Science Arts & Métiers (SAM)

is an open access repository that collects the work of Arts et Métiers Institute of
Technology researchers and makes it freely available over the web where possible.

This is an author-deposited version published in: https://sam.ensam.eu

Handle ID: .http://hdl.handle.net/10985/8642

To cite this version :

Ge SONG, Frédéric ALIZARD, Jean-Christophe ROBINET, Xavier GLOERFELT - Global and

Koopman modes analysis of sound generation in mixing layers - Physics of Fluids - Vol. 25,
p.124101 - 2013

Any correspondence concerning this service should be sent to the repository

Administrator : [email protected]
Global and Koopman modes analysis of sound generation in mixing layers AIP/123-QED

Global and Koopman modes analysis of sound generation in mixing layers

G. Song + , F. Alizard † , J.-C. Robinet + , X. Gloerfelt +1

Laboratoire DynFluid, Arts et Métiers ParisTech+ and CNAM† ,

151 Boulevard de l’Hopital 75013 Paris.

(Dated: 29 September 2014)

It is now well established that linear and nonlinear instability waves play a significant
role in the noise generation process for a wide variety of shear flows such as jets
or mixing layers. In that context, the problem of acoustic radiation generated by
spatially growing instability waves of two-dimensional subsonic and supersonic mixing
layers are revisited in a global point of view, i.e. without any assumption about the
base flow, in both a linear and a nonlinear framework by using global and Koopman
mode decompositions. In that respect, a timestepping technique based on disturbance
equations is employed to extract the most dynamically relevant coherent structures
for both linear and nonlinear regimes. The present analysis proposes thus a general
strategy for analysing the near-field coherent structures which are responsible for the
acoustic noise in these configurations. In particular, we illustrate the failure of linear
global modes to describe the noise generation mechanism associated with the vortex
pairing for the subsonic regime whereas they appropriately explain the Mach wave
radiation of instability waves in the supersonic regime. By contrast, the Dynamic
Mode Decomposition (DMD) analysis captures both the near-field dynamics and the
far-field acoustics with a few number of modes for both configurations. In addition,
the combination of DMD and linear global modes analyses provides new insight
about the influence on the radiated noise of nonlinear interactions and saturation of
instability waves as well as their interaction with the mean flow.

PACS numbers: 47.20.Ky, 43.28.+h, 45.30.+s, 47.20.Ft

Keywords: global linear stability, computational aeroacoustics, modal decomposition,
Koopman modes, mixing layers

1
Global and Koopman modes analysis of sound generation in mixing layers

I. INTRODUCTION

Sound generation in a flow is of key importance for a wide range of industrial applications,
such as noise in aircraft or ground transport. In such configurations, inhomogeneities in the
aerodynamic field are most often at the origin of the sound. This is known as aeroacous-
tics. Strong gradient zones where the flow is sheared and distorted are generally associated
with a complex dynamics, leading eventually to a large range of time and spatial scales in
which successive bifurcations may lead to turbulence. This vortex dynamics is also usu-
ally responsible for significant acoustical radiation in the far field. Nevertheless, from the
point of view of instability theory, vortical structures may be regarded in Fourier-Laplace
space as a collection of instability waves with distinct frequencies and wavelengths. In the
compressible framework, the vortex dynamics can generate sound. How then connect the
instability waves to acoustic radiation? As a canonical flow model to investigate flow noise
mechanisms, the shear layer has been successfully used in computational aeroacoustics. The
shear layer is thus a good prototype for studying the role of instabilities in the mechanism of
acoustic emission. The objective of this paper is to propose a modern computation method
of instabilities that could potentially be applied to complex flows and also to show that this
method calculates the resulting acoustic radiation. Two shear layers, one subsonic and the
other supersonic, as in the study by Cheung & Lele8 , are reviewed in the context of global
instability.

The laminar free shear layer has been extensively investigated both from stability4,5,24,36 or
numerical simulation27,34,39 points of view for various flow regimes (incompressible, subsonic
or supersonic). The study of the associated acoustic radiation by numerical simulations was
10
undertaken in the 90s by Colonius et al. and Bogey & Bailly7 for the two-dimensional
mixing layer. However, the link between the development of instability waves and acoustic
radiation is not yet directly studied.

The first analyzes linking instabilities to the acoustic radiation for a shear layer were
made by Tam & Morris49 and Tam & Burton47 using inviscid local and parallel stability
theory with an asymptotic expansion. The evolution of acoustic waves in the far field was
adequately described. All these studies on this subject have shown that linear stability
analysis did not allow to capture the acoustic radiation for a free shear layer in the subsonic
regime. It is thus necessary to take into account nonlinear processes. For a supersonic shear

2
Global and Koopman modes analysis of sound generation in mixing layers

layer, on the contrary, the linear stability theory predicts well the Mach wave radiation.
Cheung & Lele8 extended these approaches by studying the non-parallel effects as well as
the nonlinear regime with the linear and nonlinear Parabolized Stability Equations (PSE) on
the near-field dynamics and the resulting acoustic radiation of two-dimensional compressible
shear layers.
They also showed that nonlinear PSE method could predict the nonlinear dynamics in the
near field, and in particular the vortex pairings observed in a subsonic shear layer. Although
the vortex-pairing sound can be accurately captured in the near field, the linear or nonlinear
PSE methods do not properly restitute the acoustic field in the far field. This limitation is
mainly due to the homogeneous boundary conditions imposed along the normal direction.
Another disadvantage of PSE methods, assuming that the base flow is weakly non-parallel,
is its inability to capture upstream-propagating waves that may be present in many practical
applications such as cavity flows19 or the impinging jet23 .
To address these limitations, it is necessary to analyze the dynamics through global mode
decomposition. This approach has now become classic in the incompressible regime12,51,52
but still uncommon for compressible flows and more specifically for the supersonic regime
or when the associated acoustic radiation is considered. From a methodological point of
view, there are two families of methods for computing global modes. The first approach
is a matrix-forming method, which consists in computing the eigenvalues of the Jacobian
matrix from the linearized Navier-Stokes equations. This method is usually carried out in
four steps: (i) establishment of linearized Navier-Stokes (LDE), (ii) Laplace transform of
LDE, (iii) spatial discretization of resulting stability equation, (iv) resolution of the resulting
eigenproblem. A number of problems is raised in the compressible regime especially when
the phase velocity of the mode is locally supersonic with respect to the base flow. Although
some problems can be partially solved by reversing steps (ii) and (iii), this method requires
the storage of a large matrix, which limits to relatively simple and two-dimensional (for the
base flow) configurations. Recent applications of this method in the compressible regime can
25 53
be found in Kierkegaard et al. and Yamouni et al. for open cavities, Fosas de Pando
13
et al. for airfoil flow, Nichols & Lele35 and Garnaud et al. 17,18
for jet flows and Mack &
Schmid30 for swept parabolic flow. In short, a critical step in the process of systematically
investigating stability properties by matrix-forming methods is to solve very large eigenvalue
problems and to be uniformly valid with respect to the boundary conditions.

3
Global and Koopman modes analysis of sound generation in mixing layers

The second approach is a matrix-free method, in which very large eigenvalue problems
are solved by storing only velocity fields at different times instead of large matrices. This so-
15
called time-stepper technique was popularized in fluid mechanics by Edwards et al. and
recently used for Blasius flow by Bagheri et al. 3 . There are two main advantages associated
with this method. On one hand, its cost. The modes are calculated using a matrix whose size
is the number of snapshots. Also, management of boundary conditions for the disturbances
is facilitated; it is thus possible to apply nonreflecting boundary conditions compatible with
the acoustic wave radiation. On the other hand, its extension to the nonlinear regime is
straightforward. To achieve this, it is sufficient to compute Koopman modes31,38,41 instead
of global modes by replacing the linearized Navier-Stokes solver by the full nonlinear solver.
The objective of the paper is to characterize the relationship between development of
instabilities and the resulting acoustic radiation within a global instability framework.
Wave effects of linear and nonlinear instability on the near- and far-field dynamics of two-
dimensional compressible free shear layers are studied. Two shear layers are considered,
similar to those analyzed by Cheung & Lele8 . A subsonic configuration, forced by the fun-
damental frequency and its first sub-harmonic, where nonlinear mechanisms are responsible
for the acoustic radiation (vortex pairing sound) and a supersonic configuration, forced only
by its fondamental, where the sound is produced by Mach wave radiation and mainly caused
by linear instabilities.
From a methodological point of view, a nonlinear disturbance equations (NLDE) formu-
lation of two-dimensional compressible Navier-Stokes equations is used. This approach is
32,33
commonly used for aeroacoustic computations, see Morris et al. . The advantage of this
formulation is firstly that the base flow is imposed and secondly it is very easy to move from
linear to nonlinear equations. Based on the linear or nonlinear version of the code, global
or Koopman modes are then extracted by a matrix-free method.
The paper is organized as follows. In section II, the direct, NLDE numerical methods and
algorithms to extract the global and Koopman modes are presented. Section III is dedicated
to the subsonic shear layer where the aerodynamic and the acoustic fields are computed in
§A. In particular, the NLDE solver is validated against the direct solver and the base flow
is computed. The global modes (§B) and the Koopman modes (§C) are extracted by the
time-stepping method and then compared to the NLDE solutions. Special attention is paid
to the reconstruction of the solution and its associated acoustic radiation. The supersonic

4
Global and Koopman modes analysis of sound generation in mixing layers

shear layer case is discussed in section IV in the same way as the subsonic case. Finally,
Section V provides concluding remarks.

