Numerical Heat Transfer, Part A, 40:37^54 , 2001

Copyright # 2001 Taylor & Francis

1040-7782 /01 $12.00 + .00

NUM ERICAL SIM ULATIONS OF R ESONANT

OSCILLATIONS IN A TUB E

Hua-Z hong Tang and Ping Cheng

Department of Mechanical Engineering, The Hong Kong University of Science
and Technology, Clear Water Bay, Kowloon, Hong Kong

Kun Xu
Department of Mathematics, The Hong Kong University of Science and
Technology, Clear Water Bay, Kowloon, Hong Kong

The article concerns a genuinely two-dimensional numerical study of resonant oscillation

phenomena in a gas-Ž lled tube with an isothermal wall or an adiabatic wall. The
time-dependent , axisymmetric, compressible Navier–Stokes equations in two dimensions
are solved by a new Ž nite volume method with the second-order kinetic  ux-vector splitting
( KFVS) scheme for convective terms, and a third-order Runge–Kutta method for the time
evolution. The oscillatory motion of the  uid in a closed tube is generated by a piston at one
end, and re ected by the other closed end. W eak shock waves propagating within the tube
at the resonant frequency and slightly off-resonance frequencies are numerically captured,
which are consistent with both experimental observation and previous theoretical analyses.
The interesting results of the sudden change in axial velocity near the piston and the closed
end are also presented.

INTROD UCTION
The understanding of gas oscillating behavior inside a tube is of both funda-
mental and practical interest [1^11]. Saenger and Hudson [1] observed from exper-
iments that a weak shock wave would appear in a closed tube if a piston was
oscillating at the resonant frequency on the other end of the tube. To explain this
observation, Saenger and Hudson [1] tried to construct theoretical models based
on the existence of propagating shock waves. However, their analysis was not very
successful because viscosity and heat conduction were not taken into consideration,
and the amplitude of gas oscillation became in¢nite. Later, Betchov [2] showed that
nonlinear cumulative effects could lead to ¢nite amplitude gas oscillations in a closed
tube even if viscosity and heat conduction were not considered. By combining the
Received 3 March 2000; accepted 15 February 2001.
This work was supported by an RGC grant (HKUST 6039/99E). The ¢rst author would also like to
thank the National Natural Science Foundation of China (No. 19901031 ) and the Foundation of National
Key Laboratory of Computational Physics for partial support of this work. We would like to thank Mr.
G. Q. Lu for helpful discussions.
Address correspondence to Hua-Zhong Tang, State Key Laboratory of Scienti¢c and Engineering
Computing, Institute of Computational Mathematics and Scienti¢c/Engineering Computing, Chinese
Academy of Sciences, Beijing 100080, P.R. China.

37
38 H.-Z. TANG ET AL.

NOM ENCLATUR E
Ai;j control volume R0 radius of tube
Cp speci¢c heat at constant pressure T temperature
e total energy u axial velocity
E; F convective £uxes U conservative variables
E~ ; F~ diffusive £uxes …u~ ; v~ † molecular velocity
g Maxwell^Boltzmann v radial velocity
distribution r density
k thermal conductivity coef¢cient m viscosity
K internal degree of freedom t stress tensor
L length of tube C moment function vector
M number of grid points in axial g ratio of the speci¢c heats
direction Dt grid size in time t direction
N number of grid points in radial Dx grid size in axial direction
direction Dr grid size in radial direction
p pressure x internal particle velocity
Pr Prandtl number

method of characteristics and a perturbation method with both gas compressibility

and viscosity taken into consideration, Chester [3] was able to predict the shape
and strength of the shock waves. Most recently, a number of perturbation solutions
have been published on thermoacoustic effects in a resonant tube with applications
to thermoacoustic machine. (See [4^7] and the references therein.)
Although there exist a few numerical investigations of wave propagation in a
tube [8^9], they are restricted to solving one-dimensional nonlinear Lagrangian wave
equations. As far as the authors are aware, a numerical simulation of resonant
oscillations of a two-dimensional compressible viscous £ow inside a closed tube
and its subsequent appearance of shock waves has not been carried out. There exist
several dif¢culties in the numerical simulation of such a compressible £ow by solving
fully compressible Navier^Stokes equations. First, the appearance of weak shock
waves in resonant oscillations requires a high-resolution numerical scheme so that
discontinuities can be captured sharply. Thus, application of some high-order
schemes (such as MacCormack’s scheme and the Lax^Wendroff scheme) to the cur-
rent problem would lead to numerical oscillations. Second, the extremely small
Mach number of the gas £ow in the tube would result in small spatial pressure
variations but affect the velocity ¢eld at leading order [10]. The third dif¢culty
is that it would require a great deal of computer time before the appearance of shock
waves in the simulation, thus requiring a highly accurate numerical scheme in time so
that numerical dissipation because of truncation errors would not lead to inaccurate
numerical results.
During the last two decades, various high-resolution schemes for capturing
discontinuous solutions have been developed for the Euler equations [11^13] as well
as for the Navier^Stokes equations [14,15]. These include total variation diminishing
(TVD) schemes [11] and essentially nonoscillation (ENO) schemes [12], and so on. In
the development of modern high-resolution shock capturing methods, the Monotone
Upstream-Centered Schemes for Conservation Laws (MUSCL) interpolation tech-
 RESONANT OSCILLATIONS IN A TUBE 39

