Professional Documents
Culture Documents
Abstract
Supersonically expanding stationary thermal plasma, formed by a thermal
cascaded arc is studied. Due to the low chamber pressure (20–100 Pa) the
results of continuum mechanics model can be doubtful. This is why these
results are validated against kinetic Monte Carlo simulation and
experimental data obtained by means of laser induced fluorescence. The
analysis proves that continuum mechanics is still applicable for the velocity
and temperature field predictions downstream of the shock region. However,
the shock formation and some non-equilibrium effects typical for supersonic
flow can be correctly studied only with the help of kinetic simulations. We
show that the errors in the results using continuum mechanics can be
attributed to the presence of flow gradients. These errors diminish when the
shock regions are thickened due to rarefaction, viscosity and heat
conductivity. Besides, both methods show that the effect of the chamber
geometry on the plasma flow field is important.
observed. This breakdown results in the deviation from the the application of the method to describe plasma flows is not
Maxwellian distribution characterized by a single translational straightforward. The standard approach to add particle-in–
temperature [10]. cell procedures to DSMC algorithms fails at high ionization
The description of gas motion using CFD approach degree [24]. At the same time quite simple multi-temperature
implies solving the system of Navier–Stokes equations continuum model can be easily applied for the description of
[19]. These equations represent the laws of conservation non-equilibrium plasma flows [28].
of mass, momentum and energy with addition of the A hybrid numerical technique that can switch from CFD
constitutive relationships between the pressure tensor and to DSMC techniques when it is physically needed would be a
velocity gradients and between the heat flux and temperature perfect solution to the problems mentioned, but to develop this
gradients. Using the appropriate equation of state and transport technique one needs some good criteria of CFD failure.
coefficients, we can apply this system of equations to describe The facts mentioned above have motivated us to
the whole supersonic flow. However, the reliability of the investigate the applicability of the continuum mechanics
results of such calculations can be doubtful for several reasons. approach to study supersonic cascaded arc plasma expansions.
First, the linear relationships in the constitutive equations are This kind of plasma is very well studied experimentally [5–9],
inadequate for strong gradients that are present in the expansion while a model for the whole two-dimensional axi-symmetric
and compression waves. Even though in a plasma these supersonic–subsonic gas flow has not been presented yet.
gradients can be less pronounced than in a cold gas due to Previously published results contain only one-dimensional
the increased role of thermal conductivity and viscosity at continuum models for supersonic expansions [5, 13, 14],
high temperatures [20], the applicability of CFD still needs or two-dimensional models, where the calculations were
validation. Second, the continuum approach assumes that the performed with the help of PLASIMO [16] and PHOENICS
molecular velocity distribution is close to Maxwellian. This [17] codes for only the subsonic part of the expansion.
assumption certainly results in some errors of CFD simulations In this paper, we report the results of supersonic
of the flows containing a strong expansion and shock waves. cascaded arc plasma expansion modelling and compare two
Finally, in terms of CFD applicability, the rarefaction effect computational approaches: the CFD and DSMC methods. For
on the supersonic jet flow is not so obvious [21]. On the one CFD calculations we utilize the commercially available code
hand, this effect results in the deviation from a Maxwellian FLUENT (FLUENT is a registered trademark of FLUENT
distribution, but on the other hand, the rarefaction broadens Inc. Ceterra Resource Park, 10 Cavendish Court, Lebanon,
the shock waves and diminishes the gradients. To sum up, NH 03766 USA) [19]. For the DSMC method we apply a
the applicability of the continuum mechanics approach to standard procedure developed by Bird [18]. We show that
supersonic plasma modelling should be analysed and validated both of these approaches can give two-dimensional solutions
by the comparison with experimental data and with the results for the whole region of supersonic cascaded arc expansions.
