residuals are solved to quite tight toler-
ances, with the tolerance for the worst scaled residual set at approxi-
mately 1.0 109 or less. A scaled residual is the residual equation
imbalance times its Lagrange multiplier, a measure of its importance.
Tight tolerances are required to assure that all equations (residuals) are
solved well, even when they involve, for instance, very small but impor-
tant numbers such as electrolyte molar balances.
The overall cycle time from sensing that the plant is reasonably close
to steady-state conditions, fetching and validating data, posing the
appropriately configured problem (using initialization logic based on
observed data), solving the parameter estimation case, and then the
Optimize case takes less than 5 min. This overall plant model includes
rigorous reaction rate-based models for all the major reactors, including
the primary and secondary steam reformers, high- and low-tempera-
ture water gas shift reactors, a methanator, and a multibed ammonia
synthesis reactor. The process, refrigeration, steam, and fuel/flue gas
parts of the plant are all included in the plant-wide model. All com-
pressors and their steam turbine drivers are modeled with predictive
relationships. All the recycle streams are closed and solved very effi-
ciently using the EO simultaneous solution approach.
For the CO2 capture system, the parameter cases update all signif-
icant equipment performance to match observed data. The primary
performance of this system is the CO2 absorber slip and the energy
required to regenerate the solution to semilean and lean solution
quality. The model can be made to match observed performance in
several ways. Measurements are available for the lean and semilean
solution CO2 content (with all CO2-bound species expressed as mass

percent CO2). The semilean and lean solution columns are forced to
deliver solution with the observed quality by adjusting liquid holdup
(affecting residence times and therefore reaction effects) and interfa-
cial areas (affecting mass transfer rates). Holdup and interfacial areas
are determined from correlations for specific packing families or trays
(depending on the equipment configuration) but these correlations
are not precise predictors. Adjustment can be made to only one
parameter in any one of these correlations (usually a multiplier fac-
tor), or reconciliation can be employed to distribute the model–data
mismatch among more than one correlation. Sensitivity analysis is
useful for determining reasonable ranges for the parameters (factors).
For packing the height equivalent to a theoretical plate (HETP) needs
to be reasonable to have meaningful adjustments to correlations to
match observed performance. Even lean and semilean flow rates may
have associated errors and offsets or multipliers to these flows can be
simple parameters used to match the model results to observed data.
The best way to determine the best choice of parameters to update
every Parameter case cycle is to analyze multiple data sets simulta-
neously as part of the commissioning effort. These data sets should be
at different operating conditions. A set of parameters with the least
deviation from ‘‘ideal’’ (typically a factor of 1.0) that can best match
several data sets at different conditions is desirable.
Optimization cases are done only with the overall plant model (with
any unit operations that are out of service turned off in the model
automatically using logic based on measured data). The CO2 absorber
cannot leak out too much CO2 since the absorber overhead stream goes
to a methanator that will overheat with too much CO2 in its feed. This
situation can occur rapidly and can cause the plant to have to vent
synthesis gas before the methanator—a costly incident. The overall
plant optimization case objective function is operating profit, so the
DOF in the CO2 scrubbing area and those upstream that affect the
operating profit include those that must keep the CO2 slip to an accept-
able level using the least costly utilities, while delivering the most CO2
to the urea plant. Far upstream in the process, the primary reformer feed
steam to carbon (S/C) ratio is many times most economical to minimize
(the primary reformer fuel usage is reduced), but the excess steam in the
process provides the lean solution regenerator heat, so the S/C ratio
cannot be reduced below that needed to sufficiently regenerate the
solution to allow adequate absorber performance. A high fidelity model
operating online in conjunction with a multivariable, predictive,
advanced control system makes these trade-offs often and allows the
plant to continuously operate closer to the optimum.
The Optimize case executes about 12 times per day, sufficient to

capture differences in the optimum set of operating conditions due to
ambient and cooling water temperatures, changing natural gas compo-
sition, and operational mode changes imposed by external business
decisions and feed gas availability. The maximum number of execu-
tions per day is not set by computing times, but is set by the aforemen-
tioned conditions as well as the settling time of the ammonia plant. The
advanced control system accepts new external targets after each opti-
mization cycle, moves the plant to these conditions, and holds these
conditions until new targets are received. Most external targets are
intensive conditions such as feed steam to carbon ratio, tempera-
tures, and compositions or composition ratios (i.e., synthesis gas H2
to N2 ratio). Extensive conditions such as flow rates can be contin-
uously manipulated by the control system to allow, for instance,
maximum production to be sought out, if desired, between optimi-
zation executions.
Ammonia plants are linked to the ambient conditions more closely
than many plants since the process air compressor (supplying the nitro-
gen for the synthesis gas and the oxygen for the autothermal secondary
reformer) is significantly affected by ambient temperature. Another
trade-off that the optimizer exploits is the best distribution of the very
high pressure (120 kg/(cm2 g)) steam between the ammonia plant syn-
thesis gas steam turbine driver and a similar driver of the urea plant
CO2 compressor. There is very close coupling among the S/C ratio, very
high-pressure steam generation using waste heat recovery, the CO2
removal system performance, and other key operating conditions.
20. MODELING SITE-WIDE CHEMICAL COMPLEX
Shown in Figure 10, this ammonia plant is a major part of the overall
fertilizer site complex. Other major facilities include urea plant, steam
system, and cooling water system. Most of the ammonia is used to make
granulated urea product. The other raw material for urea synthesis is
CO2 from the CO2 capture system in the ammonia plant, supplemented
with a small stream from an adjacent business. The ammonia produc-
tion and the CO2 available from the ammonia plant are never precisely
in balance, in part because of the overall stoichiometric yields of ammo-
nia and CO2 from the natural gas feedstock. CO2 is the limiting feed-
stock for the urea plant and its production rate in the ammonia plant
sets the urea plant production rate since there is no intermediate CO2
storage to buffer the urea production from the CO2 production rate.
Ammonia that is produced in excess of that which is used to make urea
[(Figure_0)TD$IG]
Figure 10 A fertilizer site complex.
is sent to refrigerated storage, and sold. The ammonia and urea plants
are not only integrated with raw material streams but are strongly
coupled by the site steam system, and less critically, through the cooling
water system. Both the synthesis gas compressor in the ammonia plant
and the CO2 compressor in the urea plant use the highest pressure
steam available at the site, 112 bar steam, generated from waste heat
recovery in the ammonia plant. These two turbines are the only con-
sumers of this ‘‘very high-pressure’’ steam. As the plants’ production
rates increase the total power demand in these turbines increases faster
than the 112 bar steam production. When the 112 bar letdown valve to
the 48 bar steam header approaches being closed, both the ammonia
and urea plants cannot increase their production rates since the utility
boiler supplies 48 bar steam—112 bar steam is only generated by waste
heat recovery. The urea plant production can be increased by venting
ammonia plant synthesis gas before the compressor (thus not increas-
ing its power demand), and increasing the front-end rate of the ammo-
nia plant to produce more CO2. This operating mode is costly, but the
overall plant economics are more favorable at these higher urea pro-
duction rates than at lower rates. The optimization model includes a
simplified but rigorously correct urea reactor to account for stoichio-
metric consumption of both ammonia and CO2. Additionally, the
model includes a detailed rigorous model of the CO2 compressor and
its steam turbine to allow trade-offs among the power demands of this
turbine, the ammonia synthesis gas turbine, and the steam generation
from waste heat recovery. Significant savings have been identified

through the strategy of increasing the steam-to-carbon ratio in the
ammonia plant, which requires more fuel usage in the primary
reformer, but generates significantly more 112 bar steam by recovering
a large portion of this energy. This allows higher ammonia and urea
plants rates to be achieved without venting synthesis gas. The optimi-
zation model includes a quite detailed model of the site steam system,
not only at the highest steam pressure levels but also of all the steam
headers, all the way down to the condensate recovery levels. The 48 bar
steam utility boiler is modeled, including its fuel and combustion air
system. The strategy employed when defining the model scope was to
include sufficient site-wide facilities to honor important constraints, to
exploit available trade-offs, and to account for the effects of the optimi-
zation system directly, and not solely through feedback.
This application illustrates that integrated, high fidelity, multiscale
models from molecular level to site-wide complex can be deployed in
nonideal online environments to deliver benefits and insight that can-
not be elucidated with simpler, less rigorous, more empirical models.
21. SUMMARY
Efficient and robust process models of chemical plants and refineries

deliver benefits by making accurate trade-offs often that cannot be
determined through other means. These models must be quite accurate
in areas that affect economics and feasibility to determine benefits that
can be captured, that while large, are small percentages of the overall
operating profit. Rigorous, mechanistic models satisfy the required
accuracy, but even these are greatly improved by using measured plant
data to update model parameters. Models of appropriate heterogeneous
fidelity and scale must be integrated to keep model development costs
down while still being able to identify actionable profitable changes in
conditions. EO simultaneous solution techniques make it possible to
solve highly complex models in industrial environments efficiently
and very robustly. The nature of the optimal solutions is that they will
be at many simultaneous constraints, so an advanced process control
application (typically model based, predictive, and multivariable) is
needed to impose the determined solution onto the operating plant.
ACKNOWLEDGMENT
The authors thank our colleagues Maurice Jett and Joseph DeVincentis
for their careful review of this manuscript.
REFERENCES
Alkaya, D. Vasantharajan, S. and Biegler, L. T., Successive quadratic programming: appli-
cations in the process industry, in ‘‘Encyclopedia of Optimization’’ (C. Floudas, P.
Pardalos, Eds.), pp. 400–413. Vol.5, Kluwer (2001).
Biegler, L. T. Advanced Nonlinear Programming Methods for Chemical Process
Optimization University of Wisconsin-Madison., O. A. Hougen Lecture, September
22 (2009).
Bishnoi, S. and Rochelle, G. T. AIChE J 48, 2788–2799 (2002).
Camolesi, V. J. Moro, L. F. L. and Zanin, A. C. Controle y Automacao 19, 128–137 (2008).
Caplow, M. J. Am. Chem. Soc 90, 6795–6803 (1968).
Chen, H. -S. and Stadtherr, M. A. AIChE J. 31, 1843–1856 (1985).
Cutler, C. R. and Perry, R. T. Comput. Chem. Eng 7, 663–667 (1983).
Danckwerts, P. V. Chem. Eng. Sci 34, 443–446 (1979).
Dumesic, J. A. Rudd, D. F. Aparicio, L. M. Rekoske, J. E. and Trevino, A. A. ‘‘The
Microkinetics of Heterogeneous Catalysis, ACS Professional Reference Book’’,
American Chemical Society (1993).
Ermatchkov, V. P erez-Salado Kamps, A. and Maurer, G. Ind. Eng. Chem. Res 45, 6081–6091
(2006).
Evans, L., The Evolution of Computing in Chemical Engineering: Perspectives and Future
Directions. CACHE Trustees 40th Anniversary Meeting, Boulder, Colorado August
8 (2009).
Fatora III, F. C. and Ayala, J. S. Successful Closed Loop Real-Time Optimization
Hydrocarbon Processing (June 1992).
Fatora III, F. C., Gochenour, G. B., Houk, B. G., and Kelly, D. N., ‘‘Closed-Loop Real-Time
Optimization and Control of a World Scale Olefins Plant’’ Paper Presented at the
National AIChE Meeting, New Orleans, Louisiana (April 1992).
Fatora III, F. C. Gochenour, G. B. and Kelly, D. N., ‘‘Modeling Ethylene Plants for Real-
Time Optimization Applications’’, Paper Presented at the National AIChE Meeting
(April 1992).
Hanley, B., Chen, C.-C., New Mass-Transfer Correlations for Packed Towers, AIChE J.
(2011) doi: 10.1002/aic.12574.
Houk, B. G. Kelly, D. N. Davenport, S. L. and Fatora III, F. C., ‘‘Closed-Loop Plant-Wide
Optimization and Advanced Control Technologies Capture Many Benefits’’, Paper
Presented at the International Conference on Productivity and Quality in the
Hydrocarbon Processing Industry (1992).
Jou, F. Y. Mather, A. E. and Otto, F. D. Ind. Eng. Chem. Process Design Dev 21, 539–544
(1982).
Kelly, D. N. Fatora III, F. C. and Davenport, S. L., ‘‘Implementation of a Closed Loop Real-
Time Optimization on a Large Scale Ethylene Plant’’ Paper Presented at the
Meeting of the Instrument Society of America’’ Anaheim, California (October
1991) .
Kisala, T. P. Trevino-Lozano, R. A. Boston, J. F. Britt, H. I. and Evans, L. B. Comput. Chem.
Eng. 11, 567–579 (1987).
Liporace, F. S. Gomes, M. V. C. Katata, A. C. Zanin, A. C. Moro, L. F. L. and Porfırio, C. R.
Comput. Aided Chem. Eng. 27, 1245–1250 (2009).
Lowery, R. P. McConville, B. Yocum, F. H. and Hendon, S. R., ‘‘Closed-Loop Real Time
Optimization of Two Bisphenol-A Plants’’, Paper Presented at the National AIChE
Meeting, Houston, Texas (March 1993).
Meixell, M. D. and Tsang, E. H., ‘‘Large Scale Ammonia Process Optimization in Real
Time’’, Paper Presented at American Control Conference (1988).
Mercang€ oz, M. and Doyle III, F. J. Comput. Chem. Eng 32, 789–804 (2008).
Mudt, D. R. Hoffman, T. W. Hendon S. R. ‘‘The Closed-Loop Optimization of a Semi-

Regenerative Catalytic Reforming Process’’ Paper Presented at the National AIChE
Meeting, Houston, Texas (March 1995).
National Research Council (NRC), ‘‘Beyond the Molecular Frontier: Challenges for
Chemistry and Chemical Engineering’’, p. 74. National Academies Press,
Washington, D.C. (2003).
Özyurt1, D. B. and Pike., R. W. Comput. Chem. Eng 28, 381–402 (2004).
Paules I.V., G. E. and Meixell Jr., M. D., ‘‘A Fundamental Free Radical Kinetic Pyrolysis
Model for On-Line Closed-Loop Plant-Wide Optimization of Olefins Plants’’,
Paper Presented at CIMPRO’94, New Brunswick, NJ (April 1994).
Rejowski Jr., R. Shah, V. Fontenot, C. E. de Tarso, P. and Santos, V. E. N. Comput. Aided
Chem. Eng 27, 351–356 (2009).
Rochelle, G. T. Science 325, 1652–1654 (2009).
Song, Y. and Chen, C. -C. Ind. Eng. Chem. Res 48, 7788–7797 (2009).
Stegelmann, C. Schiodt, N. C. Campbell, C. T. and Stoltze, P. J. Catal 221, 630–649 (2004).
Tjoa, I. B. and Biegler, L. T. Comput. Chem. Eng 15, 679–690 (1991).
Zhang, Y. and Chen, C. -C. Ind. Eng. Chem. Res 50, 163–175 (2011).
Zhang, Y. Chen, H. -S. Chen, C. -C. Plaza, J. M. Dugas, R. and Rochelle, G. T. Ind. Eng.
Chem. Res 48, 9233–9246 (2009).
CHAP TER 4
Molecular Thermodynamic
Models for Fluids of Chain-Like
Molecules, Applications in
Phase Equilibria and Micro-Phase
Separation in Bulk and
at Interface
Honglai Liu, Ying Hu, Xueqian Chen, Xingqing Xiao
and Yongmin Huang

2. Lattice Based Molecular Thermodynamic Model of 156
Polymer Systems
2.1 General framework 159
2.2 Athermal entropy of mixing 162
2.3 Residual Helmholtz energy of mixing for 163
multicomponent Ising mixture
2.4 The residual Helmholtz energy of dissociation 166
and association of polymer chains
2.5 Helmholtz energy of mixing of polymer systems 167
2.6 Comparisons with molecular simulation results 167
2.7 Equation of state for polymer systems based on 171
lattice fluid model
2.8 Applications for calculations of phase equilibria 173
3. Density Functional Theory of Lattice Based Polymer 176
Adsorption
State Key Laboratory of Chemical Engineering and Department of Chemistry, East China University of
Science and Technology, Shanghai 200237, China
ISSN 0065-2377, DOI 10.1016/B978-0-12-380985-8.00004-X All rights reserved
153
154 Honglai Liu et al.
3.1 General formalism 178

3.2 Lattice density functional for a homopolymer 179
solution
3.3 Polymer adsorption at solid–liquid interface 182
4. Meso-Structures of Polymer Melts Confined in Curved 184
Surfaces
4.1 Monte Carlo simulation for diblock copolymers 187
confined in curved surfaces
4.2 SSL theory for diblock copolymers confined in 192
ring-like curved surfaces
4.3 Application to phase separation of diblock 199
copolymer confined in two curved surfaces
4.4 Remarks on the phase separation of confined diblock 209
copolymer
5. Conclusions 210
List of Symbols 212
Acknowledgments 213
References 213
Abstract Molecular thermodynamic models based on lattice framework

have been widely applied to study the thermodynamic properties
and the phase behaviors of chain-like fluids. Recently, we have
developed a new molecular thermodynamic model by combin-
ing statistical mechanics theory with computer simulation. The
effects of branching, coordination number, chain stiffness, com-
position, hydrogen bonding and pressure on thermodynamic
properties and phase behaviors can be well described by the
new model. Satisfactory agreement is obtained between the
predicted results and Monte Carlo (MC) simulation data for mul-
ticomponent Ising and Flory–Huggins lattice systems. The model
can be used to satisfactorily correlate phase equilibria including
vapor–liquid and liquid–liquid equilibria for the mixtures of ordi-
nary fluids, polymers, and ionic liquids. Incorporated with density
functional theory (DFT) for nonuniform fluids and weighted-den-
sity approximation (WDA), the model can also be used to
describe the adsorption of polymer at solid–liquid interface
and conformation distributions at interfacial regions. For the
morphologies of micro-phase separation of diblock copolymers
confined in curved surfaces, a framework has been developed
based on the strong segregation limit (SSL) theory. The SSL pre-
dictions agree well with simulation and experimental results on
multilayer transitions. Upon comparison between theoretical
predictions and MC simulations, we have established a numerical
calculation method of the Helmholtz energy for a special struc-
ture called the complex multilayered sector column (CMSC)
Molecular Thermodynamic Models 155
structure. Finally, the detailed theoretical studies together with

simulations indicate that the CMSC structure tends to be formed
at higher thickness, while the mutually competing concentric
cylinder barrel and the sector column structures appear at
lower thickness.
1. INTRODUCTION
A large number of new functional materials used in engineering

practice have unique meso-scale structures related to the processes over
a wide range of length and time scales. These meso-scale structures
determine the properties of functional materials. To design and control
material manufacturing, it is crucial to understand the formation
mechanisms of the meso-scale structures and to investigate the effects
of different engineering factors such as the changes of flow field, tem-
perature gradient, and external field on the structures. For example,
copolymers typically with multi-scale structures can be used as tem-
plates to prepare various complex materials (Bates, 1991; Park et al.,
1997; Xia et al., 1999; Li and Huck, 2002). To realize the regulation of
the meso-scale structures of copolymer materials, there are three
fundamental questions. What are the physical conditions for micro-
phase separation to form materials with desired compositions and
meso-scale structures in different domains? How are the meso-scale
structures evolved during preparation period? What is the composition
or density profile at interface, which might determine the stability
and properties of the material? The first question can be solved by
an equilibrium thermodynamic theory (e.g., an equation of state
(EOS) or a Helmholtz energy model)(Prausnitz et al., 1999), or by com-
puter simulation such as Monte Carlo (MC) and molecular dynamics
(MD) (Sadus, 1999). The second question involves the dynamics
of meso-structure evolution that can be examined by field-based theo-
ries (e.g., time-dependent Ginzburg–Landau equation (TDGL)(Chaikin
and Lubensky, 1995), self-consistent-field theory (SCFT)(Matsen and
Barrett, 1998), dynamic density functional theory (DDFT)(Fraaije,
1993), and the cell dynamics system method (CDS))(Oono and Shiwa,
1987; Oono and Puri, 1988), or by coarse-graining simulations (e.g.,
dissipative particle dynamics (DPD)(Groot and Warren, 1997), kinetic
Monte Carlo (KMC)(Graham and Olmsted, 2010), Langevin dynamics
(LD) (Pankavich et al., 2009), and lattice Boltzmann (LB) method)
(Song et al., 2008). The last question can be addressed by density func-
tional theory (DFT)(Wu, 2006) and SCFT (Scheutjens and Fleer, 1979),
also by simulations (Chen et al., 2010; Feng et al., 2005).
In our laboratory, we have developed a multi-scale theory to study

above questions based on an off-lattice model for polymer materials
(Liu et al., 2008). We first developed a molecular thermodynamic model
(EOS and Helmholtz energy model) for square-well fluids (SWCF-EOS)
based on the statistical mechanics theory of association. Combining
SWCF-EOS with the Langevin equation and adopting a weighted-
density approximation (WDA), we then developed a DDFT based on
equation of state (EOS-based DDFT). Finally, this SWCF-EOS with
WDA was applied for polymers at interfaces. The key of the theory is
the development of the EOS or Helmholtz energy model for homoge-
neous fluid at a molecular level. The molecular parameters in model can
be obtained by the regression of experimental vapor pressure and/or
pVT data of pure substance. This EOS or Helmholtz energy model can
be used to accurately describe vapor–liquid equilibria (VLE), liquid–
liquid equilibria (LLE), and solid–liquid equilibria (SLE) for nonpolar,
polar, and associating small molecules, as well as polymers and ionic
liquids (Hu et al., 1996a; Liu and Hu, 1998; Peng et al., 2001; Peng et al.,
2002; Peng et al., 2003; Li et al., 2009a; Li et al., 2009b; Li et al., 2009c). The
EOS-based DDFT can describe the meso-scale structures of polymers
and their evolution dynamics. The effect of pressure on meso-phase
separation can be successfully predicted (Xu et al., 2007a; Xu et al., 2007b;
Xu et al., 2008a). DFT could describe the density distribution and molec-
ular conformation of polymer at interface (Cai et al., 2002; Zhang et al.,
2004; Ye et al., 2005; Ye et al., 2006; Chen et al., 2006a; Ye et al., 2007;
Chen et al., 2007; Chen et al., 2008).
In this review, we introduce another approach to study the multi-
scale structures of polymer materials based on a lattice model. We first
show the development of a Helmholtz energy model of mixing for
polymers based on close-packed lattice model by combining molecular
simulation with statistical mechanics. Then, holes are introduced to
account for the effect of pressure. Combined with WDA, this model
of Helmholtz energy is further applied to develop a new lattice DFT to
calculate the adsorption of polymers at solid–liquid interface. Finally,
we develop a framework based on the strong segregation limit (SSL)
theory to predict the morphologies of micro-phase separation of diblock
copolymers confined in curved surfaces.
2. LATTICE BASED MOLECULAR THERMODYNAMIC MODEL

OF POLYMER SYSTEMS
In a lattice model, molecules are assumed to be arranged regularly on an

array of sites or cells as shown in Figures 1 and 2, each polymer chain
consists of r segments that occupy a series of r successive sites, each

solvent molecule has one segment. A polymer system contains Nr sim-
ple cubic lattice sites, each site can be occupied by one segment of
polymer or a solvent molecule. In close-packed lattice, there is no empty
site as shown in Figure 1 corresponding to an incompressible polymer
solution. There are interaction energies between nearest-neighbor
segment-pair symbolized as e. If the segment numbers of all molecules
are equal to 1, it is the Ising lattice. For realistic lattice fluids, in addition
to the segments of polymer and solvent molecules, there are vacancies
as shown in Figure 2 corresponding to a compressible polymer solution.
[(Figure_1)TD$IG]
Figure 1 Close-packed lattice model.
[(Figure_2)TD$IG]
Figure 2 Lattice fluid model.

