Comput Mech

DOI 10.1007/s00466-012-0712-z

ORIGINAL PAPER

SUPG and discontinuity-capturing methods for coupled fluid

mechanics and electrochemical transport problems
Pablo A. Kler · Lisandro D. Dalcin · Rodrigo R. Paz ·
Tayfun E. Tezduyar

Received: 21 February 2012 / Accepted: 5 April 2012

© Springer-Verlag 2012

Abstract Electrophoresis is the motion of charged particles To demonstrate the effectiveness of the stabilized formula-
relative to the surrounding liquid under the influence of an tion, we present test computations with 1D, 2D, and 3D elec-
external electric field. This electrochemical transport process trophoretic transport problems of technological interest.
is used in many scientific and technological areas to separate
chemical species. Modeling and simulation of electropho- Keywords Electrophoresis · Fluid mechanics · Electro-
retic transport enables a better understanding of the physi- chemical transport · Finite element computation · SUPG
cochemical processes developed during the electrophoretic stabilization · Discontinuity capturing
separations and the optimization of various parameters of
the electrophoresis devices and their performance. Electro-
phoretic transport is a multiphysics and multiscale problem. 1 Introduction
Mass transport, fluid mechanics, electric problems, and their
interactions have to be solved in domains with length scales Electrophoresis is a chemical transport process where
ranging from nanometers to centimeters. We use a finite charged particles move relative to the surrounding liquid
element method for the computations. Without proper numer- under the influence of an externally applied electric field.
ical stabilization, computation of coupled fluid mechanics, The magnitude of the relative velocity depends on the elec-
electrophoretic transport, and electric problems would suffer trophoretic mobility of charged molecules. Electrophoretic
from spurious oscillations that are related to the high values mobility is an outcome of the balance between electric and
of the local Péclet and Reynolds numbers and the nonzero hydrodynamic friction forces, the latter being due to the vis-
divergence of the migration field. To overcome these com- cosity of the surrounding solution.
putational challenges, we propose a stabilized finite element The most common applications of electrophoresis are
method based on the Streamline-Upwind/Petrov-Galerkin separation processes. They are based on the separation of
(SUPG) formulation and discontinuity-capturing techniques. molecules due to differences in electrophoretic mobility of
ions under an external electric field applied at the ends
P. A. Kler (B) of capillary tubes, microchannels, or microchambers [39].
Central Division of Analytical Chemistry, Forschungszentrum
Jülich, Wilhelm-Johnen-Str, 52428 Jülich, Germany
Electrophoretic separations comprise a group of different
e-mail: p.kler@fz-juelich.de techniques, such as capillary zone electrophoresis, isoelec-
tric focusing, isotachophoresis, free-flow electrophoresis,
L. D. Dalcin · R. R. Paz and capillary electrochromatography [38,64], which are
Centro Internacional de Métodos Computacionales en Ingeniería
Instituto de Desarrollo Tecnológico para la Industria Química,
extensively used in chemical and biochemical analysis in
Universidad Nacional del Litoral—Consejo Nacional many scientific areas, such as genetics, molecular biology,
de Investigaciones Científicas y Técnicas, Colectora pharmacology, and environment monitoring.
RN 168 Km 472, S3000GLN Santa Fe, Argentina Foundations of electrophoretic separations are in the
T. E. Tezduyar
electrokinetic phenomena that develop when electrolyte
Mechanical Engineering, Rice University, MS 321, 6100 Main Street, solutions interact with charged surfaces. Generally, most sub-
Houston, TX 77005, USA stances will acquire a surface electric charge when brought

123
 Comput Mech

into contact with an aqueous (polar) medium. The effect of and Albrecht et al. [1], using 2D domains and flux corrected
any charged surface (solid walls or molecules in the solution) transport method.
in an electrolyte solution will be to influence the distribu- Numerical simulation of electrophoretic separations in
tion of nearby ions in the solution. Ions of opposite charge microfluidic chips represents a computationally challeng-
to that surface (counterions) are attracted towards the sur- ing problem. The large differences between the length scales
face while ions of like charge (coions) are repelled from involved and the multiphysics nature of the problem lead to
the surface, leading to the formation of an electric double numerical difficulties, such as multiple nonlinear problems,
layer (EDL) [28]. The EDL is a region close to the charged excessive number of degrees of freedom, and global matrices
surface where there exists an excess of counterions over ill-conditioned because of the high aspect ratios [29]. There-
coions to neutralize the surface charge. When an external fore, implementations for parallel computing and advanced
electric field is applied in the axial direction of a channel, preconditioning techniques are imperative to achieving accu-
as in the case of electrophoretic separation, the electrical rate numerical results with low computation times. Tsai
forces acting on unbalanced ions drag the surrounding liquid, et al. [63] presented a 2D parallel finite volume algorithm
and electroosmotic flow (EOF) and electrophoresis develop. to solve EOF in L-shaped microchannels. Chau et al. [11]
Consequently, when modeling electrophoretic transport, it reported 3D parallel computation of electrophoretic pro-
is necessary to include the EDL and EOF to have a com- cesses for free-flow electrophoresis using a finite difference
prehensive approach. This comprehensive approach has to method. Kler et al. [31,32] presented finite element simula-
include the different effects present in the electrophoretic tions for electroosmotic flow, capillary zone electrophoresis,
transport, such as diffusion, advection, and reaction, as in isoelectric focusing, and 2D electrophoresis in 3D complex
the classical transport equation. Moreover, transport due to domains.
electric forces must also be considered. This effect is known It is well known that when solving advection-dominated
as migration. transport problems, numerical stabilization techniques are
Modeling and simulation of electrophoretic separation is needed to avoid spurious oscillations in the solution [8,27].
valuable because it can provide a better understanding of the Generally, these undesirable effects are related to the high
fundamentals of the physical and chemical processes stud- values of the nondimensional numbers representing the ratio
ied. Also, simulation would help with optimizing the device of the advection to diffusion, such as the local Péclet and
designs and operation parameters, reducing the risk of spend- Reynolds numbers. Also, in electrophoretic transport prob-
ing time or money on flawed devices or potentially dangerous lems the strong coupling between the fluid and electric
experiments. problems generates a nonzero-divergence advection field
Early mathematical models of electrophoresis were devel- due to the migration. When the migration field acts on a
oped by Saville et al. [43]. These 1D models are valid nonuniform concentration field, it generally results in high
for monovalent analytes in a stagnant electrolyte solution, gradients or even discontinuities in the concentration of dif-
without EOF. More complex models of conventional elec- ferent ions. This requires additional stabilization in the form
trophoresis were later reported in [5,22,62]. These models of discontinuity-capturing (DC) techniques. In addition, a
are useful for capillary electrophoresis where the system “pressure stabilization” is needed in finite element computa-
is inherently 1D due to the uniform section of the capil- tion of incompressible flows with equal-order interpolation
lary and the high aspect ratio. These kind of models can functions for velocity and pressure.
be simulated by using 1D finite differences as is the case One of the most popular stabilized finite element for-
in the cited papers. Finite element formulations for electro- mulations for flow and transport problems is the Stream-
phoretic transport and 2D computations were presented by line-Upwind/Petrov-Galerkin (SUPG) formulation. It was
Ganjoo and Tezduyar [19] and Ganjoo et al. [20]. That work introduced for advection–diffusion problems and incom-
targeted capillary and free-flow electrophoresis assuming a pressible flows in [8] and for compressible flows in [27].
given velocity field. In the case of electrophoretic separation The Pressure-Stabilizing/Petrov-Galerkin (PSPG) was intro-
developed in microfluidic chips, there exist multiple channels duced in [48,53]. An earlier version of the pressure stabil-
with different sections; such systems require more complex ization, for the Stokes flow, was introduced in [25]. The
domains. Such simulations started with the work of Patan- space–time SUPG/PSPG formulations, as different versions
kar and Hu [37] and Ermakov et al. [18], using 2D finite of the Deforming-Spatial-Domain/Stabilized Space–Time
element models. Chatterjee [10] developed a 3D finite vol- (DSD/SST) method, were introduced in [47–49,51,52,58].
ume model to study several applications in microfluidics, Supplementing the SUPG formulation with DC and
including isoelectric focusing. Recently, Barz [3] developed shock-capturing techniques goes almost as far back as the
a fully coupled model for electrokinetic flow and migration development of the SUPG formulation. Use of such addi-
in microfluidic devices employing a 2D finite element model. tional stabilization in the context of advection–diffusion
Other examples of related work are those of Shim et al. [45] problems and compressible flows was reported in [24,26,55].