II. NUMERICAL METHODS AND ALGORITHMS

A. Direct solvers

1. Full Navier-Stokes equations

The two-dimensional compressible Navier-Stokes equations along with the ideal gas equa-
tion of state are used as a mathematical model to describe the dynamics of the mixing layer.
The governing equations in conservative form may be written in a Cartesian frame as
∂U ∂Ee ∂Fe ∂Ev ∂Fv
+ + − − =0 (1)
∂t ∂x ∂y ∂x ∂y
U is the vector of conservative variables U = (ρ, ρu, ρv, e)T , where ρ is the fluid density and
u, v are the velocity components in the two directions. The total energy per unit volume of
fluid is defined as
1 p 1
e = ρ[cv T + (u2 + v 2 )] = + ρ(u2 + v 2 ) (2)
2 γ−1 2
where p, T , cv and γ are the pressure, temperature, specific heat at constant volume and
specific ratio, respectively. The ratio of specific heats γ = cp /cv is taken to be 1.4. Note that,
in (2), the ideal gas law p = ρrT is used to relate between the thermodynamic variables,
where the gas constant r is fixed to 287.06 J.kg−1 .K−1 . The specific heat at constant pressure
cp and specific heat at constant volume cv are defined as cv = cp /γ and cp = rγ/(γ − 1).
Ee and Fe are the inviscid convective fluxes in the two directions, while Ev and Fv are the
fluxes including both viscous stresses and thermal conduction:

Ee = (ρu, ρu2 + p, ρuv, (e + p)u)T

Fe = (ρv, ρuv, ρv 2 + p, (e + p)v)T

Ev = (0, τxx , τxy , uτxx + vτxy − qx )T

Fv = (0, τyx , τyy , uτyx + vτyy − qy )T

The viscous stress tensor is defined for a Newtonian fluid with Stokes’ hypothesis as

4 ∂u 2 ∂v ∂u ∂v 4 ∂v 2 ∂u
τxx = µ − ; τxy = τyx = µ + ; τyy = µ −
3 ∂x 3 ∂y ∂y ∂x 3 ∂y 3 ∂x

5
Global and Koopman modes analysis of sound generation in mixing layers

The dynamic molecular viscosity µ is approximated with Sutherland’s law:

23
T T0 + 110.4
µ(T ) = µ0
T0 T + 110.4

with the reference point T0 =273.15 K and µ0 =1.711×105 kg.m−1 .s−1 . The heat flux com-
ponents model thermal conduction with Fourier’s law:

∂T ∂T
qx = −κ ; qy = −κ ;
∂x ∂y

where κ is the coefficient of thermal conductivity. The Prandtl number is P r = µcp /κ and
we assume a constant value of P r = 0.723.

2. Nonlinear disturbance equations

Nonlinear disturbance equations (NLDE) have first been proposed by Morris et al.32 .
By decomposing the instantaneous flow variable into a mean flow and fluctuating part,
they obtained the disturbance equations. In their work, the mean flow are calculated with
a Reynolds averaged Navier-Stokes (RANS) solver, and the perturbations quantities are
determined directly. This NLDE approach has been used rather satisfactorily to calculate
the acoustic fields in a supersonic jet33 or a hot jet28 .
In the present stability study, a perturbative version of the Navier-Stokes equations allows
a better control of the base flow and the investigation of the linear regime by simply omitting
the nonlinear terms in the equations. To investigate the behavior of perturbations about the
base flow, the flow vector q is decomposed into the steady base flow qb and a perturbation
q′ :
q = qb + q′ (3)

Note that the base flow is non-bifurcated and an equilibrium solution of (1)45 . Substitution
of (3) into (1) results in a set of base flow and perturbation terms. After rearrangement of
these terms, the NLDE in conservative form can be written as

∂U′ ∂Ee ′ ∂Fe ′ ∂Ev ′ ∂Fv ′

+ + − − =0 (4)
∂t ∂x ∂y ∂x ∂y

where U′ = (ρ′ , ρb u′ + ρ′ ub + ρ′ u′ , ρb v ′ + ρ′ vb + ρ′ v ′ , e′ )T . The convective perturbation fluxes

6
Global and Koopman modes analysis of sound generation in mixing layers

Ee ′ and Fe ′ containing both linear and nonlinear terms are given by

 
′ ′ ′ ′
 ρb u + ρ ub + ρ u 
′ 2 ′ ′ ′ ′2 ′ ′2 ′
ρ ub + 2ρb ub u + 2ρ ub u + ρb u + ρ u + p
 
′
 
Ee =  ,
 ρb (u′vb + ub v ′ ) + ρb u′ v ′ + ρ′ (u′v ′ + u′ vb + ub v ′ ) + ρ′ ub vb 
 
 
u′ (eb + pb ) + ub (e′ + p′ ) + u′ (e′ + p′ )

 
 ρv ′ + ρ′ vb + ρ′ v ′ 
′ ′ ′ ′ ′ ′ ′ ′ ′ ′
 ρb (ubv + vb u ) + ρb u v + ρ (ub v + vb u + u v ) + ρ ubvb 
 
′
Fe = 
 ,
′ 2 ′ ′ ′ ′2 ′ ′2 ′
ρ vb + 2ρb vb v + 2ρ vb v + ρb v + ρ v + p

 
 
v ′ (eb + pb ) + vb (e′ + p′ ) + v ′ (e′ + p′ )

with • : the nonlinear terms

and the viscous perturbation stresses Ev ′ and Fv ′ are given by

Ev ′ = (0, τxx
′ ′
, τxy , u′ τxx
b
+ u′ τxx
′
+ v ′ τxy
b
+ v ′ τxy
′ ′
+ ub τxx ′
+ vb τxy − qx ′ )

Fv ′ = (0, τyx
′ ′
, τyy , u′ τyx
b
+ u′τyx
′
+ v ′ τyy
b
+ v ′ τyy
′ ′
+ ub τyx ′
+ vb τyy − qy ′ )

The components of the perturbation viscous stress tensor are:

4 ∂u′ 2 ∂v ′
′
∂v ′ 4 ∂v ′ 2 ∂u′

′ ′ ′ ∂u ′
τxx = µ − ; τxy = τyx = µ + ; τyy = µ −
3 ∂x 3 ∂y ∂y ∂x 3 ∂y 3 ∂x
Similarly for the components of the base-flow viscous stress tensor9 :

b 4 ∂ub 2 ∂vb b b ∂ub ∂vb b 4 ∂vb 2 ∂ub
τxx = µ − ; τxy = τyx = µ + ; τyy = µ −
3 ∂x 3 ∂y ∂y ∂x 3 ∂y 3 ∂x
and the perturbation heat flux components are written as
∂T ′ ∂T ′
qx ′ = −κ ; qy ′ = −κ
∂x ∂y
As soon as the conservative perturbation vector U′ is obtained from the solution of (4),
the velocity perturbation u′ , v ′ may be obtained from the definition of U′ and the base flow
values, while the fluctuating pressure p′ may be deduced from
p′
e′ = + ρb (u′ub + v ′ vb ) + ρ′ (u′ ub + v ′ vb )
γ−1
(5)
1 2 2 1 1 2 2
+ ρb (u′ + v ′ ) + ρ′ (ub 2 + vb 2 ) + ρ′ (u′ + v ′ )
2 2 2

7
Global and Koopman modes analysis of sound generation in mixing layers

3. Algorithm and boundary treatment

Two main difficulties have to be considered when writing a computational aeroacoustics

(CAA) code: to propose an efficient algorithm for the acoustic wave propagation and to
use accurate boundary conditions for the truncated computional domain. In fact, acoustic
waves have some typical features that should be taken into account: their nondissipative
and nondispersive character, the extremely low amplitude of acoustic perturbations, and
the need to resolve high frequencies accurately. Therefore, the NLDE are advanced in
time using an explicit six-substep Runge-Kutta scheme, with coefficients optimized in the
frequency space and fourth-order formal accuracy6 . An 11-point-stencil centered Dispersion-
Relation Preserving (DRP) scheme, initially developed by Tam & Webb50 , is used to obtain
the spatial derivatives. To avoid the so-called grid-to-grid oscillations due to the centered
finite-difference scheme, an 11-point-stencil centered selective filter is introduced in order to
filter out non-physical high-frequency oscillations6 .

Non-reflecting conditions are required to mimic an infinite continuous medium. To this

end, a set of radiation and outflow boundary conditions, proposed by Tam & Dong48 , are
used. In particular, a sponge zone combining grid stretching and a Laplacian filter is applied
at the outflow boundary (see21 for details).

To trigger the convective flow instabilities, a forcing is introduced in the inlet plane. In
particular, for the inlet condition, the radiation conditions, expressed in the polar coordinates
(r, θ) centered at the center of the computational domain, are modified at the inlet boundary
as
   
ρ − ρb ρ′in

u−u 
 
 u′ 

1 ∂ ∂ 1 b  1 ∂ ∂ 1  in 
+ +  = + + (6)

Vg ∂t ∂r r Vg ∂t ∂r r
  
 v − vb   v′ 
   in 
p − pb p′in

p
where Vg = ub cos θ + vb sin θ + c2b − (vb cos θ − ub sin θ)2 denotes the group velocity of
acoustic waves, cb being the local sound speed calculated from base-flow values. The vector
q′ in = (ρ′in , u′in , vin
′
, p′in )T assembles the density, velocity components, and pressure fluctu-
ations to be imposed at the inflow. It is explicitely defined by the solutions of local linear
stability analysis as described in Appendix A, so the right-hand side of (6) is known.

8
Global and Koopman modes analysis of sound generation in mixing layers

B. Global and Koopman modes

1. Iteratives techniques based on snapshots

In classical dynamical system theory, the evolution of flow variables q′ defined on a state
space D may be expressed as
∂q′
= F (q′ ) (7)
∂t
From a time-stepper point of view, the system (7) is rewritten as

q′ (t + ∆t) = B (∆t) q′ (t) (8)

where t represents the time evolution and B (∆t) is a propagator for a time step ∆t. In
this context, a linear snapshot-to-snapshot mapping is associated with B (∆t), where a data
sequence is generated by the time integration of the dynamical system (7). The Nyquist
criterion applied to ∆t determines which structures are solved in the frequency domain.
In a time-stepper framework, the coherent structures are associated with eigenmodes of
B (∆t), called the Ritz eigenvalues. Hence, this theory can be applied in both a linear and
nonlinear framework.
Let us consider a set of N snapshots of (7), separated in time by ∆t

S N = (S1 , S2 , ....SN ) (9)

where the propagator B (∆t) maps an instantaneous Si onto the next one Si+1 . In a linear
theory, B (∆t) is linked to the Jacobian matrix about a steady state, referred to as A, via

B (∆t) = eA(∆t)

The eigenmodes of the propagator yield the so-called global modes. In a nonlinear frame-
43
work, the eigenmodes of B (∆t) are connected to the Koopman modes (see Schmid et al.
and Mezic31 ). Arnoldi-type algorithms are therefore suitable for obtaining an approximation
of the dominant eigenmodes of the propagator by using (9).
⊥
In a linear framework, an orthonormalized basis S N of S N is constructed using a
⊥
modified Gram-Schmidt orthogonalization algorithm. A projection of B (∆t) onto S N
yields the system:
⊥ ⊥
B (∆t) S N = SN H

9
Global and Koopman modes analysis of sound generation in mixing layers

where H is an N × N upper Hessenberg matrix. By increasing the number of snapshots N,

the eigenmodes of H converge toward the dominant eigenmodes of B (∆t).

In a nonlinear framework, the Dynamic Mode Decomposition algorithm yields an approx-

imation of the Koopman modes41 . The N-th snapshot is expressed as a linear combination
of (S1 , S2 , ..., SN −1 ). The projection of B (∆t) onto S N leads thus to

B (∆t) S N = S N C

with C an N ×N companion matrix, whose eigenvalues approximate the dominant Koopman

modes. In this context, the expression ’dominant modes’ is used for the most observable
modes. One may remark that the algorithm associated with the global modes could be
recovered from a QR decomposition of the companion matrix C.