The number of vacancies is related to the pressure of the polymer

system.
People have recognized the problems associated with lattice mod-
els over half a century. The most prominent one is that it is difficult to
obtain concise and accurate analytical expressions for Helmholtz
energy of mixing and other thermodynamic properties. Even for
the simplest Ising lattice, only the one-dimensional (1D) and two-
dimensional (2D) lattice can be rigorously solved on the basis of
statistical mechanics. The three-dimensional (3D) lattice has yielded
so far to rigorous analysis only by way of series expansion (Hill,
1956). An alternative way other than traditional statistical mechanical
derivation might be needed. Toward this end, introducing molecular
simulation into statistical mechanics to establish models seems to be
useful (Hu and Liu, 2006). In off-lattice approaches, a good example
is the well-known Carnahan–Starling equation for hard-sphere fluids
(Carnahan and Starling, 1969). The equation was established by a
linear combination of the PY pressure equation and compressibility
equation; the coefficients were judged by simulation results. The
EOS for hard-sphere chain fluids developed in our laboratory is
another example (Hu et al., 1996a). Based on the sticky-point model
of Cummings, Zhou, and Stell (Cummings and Stell, 1984; Stell and
Zhou, 1989; Zhou and Stell, 1992), an expression of residual
Helmholtz energy in terms of cavity correlation function (CCF) was
established where the nearest-neighbor CCF was derived from
rigorous Tildesley–Streett equation (Tildesley and Streett, 1980)
inducted from simulation of hard dumbbells, the next-to-nearest-
neighbor CCF was obtained by fitting simulation of trimers. A con-
cise EOS similar to the Carnahan–Starling equation was finally
obtained for hard-sphere chain fluids. To develop the Helmholtz
energy model of mixing for polymer systems based on lattice, we
adopted the similar approach, that is, combining molecular simula-
tion with statistical mechanics.
A wide variety of theories have been developed for polymer solu-
tions over the later half of the last century. Among them, lattice
model is still a convenient starting point. The most widely used
and best known is the Flory–Huggins lattice theory (Flory, 1941;
Huggins, 1941) based on a mean-field approach. However, it is
known that a mean-field approximation cannot correctly describe
the coexistence curves near the critical point (Fisher, 1967; Heller,
1967; Sengers and Sengers, 1978). The lattice cluster theory (LCT)
developed by Freed and coworkers (Freed, 1985; Pesci and Freed,
1989; Madden et al., 1990; Dudowicz and Freed, 1990; Dudowicz
et al., 1990; Dudowicz and Freed, 1992) in 1990s was a landmark.
Similar to Mayer’s theory for nonideal gases, they developed a dou-

ble expansion in power series with respect to the reciprocals of coor-
dination number z and temperature T. However, because of the
complexity of the expansion, lengthy equations are involved even
for truncation at the first order or at the second order, its practical
usage is largely limited. For some cases, the results are unsatisfactory,
for example, the predicted critical compositions of binary polymer
solutions exhibit unrealistic kink in contradictory with simulation
(Madden et al., 1990; Yan et al., 1996; Panagiotopoulos and Wong,
1998). Also the predictions do not match well with simulated for
ternary polymer systems (Jiang et al., 1997). Hu et al. (Hu et al.,
1991a; Hu et al., 1991b; Hu et al., 1996b) developed a revised Freed
Theory (RFT) by using an effective chain-insertion probability for
entropy and a series expansion for energy. The former is used to
improve the mean-field Flory–Huggins entropy term, while the latter
is adopted to account for interactions between more than two seg-
ments. The coefficients in the model are determined by a few simu-
lation data and referred to as a rigorous LCT. The agreement with
simulation for spinodals and binodals is excellent, much better than
the classical Flory–Huggins theory. The comparisons with calculated
binodals and critical coordinates by LCT for binary and ternary sys-
tems over a wide range of chain lengths indicate that the two theories
give almost same results. Lambert et al. (Lambert et al., 1993) and Bae
et al. (Chang et al., 1998) have made similar improvements. Chen
et al. (Chen et al., 2000; Chen et al., 2005) extended RFT to random
copolymer solutions, Bae et al. (Chang and Bae, 2003; Oh and Bae,
2010a) extend their model to polymer blend systems.
Recently, we developed a new close-packed lattice model for poly-
mer systems (Yang et al., 2006a; Yang et al., 2006b; Xin et al., 2008a) based
on Zhou–Stell theory (Stell and Zhou, 1989; Zhou and Stell, 1992). This
model was extended to random copolymer systems (Xin et al., 2008b)
and used to develop an EOS based on lattice fluid model (Xu et al.,
2008b; Xu et al., 2009a).
2.1 General framework

We start from a simple cubic lattice system containing Nr sites with a
coordination number z = 6. The lattice is filled with K components of
chain molecules. Ni and ri are the number of molecules and chain length
of component i, respectively. ri = 1 means that thePith component is a
solvent. The total number of lattice sites is N r ¼ Ki¼1 N i ri . Only the
nearest-neighbor interactions are considered.
According to Gibbs–Helmholtz relationship, the Helmholtz energy
of mixing of system can be obtained by integrating the internal energy
of mixing against the reciprocal of temperature:
Z1=T
Dmix F=T ¼ ðDmix F=TÞ1=T!0 þ ðDmix UÞdð1=TÞ ð1Þ
0
(DmixF/T)1/T!0 is the Helmholtz energy of mixing of system at temper-

ature approaching to infinity. The interactions between segments have
no effect on the thermodynamic properties at infinite temperature,
(DmixF/T)1/T!0 is the negative of the athermal entropy of mixing
DmixS0. The second term on the right-hand side of Equation (1) is
the residual contribution of the Helmholtz energy of mixing (DFr)
due to the interactions between segments that include the internal
energy of mixing and the entropy arose from the interactions between
segments. Equation (1) can be further expressed as
Dmix F ¼ TDmix S0 þ DFr ð2Þ
To obtain the residual Helmholtz energy of mixing DFr, we design a

three-step process as shown in Figure 3 (Xin et al., 2008c): (1) dissociate
the pure chains to form pure monomers; (2) mix monomers and solvent
to form an Ising mixture; and (3) associate the monomers into chain
molecules. The residual Helmholtz energy of mixing DFr can be
expressed as
DFr ¼ DFr1 þ DFr2 þ DFr3 ð3Þ
[(Figure_3)TD$IG]
Figure 3 Sketch for the mixing process of chain-like molecular systems (Xin et al.,
2008c).
The second term on the right-hand side of Equation (1) is the resid-
ual Helmholtz energy of mixing for an Ising mixture (DmixFrIsing). Its
expression will be discussed below.
For the contributions of step (1) and step (3) to the residual
Helmholtz energy of mixing (DFr1 + DFr3), the bond energies in-
volved in the dissociation step (1) and the association step (3) are
mutual compensated; therefore, they need not be considered in this
scheme. However, their environments in the dissociation step (1)
and the association step (3) are different, the entropy changes arisen
from the dissociation and association of polymer chains are then
different. In the step (1), dissociation occurs in pure substance, the
environment has no change before and after dissociation, the
entropy change is thus zero. While in the step (3), the association
of polymer chains occurs in mixtures, the monomer of Ising mixture
should be first arranged in the same conformation of polymer solu-
tion then bonded into chain-like molecules, the entropy is arisen
from this configuration change of the Ising mixture. The formalism
of chemical association through the CCF y in our previous work
(Hu et al., 1996a) based on the sticky-point model of Cummings,
Zhou, and Stell (Cummings and Stell, 1984; Stell and Zhou, 1989;
Zhou and Stell, 1992) can be employed. The total residual Helmholtz
energy of mixing in the step (1) and step (3) is given by (Yang et al.,
2006a; Xin et al., 2008a)
X
K
ðri Þ
DFr1 þ DFr3 ¼ kTN i lngi ð4Þ
i¼1
where g(r2) is the r-particle correlation function.

Currently, it is formidable, if not impossible, to derive an explicit
expression for g(r2) solely by statistical mechanics. In off-lattice
space, Zhou and Stell (Zhou and Stell, 1992) adopted a linear ap-
proximation and simplified the r-particle CCF by using the nearest-
neighbor two-particle CCF. In our previous work for hard-sphere
chain fluids, besides the nearest-neighbor two-particle CCF, we used
the next to nearest correlations determined by simulation (Hu et al.,
1996a). In this work, we follow the similar approach. If we adopt
Kirkwood’s superposition approximation, we can write g(r2) =
(g(2))r21, where g(2) is the radial distribution function. However, this ap-
proximation neglects the long-range correlations. We then introduce a
parameter l into the exponential to account for the long-range cor-
relations beyond the close contact pair, gðr2 Þ ¼ ðgð2Þ Þr2 1þl , where
g(2) and l are to be determined. Then, Equation (4), the contribution
of dissociation and association of polymer chains (DmixFrChain), is
rewritten as
Dmix Frchain DFr1 þ DFr1 X

K
ðri 1 þ li Þ ð2Þ
¼ ¼ fi lngi ð5Þ
N r kT N r kT i¼1
ri
Finally, we have the expression for the Helmholtz function of

mixing DmixF,
Dmix F Dmix S0 Dmix FrIsing Dmix Frchain

¼ þ þ ð6Þ
N r kT Nr k N r kT N r kT
2.2 Athermal entropy of mixing

Among various theories for the athermal entropy of mixing
DmixS0, Staverman and Guggenheim’s athermal entropy of mixing
(Guggenheim, 1952) is proved to be the best by comparison with sim-
ulation (Hu and Liu, 2006; Yang et al., 2006a). As shown in Figure 4
for chemical potentials of an athermal binary mixture with r1 = 1 and
r2 = 4, and in Figure 5 for the generalized plot for the probabilities
of 1–1 pairs versus surface fraction for athermal binary mixtures of
r1 = 1 and various chain lengths r2, the predictions from Staverman–
Guggenheim’s theory are perfect, and much better than that from the
Flory–Huggins theory. Therefore, we adopt Staverman–Guggenheim’s
athermal entropy of mixing (Guggenheim, 1952) for DmixS0,
Dmix S0 Dmix SGuggenheim X

K
fi zX K
q ui
¼ ¼ ln fi þ fi i ln ð7Þ
Nrk Nrk i¼1
ri 2 i¼1 ri fi
[(Figure_4)TD$IG]
Figure 4 Chemical potentials of an athermal mixture with r2 ¼ 4.

[(Figure_5)TD$IG]
Figure 5 Probabilities of 1–1 pairs of athermal mixtures with different chain lengths
(Yang et al., 2006a).
where, z is the coordination number of the lattice. fi and ui are the

P of component i, respectively,
volume fraction and surface fraction P which
can be calculated by fi ¼ N i ri = Kj¼1 N j rj and ui ¼ N i qi = Kj¼1 N j qj , qi is
the surface area parameter defined as qi = [ri(z 2) + 2]/z.
2.3 Residual Helmholtz energy of mixing for multicomponent

Ising mixture
The Ising lattice is the simplest lattice model in which each molecule
occupies a single site. Nevertheless, as mentioned above, due to math-
ematical difficulties, a rigorous analysis of the 3D Ising lattice has so far
achieved only by means of series expansion (Hill, 1956). When the chain
length is equal to 1, the Flory–Huggins theory and the Freed’s LCT can
also be used in Ising lattice (Hu et al., 1991a; Hu et al., 1991b; Hu et al.,
1996b). Other theories for the Ising lattice are Bragg–Williams approx-
imation (Bragg and Williams, 1934), Bethe-approximation (Bethe and
Wills, 1935), and Kikuchi’s cluster variation method (Kikuchi, 1951),
which have been employed to investigate the phase behavior of solid
alloys. All these mean-field theories show comparable accuracy with
the Flory–Huggins theory for the Ising lattice.
For a binary Ising lattice, we introduced a nonrandom factor that
was observed from simulation to have a linear relation with composi-
tion. The characteristic parameter of the linear relation was found by
combining a series expansion and the infinite dilution properties. On
this basis, an accurate expression for the Helmholtz energy of mixing
was developed, which can accurately reproduce the critical point and
coexistence curve of Ising lattice (Hu and Liu, 2006; Yan et al., 2004). This
model has been extended to multicomponent Ising mixture (Yang et al.,
2006ca). The predicted internal energy of mixing for ternary and qua-
ternary systems match accurately with simulation results. The pre-
dicted liquid–liquid phase equilibria for ternary systems are in nearly
perfect agreement with simulation results, and substantially improved
against the Flory–Huggins theory and the LCT.
For a simple cubic Ising lattice with a total Nr sites is occupied by K
types
PK of molecules, eachP molecule occupies one site, the constrains
K
i¼1 N i ¼ N r and 2N ii þ j6¼i N ij ¼ zN i are satisfied, where Ni is the
number of molecules of component i, Nij is the number of i–j contact
pairs, and z = 6 is the coordination number of the lattice. The internal
energy of mixing of the Ising system can be calculated by
1X K X K
Dmix U ¼ N ij 2ij ð8Þ
4 i¼1 j¼1
where 2ij = eii + ejj 2eij is the exchange energy between components i
and j. To calculate Nij, a nonrandom factor fij is defined by Nij/2Nii =
fijxj/xi, where xi = Ni/Nr is the mole fraction of component i. Then, we
get the number of i–j contact pairs
XK
N ij ¼ zN r xi xj f ij = xf
k¼1 k ik
ð9Þ
with fii = 1. By substituting Equation (9) into Equation (8), the internal
energy of mixing is
zN r X
K XK XK
Dmix U ¼ xi xj f ij 2ij = xf
k¼1 k ik
ð10Þ
4 i¼1 j¼1
The nonrandom factor fij characterizes the degree of deviation from

ideal mixing, and its numerical value can be estimated directly by
simulation. It is shown that 1/fij has a fairly well linear relation with
mole fraction. For binary Ising lattice, the expression of fij was obtained
by combining simulation and statistical mechanics (Yan et al., 2004). For
the multicomponent Ising lattice, a generalized expression has been
proposed as
X ~
K
eij þ e~ik e~jk
1=f ij ¼ xk exp ð11Þ
k¼1
2
eij ¼ b2ij is the reduced exchange energy with b = 1/kT.

where ~
The Helmholtz energy of mixing can be obtained by substituting

Equation (11) into Equation (10) and integrating Equation (10) with
Gibbs–Helmholtz equation against the reciprocal of temperature. In
the integration, we first expand it to a polynomial of the reciprocal of
temperature for ternary Ising system (K = 3), then integrate and finally
obtain the expression of Helmholtz energy of mixing by Gibbs–
Helmholtz equation. The complete expression is very long and compli-
cated even truncated at the third order of the reduced energy. For
simplification and practical use, the expression is truncated at the sec-
ond order of the temperature, with a constant c introduced to maintain
accuracy. Finally, we have
Dmix F X
K
zX K X K
z XK X K
¼ xi lnxi þ eij
xi xj~ e2ij
xi xj ~
N r kT i¼1
4 i¼1 j¼1
16 i¼1 j¼1
0 12
zX K X K X K
cz @X K X K
þ eij~eik
xi xj xk ~ eij A
xi xj~ ð12Þ
8 i¼1 j¼1 k¼1 16 i¼1 j¼1
By reproducing the critical point of binary Ising system, we get c = 1.1.

Figure 6 shows the coexistence curves for binary Ising lattice predicted
by this model, the Flory–Huggins model, and the Freed model. The
predicted results of our model are in nearly perfect agreement with
simulation data (Yan et al., 1996). Figure 7 is the 3D diagram for the
internal energy of mixing for a ternary Ising system with ~ e12 ¼ 0:4,
e13 ¼ 0:3, and ~e23 ¼ 0:2. All the simulation data (open squares) are
~
distributed on the curved surface calculated by Equation (10).
[(Figure_6)TD$IG]
Figure 6 Coexistence curves of binary Ising lattice (Yan et al., 2004).

[(Figure_7)TD$IG]
Figure 7 Internal energy of mixing for a ternary Ising lattice (Yang et al., 2006ca).
P
By subtracting the Helmholtz energy of ideal mixing ( Ki¼1 xi ln xi ),
Equation (12) can be used to calculate the residual Helmholtz energy of
mixing (DmixFrIsing) of monomer mixture of polymer system in step 2 of
Figure 1. However, the mole fraction xi should be replaced by volume
fraction fi. We have
Dmix FrIsing zX K X K
z X K X K
¼ fi fj~eij ff~e2
N r kT 4 i¼1 j¼1 16 i¼1 j¼1 i j ij
0 12
z XK X K X K
cz X
K XK
þ fff ~ eij~eik @ ff~ eij A ð13Þ
8 i¼1 j¼1 k¼1 i j k 16 i¼1 j¼1 i j
2.4 The residual Helmholtz energy of dissociation

and association of polymer chains
The residual Helmholtz energy due to the dissociation of polymer
chains in pure state and the association of polymer chains in mix-
ture state can be calculate by Equation (5). The pair correlation
functions of component i in Pthe corresponding Ising lattice system
ð2Þ
are calculated by gi ¼ 1= Kj¼1 fj f ij (Liu et al., 2007). The residual
Helmholtz energy of dissociation and association of polymer chains
is then
0 1
Dmix Fchain
r XK
ðri 1 þ li Þ XK
¼ fi ln@ fj f ij A ð14Þ
N r kT i¼1
ri j¼1
2.5 Helmholtz energy of mixing of polymer systems

Finally, by substituting Equations (7), (13), and (14) into Equation (7),
we obtain the Helmholtz energy of mixing for multicomponent poly-
mer systems
Dmix F X
K
fi zX K
q ui z X
K X
K
¼ lnfi þ fi i ln þ f f ~eij
N r kT i¼1
ri 2 i¼1 ri fi 4 i¼1 j¼1 i j
z XK X K
zX K X K X K
fi fj~e2ij þ fff ~ eij~eik
16 i¼1 j¼1 8 i¼1 j¼1 k¼1 i j k
0 12 0 1
cz @X K X K X K
ðr 1 þ l Þ XK
f f ~eij A þ fi ln@ fj f ij A ð15Þ
i i

16 i¼1 j¼1 i j i¼1
ri j¼1
The parameter l characterizing the long-distance correlations was

determined by MC results of critical temperatures and compositions for
two binary polymer systems with chain lengths r1 = 1, r2 = 4 and 200
(Yan et al., 1996; Panagiotopoulos and Wong, 1998). The long-range
correlations for a branched polymer are apparently different from
those of a linear polymer with the identical molecular weight, although
they both have r 1 neighboring pairs. Similar to Hawker et al.
(Hawker et al., 1991), a parameter Db = (N? + Nh 2)/r has been intro-
duced to characterize the degree of branching (Yang et al., 2006b), where
N? represents the number of ways in which three bonds meet up at a
lattice site (Nemirovsky et al., 1987); Nh is the number of head units of a
polymer chain. By this definition, the degree of branching of linear
polymers is naturally equal to zero. The increase in the degree of
branching Db enhances the segment contacts and the long-range corre-
lations. Finally, the parameter l can be calculated by
zðri 1Þðri 2Þð1 þ Dbi Þ
li ¼ ð0:1321ri þ 0:5918Þ ð16Þ
6r2i
This equation can be used for polymers of arbitrary Db in a lattice

with arbitrary z.
2.6 Comparisons with molecular simulation results

The above molecular thermodynamic model for polymer systems
has been widely tested by comparing with simulation results
(Yang et al., 2006a; Xin et al., 2008a). Figure 8 shows the comparisons
between predicted critical temperature and critical volume fraction
for binary polymer solutions at different chain lengths of with the
[(Figure_8)TD$IG]
Figure 8 Chain-length dependence of the reduced critical temperature and the

critical volume fraction. Square and triangle: MC data; solid line: this work; dot-dashed
line: this work with l ¼ 0; dash line: Flory–Huggins’s theory; dotted line: Freed theory
simulated results of Yan et al. (Yan et al., 1996) and Panagiotopoulos

and Wong (Panagiotopoulos and Wong, 1998). Figure 9 shows the
comparisons between simulated coexistence curves (Yan et al., 1996;
Panagiotopoulos and Wong, 1998) with predictions of our model for
binary polymer solutions with chain lengths up to r2 = 600. Figure 10
shows the comparison between simulated spinodals (Rodriguez et al.,
1992) with predictions by various theories for systems with r2 = 18 and 60.
[(Figure_9)TD$IG]
Figure 9 Coexistence curve of binary polymer solutions with different chain lengths.
From bottom to top: the chain lengths are 1, 2, 4, 8, 16, 32, 64, 100, 200, and 600,
respectively.
[(Figure_0)TD$IG]
Figure 10 Coexistence curve of binary polymer solutions with chain length r2 ¼ 18

and 60. Open squares: MC data; solid line: this work; dash line: Flory–Huggins theory;
dotted line: Freed theory.
Figure 11 shows the coexistence curves of branched polymers L, B2,

and D with r2 = 65 and different architectures. The predictions from
this model agree very well with MC data (Arya and Panagiotopoulos,
2005), particularly at high-polymer volume fractions. Figure 12 illus-
trates the comparisons between calculated binodal curves by the RFT
(Chen et al., 2005), this work (Xin et al., 2008b), and corresponding
simulation data (Chen et al., 2000; Chen et al., 2005) for random
[(Figure_1)TD$IG]
Figure 11 Coexistence curves of branched polymer solutions with r2 ¼ 65 and

different architectures (Yang et al., 2006b). MC data: squares (L), crosses (B2), triangles
(D); solid line: this work; dotted line: LCT.
[(Figure_2)TD$IG]
Figure 12 Coexistence curves for lattice random copolymers with different chain
compositions eAA:eAB:eBB ¼ 1.0:0.8:0.6, rs ¼ 1, rp ¼ 32(Xin et al., 2008b). Solid line: this
work; dotted line: RFT.
copolymer systems where the chain length is held fixed at 32 but the
chain composition varies, the employed interaction energy parameters
are eAA:eAB:eBB = 1.0:0.8:0.6. Figure 13 shows the predicted LLE of
ternary polymer solutions with type 1 and type 2 phase separation.
The MC data are taken from (Jiang et al., 1997; Xin et al., 2008a; Liu et al.,
2007). The liquid–liquid phase equilibria of ternary chain-like mixtures
predicted by this model are in good agreement with MC simulation
[(Figure_3)TD$IG]
Figure 13 Liquid–liquid equilibria phase diagrams of ternary polymer solutions

(Xin et al., 2008a). Open circles: the simulated results; dotted lines: Flory–Huggins;
short-dot lines: RFT; solid lines: this work.
[(Figure_4)TD$IG]
Figure 14 Normalized internal energy of mixing for a binary polymer solutions with
r2 ¼ 64(Yang et al., 2006a) and 100. MC data: kT/2 ¼ 4 (open square); kT/2 ¼ 10 (open
triangle); kT/2 ¼ 10 (open diamond); solid lines: this work; dotted line: Freed theory;
dash line: Flory–Huggins theory.
results. All three types of phase separations of Treybal classification

can be described satisfactorily. Figure 14 shows the comparisons
between simulated internal energies of mixing zDmixU/2Nref1f2
(Yang et al., 2006a) with those predicted by various theories for systems
with r2 = 64 and 100. The predicted critical temperatures and critical
compositions, spinodals and coexistence curves as well as the internal
energies of mixing for systems with various chain lengths are in satis-
factory agreement in comparison with simulation results.
2.7 Equation of state for polymer systems based on lattice

fluid model
To account for the volume (pressure) effect, a lattice fluid model
based on the Flory–Huggins theory was first proposed by Sanchez
and Lacombe (Sanchez and Lacombe, 1976; Lacombe and Sanchez,
1976) by assuming complete randomness in the distribution of mole-
cules and holes on the lattice. Hu and coworkers (Hu et al., 1992)
adopted a two-step mixing process to introduce holes to the RFT
and established another lattice fluid model. Recently, Shin et al.
(Shin and Kim, 2006) extended a new quasichemical nonrandom lat-
tice fluid model to describe VLE of mixtures. Panayiotou et al.
(Panayiotou et al., 2004) developed a nonrandom hydrogen-bonding
model, which was applicable to various systems including nonpolar
systems and highly nonideal systems with strong specific interac-
tions, and Tsivintzelis et al. (Tsivintzelis et al., 2006) derived new
analytical expressions for the nonrandomness factor, which was used
to describe nonrandomness in mixtures, and compared it with others
corresponding expressions.
[(Figure_5)TD$IG]
Figure 15 Process of two-step mixing (Xu et al., 2008b).
To extend a close-packed lattice model Equation (15) to a lattice fluid

model, we adopt a two-step process as shown in Figure 15 to establish
an EOS (Hu et al., 1992). In the first step, pure chain molecules at close-
packed lattice are mixed to form a close-packed mixture. In the second
step, the close-packed mixture is mixed with N0 holes to form an
expanded realistic system with volume V at temperature T and pressure
p. According the two-step process, the Helmholtz energy of mixing can
be expressed as
Dmix F ¼ Dmix FI þ Dmix FII ð17Þ
Following (Hu et al., 1992), the same model is applied for both steps.
In the first step, the Helmholtz energy of mixing DmixFI is calculated by
Equation (15). In the second step, the close-packed mixture is consid-
ered to be a pseudopure substance ‘‘a’’, its average segment number ra,
and the segment–segment
PK interaction PK parameter eaa are esti-
PKenergy
mated by r1 a ¼ i¼1 fi =ri and eaa ¼ i¼1 j¼1 u i u j eij with eij = (1 kij)
(eiiejj)1/2. kij is an adjustable parameter and can be correlated from
experimental data. Then, N0 holes are mixed with pseudopure sub-
stance ‘‘a’’, and the Helmholtz energy of mixing DmixFII in this step is

Dmix FII 1 ~ Þlnð1 r
r~ ~ z
~ Þ þ lnr
¼ ð1 r þ
N r kT ~
r ra 2

qa qa
ð1 r ~ Þln½1 þ ðqa 1Þr ~ þ r ~ ln
ra 1 þ ðqa 1Þr ~

z cz 1 þ la 1 þ ð1 r ~ ÞD
þ ð1r ~
~ Þr ð1 r ~ 2 ra
~ Þ2 r ~ ln
r ð18Þ
2T~ 4T~2 ra 1 þ ð1 r ~ Þr
~D
where D ¼ expð1=TÞ ~ 1, la is the parameter characterizing the long-

range correlations between monomers in the pseudopure substance ‘‘a’’
beyond the close contact pairs and can be obtained by Equation (16).
Using classical thermodynamics, we obtain the EOS for mixtures (Xu
et al., 2008b; Xu et al., 2009a)

~ z z 1 z 2 cz
~ ~
p ¼ T lnð1 rÞ þ ln ~
1 rþ1 r ~ ð19Þ
2 2 ra 2 ~
4T
ra 1þ la ~ 2 ½1þð1 r ~ ÞD2 1

~ 4r
ð3r
4
~ þr
~ Þþ
3 2
~
Tr
ra ~ ÞD½1þð1 r
½1þð1 r ~ Þr
~ D
The reduced temperature T,~ reduced pressure ~ p, and reduced den-

~ ¼ kT=eaa , ~
~ are defined by T
sity r p ¼ pn =eaa , and r
~ ¼ N r n =V, respec-
tively. Where n is the hard-core volume of a site or a segment.
2.8 Applications for calculations of phase equilibria

Applying the lattice model to practical systems, we have to introduce
physically meaningful temperature dependence for the energy param-
eter due to the oriented interactions between segments. The double-
lattice model previously proposed by the authors can provide this
relationship (Hu et al., 1991a; Hu et al., 1991b; Hu et al., 1996b). For each
type of i–i, j–j, and i–j segment–segment pairs, we imagine a secondary
lattice to account for the additional Helmholtz energy originated from
various possible arrangements of the oriented and nonoriented inter-
actions. For i–j pairs, the total number of sites in the corresponding
secondary lattice Nrij relates to the number of i–j pairs Nij by the con-
servation relation, Nrij z/2 = Nij. This secondary lattice is an Ising lattice
in which the number of sites available for oriented interactions is Nrijhij,
while the remaining Nrij(1 hij) cannot participate in oriented i–j inter-
actions as shown in Figure 16.
[(Figure_6)TD$IG]
Figure 16 A schematic representation of double lattice model for oriented

interactions between molecules i and j (Hu et al., 1991b).
The additional Helmholtz energy responsible for this secondary

lattice can be expressed by using Equation (12) for binary Ising mix-
ture with x1, x2 replaced by hij and (1 hij). Finally, we obtain the
temperature-dependent interchange energy that is quadratic to the
inverse temperature.