123
Comput Mech

The additional stabilization techniques introduced in [55] tational acceleration, E is the electric field strength, T =
included the “DRD” stabilization for advection–diffusion– − pI + μ(∇u + ∇uT ) is the stress tensor, and μ is the vis-
reaction problems with dominant reaction terms. The original cosity. The last term on the right hand side of Eq. (1) is the
DRD stabilization and its newer versions were used exten- contribution of the electrical forces to the momentum bal-

sively in flow and transport problems with dominant reac- ance, where ρel = F j z j c j is the electric charge density
tion or reaction-like terms (see [12–16,35,36,55,56]). Other of the electrolyte solution, obtained as the summation over
DC and shock-capturing techniques introduced in conjunc- all j-type ions, with valence z j and concentration c j , and F
tion with the SUPG formulation include the Discontinuity- is the Faraday constant. Equation (2) is the incompressibility
Capturing Directional Dissipation (DCDD) [49] and Y Zβ constraint.
shock-capturing [50,59], which were also used extensively
(see [4,40–42,59–61]). 2.2 Electric field
In the stabilized finite element formulations, an embed-
ded stabilization parameter, which is known as “τ ”, plays Charge distribution in an electrolyte solution with solvent
an important role. This stabilization parameter involves a permittivity  is related to the electric field E as
measure of the local length scale (also known as “element
length”) and other parameters such as the local Reynolds ∇ · E = ρel , (3)
and Courant numbers. Various element lengths and τ s were
where the electric field is obtained from the associated elec-
proposed starting with those in [8] and [27], followed by
tric potential Φ as
the one introduced in [55]. Various new τ definitions were
proposed in recent years, including those with a second, dif- E = −∇Φ. (4)
fusive element length scale [49], those based on the element-
level matrices [54], and those based on the element-level Modeling electrophoresis problems requires special con-
vectors [54], which address the difficulties reported at small sideration for the electric field since it involves different
time step sizes (see [23,54]). contributions in the flow domain, and because it is strongly
The SUPG formulations for the electrophoretic separa- influenced by the presence of nonuniform electrolyte con-
tion process were reported by Ganjoo and Tezduyar [19] and centrations. Here, a wall-fitted coordinate system (η, ξ1 ,ξ2 ) is
Ganjoo et al. [20]. Those studies did not have a DC term in used, where η is the coordinate normal to the solid boundary,
the stabilized formulation, and the flow field was assumed while ξ1 and ξ2 are the two orthogonal coordinates tangent
to be given. Here we present a stabilized formulation for the to the wall. The first contribution to the electric field comes
coupled fluid mechanics and electrochemical transport prob- from the presence of electrostatic charges at the solid–liquid
lem. The stabilization consists of the SUPG formulation, with interfaces. The interfacial charge has an associated electric
suitable τ definitions, and a DC term that is closely related potential ψ that decreases steeply in the η-direction due to
to the DCDD terms proposed in [49,57,59,60]. the screening produced by counterions and other electrolyte
ions in the solution. The value of ψ at the interface [39]
depends on the wall–solution interactions and it is known as
the electrokinetic potential ζ [7]. There is also a potential φ
2 Governing equations
in the flow domain, which arises when a potential difference
φ is externally applied to drive the electrophoresis and/or
The mathematical model for electrophoretic transport
induce EOF. As the channel walls are supposed be perfectly
includes the fluid dynamics, electric field, mass transport,
insulating, there are no components of the applied electric
chemical reactions, and the coupling of all these.
field normal to the wall. So that, near the walls, φ varies only
in the ξ1 –ξ2 plane and the total electric potential Φ can be
2.1 Flow field written as

In the framework of continuum fluid mechanics, fluid veloc- Φ(η, ξ1 , ξ2 ) = ψ(η) + φ(ξ1 , ξ2 ). (5)
ity u and pressure p are governed by the Navier–Stokes equa-
This superposition is valid if the applied electric field is small
tions of incompressible flows [33,39]:
compared to the EDL electric field, as often happens in prac-
 
∂u tice. Introducing the electric potential into Eq. (3) leads to
ρ + u · ∇u = ∇ · T + ρg + ρel E, (1) the following equation for the electric potential referred in
∂t
∇ · u = 0. (2) the wall-fitted coordinate system:

∂ 2ψ  ∂ 2φ
2
Equation (1) represents the conservation of momentum for ρel
+ =− . (6)
a Newtonian fluid, where ρ is the density, g is the gravi- ∂η 2 ∂ξi2 
i=1

123
 Comput Mech

The second term on the left hand side of Eq. (6) is non- Here Ω j is the ionic mobility. The terms in parenthesis in
zero in electrophoresis problems because the concentration Eq. (9) represent the electric current density i, which accounts
∂φ
gradients in the fluid induce variations on ∂ξi
inside the chan- for the ion fluxes due to flow advection, electrical forces, and
nel. However, ∂2φ
are several orders of magnitude smaller diffusion. We note that Eq. (9) reduces to ∇ 2 φ = 0 (i.e., the
∂ξi2 Laplace’s equation for the applied potential), which has his-
∂2ψ
than ∂η2
(see also [17,34,46]), which allows splitting the torically been used for electric potential calculations, only
computation of the electric field into two parts as explained if the electrolyte concentration and mobility are perfectly
below. uniform and stagnant.