of kinetic models. Another solution could be the use of the In the literature, the DSMC method is often applied to
improved CFD approach, which is the Burnett formulation simulate the gas flows when experimental validation is a rather
[22]. Unfortunately, this formulation is too complicated to be difficult task. As an example of this situation one can mention
applied in most of the practical problems. non-stationary plasma expansion encountered in pulsed laser
The recent development of kinetic gas models such as deposition [29]. From the physical point of view, stationary
direct simulation Monte Carlo (DSMC) [18] and molecular plasma expansion is a similar problem [30], but it facilitates the
dynamics (MD) [23] methods has progressed quickly due to experimental validation. This is why such flows can be used to
the improvement of computer hardware capabilities. These check the feasibility of the method to simulate the expanding
methods have been tested on shock wave problems and gas flow in general. With this in mind, here we validate
both of them have demonstrated excellent agreement with the modelling results on the stationary gas temperature and
experiments. This observation can be explained by the fact velocity profiles against the experimental data obtained with
that both methods simulate the solution of the Boltzmann the help of a laser-induced fluorescence technique described
equation and are based on the microscopic representation of in detail elsewhere [10–12].
a gas. The difference between the two methods lies in the In this paper, we restrict ourselves to the case where the
calculations of the dynamics of the system. In DSMC methods, static pressure in the plasma formation source, i.e. Pin =
unlike the MD method, the trajectories of particles are not 0.2 atm is 200–1000 times higher than the static pressure
computed exactly. This method uses instead a discretization in the expanding chamber, i.e. Pch = 100–20 Pa. It was
of time and space and a description of the evolution of an proved that the character of the very first region of the
ensemble of particles according to the stochastic rules. The supersonic expansion is similar to the adiabatic expansion
DSMC method has become a widespread modelling tool to of a gas from a high-pressure reservoir through an orifice
describe gases in the transition regime between the continuum into vacuum [7, 10–12]. The expansion into a low-pressure
and the free-molecular flow [18, 23–27]. However the time (not vacuum) atmosphere, however, consists of the region of
required for DSMC calculations is inversely proportional to adiabatic expansion followed by a stationary normal shock
the Knudsen number. Because of this fact, the method can wave and a subsonic relaxation zone (figure 1). We can
be ineffective to describe the full nozzle-jet flow. Besides, conclude that the supersonic plasma expansion into low-
unlike the continuum mechanics approach, there have been pressure atmosphere represents a combination of three sorts
only a few attempts to apply the DSMC method to supersonic of flows: a plasma expansion into vacuum [31–34], a shock
flows taking into account the plasma effects. Due to the large wave formation [20, 35, 36] and a subsonic plasma jet [37–41].
discrepancy between the plasma and the collision length scales, In our simulations, however, we do not separate the regions
1363
S E Selezneva et al
plasma with a gas temperature T ≈ 1 eV. The pressure in , F and G are defined as
where the vectors W
the arc channel is slightly sub-atmospheric (∼0.1–0.5 atm).
The total power given to the arc varies from 2 to 5 kW. The
ρ
ρ V
0
plasma expands supersonically into the low pressure (∼20 Pa)
ρu ρ V u + P î
τxi
chamber. The last plate, where the nozzle is placed, acts ˆ
W = ρυ , F = ρ V υ + P j , G = τyi
as an anode. The nozzle has a straight channel with a
ρw
ρ V w + P k̂
τzi
diameter of 6 mm and a length of 10 mm. The last part of
ρE ρV E + P V τ V
ij j + q
the nozzle (5 mm) makes an angle of 45˚ with the channel.
The deposition chamber is a cylindrical stainless steel vessel
with an inner diameter of 32 cm. During the deposition, and S vector can contain source terms such as body forces
an admixture of molecular gas (e.g. silane or acetylene) is and energy sources. Here ρ, V , E and P are the density,
1364
Stationary supersonic plasma expansion
velocity, total energy per unit mass, and static pressure of The computational domain used for the modelling is
the fluid, respectively; u, υ, ω and x, y, z are the velocity demonstrated in figure 3. This domain consists of two
components and the coordinates in the directions of î, jˆ, k̂ cylinders. The first cylinder represents a part of the arc flow
coordinate vectors. τ is the viscous stress tensor and q is the channel with the diameter of 4 mm and the length of 10 mm.
heat flux. Total energy E is related to the total enthalpy H , or The second cylinder represents the expansion chamber. The
to the static enthalpy h by the following expressions: radius of the chamber is 162.25 mm, its length is 800 mm. We
neglect the presence of the divergent part of the nozzle. This
p |V |2 simplification is due to the fact that in reality the nozzle walls
E=H− , where H =h+ .