nique [13] has played an important role. We refer the readers to [16, 17] and the
references therein for a detailed introduction to high-resolution methods. The main
features of a high-resolution scheme include the following:
. at least second-order accuracy for smooth solutions,
. sharp resolution of discontinuities without excess smearing, and
. the absence of spurious oscillations in the computed solution.
To capture accurate discontinuity solutions in aerodynamic applications, most
high-resolution methods have intrinsically implemented approximate or exact
Riemann solvers. This requires considerable work and cost for the simulation of
any physical problems. Because of this, the £ux-vector splitting methods and other
high-resolution shock capturing methods without implementation of Rieman
solvers, such as gas-kinetic methods, have been developed in recent years [18^21].
In this article, ¢rst we use a high-resolution shock-capturing scheme for the
unsteady axisymmetric compressible Navier^Stokes equations to simulate numeri-
cally resonant oscillation phenomena in a gas-¢lled tube with an isothermal wall
or an adiabatic wall. In the construction of the current scheme, the kinetic £ux-vector
splitting and and the initial reconstruction technique [13] are used for the convective
terms, second-order central difference for derivatives in the diffusive terms, and the
source terms. For time evolution, we use a third-order Runge^Kutta method to relax
the time constraints.

GOVER NING EQUATIONS

Let x and r denote the axial and radial coordinates; t is the time; u and v denote
the velocity components in x- and r-directions, respectively; and r, e, and p denote
the £uid density, the total energy per unit volume, and the pressure, respectively.
The axisymmetric Navier^Stokes equations for a compressible £uid can be written
in the following conservative form:

@rU @r…E ¡ E~ † @r…F ¡ F~ †

‡ ‡ ˆ S …U † …1†
@t @x @r
where the conservative variables U , the source term S …U †, convective £uxes E and F ,
and the diffusive £uxes E~ and F~ , are de¢ned by
U ˆ ‰ r; ru; rv; eŠT
S …U † ˆ ‰ 0; 0; p ¡ tyy ; 0ŠT
E …U † ˆ ‰ ru; ru2 ‡ p; ruv; u…e ‡ p†ŠT
" #T
@T
E~ …U † ˆ 0; txx ; txr ; utxx ‡ vtxr ‡ k
@x
F …U † ˆ ‰ rv; ruv; rv2 ‡ p; v…e ‡ p†ŠT
" #T
@T
F~ …U † ˆ 0; trx ; trr ; utrx ‡ vtrr ‡ k …2†
@r
40 H.-Z. TANG ET AL.

In Eq. (2), txx , txr , and trr denote the components of the stress tensor given by

txx ˆ …2m ‡ m~ †
( )
@u
@x
v @v
‡ m~ ‡
r @r
@
( )
v
trr ˆ …2m ‡ m~ † ‡ m~ ‡
@r
v
r @x
@u

( )
v
tyy ˆ …2m ‡ m~ † ‡ m~
r
@v @u
‡
@r @x

txr ˆ trx ˆ m
( )
@u @v
‡
@r @x
…3†

Here m~ ˆ ¡2=3m, and m and k denote the viscosity and the thermal conductivity
coef¢cients, respectively, and T is the gas temperature. The viscosity of the £uid
is de¢ned by Sutherland’s law
1 :5
m ˆ m0 ( )
T
T0
T0 ‡ C
T ‡C
…4†

where T0 and m0 are reference values at standard sea level conditions, and C is a given
constant that is equal to 110. If the Prandtl number is assumed to be constant
(approximately equal to 0.71 for calorically perfect air), thermal conductivity
can be calculated from the equation Pr ˆ mCp =k, where Cp is the speci¢c heat at
constant pressure.
For a perfect gas, the pressure is related to the total energy e by
p ˆ …g ¡ 1†‰e ¡ 12 r…u2 ‡ v2 †Š