2.2.1 Electric double layer 2.3 Mass transport and chemistry

According to the previous analysis, the EDL potential is gov- The mass transport of weakly concentrated sample molecules
erned by and buffer electrolyte components can be modeled by a lin-
∂ 2ψ ρel ear superposition of the migrative, advective, and diffusive
=− . (7) transport mechanisms, plus a source term due to the chemi-
∂η 2 
cal reactions. In a non-stationary mode, the concentration of
Nevertheless, computational requirements are very large each j-type species is governed by [39]
when an entire electrophoretic device is modeled. In that
sense, we simplify the computation of the EOF by intro- ∂c j
+ ∇ · (−z j Ω j ∇φc j + uc j − D j ∇c j ) − r j = 0, (11)
ducing the so-called thin EDL approximation [6,9], where ∂t
the EOF is regarded as an electrically induced slip velocity
where r j is the reaction term. Different electrolytes (acids,
(electroosmotic velocity, ueo ) in the direction of the applied
bases, and ampholytes), analytes, and particularly the hydro-
electric field. This velocity is determined by the Helmholtz–
gen ion have to be considered. In electrolyte chemistry the
Smoluchowski equation:
processes of association and dissociation are much faster than
ζ E the transport processes, hence, adopting chemical equilib-
ueo = − . (8)
μ rium constants to model the reactions of weak electrolytes
is a suitable approximation [2]. The strong electrolytes are
This approximation also implies that ρel ≈ 0 in the fluid
considered as completely dissociated. Expressions for r j in
outside the EDL, meaning that the last term on the right-
modeling the different reactive phenomena and their approx-
hand side of Eq. (1) is negligible, i.e. ρel E ≈ 0. Thus, the
imations require a detailed explanation that exceeds the scope
electroosmotic velocity is included in the fluid problem as
of the present work. Some of the most common expressions
a boundary condition, reducing significantly the computa-
are presented in the Appendix, and a more complete descrip-
tional demand. The simplification is appropriate considering
tion can be found in [30].
that ψ(η) decreases to zero within 1–10 nm from the wall,
while the cross-sectional channel or capillary dimensions are
typically 20–200 µm.
3 Finite element formulation
2.2.2 Bulk fluid
In this section we first present the finite element formula-
Based on the considerations above, the electric potential φ tions for the fluid mechanics, charge conservation, and mass
has to be computed using the charge conservation equation transport. Following that, the coupling of these equations is
in steady state [39]: described.
⎛ ⎞
N
3.1 Navier–Stokes equations of incompressible flows
∇ · ⎝−σ ∇φ − F z j D j ∇c j + ρel u⎠ = 0, (9)
j=1
We use the SUPG/PSPG formulation [48,53] of Eqs. (1) and
where D j is the diffusion coefficient for the species j, N (2). Considering a computational domain Ω with boundary
is the number of different species present in the electrolyte Γ divided into n el finite elements Ω e , e = 1, . . . , n el , E is
solution, and σ is the electrical conductivity: the set of those elements and H 1h is the finite-dimensional
space defined as

N
σ =F z j 2Ω j c j . (10)
j=1
H 1h = θ h |θ h ∈ C 0 (Ω), θ h |Ω e ∈ P 1 , ∀Ω e ∈ E , (12)

123
Comput Mech

with P 1 representing the first-degree polynomials. The func- 3.2 Electric problem
tion spaces for the trial and test functions are defined as
For the finite element formulation of Eq. (9), the function
Suh = { uh |uh ∈ (H 1h )n sd , uh =D
˙ on ΓD }, spaces for the trial and test functions are defined as
Vuh = { wh |wh ∈ (H 1h )n sd , wh =0
˙ on ΓD },
Sφh = { φ h |φ h ∈ H 1h , φ h =ϕ
˙ h on Γϕ },
S hp = { q|q ∈ H 1h }, (13)
Vφh = { v h |v h ∈ H 1h , v h =0
˙ on Γϕ }. (18)
where n sd is the number of space dimensions. It is assumed
that both Dirichlet- and Neumann-type boundary conditions It is assumed that only Dirichlet-type boundary condi-
could be imposed at different segments of the boundary Γ : tions are imposed at one or more segments Γϕ of the
boundary Γ :
uh = D on ΓD ,
φ = ϕ on Γϕ . (19)
n · Th = h on Γh , (14)
The Galerkin formulation of the problem can be written
where ΓD and Γh are the complementary subsets of Γ , and as follows: find φ h ∈ Sφh , such that ∀v h ∈ Vφh :
uh includes the electroosmotic velocity ueo , given by Eq. (8).
The SUPG/PSPG formulation is written as follows: find
uh ∈ Suh and p h ∈ S hp , such that ∀wh ∈ Vuh , ∀q h ∈ S hp : ∇v h · σ h ∇φ h dΩ
Ω
 ⎛ ⎞
∂uh 
N
wh · ρ + uh · ∇uh dΩ = ∇v h · ⎝−F z j D j ∇chj + ρelh uh ⎠ dΩ.
∂t (20)
Ω j=1
 Ω
+ ∇wh : Th dΩ − wh · ρg + ρelh E dΩ
Ω Ω
3.3 Mass transport problem
n el
1 
+ τSUPG uh · ∇wh + τPSPG ∇q h For the finite element formulation of Eq. (11), the function
ρ spaces for the trial and test functions are defined as
e=1Ω e
  
∂uh S hj = { chj |chj ∈ H 1h , chj =C
˙ j on ΓC j },
· ρ +u · ∇u −∇ · T − ρg−ρel E dΩ
h h h h
∂t
V jh = { v h |v h ∈ H 1h , v h =0
˙ on ΓC j }. (21)
+ q ∇ · u dΩ =
h h
w · h dΓ
h
(15)
It is assumed that both Dirichlet- and Neumann-type bound-
Ω Γh
ary conditions could be imposed at different segments of the
The stabilization parameters in Eq. (15) are defined as boundary Γ :