ρ 2 work as anode and the plasma dissipating some Ohmic power
is not quenched much in the divergent part. In the model, we
Perfect gas law state equation is added to the system. do not take into account the Ohmic heating, hence the plasma
We study the axi-symmetric flow. The considered flow quenching can be unrealistically high if we take into account
is mostly supersonic and the coupling between momentum the presence of the divergent part of the nozzle. However, note
and energy equations of a Navier–Stokes system is very that this simplification can cause some errors in the results of
strong. This is why we choose a coupled technique solving both kinetic and continuum models.
simultaneously the equations written in matrix form. The The boundary conditions are as follows: the temperature is
Navier–Stokes equations become numerically very stiff at 12 000 K at the inlet. At the wall we use a constant condition
low Mach number due to the disparity between the fluid for the temperature (300 K) and non-slip velocity boundary
velocity and the speed of sound. The numerical stiffness condition. In this paper we do not study in detail the near-wall
of the system of equations under these conditions results flow, and so we do not set at the walls physically more correct
in poor convergence rates. This difficulty is overcome in conditions of temperature jump and velocity slip. The param-
FLUENT’s coupled solver by employing a technique called eters defining the supersonic jet structure are the nozzle Mach
time-derivative preconditioning [19]. A k–ε model [42] number M and the ratio of the static pressure at the chamber
modified by renormalization group method (RNG) is applied inlet Pin to the ambient chamber pressure Pch . For the jets
for turbulence simulation because it predicts the properties exhausting into the low pressure chamber from a straight noz-
of axisymmetric jet better than a usual k–ε model. For zle the condition M ≈1 is valid. During the experiments the
thermodynamic and transport properties of hot argon we static pressures of the jets were between 0.1 and 0.2 atm, while
use simple expressions valid for gas temperatures T the pressure in the chamber was 20 or 100 Pa. All experiments
12 000 K. The power-law temperature dependence is used for were performed at a flow rate through the arc of 3 standard
the viscosity, µ, approximation: litres per minute (SLM). At the inlet boundary we set the mass
0.72 flow rate corresponding to the given volume flow rate and we
T
µ = µ0 , (1) assume that the temperature and velocity profiles are radially
T0 uniform. This approximation is possible, because we do not
where the reference temperature T0 = 273.11 K, the focus our attention on the flow inside the arc channel. We uti-
corresponding reference viscosity of argon is µ0 = 2.125 × lize a part of this channel to obtain more realistic profiles at the
10−5 kg m−1 s−1 . expansion chamber inlet. The exact value of turbulence inten-
The thermal conductivity λ is calculated as follows: sity is unknown at the arc inlet boundary. Despite the high flow
velocity in the arc channel (1500 m s−1 ) the Reynolds number
µCp is moderate due to the high temperature and small radius of the
λ= , (2)
Pr channel. This is why the intensity of turbulence is supposed
to be 1% at the inlet boundary. The pumping exit, which is in
where Pr = 0.7, the Prandtl number and Cp is the heat reality a circular hole, in two-dimensional model can be repre-
capacity. The heat capacity is modelled as a piecewise linear sented as a slit with a width of 20 mm at a distance of 600 mm
function of static temperature using the tables in [43]. We
repeated the calculations using more accurate expressions for
viscosity and thermal conductivity obtained from the kinetic
theory [44, 19]. The difference in the temperature profiles
never exceeded 2% everywhere, however, the computational
time increased. This is why we utilize the expressions (1) and
(2) in all the calculations presented in this article.
In our models we assume that the gas is composed only of
argon atoms, neglecting the presence of electrons and ions. We
also neglect the effect of ionization and recombination on the
flow field. Because of the fact that the plasma is quenched in
the expansion, the ionization is practically absent in this flow.
The recombination however can be significant. We studied
the effect of recombination by repeating the calculations
using a two-temperature model described in [28]. It was
found that recombination affects significantly only the electron
temperature field, not the gas temperature and velocity fields. Figure 3. Contours of the translational temperature predicted by
The detailed results of this study will be reported later. DSMC for 20 and 100 Pa chamber pressures.
