Professional Documents
Culture Documents
Knu Visco
Knu Visco
net/publication/226124671
CITATIONS READS
58 1,082
4 authors, including:
Eugene D Skouras
University of Peloponnese
124 PUBLICATIONS 628 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Eugene D Skouras on 01 June 2014.
RESEARCH PAPER
123
Microfluid Nanofluid
particle-based technique, which has proven quite valuable proportional to k. An equally simple kinetic theory result,
method for the simulation of gas flow under rarefied con- proposed by Maxwell, for the bulk viscosity is (Pollard and
ditions (Bird 1998; LeBeau and Lumpkin 2001). In the Present 1948)
DSMC technique, the description of the system is given 1
through the definition of the positions and velocities of the l0 ¼ qck: ð4Þ
3
gas molecules through a sample of representative particles.
The system is evolving in discrete time increments during Alexander et al. (1998) and Hadjiconstantinou (2000)
which the particle translation step and the inter-particle examined the effects of the discretization error in DSMC
collision step are decoupled. The employment of only a calculations of the viscosity using the Green–Kubo theory.
small set of representative particles in the simulation and The resulting expressions are
the stochastic treatment of collisions are the key features of rffiffiffiffiffiffiffiffiffi !
5 mkT 16 L2y
the method and offer computational efficiency in compar- l¼ 1þ ð5Þ
ison to the more detailed molecular dynamics technique. It 16d2 p 45p k2
has been shown that the limiting behavior of DSMC, when
following a correction from Lilley and Sader (2009), and
the number of particles tends to infinity and the time step
rffiffiffiffiffiffiffiffiffi" #
and cell size become vanishingly small, coincides with the 5 mkT 32 ðcm DtÞ2
solution of the Boltzmann equation (Wagner 1992). l¼ 1þ ; ð6Þ
16d2 p 150p k2
The objective of the current study is to investigate the
dependence of the dynamic viscosity of a gas in the tran- where d is the collision diameter, Ly is the width of the cell in
sition regime on the Knudsen number using DSMC cal- the direction normal to the flow direction, Dt is the time step
culations and provide an analytical expression for such and cm is the most probable speed of the gas molecules.
dependence. This expression would be extremely useful in Lilley and Sader (2009) investigated gas flow in the
phenomenological descriptions and numerical computa- Knudsen layer for small Kn values and suggested a power-
tions of rarefied flows in narrow channels or pores. Care law dependence of viscosity on the dimensionless distance
must be taken to avoid confusion between flow systems from the solid surface ye; which has the form
that involve dilute gases and systems involving gases
within tight passages. In the former case, the molecule l0 ye1d
le ðe
yÞ ¼ ð7Þ
collisions are rare in the absence of walls whereas in the Cd
latter, the molecule–molecule collision frequency is where l0 is the bulk dynamic viscosity, and d and C are
reduced in the presence of walls. Although numerous parameters that are obtained by linear regression of the
efforts have been made to provide expressions for the gas results of the linearized Boltzmann equation for Kramer’s
viscosity as a function of density, the same is not true for problem and verified by the authors with DSMC for
the dependence of the viscosity on the channel dimensions Kn = 0.06.
and, eventually, on the Knudsen number. Beskok and Karniadakis (1999) suggested that a Bo-
The Chapman–Enskog theory of dilute gases (Chapman sanquet-type of expression for the viscosity in the transi-
and Cowling 1970) provides for the hard sphere model the tion regime and conducted numerical computations of flow
following expression for the bulk viscosity in cylinders and channels using the Navier–Stokes formu-
rffiffiffiffiffiffiffiffiffiffiffi lation complemented with a slip boundary condition. They
5 2pkT
l0 ¼ qk; ð1Þ found good agreement of their model with DSMC results
16 m
and with the linearized Boltzmann solution. Supportive
where k is the Boltzmann constant, T is the absolute arguments on the validity of an expression for the effective
temperature, m is the molecular mass, and q is the mass viscosity l0 in the form
density. Considering that the mean thermal speed is given by 1
rffiffiffiffiffiffiffiffi le ¼ l0 ð8Þ
8kT 1 þ aKn
c¼ ð2Þ
pm were presented in that study, where l0 is the bulk viscosity.