~eij ¼ 2ij =kT ¼ de1ij þ deð2Þij =kT þ deð3Þij =ðkTÞ2 =kT ð20Þ
The lattice fluid molecular thermodynamic model described above

has been used to calculate phase equilibria including VLE and LLE
for systems containing ordinary fluids, polymers, and ionic liquids.
Figure 17 shows the calculated spinodal curves and liquid–liquid co-
existence curves of PS/cyclohexane systems with temperature-
independent energy parameters (Yang et al., 2006a). Figure 18 shows
the calculated liquid–liquid coexistence curves of tert-butyl acetate/PS
and water/poly(ethylene glycol) systems with temperature-dependent
[(Figure_7)TD$IG]
Figure 17 Spinodal curves and coexistence curves of PS/cyclohexane systems

[(Figure_8)TD$IG]
Figure 18 Coexistence curves of tert-butyl acetate/PS and water/poly(ethylene

glycol) systems (Yang et al., 2006cb).
[(Figure_9)TD$IG]
Figure 19 Vapor–liquid equilibria for system of propanol þ [Me3BuN][NTf2] and

liquid–liquid equilibria for system of [Rnmim][PF6] þ Butan-1-ol (Yang et al., 2006d).
energy parameters (Yang et al., 2006cb). Figure 19 shows the VLE for
x1{propanol} + (1 x1){[Me3BuN][NTf2]} and LLE for x2{[Rnmim][PF6]}
+ (1 x2){Butan-1-ol}(Yang et al., 2006d). Combining the Flory’s
Gaussian chain model for the elastic contribution due to the cross-link
between polymers, this model was also used to calculate the swelling
behavior of temperature- and/or solvent-sensitive hydrogels in pure or
mixed solvents (Huang et al., 2008; Zhi et al., 2010). Figure 20 is the
swelling curve (swelling ratio, SR) of PNIPAm gels in ethanol/water
mixed solvents at different temperatures.
The EOS based on the lattice fluid model has also be used to describe
thermodynamic properties such as pVT behaviors, vapor pressures and
liquid volumes, VLE and LLE of pure normal fluids, polymers and ionic
[(Figure_0)TD$IG]
Figure 20 Swelling ratio of PNIPAm gels in ethanol/water mixed solvents (Zhi et al.,
2010).
[(Figure_1)TD$IG]
Figure 21 Phase diagrams for mixtures of CHF3 and [C4mim][PF6](Xu et al., 2009b) and
the LLE for cyclohexane (1) + 2-butanone (2) + [C6mim][PF6] (3) mixture at 298.15 K.
liquids, and their mixtures (Xu et al., 2008b; Xu et al., 2009a; Xu et al.,
2009b). Usually, the parameters in an EOS are obtained by correlating
experimental pVT behavior, vapor pressure, and liquid volume data.
For the VLE of binary mixtures, only one adjustable binary interaction
parameter was used. The solubility of gas in polymer and ionic liquid
can be calculated up to a high pressure. The VLE and LLE of ternary
systems containing ionic liquid can be accurately predicted. Figure 21
shows the isothermal pTx phase diagrams for mixtures of CHF3 and
[C4mim][PF6] and LLE binodal curves for cyclohexane (1) + 2-butanone
(2) + [C6mim][PF6] (3) mixture at 298.15 K.
3. DENSITY FUNCTIONAL THEORY OF LATTICE BASED

POLYMER ADSORPTION
Polymer adsorption at interface plays a key role in many traditional

technical fields such as paints, coatings, surface lubricants, ceramics,
and adhesives, as well as many emerging areas including self-assem-
bly of functional nanostructures. The layers formed by polymer
adsorption have rich structural features because long flexible mole-
cules can adopt a large number of conformations resulting in signif-
icant entropic effect. Adsorption of a flexible macromolecule onto an
impenetrable surface causes a competition between the reduction in
conformational entropy and the energetic compensation of favorable
binding. The counterbalance of this competition dominates the molec-
ular conformations and properties of adsorbed polymers. Using dif-
ferent approximations, a series of theoretical models for polymer
solutions have been developed based on lattice or off-lattice model.
The lattice model provides a simple but effective method for fluids
where the packing effects are not important. Neglecting the packing
effects of segments near surface, the adsorption behavior of polymer
is exclusively determined by the coupling of conformational entropy
and segmental interaction energy.
A number of lattice-based theories have been developed for poly-
mer adsorption (Scheutjens and Fleer, 1979; Simha et al., 1953; Ash
et al., 1970; Helfand, 1975; DeMarzio and Rubins, 1971; Scheutjens and
Fleer, 1980; de Gennes, 1980a). Among them, a well-known adsorp-
tion theory was proposed by Scheutjens and Fleer (SF) (Scheutjens
and Fleer, 1979; Scheutjens and Fleer, 1980). By approximating the
partition function of polymer solutions at interface and adopting a
matrix method (DeMarzio and Rubins, 1971), they successfully
extended the Flory–Huggins theory (Flory, 1941; Huggins, 1941) to
predict the adsorption profiles of polymer near solid surface. The
mean-field approximation adopted is reasonable at high densities
but invalid at low densities (Hu et al., 1991a; Hu et al., 1996b;
Janssen and Nies, 1997). Another approach to polymer adsorption is
based on integral equation theory that directly describes the correla-
tions of polymer segments with adsorbing wall (Janssen and Nies,
1997). By neglecting packing effects, Janssen et al. used the discretized
polymer–reference interaction site model to calculate the adsorption
profiles. With an appropriate closure relation and approximation of
the intramolecular two segments distribution function, they obtained
improved predictions. To describe the structural and thermodynamic
properties of inhomogeneous fluids, DFT is a robust method with the
Helmholtz energy density functional of system as its starting point.
Over the last two decades, DFT has been successfully applied to
inhomogeneous polymeric systems, and most of studies have been
focused on the continuum free-space model (Cai et al., 2002; Zhang
et al., 2004; Ye et al., 2005; Ye et al., 2006; Chen et al., 2006a; Ye et al.,
2007; Chen et al., 2007; Chen et al., 2008; Woodward et al., 1991; Kierlik
and Rosinberg, 1992; Yethiraj and Woodward, 1995; Zhou and Zhang,
2001; Yu and Wu, 2002; Patra and Yethiraj, 2000). In the mean time,
lattice density functional theory (LDFT) has been proposed for the
Ising lattice and for the adsorption of lattice gas. The pioneer work of
LDFT is the Ono–Kondo equation (Ono and Kondo, 1960) based on
the mean-field approximation. More rigorous treatment is to connect
the Helmholtz energy functional with the nonideal interactions
through the classical approximation in analogy to the continuum
counterpart (Nieswand et al., 1993; Reinhard et al., 2000; Prestipino
and Giaquinta, 2003). Ono–Kondo LDFT has been extended to inves-
tigate lattice fluids with various types of molecular structures and
intermolecular interactions, including dimers (Aranovich et al., 1999;
Aranovich et al., 2000; Chernoff et al., 2002; Chen et al., 2006b). The
extension of LDFT to polymer adsorption has been a challenge.
Recently, we presented a LDFT for polymer solutions on a 3D lattice
on the basis of an analytical expression of ideal Helmholtz energy
density functional for polymer systems (Chen et al., 2009). The excess
contributions in the Helmholtz energy density functional are con-
structed by adopting the close-packed molecular thermodynamic
model for polymer solutions described above with either local density
approximation (LDA) or WDA. The LDFT developed takes into account
the intrinsic energy due to the constraints of connected segments, as
well as the site exclusion originated from the excluded-volume effect,
the attractive interactions between the nearest neighbors, the long-
range correlations from the nonbonded interactions, and their cou-
plings with intrinsic energy. At a special condition, the theory can be
used to obtain the equilibrium segment-density distribution for a poly-
mer at interface. By including the correlations between polymer seg-
ments due to chain connectivity and attractive interaction, the theory
not only predicts the density profiles well but also provides the infor-
mation of adsorption conformation.
3.1 General formalism

In the lattice representation of a polymer solution, each polymer seg-
ment or solvent molecule occupies one lattice site, while the system is
regarded as a binary mixture of polymer and solvent. The Helmholtz
energy of system can be expressed as
F ¼ Fid þ DFex ð21Þ
Fid and DFex are the ideal Helmholtz energy functional and excess
Helmholtz energy functional, respectively. The ideal Helmholtz energy,
as a functional of rp(Q) and rs(q), can be written as
^ i bTSid
bFid ½rp ðQ Þ; rs ðqÞ ¼ bhX
H
id
X
¼b rp ðQÞVint ðQÞ þ rp ðQÞlnrp ðQÞ
XQ Q
þ rs ðqÞlnrs ðqÞ ð22Þ
q
Here subscripts p and s represent polymer and solvent, respectively, Q

is the coordinate of all polymers with configurations inherited and q is
that of all solvent molecules, Vint(Q) is the intrinsic energy function. If
all the potential terms in addition to the intrinsic energy of r connected
segments are taken into account, the real Helmholtz energy functional
of the system can be expressed by
bF½rp ðQÞ; rs ðqÞ bFid ½rp ðQÞ; rs ðqÞ þ bDFex ½rm ðqÞ; rs ðqÞ ð23Þ
The excess part is approximated as a functional of solvent density

distribution rs(q) and the average segment-density distribution
rm(q). The latter is related to polymer density rp(Q) via a transform as
XX
r
rm ðqÞ dðqj qÞrp ðQÞ ð24Þ
Q j¼1
with rm(q) + rs(q) = 1 for incompressible polymer solution, where d is a

Kronecker delta, d(x) d0x.
3.2 Lattice density functional for a homopolymer solution

We consider a simple cubic lattice with a coordination number z = 6.
For an incompressible polymer solution, each lattice site is occupied
by a solvent molecule or by a segment of polymer chain. The attrac-
tion interactions between the nearest-neighbor sites are character-
ized by a reduced exchange energy ~ e ¼ bðepp þ ess 2eps Þ between
a segment p and a solvent s, where eij is the attractive energy of an
i–j pair.
3.2.1 Excess Helmholtz energy functional

To construct the excess Helmholtz energy functional DFex for inhomo-
geneous fluids, the close-packed molecular thermodynamic model for
homogeneous bulk fluids described by Equation (15) is used by com-
bining the LDA and nonlocal WDA. The excess Helmholtz energy
functional is then separated into the athermal entropy of mixing and
the internal energy of mixing,
DFex ¼ Dmix Father þ Dmix Finter ð25Þ
Comparing Equation (25) with Equation (23), the approximate den-

sity functional of athermal part in Equation (25) can find its counterpart
in Equation (23) and is assumed to be
inhomogeneous
bDmix Father ! bDFex
ather ½rm ðqÞ; rs ðqÞ ð26Þ
where
X
ather ¼
bDFex fch ðrm ðqÞ; rs ðqÞÞ ð27Þ
q
This athermal excess Helmholtz energy functional can be esti-

mated via the Staverman–Guggenheim athermal entropy of mixing
(Guggenheim, 1952). fch is a function of ri (i = m, s), representing the
segment–segment (intrachain or interchain) and segment–solvent vol-
ume repulsions in athermal solution,
z har a a a i
fch ðrm ; rs Þ ¼ rm ln r r rm þ rs ln r rm þ rs ð28Þ
2 r r r r
where ar = [(z 2)r + 2]/z. Because the intersegment volume repulsions
are reflected by the ‘‘site exclusion’’ within a single lattice site, the LDA
can be used to calculate rm. The approximate density functional of the
internal energy of mixing in Equation (25) can also find its counterpart
in Equation (23). However, the WDA should be adopted. The functional
is assumed to be
inhomogeneous X
bDmix Finter ! bDFex
inter ¼ rm ðqÞf attr ðrm ðqÞ; rs ðqÞÞ ð29Þ
q
where ri ðqÞ (i = m, s) is the coarse-grained (weighted) density

defined as
X 0 0
rm ðqÞ wðjq qjÞrm ðq Þ; rs ðqÞ ¼ 1 rm ðqÞ ð30Þ
0
q
fattr(rm, rs) = fam(rm, rs) + fac(rm, rs), where fam and fac stand for the
contributions from the attractive interactions and the coupling effects
between energetic correlation and chain connectivity, respectively. The
expressions of the two functions can be obtained through the Helmholtz
energy for uniform fluids

z 1 2 1 3
f am ðrm ; rs Þ ¼ ~ers ~e rm rs ~e rm rs ð1 2rm rs Þ
2 2
ð31Þ
2 2 6
r 1 þ la eÞ 1rs þ 1
½expð~
f ac ðrm ; rs Þ ¼ ln ð32Þ
r ½expð~eÞ 1rm rs þ 1
where la is a factor accounting for the long-range correlations of poly-

mers and is calculated by Equation (16).
To construct the functional for inhomogeneous fluids and account
for the correlations, the coarse-grained densities rather than the local
densities should be used. For convenience, we use the Heaviside step
function to estimate the weighted densities. In this work, the attraction
exists between nearest neighbors only; therefore, the weighting sum-

mation constrained by the Heaviside step function Q merely runs over
the sites adjacent to site q and itself. The weighting function is therefore
given by wðxÞ Qðx 1Þ=ðz þ 1Þ.
3.2.2 Grand potential and equilibrium density distribution

For the lattice model, the grand potential V can be written as
X
V¼Fþ p ðQÞ mp rp ðQÞ
V ext ð33Þ
Q
Pr
p ðQÞ ¼
where mp is the chemical potential and V ext j¼1 n ðqj Þ is the
ext
external potential exerted on all segments. The variational principle

gives that the extremum of grand potential corresponds to the equilib-
rium state of the system, which results in
dV
¼0 ð34Þ
drp ðQÞ
where the functional derivative is equivalent to a partial derivative with
respect to rp(Q) in the discrete condition. Substituting Equation (33)
into Equation (34) and using Equation (23), the LDFT equation for
equilibrium distribution rp(Q) can be obtained as

rp ðQÞ ¼ exp bmp bVint ðQÞ bC ðQÞ ð35Þ
where
X
r
bC ðQÞ ¼ b’ðqj Þ
j¼1
" #
Xr
1 dbDFex
¼ 1 lnrs ðqj Þ þ þ next ðqj Þ ð36Þ
j¼1
r drm ðqj Þ
dbDFex dbDFex dbDFex

¼ ather
þ inter
drm ðqÞ drm ðqÞ drm ðqÞ
@fch @fch
¼ þ f ðr ðqÞ; rs ðqÞÞ
@rm ðqÞ @rs ðqÞ attr m
X
@f attr @f attr
þ rm ðq0 Þwðjq0 qjÞ ð37Þ
q0
@rm ðq0 Þ @rs ðq0 Þ
Substituting Equation (35) into Equation (24), we finally have the

LDFT equation to determine the segment-density distribution as
XX
r
rm ðqÞ ¼ dðqj qÞexp½bmp bVint ðQÞ bC ðQÞ ð38Þ
Q j¼1
In order to solve Equation (38), we have to know the expression of the

intrinsic energy function Vint(Q). This function has no contribution to
the Helmholtz energy in bulk, thereby it must satisfy the normalizing
condition
X
exp½bV int ðQÞ ¼ 1 ð39Þ
Q
For freely jointed flexible chain, the intrinsic energy can be defined by
the r-mer Mayer function of an ideal chain,
Y
r1 dðjq
jþ1 qj j 1Þ
exp½bVint ðQÞ ¼ ð40Þ
j¼1
z
From Equation (40), Equation (38) can be expressed by

X
r
rm ðqÞ ¼ exp bmp b’ðqÞ GLð jÞ ðqÞGRð jÞ ðqÞ ð41Þ
j¼1
ð jÞ ð jÞ
Here, the left and right propagator functions, GL ðqÞ and GR ðqÞ, are
calculated by the following recursive relations
ðjÞ
X dðjq0 qj 1Þ 0 ðj1Þ 0
GL ðqÞ¼ exp½b’ðq ÞGL ðq Þ ð42Þ
q0
z
ðjÞ
X dðjq0 qj 1Þ 0 ðjþ1Þ 0
GR ðqÞ¼ exp½b’i ðq ÞGR ðq Þ ð43Þ
q0
z
ð1Þ ðrÞ
where j = 1, 2, , r 1, and GL ðqÞ ¼ GR ðqÞ ¼ 1.
3.3 Polymer adsorption at solid–liquid interface

The LDFT equation near a planar solid surface can be rewritten as
rb X
r
ðjÞ ðjÞ
rm ðkÞ ¼ exp½br’b b’ðkÞGL ðkÞGR ðkÞ ð44Þ
r j¼1
where ’b = ’( 1 ), that is, the value of ’ with the bulk density rb, and
ðjÞ
the propagator function GLðRÞ ðkÞ can be calculated by Equation (42) or
Equation (43).
The segment-density distributions of train, loop, and tail form can be
calculated by LDFT,
train rtrain ¼ rm ð1Þ ð45Þ
rb X
r
ðjÞ ðr þ 1 jÞ
loop rloop ðkÞ ¼ exp br’b b’ðkÞ GA ðkÞGA ðkÞ ð46Þ
r j¼1
2rb Xr
ðjÞ ðr þ 1 jÞ
tail rtail ðkÞ ¼ exp br’b b’ðkÞ GA ðkÞGF ðkÞ ð47Þ
r j¼1
ðjÞ
where GA ðkÞ can be calculated by Equation (42) or Equation (43)
ð1Þ ðjÞ ðjÞ
with the condition GA ðkÞ ¼ dðk 1Þ and GA ð1Þ ¼ GðjÞ ð1Þ. GF ðkÞ =
ðjÞ ðjÞ
G ðkÞ GA ðkÞ.
The LDFT performs quite well at high densities and the deviations
from simulation become distinct when the chain connectivity is impor-
tant at low densities, long-chain polymers, or low temperatures.
Figure 22 plots the normalized surface coverage r(1) versus the bulk
density of an athermal (upper line) and thermal (lower line) 30-mer
polymer from the LDFT. The symbols denote the results from MC
simulation of (Janssen and Nies, 1997). It is seen that the rb dependence
[(Figure_2)TD$IG]
Figure 22 Normalized surface coverage versus the bulk density of 30-mer polymer
from LDFT (Chen et al., 2009).
[(Figure_3)TD$IG]
Figure 23 Total segment-density distributions of thermal 40 mers. rb = 0.108, 0.311,

0.51 from bottom to top (Chen et al., 2009).
of r(1) is reasonably accurate at all bulk densities, especially at rb > 0.5.

