Non-Normality and Nonlinearity in Combustion-Acoustic Interactions in Diffusion Flames
Non-Normality and Nonlinearity in Combustion-Acoustic Interactions in Diffusion Flames
I. Introduction
T HE occurrence of combustion instabilities has been a plaguing problem in the development of combustors
for rockets, jet engines, and power generating gas turbines1. Predicting and controlling combustion
instability requires an understanding of the interactions between the combustion process and the acoustic
waves. Combustion-acoustic interaction involves a feedback mechanism where, the fluctuating heat release acts
as a source of energy for the acoustic field and the latter in turn affects the combustion process and hence the
heat release rate.
Combustion instability has been observed in both premixed and non-premixed combustors. A large number
of investigations have been performed on the combustion instability in premixed systems. A comprehensive
review of this subject can be found in Lieuwen2. As compared to premixed flames, not much work is done in
combustion instability of non-premixed flames. However, most gas turbine combustors for aircraft and other
industrial applications involve non-premixed combustion; therefore, it is important to investigate combustion
instability in such systems.
Combustion instability in non-premixed flames has been analyzed by a few authors. Vance et al.3 studied the
stability of one dimensional diffusion flame by analyzing the effect of perturbations on the Burke-Schumann
flame using linear stability analysis. Chakravarthy et al.4 analyzed the heat release rate response of a non-
premixed flame to velocity disturbances numerically. They also analyzed the effect of the equivalence ratio
fluctuations. Though their results indicated nonlinearity in combustion response, the nature of the nonlinear
interactions was not discussed in detail.
Recently Tyagi et al.5 investigated the unsteady combustion response of a ducted non-premixed flame and
coupling with the duct acoustic field using numerical simulations. They considered a two-dimensional co-
flowing non-premixed flame in a uniform flow field, as in the Burke-Schumann geometry. Both finite-rate and
infinite-rate chemistry effects were examined. The one-dimensional acoustic field was simulated in the time
domain using the Galerkin method, treating the fluctuating heat release from the combustion zone as a compact
1
Graduate Student
2
Corresponding author & Professor; Member of AIAA; [email protected]
1
American Institute of Aeronautics and Astronautics
Copyright © 2007 by Koushik Balasubramanian and R. I. Sujith. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
acoustic source. The combustion oscillations are shown to cause exchange of acoustic energy between the
different natural modes of the duct over several cycles of acoustic oscillations. The authors emphasize the
nonlinear nature of the interaction. However, the role of non-normality is not discussed.
The objective of this paper is to investigate the mechanism that leads to the complex interaction between the
different acoustic modes. The exchange of acoustic energy between the different modes leading to transient
growth happens even when finite rate chemistry effects are neglected. Since not much work has been performed
on combustion instability of non-premixed flames, the present approach is to retain only the most basic aspects
of such a flame, to gain clear understanding. The configuration chosen is identical to the one in Tyagi et al.5;
i.e., a two-dimensional co-flowing non-premixed flame in the Burke-Schumann geometry, with infinite rate
chemistry model. The present paper aims to clarify the roles of non-normality and nonlinearity in combustion-
acoustic interactions in the context of such a diffusion flame.
It will be shown that the modes of the oscillatory heat release are non-normal. Further, it will be shown that
the presence of heat release makes the coupled combustion-acoustic system non-normal. The heat release
fluctuations contain a host of disturbance modes. Each mode could be exponentially decaying; however, since
the eigenvectors of the operator are non-orthogonal, the amplitudes of the various modes are interdependent. As
a result of the non-normal behavior the solutions
exhibit large transient growth, which could
potentially trigger nonlinearities in the system when
the amplitudes reach high enough values. This
phenomena has been studied in detail in the context
of turbulence by Baggett et al.6 They explain that in
the non-normal evolution, the input and output
structures (such as streamwise vortices, streaks etc.)
are different and nonlinearity closes the feedback
loop by converting some of the output into input. In a
similar manner, the interplay between transient linear
growth resulting from the non-normality of the Figure 1. Shows that even if a system is linearly
modes and “nonlinear mixing” can indeed lead to the unstable, transient growth can cause the
growth of the acoustic modes over a large number of amplitude to increase to high enough values to
cycles. The consequences of non-normality of modes trigger the nonlinear driving which causes the
have been studied in the context of instability of amplitudes to grow or saturate.
magnetic plasmas7, the formation of cyclones8 and
transition to turbulence6, 9, 10.
In this paper, we emphasize on the non-normality of the combustion-acoustic interaction. Linear stability
becomes a poor indicator of the short-term dynamics for such non-normal systems11, 12. The non-normal nature
of the evolution equations would lead to transient growth before the oscillations eventually decays. However,
there could be situations where, the short-term growth of fluctuations leads to significant amplitudes where
nonlinear effects which could cause “nonlinear driving” can play a role, making the system unstable. This is
illustrated in Fig. 1.
An operator is said to be non-normal if it does not commute with its adjoint. Such an operator has non-
orthogonal eigenvectors and it is this property which leads to transient growth. This property is illustrated in
Figs. (2) and (3) which are adapted from11. In Fig. 2 & 3, e1 and e2 represent the direction of the eigenvectors,
Φ is a vector in the functional space, which is expressed as a linear combination of the eigenvectors. Figure 2
shows that for a normal system, Φ decreases monotonically if the amplitude of the individual eigenvectors
themselves decays. On the contrary, for the non-normal system shown in Fig. 3, Φ increases even when the
amplitude of individual eigenvectors decay. However Φ decays after a sufficiently long time if nonlinear
effects do not become significant during the transient growth.
In this paper, the partial differential equations governing the combustion acoustic interaction are reduced to
ordinary differential equations using the Galerkin technique. The complete evolution equations are solved
numerically using the fourth order Runge-Kutta scheme. This system of nonlinear ODEs is similar to the
numerous dynamical systems studied in literature. Such systems show interesting dynamical behavior such as
period doubling, in which the time required for the motion of the system to repeat itself doubles again and again
as a parameter describing the system is increased. Period doubling is one frequently encountered scenario
leading to chaos. The effect of various non-dimensional parameters such as the Peclet number, flame location,
slot width to duct width ratio, ratio of the acoustic to combustion time scales, etc. on the system stability is
investigated. The complete evolution equations are linearized and the linearized equations are found to be non-
normal. The transient growth is analyzed using the method adopted in Ref. 12. This analysis helps in
understanding the role played by non-normality in making the nonlinear effects significant through short-term
amplification.
2
American Institute of Aeronautics and Astronautics
Figure 2 shows monotone decay of a normal Figure 3 shows transient growth of a normal
system. The initial state is Φ(0) = d1(0)e1-d2(0)e2, system. The initial state is Φ(0) = d1(0)e1-d2(0)e2,
and the final state Φ(t) = d1(t)e1-d2(t)e2. and the final state Φ(t) = d1(t)e1-d2(t)e2.