 − 1 chj = C j on ΓC j ,
2
1 1 1
τSUPG = τPSPG = + 2+ 2 , n · ∇chj = F j on Γ F j , (22)
τ12 τ2 τ3
h SUPG where ΓC j and Γ F j are the complementary subsets of Γ .
τ1 = ,
2||uh || The SUPG+DC formulation [49,57,59,60] for this prob-
tNS lem is given as follows: find chj ∈ S hj , such that ∀v h ∈ V jh :
τ2 = ,
2  
h2 ∂chj  
τ3 = ρ SUPG . (16) v h
+∇ · − z j Ω hj ∇φ h chj + uh chj dΩ
12μ ∂t
Ω
Here, tNS is the time step size and the element length h SUPG + ∇v h · D j ∇chj dΩ − v h r j dΩ
is defined [55] as
Ω Ω
n −1
 en nel
 
h SUPG = 2||uh || |uh · ∇ Na | , (17) + τSUPG j − z j Ω hj ∇φ h + uh · ∇v h
a=1 e=1Ω e
 h 
where n en the number of element nodes and Na is the inter- ∂c j  
× +∇ · − z j Ω j ∇φ h chj +uh chj + D j ∇chj −r j dΩ
polation function associated with element node a. ∂t

123
 Comput Mech


n el
in this work that the EDL length is considerably smaller than
+ νDC j ∇v h ∇chj dΩ = v h F jh dΓ. (23)
the channel width.1 Consequently, we can assume ρel ≈ 0
e=1Ω e ΓF j in the bulk fluid and the electric force term in Eq. (15) van-
We note that the SUPG stabilization involves the residuals ishes. The electric forces in the EDL are included in the fluid
of the governing equations, which include the second-order problem as a Dirichlet boundary condition modeled by the
term ∇ · (D j ∇chj ). In this case, for linear (triangular and tet- Helmholtz–Smoluchowski approximation for the slip veloc-
rahedral) elements, the term vanishes. For additional details, ity given by Eq. (8). To calculate this velocity, the updated
the reader is referred to [54]. In the case of the second-order electric potential (φ h ) is used.
term involving ∇ · (∇φ h ), the term is evaluated by treat- After that, the concentrations of all species (chj , for j =
ing (−z j Ω hj ∇φ h ) like a (non-incompressible) velocity field, 1 . . . N ) are updated simultaneously by solving Eq. (23),
using the updated potential (φ h ) and velocity (uh ). The equa-
where φ h comes from the solution of Eq. (20).
tions are advanced n M T steps using a smaller time step tMT :
The stabilization parameter τSUPGj for each j-species is
defined as tNS = n M T tMT , (28)
 − 1
2
1 1 1 1
τSUPG j = + + + , where, at each step with tMT , linearly-interpolated values of
τ12j τ22 τ32j τ42j φ h and uh are used. The updated concentration field enables
h SUPG j the calculation of the new conductivity field using Eq. (10)
τ1 j = ,
2||−z j Ω hj ∇φ h + uh || and the start of a new cycle by updating the electric potential.
The process described above is carried out once for each step
tMT
τ2 = , with tNS .
2
The flow field is treated as quasi-steady because the char-
h 2SUPG j acteristic time scales for advection and diffusion are consid-
τ3 j = ,
12D j erably larger than the characteristic time scale for reaction.
1 For the same reason, the electric field is steady. Consequently,
τ4 j =  . (24)
∇ · − z j Ω hj ∇φ h there exist two time steps that govern the solution of the com-
plete problem. In updating the flow field, the larger time step
Here, tMT is the time step size and the element length ( tNS ) is determined to achieve a maximum local Courant
h SUPG j for species j is defined [55] as number (CFL) close to 1.0:
h SUPG j = 2||−z j Ω hj ∇φ h + uh ||  
n −1 ||−z j Ω hj ∇φ h + uh || tNS
 en
  CFL = max . (29)
e∈E , 1≤ j≤N h SUPG j
× | − z j Ω j ∇φ + u · ∇ Na |
h h h
. (25)
a=1
The smaller time step ( tMT ) used in updating the concen-
The DC parameter νDCj for species j is defined as tration fields is determined by the reaction coefficient that
 2 models the fastest reaction.
h DC j ||∇c h ||h
j DC j
νDC j = ||−z j Ω hj ∇φ h + uh || , (26)
2 cref j
where cref j is a reference value for the concentration of the 4 Numerical examples
j-species, and
 nen −1 The first numerical experiment is a 1D simulation of a

h DC j = 2||∇chj || |∇chj · ∇ Na | . (27) capillary electrophoresis assay, where we test different sta-
a=1 bilization parameters. Following that, a 2D application exam-
ple with several species and different advection fields acting
3.4 Coupling on the transported sample along different directions is pre-
sented. Finally a 3D test where the main flow is split and
Starting from an initial or previous state of all the fields (elec- recombined to obtain an efficient mixing of two buffer com-
tric potential, velocity, pressure, and the concentration of ponents is studied.
each species), the first unknown to be updated is the elec-
tric potential (φ h ), using Eq. (20) with the Dirichlet-type 1 For cases where the EDL length is comparable to the channel width,
boundary condition given in Eq. (19). Next, the new velocity
we compute the charge density by using the Poisson–Boltzmann equa-
(uh ) and pressure ( p h ) are computed by advancing Eq. (15) a tion and the fluid mechanics problem does include the electric force
single time step tNS . As mentioned in Sect. 2.2, we assume term; see [30,32] for more details.

123
Comput Mech

Table 1 Physicochemical properties of the buffer components and analytes

Component pK a a pK b Mobility (m2 /Vs) Diffusivity (m2 /s)

Hydrochloric acid −2.0 − 7.91 × 10−8 2.03 × 10−9

Tetraphenylborate 5.0 − 1.80 × 10−8 4.60 × 10−10
TRIS − 8.08 2.95 × 10−8 7.60 × 10−10
Acetic acid 4.76 − 4.20 × 10−8 1.08 × 10−9
a pK represents the commonly used notation for equilibrium constants: pK = − log10 K