Eq. 1 can also be written as They suggested a Knudsen dependence of the rarefaction
5p parameter a and provided an analytical expression for this
l0 ¼ qck dependence based on the free molecular limit using two
32 ð3Þ
free parameters.
ffi 0:49qck
Guo et al. (2007, 2008) proposed a model in the Navier–
This, for a specific gas and fixed temperature, implies a Stokes formulation also, but with a ‘‘geometry’’-dependent
constant value for the viscosity given that q is inversely viscosity. In addition, they extended this concept into the
123
Microfluid Nanofluid
multiple-relaxation-time lattice Boltzmann technique using with significantly less statistical error, the inclusion of
arguments from Stops (1970) and his formula for the phenomenological models for the collision information
effective mean free path length in a channel in the transi- limits its capability of predicting the correct viscosity
tion regime. Using as argument the analogy between the coefficient value in the 0.1 \ Kn \ 0.5 region, and, thus, is
mean free path and the dynamic viscosity, an expression not expected to be suitable for the even wider range of
for the effective viscosity as a function of the distance from Knudsen numbers studied in the present study.
the wall was proposed in Guo et al. (2007), which in the Literature shows that DSMC is capable of reproducing
case of flow between two parallel plates at distance H is rarefaction effects by comparison to the solution of the
given by the equation linearized Boltzmann equation (Bird 1994). In this study,
1 y Hy DSMC is used with the goal of investigating the Knudsen
le ðyÞ ¼ l0 w þw : ð9Þ number effect on gas viscosity without resorting to any
2 k k
adjustment process or phenomenological description of
In Eq. 9, y is the distance from the wall and the function flow. In fact, the effective viscosity is obtained directly
w is given by from the flow field calculations and its shear stress-based
wðxÞ ¼ 1 þ ðx 1Þex x2 Ei ðxÞ ð10Þ definition. This is an intriguing goal on its own given the
dearth of similar attempts despite the obvious usefulness of
with Ei(x) defined as the relationship between viscosity and Knudsen number not
Z1 only in vacuum applications and MEMS but also in porous
Ei ðxÞ ¼ t1 ext dt: ð11Þ media science and technology. Once such an effective
1
viscosity is available, it can be used in phenomenological
models of flow under rarefaction conditions without having
The Information Preservation (IP) scheme (Fan and to resort to tedious simulations at the microscopic scale.
Shen 1999, 2001) was specifically developed as a
mesoscopic technique to overcome the statistical noise of
direct methods, such as the DSMC, and was applied by 2 Effective viscosity using DSMC
Roohi and Darbandi (2001) to the extension of the Navier–
Stokes slip-flow description into the early transition The case of pressure driven flow of nitrogen in a narrow slit
regime. Through the use of the IP results, a modified is considered in the present study under isothermal, low
viscosity coefficient expression was suggested and speed conditions. The hard sphere model is used as it is
introduced in the macroscopic NS description to predict adequate for the isothermal case examined here. The
both the mass flow rate and the velocity profile in 2D molecular diameter is 4.17 9 10-10 m, whereas the
micro- and nanochannel flow processes. However, the molecular mass is 4.65 9 10-26 kg. In most of the calcu-
determining factor for the accuracy of shear stress lations shown here, the inlet-to-outlet pressure ratio is
calculations in the IP scheme is the collision model that Pr = 2 and the length-to-width ratio L/H is varied between
is employed. The IP collision model of Fan and coworkers 15 and 30. Larger L/H ratios could also be investigated but
(1999, 2001) assumes that the preserved information of they would require very large computational times. The
particles is the same after collision. However, the distance H between the two plates is considered to be the
numerical tests have shown that this basic model cannot characteristic length in the definition of the Knudsen
correctly simulate the viscosity and thermal conductivity number, which was varied by varying the applied pressure
of gas flows. Sun and Boyd (2002) suggested a and, hence, the mean free path. Fully diffusive thermal
phenomenological model for the distribution of the walls were assumed. Care was taken, so that the usual
information after collision based on the collision approximations involved in the DSMC method were sat-
deflection angle. Results (Roohi and Darbandi 2001) isfied. Namely, in all cases, the mean free path to cell size
have shown that the slip velocity is overestimated, ratio was k/Dxcell [ 3, the time step was kept smaller than
whereas the maximum velocity is underestimated, for the mean collision time, and sufficient number of mole-
both of the current IP collision models when Kn [ 0.5, cules per cell ([20) were used. Calculations reported here
following comparison with the DSMC predictions. This were performed on a 200 9 12 grid, with 2 9 2 subcells in
has also been shown by Cai et al. (2000) for the velocity each cell, on a sample 10-5 m long (x) and 6 9 10-7 m
prediction of the collision IP model suggested in Fan and wide (y). Finer grids (200 9 32 and 400 9 22) have been
Shen (1999, 2001), while Roohi and Darbandi (2001) also used to ensure convergence, however without practical
demonstrate this discrepancy for the collision model of difference (\5% change in the mean value of Kn at
Sun and Boyd (2002). Although the IP solution predicts Kn = 0.1). Calculations were terminated and results were
mass flow rate and shear stress close to that of DSMC collected when the mean Knudsen number of all cells was
123
Microfluid Nanofluid
boundary conditions are similar to those used in Wu and 0,000018 Symbols: DSMC calc.