1365
S E Selezneva et al
from the nozzle (figure 3). The substrate is situated 350 mm performed for two chamber pressures: Pch = 20 and 100 Pa.
from the chamber inlet. The radius of the substrate is 50 mm. Because of the fact that the uncertainty of most experimental
The computations start on a rather coarse non-structural measurements for such flows is ∼10%, we will consider the
grid using the first-order upwind scheme. The computational agreement between the results of the two models to be good if
grid contains 231 quadrilateral cells in the first cylinder and the difference does not exceed 10%. Both models show that
1720 non-structural cells in the second cylinder. The cell size because of the low chamber pressures compared to the pressure
depends on the position in the chamber: this size is small in the discharge tube, the plasma strongly expands. This strong
in the supersonic part of the expansion and within the high- expansion leads to the formation of the jet structure with a
velocity jet region. The cell size is increased far from the supersonic barrel (figures 3, 4 and 5), ended by a stationary
axis. After performing several hundreds of iterations, the grid normal shock wave (Mach disk), which is followed by a
is adapted (120 child cells are added in the supersonic region) subsonic mixing and relaxing region. Figure 5 shows that in the
and the calculations proceed using the second-order scheme expansion, the axial Mach number rises till the value of ∼3.5
until the residuals of all the equations become less than 10−4 . if Pch = 100 Pa and till the value of ∼4.5 if Pch = 20 Pa. Then
The computations were repeated for the grid containing twice the Mach number drops in the compression region. Figure 6
as much cells in the expansion chamber. The difference in the demonstrates that in the expansion, the static pressure at the
results was never more than 5%. axis drops from 0.15 atm till the chamber pressure values. This
static pressure stays almost constant in the subsonic region.
As it is known [45], in the continuum regime the axial shock
3.2. Aspects of kinetic simulations position, xc can be evaluated by the expression
In a direct simulation Monte Carlo method [18], a gas is
composed by a number of model particles, each of them xc Pstagn
representing the average movement of much bigger number ≈ 0.67 ,
d Pch
of real particles (atoms, ions, electrons). The following two
steps model the evolution of a system:
(i) collisions of particles (elastic or inelastic),
(ii) translational movement.
Each of the model particles has coordinates in physical space,
three velocity components, and internal energy. The physical
space is discretized so that the computational grid groups
together the particles that are likely to collide. The kinetic
theory of gases is applied to calculate the probability of the
collision selection. Collision probability is calculated based
on the collision frequency and pairs are selected based on the
product of relative velocity and collision cross section.
Only elastic collisions are considered in this paper; the
gas consists only of argon atoms. Variable hard sphere (VHS)
collision model [18] is applied to simulate the collisions.
The modelling domain is the same as described in the
previous section. The quadrilateral grid containing 80 cells
in x-direction and 60 cells in y-direction is used. In both
Figure 4. Contours of the translational temperature predicted by
directions the cells contain two sub-cells. Near the inlet the FLUENT for 20 and 100 Pa chamber pressures.
cell size should be very small to capture the phenomena when
the Knudsen number is small. In this paper, however, our
purpose is to predict correctly the flow downstream of the
nozzle. The cell spacing is made progressively larger in the
direction of expansion. The geometric progression coefficients
are as follows: seven in x-direction and four in y-direction.
The number of model molecules is 106 . The time step is
made less than the residence time of the particle in the smallest
cell. At the walls we set the condition of diffusive reflection.
At the nozzle inlet plane the flow conditions are the axial
velocity 2000 m s−1 , the number density 1023 m−3 and the gas
temperature 12 000 K.
In figures 3 and 4 the results of Monte Carlo simulations Figure 5. Local Mach number axial profile predicted by FLUENT
on the gas temperature field are compared with the results for chamber pressures of 20 and 100 Pa. Here and in figures 5, 6 and
of continuum mechanics. The model calculations are 9 the coordinate x = 0 corresponds to the nozzle outlet.
