MMMMMMM

Downloaded 12/30/12 to 129.173.72.87. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.
php
SIAM J. SCI. COMPUT.

Vol. 29, No. 6, pp. 23522380
c 2007 Society for Industrial and Applied Mathematics

NUMERICAL SIMULATION OF ACOUSTIC STREAMING ON

SURFACE ACOUSTIC WAVE-DRIVEN BIOCHIPS
DANIEL KOSTER
Abstract. A novel type of microuidic biochip employs acoustic surface waves as a pumping
mechanism for uids. Piezoelectric substrate materials in the chip eciently convert an electromagnetic input signal into an elastic wave conned to the surface layer of the substrate, hence the surface
acoustic wave. The interaction of these waves with an adjoining uid volume leads to streaming patterns in the uid or of the motion of the uid as a whole. In this work we study this acoustic streaming
eect. We present a two-scale mathematical model, some theoretical results on the well-posedness of
the model problem, as well as numerical simulations based on the nite element method. The model
will include the case of free capillary uid boundaries.
Key words. scientic computing, modeling, numerical simulation, acoustic streaming, nite
elements, numerical analysis, biochips, surface acoustic waves, lab-on-a-chip
AMS subject classications. 74F10, 65M60, 65Z05
DOI. 10.1137/060676623
1. Introduction. The past decades have seen unprecedented advances in the

production of miniaturized electronic circuits, from the leviathan computers lling
entire rooms in the humble beginnings to the powerful and omnipresent microchips
of today. A similar revolution under the motto smaller is better is currently taking
place in the life sciences regarding the manipulation of uids, with the transition from
test tubes and beakers to micropipettes and microtitre plates.
In contrast to the eld of microelectronics, where the aim is still to reduce the
size of transistors, in microuidics the main concern nowadays is to manufacture
increasingly complex systems of channels with sophisticated pumps, valves, mixers,
separators, and lters, all on a scale of micrometers. A popular concept today is
known as lab-on-a-chip. The goal of a lab-on-a-chip is the integration of a complete
analysis and reaction system on a small surface ( 1 cm2 or less). A sample device is
shown in Figure 1.1.
There are many reasons for desiring a small scale. Less energy and lesser amounts
of expensive reagents are needed, reducing costs. Devices can be used in parallel and
with automatic control, useful in applications such as gene sequencing. From the
physical point of view we have more ecient heat and material transport, leading to
better yields and faster analysis.
Despite all of these promising aspects there are certain technical diculties arising in the production of microuidic components. Specically, the task of performing
controlled pumping or mixing of nano- or picoliter quantities of uids becomes dicult. For instance, the low Reynolds number of typical ows on this scale signies that
ows in this regime are laminar and not turbulent. This is an obstacle in designing a
component to mix reagents.
Received by the editors December 4, 2006; accepted for publication (in revised form) May 14,
2007; published electronically October 5, 2007. This work was partially supported by the German
National Science Foundation (DFG) within the Priority Program SPP 1253.
http://www.siam.org/journals/sisc/29-6/67662.html
Numerical Analysis and Computational Mechanics Group, University of Twente, Faculty
EEMCS, Department of Applied Mathematics, P.O. Box 217, 7500 AE Enschede, The Netherlands
(kosterdtp@ewi.utwente.nl).
2352
Copyright by SIAM. Unauthorized reproduction of this article is prohibited.
Downloaded 12/30/12 to 129.173.72.87. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS
2353
Fig. 1.1. Example of a biochip.
A novel type of microuidic biochip employs surface waves as a driving force.

The key of this technology is a pump able to position reagents on the surface of
chips or in microuidic channels without mechanical contact. This is achieved using
surface acoustic waves induced using radio frequency electric signals. The waves arise
through the use of piezoelectric substrate materials in the chip, e.g., lithium niobate.
The electromagnetic signal is eciently converted into an elastic wave conned to the
surface layer of the substrate, the hence surface acoustic wave (SAW).
The interaction of these waves with the uid leads to streaming patterns in the
uid or of the motion of the uid as a whole. It is thus possible to mix minute
amounts of uid or conduct chemical reactions through precisely controlled transport
of reagents. See, e.g., [48, 21] for further details. Microuidic pumps based on this
technology are already available on the market; see [49].
In this work we will use the compressible NavierStokes equations as a starting
point to describe uid ow. This is justied since the typical value of the Knudsen
number, dened as the ratio of the characteristic physical length scale to the molecular
mean free path length, is much less than 1. Due to the presence of widely dierent
time scales in the physical problem we will derive further model equations to describe
the uid behavior.
The resulting two subproblems describe uid ow at a microscopic time scale,
where acoustic eects are dominant, and at a macroscopic time scale, where the
streaming eects are observable. The solution of the microscopic model is necessary
to calculate the eective force terms in the macroscopic model. Our model ts into the
formal framework of heterogeneous multiscale methods as described in [14, 15]. This
general approach attempts to formalize the idea of performing microscale simulations
to close macroscale models.
Furthermore, we will consider the case of free capillary uid boundaries. This
problem is of interest because some biochip models do not employ closed uid channels. Instead these chips are constructed with a virtual channel geometry dened by
planar regions with dierent wetting properties on the chip surface. Fluids adhere to
the hydrophilic regions by surface tension and are thus contained; see [49]. Another
use of free capillary boundaries is for the passive lling of small cavities by surface
tension; see [10].
1.1. Prior work and aims of the paper. The mathematical modeling and numerical simulation of the interaction between SAWs and surrounding uids is a mostly
unexplored eld. This is mainly due to the novelty of the applied technology. The
book [24] may serve as a recent overview of small-scale uid-streaming phenomena.
2354
DANIEL KOSTER
Many numerical studies of acoustic streaming eects are already available. However, most of these use the older streaming force approach by Nyborg [32] as a model.
As an example we cite [40], where the authors simulated acoustic streaming in microuidic channels induced by disk actuators.
Another work of interest is [5] since SAWs are also employed there and simulated
with true uid-structure interaction. However, the focus of that work is the simulation
of rst order acoustic eects and not of acoustic streaming. Hence, the problem of
dierent time scales does not appear there.
The model used in this work was introduced by Bradley [6]. It provides a rigorous
treatment of boundary conditions often neglected in other works. As far as we know,
the present work contains the rst numerical simulation based on this improved model.
In this work we have extended this model to include the eect of free capillary uid
boundaries which complicates the numerical algorithm considerably. We note that
relatively few works exist which treat the simulation of compressible ows with free
capillary surfaces.
The main contribution of this paper is a viable numerical algorithm to predict the
deformation of free droplet shapes under the inuence of acoustic streaming. The numerical results suggest that the method possesses good stability properties, although
much numerical eort is currently invested for solving the microscale subproblem.
Alternative approaches for the microscale subproblem are current work.
1.2. Outline of this work. We start by presenting the physical fundamentals
in section 2. The classical NavierStokes equations of hydrodynamics serve as a basis
to develop the numerical model. The model consists of two separate systems of partial
dierential equations to describe the ow variables at pressure p and velocity v on
two dierent time scales.
Section 3 presents an analysis of the central mathematical problems arising in
the model. The problem can be characterized as a generalized instationary, compressible Stokes problem, supplied with initial and boundary conditions. We study
several dierent solution concepts: general instationary solutions, periodic solutions,
and time-harmonic solutions. We dene suitable function spaces as well as the used
concept of weak solutions within these spaces.
We will prove the existence and uniqueness of weak solutions, as well as continuous dependence on the problem data. We will additionally introduce a result
demonstrating the relationship between the dierent solution concepts.
Section 4 will introduce the extension of the prior problems to problems incorporating free capillary boundaries. Due to the complexity of the problem we restrict
ourselves to the study of appropriate numerical algorithms.
Section 5 presents an overview of the practical algorithms used to solve the discretized problems. We have used standard Lagrange-type conforming triangular and
tetrahedral nite elements.
The nal section presents results of numerical experiments. Academic test problems are formulated which serve to validate the software implementation. Problems
with realistic physical parameters are also dened which demonstrate the applicability
to real-life situations.
1.3. Notation and function spaces. In what follows we will denote by a
bounded domain in Rd . We will use standard notation from Lebesgue and Sobolev
space theory.
2355
Fig. 2.1. Working principles of a SAW-driven uidic device.
2. Derivation of the model. The biochips treated in this work contain a substrate layer consisting of a piezoelectric material, e.g., lithium niobate. Embedded on
the substrate surface are ne electrodes known as interdigital transducers (IDTs) due
to their comblike structure. When a radio frequency signal (typical frequencies are 100
MHz) is applied to the IDTs, the inverse piezoelectric eect causes a distortion of the
substrate layer. Using special crystal cuts one obtains coupled elastic-electromagnetic
waves in the substrate.
The chips used for manipulating uid are designed to favor a special type of wave
known as SAWs, also called Rayleigh waves. These waves spread out along the surface
of the substrate in the fashion of a miniature earthquake. The wavelength is dened
by the IDT geometry, and typical amplitudes of displacement are in the nanometer
regime. Rayleigh waves are restricted to a thin layer of several wavelengths on the
surface of the chip.
If the surface of the substrate is in contact with uid and if the normal component
of displacement is nonzero, part of the energy will be radiated into the uid. The thus
damped SAW is known as a leaky SAW (LSAW). The interaction between the LSAW
and the uid causes an internal streaming eect as explained below; see Figure 2.1
for an illustration. This streaming can be used to mix the uid or as a contact-free
pump. It is even possible [49] to dene a virtual uidic network on the chip surface
using hydrophobic and/or hydrophilic materials etched onto the substrate using photolithographic techniques. Here we are interested in the eect of the streaming on
free capillary boundaries.
Our approach is based on a splitting into two subproblems, with coupling being
either neglected or treated empirically. The analysis and numerical simulation of
the elastic/electromagnetic subproblem describing the creation of SAWs was treated
in [17]. In this work we will focus on the coupling of LSAWs with the uid. This
interaction will be modeled in only one direction. To be precise, we use an empiric
LSAW displacement as the input parameter for the streaming eect and neglect the
backward coupling of the uid on the elastic substrate.
2.1. Overview of the uidics problem. For the modeling of the uid ow we
use the compressible NavierStokes equations; see [27, 6]. Assume a bounded uid
domain (t) with the boundary d (t) (t) in contact with elastic walls partially