3
American Institute of Aeronautics and Astronautics
where, lc is the non-dimensional distance between the duct end and the flame location, Ti is the inlet
temperature, X i and Yi are the non-dimensional oxidizer and fuel mass fractions (with respect to the
stoichiometric values) at the inlet respectively. Since the combustion zone is much smaller compared to the
acoustic length scale, the outflow condition can be applied at infinity5. However, in the present paper, the
outflow conditions are applied at the duct end. The above boundary conditions must be supplemented by an
initial condition. In this paper it is assumed that a steady Burke Schumann flame is perturbed initially which
leads to unsteady combustion oscillations. It is possible to model the flame using other physical boundary
conditions such as flux boundary condition which was used in Ref. 5. In the present analysis, the boundary
conditions used by Chung and Law13 are used.
The above equation is solved using Galerkin technique14 and the solution can be written as,
Z=
∑∑ m n
πx
An cos(nπ y ) sin((m + 1/ 2) )Gmn (t ) + Z st
lc
(2)
where, Z st is the solution of steady Burke-Schumann equation. Galerkin technique makes use of the fact that
any function in a domain can be expressed a superposition of modal functions which form a complete basis in
that domain. The modal functions are chosen such that they satisfy the boundary conditions. However the choice
of the mode functions is not unique. It is convenient to choose an orthonormal basis though it is not necessary. It
should be emphasized that the mode functions chosen here are just an arbitrary basis and not the eigenfunctions
of the system. Clearly the mode functions chosen here satisfy the boundary conditions and they form a complete
basis. Substituting the expression for Z in (1), and integrating over the domain after multiplying by the basis
functions, the following evolution equation for Gmn (t ) is obtained,
∑
G m( n ) + u (t )
k
Wmk Gk( n ) = −
(m + 1/ 2) 2 π 2 ( n )
2
lc Pe
Gm −
n 2π 2 ( n )
Pe
Gm + [u (t ) − 1] Cm( n ) (3)
where,
lc
Wmk =
∫0
sin [ (m + 1/ 2)π x / lc ] cos [ (k + 1/ 2)π x / lc ] dx (4)
The orthogonal property of the trigonometric functions has been used to obtain the above equations. It can be
seen from (3) that the evolution equations for the amplitude of the mth mode function depends on the amplitude
of the nth mode. Further, it can be seen from Eq. (4) that the dependence of the amplitude of mth mode on the
amplitude of kth mode is different from the dependence of kth mode on the mth mode. In other words, matrix W
does not commute with its adjoint. Hence the evolution equations are non-normal. As explained in section IV,
this will lead to transient energy growth. As mentioned in section I, the eigenvectors of non-normal matrices are
not orthogonal and hence the amplitude of each mode depends on the amplitude of other modes. Equation (3)
can be written as a first order matrix differential equation by considering only a finite number of modes as
follows:
G ( n ) + u (t )[W ]G ( n ) = −[ D]( n ) G ( n ) − (u (t ) -1)C ( n ) (5)
where,
G ( n ) = ⎡⎣G1n G2n G3n ................GMn ⎤⎦
T
1 (6)
[ D]nmm = − diagnol ((m + 1/ 2) 2 π 2 / lc2 + n 2π 2 )
Pe
and the elements of the matrix W is shown in Eq. (4). The first order matrix differential equation can be solved
analytically if the velocity profile at the flame is prescribed. The solution can be written as,
t
G ( n ) (t ) = G ( n ) (0) − IF −1 (t )
∫
0
IF (t ′)C ( n ) (u (t ′) − 1)dt ′ (7)
G ( n ) (0) = 0 for an initially steady flame. IF(t) is the integrating factor of the first order matrix differential
equation
∫
⎡ ⎤
IF (t ) = exp ⎢[ D]t ′ + [W ] u (t )dt ⎥
⎣⎢ ⎦⎥
4
American Institute of Aeronautics and Astronautics
It can be inferred from the solution that Z has exponential dependence on the amplitude of acoustic
perturbations and hence the combustion response to external velocity perturbations is highly nonlinear even
though the evolution equations are linear. In a self-excited system where the combustion oscillations and
acoustic oscillations are coupled, the evolution equations themselves are nonlinear, as shown in section III. The
evolution equations when linearized yield non-modal solutions (i.e., non-orthogonal eigenvectors) which leads
to transient growth. The coupled response will be discussed in detail in the next section.
It is necessary to calculate the heat release rate to study the coupled response and it can be done as follows.
The Burke Schumann temperature field (nondimensionalised by QH / C p ) is given by 5:
Tbs = Ti + X i (Yi + Z ) ( X i + Yi ) Z ≤ 0
(8)
Tbs = Ti + Yi ( X i − Z ) ( X i + Yi ) Z ≥ 0
where, QH is the heating value of fuel per unit mass of the mixture. The above expression for temperature is an
exact solution of the energy equation in the infinite reaction rate assumption if the pressure oscillations are
negligible. This happens when the Mach number of the mean flow is very low. The heat release rate can then be
calculated from the temperature field using thermodynamic relations5. The heat release rate (nondimensionalised
by heating value of fuel ) is given by,
∫ ∫
⎛ dT G G⎞ ⎛ ∂T G G ⎞
Q c = ⎜ bs + Tbs ∇.u ⎟ dV = ⎜ bs + ∇.(Tbs u ) ⎟ dV (9)
⎝ dt ⎠ ⎝ ∂t ⎠
V V
Since Z has exponential dependence on the acoustic perturbations, the heat release rate also has exponential
dependence on the acoustic perturbations. As shown earlier, the combustion modes are non-normal, leading to
the interaction between combustion modes. Since the combustion process interacts with the acoustic field, the
above mentioned interaction between combustion modes will lead to interaction between acoustic modes. The
interaction process is further complicated if nonlinear acoustics is considered. To focus the attention on the non-
normal and nonlinear behavior of the combustion response, the authors have modeled the acoustic field using a
one-dimensional linear acoustic model as in Ref. 5.
III. Interaction between the Combustion Process and the Acoustic Field
The combustion model discussed above has been coupled with a one-dimensional linear acoustic model to
examine the growth of the acoustic pressure, velocity and the heat release fluctuations. Neglecting the effect of
mean flow and mean temperature gradient in the duct, the governing equations for the one-dimensional acoustic
field are5:
Acoustic Momentum:
∂u ′ ∂p '
ρ + =0 (10)
∂t ∂x
Acoustic Energy:
∂p ′ ∂u ′
+ = ( γ − 1) Q ′ (11)
∂t ∂x
where, p ′ is the acoustic pressure, u ′ is the acoustic velocity and, Q ′ the unsteady heat release rate. The
expression for Q ′ can be related to Q c′ through dimensional constants as Q ′ = ρ c pTref ( u0 / H ) Q c′ .
The above equations can be non-dimensionalized as follows
L u
x = La x ; t = a t ; u ′ = u0 u ′ ; p ′ = ρ0 u0 p ′ ; M = 0 ; ρ = ρ0 ρ
c0 c0
where, La is the length of the duct, the subscript ‘0’ indicates quantities in the undisturbed medium. The duct is
assumed to be open at both ends and hence the acoustic pressure vanishes at the ends. The unsteady heat release
is assumed to be a compact source, as the ratio of combustion length scale to the acoustic length scale is small5.