This Arnoldi-type method allows us to extract the most dynamically relevant coherent
structures of linear and nonlinear dynamics. These modes have the property that they
attribute to each coherent structure a spatial shape, q̂k , multiplied by a time-dependent
k
function of the form e−iω t with ω k = ωrk + iωik (ωrk its circular frequency and ωik its temporal
amplification rate) and such that

k log ω k
λ = (10)
∆t

where λk are the eigenvalues of B (∆t).

As a consequence, a similar numerical method, based on data-sequence of snapshots is
employed in the following to treat both linear and nonlinear dynamics. More details about
the linear algorithm can be found in2 .

One may remark that each method requires an orthogonal projection onto a data se-
quence. Our particular interest is to deal with compressible flow and aeroacoustics. Then,
we introduce a compressible inner product (•, •)E that includes both kinetic and internal
energies (11)
p′1 ∗ p′2
Z
1 ′∗ ′ ′∗ ′
(q′1 , q′2 )E = ρb (u1 u2 + v1 v2 ) + dV (11)
Ω 2 pb (γ − 1)

where Ω represents the computational domain, •b refers to the basic state, and ∗ denotes
the complex conjugate.

10
Global and Koopman modes analysis of sound generation in mixing layers

2. Orthogonal projection

Both global and Koopman modes are not necessarily orthogonal. In order to perform
a projection of a solution of our dynamical system (7) onto a set of eigenmodes of the
propagator, a modified Gram-Schmidt procedure is undertaken to orthogonalize the basis
(see1 and16 ). Let us denote the orthogonal basis as (q′⊥ ′⊥ ′⊥
0 , q1 , · · · , qm−1 ). An instantaneous

field may be expanded as

m−1
X
′
q (x) = ξk q′⊥
k (x) (12)
k=0

Taking the orthogonality of the basis into consideration,

ξk = q′⊥ ′

k , q0 E
(13)

Therefore, the coordinates in the basis of global/Koopman modes, referred to as Kk , are

recovered by making use of a matrix product:

K = P −1 ξ (14)

with K = (K0 , K1 , · · · , Km−1 )T , and ξ = (ξ0 , ξ1 , · · · , ξm−1 )T . The coefficients of P are given

by Pi,j = q′⊥ ′
j , qi E .

III. SUBSONIC CASE

In this section, the sound generated by a fixed vortex-pairing event in a subsonic mixing
layer is investigated. First, results from direct computations are presented. Then a linear
global-mode analysis in conducted. A non-linear Koopman-mode analysis is finally used to
reconstruct both the aerodynamic and acoustic fields.

A. Direct computations

1. Configuration and numerical specifications

The first case considered is a cold subsonic Mach 0.5/0.25 mixing layer at a Reynolds
number Re = ρ∞ Uc δω (0)/µ∞ = 1500, based on the initial vorticity thickness δω (0). The
reference velocity is Uc = (U1 + U2 )/2 where U1 =0.5c∞ and U2 =0.25c∞ are the speeds in the

11
Global and Koopman modes analysis of sound generation in mixing layers

upper and lower streams, respectively. The thermodynamic variables are the same in both
√
streams with T∞ =298 K, p∞ =101 300 Pa, ρ∞ = p∞ /(rT∞ ), c∞ = γrT∞ , and µ∞ = µ(T∞ ).
The computational domain has dimensions (Lx , Ly ) = (275, ±235)δω (0), including the
sponge zone. The mesh is built up from a nonuniform Cartesian grid of 700 by 382 grid
points in the x- and y-directions, respectively. The grid along x is uniform with spacing
∆xmin = 0.32 δω (0) up to x = 195 δω (0). The grid is then progressively stretched from
x = 195 δω (0) to x = 275 δω (0) to form the sponge zone as defined in section II A. Along
the normal direction, the grid is stretched at a rate of 1.8% symmetrically from y = 0 to
y = ±90 δω (0), with ∆ymin = 0.16 δω (0) at y = 0. The step size is then kept constant up to
±235 δω (0).

2. Base flow

First a base flow is obtained from the solution of the full Navier-Stokes equations (1)
without inlet perturbations. The initial flow is a parallel hyperbolic-tangent profile for the
streamwise velocity:
uini (y) 2y
= 1 + R tanh (15)
Uc δω (0)
where R = (U1 − U2 )/(U1 + U2 ) is the velocity ratio, equal to 1/3 in the present case. The
mean convective Mach number is Mc = Uc /c∞ = 0.375. The temperature is initialized using
the Crocco-Busemann relationship:

1
Tini (y) = T∞ + (uini (y) − U2 )(U1 − uini (y)) (16)
2cp

In the absence of inlet perturbations, a steady solution can be reached due to the convec-
tive nature of the instabilities. After a transient period, the flow is seen to relax toward a
steady state as depicted in Figure 1(a). 600 000 iterations are necessary to saturate residual
values.
The base flow spreads slowly along the streamwise direction due to the external flow
entrainment and the viscous diffusion. The streamwise velocity profile is very close to the
hyperbolic-tangent function (15), as shown in Figure 1(b). Note that several methods may
28,29
have been used to compute a base flow. For instance, Lesshafft et al. used a solution
of the steady compressible boundary layer equations. The advantage in the present study is
that the base flow is a solution of the governing equations in the discrete sense.

12
Global and Koopman modes analysis of sound generation in mixing layers

(a)
(b)
0

-2 1.2
res
-4 U/Uc 1

0.8
-6

0 2 4 6 -4 -2 0 2 4

iterations ×105 y/δω (0)

FIG. 1. Base flow for the subsonic mixing layer: (a) convergence of normalized residuals (in
logarithmic scale) for the conservative variables as a function of time iterations: ρ, • ρu,
ρv, N ρe; (b) comparison of streamwise velocity profiles of the base flow (• •) and the
hyperbolic-tangent function (15) ( ) at the inlet plane.

3. NLDE results

The local linear stability theory allows us to determine the fundamental frequency f0
at the inlet f0 = ω/2π with ω the circular frequency ≈ 2π × 0.82Uc /δω (0) (see Appendix
A). To fix the vortex pairing location, the flow is forced artificially at its fundamental
7 10
frequency f0 and its first sub-harmonic f0 /2 as in Bogey et al. and Colonius et al. . The
eigenfunctions and corresponding streamwise wavenumbers obtained from the local stability
theory are integrated in the inflow boundary condition (6). This forcing condition is applied
at every time step. The NLDE equations are run for 100 000 iterations with a timestep
δtUc /δω (0) ≃ 0.039.
The results from the NLDE solver (§II A 2) using the disturbance equations about the
base flow are compared to the solution of the non-perturbative equations (§II A 1) in Figures
2 and 3 at the same instant. The similarities are evident for both the near-field vortical flow
and the far-field acoustic field, validating the use of the NLDE with a base flow. The

13
Global and Koopman modes analysis of sound generation in mixing layers

10
y/δω (0) 0
-10
50 100 150 200
x/δω (0)

(a)

10
y/δω (0) 0
-10
50 100 150 200

x/δω (0)

(b)

FIG. 2. Direct computations for the subsonic case. The snapshots of the total spanwise vorticity
show the vortex pairing process: (a) full Navier-Stokes simulation; (b) NLDE simulation. Contours
levels range from −0.20 to −0.05 where the values are made dimensionless by Uc and δω (0).

spanwise vorticity plot in Figure 2 exhibits the saturated Kelvin-Helmholtz rolls and the
fixed vortex pairings around x ≃ 100δω (0). The acoustic pattern in Figure 3 illustrates the
rotating quadrupolar radiation7 resulting from vortex-sound generation mechanism. The
principal directions of the wavefronts depend on convection effects in the upper and lower
streams.
The directivity of the wave radiation provides a quantitative knowledge of noise emissions.
To this aim, the sound pressure level
!
p′rms
SP L = 20 log10 (17)
p′ref

where p′ref = 2 × 10−5 Pa is the reference pressure and p′rms is the root mean square (rms)
value is computed during one pairing period. The directivity is then evaluated by interpolat-
ing the values of p′rms on an arc of radius 100δω (0) centered at the apparent source location
(x, y) = (100δω (0), 0). This apparent source location corresponds to the vortex pairing zone

14
Global and Koopman modes analysis of sound generation in mixing layers

(a) (b)

y/δω (0)

x/δω (0) x/δω (0)

FIG. 3. Direct computations for the subsonic case. The fluctuating pressure over the entire domain
shows the far-field noise radiation: (a) full Navier-Stokes simulation; (b) NLDE simulation. The
pressure is made dimensionless with ρ∞ c2∞ . Contours levels range from −3 × 10−5 to 3 × 10−5 .

which is responsible for the acoustic radiation7 . The directivity is shown in Figure 4(a). For
the upper stream, the maximum of acoustic radiation appears at θ ≈ 38◦ measured from
the streamwise direction. A smaller angle θ ≈ −21◦ from the axis is noticed for the emission
lobe in the lower stream.

Lastly, the frequency content in the flow is characterized through the evolution of the
modal energy as defined by (11) but integrated only in a normal section (dV ≡ dy), plotted
in Figure 4(b). The excited modes at frequencies f0 /2 and f0 fisrt undergo an exponential
growth, in good agreement with the local stability prediction in Appendix A, underlying the
almost parallel character of the base flow. The fundamental mode saturates first and is then
overwhelmed by the subharmonic component. This modification of the dominant frequency
marks the location of the pairing event. Higher-order modes such as 3f0 /2 and 2f0 are then
triggered through nonlinear interactions.

15
Global and Koopman modes analysis of sound generation in mixing layers

(a)
(b)
110
0
105

100
SP L (dB)

-2
95 Em
90 -4
85

80 -6
-120 -80 -40 0 40 80 120 0 50 100 150 200

θ x/δω (0)
FIG. 4. NLDE computation for the subsonic case: (a) Far-field directivity for an arc of radius
100δw (0) centered at x = 100δω (0); (b) Integrated modal energy for f0 /2 ( ), f0 ( ), 3f0 /2
( ) and 2f0 ( ). The growth rates ( ) obtained from the local stability
analysis in Appendix A for f0 /2 and f0 are shown for comparison.