DANIEL KOSTER
2356
traversed by SAWs.

v
+ (v)v =
+ (v) = 0
(2.1)
t
u
v(t, x + u(t, x)) =
(t, x)
t
v = 0, p = p0 , (0) = 0
in (t), t > 0,
in (t), t > 0,
on d (t), t > 0,
for t = 0,
with the Newtonian stress tensor = (v, p) dened as

2
vi
vj
+ ( v)ij .
+
ij (v, p) = pij +
xj
xi
3
The quantities that appear are
v
p
,
u
n
uid velocity,
uid pressure,
uid density,
standard and bulk viscosity coecients,
SAW displacement from equilibrium position,
typically a harmonic function of time,
outer unit normal vector on .
The divergence of the stress tensor yields

( v).
= p + v + +
3
We will need a constitutive equation to close this system. We use an equation of
state arising from the assumption of constant entropy in time and space (see [11]).
This leads to a simple relation of the form
(2.2)
p p0 = a( 0 ) ,
where p0 , 0 are equilibrium states and a, > 0. We assume a linear dependence for
liquids such as water; thus, = 1.
2.2. Physical origin of acoustic streaming. As discussed above we assume
that ow in the uid region arises exclusively from the harmonic oscillation of the
solid boundary. When using standard acoustics theory, where nonlinear eects are
neglected and low frequencies and amplitudes are assumed, we expect the propagation
of acoustic waves in the uid. An additional eect observed in real life, however, is
that of a stationary average ow eld, not accounted for by standard acoustic theory.
This eect, known as acoustic streaming, was rst studied by Lord Rayleigh [35].
Further contributions to the study of this phenomenon were made by Lighthill, Eckart,
Nyborg, Riley, and others [36, 37, 31, 32, 28, 38, 16]. Two types of acoustic streaming
are generally distinguished today:
Rayleigh streaming is associated with the boundary layers of a uid near solid
surfaces.
2357
Quartz wind is associated with the dissipation of acoustic energy in the main
body of the uid.
Acoustic streaming eects are a result of the nonlinearity of the NavierStokes
equations (2.1) in combination with the viscosity. The main problem for the numerical
treatment of acoustic streaming is that extremely dierent time scales are involved.
Typical frequencies of the harmonic oscillation are 100 MHz, requiring a time discretization on the order of 108 s. The acoustic streaming relaxation times are of the
order 103 101 s.
2.3. Acoustic subproblem. We assume an expansion of the unknowns v, p,
and of the form

p = p(0) + p + 2 p + O 3 ,

(2.3)
= (0) + + 2 + O 3 ,

v = 0 + v + 2 v + O 3 .
The small parameter 0 < 1 is later xed as the ratio u0 /L, with u0 being the
maximal SAW displacement of the domain boundary and L being the typical length
scale used for a dimensionless formulation.
Set p(1) := p , p(2) := 2 p , and analogously , v, . All terms with index 0 are
constant in time and space and represent the known equilibrium state without SAW
driving. By gathering all terms of order O() we arrive at
(2.4a)
(2.4b)
(2.4c)
(2.4d)
(2.4e)
(0)
v (1)
(1) = 0
t
in (t), t > 0,
(1)
+ (0) v (1) = 0
t
p(1) = c20 (1) ,
u
(t, x)
v (1) (t, x) =
t
v (1) = 0, p(1) = 0
(0) = 0
in (t) t > 0,
on d (t), t > 0,
for t = 0,
where c0 is the small signal sound velocity in the uid. The boundary condition is
derived by observing that u = O() and performing a Taylor expansion:
2
v(t, x + u) = v(t, x) + (v)u + O(u ) =
u
(t, x).
t
In the future we will neglect the time dependence of arising through the SAW
oscillation. This is possible since the SAW displacement u is much smaller than the
typical length L, equivalent to the fact that 1.
This linear system describes the damped propagation of acoustic waves. These
arise solely as a reaction to the time-harmonic boundary values (2.4d), since the righthand sides are zero. Due to the parabolic character of the system and the boundedness
of we expect the solution to converge into a time-harmonic quasi-equilibrium state
for large t. Since the goal is the solution of the system (2.1) on the large time scale,
the startup eects are of less interest than this equilibrium state.
2.4. Acoustic streaming subproblem. Assuming that the above rst order
system (2.4) is solved we proceed to gather terms of second order in while treating

DANIEL KOSTER
2358
the solution of (2.4) as given data. The result is

(0)
(2.5a)
v (2)
v (1)
(2) = (1)
(0) (v (1) )v (1)
t
t
in (t), t > 0,
(2)
+ (0) v (2) = ((1) v (1) )
t
p(2) = c20 2 ,
(2.5b)
(2.5c)
in (t), t > 0,
v (2) = (v (1) )u
(2.5d)
v
(2.5e)
(2)
(2)
= 0,
= 0,
on d (t), t > 0,
(0) = 0
for t = 0.
This system can be interpreted as an instationary Stokes problem describing

creeping ow and containing mass sources and body forces.
Dene T := 2 as the period of the harmonic SAW oscillation. Since we are
interested in the acoustic streaming eects observable on the larger time scale, we take
a temporal average of the right-hand sides. More precisely, we apply the operation
a :=
1
T
t0 +T
a(t) dt.
t0
We thereby obtain the system
(0) v
(2.6)
(2)
(2) =
(1) v
(1)
(0)
(v
(1)
)v
(1)

(2)
(2) = ((1) v (1) )
+ (0) v
t
p(2) = c20 2 ,
(2)
= 0,
= 0,
in (t), t > 0,
in (t), t > 0,
(2) = (v 1 )u

v
(2)
on d (t), t > 0,
(0) = 0
for t = 0.
(2) , p(2) , (2) and

In the following we will deal only with the averaged terms v
therefore denote these without the bars for simplicity. The corresponding stationary problem is also of interest when performing simulations without free capillary
boundaries.
(2.7a)
(2.7b)
(2)
(0) v (2)
(1) v
(1)

= ((1) v (1) )
=
(2.7c)
p(2) = c20 2 ,
(2.7d)
v (2) = (v 1 )u
(0)
(v
(1)
)v
(1)
in ,
in ,
on d .
2.5. Free capillary boundaries. In addition to the problems with xed channel geometries we will also treat the case of free capillary surfaces along parts of the
boundary. To prepare this we rst rewrite (2.1) using additional boundary conditions
which are valid on a free capillary boundary f (t) between liquid and air. The new
boundary conditions are necessary to determine the normal velocity of the moving
2359
interface as well as the stress acting due to capillary forces.