1366
Stationary supersonic plasma expansion
1367
S E Selezneva et al
Figure 7. Comparison of various axial profiles. (a) Translational temperature T predicted by FLUENT, DSMC and experimentally
measured [10] for the Pch = 20 Pa. (b) Predicted by DSMC translational temperature in the axial direction, Tx , and in normal to the axis
direction, Tn , and experimentally measured temperature [10] for Pch = 20 Pa. (c) Velocity U predicted by FLUENT, DSMC, and
experimentally measured velocity [10] for Pch = 20 Pa. (d) Translational temperature T predicted by FLUENT, DSMC and experimentally
measured velocity for Pch = 100 Pa. (e) Predicted by DSMC translational temperature in the axial direction, Tx , and in normal to the axis
direction, Tn , and experimentally measured temperature [10] for Pch = 100 Pa. (f ) Velocity U predicted by FLUENT, DSMC, and
experimentally measured velocity [10] for Pch = 100 Pa.
2
temperature. The overall translational temperature T in this component u in the x-direction, Tn = u) du is
v f (
flow can be defined as the normal or circumferential temperature based on the
T = 1
(Tx + 2Tn ) , microscopic velocity component v that is normal to the
3
x-direction, U is the macroscopic flow velocity, f (u) is
where Tx = u) du is the axial or parallel
(u − U ) f ( 2
the velocity distribution function. Unlike the continuum
translational temperature based on the microscopic velocity model, the DSMC method indicates the deviations from
1368
Stationary supersonic plasma expansion
translational equilibrium. The results of calculations presented of random movement parallel to the flow direction and then
in figure 7 show that along the axis in the expansion the axial to the one normal to the flow direction. At the same time,
temperature Tx does not drop so fast as the normal temperature in expanding flow the collision rate becomes so low that
Tn does. As a result, at the end of the expansion region, the thermal equilibrium between these energy modes cannot be
temperature Tx is larger than the temperature Tn . In the normal maintained. It was found [2] that the velocity distribution
shock (figure 7), the rise of Tx precedes the shock rise of Tn and function at the axis of the expansion is well described by the
in the shock region the overshoot of Tx over Tn takes place. The ellipsoidal distribution function
ratio Tx /Tn reaches its maximum at a distance of x ≈ 50 mm
1/2
from the nozzle when Pch = 20 Pa and at a distance of m mv 2 m mu2
u) = n
f ( exp − × exp − ,
x ≈ 25 mm when Pch = 100 Pa. Radial profiles of the 2π kTn 2kTn 2π kTx 2kTx
temperatures Tx , Tn and T demonstrated in figure 8 for the case
of the chamber pressure Pch = 20 Pa, show that translational where n is the number density, m is the mass of the particle,
non-equilibrium can be found within the whole supersonic k is the Bolzmann constant. Second, in the shock wave, due
region of the jet. Immediately after the shock, where the flow to the steep gradients the equilibrium cannot be established.
velocity is subsonic, the temperatures Tx and Tn become equal In fact, in the shock region, the velocity distribution may be
to each other. In the subsonic region, the velocity distribution strongly bimodal [47], i.e. this distribution may be close to a
equilibrates and becomes close to Maxwellian. The radial sum of two parts representing the velocity distributions ahead
profiles of temperature and velocity (figure 9) illustrate rather and behind the shock wave. Moreover, as mentioned earlier,
satisfactory agreement between DSMC and FLUENT models in the rarefied flow the normal shocks and barrel shocks are
for Pch = 20 Pa. thickened and their thickness is comparable with the mean free
There are several reasons for the observed breakdown of path. This fact leads to the possibility of the background gas
translational equilibrium in supersonic expansion. First, this invasion in the supersonic region, [8, 10, 48] which can also
phenomenon can be explained by the fact that in supersonic drive the flow from equilibrium.
jets, the kinetic energy of the macroscopic movement parallel However, the FLUENT model cannot take into account
to the flow direction transfers first to the microscopic energy the translational non-equilibrium. This is the second reason
Figure 8. Radial profiles of temperatures in the axial direction, Tx , and in normal to the axis direction, Tn , and total translational
temperature, T , predicted by DSMC for a chamber pressure of 20 Pa for the axial positions (a) x = 3 cm, (b) 5 cm, (c) 7 cm, (d) 20 cm.