-void: Poiseuille type of flow
Tseng (2001). It should be noted that under these condi- 0,000016 -filled:Couette flow, uwall=1 m/s
123
Microfluid Nanofluid
Bosanquet parameter, a
formance in the lower part of the transition regime is not
encouraging for higher Kn values.
In order to simplify further the utilization of this result 2
in phenomenological approaches and computations in
MEMS or porous media, it is tempting to investigate
whether a Bosanquet-type of approximation, which is 1
known to hold with good accuracy (Pollard and Present
1948) for the diffusion coefficient in the transition regime,
reproduces satisfactorily the viscosity dependence on the 0
1 10
Knudsen number. Following the brief review of bulk vis-
Knudsen number
cosity expressions that was made in Sect. 1, the viscosity in
the absence of walls (Kn ? 0) can, in general, be Fig. 2 Variation of factor a with the Knudsen number as results from
expressed as DSMC calculations in a channel. Each set represents separate
simulation results in the channel with different end pressure values
l0 ¼ a0 qck ð15Þ
where a0 is a numerical constant, whereas at the free
A similar value (a = 2.2) was suggested by Beskok and
molecular flow limit (Kn ? ?)
Karniadakis (1999) in their effort to match their numerical
l1 ¼ a1 qcH ð16Þ results for the mass flow rate in a channel with the corre-
sponding DSMC results. It must be stressed though that the
with a? a numerical constant for infinitely large k. By
factor a is not an adjustable parameter in the present study.
analogy to the diffusion case, the Bosanquet-type of
Its value was determined from a post-process calculation
approximation would read
following DSMC simulation of flow in a channel and using
1 1 1 directly the definition of the viscosity. It also noteworthy
¼ þ : ð17Þ
le l0 l1 that the calculated viscosity approaches its bulk value for
Substituting expressions (15) and (16) in Eq. 17, one Kn \ 10-3, as expected, whereas for Kn [ 10, it takes a
gets value that is very close to the prediction of Eq. 16 with
a? = 1/3. It is, also, useful to note that the same value of a
1
le ¼ l0 ð18Þ was obtained for all length-to-width ratios examined in the
1 þ aKn present study (L/H = 15–30). Smaller L/H values would
which was also suggested in Beskok and Karniadakis lead to strong end effects whereas larger values would
(1999). Note that the factor a = a0/a? is employed in this require very large computational times.
expression to absorb the unknown numerical constants a0 Figure 3 shows the effective viscosity, non-dimension-
and a? into a single quantity. Using l0 = 1.778 9 10-5 alized with its bulk value, as a function of the dimen-
Pa s and the values of le that were calculated in the sim- sionless distance from the wall, y/H. Comparison is made
ulations of Fig. 1, one can calculate the value of a from between the DSMC results using l0 = 1.778 9 10-5 and
Eq. 18 for each Kn value. Figure 2 shows that the value of the predictions of Eq. 9 for Kn = 0.128. Although the
a is not constant but depends on the Kn value, as also noted presence of statistical noise is apparent in the DSMC
in Beskok and Karniadakis (1999). However, this depen- results, a relatively good agreement is noted with the
dence is rather weak over the majority of the transition effective viscosity values that were proposed by Guo et al.
regime, suggesting an ‘‘effective’’ value close to 2. In fact, (2007). Taking into account that the two approaches are
if one uses this value in the Bosanquet expression (18), the entirely different from each other, this agreement is quite
Kn-dependence of the viscosity follows the solid line of encouraging as it contributes significantly to the quantifi-
Fig. 1, which is seen to reproduce quite satisfactorily the cation of the viscosity variation within the Knudsen layer.