v
(2.8a)
+ (v)v =
in (t), t > 0,
(2.8b)
+ (v) = 0
in (t), t > 0,
t
u
v(t, x + u(t, x)) =
(t, x)
on d ,
(2.8c)
t
(2.8d)
n = (d 1)
on f (t), t > 0,
vn=V
on f (t), t > 0,
(2.8e)
v = 0, p = p0 , (0) = 0
(2.8f)
for t = 0.
Here we have as the constant of surface tension between water and vacuum,
the dierence to the conditions of water/air interfaces being negligible, (cf. [27]),
f (t) as the free boundary, and V as the normal velocity component of the
free boundary. Furthermore we have as the vector of curvature. By denition this
vector has magnitude H (mean curvature of (t)) and points in the direction of n
(outer unit normal vector of (t)).
The location of f (t) will change signicantly in time and now represents one
of the unknowns of the problem. To adapt the two-scale model for the inclusion of
capillary boundary eects, we include the capillary stress condition (2.8d) in both
subproblems.
For the sake of consistency we will continue to neglect the motion of the domain
when solving the rst order acoustics problem (2.4). However, we will need to take it
into account when solving the large time scale acoustic streaming subproblem (2.7).
3. Analytical results. In this section we will analyze a mathematical model of
the acoustics and acoustic streaming subproblems introduced in section 2 for a xed
domain without free boundaries. Both subproblems (2.4) and (2.7) have the following
general form: Given right-hand sides f v , fp , boundary values g, and initial values
v 0 , p0 , nd v and p satisfying
(3.1a)
(3.1b)
(3.1c)
(3.1d)
v
1 v 2 ( v) + p = f v
t
p
+ v = fp
p
t
v=g
v = v 0 , p = p0
in , t > 0,
in , t > 0,
on , t > 0,
in , t = 0,
with positive constant parameters v , p , 1 , 2 . We will consider both the timedependent and the stationary cases v = p = 0 which requires dierent methods.
As before, we assume to be a bounded Lipschitz domain. We will use the
following variational formulation of (3.1):

v
v
w + 1
v : w + 2 ( v)( w)
t

p w =
fv w
for all w C
0 (),

p
q v =
fp q
for all q C0 ().
p
q+
t
In our analysis we will assume v = p .

DANIEL KOSTER
2360
To deal with the instationary problem we will introduce three dierent solution
concepts, each requiring a separate theoretical treatment and numerical implementation. The validity of each approach is determined by the form of the problem data
(right-hand sides f v , fp and boundary conditions g):
general time-dependent solutions, requiring no assumptions on the form of
the problem data,
time-periodic solutions, requiring periodic problem data,
time-harmonic solutions, requiring problem data given as a superposition of
sine and cosine terms.
Periodic solutions are of interest, since the application to SAW-driven acoustic streaming often involves time-periodic data f v , fp , g. In the case of time-harmonic problem
data it is possible to eliminate the time dependence altogether and reduce the problem to an equivalent stationary system. This so-called quasi-stationary approach is
treated later in this section.
3.1. Solution spaces and variational formulation. For the variational formulation we dene the following spaces:
V = H 10 (),
Q = L20 (),
U = H 10 () L20 (),
H = L2 () L20 (),
U = H 1 () L20 ()
as the dual space of U.
This yields a Gelfand triple

U H U
(3.2)
with each space continuously embedded and dense in the subsequent space; see
e.g., [50].
For a given bounded time interval [0, T ] R we dene the Hilbert space
W (0, T ) = {f H 1 (0, T ; U ) | f L2 (0, T ; U )},
with the scalar product

(f1 , f2 )W (0,T ) =
(f1 (t), f2 (t))U dt +
(f1 (t), f2 (t))U dt
for all f1 , f2 W (0, T ).

Thanks to well-known continuity properties of this space we will be able to justify
the specication of initial values in H. We now dene the following bilinear forms:
(3.3a)
(3.3b)
a : V V R,

a(v, w) = 1
v : w + 2 ( v)( w),
(3.3c)
(3.3d)
b : Q V R,

b(p, v) =
p v,
(3.3e)
(3.3f)
s : U U R,
s((v, p), (w, q)) = a(v, w) + b(q, v) b(p, w).
2361
Lemma 3.1. The following properties hold for the bilinear forms dened in (3.3):
1. The form a is V -elliptic, meaning that there exist constants ca , Ca > 0 so
that for all v, w V
|a(v, w)| Ca v1 w1 ,
2
a(v, v) ca v1 .
2. There exist constants cb , Cb > 0 so that for all v V and p Q
|b(p, v)| Cb p0 v1 ,
(3.4)
b(p, v)
cb .
pQ\{0} vV \{0} p0 v1
inf
(LBB)
sup
The inequality (LBB) is the well-known Ladyzhenskaya-Babuska-Brezzi condition.

3. The form s is U -coercive (see, e.g., [29]); in other words, there are constants
cs , Cs > 0, ks R so that for all f, g U
|s(f, g)| Cs f U gU ,
(3.5)
s(f, f ) + ks f H cs f U .
(3.6)
We can choose cs = ks = ca .
Definition 3.2. Using Lemma 3.1 we dene the following bounded linear operators. Let v, w V , p, q Q, f, g U .
A L(V , V ) :
B L(Q, V ) :
(3.7)
B L(V , Q) :
S L(U, U ) :
Av, wV V = a(v, w),

Bp, vV V = b(p, v),
B v, pQQ = b(p, v),
Sf, gU U = s(f, g),
K L(Q, Q) :
K = B A1 B.
Also useful will be the bilinear form corresponding to the Schur complement operator
K:
k : Q Q R,
k(p, q) = (Kp, q)0 .
3.2. Instationary solutions. We can now formulate the variational formulation of (3.1).
Problem 3.3 (weak formulation of the instationary problem). Let u0 = (v 0 , p0 )
H and f = (f v , fp ) L2 (0, T ; U ). Find u = (v, p) W (0, T ) so that
(P)
w, u (t)U U + s(w, u(t)) = w, f (t)U U

u(0) = u0
for all w U, f.a.a.1 t (0, T ),

in H.
We remark that Problem 3.3 is well-posed due to the properties of the solution
space W (0, T ).
1 The abbreviation f.a.a. will stand for for almost all with respect to the used Lebesgue
measure.

DANIEL KOSTER
2362
Lemma 3.4. The problem (P) can be equivalently formulated as

Find (v, p) W (0, T ) so that
w, v (t)V V + a(w, v(t)) + b(p(t), w) = w, f v (t)V V
(q, p (t))0 b(q, v(t)) = (q, fp (t))0
(P )
for all w V ,
for all q Q,
f.a.a. t (0, T ),
v(0) = v 0
in L2 (),
p(0) = p0
in Q.
Proof. Assume (P ) holds. Set w = (w, q), u0 = (v 0 , p0 ), and sum the rst two
equations to get (P). Now assume (P) holds. Select rst = (w, 0) to obtain the
rst variational equality in (P ) and then = (0, q) for the second one. The initial
condition is clear.
The stationary problem corresponding to (3.1) can be interpreted as a compressible Stokes system.
Problem 3.5 (weak formulation of the stationary problem). Let f = (f v , fp )
U . Find u = (v, p) U so that
(Pstat )
a(w, v) + b(p, w) = w, f v V V

b(q, v) = (q, fp )0
for all w V ,
for all q Q.
An equivalent reformulation of (Pstat ) using the form s is also possible using the
same methods as in Lemma 3.4.
Remark 3.6. The choice of Q as pressure space is a real restriction as the following
example shows.
Given = (0, 1)2 , solve the problem
(3.8)
v
(t) v ( v) + p = 0
t
p
(t) + v = 1
t
v(x, t) = 0
v 0 (x) = 0,
in ,
in ,
on ,
p0 (x) = 0.
We immediately verify that a solution is given by v(x, t) = 0, p(x, t) = t, which does

not lie in W (0, T ).
This shows that working with Q = L20 () is too restrictive in this setting. However, we use this space for simplicity in this presentation. An extension of the existence and uniqueness theorems presented later to the case of generalized solutions
(v, p) sought in the larger space H 1 () L2 () may be found in [26].
Since the Ladyzhenskaya-Babuska-Brezzi condition (LBB) is fullled we can deduce the existence and uniqueness of solutions of Problem 3.5.
Lemma 3.7. Problem (Pstat ) has a unique solution (v, p) U . The following a
priori estimate holds:
v1 + p0 C(f v V + fp 0 ),
with a constant C > 0.
Proof. Given the properties of the bilinear forms in Lemma 3.1, we can use the
saddle point theory specied in [34]. Additional references are [19, 25, 7].
2363
We observe that (P) denes a parabolic problem due to the properties of s(, )
explained in Lemma 3.1. We cite the following theorem.
Theorem 3.8 (existence and uniqueness of (P)). For 0 < T < the problem
(P) possesses a unique solution (v, p) W (0, T ). Furthermore, we have the following
stability estimate:

(v, p)W (0,T ) C (v 0 , p0 )H + (f v , fp )L2 (0,T ;U ) .
The constant C will generally depend on T .
Proof. [50, Theorem 26.1].
We also state the following result for time regularity of the solution.
Theorem 3.9 (time regularity of the solution of (P)). Assume u = (v, p) as the
solution of the parabolic problem (P) stated in the following form:
u + Su = f
a.e. in (0, T ),
u(0) = u0
for t = 0.
Let 0 < k N. We assume f H k (0, T ; U ) and u0 fullling the following compatibility conditions:
w0 := u0 U,
dj1 f
(0) Swj1 U, j = 1, . . . , k 1,
dtj1
dk1 f
wk := k1 (0) Swk1 H.
dt
wj :=
We then have
u H k (0, T ; U )
and
dk+1 u
L2 (0, T ; U ).
dtk+1
Proof. [50, Theorem 27.2].