At low density (rb < 0.5), the overestimation of LDFT is somewhat
distinct. Figure 23 shows comparisons between theoretical prediction
and MC data for the total segment-density distributions of a thermal
case with r = 40.
4. MESO-STRUCTURES OF POLYMER MELTS CONFINED

IN CURVED SURFACES
Considerable attention has been paid to the self-aggregation of diblock

copolymers primarily due to their ability to form periodical spatial
meso-structures such as the body-centered cubic, the hexagonal, and
the lamellar structures Both theoretical and experimental approaches
have demonstrated that the morphologies of block copolymer melts are
essentially controlled by molecular architecture, polymerization index,
composition, external fields, and many others. These micro-phase
structures can be used as templates to prepare duplicated nanomater-
ials. For instance, Hashimoto’s group (Mita et al., 2008) has produced a
polymer template with the macroscopic orientation of hexagonally
packed cylinders by imposing a moving temperature-gradient field.
Rider (Rider et al., 2008) have migrated polystyrene-b-poly(ferroceny-
lethylmethylsilane) diblock copolymers into silica colloidal crystals and
inverse silica colloidal crystals.
The formation mechanism of micro-phase structures for block copo-
lymers essentially roots in a delicate balance between entropic and
enthalpic contributions to the Helmholtz energy. Several theories have
been developed. Leibler (Leibler, 1980) calculated the mean-field ther-

modynamic potential by using the random phase approximation. The
Strong Segregation Limit (SSL) theory, originally proposed by Helfand
and Wasserman (Helfand and Wasserman, 1976), was further devel-
oped by Ohta and Kawasaki (OK) using this approach (Ohta et al., 1986;
Ohta and Kawasaki, 1990). The key point in OK theory is that the
Helmholtz energy of copolymer melt with micro-phase separation
can be divided into two parts: one is contributed by the short-range
interaction and another by the long-range interaction, that is, the enthal-
pic and entropic contribution, respectively. Moreover, the Helmholtz
energy density of the enthalpic contribution was assumed to be propor-
tional to the Flory–Huggins interaction parameter between the immis-
cible blocks, while that of the entropic contribution is scaled to the
polymerization index inversely. Later, Fredrickson and Helfand
(Fredrickson and Helfand, 1987) extended SSL theory by taking account
of fluctuation effects. Following OK theory, SSL theory was improved
further to deal with simple graft and star copolymers (Anderson and
Thomas, 1988), and calculate the phase diagrams of ABC triblock copo-
lymers (Zheng and Wang, 1995). By utilizing the analytical calculation,
instead of the density expansion method in OK theory, Semenov
(Semenov, 1989; Semenov et al., 1996) studied both micellization and
phase separation in the SSL case. Likhtman (Likhtman and Semenov,
1994; Likhtman et al., 1999) extended Semenov’s theory for the stability
of ordered bicontinuous double diamond structure and the surface
deformations of polymeric brushes in solution.
The external potentials or geometry confinements greatly influence
the morphologies of block copolymer melts. The lamellar phase of AB
diblock copolymer melts confined between two flat plates has been
studied by Turner (Turner, 1992). He found that for surfaces with weak
selectivity, either parallel or vertical lamellar structures can be formed
depending on the extent of frustration between film thickness and bulk
lamellar period. This model was further improved by Walton et al.
(Walton et al., 1994) to calculate the Helmholtz energy of both sym-
metrical and asymmetrical thin films. They found that a critical layer
number exists, and above which only the parallel morphology exists
symmetrically; whereas, below this number, either vertical or parallel
lamellar symmetrical structures can be predicted depending on the
deformation of chain. The experiments on the P(S-b-MMA) thin films
confirmed their predictions (Lambooy et al., 1994; Kellogg et al., 1996).
More recently, many studies have been reported on the micro-phase
separation of block copolymers with multidimensional confinements
including cylindrical and spherical shapes. Shin and Xiang’s group
(Shin et al., 2004; Xiang et al., 2004; Xiang et al., 2005a) observed the
phase behavior of polystyrene-block-polybutadiene by transmission

electron microscopy (TEM). Via a capillary action, the diblock copoly-
mer was drawn into nanoporous alumina membranes to produce free-
standing nanorods with a variety of pore diameters. The similar way
was also adopted by Sun et al. (Sun et al., 2005) to prepare polystryrene-
block-poly(methyl methacrylate) nanorods, in which the diameter-
dependence of the morphologies was systematically investigated by
varying the pore diameters of templates from 400 nm down to 25 nm.
More experimental results on the cylindrical confinement of block copo-
lymers were reported in (Xiang et al., 2005b; Ma et al., 2006). More
researches focused on simulation and the theoretical analysis (He
et al., 2001; Sevink et al., 2001; Feng and Ruckenstein, 2006a; Feng and
Ruckenstein, 2006b; Chen et al., 2006c; Chen and Liang, 2007; Li et al.,
2006; Li and Wickham, 2006; Sevink and Zvelindovsky, 2008; Xiao et al.,
2007; Zhu and Jiang, 2007; Yu et al., 2006; Yu et al., 2007a; Yu et al., 2007b;
Yu et al., 2007c; Wang et al., 2008a; Wang et al., 2008b; Wang, 2007; Han
et al., 2008). A wide variety of morphologies for block copolymers
confined in a cylinder such as the so-called ‘‘multibarrel-layer,’’ ‘‘dart-
board,’’ or ‘‘concentric cylinder barrel’’ was predicted by simulation
(He et al., 2001) and DDFT (Sevink et al., 2001). As consistent with
experimental results, Yu et al. (Yu et al., 2006; Yu et al., 2007a; Yu et
al., 2007b; Yu et al., 2007c) systematically investigated the morphologies
of AB diblock copolymers under multidimensional confinements via
the lattice MC simulation. Inspired from Wang’s work (Wang, 2007), a
simple model for the layer thickness of concentric lamellae in both 2D
and 3D confined systems was proposed for symmetrical diblock copo-
lymers (Yu et al., 2007b; Yu et al., 2007c), where the accurate layer
thicknesses was calculated based on the lattice model and compared
with Wang’s theoretical prediction. In a ring-like curved confinement,
Han et al. (Han et al., 2008) studied the effect of disperse index on the
morphology of diblock copolymers and found that such effect acts
directly on the formation of concentric cylinder structures.
In general, if a ring-like curved surface selects one of the blocks
strongly, parallel lamellar structures (or concentric-ring barrel struc-
tures) will occur. If the surface is neutral or weakly preferential,
however, more complex structures such as the sector column and the
multilayer sector column will form. Theoretically, the molecular ther-
modynamic model described in Section 2 can be used to calculate the
Helmholtz energy of this system, and the equilibrium complex struc-
tures can be obtained by minimizing the Helmholtz energy. However,
the interfacial Helmholtz energy should be accounted in this case due to
the existence of interfaces among different meso-domains. In this work,
the Semenov’s approach is adopted to deduce the Helmholtz energy of
AB diblock copolymers confined in ring-like curved surfaces. Both

strong and weak preferences to different block from the curved surfaces
will be discussed. Especially, the Helmholtz energy confined in a nano-
pore can be obtained once the interior radius approaches to zero. For a
strong preferred surface in a nanopore, to predict structural lamellar
transition, MC simulations are carried out to investigate symmetrical
and asymmetrical concentric cylinder barrel structures, and the
Helmholtz energy profiles are plotted as a function of dimensionless
radius. However, for a weak preferred surface in a nanopore, the dis-
cussions are mainly done for symmetrical parallel lamellar and sector
column phases. Comparing this graph with simulation results, we
found a conflict of compatibilities caused by neglecting other possible
morphologies. Consequently, the topological morphology of complex
multilayered sector column (CMSC) structure is extracted theoretically
from MC simulation.
4.1 Monte Carlo simulation for diblock copolymers confined

in curved surfaces
In order to develop a theoretical method for describing the meso-
structures of diblock copolymer confined in curved surfaces, MC sim-
ulation was first used to find the possible phase separation structures of
diblock copolymer melt.
4.1.1 Diblock copolymers confined in cylindrical pore

4.1.1.1 Strong preference to block copolymer. A cylindrical pore with a
size of Rex Lz was used in the simulation, where Rex is the exterior
radius of pore and Lz = 50 is the length of pore. The periodic boundary
condition was applied in the axial direction eAB = 0.5kBT is the
interaction energy between A and B segments, while eAS = eBS =
1.0kBT is the interaction energy between the block and the exterior
pore surface. Those interaction energy parameters suggest that the
interaction between blocks and B is repulsive, the surface has strong
attractive interaction with block A and strong repulsive interaction with
block B. The volume concentration of the A5B5 diblock copolymer is
90%.
In order to determine the thickness of each layer for the concentric
cylinder barrel structure, the radial order parameter, c(r) = hfA(r)
fB(r)i/hfA(r) + fB(r)i, was introduced, where fA(r) is the density of
A blocks at r, and < > denotes the ensemble average. c(r) > 0 means the
rich A layer at r and c(r) < 0 otherwise. Figure 24a and b plot c(r) with
Rex/L0 = 1.39 and 1.80 corresponding to Nlayer = 1.5 and 2, respectively.
It is shown that the symmetrical cylinder barrel in Figure 24b possesses
[(Figure_4)TD$IG]
Figure 24 c(r) profiles at varying Rex/L0: (a) Rex/L0 ¼ 1.39, Nlayer ¼ 3/2 and (b) Rex/
L0 ¼ 1.80, Nlayer ¼ 2.
the same fringe thickness at both outmost and innermost layer, and at
each middle layer, whether it belongs to A domain or not, nearly twice
the thickness of the fringe layers. As for the asymmetrical structure in
Figure 24a, the alternative ABAB concentric cylinder barrel can be
considered as a combination of symmetrical ABA and AB with half a
period. All the remarks can be confirmed through the data in Figure 24.
The morphology transition of multilayered structures can be quan-
titatively determined byc(r). Figure 25a reveals the relationship
between Nlayer and Rex/L0. It can be seen that symmetrical structure
occurs at about Rex/L0 = n, while asymmetrical structure at about
Rex/L0 = n + (1/2), where n is an integer. Before or after these transition
points, the excess relaxation or compression in each layer space causes
the deviation from the characteristic period L0. The circles with sparse
bias in Figure 25a indicate the transition points at Rex/L0 = 1.14 and 1.72,
corresponding to Nlayer = 1 and 1.5, respectively. Due to the perturbed
frustration between Rex and L0 before these transition points, a signif-
icant deformation, instead of the perfect circle, occurs in the center
phase of the pores, as shown in Figure 25b and c. In Xiang’s experiments
(Xiang et al., 2004), the deformation in the center phase for either PS or
PBD was also observed by TEM, which agrees well with our simulation.
4.1.1.2 Weak preference to block copolymer. Figure 26 shows the MC

simulated morphologies of A5B5 diblock copolymers confined in
cylindrical nanopores with the different exterior radii Rex and eAS =
eBS = 0. It is shown that the lamellar structure forms parallel to the
pore axis. However, there is a little difference between small and
large Rex. The lamellar structure is not bended at small Rex as shown
in Figure 26a, while it is bended like the wave-shape at large Rex as
[(Figure_5)TD$IG]
Figure 25 (a) For concentric cylinder barrel, Nlayer versus Rex/L0 is plotted. The
circles with sparse bias mark the transition points at Rex/L0 ¼ 1.14 and 1.72,
respectively. Parts b and c are the snapshots corresponding to Nlayer ¼ 1 and 1.5,
respectively.
shown in Figure 26b–e. There are four small collective phase regions
approximated by the concentric square column structure phase. The
statistical average values of gyration radius of diblock copolymer
chains show that the plane (x, y) component is more than 4.0 and the
perpendicular component is nearly 1.0. This indicates that AB diblock
copolymer chains are relatively compressed in the z direction, whereas
they are relatively stretched and ellipse-like along the plane (x, y).
Additionally, the up-and-down periodical change in gyration radius
of diblock copolymer chains with Rex is observed, which implies that
even though an ellipse-like packing along the plane (x, y), AB diblock
copolymer chains would also shrink and spread alternately with the
increase of Rex. It is consistent with the prediction by SSL theory
shown later.
[(Figure_6)TD$IG]
Figure 26 Micro-phase morphologies of A5B5 diblock copolymer melts in cylindrical

nanopore as the neuter exterior surface when varying the exterior radius Rex. eAB ¼
0.5kBT; eAS ¼ eBS ¼ 0. Red: A block; blue: B block. (a) Rex ¼ 9; (b) Rex ¼ 30; (c) Rex ¼ 35; (d)
Rex ¼ 40; (e) Rex ¼ 45.
4.1.2 Diblock copolymers confined between two concentric

curved surfaces
When the confined surfaces suffer from a weak interaction with block
copolymer, either parallel or vertical lamellar structures for AB diblock
copolymer systems under flat and curved confinements could exhibit,
as shown in Figure 27. From theoretical predictions (Turner, 1992;
Walton et al., 1994) and simulations (Wang et al., 2000; Yin et al., 2004),
the frustration between d and L0 could result in the alternative appear-
ance of parallel lamellar and vertical lamellar structures under flat
confinements. A question is naturally arisen: can both concentric cylin-
der barrel and sector column structures appear under the curved
confinement?
The morphologies of AB diblock copolymers confined between
two concentric curved surfaces with exterior radii Rex and interior radii
Rin have been investigated via MC simulations in our previous work
[(Figure_7)TD$IG]
Figure 27 Schematic illustrations of lamellar morphologies for AB diblock

copolymers under flat and curved confinements. The concentric cylinder barrel
structure under curved confinement corresponds to the parallel lamellar structure
under flat confinement (left). The sector column structure corresponds to the vertical
lamellar structure (right).
(Xiao et al., 2007). The results indicate that the regular parallel lamella
structure (the concentric cylinder barrel structure) appears if d and L0
are compatible in both flat and curved confinements, while a vertical or
distorted vertical lamellae structure (sector column structure) is
observed otherwise. Upon increasing the curvature of the exterior sur-
face K = 1/Rex, the compatibility span of d and L0 becomes smaller;
consequently, the formation of the parallel lamella structure (or the
concentric cylinder barrel structure) is more difficult. Our MC results
in this work are shown in Figure 28 and more details can be referred to
(Xiao et al., 2007).
The effect of curvature on the morphology transition of AB diblock
copolymers by MC simulation is illustrated in Figure 29 (correspond-
ing to Figure 8 in (Xiao et al., 2007)). When d/L0 = 0.735, d and L0 are
compatible from Figure 28a, a parallel lamellar structure occurs at
Nlayer = 1 and K = 0 as observed from Figure 29a. When K increases,
the incompatibility between d and L0 becomes more acute, and more
vertical lamellar domains along with the thickness emerge as shown
in Figure 29b–d. Figure 29c shows the typical coexisting vertical and
parallel lamellar structure of AB diblock copolymers. A series of
simulations prove that the distorted vertical lamellar structure, rather
than the perfect sector column and concentric cylinder barrel struc-
tures, forms more readily at the higher K because of its lower
Helmholtz energy, which is also classified as the incompatible region
in Figure 28b–c.
[(Figure_8)TD$IG]
Figure 28 Effect of d/L0(AB) (MC simulation result origins from Figure 9 in (Xiao et al.,
2007)) on morphology transitions of AB copolymers at different curvature K of (a) 0;
(b) 0.03125; (c) 0.04167; and (d) 0.0625. ‘‘inco’’ means ‘‘incompatible region,’’ while
‘‘co\n’’ is a compatible region between d and L0(AB) with Nlayer ¼ n.
[(Figure_9)TD$IG]
Figure 29 Morphologies of AB diblock copolymers confined in symmetry surfaces

with various K at d/L0(AB) ¼ 0.735. Red: A blocks; blue: B blocks. K: (a) 0; (b) 0.03125;
(c) 0.04167; and (d) 0.0625.
4.2 SSL theory for diblock copolymers confined in ring-like

curved surfaces
According to the MC simulation results mentioned above, if the diblock
copolymer melts are confined in cylindrical pore or between two con-
centric curved surfaces with strong preference to one of the blocks, the
concentric-ring barrel structures can be observed. However, if the in-

teractions between polymer block and surfaces are weak, the sector
column structures or more complex structures may appear. Those irreg-
ular complex structures can be considered as the combination of simple
regular phase separation structures. In order to theoretically describe
those structures, the Helmholtz energies of simple regular phase sepa-
ration structures are first derived by using SSL theory in the following
sections.
The SSL theory regards chemically distinct blocks to be completely
immiscible. The Helmholtz energy of the molten diblock copolymer
system consists of two contributions: one is the interfacial energy Fint
and the other the elastic energy of blocks Fst. In Semenov’s SSL theory
(Likhtman and Semenov, 1994; Semenov, 1985), the excess Helmholtz
energy per copolymer chain is simplified and expressed as:
Z
S0 3p2
F ¼ N s AB þ zðrÞ2 dr ð48Þ
V0 8f 2a Na2 V 0
½ a
where N is the polymerization degree of block copolymers, s AB is the

interfacial tension of A and B domains expressed by the Flory–Huggins
parameter x as s AB = a(x/6)1/2, V0 is the volume of a Wigner–Seitz cell,
S0 is the interfacial area per cell, fa is the volume fraction of a block and a
is the statistical bond length. The model system contains n monodis-
persed diblock copolymer chains confined in two concentric curved
surfaces with a size of Rex Rin Lz, where Lz is the length at z
direction, Rex and Rin are the exterior and interior radii. The thickness
is d = Rex Rin, and the curvature is K = 1/Rex. The symmetric exterior
and interior surfaces prefer to attract A segments but are repulsive to B
segments. s AS (or s BS) defines the interfacial tension between A domain
(or B domain) and the two surfaces. In this work, the average concen-
tration of segments r(=nN/V) is assumed to be unity.
4.2.1 Helmholtz energy of symmetrical parallel lamellar confined

in ring-like curved surfaces
Figure 30 shows the symmetrical concentric-ring barrel structure
(A–B–A)m confined between two ring-like curved surfaces (cylinders).
It is assumed that the fringe thickness is the same for both the exterior
and interior layers, while that of the middle layers is nearly twice in size
regardless of A domain or B domain. This assumption has been proved
to be valid via MC simulation mentioned above.
The length of the circular lamellar period is L. A variable d is set to
denote the thickness of the barrel structure, d = mL, m is the number of
periods. From Figure 30, the interfacial Helmholtz energy of a barrel
[(Figure_0)TD$IG]
Figure 30 Symmetrical concentric-ring barrel structure (left) confined in the ring-

like surface and a unit cell with one lamellar period L (right).
structure characterized by d or m per copolymer chain can be expressed

as
" #
Sym 1 X m
f f
Fint ¼ 2p RðlÞ L þ RðlÞ L þ L Lz
nc l¼1 2 2
2p ð49Þ
s AB þ ð2Rex dÞLz s AS
nc
pLz ð2Rex dÞ pLz ð2Rex dÞ
¼ 2ms AB þ 2s AS
nc nc
where nc is the number of copolymer chains, subscripts S and ex denote

the wall surface and the exterior surface, Lz is the length of cylinder, and
f is the asymmetrical parameter of diblock copolymers. Obviously,
f cancels out in Equation (49), which indicates that the asymmetry of
AB diblock copolymers has no effect on the interfacial Helmholtz
energy. Equation (49) can be further simplified with r = ncN/V.
Sym 2N
Fint ¼ ðms AB þ s AS Þ ð50Þ
rd
As for the elastic Helmholtz energy of a single chain, applying
Equation (48) to A block gives
Z
3p2 Lz X
m
FAst ¼ zðr Þ2 dr ð51Þ
8f 2 Na2 V l¼1
½SA ðlÞ
The conformational integral is calculated by

Z fL=2
R 2 fL
½SA ðlÞ zðrÞ dr ¼ 2p RðlÞ þ r r2 dr
0 2
Z fL=2
fL
þ 2p RðlÞ L þ r r2 dr
0 2
p f 3 L3
¼ ½2RðlÞ L ð52Þ
12
By combining Equation (52) with Equation (51), the elastic Helmholtz
energy of A block can be obtained as
p3 Lz fL3
st ¼
FA mð2Rex mLÞ ð53Þ
32Na2 V
Similarly, the elastic Helmholtz energy of B block can be obtained from
Equation (48). For a unit cell shown in Figure 30, the distortion can be
estimated by

f þ xl f
ð f þ xl Þ 2RðlÞ L ¼ 2RðlÞ L ð54Þ
2 2
where xl is the distortion coefficient of B block in the lth cell. Hereby,

the conformational integral is extended as
Z 3 3
x L f x 4 L4 L3 x 3
zðr Þ2 dr ¼ 2p l RðlÞ L l þ 1f l
24 2 64 3 2
½SB ðlÞ

f L4 x 4
RðlÞ L þ L þ 1f l ð55Þ
2 4 2
The elastic Helmholtz energy of B block can be written as

Xm Z
3p2 Lz
FBst ¼ zðrÞ2 dr ð56Þ
8ð1 fÞ2 Na2 V l¼1
½SB ðlÞ
By combining Equation (53) with Equation (56), the elastic

Helmholtz energy of AB diblock copolymer chain can be obtained.
Finally, after substituting Equations (50), (53), and (56) into
Equation (48), the excess Helmholtz energy FSym per copolymer
chain can be further simplified and formulated as a function of d

and m,
2N f p2 2
FSym ¼ ðms AB þ s AS Þ þ L þ FBst ð57Þ
rd 32Na2
4.2.2 Helmholtz energy of asymmetrical parallel lamellar confined

in ring-like curved surfaces
Figure 31 shows the structure of an asymmetrical concentric-ring barrel
confined in ring-like curved confinements. Different from the symmet-
rical one, the exterior and interior rings in the asymmetrical system are
not the same. In order to calculate the free energy, the asymmetrical ring
barrel is considered as (A–B–A)M–A–B, which is composed of a sym-
metrical multilayered structure and an asymmetrical one with half a
period. In this way, the Helmholtz energy FAsy can be separated into
two parts: one is FSym of the symmetrical multilayered structure and the
other is the Helmholtz energy of the asymmetrical one with half a
period. By setting d = ML + L/2, M is the number of symmetrical
periods. The symmetrical part FSym has been presented in above section.
Because the innermost cell of the cylinder is one A–B unit with half a
period L/2, the distortion coefficients of A and B blocks, f0 and x0 , must
accord with the following equations
0
!
4M þ 1 0 4M þ f
f 2Rex d ¼ f 2Rex d ð58Þ
2M þ 1 2M þ 1
and
0 0
f þx ¼1 ð59Þ
[(Figure_1)TD$IG]
Figure 31 Asymmetrical structure of concentric-ring barrel (left) and the inner cell
with half an period (right).
The interfacial contribution to the Helmholtz energy per copolymer

chain can be expressed as
" #
Asy 1 X M
fL fL
Fint ¼ 2p RðlÞ þ RðlÞ L þ Lz s AB
nc l1 2 2
" " 0
# #
1 fL
þ 2p RðMÞ L Lz
nc 2
2pLz
s AB þ ½ðs AS þ s BS ÞRex s BS d
nc !
0
2Ms AB N 2M 2s AB N 2M þ f
¼ 2Rex d þ R d
rdð2Rex dÞ 2M þ 1 rdð2Rex dÞ ex 2M þ 1
2N½ðs AS þ s BS ÞRex s BS d
þ
rdð2Rex dÞ
ð60Þ
Asy
The elastic Helmholtz energy Fst of AB diblock copolymer chain can
Sym ðAÞAsy ðBÞAsy
be expressed by a sum of Fst , Fst , and Fst as
Asy Sym ðAÞAsy ðBÞAsy
Fst ¼ Fst þ Fst þ Fst ð61Þ
Sym ðaÞAsy
where Fst is the contribution from the symmetric part, Fst denotes
the elastic Helmholtz energy of a block in the innermost cell. According
Sym
to Equations (53) and (56), Fst can be expressed as
Sym p2 fL3 BðSymÞ
Fst ¼ Mð2Rex MLÞ þ Fst ð62Þ
32Na2 dð2R ex dÞ
BðSymÞ BðSymÞ 3p
When M = 0, Fst ¼ 0; otherwise, Fst ¼
PM Ð 2 2
8ð1 fÞ Na2 dð2Rex dÞ
zðrÞ dr.
l¼1 ½S ðlÞ
B
In the A–B unit with half a period, the elastic Helmholtz energy of each
block can be evaluated by
" 0
! 03 04
#
ðAÞAsy 3p2 2M þ f f L3 f L4
Fst ¼ 2 Rex L þ ð63Þ
4f Na2 dð2Rex dÞ 2 24 64
and
" 0
! 03 04
#
ðBÞAsy 3p2 x x L3 x L4
Fst ¼ Rex d þ L ð64Þ
4ð1 fÞ2 Na2 dð2Rex dÞ 2 24 64
By combining Equations (62), (63), and (64) with Equation (61), the
elastic Helmholtz energy of AB diblock copolymer chain can be ob-
tained. Finally, we have FAsy by adding Equation (50) to Equation (51).
4.2.3 Helmholtz energy of sector column confined in ring-like

curved surfaces
As shown in Figure 32, the sector column phase can also be considered
as the distortion of the vertical lamellae phase in the flat plates system,
which is a counterpart to the above parallel lamellae. With the assump-
tion of the repeated number of basic A–B unit in the cylinder, and the
interval angle of basic unit, u ¼ 2p=M, M is the number of sectors, the
interfacial contribution per copolymer chain can be expressed as:
2MdLz s AB 1
int ¼
FSec þ ½2f pð2Rex dÞLz s AS þ 2ð1 fÞpð2Rex dÞLz s BS
n n
2MN s AB 2N
¼ þ ½s þ fðs AS s BS Þ
prð2Rex dÞ rd BS
ð65Þ
Similar to the abovementioned method, the entropic contribution of

single chain of sector column phase can be written as
Z
3p2 MLz
FAst ¼ 2 zðrÞ2 dr ð66Þ
2
4f Na V
0
½S A
[(Figure_2)TD$IG]
Figure 32 Sector column structure (left) and the unit cell with half period of A
domain (right).
R R f u=2 R Rex 3 2
with 0
½S A
zðrÞ2 dr ¼ 0 Rin R sin ’dRd’ ¼ ð1=16ÞðRex Rin Þðf u
4 4
sinf uÞ and the Taylor polynomial expansion, sin x = x x /3 ! + O(x3),

3
Equation (66) can be simplified as
pMðR2ex þ R2in Þf u3 p4 fð2R2ex 2Rex d þ d2 Þ

FAst ¼ ¼ ð67Þ
128Na2 16Na2 M2
The same way can be used to derive the entropic contribution of B
block.
The Helmholtz energy FSec of sector column phase in SSL is the sum
a
int and Fst . Furthermore, considering M = p(2Rex d)/L0 with L0 ¼
of FSec
2ðs AB =3rÞ1=3 ðaNÞ2=3 is the bulk lamellae period, we get the simplified
expression Equation (65) of sector column structure:
1=3
3N s 2AB 2N p2 3N s 2AB 1=3
FSec ¼ þ ½s þ fðs s Þ þ
r2 a2 rd BS AS BS
12 r2 a2
ð68Þ
2 2Kd þ K2 d2