Equations (10)-(11) can be rewritten as,
∂u ′ ∂p ′
γM + =0 (12)
∂t ∂x
L Q H δ ( x − x f )
∂p ′ ∂u ′
+γM = (γ − 1) a av 2 (13)
∂t ∂x c0 ρ0 c0 La
5
American Institute of Aeronautics and Astronautics
where, Q av is the averaged heat release over the combustion zone, and xf is the flame location.
The acoustic field is solved in time domain using the Galerkin technique15, 16. The acoustic pressure and
acoustic velocity can be written as:
∑ ∑
N N
γM
u′ = η j cos( jπ x) and p ′ = η j sin( jπ x ) (14)
jπ
j =1 j =1
The velocity used in Eq. (1) is the sum of acoustic velocity evaluated at the flame location x f and the base flow.
The mode functions chosen for the acoustic variables satisfy the boundary conditions and form a complete basis
as well. The orthogonal property of trigonometric function makes the analysis simpler. The following evolution
equations for the acoustic field are obtained after substituting the expressions for acoustic velocity and pressure
in the linear acoustic equations and integrating over the duct (domain for the acoustic field) after multiplying by
the basis functions,
dη j
= η j (15a)
dt
dη j 2k j Q av 1
+ k 2
η = − sin( jπ x f ) (16a)
(Ti + Tad ) / 2 γ
j j
dt 2
where Ti is the inlet temperature, Tad is the adiabatic flame temperature and kj is the wave number corresponding
to the jth mode.
Further, the role of damping on the growth and saturation of oscillations is also studied. The effect of
damping has been included in the linear acoustic equations. The damping model used in this paper is same as
that in Ref. 17. In the presence of damping the above set of equations can be modified as follows17:
dη j
= η j (15b)
dt
dη j 2k j Q av 1
+ 2ξ j ω jη j + k 2j η j = − sin( jπ x f ) (16b)
dt 2
(Ti + Tad ) / 2 γ
where, ω j is the wave number of the jth mode, ω1 is the wave number of the first mode. Damping is higher for
higher frequencies and hence the higher frequencies in the oscillations decay rapidly, when there is no
mechanism to drive the higher frequencies. Damping occurs due to acoustic boundary layer losses, which can
modeled as a volumetric source term, due to sound radiation losses at the ends and convection of sound by the
mean flow. The heat release rate can be written in matrix form using Eq. (9) as,
Q av =
∑∑ dGnm
dt ∑∑
( Rnm ) + u (t )Gnm J nm + Qst ( u (t ) − 1)
(18)
dG
= [ R] + u (t ) [ J ] G + Qst ( u (t ) − 1)
dt
where,
G = ⎡⎣G (1) G (2) G (3) ...G ( n ) ...G ( N ) ⎤⎦
T
lc 1
∫∫
⎡ Yi X i ⎤ ⎛ m + 1/ 2 ⎞
Rnm = ⎢θ (− Z ) − θ (Z ) ⎥ sin ⎜ π x ⎟ cos ( nπ y ) dxdy
⎣ X i + Y1 X i + Y1 ⎦ ⎝ lc ⎠
0 −1
1
∫
⎡ Yi Xi ⎤
J nm = ⎢θ (− Z (lc , y )) − θ ( Z (lc , y )) ⎥ (−1) cos ( nπ y ) dxdy
m
⎣ X i + Y1 X i + Y1 ⎦
−1
The expanded form of the matrices are presented in Appendix (1). The above system of equations is solved
using Runge-Kutta method. In the above equation BNL is the matrix which leads to nonlinear interaction and
BNN is the matrix which leads to non-normal growth of the oscillations. The above equation has the same form
as that of the evolution equations discussed in Ref. 6, 9 & 10 and hence it is expected that the above system
might show similar behavior as discussed in these references. Further, due to the non-normal and nonlinear
nature of the system, it is expected that the system will exhibit chaotic and fractal behavior.
Equation (19) when linearized yields:
dχ dχ
M + BNN χ = 0 or + L χ = 0 , where L = M-1BNN (20)
dt dt
In the above equation L is called the stability matrix12 and it is purely non-normal. When the eigenvalues of
matrix L are stable, then the oscillations of the linearized system eventually decay. However, due to the non-
orthogonality of the eigenvectors, it is possible for the linearized system to exhibit large transient growths
triggering the nonlinearities as shown in Fig. 1. In the next section, we will use Eq. (20) to study the transient
growth of the above system of equations to analyze the stability of combustion-acoustic interaction.
The thermo-acoustic system described in section III has a non-normal evolution, which leads to transient
energy growth. As discussed earlier, transient growth plays an important role in amplifying the initial
disturbances to a high enough value where nonlinear effects can play a significant role. Hence it is necessary to
identify the initial conditions for which the transient growth is maximum. Schmidt and Henningson12 gives a
detailed discussion on the analysis of transient growth, in the context of transition to turbulence in shear flows.
They analyzed the stability of shear flows by studying the energy growth of the system. However, their analysis
is quite general and can be applied to thermoacoustic systems as well. In this analysis, the initial disturbances
which maximize the growth at a particular instant of time is determined12. Further, in this analysis the
development of a general solution of the linearized system is considered rather than the individual eigenmodes
of the system. The evolution operator (matrix) of the coupled combustion-acoustic system described by Eq. (20)
is non-normal as, LL† ≠ L† L . The solution of the system of equations in (20) can be written in the operator form
as follows12:
η (t ) = exp(− Lt )η (0) = S −1 exp(− L D t ) Sη (0) (21)
where, L is the stability operator, S is the similarity transformation that diagonalizes L, LD is the diagonal form
of L. Since L is non-normal, S is non-unitary indicating that it is not possible to make the eigenvectors
perpendicular by a simple rotation or by a different choice of basis functions. The maximum amplification of the
energy density is defined as12,
η (t )
2
G (t ) = max = exp(− Lt )
2
(22)
η (0)
η 2
The initial condition that maximizes the growth factor is different for different times and hence the maximum
growth (corresponding to this initial condition) is the envelope of the energy evolution of individual initial
conditions12. The expression in (22) is maximized for various instants of times over all possible initial
conditions. This is achieved through singular value decomposition. The initial condition corresponding to the
maximum singular value is chosen as the initial condition leading to maximum growth rate. The maximum
growth function in a particular time interval [0, t] is defined as:
Gmax = max G (t )
t
If there exists an unstable eigenvalue, then the above maximum value is infinite. This corresponds to the linearly
unstable system. The Gmax values for various parameters are calculated and the region with large transient
growth is identified. The Gmax = 1 contour separates parameter combinations for which transient growth may
occur from ones for which energy decay occurs. When Gmax = 1 , then the energy at any instant is less than the
initial energy of the system. Hence, a system which is almost linear initially will behave like a linear system
7
American Institute of Aeronautics and Astronautics
throughout its evolution as there is no amplification of the oscillations to trigger nonlinear effects, when
Gmax = 1 . This enables us to study the range of parameters for which the system is linearly stable, linearly
unstable and region where transient growth is significant. The dependence on various parameters was
systematically studied and physical arguments based on the analysis are presented in the next section.