3.3. Periodic solutions. For the acoustics subproblem (2.4) we generally have
a time-periodic driving force u. It is therefore of interest to reformulate Problem 3.3
for right-hand sides f which have period T and search for periodic solutions. This
changes the character of the problem from an initial value problem to a periodic
problem, where u(0) and u(T ) are unknown a priori.
Later in this section we will compare the solution u of the initial value problem (P)
supplied with a periodic forcing f to the solution uper of the corresponding periodic
problem. We expect the solution of the initial value problem to converge in some sense
towards a periodic equilibrium state, that is, uper , for large times. This phenomenon
is known as forced oscillation in physics. In [26] we present a theorem conrming this
convergent behavior.
We rst specify what is meant by a periodic solution. The denition may be
surprising at rst since periodicity is not mentioned explicitly. The characterization
given in Proposition 3.11 will justify the terminology.
Problem 3.10 (periodic solutions). Let 0 < T < and f = (f v , fp )
L2 (0, T ; H). Find u = (v, p) W (0, T ) so that
(Pper )
w, u (t)U U + s(w, u(t)) = (w, f (t))H

u(0) = u(T )
for all w U, a.e. in (0, T ),

in H.
2364
DANIEL KOSTER
Note that u(0) and u(T ) are unknowns, not data of the problem. We demand
more regularity in space, i.e., f (t) H. This will enable us to use elementary results
of Fourier analysis (see [1]) for the derivation of a priori estimates.
Proposition 3.11. Assume that the right-hand side f is extended to a periodic
function f by
f (t + kT ) = f (t)
for t [0, T ), k Z.
A solution u of (Pper ) can then be uniquely extended to a periodic solution u

fullling
w, u
(t)U U + s(w, u
(t)) = (w, f (t))H
for all w U, f.a.a. t R.
Proof. The existence and uniqueness Theorem 3.8 implies that the solution u
with initial values u(0) is well dened for an arbitrarily large interval [0, ]. Dene
z(t) := u(T + t) u(t). The function z is a solution of the following initial value
problem:
z (t) + Sz(t) = 0,
z(0) = 0.
The a priori estimate of Theorem 3.8 then implies z = 0, and therefore u(t) = u(T +t).
A similar argument shows that the solution can be uniquely extended backwards in
time.
Theorem 3.12 (existence and uniqueness of (Pper )). The problem (Pper ) has a
unique solution (v, p) W (0, T ). We have the stability estimate
(v, p)W (0,T ) C (f v , fp )L2 (0,T ;H) .
The constant C will generally depend on the period T .
The idea of the proof is to use the standard Galerkin method. In a rst step
we establish stability properties of corresponding nite-dimensional problems. These
are extended to the full problem using standard compactness results. For the nitedimensional problem we used Fourier analysis to eliminate the time dependence of
the periodic problem. The key elements of the stability proof are elaborate estimates
on the eigenvalues of the compressible Stokes operator S dened in (3.7). These were
taken from [4]. For the complete proof see [26, Theorem 3.2.6].
3.4. Quasi-stationary approach. In the physical applications the forcing terms
are typically given by a harmonic signal:
(3.9)
(x) cos(t) + f
(x) sin(t),
f v (t, x) = f
vc
vs

fp (t, x) = fpc (x) cos(t) + fps (x) sin(t),
c (x) cos(t) + g
s (x) sin(t),
g(t, x) = g
with a frequency = 2/T > 0. As proven in [26, Theorem 3.2.10] the solution of
the acoustic system will converge against a periodic equilibrium state for t . As
will become clear later this periodic equilibrium state must also be time-harmonic.
Since the equilibrium state is what we are interested in when solving the full problem,
we may attempt to split the spatial and temporal dependence of the acoustic eld
solution. This eliminates the need for a time discretization. To be precise, we consider
the following ansatz:
(3.10)
c (x) cos(t) + v
s (x) sin(t),
v(t, x) = v
p(t, x) = pc (x) cos(t) + ps (x) sin(t),
2365
with a frequency equal to the driving frequency of the boundary conditions and
right-hand sides in (3.9). This idea is standard for solving time-harmonic problems,
although a decomposition using the complex terms exp(it), exp(it) is usually
used instead.
After substituting these formulas for v and p in system (3.1) and using the linear
independence of sine and cosine functions, we arrive at four new eld equations in the
i , pi , i = c, s:
unknowns v
,
s 1
c ) +
v c 2 ( v
pc = f
v v
vc
,
c 1
s ) +
v v
v s 2 ( v
ps = f
vp
c = fpc ,
p ps + v
s = fps
p pc + v
c ,
c = g
v
s = g
s
v
(3.11)
in ,
on .
We are now faced with solving only one stationary system at the price of twice as
many unknowns as before. A variational formulation of this system is derived as
before. This variational formulation can be analyzed using well-known results for
saddle point problems; see [34, Theorem 2.2], [25], and also [7].
4. Methods for free boundary problems. This section will treat the discretization of free capillary problems associated with (3.1). Most of the available
literature for free capillary boundary problems concentrates on the incompressible
(nonlinear) NavierStokes system; see [3] and references therein. It is interesting to
note that the nonlinear convection term, together with the assumption of incompressibility, may actually serve to prove the stability of numerical schemes; see [3, p. 6].
Analytical results of existence and uniqueness, even for the more well known case of
incompressible NavierStokes, are applicable only to special domains or small times or
rely on function spaces with high regularity properties unsuitable for a nite element
method (FEM) discretization; see [30, 45, 42].
We will treat the case of Dirichlet and free capillary boundary conditions specied
on subsets d (resp., f ) of , d f = . We will assume that d is a topologically
closed surface, implying that the contact line between the free surface and walls
remains xed. We note that this hypothesis is not always valid in physical situations.
However, moving contact lines cause problems in mathematical formulations due to
the incompatibility of boundary conditions; see the rst section of [45].
This yields the following system:
(4.1a)
(4.1b)
(4.1c)
(4.1d)
(4.1e)
(4.1f)
(4.1g)
(4.1h)
v
= fv
t
p
p
+ v = fp
t
v=g
n = (d 1)
vn=V
in (t), t > 0,
in (t), t > 0,
on d ,
on f (t), t > 0,
on f (t), t > 0,
(0) = 0 ,
v(t = 0, x) = v 0
p(t = 0, x) = p0
in 0 ,
in 0 ,

DANIEL KOSTER
2366
with

ij = pij + 1
V
>0
vi
vj
+
xj
xi

+ (2 1 )( v)ij
the stress tensor,
the normal velocity of f (t),

the vector of mean curvature,
the coecient of surface tension.
We are interested in a sharp interface arbitrary Lagrangian Eulerian (ALE) approach, where the domain essentially follows the motion of the uid; see [23, 22] and
references therein. This seems most promising because of its simplicity, at the price
of not being able to deal with domain topology changes, for example, due to the formation of cusps. Examples of more general methods are volume-of-uid [20] or the
more recent level set methods [33, 47, 44, 46, 43]. In the standard physical situations
considered we do not expect topology changes.
4.1. Variational formulation including curvature. To incorporate free capillary boundaries in the nite element framework, we will need a variational formulation of the curvature term [12, 13].
We assume Rd to be a subset of an orientable (d 1)-dimensional smooth
manifold M and : U , U an open subset of Rd1 , as a local parameterization of
. We dene gij = i j as the metric tensor of and g ij = (g 1 )ij as its inverse.
For functions f : R, f : Rd we introduce the following dierential
operators on :
d1