B. Linear analysis: global modes and LDE simulation

1. Global modes: numerical parameters and spectrum

In this section, the nonlinear terms in (4) are omitted and the equations are referred to
as linear disturbance equations (LDE). The sequence of snapshots is initiated by a localized
impulse in the mixing layer. Even with a divergence-free initial velocity field, it is important
to specify properly the thermodynamic variables p′ and ρ′20,37 to avoid spurious noise. Hence,
we use as an initial condition a Taylor’s vortex with a given amplitude A = 10−6 :
2 2
u′ = A (y − y0 ) eΛR , v ′ = −A (x − x0 ) eΛR
(18)
q
where R = (x − x0 )2 + (y − y0 )2 and Λ = −ln (2/b2 )

The Gaussian half-width b is taken as 3∆ymin and the vortex is centered at (x0 , y0) =
(20δω (0), 0). The initial pressure is known analytically:
A2 2ΛR2
p′ = −ρ∞ e (19)
4Λ

16
Global and Koopman modes analysis of sound generation in mixing layers

0.02

ωi -0.02

-0.04

-0.06

-0.08
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2

ωr
FIG. 5. Global spectrum of the subsonic mixing layer is shown. The eigenvalues are dimensionless
by δω (0) and Uc .

As disturbances grow in amplitude and propagate downstream, the sequence of snapshots

separated by a sampling period ∆t is carried out. The sampling period in the present
investigation is fixed to a value which guarantees a resolution up to 2f0 (Nyquist criterion).
The snapshots are realized for a window extending from x = 0 to x = 115δω (0) and y =
±45δω (0). In the following, N = 240 snapshots are used to extract the dominant global
modes. The number of snapshots are chosen to satisfy a minimal residual value for the
Arnoldi algorithm. In addition, one may precise that the base flow is slighty nonparallel
which prevents ill-conditioned operator11 .

The global spectrum is depicted in Figure 5. One may observe that all the global modes
are damped temporally, illustrating the well-known noise amplifier behaviour of a coflow
mixing layer. Figure 6 shows two samples of eigenvectors at two different circular frequen-
cies. The wavelength is seen to decrease as the circular frequency increases and a large
spatial amplification is observed for both frequencies. These remarks are consistent with the
convective nature of instabilities arising in the mixing layer.

By plotting the pressure component of global modes no acoustic radiation originating

from the shear layer is visible in contrast with the NLDE simulation. To illustrate this last
comment, the relative phase velocities of the two global modes displayed in Figure 6 are
computed for the lower and upper streams. For that purpose, we use a weakly nonparallel
assumption to derive an expression of the wavenumber αr along the streamwise direction.

17
Global and Koopman modes analysis of sound generation in mixing layers

(a)

y/δω (0)

(b)

y/δω (0)

x/δω (0)
FIG. 6. Pressure fluctuations associated with eigenvalues marked by (a) • and (b) N in Figure 5.

Hence, (20) is evaluated at the maximum value of |v̂| along x as

∂ −1 v̂i
αr = tan (20)
∂x v̂r

where •r and •i denote the real and imaginary parts, respectively. The relative phase velocity

18
Global and Koopman modes analysis of sound generation in mixing layers

0.16 0.16

0.14 0.14

Mr,k
0.12 0.12

0.1 0.1
50 60 70 80 90 100 50 60 70 80 90 100

x/δω (0) x/δω (0)

(a) (b)

FIG. 7. Phase speeds Mr,1 ( ) and Mr,2 ( ) corresponding to the eigenvalues marked by:
(a) N and (b) • in the spectrum of Figure 5.

is defined as
|ωr /αr − Uk |
Mr,k = (21)
c∞
where k = 1, 2 for the upper and the lower stream respectively and c∞ denotes the speed of
sound. The relative phase velocities of the global modes of Figure 6 are plotted in Figure
7 for the lower and upper streams. As expected, each global mode has a subsonic phase
velocity relative to the ambient sound speed. Therefore, these instability waves can not
generate sound47 . A similar behaviour is observed for all global modes of the spectrum
shown in Figure 5.

2. Response to a harmonic forcing and noise emissions

Lastly, the method of artifical excitation with real frequency is classically adopted to
understand the noise generation process in open flows. For instance, the subsonic mixing
layer is forced at the fundamental and subharmonic frequencies in the NLDE simulation,
presented in the section III A 3, to get insight into noise emissions. Since the pioneering
work of Cossu & Chomaz11 , it is well established that temporally damped global modes may
lead to transient growth. To characterize this growth, the transfer function associated with
the response to a harmonic forcing may be investigated42 . Note that the harmonic forcing is
based on an appropriate superposition of the previously computed global modes. Since each
global mode is individually not radiating, a similar behaviour is expected for the summation
of modes.

19
Global and Koopman modes analysis of sound generation in mixing layers

To model the external excitation, a small-amplitude forcing term f = f̂e−iωf t is added to

the linear system, characterized by its angular frequency ωf and its spatial structure f̂. The
flow being globally stable, the asymptotic response is governed by
m−1
′
X
−iωf t Kkf
q = Kk q̂k e with Kk = (22)
k=0
iωf − iωk

where the expansion coefficients of f̂ into a global modes basis have been noted Kkf . To
measure the response of our linear system to a harmonic forcing at a real frequency ωf , we
introduce the following quantity:
kq′ k2
R(ωf ) = 2 (23)
f̂
In practice, the forcing term takes the form of a wavepacket centered near x/δω (0) = 40
derived from an instantaneous snapshot of the LDE simulation, which was used in the
previous section to extract the global modes. An orthogonal projection is employed to
recover the expansion coefficients. A total of m = 62 global modes are considered, including
positive and negative circular frequencies. The transfer function is illustrated in Figure 8(a).
We observe a broadband hump centered around a dimensionless forcing frequency ωf ≈ 0.67,
which is somewhat lower than the fundamental frequency as computed by a local stability
theory at the inlet (see Appendix A). This shift may be attributed to the thickening of the
shear layer along the streamwise direction for the base flow.
The fluctuating pressure associated with the asymptotic response is depicted for the sub-
harmonic in Figure 8(b). As expected from the analysis of individual global modes, no
acoustic radiation is observed for the global response at f0 /2. Similar observation could be
made for all forcing frequencies since the present study is restricted to the linear regime.
Hence, this analysis illustrates the ability of the global modes to evaluate the linear ampli-
fication of perturbations in subsonic mixing layer but also exhibits its failure to represent
the noise radiation observed in the NLDE simulation.

3. Comparison between LDE and NLDE simulations

To further illustrate the conclusions given by our global modes analysis, a LDE simulation
is performed by considering the forcing condition detailed in section III.A.3. The result is
displayed in Figure 9. From the pressure field, in agreement with the global modes analysis,

20
Global and Koopman modes analysis of sound generation in mixing layers

(a) (b)

f0
1

0.8

R 0.6
y/δω (0)
0.4

0.2

0
0.4 0.6 0.8 1 1.2

ωf x/δω (0)

FIG. 8. Linear response to the forcing term based on 62 global modes: (a) Transfer function
normalized by its maximum value. The dimensionless fundamental circular frequency at the inlet
is referenced with a vertical dashed line; (b) Spatial structure associated with the response at f0 /2.

no acoustic radiation appears. In addition, the influence of the nonlinear terms is character-
ized through the evolution of the integrated energy in a normal section for disturbances, at
a given time. In particular, a significant gap between the LDE and the NLDE simulations
is observed at the pairing location giving strong evidence that the nonlinear terms are not
small enough to be neglected.

C. Nonlinear model

In this section, the subsonic mixing layer is used to illustrate the possibility to extract
the coherent structures responsible for the radiated noise in a nonlinear framework, from a
DMD analysis.

1. DMD analysis: Koopman modes

For that purpose, the DMD analysis is applied to the NLDE database, described in
section III A, over the entire domain, referred to as Df , and over a subdomain, noted Ds ,
extending from x = 0 to 250 δω (0) and from y = −15 δω (0) to y = 15 δω (0). This subdomain
is thus dedicated to the analysis of the vortical structures in the source region. A number

21
Global and Koopman modes analysis of sound generation in mixing layers

(b)

6
y/δω (0)
4

2
E
0

-2

0 50 100 150 200

x/δω (0) x/δω (0)

(a)
FIG. 9. LDE simulation for the subsonic case; (a) the pressure field is shown; (b) integrated
energy associated with disturbances for NLDE( ) and LDE ( ) simulations are plotted in
a logarithmic scale.

of N = 210 instantaneous snapshots of the density, velocity and pressure fluctuations are
recorded at a sampling rate ∆tUc /δω (0) = 0.6358, which corresponds approximately to 12
points per period at the fundamental frequency f0 .
The eigenvalues of the Koopman operator associated with Df and Ds are shown in Figure
10(b) and 11(b), respectively. Nearly all the Ritz values are found to lie on the unit circle
|λj | = 1, which means that the sample points lie on or near an attracting set38 . Figures
10(a) and 11(a) display the normalized energies of the extracted modes as a function of the
frequency. Each mode represented by its norm is depicted with a vertical line and appear as
a complex conjugate pair. The two dominant peak frequencies containing the highest energy
are then identified and marked with the same colours as the corresponding eigenvalues in
Figures 10(b) and 11(b). These two modes are associated with the first subharmonic f0 /2
and the fundamental mode f0 . Higher frequencies are also excited, corresponding to sum

22
Global and Koopman modes analysis of sound generation in mixing layers

(a) (b)
1 1

0.8
0.5

0.6
E/Emax λi 0
0.4

-0.5
0.2

0 -1
-1.5 -1 -0.5 0 0.5 1 1.5 -1 -0.5 0 0.5 1

ωr λr

FIG. 10. DMD analysis performed over the entire computational domain Df : (a) Energy spectrum;
(b) Ritz values. The values are made dimensionless by δω (0) and Uc .

(a) (b)
1 1

0.8
0.5

0.6
E/Emax λi 0
0.4

-0.5
0.2

0 -1
-1.5 -1 -0.5 0 0.5 1 1.5 -1 -0.5 0 0.5 1

ωr λr

FIG. 11. DMD analysis performed over the subdomain Ds : (a) Energy spectrum; (b) Ritz values.
The values are made dimensionless by δω (0) and Uc .

and difference of the frequencies. For instance, the nonlinear interaction of f0 /2 and f0
results in a peak at 3f0 /2, and so on. One may notice that a few modes dominate the
energy spectrum and that the energy distribution among the Koopman modes is roughly
similar for Df and Ds .