4 4Kd þ K2 d2
In this work, we assume s BS = s AS. Equation (68) indicates that

FSec is a function of both thickness d and curvature K.
4.3 Application to phase separation of diblock copolymer

confined in two curved surfaces
The Helmholtz energy expressions derived above for symmetrical and
asymmetrical parallel lamellae (concentric-ring barrel) and sector col-
umn structures can be extended to flat surface systems. The compar-
isons between the curved and flat systems will not be introduced in
detail here; however, some conclusions can be shared as follows: the
concentric-ring barrel structure is the same as the parallel lamellar
essentially; the difference exists in the bending Helmholtz energy that
is zero in the flat confinement; when the curvature K increases, the
change of bending Helmholtz energy must not be neglected.
Consequently, it is well founded that the flat confinement is merely a
special case of the ring curved confinement with K ! 0.
4.3.1 Strong preference of the curved surfaces to block copolymer

The morphologies of polystyrene-b-polybutadiene (PS-b-PBD) diblock
copolymers confined in a nanopore were observed by Shin and Xiang
(Shin et al., 2004), in which a lot of attentions were paid to the layer
number of the concentric cylinder barrel structure as a function of the
nanopore radius. In this work, the symmetrical and asymmetrical
parallel lamellar mentioned in Section 4.2 are adopted to study this

problem.
In a nanopore confinement, the Helmholtz energy of both symmet-
rical and asymmetrical concentric cylinder barrel phases can be
obtained by taking d ! Rex in Equation (57) and Equation (60)–(64).
In order to compare with the above MC simulation quantitatively, the
same parameters were selected. In addition, the statistical bond length
in SSL (a = 1.29) was calculated by an ensemble-average bond length
over all the collected configurations in MC simulation.
Figure 33 plots the Helmholtz energy profiles of both symmetrical
and asymmetrical cylinder barrels along with Rex/L0. Both are similar to
the undulated wave. At a small Rex/L0, the fluctuations of Helmholtz
energy are intensified because of the bending. With increasing Rex/L0,
the two curves oscillate periodically, indicating the periodic adjustment
of the multilayer number. Meanwhile, the Helmholtz energy at a higher
Rex/L0 converges to the intrinsic state of the bulk phase. The corre-
sponding layer numbers Nlayer of the two curves are also plotted syn-
chronously, in which the transitions of both structures are indicated at a
[(Figure_3)TD$IG]
Figure 33 Helmholtz energy profiles and Nlayer versus Rex/L0. Line 1: asymmetrical
structure; line 2: symmetrical structure.
certain Rex/L0. Line 1 denotes the asymmetrical structure, while line 2

indicates the symmetrical one. It is obvious that the Nlayer transition of
line 1 occurs at n + (1/2), while that of line 2 merely at n, where n is an
integer. There are many cross-points between line 1 and line 2 in
Figure 33, which implies the mutual transition between the two
morphologies at a certain Rex/L0. By minimizing the Helmholtz energy
before and after the cross-points, a continuous profile of the total energy
can be plotted as a function of Rex/L0, and the transition of Nlayer along
with Rex/L0 in Figure 34. In the interval [0.4–5.5] of Rex/L0, the exposed
structure with higher frequencies is the asymmetrical cylinder barrel
with Nlayer = n + (1/2), which is very consistent with simulations shown
in Figure 25a.
For the direct comparison with our MC simulation, SSL theory and
the results of other simulations and experiment quantitatively, a
detailed summarization for Nlayer is listed as a function of Rex/L0 in
Table 1. There are several sources of different simulations and experi-
mental data taken from Wang (Figure 3 in (Wang, 2007)), Sun et al.
[(Figure_4)TD$IG]
Figure 34 Total Helmholtz energy profile and the corresponding layer transition are
plotted as a function of Rex/L0, which are obtained by minimizing free energy of two
kinds of the morphologies in Figure 33.
Table 1 Nlayer versus Rex/L0 in MC simulation, SSL theory and experiments
202
Rex/L0 Nlayer Rex/L0 Nlayer

MC SSL Wang’s Sun et al.’s Xiang et al.’s MC SSL Wang’s Sun et al.’s Xiang et al.’s
sim exp exp sim exp exp
0.43 0.500 0.32 0.5 2.93 3

Honglai Liu et al.
0.54 0.5 3.04 3

0.65 0.5 3.15 3
0.76 0.5 3.26 3
0.82 0.87 0.791 0.77 0.95 (Nlayer = 0.5) 1.0 3.37 3.5
0.90 0.98 1.067 1.0 3.48 3.5
0.98 1.09 1.0 3.59 3.5
1.06 1.0 3.70 3.5
1.14 1.0 3.80 3.5
3.91 3.5
1.23 1.20 1.344 1.30 (Nlayer = 1.0) 1.5
1.31 1.30 1.5 4.02 4.05 4
1.39 1.41 1.5 4.13 4
1.47 1.52 1.500 1.60 1.5 4.24 4
1.55 1.63 1.5 4.35 4
1.64 1.74 1.5
1.72 1.5
(Continued)
TABLE 1 (continued)
Rex/L0 Nlayer Rex/L0 Nlayer

MC SSL Wang’s Sun et al.’s Xiang et al.’s MC SSL Wang’s Sun et al.’s Xiang et al.’s
sim exp exp sim exp exp
1.80 1.85 2.0 4.46 4.5

1.88 1.96 2.0 4.56 4.5
1.96 2.06 2.0 4.67 4.5
2.04 2.17 2.0 4.78 4.5
2.13 2.0 4.89 4.5
5.00 4.5
2.21 2.28 2.31 2.5 5.11 5.13 5

2.39 2.5 5.22 5
2.50 2.5 5.33 5
2.61 2.5 5.43 5
2.72 2.5
2.83 2.5 ... ...
Molecular Thermodynamic Models
203
(Figure 4 in (Sun et al., 2005)) and Xiang et al. (Figure 1 in (Shin et al.,
2004) and Figure 2 in (Xiang et al., 2005a)). It is worth noticing that the
diameter D in their simulations and experiments is related to the dimen-
sionless diameter D/L0, rather than Rex/L0 employed in our definition.
The Nlayer values in the parentheses represent the experimental data
that are different from our calculations, and all other data without the
parentheses are identical to this work.
A significant consistency is observed between MC simulation and
SSL. Wang (Wang, 2007) also studied the layer transitions in nanopore
by MC simulation and strong stretching theory. Sun et al. (Sun et al.,
2005) reported the PS-PMMA diblock copolymers confined in a nano-
pore, Xiang et al. (Shin et al., 2004; Xiang et al., 2005a) investigated the
effect of curved confinement on the morphologies of PS-PBD diblock
copolymers. From Table 1, it is found that our predictions are more
closely consistent with Wang’s simulation results and Sun et al.’s
experimental data, but a relative deviation occurs from Shin et al.’s
and Xiang et al.’s data at the small Rex/L0. At Rex/L0 = 1.30, for
example, Nlayer = 1.5 in our SSL calculation, but Nlayer = 1.0 in Shin
et al.’s and Xiang et al.’s experiments for the concentric cylinder
barrel structure. One possible reason may lie in the inaccuracy in their
experiments (not a well-defined round nanopore at the small Rex/L0.
Finally, theoretical calculation is found to be very consistent with
simulation, and both give good predictions in comparing with exper-
imental results.
4.3.2 Weak preference of the curved surfaces to block copolymer

4.3.2.1 Mutual transformation between the competing morphologies.
Figures 35 and 36 show the Helmholtz energies of the concentric
cylinder barrel and sector column structures of diblock copolymer
melts confined in ring-like curved surfaces as a function of the
thickness d. Lines 1 and 2 in Figure 35 are the Helmholtz energies
of the concentric cylinder barrel structures for s AS = 0 and 0.05,
respectively. The lower Helmholtz energies are attributed to more
negative s AS that implies more stable micro-phase structures. Line
3 in Figure 35 is the Helmholtz energy of the sector column structures
at different K values, which reveals a rapid increase at large d. At a
small d, the Helmholtz energy of the sector column phase is mainly
dominated by interfacial energy. As d increases, however, the
Helmholtz energy is gradually dominated by conformational
entropy. As compared lines 1 and 2 with line 3 in Figure 35a and b,
there is a drastic competition occurring at a small d caused by
minimizing the Helmholtz energies of two structures. This indicates
a notable periodical transformation from the concentric cylinder
[(Figure_5)TD$IG]
Figure 35 Helmholtz energies predicted by SSL for AB diblock copolymers under

curved confinements as a function of the thickness d (f ¼ 0.5, xN ¼ 30, and N ¼ 20) at
curvatures of K ¼ (a) 0.02 and (b) 0.03. Lines 1 and 2: concentric cylinder barrel
structures at s AS ¼ 0 and 0.05, respectively; line 3: sector column phase; line 4:
complex multilayered sector column phase discussed in Section 4.3.2.3.
barrel to the sector column structures and vice versa. It is hard to form
the sector column phase at a high d due to the increasing Helmholtz
energy. In order to examine the effect of curvature on the morphology
transition for AB diblock copolymer films, Figure 36a summaries lines
[(Figure_6)TD$IG]
Figure 36 (a) Helmholtz energy predicted by SSL for AB diblock copolymers

confined in the different curving surfaces as a function of d at s AS ¼ 0.05 (f ¼ 0.5,
xN ¼ 30, and N ¼ 20). Lines 1–3: concentric cylinder barrel phase at K ¼ 0.01, 0.02, and
0.03, respectively; lines 4–6: sector column phase at K ¼ 0.01, 0.02, and 0.03,
respectively; lines 7–9: complex multilayered sector column phase at K ¼ 0.01, 0.02,
and 0.03, respectively. (b) A sketch of morphology transition for a set of competing
structures between the concentric cylinder barrel phase and the sector column phase
in part (a), where ‘‘sec’’ means the sector column phase, and ‘‘con\n’’ means the
concentric cylinder barrel structure with Nlayer ¼ n.
(2 and 3) in Figure 35a and b. For the concentric cylinder barrel

structures, line 1 in Figure 36a corresponds to K = 0.01; lines (2) and
(3) in Figure 36a correspond to line 2 in Figure 35a and b. While lines
4–6 in Figure 36a are for the sector column phase with the same K =
0.01 and correspond to line 4 in Figure 35a and b. In contrast to lines
4–6 in Figure 36a, it is clear that the upward tendency of sector
column structures becomes more sensitive along with increasing K
values. With increasing K from 0.01 to 0.03, the thickness d of
periodical transformation decreases gradually from 23 to 14. The
transformation between concentric cylinder barrel and sector
column structures is summarized graphically in Figure 36b.
From Figure 36, when K and d increase, the Helmholtz energy of the
sector column phase rises up rapidly. If K ! 0 in Equation (68) (the
vertical lamellae structure under the flat confinement), the Helmholtz
energy approaches to a constant and is independent of d because the
second term of Equation (68) is zero. In Semenov’s SSL, the chain con-
figurations are described by the Alexander-de Gennes approximation
(de Gennes, 1980b) in which the chain stretch is assumed to be normal to
the phase interface. At a large value of Kd, the copolymer chains near the
curved surfaces stretch and shrink excessively with a sharp increase of
conformational entropy. Obviously, a new phase will come up nearby
the exterior surface for a large d resulting in a limited extensibility of the
copolymer chains.
4.3.2.2 Complex multilayered sector column structure. In Section

4.32.1, we have estimated the Helmholtz energies of both the
concentric cylinder barrel and the sector column structures by SSL
theory. It is found that, with increasing the curvature K, SSL theory
predicts that the compatible region will increase, this result is in
contradiction to MC simulation. This conflict between SSL theory and
MC simulation is probably caused by the neglect of some possible
morphologies in SSL theory. One of the possible morphologies is the
‘‘CMSC’’ structure just as mentioned above. This new phase region
might appear near the curved surfaces at a large d.
From the above MC simulation results, a suppositional concentric
square column structure is constructed, which is not observed in MC
simulation (Yu et al., 2007a) but can be considered as a basic block of the
CMSC structure. We plot the same four architectures together forming a
close-packed structure in Figure 37. Under curved confinements, the
topological CMSC structure can be considered as the cross-sectional
concentric circle that truncates the four close-packed square columns
with the exterior radii Rex, and the center is at the common corner of the
four squares. Among the four square columns, the phase structures of
[(Figure_7)TD$IG]
Figure 37 Under curved confinement, when concentric circle scans the close-packed
square columns of exterior radius Rex, the topological structure of CMSC would be
considered to be a cross-section of concentric circle, whose center is at the common
corner of the four squares, exterior and interior radius are Rex and Rin, respectively.
neighboring square columns are antisymmetry, while that of diagonal

square columns are symmetry. Because of the geometrical confinement
of the 2D concentric circles, the Helmholtz energy of the CMSC struc-
ture consists of three parts: a term of concentric square columns, a
surface tension term, and the bending term from the curved surfaces:
pffiffiffi !4
N 3p2 2L
FCMSC ¼ FSquare ðRex Þ þ ðs AS þ s BS Þ 2

rd 2
64Na Rex 2
2 ð69Þ
d

Rex
When both the exterior and interior surfaces of the confined cylinder
are neutral, the second term in Equation (69) equals to zero. If the
exterior radius Rex is fixed, FCMSC is only a function of d. Fsquare(Rex)
is the Helmholtz energy of concentric square column phase.
Sym
The Helmholtz energy per copolymer chains FSquare of the symmet-
rical concentric square column structure can be expressed as
pffiffiffi
Sym 2 2N s AB 8m2 þ 4m þ 1
FSquare ¼
rRex 8m þ 2

3p2 L3 2m þ 1 1 1 L
þ Rex mþ mL þ FBst ð70Þ
64Na2 R2ex 6 6 2 32
Sym
FSquare is a function of the exterior radius Rex as well as the layer
number m. Furthermore, if m = 0, FBst ¼ 0; while m 6¼ 0, FBst =
m R
P
3p 2 2
4Na2 R2 ex
½S ðlÞ zðrÞ dr, where
B
l¼1
Z
2 1 x3l L3 L x4 L4
zðrÞ dr ¼ RðlÞ l
4 3 4 8
½SB ðlÞ

ð 1 x l Þ 3 L3 3 ð1 xl Þ4 L4
þ RðlÞ L þ ð71Þ
3 4 4
and

1 2x þ 1 1
xl þ 2RðlÞ l L ¼ 2RðlÞ L ð72Þ
2 4 4
Similar to the asymmetrical cylinder barrel structure, the asymmet-

rical square column is considered as a combination of the symmetrical
multilayered structure with an asymmetrical one with half a period.
Asy
The Helmholtz energy FSquare of the asymmetrical concentric square
column structure is separated into two parts: one is the Helmholtz
Sym
energy FSquare of symmetrical multilayered structure and another is
Asy
the Helmholtz energy FSquare of an asymmetrical one with half a period
expressed by
Asy Asy Asy
FSquare ¼ Fint þ Fst
pffiffiffi 0
2 2s AB N 8M2 þ 12M þ 4x þ 3 3p2 L3
¼ þ
rRex 8M þ 6 64Na2 R2ex

2M þ 1 1 1 L BðSymÞ
Rex Mþ ML þ Fst
6 6 2 32
" 0
! 03 04
# 04
3p3 4M þ 2f þ 1 f L3 f L4 p2 x L4
þ R ex L þ þ
4Na2 R2ex 4 12 32 128Na2 R2ex
ð73Þ
BðSymÞ BðSymÞ
If M = 0, Fst ¼ 0, while M 6¼ 0, Fst ¼ ð3p2 =4Na2 R2ex Þ
M R
P 2
½SB ðlÞ zðrÞ dr. In the limit of d ! Rex, f 0 and x0 could be solved by
l¼1
02 0
2f 4f þ 1 ¼ 0 and f 0 + x0 = 1. For a given Rex, the minimization of
Asy
FSquare is performed with respect to the optimization of m.
4.3.2.3
Helmholtz energy of CMSC structure. In Section 4.32.1, the
Helmholtz energies of the two competing structures, that is, the
concentric cylinder barrel and the sector column structures, have been
discussed by SSL. Line 4 in Figure 35a and b denotes the Helmholtz
energy FCMSC of the CMSC structure, which slightly declines with
increasing d. By comparing with other Helmholtz energy profiles, it is
easy to find that the CMSC structure tends to form at a high d, while the
concentric cylinder barrel and the sector column structures occur
mostly at a small d. This is consistent with MC simulation results.
Further analyzing the profiles in Figure 35a and b, we see the line 4 is
lower than other lines at both higher K and larger d indicating that the
CMSC structure is more stable. However, the CMSC structure is not
observed at small enough Rex in Figure 36a.
Line 7 in Figure 36 indicates the occurring of the CMSC structure at
K = 0.01; while lines 8 and 9 are at K = 0.02 and 0.03, respectively,
corresponding to line 4 in Figure 35a and b. From lines 7–9, one can
find that the origo of the Helmholtz energy profiles rises up with
increasing curvature K, despite a visible downtrend of the Helmholtz
energy FCMSC. Due to the influence of bending, higher Helmholtz
energy profiles of square column phase are observed at smaller Rex
and larger FCMSC. As a consequence, the CMSC structure is very hard
to form at a sufficiently small Rex. Under these conditions, the morphol-
ogy in Figure 36a is different from others.
4.4 Remarks on the phase separation of confined diblock

copolymer
In this work, a framework of the SSL theory for diblock copolymer melts
confined in ring-like curved surfaces has been proposed. When the
curvature approaches to zero, it reduces to the well-known SSL theory
for the parallel lamellar phases. In the case of the equal confined thick-
ness to the exterior radius, it can also be extended to the system with a
nanopore confinement. Moreover, the Helmholtz energy of the concen-
tric cylinder barrel, sector column and CMSC phases in 2D confine-
ments based on this SSL theoretical framework can be evaluated in the
convenient manner. The calculated results show that the diblock copol-
ymer melts exhibit a layer-type transition with a similar mechanism,
regardless of ring-like curved surfaces, planar surfaces, and nanopores.
In this work, we have focused on strong surface preference with only
mutual transition between the symmetrical and asymmetrical layer-
type structures. SSL theory and MC simulation are further applied to
investigate the self-assembled morphology of diblock copolymers con-
fined in the nanopore. MC simulation shows that the Nlayer of the
concentric cylinder barrel changes with respect to the extent of frustra-
tion between the exterior radius Rex and the bulk lamellar period L0.
Simultaneously, the predictions of SSL theory also show that both
symmetrical and asymmetrical structures occur periodically by mini-

mizing the Helmholtz energy before and after the cross-points with
increasing Rex/L0. SSL predictions are consistent with MC simulations.
Although there are some marginal discrepancies for PS-PBD diblock
copolymers, SSL theory for confinement presented in this work shows
good agreement with experimental data for PS-PMMA diblock copoly-
mers reported in the literature.
For the surfaces with weak preference, it can be expected for more
complex structures with the competition of Helmholtz energy, leading
to the morphology transition with the multilayer type. First, two com-
peting micro-phase morphologies of the concentric cylinder barrel and
the sector column structures exist for AB diblock copolymers confined
between curved surfaces, which resembles the cases of the parallel and
vertical lamellar structures confined in planar surfaces. However, the-
oretical predictions are inconsistent with simulations results due to the
neglect of some complicated morphologies in the theoretical frame-
work. Second, we examine a representative complicated morphology
of the CMSC structure. From MC simulation, the topological character-
istic relationship between the CMSC phase and the concentric square
column structure phase is established. Finally, the predicted results are
in good agreement with simulation results that the CMSC structure
occurs at a higher thickness d, and both the mutually competing con-
centric cylinder barrel and the sector column structures at a lower d. The
reason why the CMSC structure hardly forms at a lower Rex is also
unveiled in the theory, which is confirmed by MC simulation.
5. CONCLUSIONS
Many polymer blends or block polymer melts separate microscopically

into complex meso-scale structures. It is a challenge to predict the multi-
scale structure of polymer systems including phase diagram, mor-
phology evolution of micro-phase separation, density and composition
profiles, and molecular conformations in the interfacial region between
different phases. The formation mechanism of micro-phase structures
for polymer blends or block copolymers essentially roots in a delicate
balance between entropic and enthalpic contributions to the Helmholtz
energy. Therefore, it is the key to establish a molecular thermodynamic
model of the Helmholtz energy considered for those complex meso-
scale structures. In this paper, we introduced a theoretical method
based on a lattice model developed in this laboratory to study the
multi-scale structure of polymer systems. First, a molecular thermody-
namic model for uniform polymer system is presented. This model can
be successfully used to calculate the pVT behavior, VLE and LLE for
polymer systems. Combined with DFT, this molecular thermodynamic
model can also be used to describe the adsorption of a polymer at a
solid–liquid interface and the molecular conformation distribution at
interface regions. For the meso-structure of diblock copolymer melts
confined in curved surfaces, a SSL theory is presented to study the
morphological transition mechanism. The theoretical predicted results
are in agreement with MC simulations and experimental observations.
It means that the model parameters in the Helmholtz energy model can
be obtained by correlating the pVT behavior, VLE and LLE of polymer
systems. They then can be used, combining with DFT and SSL theory, to
predict the adsorption behavior of polymers at solid–liquid interfaces
and the meso-structure of polymer melts.
To establish the molecular thermodynamic model for uniform sys-
tems based on concepts from statistical mechanics, an effective method
by combining statistical mechanics and molecular simulation has been
recommended (Hu and Liu, 2006). Here, the role of molecular simula-
tion is not limited to be a standard to test the reliability of models. More
directly, a few simulation results are used to determine the analytical
form and the corresponding coefficients of the models. It retains the
rigor of statistical mechanics, while mathematical difficulties are
avoided by using simulation results. The method is characterized by
two steps: (1) based on a statistical–mechanical derivation, an analytical
expression is obtained first. The expression may contain unknown
functions or coefficients because of mathematical difficulty or some-
times because of the introduced simplifications. (2) The form of the
unknown functions or unknown coefficients is then determined by
simulation results. For the adsorption of polymers at interfaces, simu-
lation was used to test the validity of the weighting function of the WDA
in DFT. For the meso-structure of a diblock copolymer melt confined in
curved surfaces, we found from MC simulation that some more com-
plex structures exist. From the information provided by simulation,
these complex structures were approximated as a combination of sim-
ple structures. Then, the Helmholtz energy of these complex structures
can be calculated by summing those of the different simple structures.
The macroscopic properties such as mechanical behavior of block
copolymers or polymer blends depend directly on the relative concen-
trations of different constituents and their meso-structures. How to
predict the exact macroscopic properties of polymer blends or block
copolymers with meso-phase separation structures from pure com-
ponent properties remains a big challenge. Some theoretical efforts
have been explored. For example, Buxton et al. found that the defor-
mations and fractures of polymer blends can be described by the
micromechanical lattice spring model (LSM). Here, the information of

the micro-phase separation of polymer blends obtained from the Cahn–
Hilliard (CH) model, the lattice Boltzmann model (LBM), or the MD
method are taken as the input of the LSM (Buxton and Balazs, 2005;
Buxton and Balazs, 2004; Buxton and Balazs, 2003; Buxton et al., 2005).
LIST OF SYMBOLS
a the statistical bond length of polymer