The acoustic modes are normal in the absence of combustion. However, the presence of combustion makes
the system non-normal as can be seen from the following physical arguments. A small disturbance in velocity
causes the combustion process to become unsteady and it in turn acts as the source of acoustic oscillations. The
acoustic field in a duct is driven by combustion when the combustion oscillations are in phase with the acoustic
oscillations. The phase difference is due to a time lag between the two processes; this time lag itself varies with
time and hence the phase difference between the combustion and the acoustic oscillations depends on the phase
difference between the two processes at an earlier time. Combustion drives that mode of acoustics which is in
phase with the combustion process. Since the phase difference between the combustion and particular mode of
the acoustic field depends on the phase difference at an earlier time, the interaction would depend on that mode
of the acoustic field which was in phase with combustion at an earlier time. Hence the mode which is excited at
a particular instant of time depends on which mode got excited at an earlier time. In a general situation, this
would lead to a complicated interaction between the various modes which would lead to non-orthogonal
behavior of the eigenmodes. This is the defining characteristic of non-normal systems.
The non-normality of combustion process is present even in the absence of the feedback from the acoustic
field. This is because the non-normality in combustion process is due to convection which transports heat from a
“hot” region to a region which is colder. This process is different from diffusion as diffusion is a gradual process
in which the direction of transport is the same as the direction of the thermal gradient unlike convection in
which it is the direction of the flow. Hence the base flow convects heat from a hot region to a cold region where
the relative amplitude of heat release oscillations increases. This causes interaction among various combustion
modes. Hence, it can be seen that the eigenmodes are non-orthogonal in this case as well, as two different
oscillatory modes, i.e., the oscillatory conduction and the oscillatory convection, are interdependent. This
behavior is analogous to the kind of behavior in vortical flows that leads to formation of streaks6 at the onset of
transition. The feature discussed above can also be thought of an entropy wave as entropy gets convected with
the flow.
⎡ ∂s ∂s ⎤ 1 ⎡ ∂2 s ∂2 s ⎤
T ⎢ + u (t ) ⎥ = − ⎢ + ⎥ + Qδ ( Z ) (23)
⎣ ∂t ∂x ⎦ Pe ⎣ ∂x 2 ∂y 2 ⎦
The flame sheet acts as a localized source of entropy disturbance and Eq. (23) describes the propagation of this
disturbance to other regions in space. The entropy disturbance propagates with the flow velocity and also
diffuses to other regions. This movement of entropy disturbance is similar to that of the advection of vorticity
disturbance discussed in Ref. 9. It has been discussed in Ref. 9 that such a movement causes the occurrence of
structures such as streaks. The advection of disturbance causes the various eigenmodes to be interrelated and
this causes the system to be non-normal. Hence the non-normality of the combustion process is essentially due
to the movement of entropy disturbances from regions of high entropy to low entropy.
The variation of heat release rate and the acoustic velocity at the flame location for different choices of the
parameters are presented in the following subsections and their effects on the combustion acoustic system is
discussed in detail. Test numerical simulations were performed for different choices for the number of acoustic
and combustion modes used in the Galerkin technique and the number of modes was chosen (for subsequent
simulations) such that the solution does not change significantly if additional modes are introduced. It was
observed that even 900 combustion modes (30 for y dependence and 30 for x dependence) and 4 acoustic modes
are sufficient. To be on the conservative side, 2500 combustion modes (50 for the y dependence and 50 for x
dependence) and 6 acoustic modes were taken for the calculations presented subsequently.
A. Triggering of Nonlinearities
If the system is non-normal as well as nonlinear, then there exists some initial condition for which the
oscillation decays and some initial conditions for which the oscillations grow due to nonlinear triggering. In this
example, this feature is captured by a flame located at ¾ duct length when the fuel slot width is 0.25. The Peclet
number was chosen as 5.0, X i was chosen as 3.2, Yi was chosen as 3.2/7 and the duct length to duct width ratio
was chosen as 25.0. For reasons that will be clear later, damping is assumed to be absent in this example. The
choice of initial conditions is shown in the figure caption. Figures 5 a and b show that there are several periods
of short time growth, even though the oscillation decays. This clearly shows that transient growth is not
8
American Institute of Aeronautics and Astronautics
sufficient to trigger the nonlinearities. Figure 5c shows the variation of heat release rate with velocity also
known as the phase portrait. It is
clear that the phase portrait is
elliptically spiraling inwards, which
indicates that the response of heat
release rate to velocity fluctuations
is almost linear. The elliptical nature
also indicates that there is a phase
difference between the heat release
rate and the velocity perturbations.
Figure 6a clearly shows that for
the same thermoacoustic system, for
a different initial condition, the
oscillation grows and saturates. This
is due to the fact that transient
growth causes the amplitudes to
increase to a value sufficient to
cause nonlinearity. It is clear from
the heat release rate oscillations in
Fig. 6b that after some time, the
mean value of heat release rate
changes with time, clearly indicating
nonlinear behavior. The nonlinear
response of the heat release rate can
also be clearly seen from the fractal
behavior exhibited by the phase
portrait. Such fractal behavior,
known as folds, has been discussed
in the context of premixed flames18.
In Ref. 18, the effect of time delay
on the system is studied and it can
be seen that the stroboscopic map
shows Rossler bands. This fold
behavior can be observed when the
attractor is similar to that of a
Lorentz attractor. Further the fractal
behavior indicates that the system is
highly sensitive to initial conditions.
Another interesting feature shown
by the system is saturation in the
absence of damping. It is the
accepted notion that when both
driving and damping forces are
present, then the amplitudes can
saturate when driving balances
damping2. However, this example
clearly shows that saturation is
possible even in the absence of
damping. Saturation can occur when
the phase difference between the
pressure oscillations and the heat
release rate becomes 90 degrees.
When the heat release rate and
pressure oscillations differ by a
Figure (5) shows (a) the evolution of acoustic velocity (b) heat phase of 90 degrees then at that time
release and (c) phase portrait at the flame location for a case instant the growth rate of energy is
where the initial condition corresponds to excitation of the zero and hence further growth is not
first duct mode; η1 (0) = 0.1 ; xf=3/4; Pe= 5.0, X i = 3.2, Yi = possible. Hence when the phase
3.2/7, La/(2H)= 25. difference becomes 90 degrees and
if it remains a constant then the
9
American Institute of Aeronautics and Astronautics
oscillations will saturate. This can be inferred from Fig. 6d which shows that the phase difference is 90 degrees
when the oscillation saturates. Saturation of the oscillations also indicates the existence of an attractor (as the
system tends towards a limit cycle oscillation). The fractal dimension of the attractor for this case is found to be
1.68. The fractal dimension is obtained using box counting technique. There are various definitions of fractal
dimension. In this paper we have referred to the box-counting dimension as the dimension of the fractal. The
definition of box-counting dimension is explained in Appendix 2.