NUM ERICAL M ETH OD S

Space D iscretizations
Let xi ˆ iDx and rj ˆ j Dr (i ˆ 0; 1; . . . ; M; j ˆ 0; 1; . . . ; N) denote grid points
in x- and r-directions, respectively, where Dx ˆ L=M and Dr ˆ R0 =N; then the com-
putational domain ‰0; LŠ £ ‰0; R0 Š is subdivided into quadrilaterals as shown in
Figure 1.
Integrating Eq. (1) over a control volume Ai;j ˆ ‰xi¡1=2 ; xi‡1=2 Š £ ‰rj¡1=2 ; rj‡1=2 Š
gives
ZZ Z rj‡1=2
@
rU dx dr ‡ r……E ¡ E~ †i‡1=2 ¡ …E ¡ E~ †i¡1=2 † dr
@t Ai;j rj¡1 =2
Z xi‡1=2
‡ …rj ‡1=2 …F ¡ F~ †j ‡1=2 ¡ rj ¡1=2 …F ¡ F~ †j ¡1=2 † dx
xi¡1=2
ZZ
ˆ S …U † dx dr
Ai ; j
 RESONANT OSCILLATIONS IN A TUBE 41

Figure 1. Grid points (.) and control volume Ai;j surrounded by the dashed lines.

which can be approximated using a midpoint formula such as

@ rj
rj Ui;j …t† ‡ ‰…E ¡ E~ †i ‡1=2;j ¡ …E ¡ E~ †i¡1;2;j Š
@ t D x
1
‡ ‰rj‡1=2 …F ¡ F~ †i;j ‡1=2 ¡ rj ¡1=2 …F ¡ F~ †i;j¡1=2 Š
Dr Z Z
1
ˆ S …U † dx dr …5†
DxDr Ai;j

where Ui;j …t† denote the cell averaged conservative variables de¢ned by
Z xi‡1=2 Z rj ‡1=2
1
Ui;j …t† ˆ U …x; r; t† dx dr
DxDr xi¡1=2 rj ¡1=2

?twb=.60w>The approximation given by Eq. (5) keeps second-order accuracy in

space. Based on the conservative variables at the cell center, the convective
and diffusive £uxes across cell interfaces have to be evaluated according to the
gas evolution models.
In the following we are going to construct the convective £uxes across each cell
interface of the control volume Ai;j . Following the collisionless gas evolution model
[21], the macroscopic £uxes E …U † and F …U † can be split into two parts, that is, posi-
tive and negative £uxes

E …U † ˆ E …U †‡ ‡ E …U †¡ F …U † ˆ F …U †‡ ‡ F …U †¡ …6†

where E § and F § are given by

42 H.-Z. TANG ET AL.

Z Z §1
E § …U † ˆ hC; u~ gi§ ² § u~ Cg d u~ d v~ d x
RK ‡ 1 0

Z Z §1
F § …U † ˆ hC; v~ gi§ ² § v~ Cg d v~ d u~ d x …7†
RK ‡ 1 0

In Eq. (7), R ˆ …¡1; 1 †, …u~ ; v~ † denote molecular velocity components in x- and

r-directions; d x ˆ x1 ; . . . ; d xK and x1 ; . . . ; xK are the components of the internal par-
ticle velocity in K dimension space. For the £ow movement in the two-dimensional
case, K is related to the ratio of speci¢c heat g through the relation
K ˆ …3 ¡ g†=…g ¡ 1† ¡ 1. Here g is the Maxwell^Boltzmann distribution for the equi-
librium state
…K ‡2†=2

()l 2 2 2
gˆr e¡l‰…u~ ¡u† ‡…v~ ¡v† ‡x Š
…8†
p

and C is the moment function vector

" #T
u~ 2 ‡ v~ 2 ‡ x2
C ˆ 1; u~ ; v~ ; …9†
2

where x2 ˆ x21 ‡ ¢ ¢ ¢ ‡ x2K , and l is a function of temperature such that l ˆ r=2p. The
splitting £uxes E § and E § have the explicit forms
0 1
rh~ui§
B rh~u2 i§ C
E § …U † ˆ B C …10†
@ rh~ui§ h~vi A
1 2
u3 i§ ‡ h~ui§ …h~v2 i ‡ hx i††
2 r…h~

and
0 1
rh~vi§
B rh~vi§ h~ui C
F § …U † ˆ B
@ v 2 C
A …11†
rh~ i§
1 3 2 2
v v u
2 r…h~ i§ ‡ h~ i§ …h~ i ‡ hx i††