c empirical constants in the model of Helmholtz energy of mixing
in Equation (17)
d the thickness of block copolymer membrane confined between
two concentric curved surfaces
Db the branching degree of a branched polymer
F nonrandom factor defined in Equation (12) or composition of
diblock copolymer
F Helmholtz energy
g radial distribution function
G propagator function in Equations (59)–(61)
k Boltzmann constant
K number of component in mixture or the curvature of cylindrical
pore defined by K = 1/Rex
L the length of cylindrical pore
m the number of periods in the lamellar
N the number of molecule in the fluid mixture
Nh the number of head units of a polymer chain
Nr total number of sites in lattice model
N? the number of ways in which three bonds meet at a lattice site
P pressure
q surface area parameter
q the coordinate of all solvent molecules or polymer segment
Q the coordinate of all polymers with configurations inherited
r chain length of polymer
R the radius of cylindrical pore
S entropy
T temperature (K)
U internal energy
v the external potential exerted on a segment of polymer
n the hard-core volume of a segment of polymer or solvent
V volume or the external potential exerted on all segments
w weighting function in DFT
x mole fraction or the distortion coefficient of block
z coordination number of a lattice
x Flory–Huggins parameter
e interaction energy between segments of polymer or between the
segment of polymer and solvent
f volume fraction
V the grand potential
h surface fraction of a segment of polymer that can participate in
oriented interaction
C function defined by Equation (54)
l a parameter accounting for the long-range correlations beyond
the close contact pair
m chemical potential
u surface fraction
r density of fluid
s the interfacial tension
2 the exchange energy between segments of different components
ACKNOWLEDGMENTS
This work is supported by the National Natural Science Foundation
of China (Project No. 20736002), Program for Changjiang Scholars
and Innovative Research Team in University (No. IRT0721), and the
111 Project (No. B08021). The authors also would like to thank
Dr. Jianwen Jiang of Department of Chemical and Biomolecular
Engineering of National University of Singapore for his comments
and revisions in the preparation of this paper.
REFERENCES
Anderson, D. M. and Thomas, E. L., Macromolecules 21, 3221 (1988).
Aranovich, G. L., Donohue, P. S. and Donohue, M. D., J. Chem. Phys. 111, 2050 (1999).
Aranovich, G. L. and Donohue, M. D., J. Chem. Phys. 112, 2361 (2000).
Arya, G. and Panagiotopoulos, A. Z., Macromolecules 38, 10596 (2005).
Ash, S. G., Everett, D. H. and Findenegg, G. H., Trans. Faraday Soc. 66, 708 (1970).
Bates, F. S., Science 251, 898 (1991).
Bethe, H. A. and Wills, H. H., Proc. R. Soc. A150, 552 (1935).
Bragg, W. L. and Williams, E. J., Proc. R. Soc. A145, 699 (1934).
Buxton, G. A. and Balazs, A. C., Phys. Rev. E 67, 031802 (2003).
Buxton, G. A. and Balazs, A. C., Phys. Rev. B 69, 054101 (2004).
Buxton, G. A. and Balazs, A. C., Macromolecules 38, 488 (2005).
Buxton, G. A., Verberg, R., Janow, D. and Balazs, A. C., Phys. Rev. E 71, 056707 (2005).
Cai, J., Liu, H. L. and Hu, Y., Fluid Phase Equilib 194-197, 281 (2002).
Carnahan, N. F. and Starling, K. E., J. Chem. Phys. 51, 635 (1969).
Chaikin, P. M. and Lubensky, T. C., ‘‘Principles of Condensed Matter Physics’’.
Cambridge, Cambridge University Press (1995).
Chang, B. H. and Bae, Y. C., Chem. Eng. Sci. 58, 2931 (2003) Chang B. H., Bae Y. C., J. Polym.
Sci. B: Polym. Phys., 42, 1532(2004).
Chang, B. H., Ryu, K. O. and Bae, Y. C., Polymer 39, 1735–1739 (1998).
Chen, P. and Liang, H., Macromolecules 40, 7329 (2007).
Chen, T., Liu, H. L. and Hu, Y., Macromolecules 33, 1904 (2000).
Chen, T., Peng, C. J., Liu, H. L. and Hu, Y., Fluid Phase Equilib 233, 73 (2005).
Chen, H. Y., Ye, Z. C., Peng, C. J., Liu, H. L. and Hu, Y., J. Chem. Phys 125, 204708 (2006).
Chen, Y., Aranovich, G. L. and Donohue, M. D., J. Chem. Phys. 124, 134502 (2006) Chen Y.,
Aranovich G. L., Donohue M. D., J. Chem. Phys., 127, 134903 (2007).
Chen, P., He, X. and Liang, H., J. Chem. Phys. 124, 104906 (2006).
Chen, H. Y., Ye, Z. C., Cai, J., Liu, H. L., Hu, Y. and Jiang, J. W., J. Phys. Chem. B 111, 5927
(2007).
Chen, H. Y., Cai, J., Ye, Z. C., Peng, C. J., Liu, H. L., Hu, Y. and Jiang, J. W., J. Phys. Chem. B
112, 9568 (2008).
Chen, X. Q., Sun, L., Liu, H. L., Hu, Y. and Jiang, J. W., J. Chem. Phys. 131, 044710 (2009).
Chen, H. Y., Chen, X. Q., Ye, Z. C., Liu, H. L. and Hu, Y., Langmuir 26, 6663 (2010).
Chernoff, M. P., Aranovich, G. L. and Donohue, M. D., J. Chem. Phys 116, 9395 (2002).
Cummings, P. T. and Stell, G., Mol. Phys. 51, 253 (1984).
de Gennes, P. G., Macromolecules 13, 1069 (1980).
de Gennes, P. G., Macromolecules 13, 1069 (1980).
DeMarzio, E. A. and Rubins, R. J., J. Chem. Phys. 55, 4318 (1971).
Dudowicz, J. and Freed, K. F., Macromolecules 23, 1519 (1990) Dudowicz J., Freed K. F.
Macromolecules, 24, 5076, 5096, 5112(1991).
Dudowicz, J. and Freed, K. F., Theor. Chim. Acta 82, 357 (1992).
Dudowicz, J., Freed, K. F. and Madden, W. G., Macromolecules 23, 4803 (1990).
Feng, J. and Ruckenstein, E., Macromolecules 39, 4899 (2006).
Feng, J. and Ruckenstein, E., J. Chem. Phys. 125, 164911 (2006).
Feng, J., Liu, H. L. and Hu, Y., Mol. Simul. 31, 731 (2005).
Fisher, M. E., Rep. Prog. Phys. 30, 615 (1967).
Flory, P. J., J. Chem. Phys. 9, 660 (1941) Flory P. J., J. Chem. Phys., 10, 51(1942).
Fraaije, J. G. E. M., J. Chem. Phys. 99, 9202 (1993).
Fredrickson, G. H. and Helfand, E., J. Chem. Phys 87, 697 (1987).
Freed, K. F., J. Phys. A: Math. Gen. 18, 871 (1985).
Graham, R. S. and Olmsted, P. D., Faraday Discuss. 144, 71 (2010).
Groot, R. D. and Warren, P. B., J. Chem. Phys. 107, 4423 (1997).
Guggenheim, E. A., ‘‘‘‘Mixtures’’’’. Oxford University Press, Oxford (1952).
Han, Y., Cui, J. and Jiang, W., Macromolecules 41, 6239 (2008).
Hawker, C. J., Lee, R. and Frecbet, J. M. J., J. Am. Chem. Soc 113, 4583 (1991).
He, X., Song, M., Liang, H. and Pan, C., J. Chem. Phys. 114, 10510 (2001).
Helfand, E., J. Chem. Phys 63, 2192 (1975) Helfand E., Macromolecules, 9, 307(1976).
Helfand, E. and Wasserman, Z. R., Macromolecules 9, 879 (1976) Helfand E., Wasserman Z.
R., Macromolecules, 11, 960(1978); Helfand E., Wasserman Z. R., Macromolecules, 13,
994(1980).
Heller, P., Rep. Prog. Phys. 30, 731 (1967).
Hill, T. L., ‘‘Statistical Mechanics’’. McGraw Hill, New York (1956).
Hu, Y. and Liu, H. L., Fluid Phase Equilib. 241, 248 (2006).
Hu, Y., Lambert, S. M., Soane, D. S. and Prausnitz, J. M., Macromolecules 24, 4356 (1991).
Hu, Y., Liu, H. L., Soane, D. S. and Prausnitz, J. M., Fluid Phase Equilib 67, 65 (1991).
Hu, Y., Ying, X. G., Wu, D. T. and Prausnitz, J. M., Fluid Phase Equilib 83, 289 (1992).
Hu, Y., Liu, H. L. and Prausnitz, J. M., J. Chem. Phys. 104, 396 (1996).
Hu, Y., Liu, H. L. and Shi, Y. H., Fluid Phase Equilib 117, 100 (1996).
Huang, Y. M., Jin, X. C., Liu, H. L. and Hu, Y., Fluid Phase Equilib 263, 96 (2008).
Huggins, M. L., J. Chem. Phys. 9, 440 (1941) Huggins M. L., J. Phys. Chem., 46, 151(1942).
Janssen, R. H. C. and Nies, E., Langmuir 13, 2784 (1997).
Jiang, J. W., Yan, Q. L., Liu, H. L. and Hu, Y., Macromolecules 30, 8459 (1997).
Kellogg, G. J., Walton, D. G., Mayes, A. M., Lambooy, P., Russell, T. P., Gallagher, P. D. and
et al. Phys. Rev. Lett. 76, 2503 (1996).
Kierlik, E. and Rosinberg, M. L., J. Chem. Phys. 97, 9222 (1992) Kierlik E., Rosinberg M. L., J.
Chem. Phys., 99, 3950(1993).
Kikuchi, R., Phys. Rev. 81, 988 (1951) Kikuchi R., J. Chem. Phys., 47, 1664(1967).
Lacombe, R. H. and Sanchez, I. C., J. Phys. Chem 80, 2568 (1976).
Lambert, S. M., Soane, D. S. and Prausnitz, J. M., Fluid Phase Equilib 83, 59 (1993).
Lambooy, P., Russell, T. P., Kellogg, G. J., Mayes, A. M., Gallagher, P. D. and Satija, S. K.,
et al. Macromolecules 13, 1602 (1980).
Li, H. and Huck, W. T. S., Curr. Opin. Solid State Mater. Sci. 6, 3 (2002).
Li, W. and Wickham, R. A., Macromolecules 39, 8492 (2006).
Li, W., Wickham, R. A. and Garbary, R. A., Macromolecules 39, 806 (2006).
Li, J. L., He, H. H., Peng, C. J., Liu, H. L. and Hu, Y., Fluid Phase Equilib 276, 57 (2009).
Li, J. L., He, H. H., Peng, C. J., Liu, H. L. and Hu, Y., Fluid Phase Equilib. 286, 8 (2009).
Li, J. L., Tong, M., Peng, C. J., Liu, H. L. and Hu, Y., Fluid Phase Equilib. 287, 50 (2009).
Likhtman, A. E. and Semenov, A. N., Macromolecules 27, 3103 (1994).
Likhtman, A. E., Anastasiadis, S. H. and Semenov, A. N., Macromolecules 32, 3474 (1999).
Liu, H. L. and Hu, Y., Ind. Eng. Chem. Res. 37, 3058 (1998).
Liu, H. L., Yang, J. Y., Xin, Q. and Hu, Y., Fluid Phase Equilib 261, 281 (2007).
Liu, H. L., Xu, H., Chen, H. Y., Peng, C. J. and Hu, Y., Struct. Bonding 131, 109 (2008).
Ma, M., Krikorian, V., Yu, J. H., Thomas, E. L. and Rutledge, G. C., Nano Lett. 6, 2969 (2006).
Madden, W. G., Pesci, A. I. and Freed, K. F., Macromolecules 23, 1181 (1990).
Matsen, M. W. and Barrett, C., J. Chem. Phys. 109, 4108 (1998).
Mita, K., Tanaka, H., Saijo, K., Takenaka, M. and Hashimoto, T., Macromolecules 41, 6787
(2008).
Nemirovsky, A. M., Bawendi, M. G. and Freed, K. F., J. Chem. Phys 87, 7272 (1987).
Nieswand, M., Dieterich, W. and Majhofer, A., Phys. Rev. E 47, 718 (1993).
Oh, S. Y. and Bae, Y. C., Eur. Polym. J. 46, 1328 (2010 a) Oh S. Y., Bae Y. C., J. Phys. Chem. B,
114, 8948(2010).
Ohta, T. and Kawasaki, K., Macromolecules 23, 2413 (1990).
Ohta, T. and Kawasaki, K., Macromolecules 19, 2621 (1986).
Ono S., and Kondo S. Molecular theory of surface tension in liquids, in ‘‘Encyclopedia of
Physics’’ (S. Fl€
ugge, Ed.), Springer, Berlin (1960).
Oono, Y. and Puri, S., Phys. Rev. A 38(434), 1542 (1988).
Oono, Y. and Shiwa, Y., Modern Phys. Lett. B 1, 49 (1987).
Panagiotopoulos, A. Z. and Wong, V., Macromolecules 31, 912 (1998).
Panayiotou, C., Pantoula, M., Stefanis, E., Tsivintzelis, I. and Economou, I. G., Ind. Eng.
Chem. Res. 43, 6952 (2004).
Pankavich, S., Shreif, Z., Miao, Y. and Ortoleva, P., J. Chem. Phys. 130, 194115 (2009).
Park, M., Harrison, C., Chainkin, P. M., Register, R. A. and Adamson, D. H., Science 276,
1401 (1997).
Patra, C. N. and Yethiraj, A., J. Chem. Phys. 112, 1579 (2000) Patra C. N., Yethiraj A., J. Chem.
Phys., 118, 4702(2003).
Peng, C. J., Liu, H. L. and Hu, Y., Chem. Eng. Sci. 56, 6967 (2001).
Peng, C. J., Liu, H. L. and Hu, Y., Ind. Eng. Chem. Res 41, 862 (2002).
Peng, C. J., Liu, H. L. and Hu, Y., Fluid Phase Equilib. 206(127), 147 (2003).
Pesci, A. I. and Freed, K. F., J. Chem. Phys 90, 2017 (1989).
Prausnitz, J. M., Lichtenthaler, R. N. and de Azevedo, E. G., ‘‘Molecular Thermodynamics
of Fluid-Phase Equilibria, 3rd ed. Prentice Hall PTR (1999).
Prestipino, S. and Giaquinta, P. V., J. Phys. Condens. Matter 15, 3931 (2003).
Reinhard, J., Dieterich, W., Maass, P. and Frisch, H. L., Phys. Rev. E 61, 422 (2000).
Rider, D. A., Chen, J. I. L., Eloi, J. C., Arsenault, A. C., Russell, T. P., Ozin, G. A. and et al.
Macromolecules 41, 2250 (2008).
Rodriguez, A. L., Freire, J. and Horta, A., J. Phys. Chem 96, 3954 (1992).
Sadus, R. J., ‘‘Molecular simulation of fluids, theory, algorithms and object-orientation’’.
Amsterdam, Elsevier (1999).
Sanchez, I. C. and Lacombe, R. H., J. Phys. Chem. 80, 2352 (1976) Sanchez I. C., Lacombe R.
H., Macromolecules, 11, 1145(1978).
Scheutjens, J. M. H. M. and Fleer, G. J., J. Phys. Chem. 83, 1619 (1979).
Scheutjens, J. M. H. M. and Fleer, G. J., J. Phys. Chem. 84, 178 (1980).
Semenov, A. N., Sov. Phys.-JEPT (Engl. Transl.) 61, 733 (1985).
Semenov, A. N., Macromolecules 22, 2849 (1989).
Semenov, A. N., Bonet-Avalos, J., Johner, A. and Joanny, J. F., Macromolecules 29, 2179
(1996).
Sengers J. V., and Sengers J. M. in ‘‘Progress in Liquid Physica’’ (C. A. Croxton, Ed.),
Chapter 4. Wiley, New York (1978).
Sevink, G. J. A. and Zvelindovsky, A. V., J. Chem. Phys. 128, 084901 (2008).
Sevink, G. J. A., Zvelindovsky, A. V., Fraaije, J. G. E. M. and Huinink, H. P., J. Chem. Phys.
115, 8226 (2001).
Shin, M. S. and Kim, H., Fluid Phase Equilib 246, 79 (2006).
Shin, K., Xiang, H., Moon, S. I., Kim, T., McCarthy, T. J. and Russell, T. P., et al. J. Phys.
Chem. 57, 584 (1953).
Song, K. X., Jia, Y. X., Sun, Z. Y. and An, L. J., J. Chem. Phys. 129, 144901 (2008).
Stell, G. and Zhou, Y. Q., J. Chem. Phys. 91, 3618 (1989).
Sun, Y., Steinhart, M., Zschech, D., Adhikari, R., Michler, G. H. and G€ osele, U., et al. Mol.
Phys. 41, 85 (1980).
Tsivintzelis, I., Dritsas, G. S. and Panayiotou, C., Ind. Eng. Chem. Res. 45, 7264 (2006).
Turner, M. S., Phys. Rev. Lett. 69, 1788 (1992).
Walton, D. G., Kellogg, G. J., Mayes, A. M., Lambooy, P. and Russell, T. P., Macromolecules
27, 6225 (1994).
Wang, Q., J. Chem. Phys 126, 024903 (2007).
Wang, Q., Yan, Q. L., Nealey, P. F. and de Pablo, J. J., J. Chem. Phys 112, 450 (2000).
Wang, Z., Li, B., Jin, Q., Ding, D. and Shi, A., Macromol. Theory Simul 17, 86 (2008).
Wang, Z., Li, B., Jin, Q., Ding, D. and Shi, A., Macromol. Theory Simul 17, 301 (2008).
Woodward, C. E., J. Chem. Phys 94, 3183 (1991).
Wu, J. Z., AIChE J. 52, 1169 (2006).
Xia, Y., Rogers, J. A., Paul, K. E. and Whitesides, G. M., Chem. Rev. 99, 1823 (1999).
Xiang, H., Shin, K., Kim, T., Moon, S. I., McCarthy, T. J. and Russell, T. P., et al. J. Polym. Sci.
B: Polym. Phys. 43, 3377 (2005).
Xiang, H., Shin, K., Kim, T., Moon, S. I., McCarthy, T. J. and Russell, T. P., et al.
Macromolecules 38, 1055 (2005).
Xiao, X. Q., Huang, Y. M., Liu, H. L. and Hu, Y., Macromol. Theory Simul. 16, 732 (2007).
Xin, Q., Peng, C. J., Liu, H. L. and Hu, Y., Ind. Eng. Chem. Res 47, 9678 (2008).
Xin, Q., Peng, C. J., Liu, H. L. and Hu, Y., Fluid Phase Equilib 267, 163 (2008).
Xin, Q., Xu, X. C., Huang, Y. M., Peng, C. J., Liu, H. L. and Hu, Y. Sci. China B 38, 947 (2008).
Xu, H., Liu, H. L. and Hu, Y., Macromol. Theory Simul. 16, 262 (2007).
Xu, H., Liu, H. L. and Hu, Y., Chem. Eng. Sci. 62, 3494 (2007).
Xu, H., Wang, T. F., Huang, Y. M., Liu, H. L. and Hu, Y., Ind. Eng. Chem. Res. 47, 6368 (2008).
Xu, X. C., Liu, H. L., Peng, C. J. and Hu, Y., Fluid Phase Equilib. 265, 112 (2008).
Xu, X. C., Peng, C. J., Cao, G. P., Liu, H. L. and Hu, Y., Ind. Eng. Chem. Res. 48, 7828 (2009).
Xu, X. C., Peng, C. J., Huang, Y. M., Liu, H. L. and Hu, Y., Ind. Eng. Chem. Res. 48, 11189
(2009).
Yan, Q. L., Liu, H. L. and Hu, Y., Macromolecules 29, 4066 (1996).
Yan, Q. L., Liu, H. L. and Hu, Y., Fluid Phase Equilib 218, 157 (2004).
Yang, J. Y., Yan, Q. L., Liu, H. L. and Hu, Y., Polymer 47, 5187 (2006).
Yang, J. Y., Peng, C. J., Liu, H. L., Hu, Y. and Jiang, J. W., Fluid Phase Equilib 244, 188 (2006).
Yang, J. Y., Xin, Q., Sun, L., Liu, H. L., Hu, Y. and Jiang, J. W., et al. Fluid Phase Equilib. 249,
192 (2006).
Yang, J. Y., Peng, C. J., Liu, H. L. and Hu, Y., Ind. Eng. Chem. Res 45, 6811 (2006).
Ye, Z. C., Cai, J., Liu, H. L. and Hu, Y., J. Chem. Phys. 123, 194902 (2005).
Ye, Z. C., Chen, H. Y., Cai, J., Liu, H. L. and Hu, Y., J. Chem. Phys. 125, 124705 (2006).
Ye, Z. C., Chen, H. Y., Liu, H. L., Hu, Y. and Jiang, J. W., J. Chem. Phys. 126, 134903 (2007).
Yethiraj, A. and Woodward, C. E., J. Chem. Phys 102, 5499 (1995).
Yin, Y., Sun, P., Chen, T., Li, B., Jin, Q., Ding, D. and et al. Chem Phys Chem 5, 540 (2004).
Yu, Y. X. and Wu, J. Z., J. Chem. Phys. 117, 2368 (2002).
Yu, B., Sun, P., Chen, T., Jin, Q., Ding, D. and Li, B., et al. J. Chem. Phys 126, 204903 (2007).
Yu, B., Sun, P., Chen, T., Jin, Q., Ding, D., Li, B. and et al. J. Chem. Phys 127, 114906 (2007).
Yu, B., Li, B., Jin, Q., Ding, D. and Shi, A., Macromolecules 40, 9133 (2007).
Zhang, S. L., Cai, J., Liu, H. L. and Hu, Y., Mol. Simul. 30, 143 (2004).
Zheng, W. and Wang, Z. G., Macromolecules 28, 7215 (1995).
Zhi, D. Y., Huang, Y. M., Han, X., Liu, H. L. and Hu, Y., Chem. Eng. Sci. 65, 3223 (2010).
Zhou, Y. Q. and Stell, G., J. Chem. Phys. 96(1504), 1507 (1992).
Zhou, S. Q. and Zhang, X. Q., Phys. Rev. E 64, 011112 (2001).
Zhu, Y. and Jiang, W., Macromolecules 40, 2872 (2007).
S U B J E C T IN D E X
AB diblock copolymers makeup stream, 145

under curved confinement, MC optimization cases, 147
simulation of overall cycle time of, 146
CMSC structure. See Complex separation models for absorber and
multilayered sector column regenerators, 145
structure Aspen Plus EO model
between concentric curved surfaces, for ammonia plant. See Aspen Plus EO
190–192 ammonia plant
cylindrical pores, 187–190 for MDEA/PZ/CO2 capture unit, 143
Helmholtz energies, 204–206 Asymmetrical concentric-ring barrel
under flat confinements, 190, 191 structure, Helmholtz energy of,
Helmholtz energy of, 186–187 196–198
morphologies of, 186, 188 Asymmetrical concentric square column
MC simulated, 188–190 structure, Helmholtz energy of, 208
vertical and parallel lamellar structure Athermal entropy of mixing, 162–163
of, 191, 192 Athermal mixture
Absorber and regenerators, separation chemical potentials of, 162
models for, 145 probabilities of 1–1 pairs of, 163
Aggregates, gas–solids interphase Atomistic clusters, mapping of, 88
momentum transfer, 30–31 Atomistic MD simulations, 93
Air bearing of HDI, flow inside, 109–112 Atomistic/molecular-level modeling,
Alkanolamine solution, CO2 capture by, 76–81
136–137 and integration, 87–89
Ammonia plants, 143
Aspen Plus EO model for. See Aspen
Plus EO ammonia plant Binary ising lattice, coexistence curves of,
Aspen Plus EO ammonia plant 166
blocks in, 144 Binary polymer solutions
CO2 capture system coexistence curve of, 168–169
optimization cases, 147–148 normalized internal energy of mixing
parameter cases, 146–147 for, 171
S/C ratio, 147 Block copolymer melts. See also AB diblock
execution times for, 146 copolymers
gas composition optimization, 144 Helmholtz energy of, 185
issues related to model specification in, micro-phase separation with multidi-
145 mensional confinements, 185–186
219
220 Subject Index
morphologies of Close-packed lattice model, 157