Figure (6) shows (a) evolution of acoustic velocity at the flame location (b) heat release evolution and (c)
phase portrait for a case with the the initial condition η1 (0) = 0.5,η1 (0) = 0.5 ; Figure (6d) shows that phase
difference between heat release and acoustic pressure is 90 degrees when the amplitude saturates. Xf = 3/4
Pe = 5.0, X i = 3.2, Yi = 3.2/7 , La/2H= 25.
Wicker et al.19 defines triggering combustion instability in the context of their study of combustion
instabilities in rocket motors as “initiation of unstable pressure oscillations by a finite amplitude pulse in a
system that is otherwise stable to small perturbations”. Recent studies have attributed triggering to nonlinear
combustion19-21. These studies include the effect of nonlinear acoustics as well, although they state that
nonlinear acoustics alone is insufficient to cause triggering. The present study highlights the role of non-
normality in the occurrence of triggering combustion instabilities. The transient growth arising due to the non-
normality of the thermoacoustic system triggers nonlinearities when the amplitude reaches high enough values,
resulting in further growth of the oscillations.
10
American Institute of Aeronautics and Astronautics
t < 30 and 30 < t < 60 respectively. It is clear that the dominant frequency during the first half dies and is absent
Figure (7a) shows the evolution of acoustic velocity at the flame location with time. It can be seen that
there is transfer of energy between various modes. It can be seen that the excited mode decays initially
and there is growth of higher modes. Figure (7b & c) show the heat release rate spectra for the first and
second half of the evolution, respectively. The dominant non-dimensional frequency is 1.38, which
interestingly does not correspond to any natural frequency of the duct. Figures (7d-f) show that the while
the first mode decays, the second and third mode grows. It can also be seen that the first mode starts
growing again after some time. The amplitudes of the higher modes are negligible. The initial condition
corresponds to excitation of the first duct mode; η1 (0) = 0.1,η1 (0) = 0.1 . La/2H = 12.5.
in the second half of the evolution. It can be seen that there are several peaks in the second half of the evolution
indicating the generation of higher frequencies. It can be seen that the dominant non-dimensional frequency is
1.38 in the second half of the evolution. Interestingly this frequency does not correspond to any natural
frequency of the duct. This feature of combustion instabilities occurring at frequencies that are far from the
11
American Institute of Aeronautics and Astronautics
natural acoustic freqeuncies was discussed in the context of dump combustors by Matveev and Culick15, 22 and
experimentally observed by Schadow and Gutmark23. Figures 7d-f show the evolution of the first three modes. It
is known that for a non-normal system “pseudoresonance” may occur at frequencies far from the spectrum9, 28. It
can be seen that the while the first mode decays, the second and third mode grows. It can also be seen that the
first mode starts growing again after some time. This indicates that the there is transfer of energy from the first
mode to the second and third modes causing the first (excited) mode to decay initially. After the second and
third mode gains sufficient energy from the excited mode, they transfer back some energy to the first mode and
hence the first mode starts growing again. The net effect of all these energy transfer causes the acoustic velocity
to saturate eventually. This feature has been discussed in the context of turbulence and it is known as
“bootstrapping”10, 24. Yoon et al.25 has discussed “bootstrapping” in the context of Rijke tube using an ad hoc
nonlinear model for the heat release rate. However, their ad hoc model does not predict saturation.
The occurrence of harmonic frequency generated by oscillating flames was experimentally observed by
Lang26. The generation of higher harmonics is usually completely attributed to the nonlinear nature of the
interaction. However it is found that non-normality plays a significant role in this interaction by transferring
energy from one mode to another.
Figure (8a) shows the evolution of acoustic velocity (b) heat release (c) phase portrait for a flame
which is located at ¼ duct length. The ellipse gradually distorts as the amplitude increases and
phase portrait indicates a strong nonlinear behavior at higher amplitudes. Pe = 10.0, α = 0.1.
La/2H= 12.5 The initial condition corresponds to excitation of the first duct mode; η1 (0) = 0.1 .
Figure (8d) shows the phase difference between pressure oscillations and heat release rate
12
American Institute of Aeronautics and Astronautics
The effect of flame location can be studied by changing the flame location and keeping the other parameters
fixed. In the following, the nature of thermoacoustic oscillations for a flame located at ¾ duct is studied. Figures
9 a and b show that the acoustic and heat release rate oscillations decay with time. However, the response of the
heat release rate oscillation to velocity is not linear. This is clear from the phase portrait shown in Fig. 9c. The
phase portrait indicates a chaotic behavior. Unlike the previous situation, the response is nonlinear even at “low”
amplitudes. Figure 9d shows that the phase difference between the pressure oscillations and the heat release rate
oscillations is such that the energy growth rate is negative.
Figure (9a) shows the acoustic velocity evolution (b) heat release (c) phase portrait for a flame
located at 3/4th duct decays with timeFigure (9d) shows the phase difference between pressure and
heat release rate oscillations. Pe = 10.0, α = 0.1. The initial condition corresponds to excitation of
the first duct mode; η1 (0) = 0.1 . La/2H= 12.5.
The key difference between the two cases is that in the first case the initial condition is such that the
acoustic perturbation in one case begins with the positive cycle and with the negative cycle in the other case.
Hence in one case the perturbation adds to the base flow and in the other case has a direction opposite to the
base flow. In Fig. 8d the phase difference between the pressure oscillation and the heat release rate is such that
the acoustic energy of the system increases. Whereas, in Fig. 9, the initial acoustic energy of the system
decreases as the phase difference is greater than 90 degrees.
The effect of Peclet number on the interaction between the combustion and acoustic processes is studied in
this example. Figures 8 and 10 shows the variation of heat release rate and the acoustic velocities for different
Peclet numbers. The flame is located at quarter duct for both the case. Figure 10a shows the variation of acoustic
velocity with time for Peclet number of 5. It is clear that the acoustic velocity increases with time. Figure 10b
shows the corresponding heat release rate and Figure 10c shows the corresponding phase portrait of the heat
release and the acoustic velocity. The oscillations are linearly unstable which can be clearly inferred from Figs.
10 a, b and the phase portrait in Fig. 10c. A linearly unstable system has an infinite growth factor. However, the
situation is completely different when the Peclet number is 10.0 (higher). Figures 8 a and b that the
thermoacoustic oscillations are unstable in this case as well. However, it can be seen from Figures 8b and the
phase portrait in Figure (8c) that the growth is non-normal as well as nonlinear, unlike the previous case.
13
American Institute of Aeronautics and Astronautics
Figure (11a) and (11b) show the evolution of
Figure (10a) shows the evolution of acoustic the acoustic velocity and heat release rate.
velocity (b) heat release (c) phase portrait for a Figure (11c) shows that the phase portrait. α
flame located at ¼ duct; Pe = 10.0, α = 0.1, La/2H = 0.25; La/2H = 12.5. Other conditions are
= 12.5. The initial condition corresponds to same as that of Figure (9).
excitation of the first duct mode; η1 (0) = 0.1 .