( f ) :=
g ij i (f )j
i,j=1
as the tangential derivative of f ,

d1

( f ) :=
g ij i (f ) j
i,j=1
as the tangential divergence of f , and

( f ) := ( f ) =
d1

i,j=1

1
i
det g g ij j (f )
det g
as the LaplaceBeltrami operator of f ; see, for instance, [2, 9]. We have the following
identity:
id = ,
where is the mean curvature vector on . Let be a smooth vector valued function
dened on , with = 0 on the relative boundary relbd() in M . Using partial
integration on we get

(4.2)
dHd1 = ( id ) dHd1

d1
= id : dH
= dHd1 .
2367
There is no boundary term in the partial integration because = 0 on relbd(). We

now return to the system (4.1). If we test (4.1a) with a smooth function dened
on (t)
and fullling = 0 on d , we get

v
fv =
( )
v
t

=
()
:

=
() n dHd1
: .
Thanks to (4.2) and the boundary condition (4.1d) the boundary integral is equal to

() n dHd1 =
(n) dHd1 = (d 1)
dHd1 .
The second integral evaluates to

: =
p( )
1
2
D(v) : D()

(2 1 ) ( v)( ),
with the deformation tensor D(v) := v +v T . Combining the last equations nally
yields

(4.3) v
v 1
+
t
2

D(v) : D() + (2 1 ) ( v)( )

p( ) =
f v (d 1)
dHd1 .
This will be the starting point of the discrete formulation of the free boundary problem
in the following section.
We will rst review the ALE approach used to treat the motion of the free capillary
surface. The basic idea is to use a special nite element space dened on a reference
triangulation which will contain the coordinate function dening the mesh nodes
at any time step. We will use a time-marching procedure as before to calculate
the evolution of the velocity and pressure elds. The velocity at the free capillary
boundary will be used to update the mesh position for the next time step.
4.2. ALE formulation. For the beginning we consider the extension of the
Rd be
evolution problem (P) to a time-dependent but given domain (t). Let
a bounded reference domain. We dene a time-dependent continuous transformation
to the current domain (t):
to map
Rd ,
: [0, T ]
= (t),
(t, )
G = {(t, x) [0, T ] Rd | x (t)}.

DANIEL KOSTER
2368
We dene a reverse transformation as follows:
: G ,
)) = x
.
(t, (t, x
We will assume that the following natural requirements for a proper space-time
domain G are fullled.
Assumption 4.1 (admissible transformation). Let t [0, T ]. Then
(t) is injective, and
(t, ) :
inf x det Dx (t, x
) > 0.
Here Dx (t, x
) stands for the Jacobian of with respect to the space variable.
For a given function : G R we can therefore dene a related reference function
as
R,
: [0, T ]
(t,
x
) = (t, (t, x
)).
Let and : G R be suciently smooth to apply the chain rule of dierentiation. We have
(4.4)
Dt (t,
x
) = Dt (t, (t, x
)) + Dx (t, (t, x
))Dt (t, x
)
),
= Dt (t, x) + Dx (t, x)Dt (t, x
where x = (t, x
). The second term containing spatial derivatives of is actually
omitted in the numerical discretization, as it is of higher order with respect to the
expansion parameter and since we are mostly interested in a stationary limit. See
[26] for details.
For a discretization in space as well as in time we assume that we have a niteh U
dened on
= 0 . Furthermore, we assume
dimensional reference subspace U
that a sequence of discrete time steps 0 = t0 < t1 < < tN = T are given with
n , n : n .
corresponding mesh transformations nh :

h
We now seek the discrete solution
h }
unh Uh,n = {h nh | h U
for n = 1, . . . , N .
4.3. Algorithm. We will now dene an algorithm to discretize the free boundary
with a conforming triangulation
problem (4.1). We assume that a reference domain
T is given. We dene f as the part of the boundary where free capillary

d
as the part with Dirichlet boundary
boundary conditions are prescribed and
d to be a relatively closed surface in
and require
=
conditions. We choose
f
d,
f
d = .
We assume that T is chosen in such a way that the boundary faces of the simplices
of dimension d 1 of
f . We split the
S T induce a conforming triangulation R
simplices of T into two sets

f = },
Ti := {S T | S
Tf := T \ Ti .
We thus have the property
= {T | T is d 1 dim. simplex, T = S ,
S Tf }.
R
2369
We dene the nite element spaces for the discrete velocity, pressure, and mesh position in the interior and at the free boundary as follows:
d | wh |S P2 (S)d for all S T , wh | = 0},
V h = {wh C 0 ()
d
h = {qh C 0 ()
| qh |S P1 (S) for all S T },
Q
h = {h C 0 ()
d | h |S P1 (S)d for S Ti and h |S P2 (S)d for S Tf },
M
h = {xh C 0 (
f )d | xh |T P2 (T )d for T R}.
R
In other words, we use the TaylorHood element of order 1 for velocity and
pressure, piecewise linears for the interior simplices, and piecewise quadratics for the
boundary simplices. The choice of piecewise quadratics implies the use of curved
simplices at the free boundary which is desirable for precise modeling of discrete
curvature; cf. [3].
h at time step n, we dene the transformed mesh
Given a transformation nh M
and spaces as follows:
d ) (stays xed),
d := 0h (
T n := {n (S) | S T },
n := nh (),
f ),
f,n := n (
h
(4.5)
Tin
:=
R :=
n
{nh (S)
{nh (T )
| S Ti },
| T R},
Tfn
nh
h nh | w
h V h },
V h,n := {w
h },
h M
Mh,n := {
xh n | x
h
:=
:=
Qh,n :=
h
{nh (S) | S Tf },
(nh )1 : n ,
h },
{
qh nh | qh Q
h }.
h R
Rh,n := {
xh nh | x
Since the domain will change with time, we dene time-dependent bilinear forms
for a variational formulation:
on the domain n = nh ()
an : V h,n V h,n R,

1
an (v h , wh ) =
D(v h ) : D(wh ) + (2 1 )
( v h )( wh ),
2 n
n
(4.6)
bn : Qh,n V h,n R,

bn (qh , v h ) =
qh v h ,
n
dn : Rh,n Rh,n R,
dn (v h , wh ) = n (d 1)

f,n
f,n v h f,n wh dHd1 .
Two more tools are needed to dene the discretized free boundary algorithm: a
discrete outer unit normal vector nnh Rh,n to the mesh boundary f,n and a suitable
h M
h M
h.
extension operator Eh : R
n
The function nh Rh,n is dened by setting the values in the nodes of the
quadratic Lagrange mesh G2 of f,n . If T Rn is a d 1-dimensional simplex, we
set
T := nT Hd1 (T ),
n
P (x) := {T Rn | x T }.

DANIEL KOSTER
2370
Here nT is the (well-dened) outer unit normal vector on T . Now dene nnh Rh,n
as

T P (x) n T
if x f,n ,
n
nh (x) := | T P (x) n T |
0
otherwise
for all nodes x of G2 . This means that the (well-dened) normal on faces is averaged
to dene values on corners or edges.
The extension operator Eh is used to dene the interior nodes of n , based on
the position of f,n and the old mesh n1 :
nh = Eh (nh | f , n1
).
h
This is necessary, since the boundary condition (4.1e) prescribes only the motion of
the free boundary. The interior nodes should be selected in such a manner as to
guarantee a good quality of the mesh triangulation T n .
h and v 0 V h,0 ,
Algorithm 4.2 (free boundary discretization). Let 0h M
h
n
0
n
ph Qh,0 , as well as f v , fp as locally integrable functions on Rd , n N, and g h
as the extension by zero of given Dirichlet boundary conditions for the velocity. Loop
over n = 1, 2, . . . :
1. Solve for (
v nh , pnh ) (g h + V h,n1 ) Qh,n1 :
(4.7)
1
n

n1
nh + an1 ( h , v
nh ) + dn1 ( h , v
nh ) + bn1 (
h v
pnh , h )

1
h v n1
=
+
h f n1
v
h
n n1
n1

(d 1)
f,n1 h dHd1
f,n1
for all h V h,n1 and

(4.8)
1
n

n1
h)
h pnh bn1 (h , v
=
1
n

n1

h pn1
+
h
n1
h fpn1
for all h Qh,n1 .