effect of disperse index on, 186 Cluster accelerations, X-ray measurement
factors controlling, 184, 185 of, 6–7
variety of, 186 Clustering, 13
Boltzmann transport equation Cluster velocity series determination, 7
and SRS models, 91 Coarse-grained, bead-spring model of
Bonding mechanism between PFPEs and PFPE lubricant films, 104–105
overcoat, 72 with flat surface assumption, 105–106
BTE. See Boltzmann transport equation potential energy characteristics, 106
Bubble columns Coarse-graining methods
physical explanation of regime meso-scale—continuum levels, 91–92
transition in, 41–42 molecular—meso-scale levels, 89–90
total energy dissipation and, 40 quantum—atomistic/molecular levels,
Bubble phenomenon in situ, 99–100 87–89
Cobalt (Co)-based magnetic alloys, 69
CO2 capture
Carnahan–Starling equation for by aqueous alkanolamine solution,
hard-sphere fluids, 158 136–137
Catalyst (particle), reaction mechanism with aqueous MDEA/PZ solution
over, 3 aqueous phase reactions, 138
CFB combustors chemical species considered for, 138
components of, 46 mechanism of action, 137
EMMS-based multi-scale CFD thermophysical properties of, 139
simulation with chemical absorbent, 142
coal combustion, 51 process modeling
hydrodynamics, 48–49 absorber and stripper, 140
seesaw phenomenon, 50 and ammonia plant, 143–148
solid fluxes, 49–50 Aspen Plus EO model for, 143–144
scale-up and optimal design of, 47 equilibrium-stage models for, 141
CFB risers performance correlations, 143
components of, 21 rate-based multistage separation
ETH riser, 22–23 models for, 141–142
IPE riser, 21 Coexistence curves
simulations of, 21 of binary polymer solutions, 168–169
voidage profiles of, 21–22 of branched polymer solutions, 169
CFD simulations, 47 of lattice random copolymers, 170
Chain-like molecular systems, mixing of tert-butyl acetate/PS and water/poly
process of, 160 (ethylene glycol) systems, 174
Chemical engineering, multi-scale Complex multilayered sector column
characteristics of, 3 structure
Chemical reactors under curved confinements, 206–207
multi-scale characteristics of, 2–4 Helmholtz energy of, 207–209
need for scale-up of, 4–5 Computational fluid dynamics, 3
‘‘overall’’ reaction behavior of, 4 correlative. See correlative multi-scale
scales involved in, 3 CFD
Chemical supply chain, multiscale process space resolution of, 10
modeling of, 122 Continuously stirred tank reactor (CSTR)
Classical molecular simulation methods, model, 10
76 Correlative multi-scale CFD
Classic chemical engineering models, 10 challenges associated with, 14–15
Subject Index 221
computation cost effectiveness of, 13 of gas–solid suspension, 11

paradigms for, 12 limitation on scalability, 10–11
for single-phase turbulent flows, 12 Disk overcoat and PFPEs, interaction
subparticle simulations using, 12–13 between, 71–72
and variational, comparison bonding mechanism between, 72
between, 17 DDPA-S, DDPA-D, and ZTMD, 72
Curved surfaces, MC simulation of diblock DMFC. See Direct methanol fuel cell
copolymers confined in Drag coefficient in CFB, 8–9
between concentric curved surfaces, Dry surfaces
190–192 nanotribology of, 67
cylindrical pores, 187–190 Dual-Bubble-Size (DBS) model for
gas–liquid two-phase flow in bubble
columns
DDFT based on equation of state calculation on structure parameters and
(EOS-based DDFT) total gas holdup, 41
applications of, 156 CFD simulation, 42–43
Degrees of freedom (DOFs), 126 components of, 40
in optimization, 127 regime transition in bubble columns,
Dense ‘‘cluster’’ phase, 5 41–42
velocities with respect to, 6 Dynamic structure, 5
Dense-phase momentum balance, 25
Density functional theory (DFT), 75
Diblock copolymers Electrostatic effects, modeling of,
confined in curved surfaces, MC 76–77
simulation of Embedded solution strategy, 131–132
between concentric curved surfaces, EMMS-based multi-scale CFD
190–192 flow regime diagrams of CFB, 32
cylindrical pores, 187–190 industrial applications
confined in ring-like curved surfaces, CFB boiler, 46–51
SSL theory for, 192 fluid catalytic cracking, 43–46
Helmholtz energy of asymmetrical EMMS model. See Energy-minimization
parallel lamellar, 196–198 multi-scale model
Helmholtz energy of sector column, Endbead density profiles for PFPEs,
198–199 106–107
Helmholtz energy of symmetrical Energy-minimization multi-scale model
parallel lamellar, 193–196 application of
under flat and curved confinements, choking point prediction in fast-
190, 191 fluidization, 26
Helmholtz energy of, 186–187 mass/heat transfer and reactions,
morphologies of, 186, 188 35–40
phase separation of confined, and CFD, coupling of
209–210 two-step scheme for, 27–29
vertical and parallel lamellar structure voidage profile and, 30
of, 191, 192 closure of, 26
Dilute ‘‘broth’’ phase, 5 formulation of, 25–26
velocities with respect to, 6 meso-scale heterogeneity of, 24–25
Dilute-phase momentum balance, 25 Equation-oriented (EO) modeling, 121
Direct methanol fuel cell, 64 embedded solution strategy, 131
fuel in, 65 EO model-based RTO applications,
vs. hydrogen fuel cells, 65 134
Direct numerical simulations and sequential modular modeling,
computational demand of, 11 123
222 Subject Index
Equilibrium-based models, absorber and components of, 67, 68

regenerators, 145 as data storage systems, 67
ETH CFB simulation, 31 headdisk interface of
components of, 68, 69
cross-sectional diagram of, 70
FCC. See Fluid catalytic cracking lubricant film, 70
FCC–air system, 27 multi-scale integration
flow regime diagrams for, 32–33 atomistic simulations for, 103
HD for, 30 coarse-grained MD models for,
heterogeneity index for, 30 103–107
Fertilizer site complex meso-scale/continuum level, 109–112
major facilities of, 148–149 simple reactive sphere model for,
optimization model 108–109
CO2 compressor, 149–150 nanotribology in, 69
of site steam system, 150 read/write head, 73
urea reactor, 149 structure of, 69
schematic representation of, 149 Harmonic potential energy, 77
Flory–Huggins lattice theory, 158 HDD. See Hard disk drive
Flow regime diagrams of CFB HDI. See Head disk interface
for air–FCC system and air–HGB Head disk interface
system, 32–33 components of, 68, 69
apparent and intrinsic, 33–34 cross-sectional diagram of, 70
dependency on riser height, 34 flow inside air bearing of
Fluid catalytic cracking, 43 Knudsen number flow regime,
Fluidized bed, factors affecting meso-scale 109–110
clusters in, 15 slip velocity on wall, 110
Fluidized bed reactors lubricant film, 70
multiphase flow in, 10–11 magnetic head in, 73–74
range of solids fraction, 5 multi-scale modeling of, 101
single particle for, 4 Heat transfer, EMMS model application in,
Fluidized systems simulated, physical 38–39
properties of, 18 Helmholtz energy model of mixing
Fluid–particle interactions, 4, 11 applications of, 156
Fomblin Z derivatives, 71 expression for, 159–160
molecular parameters in, 156
for multicomponent Ising mixture,
Gas and solid phases, slip velocity 163–166
between, 13 for polymers based on close-packed
Gas–liquid systems lattice model, 159–162
in bubble columns, 40 Helmholtz energy of mixing
EMMS modeling of, 40–43 of polymer systems, 167
Gas–solid suspensions for two-step mixing process, 172
direct numerical simulations of, 11 Heterogeneity index, 6
heterogeneous structures in, 13 Heterogeneous structures in gas–solid
Gas–solid systems suspensions, 13
DNS simulation of, 14 Hierarchical multi-scale model, 84
meso-scale effects of, 14 structure of
‘‘Global reaction’’, 3 atomistic/molecular level, 76–81
meso-scale/continuum level,
81–83
Hard disk drive process-scale level, 83–84
commercialized lubricant for, 70, 71 quantum level, 75–76
Subject Index 223
Homopolymer solution, lattice density Lattice cluster theory, 158–159

functional for Lattice density functional theory
equilibrium density distribution, 182 for homopolymer solution
excess Helmholtz energy functional, equilibrium density distribution, 182
179–181 excess Helmholtz energy functional,
grand potential, 181 179–181
at solid–liquid interface, 182–184 grand potential, 181
Hydrogen PEFC at solid–liquid interface, 182–184
components of, 63, 64 for polymer adsorption, 177–178
uses of, 63 for segment-density distributions,
working principle of, 63, 64 183–184
Lattice fluid model, 157
EOS based on, 175
IBM 3370 head, 73 Lattice model. See also Ising mixture
Industrial process models applications for phase equilibria
applications, 134–135 calculations, 173
critical success factors for successful, 135 lattice fluid molecular
fidelity of, 130–131 thermodynamic model, 174
maintenance of, 133–134 Flory–Huggins lattice theory, 158
for monitoring equipment/process grand potential for, 181
performance, 126 molecule arrangement in, 156–157
objectives, 124 problems associated with, 158
offline and online usage, 132–133 LBM. See Lattice Boltzmann method
for optimization, 127–128 LCT. See Lattice cluster theory
parameter estimation with, 126 LDFT. See Lattice density functional theory
and process economics, 132 LDFT equation
for reconciliation, 126–127 for equilibrium distribution, 181–182
scope of, 130 near a planar solid surface, 182–183
for simulation studies, 125–126 Lennard–Jones potential, 76
variables, 124–125 Linear programming (LP) models, 121
Interphase forces and reactor behavior, 8 Liquid film for CO2 capture with chemical
Intrinsic flow regime diagram for air–FCC absorbent, 141–142
system, 33 Liquid–liquid equilibria
‘‘Intrinsic reaction’’, 3–4 phase diagrams of ternary polymer
Ising mixture, 163 solutions, 170–171
Helmholtz energy of mixing for, 165–166 for [Rnmim][PF6] + Butan-1-ol system,
internal energy of mixing of, 164 175
Lubricant films
characteristics of ideal, 70
Knudsen number first line of protection from mechanical
of air bearing of HDI, 109 damage, 70
normalized velocity profiles at various PFPEs, 71
values of, 110–111 chemical structure of, 71
streamlines of cavity flow at, 111–112 and disk overcoat, interaction
between, 71–72
bonding mechanism between, 72
Lattice Boltzmann method, 83 DDPA-S, DDPA-D, and ZTMD, 72
kinetic models, 82
as multi-scale simulation tool, 81
for porous media flow simulation, 97 Macro-scale, 4
REV, 99 Magnetic head slider, 74
and SRS models, 91 Mass transfer
224 Subject Index
in CFB, 8–9 Monte Carlo simulation, 78

EMMS model application in, 35–38 of CMSC structure, 206, 209
MC simulation. See Monte Carlo of diblock copolymers confined in
simulation curved surfaces
MD. See Molecular dynamics between concentric curved surfaces,
MDEA. See n-Methyldiethanolamine 190–192
MDEA–CO2–water system, CO2 partial cylindrical pores, 187–190
pressures for, 139–140 Nlayervs. Rex/L0 in, phase separation of
MD simulation, atomistic, 87, 89 diblock copolymer, 201–204, 209
Meso-scale, 4 and SSL theory, conflict between, 206
Mesoscale clusters and dispersed particles, MRE. See Modified Reynolds equation
exchange between, 8 Multiphase chemical reactor, 2–4
Meso-scale/continuum-level modeling Multiphase flow in fluidized bed reactors,
tool, 81–83 10–11
Meso-scale modeling, macro-scale Multiphenomena in gas diffusion layer,
influence into, 15 97–102
Meso-scale structures, 2 ‘‘Multi-scale CFD’’
classic chemical engineering models for, applications
10 periodic domain simulations,
of copolymer materials, 155 16–21
critical effect of, 8–9 scope of, 23–24
drag coefficient and mass transfer for simulations of risers and validations,
CFB due to, 8–9 21–23
in gas–solid suspensions, 13 correlative
particle behaviour in, 4 challenges associated with, 14–15
related to processes, 155 computation cost effectiveness of, 13
spatiotemporal features of paradigms for, 12
dynamic characterizations, 6–8 for single-phase turbulent flows, 12
time-averaged characterization, 5–6 subparticle simulations using,
TFM grid refining and, 23–24 12–13
two-phase description of, 5–6 definition, 12
n-Methyldiethanolamine, 145 variational
CO2 capture with, 137–138 challenges to, 15–16
molecular structure of, 137 definition of, 15
Microkinetics-based reactor models, 135 scale separation condition in, 15
Micro-phase structure formation Multi-scale modeling
mechanism for block copolymers, 184 approaches, candidates for evaluating
Micro-scale, 4 HDD system, 66–74
MIP reactor, industrial PEFC, 63–66
flow regime diagram of, 46 at atomistic/molecular level, 76–81
simulation of, 45 bridging methodology, 85–87
solids volume fraction in laboratory- of chemical supply chain, 122
scale cold model of, 44 components, 74–75
Mixing process of chain-like molecular demand for research in, 113
systems, 160 at meso-scale/continuum level, 81–83
Modified Reynolds equation, 109 as multidisciplinary analysis paradigm,
Molecular dynamics, 78 60
and MC, 78 at process-scale level, 83–84
molecular motion in, 79–81 publications on, 62
Molecular system, 75 at quantum level, 75–76
Molecular thermodynamic model, 156 schematic description of, 61
Subject Index 225
Multi-scale models, 122 grid size estimation, 17

coupling of physical properties, 17–18
bridging procedure for, 85–87 time-averaged dimensionless slip
challenges associated with, 86 velocity
coarse-graining methods. See grid resolution effects on, 18–19
Coarse-graining methods periodic domain size effects on,
need for developing, 61 19, 21
publications on, 62 two-phase flow, 18
time and length scales in, 62 PFPE lubricant films, coarse-grained,
Multi-scale simulation, 61 bead-spring model of, 104–105
Multi-scale structures, 2 PFPE molecule
oligomeric, rigid units of, 89–90
PFPE Zdol molecule, molecular model
Nafion1, 65 of, 105
Nanoanalysis, advances in, 60 radius of gyration of, 108
Nanopore PFPEs, functional and nonfunctional
Helmholtz energy confined in, 187 endbead density profiles for, 106–107
layer transitions in, 204 spreading profile of SRS models with,
Negative pressure heads, 73 107
Nonlinear nonequilibrium system, 16 PFPE systems, 71
and disk overcoat, interaction between,
71–72
Objective functions, 127 bonding mechanism between, 72
Ono–Kondo equation, 177 DDPA-S, DDPA-D, and ZTMD, 72
Optimization methods, 127–128 molecular conformation of, 108
Physical system, multi-scale/holistic
interpretation of, 101
Parameter cases, 128–129 Piperazine, 137
Particle–particle interaction, 11 Plug-flow model, 10
PEFC. See Polymer electrolyte fuel cells PNIPAm gels, swelling ratio of, 175
PEFC-based power plant, process-level Polymer adsorption
model of, 64 based on lattice or off-lattice model, 176
PEFC model at interface, importance of, 176
device-level, 102 lattice-based theories for, 177
multiphenomena in gas diffusion layer, DFT, 177
97–102 general formalism for, 178–179
polymer electrolyte membrane LDFT, 177–182
ab initio models of, 93 at solid–liquid interface, 182–184
composition of, 92–93 Polymer chains, residual Helmholtz
water uptake variation in, 94–96 energy of dissociation and association
process-level, 102–103 of, 166–167
water management strategies in, 65–66 Polymer electrolyte fuel cells
PEM. See Polymer electrolyte membrane component of, 63
PEM materials design, 66
functions, 65 key issues in making paradigm
Nafion1, 65 shift in, 66
PEM systems, water management issues Polymer electrolyte membrane, 63
in, 66 ab initio models of, 93
Periodic domain simulations, periodic 2D components of, 92
domains, 16 composition of, 92–93
domain-size dependency of, 20 water uptake variation in, 94–96
grid-size dependency of, 20–21 Polymer systems
226 Subject Index
based on lattice fluid model, equation of Helmholtz energy of symmetrical parallel

state for, 171–173 lamellar confined in, 193–196
close-packed lattice model for, 159 ROMs. See Reduced-order models
comparisons with molecular simulation
results
coexistence curves, 168–170 Sector column structure
critical temperature and critical Helmholtz energy of, 198–199
volume fraction, 167–168 Semilean and lean solution columns, 147
liquid–liquid phase equilibria, Sequential modular (SM) modeling and
170–171 equation-oriented (EO) modeling, 123
Helmholtz energy of, 178 ‘‘Simulate’’ cases, 128
Helmholtz energy of mixing of, 167 Slip velocity
Polystyrene-b-polybutadiene (PS-b-PBD) asymptotic, 19
diblock copolymers grid resolution effects on, 18–19
confined in nanopore periodic domain size effects on, 19
comparison with MC simulation and Solid particles, heterogeneity in, 6
SSL theory, 201–204 SRS models, 89–90, 91
Helmholtz energy profiles of, 200–201 with spins, 108
morphologies of, 199 spreading profile of PFPEs, 107–108
Primary reformer feed steam to carbon SSL theory. See Strong Segregation Limit
(S/C) ratio, 147 theory
Process economics, 132 Static structures, microscale difference of, 5
Process-scale models, 83–84 Strong Segregation Limit theory, 185
PS/cyclohexane systems, spinodal curves for diblock copolymers confined in
and coexistence curves of, 174 ring-like curved surfaces, 193–199
PZ. See Piperazine Helmholtz energies predicted by,
205–206
Nlayervs. Rex/L0 in, 201–204
Quantum level models phase separation of confined diblock
coupling of, 87–89 copolymer, 209–210
quantum level models, 75–76 Subgrid structure modeling, 23
Subparticle simulations, 12–13
Symmetrical concentric-ring barrel
Real-time optimization (RTO) applications, structure
134 Helmholtz energy of, 193–196
Reconcile case, 129 Symmetrical concentric square column
Reconciliation models, 126–127 structure, Helmholtz energy of,
Reduced-order models 207–208
approximation errors, 87
different forms of, 86–87
linking models at various scales using, Ternary Ising lattice
85–86 internal energy of mixing for, 166
role of, 86 Ternary polymer solutions, liquid-liquid
Representative elementary volume (REV) equilibria phase diagrams of, 170–171
method, 97, 98 TFM. See Two-fluid model
Reverse Monte Carlo (RMC) techniques, 88 Thermal 40 mers, total segment-density
Ring-like curved surfaces, 192 distributions of, 184
Helmholtz energy of asymmetrical parallel Tribology, 66–67
lamellar confined in, 196–198 Turbulent flows, transfer of energy in, 15
Helmholtz energy of sector column Two-fluid model. See also Periodic domain
confined in, 198–199 simulations, periodic 2D domains
Subject Index 227
applicability for for bubbling fluidized challenges to, 15–16

bed, 23 and correlative, comparison between, 17
fine-grid and coarse-grid, 11 definition of, 15
grid refining and meso-scale structures, scale separation condition in, 15
23–24
Two-step mixing process, 172
Water uptake variation in PEM, 94–96
Vapor–liquid equilibria for propanol +

[Me3BuN][NTf2] system, 175 Ztetraol multidentate, 72
Variational multi-scale CFD ZTMD. See Ztetraol multidentate
CONTENTS OF VOLUMES IN THIS SERIAL
Volume 1 (1956)
J. W. Westwater, Boiling of Liquids
A. B. Metzner, Non-Newtonian Technology: Fluid Mechanics, Mixing, and Heat Transfer
R. Byron Bird, Theory of Diffusion
J. B. Opfell and B. H. Sage, Turbulence in Thermal and Material Transport
Robert E. Treybal, Mechanically Aided Liquid Extraction
Robert W. Schrage, The Automatic Computer in the Control and Planning of Manufacturing
Operations
Ernest J. Henley and Nathaniel F. Barr, Ionizing Radiation Applied to Chemical Processes and to
Food and Drug Processing
Volume 2 (1958)
J. W. Westwater, Boiling of Liquids
Ernest F. Johnson, Automatic Process Control
Bernard Manowitz, Treatment and Disposal of Wastes in Nuclear Chemical Technology
George A. Sofer and Harold C. Weingartner, High Vacuum Technology
Theodore Vermeulen, Separation by Adsorption Methods
Sherman S. Weidenbaum, Mixing of Solids
Volume 3 (1962)
C. S. Grove, Jr., Robert V. Jelinek, and Herbert M. Schoen, Crystallization from Solution
F. Alan Ferguson and Russell C. Phillips, High Temperature Technology
Daniel Hyman, Mixing and Agitation
John Beck, Design of Packed Catalytic Reactors
Douglass J. Wilde, Optimization Methods
Volume 4 (1964)
J. T. Davies, Mass-Transfer and Inierfacial Phenomena
R. C. Kintner, Drop Phenomena Affecting Liquid Extraction
Octave Levenspiel and Kenneth B. Bischoff, Patterns of Flow in Chemical Process Vessels
Donald S. Scott, Properties of Concurrent Gas–Liquid Flow
D. N. Hanson and G. F. Somerville, A General Program for Computing Multistage Vapor–Liquid
Processes
Advances in Heterocyclic Chemistry, Volume 102 Ó 2011 Elsevier Inc.

ISSN 0065-2725, DOI All rights reserved
229
230 Contents of Volumes in this Serial
Volume 5 (1964)
J. F. Wehner, Flame Processes—Theoretical and Experimental
J. H. Sinfelt, Bifunctional Catalysts
S. G. Bankoff, Heat Conduction or Diffusion with Change of Phase
George D. Fulford, The Flow of Lktuids in Thin Films
K. Rietema, Segregation in Liquid–Liquid Dispersions and its Effects on Chemical Reactions
Volume 6 (1966)
S. G. Bankoff, Diffusion-Controlled Bubble Growth
John C. Berg, Andreas Acrivos, and Michel Boudart, Evaporation Convection
H. M. Tsuchiya, A. G. Fredrickson, and R. Aris, Dynamics of Microbial Cell Populations
Samuel Sideman, Direct Contact Heat Transfer between Immiscible Liquids
Howard Brenner, Hydrodynamic Resistance of Particles at Small Reynolds Numbers
Volume 7 (1968)
Robert S. Brown, Ralph Anderson, and Larry J. Shannon, Ignition and
Combustion of Solid Rocket Propellants
Knud Østergaard, Gas–Liquid–Particle Operations in Chemical Reaction Engineering
J. M. Prausnilz, Thermodynamics of Fluid–Phase Equilibria at High Pressures
Robert V. Macbeth, The Burn-Out Phenomenon in Forced-Convection Boiling
William Resnick and Benjamin Gal-Or, Gas–Liquid Dispersions
Volume 8 (1970)
C. E. Lapple, Electrostatic Phenomena with Particulates
J. R. Kittrell, Mathematical Modeling of Chemical Reactions
W. P. Ledet and D. M. Himmelblau, Decomposition Procedures foe the Solving of Large Scale
Systems
R. Kumar and N. R. Kuloor, The Formation of Bubbles and Drops
Volume 9 (1974)
Renato G. Bautista, Hydrometallurgy
Kishan B. Mathur and Norman Epstein, Dynamics of Spouted Beds
W. C. Reynolds, Recent Advances in the Computation of Turbulent Flows
R. E. Peck and D. T. Wasan, Drying of Solid Particles and Sheets
Volume 10 (1978)
G. E. O’Connor and T. W. F. Russell, Heat Transfer in Tubular Fluid–Fluid Systems
P. C. Kapur, Balling and Granulation
Richard S. H. Mah and Mordechai Shacham, Pipeline Network Design and Synthesis
J. Robert Selman and Charles W. Tobias, Mass-Transfer Measurements by the Limiting-Current
Technique
Contents of Volumes in this Serial 231
Volume 11 (1981)
Jean-Claude Charpentier, Mass-Transfer Rates in Gas–Liquid Absorbers and Reactors
Dee H. Barker and C. R. Mitra, The Indian Chemical Industry—Its Development and Needs
Lawrence L. Tavlarides and Michael Stamatoudis, The Analysis of Interphase Reactions and Mass
Transfer in Liquid–Liquid Dispersions
Terukatsu Miyauchi, Shintaro Furusaki, Shigeharu Morooka, and Yoneichi Ikeda, Transport
Phenomena and Reaction in Fluidized Catalyst Beds
Volume 12 (1983)
C. D. Prater, J, Wei, V. W. Weekman, Jr., and B. Gross, A Reaction Engineering Case History: Coke
Burning in Thermofor Catalytic Cracking Regenerators
Costel D. Denson, Stripping Operations in Polymer Processing
Robert C. Reid, Rapid Phase Transitions from Liquid to Vapor
John H. Seinfeld, Atmospheric Diffusion Theory
Volume 13 (1987)
Edward G. Jefferson, Future Opportunities in Chemical Engineering
Eli Ruckenstein, Analysis of Transport Phenomena Using Scaling and Physical Models
Rohit Khanna and John H. Seinfeld, Mathematical Modeling of Packed Bed Reactors: Numerical
Solutions and Control Model Development
Michael P. Ramage, Kenneth R. Graziano, Paul H. Schipper, Frederick J. Krambeck, and
Byung C. Choi, KINPTR (Mobil’s Kinetic Reforming Model): A Review of Mobil’s Industrial
Process Modeling Philosophy
Volume 14 (1988)
Richard D. Colberg and Manfred Morari, Analysis and Synthesis of Resilient Heat Exchange
Networks
Richard J. Quann, Robert A. Ware, Chi-Wen Hung, and James Wei, Catalytic Hydrometallation
of Petroleum
Kent David, The Safety Matrix: People Applying Technology to Yield Safe Chemical Plants and
Products
Volume 15 (1990)
Pierre M. Adler, Ali Nadim, and Howard Brenner, Rheological Models of Suspenions
Stanley M. Englund, Opportunities in the Design of Inherently Safer Chemical Plants
H. J. Ploehn and W. B. Russel, Interations between Colloidal Particles and Soluble Polymers
Volume 16 (1991)
Perspectives in Chemical Engineering: Research and Education
Clark K. Colton, Editor
Historical Perspective and Overview
L. E. Scriven, On the Emergence and Evolution of Chemical Engineering
Ralph Landau, Academic—industrial Interaction in the Early Development of Chemical Engineering
James Wei, Future Directions of Chemical Engineering
Fluid Mechanics and Transport
L. G. Leal, Challenges and Opportunities in Fluid Mechanics and Transport Phenomena
William B. Russel, Fluid Mechanics and Transport Research in Chemical Engineering
J. R. A. Pearson, Fluid Mechanics and Transport Phenomena
Thermodynamics
Keith E. Gubbins, Thermodynamics
J. M. Prausnitz, Chemical Engineering Thermodynamics: Continuity and Expanding Frontiers
H. Ted Davis, Future Opportunities in Thermodynamics