1369
S E Selezneva et al
Figure 9. Radial profiles for a chamber pressure of 20 Pa for the axial positions x = 3, 5, 7 and 20 cm: (a) temperature (FLUENT),
(b) velocity (FLUENT), (c) temperature (DSMC) and (d) velocity (DSMC).
why this method predicts the shock strength and position with any discontinuities in the supersonic region, therefore this
a bigger error than the kinetic DSMC method. number can be recommended as a criterion of continuum
To find the criteria of CFD failure we should evaluate the failure.
Knudsen number. At the axis of supersonically expanding Downstream of the normal shock region, the jet flow
flow, we can define the local Knudsen number as follows [25]: velocity decreases and becomes subsonic. The subsonic flow
no longer contain gradients as strong as one can find in the
l ∂Q
Kn = , pre-shock region, and the local Knudsen numbers decrease
Q ∂x in this flow. Besides, as it can be seen from figure 11,
the flow does not expand much in the subsonic region.
where l is the local mean free path, Q is a flow property
(density or temperature). Figure 10 shows the axial profile of The facts mentioned lead to the equilibration of the velocity
Knudsen numbers Kn T and Kn D, calculated by FLUENT distributions. The axial velocity and temperature profiles are
using as a flow property Q the translational temperature and very well predicted by both models downstream of the shock
density, respectively. The Knudsen numbers defined in this region when Pch = 20 Pa. These profiles are in excellent
way reflect the fact that this number can be increased not only agreement with the measured data. When Pch = 100 Pa, due
due to the pressure decrease, i.e. an increase of the mean free to the larger FLUENT errors in the shock region, downstream
path l, but also due to the increase of the flow gradients. For of the shock the FLUENT performance is not as good as the
example, it can be expected that the maximal Knudsen number DSMC performance.
Pch = 20 Pa is five times as much that when Pch = 100 Pa. From the analysis of the stream function fields it is
FLUENT calculations show, however, that due to the larger obvious that the chamber geometry plays an important role
flow gradients in the case of Pch = 100 Pa compared with in flow pattern formation: there is a big recirculation zone
Pch = 20 Pa, the difference in the maximal Knudsen number that compresses the jet and prevents it from expanding in
between these flow regimes is reduced. The maximal Knudsen subsonic region. Both models prove that in the subsonic region
number Kn D in the case of chamber pressure Pch = 20 Pa is the plasma column is almost straight (figures 3 and 4). We
only twice as much as that when Pch = 100 Pa. The results note that the results of the present models for subsonic region
of our calculations support the conclusion that the breakdown are in good agreement with the flow pattern predicted by the
of continuum takes place when the Knudsen number Kn D previously reported simulations performed with the help of
exceeds 0.05. The Knudsen number Kn D does not contain PHOENICS [17].
1370
Stationary supersonic plasma expansion
1371
S E Selezneva et al
[4] Juchmann W, Luque L and Jeffries J B 2000 Appl. Opt. 39 [27] Boyd I D and Chen G 1995 Prediction failure of the continuum
3704 fluid equations in the transitional hypersonic flows Phys.
[5] Kroesen G M W 1988 PhD Thesis University of Technology, Fluids 7 210
Eindhoven, The Netherlands [28] Selezneva S E, Rajabian M, Gravelle D and Boulos M I 2001
[6] Kessels W M M 2000 PhD Thesis University of Technology, J. Phys. D 34 3862
Eindhoven, The Netherlands [29] Urbassek H M and Sibold D 1993 Phys. Rev. Lett. 70 1886
[7] van de Sanden M C M, de Regt J M and Schram, D C 1994 [30] Bulgakov A V and Bulgakova N M 1998 J. Phys. D: Appl.
Plasma Sources Sci. Technol. 3 501 Phys. 31 693–703
[8] van de Sanden M C M, van den Bercken R and Schram D C [31] Ashkenas H and Sherman F S 1966 Rarefied Gasdynamics
1994 Plasma Sources Sci. Technol. 3 511 vol 4 (New York: Academic)
[9] Meulenbroeks R F G, Engeln R A H, Beurskens M N A, [32] Ashkenas H and Sherman F S 1966 The structure and
Paffen R M J, van de Sanden M C M, van der Mullen J A M utilisation of free jets in low density wind tunnel: Int. Symp.
and Schram D C 1995 Plasma Sources Sci. Technol. 4 74 on Rarefied Gas Dynamics vol 2 (suppl 3) p 84
[10] Engeln R, Mazouffre S, Vankan P, Schram D C and Sadeghi N [33] Chou Y S and Talbot L 1967 AIAA J. 5 2166
2001 Plasma Sources Sci. Technol. 10 595 [34] Fraser R B, Robben F and Talbot L 1971 Phys. Fluids 14 2317
[11] Mazouffre S, Boogaarts M G H, Bakker I S J, Vankan P, [35] Shanmugasundaram V and Murty S S R 1978 J. Plasma Phys.