23
Global and Koopman modes analysis of sound generation in mixing layers

To represent the vorticity and pressure modes associated with each energy peak, the
Koopman modes are projected onto an instantaneous snapshot from the NLDE simulation.
The pressure fluctuation components are displayed in Figure 12 for the modes at f0 and
f0 /2 from the analysis on Df . The acoustic waves at f0 /2 emanate from the region where
the pairings occur. Besides, the mode f0 produces a weak far-field noise compared to the
mode at f0 /2. Similar observations could also be made for 3f0 /2 or 2f0 . Therefore, the
first subharmonic mode contributes mainly to the acoustic radiation in the mixing layer.
Very different results would be obtained with a proper orthogonal decomposition (POD).
For instance, Druault et al.14 show that the most energetic POD modes are not associated
with the main noise radiation, which can be obtained from a series of higher-order POD
modes. As underlined by Schmid41 , higher-order POD modes, corresponding to a smaller
energetic content, can still play a significant dynamic role within the snapshot sequence.
In addition, the acoustic behaviour observed in Figure 12 gives some insight about the
slight modification of the relative magnitudes of Koopman modes for Df and Ds as depicted
in Figures 10 (a) and 11 (a). As discussed above, the noise is mainly produce by the
subharmonic f0 /2. As a consequence, a larger contribution of the pressure component with
respect to the inner product (11) is expected for f0 /2 by considering Df which includes the
radiated noise. This additional contribution is not captured by Ds where the DMD analysis
is restricted to the vortical structures in the near field. For instance, the ratio between the
energy peaks associated with f0 and f0 /2 is ≈ 0.72 for Ds and ≈ 0.56 for Df . Finally, it
is worth noting from the Koopman modes magnitudes that the zero frequency mode plays
a significant role in the dynamics with a magnitude of the same order as 2f0 . The total
vorticity, which includes the base flow, for the zero frequency mode and the time-averaged
flow are shown in Figure 13 for Df . A perfect match is observed, which clearly identifies the
zero frequency mode as the mean flow correction. In particular, it illustrates the nonlinear
effects associated with the mixing process, with a strong modification of the total vorticity
near the vortex pairing position. This nonlinear mechanism will lead to a modification of
the instability waves behaviour which can not be captured with a linear theory.
Finally, to gain further insight into the revelancy of the subdomain analysis, the far-field
acoustic noise is recovered via a Kirchhoff’s extrapolation method from a surface enclosing
the source (see Appendix B). This surface is composed of both the upper and lower bound-
aries of Ds . For the computation of pressure fluctuations and their derivatives in (B1), we

24
Global and Koopman modes analysis of sound generation in mixing layers

(a) (b)

y/δω (0)

x/δω (0) x/δω (0)

FIG. 12. Real part of the non-dimensional pressure fluctuations for Koopman modes calculated
on the entire domain Df corresponding to: (a) f0 /2; (b)f0 . A projection on an initial snapshot
is used to obtain the amplitudes. Contours levels range from −3 × 10−5 to 3 × 10−5 in steps of
5 × 10−6 in plot (a) and from −3 × 10−6 to 3 × 10−6 in steps of 5 × 10−7 in plot (b).

use the projection of the pressure field in the source region:

p′ (x, t) = Kk pk (x) e−iωk t (24)

with pk the pressure component of the corresponding Koopman mode and Kk a weight-
ing coefficient associated with the scalar product. As observed in Figure 14, the pressure
predicted in the far field from the analysis based on Ds and formulae (24) and (B1) is in
agreement with the decomposition based on the full domain Df .
In particular, the directivities associated with f0 /2, shown in Figures 15, indicate that
the far-field radiation evaluated from Df and Ds are of the same order of magnitude and that
the directions of radiation in the upper and lower streams are in good agreement. It is worth
noting that it hightlights the ability of the DMD analysis performed on the entire domain to
extract accurately the dominant vortical structures together with the main contribution to
the noise radiation. In addition, the Koopman modes derived from the subdomain analysis

25
Global and Koopman modes analysis of sound generation in mixing layers

10
y/δω (0) 0
-10
50 100 150 200
x/δω (0)

(a)

10
y/δω (0) 0
-10
50 100 150 200

x/δω (0)

(b)

FIG. 13. The total vorticity for the zero frequency mode with respect to Df (a) and the time-
averaged flow (b) are shown. Contours levels range from −0.20 to −0.05.

are sufficiently accurate to be combined with a Kirchhoff’s extrapolation method to recover

the far-field acoustics.

We have demonstrated that the DMD analysis applied to a near-field region or to the
full domain is capable of describing accurately the coherent structures which contributes
mainly in terms of dynamics as well as their noise radiation. In particular, the comparison
between the subdomain, corresponding to the near-field and the full domain analysis allows
to distinguish the contribution of the dominant coherent structures which are involved in
the hydrodynamic behaviour in the near field and their impact on the acoustic noise in the
far field.

A final observation can be made: when forcing the flow with f0 and f0 /2, the structures
associated with the Koopman modes can radiate sound directly to the far field which totally
differs from the linear global modes where no radiation occurs from the global flow response
to harmonic forcing at f0 and f0 /2. This last comment can be highlighted through the Figure
16, where cross sections of the modulus of the pressure eigenfunctions for the Koopman

26
Global and Koopman modes analysis of sound generation in mixing layers

y/δω (0)

x/δω (0)

FIG. 14. Real part of the non-dimensional pressure fluctuations for Koopman modes calculated
on the subdomain Ds corresponding to f0 /2. The Kirchhoff wave extrapolation method from the
boundaries of the subdomain Ds is used to determine the far-field acoustic noise. Contours levels
range from −3 × 10−5 to 3 × 10−5 in steps of 5 × 10−6 .

mode associated with f0 /2 and for the global mode for ωr ≈ 0.4 (which is closed to f0 /2) at
x/δω (0) = 110 are plotted. A clear exponential decay is visible for the global mode, whereas
the Koopman mode reveals bounces after |y| > 30δω (0) which are linked to the acoustic
emissions. In the subsonic case, the sound has a purely non linear origin, namely the pairing
event.

Hence, being able to identify the different structures and their associated noise radiation
separately, let us now consider the interactions between different eigenmodes of the Koopman
operator that can potentially generate acoustic radiation.

27
Global and Koopman modes analysis of sound generation in mixing layers

110

105

100
SP L(dB)
95

90

85

80
-120 -80 -40 0 40 80 120

θ
FIG. 15. Far-field directivity obtained by the DMD analysis for an arc of radius 100δω (0) centered
at x = 100δω (0) for the subharmonic mode f0 /2: ( ) full domain Df ; ( ) subdomain Ds
+ Kirchhoff’s extrapolation method.

2. Reconstruction of the near- and far-field dynamics

Now, to assess the ability of the DMD analysis to reproduce both the entire shear layer
dynamics in the near and far fields, a reconstruction is carried out from a selection of the
dominant eigenmodes. Considering the non-orthogonality of the Koopman modes, a prior
orthogonal projection onto the Koopman modes basis is used as described in the section
II B 2. By increasing gradually the number of modes, the difference between an instantaneous
snapshot from the NLDE simulation and the projection is made lower than a residual value
of 10−4 (based on a norm associated with the scalar product (11)) for 7 modes including the
zero-frequency mode. One may precise that both modes with positive frequency and their
complex conjugate are taken into account in the projection.
As before, we first examine the ability of the DMD analysis to capture the hydrodynamic
motions in the near field. A qualitative assessment of the vortex pairing mechanism is
provided by the total spanwise vorticity plots at four equidistant instants during one period
in Figure 17. When compared to the field computed directly (Figure 2), we see that both

28
Global and Koopman modes analysis of sound generation in mixing layers

40

20

y/δω (0) 0

-20

-40
-3 -2 -1 0
10 10 10 10

FIG. 16. Comparison of linear and nonlinear models through absolute value of pressure cross
sections for the subsonic mixing layer for the subharmonic at x/δω (0) = 110: ( ) Koopman
mode; ( ) global mode. The profiles are made dimensionless by their maximum value.

the roll-up and vortex pairing events are quite well predicted by the DMD analysis. In
particular, the projections based on Ds and Df provide similar results.

In order to further assess the accuracy of the DMD model to represent the near field
dynamics, a time signal for the normal velocity field taken at the vortex pairing location
(x = 100, y = 0) δω0 is plotted in Figure 18 by increasing the number of Koopman modes
from 1 to 7 (plus their complex conjugate). A very good agreement with NLDE reference is
already obtained with only 7 modes for both Ds and Df .

Let us now focus on the acoustic behaviour prediction in the far field. The pressure
perturbations from NLDE simulation serving as the reference is plotted in Figure 3(b).
The projected fluctuating pressure from DMD is depicted in Figure 19(a). The acoustic
patterns are very similar by using the same levels and the acoustic waves emanate from the
same apparent source location around (x, y) = (100δω (0), 0). The direction and intensity of
acoustic waves can be further validated by plotting the directivity. To this aim, the projected

29
Global and Koopman modes analysis of sound generation in mixing layers

(a) (b)
10 10
y/δω (0) 0 0
-10 x
-10

10 10
y/δω (0) 0 0
-10 -10

10 10
y/δω (0) 0 0
-10 -10

10 10
y/δω (0) 0 0
-10 -10
50 100 150 200 50 100 150 200

x/δω (0) x/δω (0)

FIG. 17. Follow-up of the vortex pairing in the near-field region from a projection onto a Koopman-
mode basis with 7 modes associated with: (a) the subdomain region; (b) the entire flow field. Same
levels as in Figure 2.

(a) (b)

0.2 0.2

v/Uc 0 0

-0.2 -0.2

0 10 20 30 40 0 10 20 30 40

t × Uc /δω (0) t × Uc /δω (0)

FIG. 18. Temporal evolution of the vertical velocity component taken at x/δω (0) = 100, y/δω (0) =
0 extracted from a reconstructed flow field onto the dominant Koopman modes over: (a) the entire
domain Ds ; (b) the subdomain Df . Reconstructions with 1 mode ( ); 2 modes ( ); 7
modes ( • ); NLDE results ( ).

fluctuating pressure defined by

m−1
X
′
p (x, t) = Kk pk (x) e−iωk t (25)
k=0

with m − 1 = 13, is advanced in time during one pairing period to compute the rms values.
Figure 19(b) shows a very good agreement with the NLDE both in term of magnitude and

30
Global and Koopman modes analysis of sound generation in mixing layers

(a)

(b)

110

105

100

SP L(dB)
y/δω (0) 95

90

85

80
-120 -80 -40 0 40 80 120

x/δω (0)

FIG. 19. Reconstruction obtained for 7 Koopman modes: (a) Real part of the pressure fluctuations;
(b) Far-field directivity for an arc of radius 100δw (0) centered at x = 100δω (0): ( ) NLDE;
( • ) Koopman-mode reconstruction.

direction of the wavefronts.

Hence, it is shown that the DMD analysis can provide a low-dimensional representation
of the data generated by an aeroacoustic simulation, using a weigthed sum of few eigenfunc-
tions. Moreover, the analysis associated with Ds appears to be sufficiently accurate to be
combined with a Kirchhoff extrapolation method to obtain the far-field acoustics. This last
remark offers some perspectives in terms of data reduction associated with large aeroacoustic
database.

IV. SUPERSONIC CASE

In this section, the sound generated by Mach wave radiation in a supersonic mixing layer
is investigated.