Kinetics, Catalysis, and Reactor Engineering
Alexis T. Bell, Reflections on the Current Status and Future Directions of Chemical Reaction
Engineering
James R. Katzer and S. S. Wong, Frontiers in Chemical Reaction Engineering
L. Louis Hegedus, Catalyst Design
Environmental Protection and Energy
John H. Seinfeld, Environmental Chemical Engineering
T. W. F. Russell, Energy and Environmental Concerns
Janos M. Beer, Jack B. Howard, John P. Longwell, and Adel F. Sarofim, The Role of Chemical
Engineering in Fuel Manufacture and Use of Fuels
Polymers
Matthew Tirrell, Polymer Science in Chemical Engineering
Richard A. Register and Stuart L. Cooper, Chemical Engineers in Polymer Science: The Need for an
Interdisciplinary Approach
Microelectronic and Optical Material
Larry F. Thompson, Chemical Engineering Research Opportunities in Electronic and Optical
Materials Research
Klavs F. Jensen, Chemical Engineering in the Processing of Electronic and Optical Materials: A
Discussion
Bioengineering
James E. Bailey, Bioprocess Engineering
Arthur E. Humphrey, Some Unsolved Problems of Biotechnology
Channing Robertson, Chemical Engineering: Its Role in the Medical and Health Sciences
Process Engineering
Arthur W. Westerberg, Process Engineering
Manfred Morari, Process Control Theory: Reflections on the Past Decade and Goals for the Next
James M. Douglas, The Paradigm After Next
George Stephanopoulos, Symbolic Computing and Artificial Intelligence in Chemical Engineering:
A New Challenge
The Identity of Our Profession
Morton M. Denn, The Identity of Our Profession
Volume 17 (1991)
Y. T. Shah, Design Parameters for Mechanically Agitated Reactors
Mooson Kwauk, Particulate Fluidization: An Overview
Volume 18 (1992)
E. James Davis, Microchemical Engineering: The Physics and Chemistry of the Microparticle
Selim M. Senkan, Detailed Chemical Kinetic Modeling: Chemical Reaction Engineering of the Future
Lorenz T. Biegler, Optimization Strategies for Complex Process Models
Volume 19 (1994)
Robert Langer, Polymer Systems for Controlled Release of Macromolecules, Immobilized Enzyme
Medical Bioreactors, and Tissue Engineering
J. J. Linderman, P. A. Mahama, K. E. Forsten, and D. A. Lauffenburger, Diffusion and Probability
in Receptor Binding and Signaling
Rakesh K. Jain, Transport Phenomena in Tumors
R. Krishna, A Systems Approach to Multiphase Reactor Selection
David T. Allen, Pollution Prevention: Engineering Design at Macro-, Meso-, and Microscales
John H. Seinfeld, Jean M. Andino, Frank M. Bowman, Hali J. L. Forstner, and Spyros Pandis,
Tropospheric Chemistry
Volume 20 (1994)
Arthur M. Squires, Origins of the Fast Fluid Bed
Yu Zhiqing, Application Collocation
Youchu Li, Hydrodynamics
Li Jinghai, Modeling
Yu Zhiqing and Jin Yong, Heat and Mass Transfer
Mooson Kwauk, Powder Assessment
Li Hongzhong, Hardware Development
Youchu Li and Xuyi Zhang, Circulating Fluidized Bed Combustion
Chen Junwu, Cao Hanchang, and Liu Taiji, Catalyst Regeneration in Fluid Catalytic Cracking
Volume 21 (1995)
Christopher J. Nagel, Chonghum Han, and George Stephanopoulos, Modeling Languages:
Declarative and Imperative Descriptions of Chemical Reactions and Processing Systems
Chonghun Han, George Stephanopoulos, and James M. Douglas, Automation in Design: The
Conceptual Synthesis of Chemical Processing Schemes
Michael L. Mavrovouniotis, Symbolic and Quantitative Reasoning: Design of Reaction Pathways
through Recursive Satisfaction of Constraints
Christopher Nagel and George Stephanopoulos, Inductive and Deductive Reasoning: The Case of
Identifying Potential Hazards in Chemical Processes
Keven G. Joback and George Stephanopoulos, Searching Spaces of Discrete Soloutions: The Design
of Molecules Processing Desired Physical Properties
Volume 22 (1995)
Chonghun Han, Ramachandran Lakshmanan, Bhavik Bakshi, and George Stephanopoulos,
Nonmonotonic Reasoning: The Synthesis of Operating Procedures in Chemical Plants
Pedro M. Saraiva, Inductive and Analogical Learning: Data-Driven Improvement of Process
Operations
Alexandros Koulouris, Bhavik R. Bakshi and George Stephanopoulos, Empirical Learning
through Neural Networks: The Wave-Net Solution
Bhavik R. Bakshi and George Stephanopoulos, Reasoning in Time: Modeling, Analysis, and
Pattern Recognition of Temporal Process Trends
Matthew J. Realff, Intelligence in Numerical Computing: Improving Batch Scheduling Algorithms
through Explanation-Based Learning
Volume 23 (1996)
Jeffrey J. Siirola, Industrial Applications of Chemical Process Synthesis
Arthur W. Westerberg and Oliver Wahnschafft, The Synthesis of Distillation-Based Separation
Systems
Ignacio E. Grossmann, Mixed-Integer Optimization Techniques for Algorithmic Process Synthesis
Subash Balakrishna and Lorenz T. Biegler, Chemical Reactor Network Targeting and Integration:
An Optimization Approach
Steve Walsh and John Perkins, Operability and Control inn Process Synthesis and Design
Volume 24 (1998)
Raffaella Ocone and Gianni Astarita, Kinetics and Thermodynamics in Multicomponent Mixtures
Arvind Varma, Alexander S. Rogachev, Alexandra S. Mukasyan, and Stephen Hwang,
Combustion Synthesis of Advanced Materials: Principles and Applications
J. A. M. Kuipers and W. P. Mo, van Swaaij, Computional Fluid Dynamics Applied to Chemical
Reaction Engineering
Ronald E. Schmitt, Howard Klee, Debora M. Sparks, and Mahesh K. Podar, Using Relative Risk
Analysis to Set Priorities for Pollution Prevention at a Petroleum Refinery
Volume 25 (1999)
J. F. Davis, M. J. Piovoso, K. A. Hoo, and B. R. Bakshi, Process Data Analysis and Interpretation
J. M. Ottino, P. DeRoussel, S., Hansen, and D. V. Khakhar, Mixing and Dispersion of Viscous
Liquids and Powdered Solids
Peter L. Silverston, Li Chengyue, Yuan Wei-Kang, Application of Periodic Operation to Sulfur
Dioxide Oxidation
Volume 26 (2001)
J. B. Joshi, N. S. Deshpande, M. Dinkar, and D. V. Phanikumar, Hydrodynamic Stability of
Multiphase Reactors
Michael Nikolaou, Model Predictive Controllers: A Critical Synthesis of Theory and Industrial Needs
Volume 27 (2001)
William R. Moser, Josef Find, Sean C. Emerson, and Ivo M, Krausz, Engineered Synthesis of
Nanostructure Materials and Catalysts
Bruce C. Gates, Supported Nanostructured Catalysts: Metal Complexes and Metal Clusters
Ralph T. Yang, Nanostructured Absorbents
Thomas J. Webster, Nanophase Ceramics: The Future Orthopedic and Dental Implant Material
Yu-Ming Lin, Mildred S. Dresselhaus, and Jackie Y. Ying, Fabrication, Structure, and Transport
Properties of Nanowires
Volume 28 (2001)
Qiliang Yan and Juan J. DePablo, Hyper-Parallel Tempering Monte Carlo and Its Applications
Pablo G. Debenedetti, Frank H. Stillinger, Thomas M. Truskett, and Catherine P. Lewis, Theory
of Supercooled Liquids and Glasses: Energy Landscape and Statistical Geometry Perspectives
Michael W. Deem, A Statistical Mechanical Approach to Combinatorial Chemistry
Venkat Ganesan and Glenn H. Fredrickson, Fluctuation Effects in Microemulsion Reaction Media
David B. Graves and Cameron F. Abrams, Molecular Dynamics Simulations of Ion–Surface
Interactions with Applications to Plasma Processing
Christian M. Lastoskie and Keith E, Gubbins, Characterization of Porous Materials Using
Molecular Theory and Simulation
Dimitrios Maroudas, Modeling of Radical-Surface Interactions in the Plasma-Enhanced Chemical
Vapor Deposition of Silicon Thin Films
Sanat Kumar, M. Antonio Floriano, and Athanassiors Z. Panagiotopoulos, Nanostructured
Formation and Phase Separation in Surfactant Solutions
Stanley I. Sandler, Amadeu K. Sum, and Shiang-Tai Lin, Some Chemical Engineering
Applications of Quantum Chemical Calculations
Bernhardt L. Trout, Car-Parrinello Methods in Chemical Engineering: Their Scope and potential
R. A. van Santen and X. Rozanska, Theory of Zeolite Catalysis
Zhen-Gang Wang, Morphology, Fluctuation, Metastability and Kinetics in Ordered Block
Copolymers
Volume 29 (2004)
Michael V. Sefton, The New Biomaterials
Kristi S. Anseth and Kristyn S. Masters, Cell–Material Interactions
Surya K. Mallapragada and Jennifer B. Recknor, Polymeric Biomaterias for Nerve Regeneration
Anthony M. Lowman, Thomas D. Dziubla, Petr Bures, and Nicholas A. Peppas, Structural and
Dynamic Response of Neutral and Intelligent Networks in Biomedical Environments
F. Kurtis Kasper and Antonios G. Mikos, Biomaterials and Gene Therapy
Balaji Narasimhan and Matt J. Kipper, Surface-Erodible Biomaterials for Drug Delivery
Volume 30 (2005)
Dionisio Vlachos, A Review of Multiscale Analysis: Examples from System Biology, Materials
Engineering, and Other Fluids-Surface Interacting Systems
Lynn F. Gladden, M.D. Mantle and A.J. Sederman, Quantifying Physics and Chemistry at
Multiple Length- Scales using Magnetic Resonance Techniques
Juraj Kosek, Frantisek Steěpanek, and Miloš Marek, Modelling of Transport and
Transformation
Processes in Porous and Multiphase Bodies
Vemuri Balakotaiah and Saikat Chakraborty, Spatially Averaged Multiscale Models for Chemical
Reactors
Volume 31 (2006)
Yang Ge and Liang-Shih Fan, 3-D Direct Numerical Simulation of Gas–Liquid and Gas–Liquid–
Solid Flow Systems Using the Level-Set and Immersed-Boundary Methods
M.A. van der Hoef, M. Ye, M. van Sint Annaland, A.T. Andrews IV, S. Sundaresan, and J.A.M.
Kuipers, Multiscale Modeling of Gas-Fluidized Beds
Harry E.A. Van den Akker, The Details of Turbulent Mixing Process and their Simulation
Rodney O. Fox, CFD Models for Analysis and Design of Chemical Reactors
Anthony G. Dixon, Michiel Nijemeisland, and E. Hugh Stitt, Packed Tubular Reactor Modeling
and Catalyst Design Using Computational Fluid Dynamics
Volume 32 (2007)
William H. Green, Jr., Predictive Kinetics: A New Approach for the 21st Century
Mario Dente, Giulia Bozzano, Tiziano Faravelli, Alessandro Marongiu, Sauro Pierucci and
Eliseo Ranzi, Kinetic Modelling of Pyrolysis Processes in Gas and Condensed Phase
Mikhail Sinev, Vladimir Arutyunov and Andrey Romanets, Kinetic Models of C1–C4 Alkane
Oxidation as Applied to Processing of Hydrocarbon Gases: Principles, Approaches and
Developments
Pierre Galtier, Kinetic Methods in Petroleum Process Engineering
Volume 33 (2007)
Shinichi Matsumoto and Hirofumi Shinjoh, Dynamic Behavior and Characterization of Automobile
Catalysts
Mehrdad Ahmadinejad, Maya R. Desai, Timothy C. Watling and Andrew P.E. York,
Simulation of Automotive Emission Control Systems
Anke G€ uthenke, Daniel Chatterjee, Michel Weibel, Bernd Krutzsch, Petr Kocı, Miloš Marek,
Isabella Nova and Enrico Tronconi, Current Status of Modeling Lean Exhaust Gas
Aftertreatment Catalysts
Athanasios G. Konstandopoulos, Margaritis Kostoglou, Nickolas Vlachos and Evdoxia
Kladopoulou, Advances in the Science and Technology of Diesel Particulate Filter Simulation
Volume 34 (2008)
C.J. van Duijn, Andro Mikelic, I.S. Pop, and Carole Rosier, Effective Dispersion Equations for
Reactive Flows with Dominant Peclet and Damkohler Numbers
Mark Z. Lazman and Gregory S. Yablonsky, Overall Reaction Rate Equation of Single-Route
Complex Catalytic Reaction in Terms of Hypergeometric Series
A.N. Gorban and O. Radulescu, Dynamic and Static Limitation in Multiscale Reaction Networks,
Revisited
Liqiu Wang, Mingtian Xu, and Xiaohao Wei, Multiscale Theorems
Volume 35 (2009)
Rudy J. Koopmans and Anton P.J. Middelberg, Engineering Materials from the Bottom Up –
Overview
Robert P.W. Davies, Amalia Aggeli, Neville Boden, Tom C.B. McLeish, Irena A. Nyrkova, and
Alexander N. Semenov, Mechanisms and Principles of 1 D Self-Assembly of Peptides into
b-Sheet Tapes
Paul van der Schoot, Nucleation and Co-Operativity in Supramolecular Polymers
Michael J. McPherson, Kier James, Stuart Kyle, Stephen Parsons, and Jessica Riley,
Recombinant Production of Self-Assembling Peptides
Boxun Leng, Lei Huang, and Zhengzhong Shao, Inspiration from Natural Silks and Their Proteins
Sally L. Gras, Surface- and Solution-Based Assembly of Amyloid Fibrils for Biomedical and
Nanotechnology Applications
Conan J. Fee, Hybrid Systems Engineering: Polymer-Peptide Conjugates
Volume 36 (2009)
Vincenzo Augugliaro, Sedat Yurdakal, Vittorio Loddo, Giovanni Palmisano, and Leonardo
Palmisano, Determination of Photoadsorption Capacity of Polychrystalline TiO2 Catalyst in
Irradiated Slurry
Marta I. Litter, Treatment of Chromium, Mercury, Lead, Uranium, and Arsenic in Water by
Heterogeneous Photocatalysis
Aaron Ortiz-Gomez, Benito Serrano-Rosales, Jesus Moreira-del-Rio, and Hugo de-Lasa,
Mineralization of Phenol in an Improved Photocatalytic Process Assisted with Ferric
Ions: Reaction Network and Kinetic Modeling
R.M. Navarro, F. del Valle, J.A. Villoria de la Mano, M.C. Alvarez-Galvan, and J.L.G. Fierro,
Photocatalytic Water Splitting Under Visible Light: Concept and Catalysts Development
Ajay K. Ray, Photocatalytic Reactor Configurations for Water Purification: Experimentation and
Modeling
Camilo A. Arancibia-Bulnes, Antonio E. Jiménez, and Claudio A. Estrada, Development and
Modeling of Solar Photocatalytic Reactors
Orlando M. Alfano and Alberto E. Cassano, Scaling-Up of Photoreactors: Applications to
Advanced Oxidation Processes
Yaron Paz, Photocatalytic Treatment of Air: From Basic Aspects to Reactors
Volume 37 (2009)
S. Roberto Gonzalez A., Yuichi Murai, and Yasushi Takeda, Ultrasound-Based Gas–Liquid
Interface Detection in Gas–Liquid Two-Phase Flows
Z. Zhang, J. D. Stenson, and C. R. Thomas, Micromanipulation in Mechanical Characterisation of
Single Particles
Feng-Chen Li and Koichi Hishida, Particle Image Velocimetry Techniques and Its Applications in
Multiphase Systems
J. P. K. Seville, A. Ingram, X. Fan, and D. J. Parker, Positron Emission Imaging in Chemical Engineering
Fei Wang, Qussai Marashdeh, Liang-Shih Fan, and Richard A. Williams, Electrical Capacitance,
Electrical Resistance, and Positron Emission Tomography Techniques and Their Applications in
Multi-Phase Flow Systems
Alfred Leipertz and Roland Sommer, Time-Resolved Laser-Induced Incandescence
Volume 38 (2009)
Arata Aota and Takehiko Kitamori, Microunit Operations and Continuous Flow Chemical
Processing
gıral and Han J.G.E. Gardeniers, Microreactors with Electrical Fields
Anıl A
Charlotte Wiles and Paul Watts, High-Throughput Organic Synthesis in Microreactors
S. Krishnadasan, A. Yashina, A.J. deMello and J.C. deMello, Microfluidic Reactors for
Nanomaterial Synthesis
Volume 39 (2010)
B.M. Kaganovich, A.V. Keiko and V.A. Shamansky, Equilibrium Thermodynamic Modeling of
Dissipative Macroscopic Systems
Miroslav Grmela, Multiscale Equilibrium and Nonequilibrium Thermodynamics in Chemical
Engineering
Prasanna K. Jog, Valeriy V. Ginzburg, Rakesh Srivastava, Jeffrey D. Weinhold, Shekhar Jain,
and Walter G. Chapman, Application of Mesoscale Field-Based Models to Predict Stability of
Particle Dispersions in Polymer Melts
Semion Kuchanov, Principles of Statistical Chemistry as Applied to Kinetic Modeling of Polymer-
Obtaining Processes
Volume 40 (2011)
Annette Karmiloff-Smith, Birkbeck Centre for Brain & Cognitive Development, University of
London, London, UK
Deborah J. Fidler, Erika Lunkenheimer, and Laura Hahn, Colorado State University, Fort Collins,
Colorado, USA
Victoria C.P. Knowland and Michael S.C. Thomas, Developmental Neurocognition Lab, Centre for
Brain and Cognitive Development, Department of Psychological Sciences, Birkbeck, University of
London, London, UK
Donald B. Bailey, RTI International, USA
Heather Cody Hazlett and Anne C. Wheeler, University of North Carolina at Chapel Hill, Chapel
Hill, North Carolina, USA
Jane E. Roberts, University of South Carolina, South Carolina, USA
Lisa Daunhauer and Deborah Fidler, Occupational Therapy Department, Colorado State
University, Colorado, USA
Nancy Raitano Lee, Katherine C. Lopez, Elizabeth I. Adeyemi, and Jay N. Giedd, Child
Psychiatry Branch, National Institute of Mental Health, NIH, Bethesda, Maryland, USA
Susan L. Hepburn and Amy Philofsky, University of Colorado Denver Health Sciences Center,
Denver, Colorado, USA
Deborah Fidler and Laura Hahn, Colorado State University, Colorado, USA
Lisa G. Shaffer, Signature Genomic Laboratories, Spokane, Washington, USA
Susan L. Hepburn and, Eric J. Moody, Department of Psychiatry, JFK Partners, Colorado
Intellectual and Developmental Disabilities Research Center (IDDRC), University of Colorado
School of Medicine, Aurora, Colorado, USA
Karen Riley, Child, Family and School Psychology, The University of Denver, Denver, USA

(Advances in Chemical Engineering 40) Guy B. Marin (Eds.) - Multiscale Simulation and Design-Elsevier - Academic Press (2011)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Advances in Chemical Engineering 40) Guy B. Marin (Eds.) - Multiscale Simulation and Design-Elsevier - Academic Press (2011)

Uploaded by

Copyright:

Available Formats

Academic Press is an imprint of Elsevier

Linacre House, Jordan Hill, Oxford OX2 8DP, UK

First edition 2011

Copyright Ó 2011, Elsevier Inc. All rights reserved

No part of this publication may be reproduced, stored in a retrieval system

For information on all Academic Press publications

Printed and bound in USA

Multi-scale Simulation and Design

The increasing importance of multi-scale computation in engineering,

models from molecular level to site-wide chemical complex or enterprise

Amsterdam * Boston * Heidelberg * London * New York * Oxford

Lorenz T. Biegler, Department of Chemical Engineering, Carnegie Mellon

Chau-Chyun Chen, Aspen Technology, Inc., Burlington, MA 01803, USA

Xueqian Chen, State Key Laboratory of Chemical Engineering and Department of

Pil Seung Chung, Department of Chemical Engineering, Carnegie Mellon

Boyd Gochenour, Aspen Technology, Inc., Burlington, MA 01803, USA

Ying Hu, State Key Laboratory of Chemical Engineering and Department of

Yongmin Huang, State Key Laboratory of Chemical Engineering and

Myung S. Jhon, Department of Chemical Engineering, Carnegie Mellon

Honglai Liu, State Key Laboratory of Chemical Engineering and Department of

Milo D. Meixell, Aspen Technology, Inc., Burlington, MA 01803, USA

Xingqin Xiao, State Key Laboratory of Chemical Engineering and Department of

Contents 1. Meso-Scale Structure—A Common Challenge for 2

Abstract Meso-scale structure is critical to characterize complex systems in

1. MESO-SCALE STRUCTURE—A COMMON CHALLENGE

Chemical engineering encompasses a broad spectrum of scales with

Figure 1 Multi-scale characteristics of chemical engineering and its multilevel

ranging from molecules to the reactor (Dudukovic, 2009; Li and Kwauk,

without mass transfer, is normally termed as the ‘‘intrinsic reaction’’

1.2 Micro-, meso-, and macro-scales

the effects of the meso-scale structures. To this end, in-depth under-

1.3 Meso-scale structure—spatiotemporal features

1.3.1 Time-averaged characterization

Li (1991). All of these, however, suggest that reducing the meso-scale

1.3.2 Dynamic characterization

Figure 2 X-ray measurement of cluster acceleration (Meng et al., 2009).

along the riser wall of a CFB. As shown in Figure 2, by cross-correlation

Besides the time-averaged and the dynamic characterizations, the

1.4 Critical effect of meso-scale structure

2. MULTI-SCALE CFD—SOLUTIONS FOR HIGH

To capture the meso-scale structure and/or to predict its effects, various

2.1 Single-scale approach—its limitation on predictability

2.2 Direct numerical simulations—its limitation on scalability

space resolution of the governing equations should be smaller than the

2.3 Multi-scale CFD approach—a compromise of predictability

tumbling’’ process describing the instability of sedimentation of few

Figure 4 Snapshot from a dynamic DNS simulation of two-dimensional gas–solid

meso-scale effects of gas–solid systems, which is otherwise a formidable

above, the meso-scale structure is characterized with dynamic varia-

broad range of scales for different systems. For a thermodynamic equi-

2.3.3 Applications of the multi-scale CFD

Correlative (e.g., closure for Variational (e.g., coupling

Characteristics * Clear scale separation is * Clear scale separation is

Table 2 Physical properties of the fluidized systems simulated

Group A Group B Group D

Particle diameter dp, m 75 300 1,020

Figure 5 Effect of grid resolution (l) on the time-averaged dimensionless slip

particles are more uniformly distributed. More efforts are needed to

confirming that the Model M accurately captures the mechanism

2.3.3.2 Simulations of risers and validations. Two CFB risers were

Figure 8 Axial profiles of cross-sectionally averaged voidage under different grid

2.3.3.3 Scope of applications. In literature, there are some other reports

3. MESO-SCALE MODELING—THE KEY TO MULTI-SCALE