14
American Institute of Aeronautics and Astronautics
The difference in behavior for different Peclet numbers, can be understood from the fact that the non-
normal nature of the system becomes significant at higher Peclet numbers as the rate at which heat is transported
from “hot region” to a colder region increases with increase in convective velocity. Hence the heat change at a
colder region occurs very rapidly and it can occur at very high frequencies which will gradually reduce to a
lower frequency. Hence more number of modes get interdependent as the Peclet number increases.
In this example the effect of slot width on the combustion acoustic interaction is studied. The fuel slot width
was chosen to be 0.25. All other conditions and parameters are same as that of Fig. 9. It can be seen from Figs.
11 a and b that the acoustic velocity and heat release rate oscillations grow with time. However this growth is
almost a linear growth which can be observed from the elliptical nature of phase portrait shown in Fig. 11c. It
should be noted that Fig. 9c (fuel slot width of 0.1) showed a nonlinear behavior, whereas the present case
shows almost linear behavior. Hence the thermoacoustic system is highly sensitive to changes in fuel slot width.
The effect of slot width in the context of damping is studied in a later example.
The duct length is a measure of the acoustic length scale whereas the duct width is a measure of the
combustion length scale. The ratio of duct length to combustion length is found to be an important parameter in
the combustion acoustic interaction. The importance of this parameter has been discussed in Ref. 5. If the Mach
number is fixed, then this ratio is a measure of the acoustic time scale to combustion time scale. Figures 12a and
b show the evolution of the acoustic velocity and heat release rate with time respectively for a duct length to
duct width ratio of 25.0. All other conditions and parameters are same as that of the previous example. The
oscillations decay exponentially in the initial stages indicating a linear behavior. It can be seen that there is
transient growth of the oscillations after it decays. It can also be observed that the higher harmonics start
dominating after sometime. The evolution clearly indicates a possibility of chaotic behaviour, which can be
studied from the phase portrait. In the phase portrait shown in Fig. 12c it can be observed that the trajectory
shifts to another ellipse which is different from the initial elliptical trajectory and eventually the system becomes
chaotic. Daw et al.27 studied chaos in thermal pulse combustion. They attributed to the onset of chaos in model
combustors to period doubling bifurcation. It can be seen that in Fig. 9 there is no transient growth, whereas Fig.
13 shows that there is transient growth of oscillations. Hence, it can be inferred that the non-normal nature of the
evolution equations is sensitive to changes in the duct length to duct width ratio.
G. Effect of Damping
Although damping was neglected in section VA to show that saturation is possible even in its absence,
realistic modeling of thermoacoustic systems require damping to be incorporated. In this section, the effect of
damping on thermoacoustic oscillations is investigated.
First, the evolution of oscillations in an undamped system is studied. Figures 8 a and b show the acoustic
oscillations and heat release rate oscillation for a flame located at ¼ duct length when the slot width is 0.1. It is
seen that the oscillations in this undamped system grow indefinitely. Figure 8c shows the variation of heat
release with velocity (phase portrait). It is clear that the heat release response becomes nonlinear as the velocity
amplitude increases. However the situation is different when damping is introduced in an unstable undamped
system.
When damping is introduced in the same system, the amplitude of the oscillation saturates. This behavior is
as expected and it is clearly seen in Fig. 13a. Figure 13b shows the heat release rate spectra. It is clear that the
higher modes are not completely damped. This behavior is counter intuitive as one would expect the high
frequencies to damp much faster than the lower frequencies. This could be the result of nonlinear driving as well
as non-normality. However, the role played by each of this mechanism is not yet clearly understood. Figure 13c
shows the variation of phase difference with time. It is clear that the phase difference is not 90 degrees when the
amplitudes saturate. As we approach the limit cycle the phase difference tends to a value which will result in
just adequate driving to balance the damping in the system. Though damping is not essential for saturation to
occur, damping plays an important role in the evolution of the perturbation. The above examples show how
damping can cause oscillations to saturate.
15
American Institute of Aeronautics and Astronautics
Figure (13a) shows the evolution of acoustic
Figure (12a) and (12b) show the evolution of the velocity (b) heat release spectra (c) phase for
acoustic velocity and heat release rate with time a flame which is located at ¼ duct length in
respectively. Figure (12c) shows the trajectory the presence of damping. All other conditions
shifts to another ellipse which is different from are same as the previous case. The initial
the initial elliptical trajectory and eventually the condition corresponds to excitation of the first
system becomes chaotic. α = 0.25. La/2H = 25.0. duct mode; η1 (0) = 0.1 . La/2H = 12.5
All other conditions are same as in the previous
example (Figure 11).
16
American Institute of Aeronautics and Astronautics
In the following example, one of the parameters is changed to show that the effect of damping depends on
the various parameters of the system. The fuel slot width is changed in the following example to show a case
where the oscillations decay to zero rapidly. Figures 14 a and b show that the oscillations are completely
damped. The flame was located at ¼ duct length. The fuel slot width was chosen as 0.25. The Peclet number
was chosen to be 10.0, X i was chosen as 3.2/7, Yi and was chosen as 3.2. The phase portrait shown in Fig.
(14c) is an ellipse spiralling inwards which indicates that the system is stable. Figure 13 shows a highly
nonlinear behavior in the presence of damping where as the present case shows a linear behavior. This
difference in the behavior is due to the high sensitivity of the system to various parameters.
The effect of various parameters on the combustion-acoustic interaction has been investigated in the above
examples. It has been found that the combustion acoustic interaction is highly sensitive to initial conditions as
well as various geometric parameters such as duct length to duct width ratio, slot width, flame location, etc and
various non-dimensional parameters such as Peclet number, equivalence ratio etc.
H. Transient Growth
In this example the variation of growth factor with duct length to duct width ratio is studied. This ratio is
found to be critical in combustion instability5. Figure 15 shows the variation of the maximum growth factor with
duct length to duct width ratio for different Peclet numbers, for a flame that is located at quarter duct. It is
observed that the growth factor increases monotonically. This implies that the growth factor becomes larger as
the flame becomes more compact. The heat release rate is considered to be a point source in our analysis. The
strength of this point source increases as the flame becomes more and more compact. This causes the flame to
be a stronger source for the acoustics. Hence growth factor is larger for a compact flame.
I. Pseudospectra
As demonstrated in the earlier sections, the thermo-acoustic acoustic oscillations exhibit a significant
sensitivity to small changes. The sensitivity is caused by the presence of non-orthogonal eigenvectors. This
determines the phenomenon of sensitivity as a property of the operator which is non-normal. Trefthen and
Embree28 introduced the concept of ε−pseudoeigenvalues to analyze the behavior of evolution governed by non-
normal operators. Pseudospectra can be defined as follows:
( zI − A)
−1
z is an ε -pseudo eigenvalue of A if it satisfies ≥ ε −1 .
There are other equivalent definitions of pseudo eigenvalues and they have been discussed in great detail in Ref.