31
Global and Koopman modes analysis of sound generation in mixing layers

A. Direct computations

1. Problem description

The case considered is a cold supersonic Mach 2.9/1 mixing layer at a Reynolds number
Re = ρ∞ Uc δω (0)/µ∞ = 1317, matching one configuration studied in Cheung and Lele8 . The
thermodynamic conditions are the same in the low- and high-speed streams, as described
for the subsonic configuration in §III A 1.
The extent of the computational domain is (Lx , Ly ) = (480, ±100)δω (0) including the
sponge zone. The Cartesian grid has 750 by 331 grid points in the x- and y-directions.
The grid is uniform in the streamwise direction with spacing ∆xmin = 0.64 δω (0) up to
x = 440 δω (0), and is then stretched with a rate of 4% to form a sponge zone. Along the
normal direction, the minimum mesh size is ∆ymin = 0.08 δω (0) at the centerline y = 0, and
a stretching with a rate of 2% is applied up to y = ±100 δω (0).

2. Base flow

The procedure to compute the base flow is the same as in the subsonic case. Equations
(15) and (16) with a mean convective Mach number Mc = Uc /c∞ = 1.95 and a velocity ratio
R = 0.487 are used to prescribe the initial conditions. The full Navier-Stokes equations (1)
are advanced without any inlet perturbation. The convergence of residual values for 600 000
iterations is depicted in Figure 20(a). A streamwise velocity profile after convergence is
compared to the analytical hyperbolic-tangent profile in Figure 20(b). The good correspon-
dance indicates that the perturbations from the local stability analysis in the inlet plane are
well supported by the computed base flow.

3. NLDE computation

As in the subsonic case, the local linear stability theory allows us to determine the funda-
mental frequency at the inlet f0 = ω/2π with ω the circular frequency ≈ 2π × 0.286Uc /δω (0)
(see Appendix A). To illustrate the sound generated by the instability waves in a supersonic
regime, the flow is forced at the fundamental frequency at each time step.
The simulation is run for 50 000 iterations with δtUc /δω (0) ≃ 0.043. After the transient

32
Global and Koopman modes analysis of sound generation in mixing layers

(a)
(b)
0
1.6

-2 1.4

res 1.2

-4 U/Uc 1
0.8
-6 0.6

0 2 4 6 0.4
-4 -2 0 2 4

iterations ×105 y/δω(0)

FIG. 20. Base flow for the supersonic mixing layer: (a) convergence of normalized residuals for the
conservative variables as a function of time iterations (see figure 1 for line legend); (b) comparison
of streamwise velocity profiles of the base flow (• •) and the hyperbolic-tangent function (15) (
) at the inlet plane.

regime, the near field vorticity is displayed in Figure 21(b) which lead to pressure fluctuations
as depicted in Figure 21(a).
In the absence of vortex pairing, the radiation pattern illustrates Mach wave radiation
in the lower stream as observed by Cheung & Lele8 in a similar supersonic mixing layer.
This behaviour is in agreement with the fact that the fundamental frequency has a subsonic
relative phase speed in the upper stream and a supersonic phase speed in the lower stream
(see Cheung & Lele8 ). To examine, how the linear model based on a global modes decompo-
sition can capture the instability waves and the acoustic behaviour observed in the NLDE
simulation, a global linear stability analysis is performed in the next section.
The evolution of the integrated energy at the frequencies f0 and 2f0 is shown in Figure
22. The fundamental mode forced at the inlet first undergoes an exponential growth and
then grows with a slower slope due to nonlinear saturation. The first harmonic is not forced
and has a very low initial amplitude. It then grows at a greater rate due to nonlinear
interactions.

33
Global and Koopman modes analysis of sound generation in mixing layers

(a)
40
20
y/δω (0) 0
-20
-40
50 100 150 200 250 300

x/δω (0)

(b)

5
y/δω (0) 0

-5
100 150 200 250 300 350

x/δω (0)

FIG. 21. (a) Far-field non-dimensional fluctuating pressure field with contours levels from −0.01
to 0.01 in steps of 0.002 (full lines: positive values, dashed lines: negatives values). (b) Total
non-dimensional spanwise vorticity with contours ranging from 0.05 to 1.5 in steps of 0.05.

B. Linear model: global spectrum, eigenfunctions

The initial perturbation described in the subsonic case is also used for the supersonic
regime. It is initially located at (x0 , y0 ) = (40δω (0), 0). The snapshots are carried out for a
window extending from x = 10δω (0) to x = 340δω (0) and y = ±75δω (0). The global modes
are obtained with a sequence of 380 snapshots which are obtained with a sampling frequency
which satisfies a resolution up to 2f0 in agreement with the Nyquist criterion.
The dimensionless eigenvalues are presented in Figure 23. One may observe that all the
global modes are damped temporally, which is consistent with the globally stable property
of such a flow. In Figure 24, the real part of the pressure fluctuation is shown for ωr ≈ 0.3,
which is closed to the forcing frequency used in the nonlinear regime (ωr ≈ 0.28), and for

34
Global and Koopman modes analysis of sound generation in mixing layers

-2

-4
Em
-6

-8

-10
0 100 200 300 400
x/δω (0)

FIG. 22. NLDE computation for the supersonic case. Integrated modal energy for f0 ( ) and
2f0 ( ).

0.02

ωi -0.02

-0.04

-0.06
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

ωr
FIG. 23. Global spectrum of the supersonic mixing layer. The eigenvalues are made dimensionless
by δω (0) and Uc .

ωr ≈ 0.4.
It is worth noting that a strong Mach wave radiation in the lower stream is also evidenced
for these global modes. To illustrate the latter point, the phase velocity relative to the
ambient sound speed for both circular frequencies is determine under the weakly nonparallel
flow assumption. In Figure 25, these quantites are plotted for the vertical velocity fluctuation
taken at y = 0. Despite some ocillations, the phase velocity is clearly supersonic for the

35
Global and Koopman modes analysis of sound generation in mixing layers

(a)
40
20
y/δω (0) 0
-20
-40
50 100 150 200 250 300

x/δω (0)

(b)

40

y/δω (0) 0

-40
50 100 150 200 250 300

x/δω (0)

FIG. 24. Real part of the pressure component for two global eigenmodes corresponding to (a) N
and (b) • in the spectrum of Figure 23. Full lines: positive values; dashed lines: negatives values.

1.5 1.5

1 1

Mr,k
0.5 0.5

0 0
50 100 150 200 250 300 350 50 100 150 200 250 300 350

x/δω (0) x/δω (0)

(a) (b)

FIG. 25. Phase speeds Mr,1 ( ) and Mr,2 ( ) of the two global eigenmodes marked by:
(a) N and (b) • in the spectrum of Figure 23.

36
Global and Koopman modes analysis of sound generation in mixing layers

lower stream, while the fluctuations in the upper stream are moving with a subsonic phase
speed for both frequencies. Consequently, these linear instability waves will lead to acoustic
radiation. Finally, let us further explore the angle of the Mach wave radiation measured
from the upstream axis for the lower stream. For the global mode whose frequency is closed
to the forcing frequency, the Mach angle is estimated at 53o near x/δω (0) ≈ 325, in good
agreement with the NLDE value. Strong similarities can be noticed between the global mode
in Figure 24(a) and the NLDE simulation in Figure 21(a).
Hence, the solution derived from a linear global stability analysis with appropriate nonre-
flecting boundary conditions gives a series of discrete supersonic waves, which radiates sound
directly through Mach wave mechanism. This supports the weakly nonparallel theory based
on matched asymptotic expansions of Tam & Burton47 dedicated to the noise generated by
spatially growing instability in supersonic mixing layers.
Finally, as discussed in a subsonic regime, these discrete instability waves may be used to
represent the effect of a localized harmonic forcing through an appropriate summation (22).
Since each individual global mode may radiate sound, the asymptotic response derived from
nonmodal interaction will produce some noise.

C. Nonlinear model: Koopman modes

A DMD analysis is performed on the NLDE database described in section IV A 3 by

taking the entire domain. A sequence of 210 snapshots is recorded with a sampling period
which satisfies the Nyquist criterion up to 6f0 . The energy spectrum for dominant modes
and the corresponding Ritz values are displayed in Figure 26.
Due to the saturated regime characterizing the NLDE database, the Ritz values are seen
to lie on a unit circle in Figure 26(b). The most observable modes are associated with the
energy peaks shown in Figure 26(a). As expected, the flow dynamics is mainly dominated by
the fondamental forcing frequency and its first harmonics. In addition, the energy spectrum
shown in Figure 26(a) clearly exhibits a different magnitude distribution for the dominant
Koopman modes compared to the one obtained for the subsonic regime. In particular, the
influence of the nonlinear effects are weaker which results in a small contribution of the har-
monics into the dynamics in comparison with the fundamental frequency. Furthermore, the
zero frequency mode contribution is almost undistinguishable. This provides some insight

37
Global and Koopman modes analysis of sound generation in mixing layers

(a) (b)
1
1

0.8 0.5

E/Emax 0.6
λi 0
0.4
-0.5
0.2

0 -1
-1.5 -1 -0.5 0 0.5 1 1.5 -1 -0.5 0 0.5 1

ωr λr

FIG. 26. DMD analysis for the supersonic mixing layer: (a) Energy spectrum; (b) Ritz values.
The values are made dimensionless by δω (0) and Uc .

y/δω (0) 0

-5
100 150 200 250 300 350
x/δω (0)

FIG. 27. Total non-dimensional spanwise vorticity based on the projection of an instantaneous
snapshot from NLDE onto the Koopman modes which correspond to f0 and 2f0 . Contours range
from 0.05 to 1.5 in steps of 0.05.

into the results depicted in Figure 22 where nonlinear effects will lead to a saturation of the
instability waves with almost no mean flow modification.
The ability of the Koopman modes to capture the near field structures is highlighted
through a reconstruction of the entire flow field based on a projection of the flow dynamics
onto the two dominant Koopman modes, f0 and 2f0 and their complex conjugates. The
total vorticity plotted in Figure 27 is almost identical to the NLDE result of Figure 21(b).
This clearly demonstrates the ability of the Koopman modes to reproduce the near field

38
Global and Koopman modes analysis of sound generation in mixing layers

(a)
40
20
y/δω (0) 0
-20
-40
50 100 150 200 250 300

x/δω (0)

(b)
40
20
y/δω (0) 0
-20
-40
50 100 150 200 250 300

x/δω (0)

FIG. 28. Non-dimensional pressure component of Koopman modes projected onto an instantaneous
snapshot from NLDE: (a) f0 levels range from −0.01 to 0.01; (b) 2f0 levels range from −0.002 to
0.002.

dynamics.
Concerning the acoustic behaviour, the pressure fields for the fundamental and its first
harmonic are displayed in Figure 28. It appears that both the fundamental and its harmonics
generate Mach wave radiation on the lower stream. These waves also propagate in the upper
side but with a lower amplitude and interferences arise due to the propagation effects through
the shear zone. In particular, the strong similarity between the Koopman mode for f0 and
the NLDE simulation is conform with the hierarchy in the energy spectrum of Figure 26(a).
In addition, the radiated noise for the Koopman mode f0 closely follows the results obtained
with the linear global stability analysis. In particular, the Mach angle in the lower stream
and the interference pattern in the upper stream are in agreement with the global mode

39
Global and Koopman modes analysis of sound generation in mixing layers

60

40

20

y/δω (0) 0

-20

-40

-60 -3 -2 -1 0
10 10 10 10

FIG. 29. Comparison of linear and nonlinear models through absolute value of pressure cross
sections for the supersonic mixing layer for the fundamental frequency at x/δω (0) = 300: ( )
Koopman mode; ( ) global mode. The profiles are made dimensionless by their maximum
value.

analysis.