4.1 1D test: capillary electrophoresis this potential difference generates an electric field that exerts
electric forces on the samples, which enables the separation
In this first test we demonstrate the effects of using the SUPG due to the differences in effective mobilities. Due to the local
and DC stabilizations in advection-dominated problems, spe- values of pH, hydrochloric acid mobility is higher than the
cifically in separation by capillary electrophoresis (CE). In a tetraphenilborate one; and consequently the displacement of
CE assay, an electric potential difference is applied between hydrochloric acid is higher.
the ends of a microchannel to obtain a longitudinal electric Figure 2 shows the sample-concentration distributions at
field. This electric field exerts forces on the charged mol- two instants and the effect of the stabilization terms on these
ecules in the solution, resulting in different electrophoretic solutions. It is clear that the stabilization reduces the spurious
velocities and enabling the separation of the sample compo- oscillations in the numerical solutions.
nents. In this example, two acidic species (hydrochloric acid
and tetraphenilborate) are separated using a low-concentra-
tion buffer of tris(hydroxymethyl)-aminoetane (TRIS) and 4.2 2D test: free-flow electrophoresis
acetic acid. Physicochemical properties of the buffer and ana-
lytes are listed in Table 1. Initial conditions for the samples In this test we demonstrate the effectiveness of the proposed
and buffer components are shown in Fig. 1a. Due to the rel- method in simulating electrophoretic transport with conju-
atively low concentration of the buffer components, samples gation of the flow advection field and migration field in var-
and their motion affect the local physicochemical properties iable directions. In this case the electrophoretic separation
of the electrolyte solution. This means that the local pres- process is simulated by using the free-flow electrophoresis
ence of the sample in a region of the capillary affects both (FFE) method.
the electric conductivity and the local acid–base equilibrium The FFE is a continuous technique for electrophoretic sep-
(i.e., the pH of the solution), as seen in Fig. 1a. The former aration. These methods provide bands across the separation
affects the local values of the electric field, and the latter alters chamber and thus a continuous supply of separated compo-
the effective mobility of the samples (see Eqs. (39) and (40)), nents at the exit of the chamber. In FFE, sample components
producing a strong dual coupling between the numerical solu- are injected into a liquid carrier with an electric field applied
tions of the electric field and the transport equations (Eqs. (9) perpendicular to the flow direction. Particles are deflected
and (11)). This can be inferred from Fig. 1b. To carry out from the flow streamlines at an angle arranged by the vector
the test, a 3000 V difference of electric potential is applied composition of the flow advection and the migration (deter-
between the end walls of a 5 cm long capillary, which is mined by the local sample mobility and the electric field
filled with the electrolytic solution. As mentioned earlier, strength). Sample components with different electrophoretic

Fig. 1 Initial conditions for the

samples, buffer components,
conductivity, and electric field. a
Samples and buffer components.
b Concentration, conductivity,
and electric field coupling

123
 Comput Mech

Fig. 2 Sample concentrations for the CE assay at t = 3 s and t = 6 s for different stabilization schemes. a Without stabilization, t = 3 s. b Without
stabilization, t = 6 s. c With SUPG, t = 3 s. d With SUPG, t = 6 s. e With SUPG+DC, t = 3 s. f With SUPG+DC, t = 6 s

mobilities have different deflections and can be collected sep- are shown in Fig. 3. The pressure field, velocity field, and the
arately at the end of the separation area. streamlines are shown in Fig. 4. The superposition of the
In this example, two aminoacids from a single flow are migration and advection effects with variable directions can
separated in two streams and redirected to two separate exit be inferred from the field lines in Fig. 3 and the streamlines
channels. Physicochemical properties of the analytes are in Fig. 4. The resulting motion of the substances will be the
listed in Table 2. superposition of these effects (plus the dispersion due to the
To develop the FFE, a 750 V electric potential difference diffusion), producing the separation of the aminoacids. Due
is applied at the outlets and the flow field is developed by to the differences in the degree of dissociation, the mobility
imposing a velocity of 1.0 cm/s at the inlet and zero pressure is negative for the aspartic acid and positive for the lysine,
at the outlets. The dimensions are 3.0 × 1.0 mm2 for the enabling the electrophoretic separation. As a result, the aspar-
chamber and 1.0 × 0.1 mm2 for the channels. Geometry of tic acid tends to move downwards to the positive potential
the FFE device, electric potential, electric field, and field lines region and enters the bottom channel, while lysine moves

123
Comput Mech

Table 2 Physicochemical properties of the analytes

Component pK a pK b Mobility (m2 /Vs) Diffusivity (m2 /s) Initial

concentration
(mM)

Aspartic acid 2.28 3.90 2.83 × 10−8 7.31 × 10−9 1.0

Lysine 9.12 10.79 2.70 × 10−8 6.98 × 10−10 1.0

the capability of the simulation tool developed for solving

3D flow and transport problems. We compare the mixing
efficiencies of the pressure driven and EOF driven mixers.
At the microscale, the fluid flow is inherently laminar, and
because of that mixing and homogenization of the electrolyte
solution results in a challenging problem from a technologi-
cal point of view [21]. In this example the mixing technique
consists of splitting and recombining the flow. First the flow
is split into two parts: the upper half and the lower half. Then,
these two parts of the flow are recombined horizontally: the
upper half fills the left half of exit channel and the lower
Fig. 3 Electric potential, electric field, and field lines. Electric poten- half fills the right half. This mixing technique is employed
tial values are represented by the background colors, and electric field to obtain a uniform buffer solution commonly used in elec-
magnitude by line colors. (color figure online)
trophoresis assays. The buffer solution is composed of equal
parts of acetic acid and TRIS. Physicochemical properties
of the buffer components employed in this example can be
found in Table 1.
The dimensions of the mixer are 100×40 µm2 for the rect-
angular section at the inlet and outlet, and 4 mm for the length
of the device. The geometry and the velocity and pressure
fields are shown in Figs. 7 and 8 for the pressure driven and
EOF driven mixers, respectively. The pressure driven flow
field is obtained by imposing a uniform velocity of 1.0 mm/s
at the inlet and zero pressure at the outlet. The EOF driven
flow field is obtained by applying an electric potential dif-
ference of 50.0 V (based on an electrokinetic potential ζ =
Fig. 4 Pressure, velocity, and streamlines. Pressure values are repre- 0.8 mV) and zero pressure at the outlet. Concentration dis-
sented by the background colors, and velocity magnitude by line colors. tributions for the buffer components and the resulting pH of
(color figure online)
the solution at different instants are shown in Figs. 9 and 10
for the pressure driven and EOF driven mixing, respectively.
upward, entering the upper channel. Aspartic acid and lysine We also evaluate the sample distributions and the pH at the
concentration distributions at different instants are shown in outlet cross section to determine which flow type provides
Fig. 5. a more efficient mixing. Figure 11 shows the concentration
Figure 6 shows the concentration distributions for aspartic and pH profiles at the outlet cross section. From the latter it
acid and lysine at the same instants as in Fig. 5, but without can be inferred that the plug-like velocity profile of the EOF
the numerical stabilization. It can be seen that the stabiliza- driven flow is less efficient than the parabolic profile of the
tion is essential when the streamlines change direction more pressure driven flow. Also, it is worth mentioning that the
rapidly. efficiency of the EOF mixing decreases due to the influence
of the migration terms, speeding up the TRIS and slowing
4.3 3D application example: split-and-recombine down the acetic acid relative to the fluid flow.
micro-mixer The equilibrium pH for the buffer solution is theoretically
determined as 6.42, which is close to the center of the pH
In this test a split-and-recombine micro-mixer inspired by range obtained at the outlet by the pressure driven mixing.
the earlier work of Schonfeld et al. [44] is simulated to show Also, the pressure driven mixing provides a narrower range