where
2 K
hx i ˆ
2l
2 2 1
h~ui ˆ u h~u i ˆ u ‡
2l
0 1 p

h~u i§ ˆ erfc…¨ lu†
2
2
0 e¡lu
h~ui§ ˆ uh~u i§ § p 

2 lp
¢¢¢¢¢¢
 RESONANT OSCILLATIONS IN A TUBE 43

m 2 m 1 m‡1 m
h~u ‡ i§ ˆ uh~u ‡ i§ ‡ h~u i§ m ˆ 0; 1; 2; . . .
2l
The formulas of h~vm i§ and h~vm i are the same as the above, with u replaced by v. Here
erfc…x† is the complementary error function
Based on the above kinetic £ux-vector splitting (KFVS) formulation (6), the
convective £uxes E …U † and F …U † at each cell interface of the control volume Ai;j
can be obtained and Eq. (5) becomes
@ rj
rj Ui;j …t† ‡ ‰G…Ui;j ; Ui‡1;j † ¡ G…Ui¡1;j ; Ui;j †Š
@t Dx
1
‡ ‰rj‡1=2 H …Ui;j ; Ui;j‡1 † ¡ rj¡1=2 H …Ui;j ; Ui;j †Š
Dr
rj 1
¡ ‰E~ i‡1;2;j ¡ E~ i¡1;2;j Š ¡ ‰rj‡1=2 F~ i;j‡1=2 ¡ rj¡1=2 F~ i;j¡1=2 Š
DxZ Z Dr
1
ˆ S …U † dx dr …12†
DxDr Ai;j
where convective £uxes G…Ui;j ; Ui‡1;j † and H …Ui;j ; U i;j ‡1 † are de¢ned as
G…Ui;j ; Ui‡1;j † ˆ E ‡ …Ui;j † ‡ E ¡ …Ui‡1;j †
H …Ui;j ; Ui;j ‡1 † ˆ F ‡ …U i;j † ‡ F ¡ …Ui;j‡1 † …13†
It is well known that the above ¢nite volume scheme is only ¢rst-order accurate in
space. To improve the accuracy of the scheme, the initial reconstruction technique
[13] has to be applied to interpolate the cell averaged variables to the cell interfaces.
For example, a linear function
Ui;j …x; r† ˆ Ui;j ‡ …Ux †i;j …x ¡ xi † ‡ …Ur †i;j …r ¡ rj †
for …x; r† 2 ‰xi¡1=2 ; xi‡1=2 Š £ ‰ rj¡1=2 ; rj‡1=2 Š …14†
can be constructed to approximate the cell averaged variables Ui;j at the beginning of
each time step, where …Ux †i;j and …Ur †i;j are the approximate slopes in the x- and
r-directions inside the control volume Aij ; that is,

…Ux †i;j ˆ
( )
@U
@x i; j
p
‡O……Dx† † …Ur †i;j ˆ
( )
@U
@r i; j
p
‡O……Dr† † p ¶ 1

To avoid overshoot and undershoot in he reconstructed initial data, the slopes

of U , that is @U =@x and @U =@r, have to be limited. In this article, the van Leer limiter
[13] is used; that is,
js‡i ;j jj si;j j
¡
…Ux †i;j ˆ …sgn …si ;j ‡ sgn …si ;j †† …15†
‡ ¡
j s‡
i;j j ‡ j si ;j j
¡

where sgn is the sign function, and

Si‡;j ˆ …Uin‡1;j ¡ Uin;j †=…xi‡1;j ¡ xi;j † s¡ n n
i ;j ˆ …Ui ;j ¡ Ui¡1;j †=…xi ;j ¡ xi¡1;j † …16†

are the corresponding slopes in the neighboring cells. Similarly, …U r †i;j can be
obtained.
44 H.-Z. TANG ET AL.

Based on the reconstruction given in Eq. (14), we can establish a high-order

spatial resolution solver for Eq. (1) as follows:
@ rj
rj Ui;j …t† ˆ ¡ ‰ G…U ~ i‡1=2;j ; U^ i‡1=2;j † ¡ G…U^ i¡1=2;j ; U^ i¡1=2;j †Š
@ t Dx
1
¡ ‰rj‡1=2 H …U ~ i;j‡1=2 ; U^ i;j‡1=2 † ¡ rj¡1=2 H …U ~ i;j¡1=2 ; U^ i;j ¡1=2 †Š
Dr
rj 1
‡ ‰E~ i‡1=2;j ‡ E~ i¡1=2;j Š ‡ ‰ rj‡1=2 F~ i;j‡1=2 ¡ rj¡1=2 F~ i;j¡1=2 Š
Dx Z Z Dr
1
‡ S …U † dx dr …17†
DxDr Ai;j
where
U
~ i‡1=2;j ˆ U i;j …xi‡1=2 ; rj † U^ i‡1=2;j ˆ Ui‡1;j …xi‡1=2 ; rj †

and U~ i;j ‡1=2 and U^ i;j ‡1=2 are de¢ned similarly. Equation (17) has second-order accu-
racy in the smooth £ow region.
Until now, the KFVS method with the combination of initial data rec-
onstruction has been presented for the discretization of convective £uxes. For
the diffusion terms, a central difference scheme is commonly used for the
discretization of diffusive £uxes. For example, the derivatives in rj E~ i‡1=2;j are
approximated by
1
rj …txx †i‡1=2;j & rj …2m ‡ m~ †i‡1=2;j …ui‡1;j ¡ ui;j † ‡ m~ i ‡1=2;j
Dx

" #
1 rj
£ …vi‡1;j ‡ vi ;j † ‡ …vi‡1;j ‡1 ¡ vi‡1;j ¡1 ‡ vi ;j ‡1 ¡ vi ;j ¡1 †
2 4D r