2. Update the position of the free boundary as follows: Dene
:= n (
v nh |f,n nn1
)nn1
Rh,n1 ,
n1
h
h
h

n
n
n1
h | f := idf,n1 +h h | f .
3. Now update the position of the entire mesh:
nh := Eh (nh | f , n1
),
h
with a suitable extension operator Eh . If nh is not an admissible mesh transformation according to Assumption 4.1, then STOP.
2371
4. Finally, set
n := nh (),
f ),
f,n := nh (
T n := {nh (S) | S T },
Rn := {n (T ) | T R},
h
nh := (nh )1 ,
nh n1
v nh := v
nh ,
h
pnh := pnh n1
nh .
h
Remark 4.3. Several remarks about the free boundary algorithm are in order.
The algorithm employs a decoupling of the bulk equations and the geometry
updates, which would otherwise be coupled in a nonlinear way.
The discretization has the following property:
(4.9)
nh ) (d 1)
dn1 ( h , v

= (d 1)

(d 1)
f,n1
f,n
f,n1
f,n1 h dHd1
nh ) dHd1 ,
f,n1 h f,n1 (idf,n1 +n v
h dHd1 ;
cf. (4.3). The capillary boundary condition is thus treated in a semi-implicit

manner, since information about the new boundary position f,n enters the
denition of the curvature term. Furthermore, note that the term dn adds
additional coercivity on the left-hand side.
The idea for this discretization of curvature was taken from [3], where it is
crucial for proving stability of a discontinuous Galerkin scheme for the free
boundary incompressible NavierStokes equations.
We have dened the update of the free boundary as

nh | f := idf,n1 +n (
n1
v nh |f,n nn1
)nn1
| f .
h
h
h
A simpler formula, motivated by (4.9), is

nh |f,n n1
| f .
(4.10)
nh | f := idf,n1 +n v
h
This method is actually used during the mesh update Algorithm 4.4 below.
However, in the case of large variations of the tangential velocity at the boundary (4.10) may lead to a quick distortion of the mesh, making the denition of
a suitable mesh extension Eh even more dicult. Moving the free boundary
only in the normal direction is justied since the set f,n remains essentially
the sameonly the nodes dening the triangulation Rn are placed dierently.
We conclude the discussion of the algorithm by presenting two implemented alh M
h M
h.
gorithms to dene the mesh extension Ehn : R
Algorithm 4.4 (mesh update algorithm 1). Given old mesh positions n1
nh V h,n1 of step 1 in Algorithm 4.2, dene nh =

Rh and the velocity eld v

DANIEL KOSTER
2372
E(nh | f , n1
) as follows:
h

nh n1
|
nh |

f := idn1 +n v

f ,
h
nh | d := 0h | d .
In other words, all interior nodes as well as the nodes on the free boundary are moved
nh .
with the velocity eld v
Algorithm 4.5 (mesh update algorithm 2). Given are old mesh positions
h as well as the new positions of the free boundary n | . Determine
n1
R
h f
h
h Mh,n1 (see (4.5)) as the solution of the elliptic system

h : h = 0
for all h Mh,n1 ,
n1
h = nh | f n1
h
on f,n1 ,
h = idd
on d .
Now set nh = E(nh | f , n1

) as
h
.
nh := h n1
h
This idea of using the Laplace operator to smooth the mesh nodes was taken from
[3]. Although simple and ecient to implement, this method has certain drawbacks;
e.g., it cannot deal with nonconvex domains without modications. Nevertheless, this
algorithm was sucient for our purposes. We refer to [39] for more sophisticated ideas
on mesh smoothing.
5. Discretization. The simulation of acoustic streaming induced by SAWs was
realized by the two-stage model already introduced in section 2. We used the second order TaylorHood nite element for the spatial discretization of the Stokes-like
acoustic and acoustic streaming subproblems. The resulting discrete saddle point
problem is solved using a CG iteration to invert the corresponding Schur complement
of the system. For details we refer to [8]. To achieve a temporal discretization we
employed either backward Euler or the CrankNicolson scheme.
We now present detailed algorithms to calculate solutions. We dierentiate between the xed mesh case and the case of free capillary boundaries.
5.1. Fixed meshes.
Algorithm 5.1 (acoustic streaming, xed domain). Given is a computational
domain Rd , physical parameters, and a periodic function u as the SAW displacement on d = .
1. Solve the acoustics subproblem (2.4) for (v (1) , p(1) ). We are interested in the
periodic equilibrium state corresponding to the periodic forcing terms. Two
methods are implemented:
(a) Perform a time discretization with xed time step size (1) > 0 together
with TaylorHood nite elements with mesh parameter h. We perform
either a xed number of time steps or iterate until a given periodicity
condition is fullled at a time tend .
(b) Employ the quasi-stationary approach (no time discretization) as described in section 3.4.
2. Calculate the right-hand terms and boundary conditions for (2.7).
2373
3. Solve the acoustic streaming subproblem for (v (2) , p(2) ). Here we have the
following alternatives:
(a) We solve the stationary problem (2.7).
(b) We solve the instationary variant (2.6) with xed time step size (2) .
This is primarily of interest when determining the relaxation time of the
acoustic streaming ow.
5.2. Free capillary boundaries. For the simulation of acoustic streaming with
free capillary boundaries, we will need physical parameters as before and additionally
a coecient of surface tension. A typical value would be = 7.3 102 N/m.
The basis of this algorithm was already presented as Algorithm 4.2. It remains
to specify the right-hand sides f nv , fpn for each time step. These are now determined
by solving the acoustics subproblem as above.
Algorithm 5.2 (acoustic streaming, free boundaries). Given is a starting domain 0 Rd and a periodic function u as the SAW displacement on d . Loop
over n = 0, 1, 2, . . . :
1. Solve the acoustics subproblem (2.4) for (v (1) , p(1) ) on the domain n up to
the periodic equilibrium state. This is done as above in Algorithm 5.1, with
either a time discretization or the quasi-stationary approach. In the former
case iterate until periodicity is reached.
(2),n
2. Calculate right-hand terms f (2),n
, fp
and boundary conditions for the
v
acoustic streaming solution at time step tn using v (1) and p(1) as above.
3. Perform step 1 of Algorithm 4.2 to determine the acoustic streaming solution
(v (2),n , p(2),n ) at time step tn with n := (2) .
4. Update the position of the mesh to determine n+1 , as described in steps 2 and
3 of Algorithm 4.2. We have the alternatives 4.4 and 4.5 as mesh smoothing
methods.
Remark 5.3. Recalculating the acoustics solution (v (1) , p(1) ) in each time step tn of
the main iteration is extremely costly. In practice we have relaxed this requirement by
(2),n
, fp
continuing to use f (2),n
for several time steps before recomputing. Numerical
v
experiments show that the additional error involved does not have a signicant eect
on the equilibrium state of the solution (n , v (2),n , p(2),n ) reached for large n.
6. Numerical results. The goal of this section is to present simulation results of
SAW-induced streaming with realistic physical parameters. The computational eort
for solving xed domain problems is still acceptable for our 3D setupthe simulation
of streaming eects including free capillary boundaries, however, is possible only in
a simple two-dimensional (2D) conguration using the current implementation. The
reason for this is that the change of the domain in each time step necessitates the
recalculation of the acoustic eld solution; see below.
All calculations were performed on AMD OpteronTM 64 bit 2.6 GHz dual-head
processors with 8 GB memory. We used the freely available ALBERTA nite element
toolbox, [41].
6.1. Fixed domains.
6.1.1. 2D simulation. We refer to (2.1) for the notation. The 2D simulation
setting is a uid-lled square cavity = [0, 1 mm]2 with SAW displacements u
prescribed on the bottom edge. The velocity v is set to zero on the other three edges.
The SAW propagates from left to right with exponential attenuation. The exact
shape assumed in this work is based on wavelengths, wave velocities, amplitudes,

DANIEL KOSTER
2374
Fig. 6.1. Detail of computational mesh and SAW displacement eld in the bottom left corner
of the domain with a computational length scale (1 unit 101 m).
Table 6.1
Numerical and physical parameters for the 2D SAW simulation.
Parameter
V
L
T
f
c0,l
c0,s
0
u0
Cd
1
2
Value and units