To further illustrate the last comment, cross-sections along the normal direction for the
modulus of the fluctuating pressure are extracted at x/δω (0) = 300 for the fundamental
Koopman mode as well as the global mode which corresponds to ωr ≈ 0.3 (close to f0 ). In
Figure 29, in contrast to the subsonic case, we can see that the strength of the radiated
pressure in lower stream for the global mode is consistent with the Koopman mode. In
addition, the pattern radiated in the upper stream exhibits also similarities. The difference
in amplitude visible in the upper side indicates discrepancies when the Mach waves propagate
through the shear region. The linear mode can indeed not represent the saturated vortices.
Overall, these results clearly show that in a supersonic regime, the radiated pressure is
directly linked to the linear spatially growing instability wave which can be represented
with a linear global-mode analysis.

40
Global and Koopman modes analysis of sound generation in mixing layers

V. CONCLUSIONS AND OUTLOOK

In this paper, the linear and nonlinear mechanisms for noise generation are revisited from
a global point of view. In this context, Krylov’s methods based on snapshots sampled from
linear and nonlinear simulations are employed to extract the dominant global modes and
Koopman modes, respectively. Forced two-dimensional mixing layers are investigated for
two different flow regimes. The subsonic regime is characterized by pairing events at a fixed
location, whereas Mach wave radiation is possible in the supersonic regime. First, a base
flow is obtained from the solution of the full Navier-Stokes solver and is used in both linear
and nonlinear disturbance equation (LDE and NLDE) solvers. This formalism allows the
use of exactly the same discretization in all methods carried out for the flow analysis.

In the subsonic regime, a linear global analysis allows a fair characterization of the Kelvin-
Helmholtz instabilities, considering the low Reynolds number of the flow. In a global frame-
work, these convective modes are stable and the wavepacket dynamics can be retrieved
through the response to an external forcing. The individual modes are not radiating sound
and the noise generation mechanism can not be reproduced since it relies on the deforma-
tion and acceleration of saturated vortices during the pairing event. By contrast, a dynamic
mode decomposition (DMD) based on the NLDE database yields an approximation of the
Koopman mode which contains the full dynamical information. It is shown in particular
that the mixing layer flow and the associated noise can be reproduced with few DMD modes.
The accuracy of the modal decomposition is demonstrated by using snapshots defined on a
subdomain only restricted to the shear region. The far field radiation can be faithfully eval-
uated with a Kirchhhoff’s wave extrapolation method, indicating that all the compressibility
effects are represented by the subdomain modes.

In the supersonic regime, the phase speed of disturbances relative to the ambient sound
speed can be supersonic and linear instabilities directly radiate sound, the so-called Mach
waves, as shown by previous weakly non-parallel stability theories. Consequently, the eigen-
functions obtained from a linear global analysis highlight Mach wave radiation which shows
good agreement with a direct computation. However, the linear analysis fails to reproduce
the complex interferences in the shear zone, associated with subtle details of the instan-
taneous nonlinear flow. On the other hand, the Koopman mode analysis allows a correct
calculation of both near and far fields. The interference pattern visible in the upper stream

41
Global and Koopman modes analysis of sound generation in mixing layers

is notably well reproduced.

Hence, the present analysis provides a general strategy to determine the linear and non-
linear noise generation mechanism associated with instability waves where classical meth-
ods, such as PSE or multiple scales techniques, are inapropriate and/or inapplicable. For
instance, it should be particularly interesting to apply such a strategy in heated or counter-
flow mixing layers46 , subsonic and supersonic hot jets44 or impinging jets23 where upstream
propagating waves may occur and more generally for three-dimensional and turbulent flows.
For the latter, efficient algorithms are needed for the extraction of the dominant modes to
deal with the storage of data collected by sampling the numerical simulation. On one hand,
one may suggest to improve the efficiency of the DMD algorithm by exploiting the parallel
structure of the numerical simulation code. On the other hand, one may also process only
subdomains of the entire flow field. In the framework of aeroacoustics, the revelancy of such
a strategy is highlighted in the present work. For turbulent flows, a similar analysis which
focuses on low frequency unsteadiness in shockwave turbulent boundary layer interaction
22
is carried out by Grilli et al. . By processing only subdomains of the numerical simula-
22
tion, Grilli et al. show that the essential characterisitics of the low-frequency dynamics
may be reduced to the superposition of the four-dominant low-frequency Koopman modes.
Finally, the recent DMD analysis performed on the H-type transition to turbulence in a
40
flat-plate boundary layer by Sayadi et al. allows the identification of three-dimensional
large-scale coherent structures which are responsible for the main contribution to the wall
skin-friction. For that purpose, a composite DMD which rearranges quantities of different
units into the column vectors associated with the data sequence of snapshots is used by
the authors. Hence, a composite DMD combining sound pressure fluctuation measurement
in the far-field and large-scale coherent structures in the near-field could be an interesting
prospect of the present analysis aiming at identifying the contribution of vortical structures
onto the radiated noise for complex turbulent flows.

Appendix A: Local linear stability analysis of the inlet plane

The disturbance vector is

q′ (x, y, t) = q̂(y) ei(αx−ωt) + c.c. (A1)

42
Global and Koopman modes analysis of sound generation in mixing layers

(a) (b)

0.01 1
0.05
2

0 0.8
0 1.5

-0.01 0.6

Re(α)
Im(α)
Re(α)
Im(α)

-0.05 1

-0.02 0.4

-0.1 0.5
-0.03 0.2

-0.15 0
0 0.5 1 1.5 2 -0.04 0
0 0.1 0.2 0.3 0.4 0.5
Re(ω) Re(ω)

FIG. 30. Dispersion relation: (a) subsonic and (b) supersonic mixing layers.

with α = αr + iαi the complex wavenumber and ω the real circular frequency. Only two-
dimensional modes are considered even if oblique modes can be more unstable at supersonic
speeds since the test cases are 2D. The local spatial stability theory can be rewritten into a
generalized eigenvalue problem:

C0 + C1 α + C2 α2 q̂ = 0, with q̂ = (ρ̂, û, v̂, p̂)

The eigenproblem is solved using a Chebyshev collocation spectral method and a QZ algo-
rithm. The base flow is approximated by the streamwise velocity profile (15), which correctly
approximates the solution for self-similar boundary-layer equations. The Crocco-Buseman
relationship (16) is used for the temperature profile.

Both real and imaginary parts of α are plotted as a function of ω in Figure 30. The
spatial amplification rate is given by −αi . Hence, the most unstable mode is obtained for
the circular frequency which leads to the largest spatial amplification rate. The inlet flow
parameters are summarized in Table I. The shape of the corresponding eigenfunctions are
shown in Figure 31. Since the analysis is linear, we need to prescribe an amplitude for the
inlet perturbations, as q′ in = εq′ with q′ given by (A1) and values of ε are reported in Table
I.

43
Global and Koopman modes analysis of sound generation in mixing layers

Case M1 M2 Mc Re δω (0)/m Mode αr αi ω ε

Subsonic 0.5 0.25 0.375 1500 1.67 10−4 fundamental 0.8234 −0.1117 0.8235 10−4
sub-harmonic 0.4006 −0.0849 0.4117 10−4

Supersonic 2.9 1. 1.95 1317 2.82 10−5 fundamental 0.2860 −0.0314 0.2860 4 10−5

TABLE I. Numerical parameters used for the boundary condition at the inlet for the subsonic and
the supersonic mixing layer. ε denotes the forcing amplitude.

(a) (b)

3 6

|u| 5.5
2.5 |v| 5
|p| |u|
|T| 4.5 |v|
|p|
2 4 |T|
Eigenfunctions
Eigenfunctions

3.5

1.5 3

2.5

1 2

1.5

0.5 1

0.5

0 0
-20 -10 0 10 20 -40 -20 0 20 40
y/δω
y/δω

FIG. 31. Eigenfunctions for the most amplified wave: (a) subsonic and (b) supersonic mixing
layers.

Appendix B: Convected Kirchhoff wave extrapolation method

To evaluate far-field noise from the near-field region, a wave extrapolation method based
on Kirchhoff’s formulation26 is used. One may consider a surface Σ, defined by the equation
f (x, t)=0, which encloses the source region. The effect of the mean flow in the observation
region is taken into account by considering the convected wave operator for uniform streams
with Mach numbers M set at the values of the lower or upper streams for the mixing layer
case. The 2-D frequency-domain convected form of the Kirchhoff formulation can be written

44
Global and Koopman modes analysis of sound generation in mixing layers

as

iβ ∂p (y, ω) (2) k k ∂rβ (2) k
Z
′
H(f )p (x, ω) = H0 rβ + 2 p (y, ω) H rβ −
4 f =0 ∂nβ β2 β ∂nβ 1 β2

∂y1 (2) Mk (x1 − y1 )
iM H exp i dΣβ (B1)
∂nβ 0 β2

where ω and k = ω/c∞ are the circular frequency and the wavenumber. x denotes the
(2) (2)
observer position and y is a source point on the extrapolation surface Σ. H0 and H1 are the
Hankel functions of second kind of zeroth and first order, respectively. The Prandtl-Glauert
√ p
transformation is employed with β = 1 − M 2 and rβ = (x1 − y1 )2 + β 2 (x2 − y2 )2 . The
notation ∂/∂nβ means (∂/∂yi )niβ , where n1β = n1 , n2β = βn2 and ni are the component
of the unit outer normal n. Similarly, dΣβ is used for the Prandtl-Glauert transform of the
surface element. H(f ) denotes the Heaviside function, zero inside the extrapolation surface
(f < 0) and one outside (f > 0). Details of this frequency-domain formulation are provided
in19 .
From an algorithmic point of view, the first step is the storage of the aerodynamic quan-
tities during one vortex pairing of the simulation or the reduced order model. In particular,
the variable p is recorded onto a horizontal surface surrounding the source region. Then,
the normal derivatives ∂p/∂y2 is evaluated from the near-field solution with the same DRP
scheme used for the direct solver in section II A. Fast Fourier transforms of p and ∂p/∂y2
allow us to obtain the sound generated in the far field by integrating (B1) with a trapezoidal
rule. Finally, the pressure p′ (x, t) in the temporal space is retrieved by an inverse Fourier
transform.