" #
1 1
rj …txr †i‡1=2;j & rj mi‡1=2;j …vi ‡1;j ¡ vi ;j † ‡ …ui ‡1;j ‡1 ¡ ui ‡1;j ¡1 ‡ ui;j ‡1 ¡ ui ;j ¡1 †
Dx 4Dr

( )
rj k
@T
@x i ‡1=2;j
&ki‡1=2;j
rj
Dx
…Ti‡1;j ¡ Ti ;j † …18†

Time D iscretizations
In the previous section we presented numerical approximation for the con-
vective and difusive terms. To keep the second-order accuracy of the scheme, a
second-order method must be used for the sorce terms. We now turn to the temporal
discretization of Eq. (17). Denoting the right-hand side of Eq. (17) as L…U † and
omitting the spatial indices, Eq. (17) can be written as
dU
ˆ L…U † …19†
dt
 RESONANT OSCILLATIONS IN A TUBE 45

The simplest single-step method for solving Eq. (19) is the explicit Euler method.
However, there is usually a step size constraint associated with the convection
and diffusive terms if the explicit Euler method is used. To overcome these
dif¢culties, we use a third-order Runge^Kutta method [22] for the time dis-
cretization.
To advance the value of U from time level n to n ‡ 1 in Eq. (19) by the
third-order Runge^Kutta method, we have

U …0† ˆ U n
U …1† ˆ U …0† ‡ DtL …U …0† †
…2 †
3 0 1 1 1
U ˆ 4 U … † ‡ 4 …U … † ‡ DtL…U … † ††
U …3† ˆ 13 U …0† ‡ 23 …U …2† ‡ DtL…U …2† ††
U n‡1 ˆ U …3† …20†

NUM ERICAL STUD Y OF RESONANT OSC ILLATIONS IN A TUB E

We now study the compressible £ow oscillating inside a closed tube with a
piston oscillating at the left end …x ˆ 0† having the velocity u ˆ l o cos…ot†. Exten-
sive computations using the above numerical method were carried out for the study
of gas oscillations in a closed tube. The tube has a total length of L ˆ 1:7018 m with
the radius of R0 ˆ 2:412 £ 10¡2 m. The tube is ¢lled with air under normal circum-
stances (T0 ˆ 291 :785 K , p0 ˆ 101325 Pa). These parameters were used in a pre-
vious article by Saenger and Hudson [1]. The sound speed corresponding to
the pressure and density of the air is a0 ˆ 332 :5317. Therefore, the ¢rst resonant
frequency of the tube becomes fr ˆ o=2p ˆ a0 =2L ˆ 100 :6 H z. The piston (at
the end of the tube) has an oscillatory amplitude of l ˆ 3:175 £ 10¡3 m [1]. The
computational domain was divided into 160 £ 40 grid points with equal cell size
2 4
Dx ˆ 1:063625 £ 10¡ m and Dr ˆ 6:0325 £ 10¡ m. A schematic description of

Figure 2. Schematic description of the tube and the coordinates.

46 H.-Z. TANG ET AL.

the tube is given in Figure 2. In the following computations, the time step size Dt is
determined by the Courant^Friedriche, Levy (CFL) condition

CCFL min fDx; Drg

Dt µ p




maxU fjuj ‡ jvj ‡ gp=pg

which comes from the explicit discretization of the convective terms, and CCF L
denotes CFL constant and is taken as 0.4.
No-slip boundary conditions for the £uid velocity are imposed at all solid
walls. Both adiabatic wall and isothermal wall boundary conditions are con-
sidered. At the symmetry axis …r ˆ 0†, the £ow conditions v ˆ 0 and
@r=@r ˆ @u=@r ˆ @p=@r ˆ 0 are imposed. Figures 3^10 show the numerical results
for the case with an adiabatic wall with the piston oscillating at a resonant frequency
of 100.6 Hz and two frequencies departing or far away from the resonant frequency,
respectively, while Figures 11 and 12 show the results for the case of an isothermal
wall. As is shown in these ¢gures, the strength of the shock wave at the resonant
frequency is weak. Because the strength of the shock wave is stronger in the case
of the adiabatic all, we shall focus our discusson on this case.