1.0 101 m/s
1.0 107 m
1.0 108 s
1.0 108 Hz
1.484 103 m/s
3.8 103 m/s
1.0 103 kg/m3
1.0 109 m
8.06 103 1/m
1.002 106 m2 /s
1.002 106 m2 /s
Description
Dimensionless velocity scale
Dimensionless length scale
Dimensionless time scale
Frequency of the SAW device
Small signal sound speed in water
Small signal sound speed in solid
Density of liquid
Maximal SAW displacement
Damping parameter of the LSAW
Kinematic viscosity of water
Kinematic bulk viscosity of water
and damping parameters of a typical lithium niobate crystal as described in [18]; see
Figure 6.1.
This setup approximates the case of a vertically oriented thin slice of water
trapped in a container and touching the SAW-traversed solid surface at the bottom.
The uid is assumed to be initially at rest, i.e., v 0 = 0, p0 = 0.
Data of the 2D SAW device simulation. The function u to describe the SAW
displacement at the lower edge is dened in dimensionless form as
(6.1)
d x1 ) sin(2(
u1 (t, x) = 0.6 exp(C
kx1 + ft)),
d x1 ) cos(2(
u2 (t, x) = exp(C
kx1 + ft))
d = Cd L,
for x , x2 = 0 with parameters = u0 /L, C
k = L/, f = f T . Note
that the discretization parameter is 0.1; confer (2.3). Additional damping terms,
not mentioned here, serve to avoid sharp jumps of the function u at x = 0 and t = 0.
The other parameters used in the simulation are given in Table 6.1.
As described in section 2 the acoustic streaming eld (v (2) , p(2) ) is calculated via
a two-stage algorithm using the intermediate acoustics eld solution (v (1) , p(1) ). We
use the standard CrankNicolson time discretization algorithm for parabolic problems
2375
Fig. 6.2. Left: Acoustics subproblem solution: pressure eld p(1) at time t = 142 1.42 s.
Right: Plot of eective force fv for the acoustic streaming solution. Clearly visible is the reection of the pressure wave at the top boundary of the domain as well as the typical Rayleigh angle
corresponding to the ratio of sound speeds in liquid and substrate.
Fig. 6.3. Acoustic streaming solution: velocity eld v (2) in mm/s.
to solve for the acoustics eld (v (1) , p(1) ) at each time step. We calculate a sucient
number of time steps in order to reach an oscillating equilibrium state.
Results of the 2D SAW device simulation. The CPU time necessary for the full
calculation was approximately two hours. See Figures 6.2 and 6.3.

DANIEL KOSTER
2376
Fig. 6.4. Left: Domain for the 3D simulation runs. Right: Acoustics subproblem solution: slab
x3 0 of the pressure eld p(1) at time t = 200 2 s.
6.1.2. 3D simulation. The parameters used for the 3D simulation are similar
to the 2D case. The domain will again be a uid-lled rectangular cavity. We set
= [0, 250 m]2 [50 m, 50 m]; see Figure 6.4. The SAW traverses the bottom
face; on the other walls, we again prescribe zero Dirichlet boundary conditions for
uid velocity.
Data of the 3D SAW device simulation. The parameters are set to the same values
as before. The function u is dened in dimensionless form as
(6.2)
d x1 ) sin(2(
u1 (t, x) = 0.6 exp(C
kx1 + ft))d(x3 ),
d x1 ) cos(2(
u2 (t, x) = exp(C
kx1 + ft))d(x3 ),
u3 (t, x) = 0.
The terms have the same meaning as in (6.1) above. What is new is the term d(x3 ),
dened as
d(x3 ) = 4 106 (x3 500)(x3 + 500).
Observing that the domain depth 100 m corresponds to the dimensionless depth of
1000 units, we have thus extended the SAW shape of the 2D simulation to a parabolic
prole over the bottom face.
Results of the 3D SAW device simulation. The CPU time necessary for the full
calculation was approximately six days. The acoustic streaming velocity values v (2)
are shown in Figure 6.5.
We additionally took a vertical cross section of the 3D mesh to dene a 2D mesh,
together with values of the SAW displacement eld u along the cross section. This
permits a comparison of a 3D solution with a corresponding 2D solution. Figure 6.6
shows that the result is qualitatively close to the 3D eld.
6.2. Free boundaries. As described above, we will restrict ourselves to performing a 2D simulation. We will describe the deformation of an originally semicircular drop of water of radius 2500 0.25 mm posed above a SAW wave of the
same shape as in the xed domain 2D simulations. The maximal SAW amplitude
was u0 = 100 nm, a much higher value compared to the xed domain simulations.
2377
Fig. 6.5. Acoustic streaming solution: slab x3 0 of the velocity eld v (2) .
Fig. 6.6. Comparison with the corresponding 2D simulation: acoustic streaming velocity eld v (2) .
The coecient of surface tension assumed in the simulations is = 7.3 102 N/m
corresponding to a water/vacuum interface.
We solved 200 microscopic time steps of length (1) = 0.1 1.0 1 ns to determine the equilibrium acoustics solution needed for each macroscopic time step. The
acoustic streaming solution was calculated using macroscopic time steps of length
(2) = 1000 10 s.
2378
DANIEL KOSTER
Fig. 6.7. Free boundary solution. Top: acoustic pressure eld p(1) at microscopic time t(1) = 19
and macroscopic time t(2) = 2.81105 . Bottom: acoustic streaming pressure eld p(2) at macroscopic
time t(2) = 2.81 105 .
One important change was done in the calculations. The mass generation rate
(2),n
and outow g (2),n n used in step 3 of Algorithm 4.2 will, in general, not be
fp
balanced. This implies a change in the pressure mean over time. In the presence of free
capillary boundaries a change in the overall pressure leads to a change in volumethis
is undesirable and nonphysical behavior. We therefore restored the balance between
outow and mass generation in the numerical calculations by multiplying g (2),n with
a correction factor. Numerical experiments performed without this correction indeed
displayed a continuously decreasing volume of the droplets.
Results of the free boundary SAW simulation. The maximal CPU time for the
calculation of 10 time steps was approximately ve hours. The observed behavior
of the drop is characterized by initial damped oscillations. A deformed equilibrium
position is obtained after a short period of time.
Figure 6.7 shows the acoustic and acoustic streaming pressure elds p(1) (resp., p(2) )
in the droplet at dierent times.
2379
REFERENCES
[1] H. W. Alt, Lineare Funktionalanalysis, Springer-Verlag, Berlin, 1999.
[2] H. W. Alt, Analysis III, Lecture notes at the University of Bonn, 2004.
nsch, Numerical Methods for the Instationary NavierStokes Equations with a Free
[3] E. Ba
Capillary Surface, Habilitation thesis, University of Freiburg, 1998.
[4] M. Benzi and V. Simoncini, On the eigenvalues of a class of saddle point matrices, Numer.
Math., 103 (2006), pp. 173196.
[5] N. D. Botkin and V. L. Turova, Mathematical models of a biosensor, Applied Mathematical
Modelling, 28 (2004), pp. 573589.
[6] C. E. Bradley, Acoustic streaming eld structure: The inuence of the radiator, J. Acoust.
Soc. Am., 100 (1996), pp. 13991408.
[7] F. Brezzi and M. Fortin, Mixed and Hybrid Finite Element Methods, Springer-Verlag, Berlin,
1991.
[8] M. O. Bristeau, R. Glowinski, and J. Periaux, Numerical methods for the NavierStokes
equations. Applications to the simulation of compressible and incompressible viscous ows,
Comp. Phys. Rep., 6 (1987), pp. 73187.
[9] M. P. D. Carmo, Dierentialgeometrie von Kurven und Fl
achen, Vieweg Verlag, Wiesbaden,
1998.
[10] K. Chakrabarty and J. Zeng, Design automation for microuidics-based biochips, ACM
Journal on Emerging Technologies in Computing Systems, 1 (2005), pp. 186223.
[11] B. Desjardins and C.-K. Lin, A survey of the compressible NavierStokes equations, Taiwanese J. Math., 3 (1999), pp. 123137.
[12] G. Dziuk, Finite elements for the Beltrami operator on arbitrary surfaces, in Partial Dierential Equations and Calculus of Variations, S. Hildebrandt and R. Leis, eds., Springer-Verlag,
Berlin, 1988, pp. 142155.
[13] G. Dziuk, An algorithm for evolutionary surfaces, Numer. Math., 58 (1991), pp. 603611.
[14] W. Engquist and B. Engquist, The heterogeneous multiscale methods, Commun. Math. Sci.,
1 (2003), pp. 87132.
[15] W. E Engquist and B. Engquist, The heterogenous multi-scale method, http://arxiv.org/abs/
physics/0205048v2, 2006.
[16] C. Eckart, Vortices and streams caused by sound waves, Phys. Rev., 73 (1947), pp. 6876.
ster, K. G. Siebert, and A. Wixforth, Numerical
[17] A. Gantner, R. H. W. Hoppe, D. Ko
simulation of piezoelectrically agitated surface acoustic waves on microuidic biochips,
Comput. Vis. Sci., 10 (2007), pp. 145161.
dler, A. Wixforth, M. Kostur,
[18] Z. Guttenberg, A. Rathgeber, S. Keller, J. O. Ra
M. Schlinder, and P. Talkner, Flow proling of a surface acoustic wave nanopump,
Phys. Rev. E, 70 (2004), paper 056311.
[19] W. Hackbusch, Elliptic Dierential Equations, Springer Ser. Comput. Mech., 1992.
[20] C. W. Hirt and B. D. Nichols, Volume of uid (VOF) method for the dynamics of free
boundaries, J. Comput. Phys., 39 (1981), pp. 201225.
[21] J. Hoffelner, H. Landes, M. Kaltenbacher, and R. Lerch, Finite element simulation of
nonlinear wave propagation in thermoviscous uids including dissipation, IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 48 (2001), pp. 779786.
[22] A. Huerta and W. K. Liu, Viscous ow with large free surface motion, Comput. Methods
Appl. Mech. Engrg., 69 (1988), pp. 277324.
[23] T. J. Hughes, W. K. Liu, and T. K. Zimmermann, Lagrangian-Eulerian nite element formulation for incompressible viscous ows, Comput. Methods Appl. Mech. Engrg., 29 (1981),
pp. 329349.
[24] G. Karniadakis, A. Beskok, and N. Aluru, Microows and Nanoows. Fundamentals and
Simulation, Springer-Verlag, Berlin, 2005.
[25] R. B. Kellogg and B. Liu, A nite element method for the compressible Stokes equations,
SIAM J. Numer. Anal., 33 (1996), pp. 780788.
ster, Numerical Simulation of Acoustic Streaming on SAW-Driven Biochips, Ph.D.
[26] D. Ko
thesis, University of Augsburg, 2006.
[27] L. D. Landau and E. M. Lifschitz, Lehrbuch der Theoretischen Physik VI: Hydrodynamik,
Akademie Verlag, Berlin, 1991.
[28] J. Lighthill, Acoustic Streaming, Academic Press, London, 1978.
[29] J.-L. Lions and E. Magenes, Non-homogeneous Boundary Value Problems and Applications,
Springer-Verlag, Berlin, 1972.
[30] P.-L. Lions and N. Masmoudi, On a free boundary barotropic model, Ann. Inst. H. Poincare,
16 (1999), pp. 373410.
2380
DANIEL KOSTER
[31] W. L. Nyborg, Acoustic streaming, in Physical Acoustics, W. Mason, ed., Academic Press,
New York, 1965, pp. 265331.
[32] W. L. Nyborg, Acoustic streaming, in Nonlinear Acoustics, M. F. Hamilton and D. T. Blackstock, eds., Academic Press, New York, 1998, pp. 207231.
[33] S. Osher and J. Sethian, Fronts propagating with curvature-dependent speed: Algorithms
based on Hamilton-Jacobi-formulations, J. Comput. Phys., 79 (1988), pp. 1249.
[34] P. Ciarlet, Jr., J. Huang, and J. Zou, Some observations on generalized saddle-point problems, SIAM J. Matrix Anal. Appl., 25 (2003), pp. 224236.
[35] L. Rayleigh, On the circulation of air observed in Kundts tubes, and on some allied acoustical
problems, Philos. Trans. R. Soc. Lond., 175 (1884), pp. 121.
[36] N. Riley, Oscillating viscous ows, Mathematika, 12 (1965), pp. 161175.
[37] N. Riley, Acoustic streaming, Theoret. Comput. Fluid Dyn., 10 (1998), pp. 349356.
[38] L. D. Rozenberg, ed., Ultrasonic Technology, Plenum Press, New York, 1971, Chapt. HighIntensity Ultrasonic Fields.
[39] M. Rumpf, A variational approach to optimal meshes, Numer. Math., 72 (1996), pp. 523540.
[40] A. Sathaye and A. Lal, Numerical simulation of acoustic streaming near embedded microstructures inside microuidic channels, in Proceedings of the IEEE Ultrasonics Symposium, 2002.
[41] A. Schmidt and K. G. Siebert, Design of Adaptive Finite Element Software: The Finite
Element Toolbox ALBERTA, Springer-Verlag, Berlin, 2005.
[42] B. Schweizer, Free boundary uid systems in a semigroup approach and oscillatory behavior,
SIAM J. Math. Anal., 28 (1997), pp. 11351157.
[43] J. A. Sethian, Level Set Methods and Fast Marching Methods, Cambridge Monogr. Appl.
Comput. Math., 1999.
[44] P. Smereka, Level Set Methods for Two-Fluid Flows, Lecture notes from a short course given
at INRIA, 1996.
[45] V. A. Solonnikov, On some free boundary problems for the NavierStokes equations with
moving contact points and lines, Math. Ann., 302 (1995), pp. 743772.
[46] M. Sussman, E. Fatemi, P. Smereka, and S. Osher, An improved level set method for incompressible two-phase ows, Comput. & Fluids, 27 (1997).
[47] M. Sussman, P. Smereka, and S. J. Osher, A level set approach for computing solutions to
incompressible two-phase ow, J. Comput. Phys., 114 (1994), pp. 146159.
[48] T. Uchida, T. Suzuki, and S. Shiokawa, Investigation of acoustic streaming excited by surface
acoustic waves, in Proceedings of the 1995 IEEE Ultrasonics Symposium, IEEE, 1995,
pp. 10811084.
[49] A. Wixforth, C. Strobl, C. Gauer, et al., Acoustic manipulation of small droplets, Anal.
Bioanal. Chem., 379 (2004), pp. 982991.
[50] J. Wloka, Partielle Dierentialgleichungen, Teubner, Stuttgart, 1982.