123
 Comput Mech

Fig. 5 Aspartic acid (left) and lysine (right) distributions at different instants. a t = 0.25 s. b t = 0.25 s. c t = 1.75 s. d t = 1.75 s. e t = 3.0 s.
f t = 3.0 s

Fig. 6 Aspartic acid (left) and lysine (right) distributions at different instants, without numerical stabilization. a t = 0.25 s. b t = 0.25 s.
c t = 1.75 s. d t = 1.75 s. e t = 3.0 s. f t = 3.0 s

Fig. 7 Pressure, velocity, and

streamlines for the pressure
driven mixing. The pictures
have been scaled by 0.2 in the
z-direction to enhance the
visualization. a Pressure field.
b Velocity field

123
Comput Mech

Fig. 8 Pressure, velocity, and

streamlines for the EOF driven
mixing. The pictures have been
scaled by 0.2 in the z-direction
to enhance the visualization.
a Pressure field. b Velocity field

Fig. 9 Acetic acid, TRIS, and pH distributions at different instants for the pressure driven mixing. The pictures have been scaled by 0.2 in the
z-direction to enhance the visualization. a t = 1.0 s. b t = 1.0 s. c t = 1.0 s. d t = 3.0 s. e t = 3.0 s. f t = 3.0 s. g t = 7.0 s. h t = 7.0 s. i t = 7.0 s

Fig. 10 Acetic acid, TRIS, and pH distributions at different instants for the EOF driven mixing. The pictures have been scaled by 0.2 in the
z-direction to enhance the visualization. a t = 1.0 s. b t = 1.0 s. c t = 1.0 s. d t = 3.0 s. e t = 3.0 s. f t = 3.0 s. g t = 7.0 s. h t = 7.0 s. i t = 7.0 s

123
 Comput Mech

Fig. 11 Acetic acid, TRIS, and pH distributions for a cross section at the outlet of the mixer. a Pressure driven. b EOF driven. c Pressure driven.
d EOF driven. e Pressure driven. f EOF driven

of pH (less than 0.7) compared to the EOF driven mixing velocity and the electric field have variable directions. The
(≈1.3). SUPG+DC formulation can effectively handle the combina-
tion of fluid and electric advection effects. The last example
helps us evaluate the mixing performance of a 3D microm-
5 Concluding remarks ixer under different flow conditions. The complexity of this
example demonstrate that the proposed formulation is suit-
We presented a stabilized finite element formulation for com- able for solving electrophoretic problems for a wide range of
puting coupled fluid mechanics and electrophoretic trans- applications.
port equations. The formulation is based on the SUPG and With the advent of more complex microfluidic devices,
DC methods developed for scalar and vector equations. We classical 1D models for electrophoresis is no longer suit-
proposed expressions for the stabilization and DC parame- able to represent both physicochemical phenomena and
ters τSUPG and νDC , accounting for the diffusion, advection, design characteristics. Consequently, more complex simu-
and reaction effects associated with the electrophoretic pro- lation frameworks are needed for design and optimization.
cesses. This work contributes to the enhancement of finite ele-
We presented some test computations and application ment simulations in the emerging technology of microfluidic
examples to asses the effectiveness of the formulation. chips.
The first example was a 1D capillary electrophoresis prob-
lem. The SUPG+DC formulation reduces substantially the Acknowledgements This work has received financial support from
spurious oscillations that are typically observed in solu- Consejo Nacional de Investigaciones Científicas y Técnicas (CONI-
CET, Argentina, grant PIP 5271/05), Universidad Nacional del Litor-
tion of advection-dominated problems without stabilization. al (UNL, Argentina, grant CAI+D 2009 65/334 and 65/238), Agencia
The second example shows a typical application of free- Nacional de Promoción Científica y Tecnológica (ANPCyT, Argentina,
flow electrophoretic separation. In these problems, the fluid grants PICT 01141/2007, PICT 0270/2008, PICT-1506/2006).

123
Comput Mech

Appendix computed as (α0 −α2 )Ω j , where α0 and α2 are the degrees of

dissociation of the anionic and cationic forms, respectively.
Acid–base reactions They can be written in terms of cH+ as
Ka Kb
For the general case, reactions associated with an ampholyte 2
cH +
AH with concentration cAH are α0 = Kb Ka Kb
, (39)
1+ cH+ + 2
cH
ka1 +
−−→ A− + H+ ,
AH ← 1
−− (30)
α2 = . (40)
ka2 Kb Ka Kb
kb1
1+ cH+ + 2
cH
AH+ −−→ +
2 ←−− AH + H ,
+
(31)
kb2 Therefore the governing equation for concentration of the
where ka1 and kb1 are the dissociation rates and ka2 and kb2 j-species in the sample plug is
are the association rates for the acidic and basic reactions, ∂c j
respectively. Then the equilibrium state is characterized by + ∇ · [−(α0 − α2 )Ω j ∇φc j + uc j − D j ∇c j ] = 0,
∂t
ka2 c −c + (41)
= A H = Ka , (32)
ka1 cAH
where the transport of analytes is coupled to their degree of
kb2 cAH cH+
= = Kb, (33) dissociation at a given pH.
kb1 cAH+
2