Figure 3. (a) Temporal variations of pressure, density, and temperature near the middle of the tube (at
x=L ˆ 0:496875 and r ˆ 0) with an adiabatic wall at resonant frequency (i.e., f ˆ fr ˆ 100 :6 H z†. (a:
p=p0 , b: r=r0 , C: T =T0 .) (b) Temporal variations of axial velocity u=lo near the middle of the tube
(at x=L ˆ 0:496875 and r ˆ 0) with an adiabatic wall at resonant frequency (i.e., f ˆ fr ˆ 100 :6 H z.
 RESONANT OSCILLATIONS IN A TUBE 47

Figures 3a and 3b show the temporal variations of pressure, density,

temperature, and axial velocity near the midpoint of the tube (at
x=L ˆ 0:496875) and at the center line of the tube …r ˆ 0†. It is noted that the tem-
poral variations of pressure, density, and temperature are in phase with each other.
The periodically sharp increase in pressure, density, and temperature shown in
Figure 3a indicates the propagation of a weak shock wave. Figure 3b shows the
axial velocity near the middle of the tube, which varies almost sinuoisdally in
the positive and negative direction periodically. The numerical results shown in
Figures 3a and 3b are very similar to those presented in Figure 1 of Betchov’s paper
[2].
Figures 4a and 4b show the corresponding temporal variations of pressure,
density, temperature, and axial velocity near the piston (x=L ˆ 3:125 £ 10¡3 and
r ˆ 0). A comparison of Figure 4a and Figure 3a shows that the period and ampli-
tude of temporal variations of pressure, density, and temperature at a location near
the piston, which are larger than those near the middle of the tube. At a small dis-
tance away from the piston, the temporal variations of axial velocity shown in
Figure 4b are almost sinusoidally in phase with the piston movement. The weak
shock wave is generated whenever there is a sudden increase in pressure and the
axial velocity is suddenly changed from zero to a negative value, which forms a
pulse-type shock wave.

Figure 4. Temperoral variations of pressure, density, and temperature near the piston (at
x=L ˆ 3:125 £ 10 ¡3 and r ˆ 0) with an adiabatic wall at resonant frequency (i.e., f ˆ fr ˆ 100:6 Hz).
(a: p=p0 , b: r=r0 , c: T =T0 .) (b) Temporal variations of axial velocity u=lo near the piston (at
x=L ˆ 3:125 £ 10 ¡3 and r ˆ 0) with an adiabatic wall at resonant frequency (i.e., f ˆ fr ˆ 100 :6 Hz).
48 H.-Z. TANG ET AL.

Figure 5. (a) Temporal variations of pressure density, and temperature near the closed end (at
x=L ˆ 0:996875 and r ˆ 0) with an adiabatic wall at resonant frequency (i.e., f ˆ fr ˆ 100:6 Hz). (a:
p=p0 , b: r=r0 , c: T =T0 .) (b) Temporal variations of velocity u=lo near the closed end (at
x=L ˆ 0:996875 and r ˆ 0) with an adiabatic wall at resonant frequency (i.e., f ˆ fr ˆ 100 :6 Hz).

The corresponding temporal variations of pressure, density, temperature, and

axial velocity near the closed end are shown in Figures 5a and 5b. A comparison
of Figures 3a, 4a, and 5a shows that (1) the period and amplitude of temporal
variatons of pressure, density, and temperature near the piston and the closed
end are almost identical, and (2) the amplitudes of these oscillations are smallest
near the middle of the tube. Figure 5b shows the temporal variation of axial velocity
near the closed end (at x=L ˆ 0:996875 and r ˆ 0), which are small except during the
occurrence of a shock wave. It is interesting to note that the axial velocity is always
nonnegative near the closed end at x=L ˆ 1 and the weak shock appears as a
pulse-type axial velocity variation.
Figure 6 shows the mesh re¢nement study of the same problem, where solid and
dashed lines denote pressure near the closed end as a function of time obtained with
160 £ 40 and 320 £ 80 mesh points, respectively. It is shown that there are only slight
differences in pressure near the valley of the curve when a re¢ned grid size was used.
We also refer the readers to [20, 21] for detailed grid convergence analyses of the gas
kinetic method.
Figure 7a gives axial velocity and temperature pro¢les at six different locations
of x=L at t ˆ 0:486 s. As shown in this ¢gure, the maximum axial velocity occurs near
the wall of the tube. Figure 7b shows that the temperature variations along the radial
direction are small for this case with an adiabatic wall.
 RESONANT OSCILLATIONS IN A TUBE 49

Figure 6. Mesh re¢nement study of pressure as a function of time near the closed end (at x=L ˆ 0:996875)
with an adiabatic wall at f ˆ 100:6 H z. Solid and dashed lines denote solutions on the ¢ne and coarse
grids, respectively.

Figure 7. Axial velocity and temperature pro¢les with an adiabatic wall at t ˆ 0:486 s at different
locations x=L at resonant frequency of f ˆ fr ˆ 100 :6 H z.

Figure 8 shows the distribution of the pressure, density, and temperature along
the tube at two different times at t1 ˆ 0:4023 s and t2 ˆ 0:4077 s, while Figure 9
shows the corresponding axial velocity. Figure 9a indicates the axial velocity at
t ˆ 0:4023 s at which time the piston is moving to the left (u…0; 0:4023 † < 0 in
Figure 8) and the shock wave is moving from the closed end …x ˆ L† to the piston
…x ˆ 0†. Figure 9b indicates the axial velocity at t ˆ 0:4077 s at which time the piston
is moving to the right …u…0; 0:4077† > 0 in Figure 8) and the shock wave is moving to
the right (u…0; 0:4077 † > 0 in Figure 8) and the shock wave is moving toward the
closed end.
50 H.-Z. TANG ET AL.