45
Global and Koopman modes analysis of sound generation in mixing layers

REFERENCES

1
F. Alizard and J.-C. Robinet. Modeling of optimal perturbations in flat plate boundary
layer using global modes: benefits and limits. Theor. Comp. Fluid Dyn., 1-4(25):147–165,
2011.
2
F. Alizard, J.-C. Robinet, and X. Gloerfelt. A domain decomposition matrix-free method
for global linear stability. Computers and Fluids, 1-4(25):147–165, 2012.
3
S. Bagheri, E. Akervik, L. Brandt, and D. S. Henningson. Matrix-free methods for the
stability and control of boundary layers. AIAA J., 47(5):1057–1068, 2009.
4
W. Blumen. Shear-layer instability of an inviscid compressible fluid. J. Fluid Mech.,
40:769–781, 1970.
5
W. Blumen, P. G. Drazin, and D. F. Billings. Shear-layer instability of an inviscid com-
pressible fluid: Part 2. J. Fluid Mech., 71:305–316, 1975.
6
C. Bogey and C. Bailly. A family of low dispersive and low dissipative explicit schemes
for noise computation. J. Comp. Physics, 194:194–214, 2004.
7
C. Bogey, C. Bailly, and D. Juvé. Numerical simulation of sound generated by vortex
pairing in a mixing layer. AIAA Journal, 38(12):2210–2218, 2000.
8
L. C. Cheung and S. K. Lele. Linear and nonlinear processes in two-dimensional mixing
layer dynamics and sound generation. J. Fluid Mech., 625:321–351, 2009.
9
T. Chyczewski, P. Morris, and L. Long. Large-eddy simulation of wall-bounded shear flow
using the nonlinear disturbance equations. In 6th AIAA/CEAS Aeroacoustics Conference,
12-14 June, Lahaina, Hawaii, AIAA Paper 2000-2007, 2008.
10
T. Colonius, SK. Lele, and P. Moin. Sound generation in a mixing layer. J. Fluid Mech.,
330:2761–2778, 1997.
11
C. Cossu and J-M. Chomaz. Global measures of local convective instabilities. Phys. Rev.
Lett., 78(23):4387–4390, 1997.
12
Ph. Meliga D. Sipp, O. Marquet and A. Barbagallo. Dynamics and control of global
instabilities in open flows: a linearized approach. Appl. Mech. Rev., 63:030801, 2010.
13
M. Fosas de Pando, D. Sipp, and P.J. Schmid. Efficient evaluation of the direct and
adjoint linearized dynamics from compressible flow solvers. J. Comp. Physics, 231:7739–
7755, 2012.
14
P. Druault, M. Yu, and P. Sagaut. Quadratic stochatic estimation of far field acoustic

46
Global and Koopman modes analysis of sound generation in mixing layers

pressure with coherent structure events in a 2D compressible plane mixing layer. Int. J.
Num. Meth. in Fluids, 62:906–926, 2010.
15
W. S. Edwards, L. S. Tuckermann, R. A. Friesner, and D. C. Sorensen. Krylov methods
for the incompressible navier-stokes equations. J. Comput. Phys., 110:82–102, 1994.
16
U. Ehrenstein, P.-Y. Passaggia, and F. Gallaire. Control of a separated boundary layer:
reduced-order modeling using global modes revisited. Theor. Comp. Fluid Dyn., 25(1):195–
207, 2011.
17
X. Garnaud, L. Lesshafft, P.J. Schmid, and J.-M. Chomaz. A relaxation method for large
eigenvalue problems, with an application to flow stability analysis. Journal of Computa-
tional Physics, 231(10):3912–3927, 2012.
18
X. Garnaud, L. Lesshafft, P.J. Schmid, and P. Huerre. Modal transient dynamics of jet
flows. Physics of Fluids., 25:044103, 2013.
19
X. Gloerfelt, C. Bailly, and D. Juvé. Direct computation of the noise radiated by a subsonic
cavity flow and application of integral methods. J. Sound Vib., 226(1):119–146, 2003.
20
X. Gloerfelt and T. Le Garrec. Generation of inflow turbulence for aeroacoustic appli-
cations. In 14th AIAA/CEAS Aeroacoustics Conference 15-17 May, Vancouver, Canada,
AIAA Paper 2008-2926, 2008.
21
X. Gloerfelt and P. Lafon. Direct computation of the noise induced by a turbulent flow
though a diaphragm in a duct at low Mach number. Comput. Fluids, 37:388–401, 2008.
22
M. Grilli, P.J. Schmid, S. Hickel, and N.A. Adams. Analysis of unsteady behaviour in
shockwave turbulent boundary layer interaction. J. Fluid Mech., 700:16–28, 2012.
23
C.-M. Ho and N. S. Nosseir. Dynamics of an impinging jet. part 1. the feedback phe-
nomenon. J. Fluid Mech., 105:119–142, 1981.
24
T. L. Jackson and C. E. Grosch. Inviscid spatial stability of a compressible mixing layer.
J. Fluid Mech., 208:609–637, 1989.
25
A. Kierkegaard, E. Åkervik, G. Efraimsson, and D.S. Henningson. Flow field eigenmode
decompositions in aeroacoustics. Computers and Fluids, 39:338–344, 2010.
26
G.R. Kirchhoff. Towards a theory of light rays. Annalen der Physik und Chemie, 18:663–
695, 1883.
27
M. Lesieur, P. Comte, E. Lamballais, O. Métais, and G. Silvestrini. Large-eddy simulations
of shear flows. Journal of Engineering Mathematics, 32:195–215, 1997.
28
L. Lesshafft. Nonlinear global modes and sound generation in hot jets. PhD thesis, Ecole

47
Global and Koopman modes analysis of sound generation in mixing layers

polytechnique, 2006.
29
L. Lesshhaft, P. Huerre, and M. Terracol. Nonlinear global modes in hot jets. J. Fluid
Mech., 554:393–409, 2006.
30
C. J. Mack and P.J. Schmid. Global stability of swept flow around a parabolic body:
features of the global spectrum. J. Fluid Mech., 669:375–396, 2011.
31
I. Mezic. Analysis of fluid flows via spectral properties of the Koopman operator. Annu.
Rev. Fluid Mech., 45:357–378, 2012.
32
P. J. Morris, L. N. Long, A. Bangalore, and Q. Wang. A parallel three-dimensional
computational aeroacoustics method using nonlinear disturbance equations. J. Comput.
Phys, 133:56–78, 1997.
33
P. J. Morris, L. N. Long, and T. E. Scheidegger. Parallel computations of high speed jet
noise. In 5th AIAA/CEAS Aeroacoustics Conference, 10-12 May, Greater Seattle, WA,
AIAA Paper 99-1873, 1999.
34
R. Moser and M. Rogers. The three-dimensional evolution of a plane mixing layer: pairing
and transition to turbulence. J. Fluid Mech., 247:275–320, 1993.
35
J.W. Nichols and S.K. Lele. Global modes and transient response of a cold supersonic jet.
J. Fluid Mech., 669:225–241, 2011.
36
S. A. Ragab and J. L. Wu. Linear instabilites in two-dimensional compressible mixing-
layer. Phys. Fluids A, 1(6):957–966, 1989.
37
J. Ristorcelli and G. Blaisdell. Consistent initial conditions for the DNS of compressible
turbulence. Physics of Fluids, 9:4–6, 1997.
38
C.W. Rowley, I. Mezic, S. Bagheri, P. Schlatter, and D.S. Henningson. Spectral analysis
of nonlinear flows. J. Fluid Mech., 641:115–127, 2009.
39
N. D. Sandham and W. C. Reynolds. Three-dimensional simulations of large eddies in the
compressible mixing layer. J. Fluid Mech., 224:133–158, 1991.
40
T. Sayadi, J.W. Nichols, P.J. Schmid, and M.R. Jovanovic. Dynamic mode decomposition
of h-type transition to turbulence. Center for Turbulence Research proceedings, 2012.
41
P. J. Schmid. Dynamic mode decomposition of numerical and experimental data. J. Fluid
Mech., 656:5–28, 2010.
42
P. J. Schmid and D. S. Henningson. Stabiblity and transition in shear flows. Springer,
2001.
43
P.J. Schmid, L. Li, M.P. Juniper, and O. Pust. Applications of the dynamic mode decom-

48
Global and Koopman modes analysis of sound generation in mixing layers

position. Theor. Comp. Fluid Dyn., 1-4(25):249–259, 2011.

44
J. M. Seiner, M. K. Ponton, and T. R. S. Bhat. Mach wave emission from a high-
temperature supersonic jet. AIAA J., 32(12):2345–2350, 1994.
45
D. Sipp and A. Lebedev. Global stability of base and mean flows: a general approach and
its applications to cylinder and open cavity flows. J. Fluid Mech., 593:333–358, 2007.
46
P. J. Strykowski and D. L. Niccum. The stability of countercurrent mixing layer in circular
jet. J. Fluid Mech., 227:309–343, 1991.
47
C. K. W. Tam and D. E. Burton. Sound generated by instability waves of supersonic flows.
Part 1. Two-dimensional mixing layer. J. Fluid Mech, 138:249–271, 1984.
48
C. K. W. Tam and Z. Dong. Radiation and outflow boundary conditions for direct com-
putation of acoustic and flow disturbances in a nonuniform mean flow. J. Comput. Phys,
4(2):175–201, 1996.
49
C. K. W. Tam and P. J. Morris. The radiation of sound by the instability waves of a
compressible plane turbulent shear layer. J. Fluid Mech, 98(2):349–381, 1980.
50
C. K. W. Tam and J. C. Webb. Dispersion-relation-preserving finite difference schemes
for computational acoustics. J. Comput. Phys, 107(2):262–281, 1993.
51
V. Theofilis. Advances in global linear instability of nonparallel and three-dimensional
flows. Progress in Aerospace Sciences, 39(4):249–315, 2003.
52
V. Theofilis. Global linear instability. Annual Reviews of Fluid Mechanics, 43:319–352,
2011.
53
S. Yamouni, D. Sipp, and L. Jacquin. Interaction between feedback aeroacoustic and
acoustic resonance mechanisms in a cavity flow: a global stability analysis. J. Fluid Mech.,
717:134–165, 2013.

49