where K a and K b are the equilibrium constants for the acidic

and basic reactions, respectively. The corresponding expres- References
sions for r j are obtained as follows:
1. Albrecht JW, El-Ali J, Jensen KF (2007) Cascaded free-flow iso-
rA− = −ka1 cA− cH+ + ka2 cAH , (34) electric focusing for improved focusing speed and resolution. Anal
Chem 79:9364–9371
rAH = ka1 cA− cH+ − ka2 cAH − kb1 cAH cH+ + kb2 cAH+ ,
2 2. Arnaud I, Josserand J, Rossier J, Girault H (2002) Finite element
(35) simulation of off-gel buffering. Electrophoresis 23:3253–3261
3. Barz DP (2009) Comprehensive model of electrokinetic flow and
rAH+ = kb1 cAH cH+ − kb2 cAH+ , (36) migration in microchannels with conductivity gradients. Micro-
2 2
rH+ = −ka1 cA− cH+ + ka2 cAH − kb1 cAH cH+ + kb2 cAH+ . fluid Nanofluidics 7(2):249–265
2 4. Bazilevs Y, Calo VM, Tezduyar TE, Hughes TJR (2007) YZβ
(37) discontinuity-capturing for advection-dominated processes with
application to arterial drug delivery. Int J Numer Methods Fluids
In Eq. (37) the water dissociation term is not included since 54:593–608. doi:10.1002/fld.1484
this reaction is several orders of magnitude faster than the 5. Bercovici M, Lele SK, Santiago JG (2009) Open source simula-
reactions given by Eqs. (30) and (31) (see [2]). Then, c O H − tion tool for electrophoretic stacking, focusing, and separation.
J Chromatogr A 1216:1008–1018
can be computed directly as
6. Berli CLA (2008) Equivalent circuit modeling of electrokinet-
Kw ically driven analytical microsystems. Microfluid Nanofluidics
cOH− = , (38) 4(5):391–399
cH+ 7. Berli CLA, Piaggio M, Deiber J (2003) Modeling the zeta potential
where K w = 10−8 mol2 m−6 is the dissociation constant for of silica capillaries in relation to the background electrolyte com-
position. Electrophoresis 24:1587–1595
pure water at 25 ◦ C. 8. Brooks AN, Hughes TJR (1982) Streamline upwind/Petrov-Galer-
kin formulations for convection dominated flows with particular
Effective charge and mobility of analytes emphasis on the incompressible Navier-Stokes equations. Comput
Methods Appl Mech Eng 32:199–259
9. Brunet E, Adjari A (2004) Generalized Onsager relations for elec-
When the concentration of the analytes is much lower than trokinetic effects in anisotropic and heterogeneous geometries.
the buffer components, its effect on the pH is negligible. Phys Rev E 69(1):016306
In such cases, considering all the ionic species involves an 10. Chatterjee A (2003) Generalized numerical formulations for multi-
physics microfluidics-type applications. J Micromech Microeng
unnecessary high computational cost. However the influence 13:758–767
of pH on the analytes must be taken into account. Thus the 11. Chau M, Spiteri P, Guivarich R, Boisson H (2008) Parallel asyn-
transport equation of these analytes includes r j = 0, and chronous iterations for the solution of a 3d continuous flow elec-
the product z j Ω j as a function of pH. For example, if the trophoresis problem. Comput Fluids 37(9):1126–1137
12. Corsini A, Iossa C, Rispoli F, Tezduyar TE (2010) A DRD finite
species is an ampholyte that obeys a reaction scheme like the element formulation for computing turbulent reacting flows in
one shown in Eqs. (30) and (31), z j Ω j is included in Eq. (11) gas turbine combustors. Comput Mech 46:159–167. doi:10.1007/
as an effective charge–mobility product [10]. This product is s00466-009-0441-0

123
 Comput Mech

13. Corsini A, Menichini C, Rispoli F, Santoriello A, Tezduyar port in microfluidic chips. Comput Methods Appl Mech Eng
TE (2009) A multiscale finite element formulation with discon- 198(30–32):2360–2367
tinuity capturing for turbulence models with dominant reactionlike 33. Li D (2004) Electrokinetics in microfluidics. Elsevier Academic
terms. J Appl Mech 76:021211. doi:10.1115/1.3062967 Press, London
14. Corsini A, Rispoli F, Santoriello A, Tezduyar TE (2006) Improved 34. MacInnes JM (2002) Computation of reacting electrokinetic flow
discontinuity-capturing finite element techniques for reaction in microchannel geometries. Chem Eng Sci 57(21):4539–4558
effects in turbulence computation. Comput Mech 38:356–364. 35. Park YJ, Deans HA, Tezduyar TE (1990) Finite element formula-
doi:10.1007/s00466-006-0045-x tion for transport equations in a mixed coordinate system: an appli-
15. Corsini A, Rispoli F, Tezduyar TE (2011) Stabilized finite element cation to determine temperature effects on the single-well chemical
computation of NOx emission in aero-engine combustors. Int J tracer test. Int J Numer Methods Fluids 11:769–790. doi:10.1002/
Numer Methods Fluids 65:254–270. doi:10.1002/fld.2451 fld.1650110605
16. Corsini A, Rispoli F, Tezduyar TE (2012) Computer modeling 36. Park YJ, Deans HA, Tezduyar TE (1991) Thermal effects on sin-
of wave-energy air turbines with the SUPG/PSPG formulation gle-well chemical tracer tests for measuring residual oil satura-
and discontinuity-capturing technique. J Appl Mech 79:010910. tion. Soc Petroleum Eng Form Eval 190:401–408. doi:10.2118/
doi:10.1115/1.4005060 19683-PA
17. Craven TJ, Rees JM, Zimmerman WB (2008) On slip velocity 37. Patankar N, Hu H (1998) Numerical simulation of electroosmotic
boundary conditions for electroosmotic flow near sharp corners. flow. Anal Chem 70(9):1870–1881
Phys Fluids 20(4):043603 38. Peng Y, Pallandre A, Tran NT, Taverna M (2008) Recent innova-
18. Ermakov S, Jacobson S, Ramsey J (1998) Computer simulations tions in protein separation on microchips by electrophoretic meth-
of electrokinetic transport in microfabricated channel structures. ods. Electrophoresis 29(1):157–178
Anal Chem 70(21):4494–4504 39. Probstein R (2003) Physicochemical hydrodynamics. An Intro-
19. Ganjoo DK, Tezduyar TE (1987) Petrov-Galerkin formulations duction, 2nd edn. Wiley-Interscience, New York
for electrochemical processes. Comput Methods Appl Mech Eng 40. Rispoli F, Corsini A, Tezduyar TE (2007) Finite element computa-
65:61–83. doi:10.1016/0045-7825(87)90183-6 tion of turbulent flows with the discontinuity-capturing directional
20. Ganjoo DK, Tezduyar TE, Goodrich WD (1989) A new formu- dissipation (DCDD). Comput Fluids 36:121–126. doi:10.1016/j.
lation for numerical solution of electrophoresis separation pro- compfluid.2005.07.004
cesses. Comput Methods Appl Mech Eng 75:515–530. doi:10. 41. Rispoli F, Saavedra R, Corsini A, Tezduyar TE (2007) Computa-
1016/0045-7825(89)90045-5 tion of inviscid compressible flows with the V-SGS stabilization
21. Hessel V, Lowe H, Schönfeld F (2005) Micromixers—a review on and YZβ shock-capturing. Int J Numer Methods Fluids 54:695–
passive and active mixing principles. Chem Eng Sci 60(8–9):2479– 706. doi:10.1002/fld.1447
2501 42. Rispoli F, Saavedra R, Menichini F, Tezduyar TE (2009) Compu-
22. Hruška V, Jaros M, Gaš B (2006) Simul 5—free dynamic simula- tation of inviscid supersonic flows around cylinders and spheres
tor of electrophoresis. Electrophoresis 27:984–991 with the V-SGS stabilization and YZβ shock-capturing. J Appl
23. Hsu MC, Bazilevs Y, Calo VM, Tezduyar TE, Hughes Mech 76:021209. doi:10.1115/1.3057496
TJR (2010) Improving stability of stabilized and multiscale formu- 43. Saville D, Palusinski O (1986) Theory of electrophoretic sepa-
lations in flow simulations at small time steps. Comput Methods rations. Part I: formulation of a mathematical model. AIChE J
Appl Mech Eng 199:828–840. doi:10.1016/j.cma.2009.06.019 32(2):207–214
24. Hughes T, Mallet M (1986) A new finite element formulation for 44. Schönfeld F, Hessel V, Hofmann C (2004) An optimised split-and-
computational fluid dynamics: IV a discontinuity-capturing oper- recombine micro-mixer with uniform ‘chaotic’ mixing. Lab Chip
ator for multidimensional advective-diffusive systems. Comput 4(1):65–69
Methods Appl Mech Eng 58(3):329–336 45. Shim J, Dutta P, Ivory C (2007) Modeling and simulation of IEF
25. Hughes TJR, Franca LP, Balestra M (1986) A new finite element in 2-D microgeometries. Electrophoresis 28:572–586
formulation for computational fluid dynamics: V. Circumventing 46. Sounart TL, Baygents JC (2007) Lubrication theory for electro-
the Babuška–Brezzi condition: a stable Petrov–Galerkin formula- osmotic flow in a non-uniform electrolyte. J Fluid Mech 576:139–
tion of the Stokes problem accommodating equal-order interpola- 172
tions. Comput Methods Appl Mech Eng 59:85–99 47. Takizawa K, Tezduyar TE (2011) Multiscale space–time fluid–
26. Hughes TJR, Mallet M, Mizukami A (1986) A new finite element structure interaction techniques. Comput Mech 48:247–267.
formulation for computational fluid dynamics: II. Beyond SUPG. doi:10.1007/s00466-011-0571-z
Comput Methods Appl Mech Eng 54:341–355 48. Tezduyar TE (1992) Stabilized finite element formulations for
27. Hughes TJR, Tezduyar TE (1984) Finite element methods for incompressible flow computations. Adv Appl Mech 28:1–44.
first-order hyperbolic systems with particular emphasis on the doi:10.1016/S0065-2156(08)70153-4
compressible Euler equations. Comput Methods Appl Mech Eng 49. Tezduyar TE (2003) Computation of moving boundaries and inter-
45:217–284. doi:10.1016/0045-7825(84)90157-9 faces and stabilization parameters. Int J Numer Methods Fluids
28. Hunter R (2001) Foundations of colloid science, 2nd edn. Oxford 43:555–575. doi:10.1002/fld.505
University Press, Oxford 50. Tezduyar TE (2004) Finite element methods for fluid dynamics
29. Karniadakis G, Beşkök A, Aluru N (2005) Microflows and nano- with moving boundaries and interfaces. In: Stein E, Borst RD,
flows: fundamentals and simulation. Springer, New York Hughes TJR (eds) Encyclopedia of computational mechanics, vol
30. Kler P (2010) Modeling and simulation of microfluidic chips for 3: Fluids, chap. 17. Wiley, New York
analytical applications. Ph.D thesis, UNIVERSIDAD NACIONAL 51. Tezduyar TE, Behr M, Liou J (1992) A new strategy for finite ele-
DEL LITORAL ment computations involving moving boundaries and interfaces—
31. Kler P, Berli C, Guarnieri F (2011) Modeling and high perfor- the deforming-spatial-domain/space–time procedure: I. The con-
mance simulation of electrophoretic techniques in microfluidic cept and the preliminary numerical tests. Comput Methods Appl
chips. Microfluid Nanofluid 10(1):187–198 Mech Eng 94(3):339–351. doi:10.1016/0045-7825(92)90059-S
32. Kler PA, López EJ, Dalcín LD, Guarnieri FA, Storti MA (2009) 52. Tezduyar TE, Behr M, Mittal S, Liou J (1992) A new strat-
High performance simulations of electrokinetic flow and trans- egy for finite element computations involving moving boundaries