MMMMMMM

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

MMMMMMM

Загружено:

Авторское право:

Доступные форматы

Downloaded 12/30/12 to 129.173.72.87. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.

SIAM J. SCI. COMPUT.

c 2007 Society for Industrial and Applied Mathematics

NUMERICAL SIMULATION OF ACOUSTIC STREAMING ON

1. Introduction. The past decades have seen unprecedented advances in the

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS

Fig. 1.1. Example of a biochip.

A novel type of microuidic biochip employs surface waves as a driving force.

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS

Fig. 2.1. Working principles of a SAW-driven uidic device.

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

with the Newtonian stress tensor = (v, p) dened as

The divergence of the stress tensor yields

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS

v(t, x + u) = v(t, x) + (v)u + O(u ) =

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

the solution of (2.4) as given data. The result is

This system can be interpreted as an instationary Stokes problem describing

We thereby obtain the system

(2) = (v 1 )u

(2) , p(2) , (2) and

v (2) = (v 1 )u

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS

interface as well as the stress acting due to capillary forces.

In our analysis we will assume v = p .

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

as the dual space of U.

This yields a Gelfand triple

(f1 (t), f2 (t))U dt +

(f1 (t), f2 (t))U dt

for all f1 , f2 W (0, T ).

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS

The inequality (LBB) is the well-known Ladyzhenskaya-Babuska-Brezzi condition.

Av, wV V = a(v, w),

w, u (t)U U + s(w, u(t)) = w, f (t)U U

for all w U, f.a.a.1 t (0, T ),

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

Lemma 3.4. The problem (P) can be equivalently formulated as

a(w, v) + b(p, w) = w, f v V V

We immediately verify that a solution is given by v(x, t) = 0, p(x, t) = t, which does

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS

Proof. [50, Theorem 27.2].

w, u (t)U U + s(w, u(t)) = (w, f (t))H

for all w U, a.e. in (0, T ),

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

A solution u of (Pper ) can then be uniquely extended to a periodic solution u

for all w U, f.a.a. t R.

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

the stress tensor,

the normal velocity of f (t),

as the tangential derivative of f ,

as the tangential divergence of f , and

Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

ACOUSTIC STREAMING ON SAW-DRIVEN BIOCHIPS

v(t, x + u) = v(t, x) + (v)u + O(u ) =

(2) = (v 1 )u

v (2) = (v 1 )u

Av, wV V = a(v, w),

w, u (t)U U + s(w, u(t)) = w, f (t)U U

a(w, v) + b(p, w) = w, f v V V

w, u (t)U U + s(w, u(t)) = (w, f (t))H