Figure 8. Axial variations of pressure, density, and temperature along center line …r ˆ 0† at ot ˆ 254:2892
(a) and ot ˆ 257:7025 (b) with an adiabatic wall at resonant freqency of f ˆ fr ˆ 100 :6 H z. (a: p=p0 , b:
r=r0 , c: T =T0 ).

Figure 9. u=lo vs. x=L at ot ˆ 254 :2892 (a) and ot ˆ 257:7025 (b) with an adiabatic wall at resonant
frequency of f ˆ f r ˆ 100 :6 Hz.

When the oscillating frequency …f ˆ 104 :7 H z† of the piston departs from the
resonant frequency …f ˆ 100 :6 H z†, the shock wave generated inside the tube
becomes weaker as shown in Figure 10a. If the oscillation frequency
…f ˆ 92:3 H z† is far away from the resonance frequency …f ˆ 100:6 H z†, the piston
only generates standing waves moving in the tube as shown in Figure 10b which
does not have a shock wave.
 RESONANT OSCILLATIONS IN A TUBE 51

Figure 10. (a) Temporal variations of pressure, density, and temperature near the closed end (at
x=L ˆ 0:996875) with an adiabatic wall at f ˆ 104 :7 H z. (a: p=p0 , b: r=r0 , c: T =T0 †. (b) Temporal
variations of pressure, density, and temperature near the closed end (at x=L ˆ 0:996875) with an adiabatic
wall at f ˆ 92:3 H z. (a: p=p0 , b: r=r0 , c: T =T0 .)

The numerical results with an isothermal wall condition are shown in

Figures 11 and 12. Similar wave structures are observed inside the tube even though
the shock waves become much weaker in this case as compared with the adiabatic
case. (See Figures 5a and 11a). This is because for the adiabatic case there is no
heat loss from the £uid to the wall, whereas for the isothermal wall there is heat
transfer from the £ow to the wall at certain locations along the wall. At the resonant
frequency of 100.6 Hz, the captured shock waves can be clearly observed. Figure 11b
shows the radial distribution of the axial velocity (left) and temperature pro¢les
(right) at t ˆ 0:391 s at different locations at resonant frequency. It is shown that
the axial velocity (left ¢gure) is positive in most locations away from the closed end,
which means that the gas at most locations at this instant is moving from the piston
to the closed end. Near the closed end, the axial velocity is negative, which implies
that the wave is re£ected at the instant. Radial temperature gradients near the wall
are negative near the piston and positive near the closed end, which indicate that
heat is transferred from the gas to the wall near the piston and heat is transfered
from the wall to the gas near the closed end. (See Figure 11a.) However, at
f ˆ 97:7 H z, which is away from the resonant frequency …f ˆ 100 :6 H z†, the weak
shock waves as shown in Figure 12a are much more easily spread out in comparison
with the adiabatic case. Figure 12b shows that the corresponding axial velocity and
temperature distributions at different locations x=L at f ˆ 99:7 H z, which have simi-
lar behavior to Figure 11b.
52 H.-Z. TANG ET AL.

Figure 11. (a) Temporal variations of pressure, density, and temperature near the closed end (at
x=L ˆ 0:996875) with an isothermal wall at resonant frequency of f ˆ fr ˆ 100:6 H z. (a: p=p0 , b:
r=r0 , c: T =T0 .) (b) Axial velocity and temperature pro¢les at t ˆ 0:391 s and at different locations
x=L with an isothermal wall at resonant frequency of f ˆ fr ˆ 100:6 H z.

D ISCUSSION AND C ONCLUSION

In this article we have presented a gas-kinetic scheme to simulate
two-dimensional resonant gas oscillations in a tube. The MUSCL-type KFVS
methods for convective terms and a third-order Runge^Kutta time discretization
were used to approximate numerically the time-dependent, axisymmetric, compress-
ible Navier^Stokes equations.
The oscillatory motion is generated by a piston at one end and re£ected by
another closed end. The dependence of the £ow distribution on the different
oscillating frequencies and wall boundary conditions are presented. At the resonant
frequency and slightly off-resonance frequencies, weak shock waves propagating
inside the tube are captured very well, even though the oscillation amplitude is small
and the £ow Mach number is very low. These results are consistent with those from
both experimental observation and previous theoretical anlyses. The results also
show the interesting phenomena of a pulse-type shock wave, which appeared in
the axial velocity near the piston and near the closed end.
 RESONANT OSCILLATIONS IN A TUBE 53

Figure 12. (a) Temporal variations of pressure density, and temperature near the closed end (at
x=L ˆ 0:996875) with an isothermal wall at a frequency of f ˆ 97 :7 H z. (a: p=r0 , b: r=r0 , c: T =T0 :†
(b) Axial velocity and temperature pro¢les at t ˆ 0:394 s and at different locations x=L with an isothermal
wall at a frequency of f ˆ 97 :7 Hz.