Engeln R and Schram D C 2001 Phys. Rev. E 64 2 16411 20 419
[12] Mazouffre S, Boogaarts M G H, van der Mullen J A M and [36] Liberman M A and Velikhovich A L 1985 Physics of Shock
Schram D C 2000 Phys. Rev. Lett. 84 2622 Waves in Gases and Plasmas, Springer Series in
[13] Beulens J J, Kroesen G M W, Vallinga P M and Schram D C Electrodynamics vol 19 (Berlin: Springer)
1989 Proc. Int. Symp. on Plasma Chemistry (ISPC-9) [37] Chang C H and Pfender E 1990 Plasma Chem. Plasma
(Pugnochiuso, Italy) Process. 10 473
[14] Beulens J J, Milojevic D, Schram D C and Vallinga P M 1991 [38] Chang C H and Pfender E 1990 Plasma Chem. Plasma
Phys. Fluids B 3 2548 Process. 10 493
[15] Burm K T A L, Goedheer W J and Schram D C 1999 Phys. [39] Chang C H and Ramshaw J D 1994 Phys. Plasmas 1 3698
Plasmas 6 2622 [40] Diliwari A H, Szekely J and Westhoff R 1990 Plasma Chem.
[16] Janssen G M 2000 PhD Thesis University of Technology, Plasma Process. 10 501
Eindhoven, The Netherlands [41] McKelliget J, Szekely J, Vardelle M and Fauchais P 1982
[17] Kessels W M M, Leroux A, Bogaarts M G H, Plasma Chem. Plasma Process. 2 317
Hoefnalels J P M, Van de Sanden M C M and Schram D C [42] Choudhury D 1993 Introduction to the renormalization group
2001 J. Vac. Sci. Technol. A 19 467 method and turbulence modeling Fluent Inc. Technical
[18] Bird G A 1994 Molecular Gas Dynamics and the Direct Memorandum TM-107
Simulation of Gas Flows (Oxford: Clarendon) [43] Boulos M I, Fauchais P and Pfender E 1994 Thermal Plasma
[19] FLUENT 5 Users Guide, 1995 Fluent Incorporated Fundamentals and Applications (New York: Plenum) vol 1
[20] Zeldovich Ya B and Raizer Yu P 1967 Physics of Shock Waves [44] Hirschfelder J O, Curtiss C F and Bird R B 1964 Molecular
and High-Temperature Hydrodynamic Phenomena Theory of Gases and Liquids (New York: Wiley)
(New York: Academic) p 515 [45] Luis K and Carlson D 1964 Normal shock position in under
[21] Muntz E P, Hamel B B and Maguire B L 1970 AIAA J. 8 1651 expanded and two-phase jet Rocket Techniques and
[22] Burnett D 1934 Proc. London Math. Soc. 40 382 Cosmonautics 2 239
[23] Mareschal M and Salomons E 1994 Transport Theory and [46] Landau L D and Lifshitz E M 1987 Fluid Mechanics 2nd
Statistical Phys. 23 281 edn (London: Pergamon)
[24] Boyd I D 1997 Phys. Fluids 9 3086 [47] Mott-Smith H M 1951 Phys. Rev. 82 885
[25] Usami M and Teshima K 1999 JSME Int. J. B 42 369 [48] Mazouffne S, Vankan P, Engeln R and Schram D C 2001
[26] Kannenberg K C and Boyd I 2000 Strategies for efficient Behaviour of the H atom velocity distribution function
particle resolution in the direct simulation Monte Carlo within the shock wave of a hydrogen plasma jet Phys. Rev.
method J. Comput. Phys. 157 727–45 E 64 066405
1372