28. The pseudospectra of normal operators are closed circles. Figure 16 shows the pseudo-spectra of a thermo-
acoustic system. The non circular nature of the pseudo spectra clearly indicates the non-normal nature of the
system. Further the non-normality becomes more pronounced in the right half plane3. This feature is also
observed in the case of Papkovich-Fadle operator, where it was attributed to a singularity in the problem28.
If a linear operator is normal, then the degree of resonant amplification that may occur in response to an
input frequency is proportional to the distance in the complex plane between the input frequency and nearest
eigenvalues. However, for a nonnormal operator the resonant amplifications may be orders of magnitude
greater. The resonances of a non-normal system are not determined by the eigenvalues alone. Such a resonance
of a non-normal system is known as pseudoresonance.
3
The authors use the stability matrix in Eq. (20) for obtaining the pseudospectra. However, the sign convention
for the operator used in this paper is opposite to that of Ref. 28. Hence there is an inversion in the Re(z) axis.
17
American Institute of Aeronautics and Astronautics
Figure 15 shows the variation of maximum growth
factor with duct length to duct width ratio for
different Peclet numbers. The growth factor
increases monotonically as duct length to duct
width ratio. ‘o’ denotes Pe = 10 and ‘+’ denotes Pe=
5. α = 0.25; xf=1/4; Xi = 3.2/7, Yi = 3.2.
18
American Institute of Aeronautics and Astronautics
VI. Conclusion
In this paper the nature of flame acoustic interaction is studied in the context of diffusion flames. It has been
observed that the various duct acoustic modes interact resulting in the exchange of energy between various
modes. This is due to the fact that the duct modes become non-normal in the presence of the flame and hence
the eigenvectors are non-orthogonal. Non-normality also leads to short time amplification even though the
individual modes decay exponentially. It is found that nonlinearity is triggered due to this transient growth and
this causes the different modes to mix. Non-normality also leads to “bootstrapping” where there is exchange of
energy between various modes causing one mode to decay when another mode is growing. In literature the
generation of higher harmonics was completely attributed to the nonlinear nature of the interaction. However it
is found that non-normality plays a significant role in this interaction by transferring energy from one mode to
another. This complicated interaction leads to interesting dynamical behavior such as chaos and fractals.
Further, it was found that saturation can occur even in the absence of damping; as the system approaches limit
cycle, the phase difference between the pressure fluctuations and heat release rate oscillations becomes 90
degrees. It has been found that the stability of the system is highly sensitive to various geometric parameters
such as duct length to duct width ratio, slot width, flame location, etc and various non-dimensional parameters
such as Peclet number, equivalence ratio, non-dimensional inlet temperature, etc. The flame location also plays
an important role in altering the stability of the system. The interaction is highly sensitive to various parameters
and the initial conditions and can result in chaos under some conditions.
The current methodology to study the onset of thermoacoustic oscillations involves looking for
exponentially growing or decaying modes by calculating the individual eigenvalues of the linearized system.
Further, the nonlinear behaviour of the combustion response is modeled using flame transfer functions which
are amplitude dependent. These approaches fail to predict phenomena such as nonlinear growth triggered by the
transient growth which results from non-normality of the thermoacoustic system, shift to higher frequencies
observed in the presence of damping when the amplitude saturates and excitation of modes that are not initially
excited.
For a non-normal system, “pseudoresonance” is shown to occur at frequencies far from the spectrum. The
stability and sensitivity of non-normal systems can be studied using the pseudospectra. The pseudospectra of
normal operators are disjoint circles. The pseudospectra of the thermoacoustic system are non-circular implying
a highly non-normal nature of the system.
The motivation behind this paper is to illuminate the role of various mechanisms in combustion instability
in conjunction with non-normality which has never been considered in the earlier study of combustion
instability. Most models that are currently used are linear and even the non-linear models are ad-hoc. The
understanding of combustion instability can be improved by building physics based models which incorporate
the various effects mentioned in this paper.
Acknowledgement
The authors wish to thank Prof. Wolfgang Polifke (Technishe Universität München), Srevatsan
Muralidharan (IIT Madras, currently at Princeton University) and Prof. Rama Govindarajan (Jawaharlal Nehru
Centre for Advanced Scientific Research, Bangalore) for their suggestions and interesting discussions.
References
[1] Mcmanus, K., Poinsot, T. and Candel, S. M. “A Review of Active Control of Combustion Instabilities”,
Progress in Energy and Combustion Sciences, Vol. 19, 1993, pp.1–29.
[2] Lieuwen, T., “Modeling Premixed Combustion-Acoustic Wave Interactions: A Review.” Journal of
Propulsion and Power Vol. 19, No. 5, 2003, pp. 765-781.
[3] Vance, R., Miklavcic, R., Wichman, I. S. “On the Stability of One-Dimensional Diffusion Flames
Established Between Plane, Parallel, Porous Walls,” Combust. Theory Modelling, Vol. 5, 2001, 147.
[4] Chakravarthy, S.R., Jamadar, N.M., Tyagi, M. International Symposium on Recent Advance in Aero
Acoustics and Active Flow combustion Control, Goa, India, 2005.
[5] Tyagi, M., Chakravarthy. S. R. and Sujith, R. I. "Unsteady Response of a Ducted Non-Premixed Flame and
Acoustic Coupling", accepted for publication in Combustion Theory and Modeling, 2006.
[6] Baggett, J. S., Driscoll, T. A. and Trefethen, L. N. “A Mostly Linear Model of Transition to Turbulence”,
Physics of Fluids, Vol. 7, 1995, pp.833-838
[7] Kerner, W. “Large Scale Complex Eigenvalue Problems”, Journal of Computational Physics, Vol. 85, 1989,
pp. 1-85.
[8] Farrell, B. F. “Optimal Excitation of Baroclinic Waves”, Journal of Atmospheric Sciences, Vol. 46, 1989,
pp. 1193-1206.
19
American Institute of Aeronautics and Astronautics
[9] Trefethen, L. N., Trefethen, A. E., Reddy, S. C. and Driscoll, T. A. “Hydrodynamic Stability Without
Eigenvalues”, Science, Vol. 261, 1993, pp. 578-584.
[10] Gebgart, T. and Grossmann, S. “Chaos Transition Despite Linear Stability”, Physical Review E, Vol. 50,
1994, 3705-3711.
[11] Handel, A. “Limits of Localized Control in Extended Nonlinear Systems” Ph. D. Thesis, School of
Physics, Georgia Institute of Technology, June 2004.
[12] Schmid, P. J. and Henningson, D. S. Stability and Transition in Shear Flows, 2001, Springer-Verlag, New
York, Berlin, Heidelberg.
[13] Chung, S. H. and Law, C. K. “Burke-Schumann Flame with Streamwise and Preferential Diffusion”,
Combust. Sci. Technol., 37, 1984, p.21.
[14] Meirovitch, L. Analytical Methods in Vibrations, Macmillan Publishing Co., Inc., New York, 1967,
Chapter 6.
[15] Dowling, A. P. “The Calculation of Thermoacoustic Oscillations”, Journal of Sound and Vibration Vol.
180, No. 4, 1995, pp.557-581.