R EF ER ENCES
1. R. A. Saenger and G. E. Hudson, Periodic Shock Waves in Resonating Gas Columns, J.
Accoust. Soc. Am. vol. 32, pp. 961^971, 1960.
2. R. Betchov, Nonlinear Oscillations of a Column of Gas, Phys. Fluid, vol. 1, pp. 205^212,
1958.
3. W. Chester, Resonant Oscillations in Closed Tubes, J. Fluid Mech., vol. 18, pp. 44^64,
1964.
4. P. Merkli and H. Thomann, Thermoacoustic Effects in Resonance Tube, J. Fluid Mech.,
vol. 70, pp. 161^177, 1975.
5. A. Gopinath, N. L. Tait, and S. L. Garrett, Thermoacoustic Streaming in a Resonant
Channel: The Time Averaged Temperature Distribution, J. Acoust. Soc. Am.,
vol. 103, pp. 1388^1405 , 1998.
6. L. Bauwens, Oscillating Flow of a Heat-Conducting Fluid in a Narrow Tube, J. Fluid
Mech., vol. 324, pp. 135^161, 1996.
7. A. Goldshtein, P. Vainshtein, M. Fichman, and C. Gut¢nger, Resonance Gas Oscillations
in Closed Tubes, J. Fluid Mech, vol. 322, pp. 147^163, 1996.
54 H.-Z. TANG ET AL.

8. C. P. Lee and T. G. Wang, Nonlinear Resonance and Viscous Disspation in an Acoustic

Chamber, J. Acous. Soc. Am., vol. 92, pp. 2195^2206, 1992.
9. L. Elvira-Segura and E. Sarabia, Numerical and Experimental Study of Finite-Amplitude
Standing Waves in a Tube at High Sonic Frequencies, J. Acous. Soc. Am., vol. 104,
pp. 708^714, 1998.
10. T. Schneider, N. Botta, K. J. Geratz, and R. Klein, Extension of Finite Volume
Compressible Flow Solvers to Multi-Dimensional, Variable Density Zero Mach
Number Flows, J. Comput. Phys., vol. 155, pp. 248^286, 1999.
11. A. Harten, High-Resolution Schemes for Hyperbolic Conservation Laws, J. Comput.
Phys., vol. 49, pp. 357^393, 1983.
12. A. Harten, B. Engquist, S. Osher, and S. R. Chakravarthy, Uniformly High-Order
Accurate Non-Oscillatory Schemes III, J. Comput. Phys., vol. 71, pp. 231^303, 1987.
13. B. van Leer, Towards the Ultimate Conservative Difference Scheme V. A Second-Order
Sequel to Godunov’s Method, J. Comput. Phys., vol. 32, pp. 101^136, 1979.
14. D. Drikakis and F. Durst, Investigation of Flux Formulae in Transonic Shock
Wave/Turbulent Boundary Layer Interaction, Int. J. Numer. Math. Fluids, vol. 18,
pp. 385^413, 1994.
15. H. C. Ye, G. H. Klopfer, and J. L. Montagne, High Resolution Shock Capturing Schemes
for Inviscid and Viscous Hypersonic Flows, J. Comput. Phys., vol. 88, pp. 31^61, 1990.
16. R. J. LeVeque, Numerical Methods for Conservation Laws, BirkhÌuser-Verlag, 1990.
17. E. F. Toro, Riemann Solvers and Numerical Methods for Fluid Dynamics, A Practical
Introduction , 2nd ed., Springer, Berlin, New York, 1999.
18. D. I. Pullin, Direct Simulation Methods for Compressible Inviscid Ideal Gas Flow, J.
Comput. Phys., vol. 34, pp. 231^244, 1980.
19. J. L. Steger and R. F. Warming, Flux-Vector Splitting of the Inviscid Gas-Dynamic
Equations with Applicatons to Finite Difference Schemes, J. Comput. Phys., vol. 40,
pp. 263^293, 1981.
20. K. Xu, L. Martinelli, and A. Jameson, Gas-Kinetic Volume Methods, Flux-Vector
Splitting and Arti¢cial Diffusion, J. Comput. Phys., vol. 120, pp. 48^65, 1995.
21. H. Z. Tang and H. M. Wu, Kinetic Flux Vector Splitting for Radiation Hydrodynamic
Equations, Computers & Fluids, vol. 29, pp. 917^933, 2000.
22. C. W. Shu, Total-Variation-Diminishing Time Discretizations, SIAM J. Sci. Statist.
Comput., vol. 9, pp. 1073^1084 , 1988.
Copyright of Numerical Heat Transfer: Part A -- Applications is the property of Taylor &
Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a
listserv without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.