DSMC results. Using Eq. 5 results in l0 = 1.41 9 10-5 An investigation of the effect of the molecule-wall
Pa s, which, in turn, leads to a value of a = 1.5. This collision law on the validity of the Bosanquet approxima-
option provides also good approximations to the calculated tion to the effective viscosity dependence on the Knudsen
viscosity except for the lower range of Kn values, as shown number is discussed next. Specifically, the simulation of
in Fig. 1. rarefied flow in the tube was repeated using different values
123
Microfluid Nanofluid
1,2 DSMC results quite satisfactorily the Kn-dependence of the viscosity over
1,1
Eq. (9) the entire range of Kn values investigated here (0.1–10),
covering, practically, the entire transition regime. The Bo-
1,0
sanquet parameter was calculated to equal 2, which is quite
0,9
similar to the value suggested independently in a previous
0,8 study (Beskok and Karniadakis 1999) that treated it as an
0,7 adjustable parameter with the goal to optimize matching
0,6 between the predictions of independent flow calculations
and those of DSMC calculations.
0,5
The simplified description of the dependence of gas
0,4
viscosity on the Knudsen number through a Bosanquet-
0,3 type of approximation with a specific parameter value
0,0 0,1 0,2 0,3
lends itself to a broad utilization in rarefied flow calcula-
Dimensionless distance from the wall, y/H
tions in MEMS and in porous media. Further study is
Fig. 3 Dependence of effective viscosity, rendered dimensionless needed though to estimate potential effects of the channel
using the bulk value as reference, on the distance from the wall. aspect ratio on the value of the Bosanquet parameter for a
Comparison between the predictions of Eq. 9 and DSMC results using more complete quantification of the Kn-dependence of the
l0 = 1.778 9 10-5 Pa s for Kn = 0.128
gas viscosity in the transition regime.
123
Microfluid Nanofluid
Fan J, Shen C (2001) Statistical simulation of low-speed rarefied gas Pollard WG, Present RD (1948) On gaseous self-diffusion in long
flows. J Comput Phys 167:393–412 capillary tubes. Phys Rev 73:762–774
Guo ZL, Shi BC, Zheng CG (2007) An extended Navier–Stokes Roohi E, Darbandi M (2009) Extending the Navier–Stokes solutions
formulation for gas flows in the Knudsen layer near a wall. to transition regime in two-dimensional micro- and nanochannel
Europhys Lett 80:24001–24006 flows using information preservation scheme. Phys Fluids
Guo ZL, Zheng CG, Shi BC (2008) Lattice Boltzmann equation with 21:082001
multiple effective relaxation times for gaseous microscale flow. Stops DW (1970) The mean free path of gas molecules in the
Phys Rev E 77:036707 transition regime. J Phys D 3:685–696
Hadjiconstantinou NG (2000) Analysis of discretization in the direct Sun Q, Boyd ID (2002) A direct simulation method for subsonic
simulation Monte Carlo. Phys Fluids 12:2634–2638 microscale gas flows. J Comput Phys 179:400–425
Karniadakis G, Beskok A, Aluru N (2005) Microflows and nanoflows: Wagner W (1992) A convergence proof for Bird’s direct simulation
fundamentals and simulation. Springer, New York Monte Carlo method for the Boltzmann equation. J Stat Phys
LeBeau GJ, Lumpkin FE III (2001) Application highlights of the 66:1011–1044
DSMC analysis code (DAC) software for simulating rarefied Wu JS, Tseng KC (2001) Analysis of micro-scale gas flows with
flows. Comput Methods Appl Mech Eng 191:595–609 pressure boundaries using direct simulation Monte Carlo.
Lilley CR, Sader JE (2009) Velocity profile in the Knudsen layer Comput Fluids 30:711–735
according to the Boltzmann equation. Proc R Soc A 464:
2015–2035
123