[16] Dowling, A. P. and Stow, S. R. “Acoustic Analysis of Gas Turbine Combustors”, Journal of Propulsion
and Power Vol. 19, No. 5, 2003, pp.751-764.
[17] Matveev, K. I. (2003b) “Thermo-acoustic Instabilities in the Rijke Tube: Experiments and Modeling”, Ph.
D. Thesis, California Institute of Technology.
[18] Fichera, A., Losseno, C., Pagano, A. “Clustering of Chaotic dynamics of a lean gas-turbine combustor”,
Applied Energy, Vol. (69) 2001, pp 101-117.
[19] Wicker, J. M., Greene, W. D., Kim, S-I and Yang, V. “Triggering of Longitudinal Combustion Instabilities
in Rocket Motors: Nonlinear Combustion Response”, Journal of Propulsion and Power, Vol. 12, 1996, 1148-
1158.
[20] Culick, F. E. C., Burnley, V. and Swenson, G. “Pulsed Instabilities in Solid-Propellant Rockets”, Journal of
Propulsion and Power, 11 (4), 1995, 657-665.
[21] Ananthkrishnan, N., Deo, S. and Culick, F. E. C. “Reduced-Order Modeling and Dynamics of Nonlinear
Acoustic Waves in a Combustion Chamber”, Combustion Science and Technology, Vol. 177, 2005, 221-247.
[22] Matveev, K. and Culick, F. E. C, “A Model for Combustion Instability Involving Vortex Shedding”,
Combustion Science and Technology, Vol. 175, 2003, 1059-1083.
[23] Schadow, K., Gutmark, E., Parr, T., Parr, K., Wilson, K., and Crump, J. “Large-scale coherent structures as
drivers of combustion instability”, Combust. Sci. Technol., Vol. 64, 1989, 167–186.
[24] J. S. Baggett, T. A. Driscoll and L. N. Trefethen “A Mostly Linear Model of Transition to Turbulence”,
Physics of Fluids, Vol. 7, 1995, pp. 833-838.
[25] H-G. Yoon, J. Peddieson and K.R. Purdy “Nonlinear Response of a Generalized Rijke Tube” International
J. of Eng. Sci., Vol. 39, 2001, pp. 1707-1723.
[26] Lang, W. “Harmonic Frequency Generation by Oscillating Flames,” Combust. Flame, Vol. 83, 1991, 253-
262.
[27] C. S. Daw, J. F. Thomas and L.L. Narayanaswami, “Chaos in Thermal Pulse Combustion”, Chaos, Vol. 5,
1995 662-670.
[28] Trefthen, L. N. and Embree, M. Spectra and Pseudospectra, 2005, Princeton University Press, Princeton
and Oxford.
20
American Institute of Aeronautics and Astronautics
Appendix 1: Expanded Form of Matrices in the Evolution Equation
The expanded form of the matrices used in Eq. (19) are presented in this appendix.
T
χ = ⎡⎣G11 G21 .....Gmn .....GMN η1 η 2 ......η K η1 η2 .......ηK ⎤⎦
2(1/ γ )
M =I+ M1
(Ti + Tad ) / 2
where, matrix M 1 can be written as a column vector and a row vector as follows,
⎡ 0 ⎤
⎢ 0 ⎥
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ 0 ⎥
M1 = ⎢
. ⎥ [ R11 R12 ......RNM × NM 0 0.........0]( NM + 2 K )×( NM + 2 K )
⎢ ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ π sin π x f ⎥
⎢ 2π sin 2π x f ⎥
⎢ ⎥
⎢ . ⎥
⎢ K π sin K π x ⎥
⎣ f ⎦
⎡ D11 ⎤
⎢ 1 ⎥
⎢ D 2 ⎥
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢ Dmn ⎥
⎢ ⎥
⎢ . ⎥
D=⎢ ⎥
⎢ . ⎥
⎢ DMN ⎥
⎢ ⎥
⎢ 0 ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ . ⎥
⎢ 0 ⎥⎦
⎣
where, D n m = − ⎡⎣(m + 1/ 2) 2 π 2 / L2 + n 2π 2 ⎤⎦ / Pe .
21
American Institute of Aeronautics and Astronautics
⎡[W (1) ] ⎤
⎢ ⎥
⎢ [W (2) ] ⎥
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢ [W ( NM ) ] ⎥
A1 = ⎢ ⎥
⎢ 0 ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢⎣ 0 ⎥⎦
Wmk =
∫
0
sin [ (m + 1/ 2)π x / L ] cos [ (k + 1/ 2)π x / L ] dx
Matrix A2 can be written as a product of a column vector and row vector as follows
⎡ 0 ⎤
⎢ 0 ⎥
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ cos(π x f ) ⎥ (1)
A2 = − ⎢ ⎡C1 NM
C2(1) .....CNM 0 0.........0 ⎤⎦
cos(2π x f ) ⎥ ⎣ ( NM + 2 K )× ( NM + 2 K )
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎢ cos( K π x f ) ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎣ 0 ⎦
where,
lc
∫
∂Z st
Cm( n ) = sin [ (m + 1/ 2)π x / lc ] cos [ (k + 1/ 2)π x / lc ] dx
∂x
0
⎡[0] [0]NM × 2 K ⎤
A3 = ⎢ NM × NM
⎣ [0]2 K × NM [ S ]2 K × 2 K ⎥⎦
22
American Institute of Aeronautics and Astronautics
where,
⎡0 0 . 0 −1 0 0 0⎤
⎢0 0 . 0 0 −1 0 0 ⎥⎥
⎢
⎢ . . . . . . . . ⎥
⎢ ⎥
0 0 0 0 0 0 0 −1⎥
S=⎢ 2
⎢π 0 0 0 0 0 0 0⎥
⎢ ⎥
⎢0 4π 2 0 0 0 0 0 0⎥
⎢ . . . . . . . . ⎥
⎢ ⎥
⎢⎣ 0 0 0 K π
2 2
0 0 0 0 ⎥⎦
2(1/ γ )
A4 = − F1
(Ti + Tad ) / 2
⎡ 0 ⎤
⎢ 0 ⎥
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ 0 ⎥
F1 = ⎢ ⎥ ⎡⎣ 0 0 ....0 cos π x f cos 2π x f ...cos K π x f 0 0....0 ⎤⎦
. ( NM + 2 K )× ( NM + 2 K )
⎢ ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ π sin π xf ⎥
⎢ 2π sin 2π x f ⎥
⎢ ⎥
⎢ . ⎥
⎢ K π sin K π x ⎥
⎣ f ⎦
The dimension of an attractor is a measure of its geometrical scaling properties. There are several
definitions for fractal dimension. In this paper the box counting dimension has been calculated. The box
counting dimension is given by,
⎡ log( N ) ⎤
DB = lim ⎢
δ → 0 log(1/ δ ) ⎥
⎣ ⎦
where N is the number of boxes of side length δ used to cover the attractor. The box counting dimension is
calculated by calculating the slope of log( N ) - log(1/ δ ) plot.
23
American Institute of Aeronautics and Astronautics