123
Comput Mech

and interfaces – the deforming-spatial-domain/space–time proce- 58. Tezduyar TE, Sathe S (2007) Modeling of fluid–structure interac-
dure: II. Computation of free-surface flows, two-liquid flows, and tions with the space–time finite elements: solution techniques. Int
flows with drifting cylinders. Comput Methods Appl Mech Eng J Numer Methods Fluids 54:855–900. doi:10.1002/fld.1430
94(3):353–371. doi:10.1016/0045-7825(92)90060-W 59. Tezduyar TE, Senga M (2006) Stabilization and shock-capturing
53. Tezduyar TE, Mittal S, Ray SE, Shih R (1992) Incompressible flow parameters in SUPG formulation of compressible flows. Com-
computations with stabilized bilinear and linear equal-order-inter- put Methods Appl Mech Eng 195:1621–1632. doi:10.1016/j.cma.
polation velocity-pressure elements. Comput Methods Appl Mech 2005.05.032
Eng 95:221–242. doi:10.1016/0045-7825(92)90141-6 60. Tezduyar TE, Senga M (2007) SUPG finite element computation
54. Tezduyar TE, Osawa Y (2000) Finite element stabilization param- of inviscid supersonic flows with YZβ shock-capturing. Comput
eters computed from element matrices and vectors. Comput Fluids 36:147–159. doi:10.1016/j.compfluid.2005.07.009
Methods Appl Mech Eng 190:411–430. doi:10.1016/ 61. Tezduyar TE, Senga M, Vicker D (2006) Computation of inviscid
S0045-7825(00)00211-5 supersonic flows around cylinders and spheres with the SUPG for-
55. Tezduyar TE, Park YJ (1986) Discontinuity capturing finite ele- mulation and YZβ shock-capturing. Comput Mech 38:469–481.
ment formulations for nonlinear convection-diffusion-reaction doi:10.1007/s00466-005-0025-6
equations. Comput Methods Appl Mech Eng 59:307–325. doi:10. 62. Thormann W, Caslavska J, Mosher R (2007) Modeling of electro-
1016/0045-7825(86)90003-4 osmotic and electrophoretic mobilization in capillary and micro-
56. Tezduyar TE, Park YJ, Deans HA (1987) Finite element pro- chip isoelectric focusing. J Chromatogr A 1155(2):154–163
cedures for time-dependent convection-diffusion-reaction sys- 63. Tsai WB, Hsieh CJ, Chieng CC (2005) Parallel computation of
tems. Int J Numer Methods Fluids 7:1013–1033. doi:10.1002/fld. electroosmotic flow in L-shaped microchannels. In: 6th world con-
1650071003 gress of structural and multidisciplinary optimization, pp 4971–
57. Tezduyar TE, Ramakrishnan S, Sathe S (2008) Stabilized formula- 4980
tions for incompressible flows with thermal coupling. Int J Numer 64. Wu D, Qin J, Lin B (2008) Electrophoretic separations on micro-
Methods Fluids 57:1189–1209. doi:10.1002/fld.1743 fluidic chips. J Chromatogr A 1184(1–2):542–559

123