Boundary Layer Linear Stability Theoryt
Boundary Layer Linear Stability Theoryt
Leslie M. Mack
Jet Propulsion Laboratory
California Institute of Technology
Pasadena, California 91109
U.S.A.
1984
Contents
Preface 9
1 Introduction 10
1.1 Historical background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2 Elements of stability theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4 Numerical Techniques 35
4.1 Types of methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Shooting methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3 Gram-Schmidt orthonormalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4 Newton-Raphson search procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1
5 Viscous Instability 38
5.1 Kinetic-energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2 Reynolds stress in the viscous wall region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
11 Forcing Theory 99
11.1 Formulation and numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
11.2 Receptivity in high-speed wind tunnels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
11.3 Reflection of sound waves from a laminar boundary layer . . . . . . . . . . . . . . . . . . . . 101
11.4 Table of boundary-layer thicknesses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
2
III Three-Dimensional Boundary Layers 105
12 Rotating Disk - A Prototype 3D Boundary Layer 106
12.1 Mean boundary layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
12.2 Crossflow instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
12.3 Instability characteristics of normal modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
12.4 Wave pattern from a steady point source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3
List of Figures
6.1 Neutral-stability curves for Blasius boundary layer: (a) F vs. R; (b) αr vs. R; (c) c vs. R;
− · −, σmax ; − − −, (A/A0 )max ; both maxima are with respect to frequency at constant R. . 42
6.2 Distribution of 2D spatial amplification rate with frequency in Blasius boundary layer at
R = 600 and 1200. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.3 Maximum 2D spatial amplification rates σmax and σ̂max as functions of Reynolds number for
Blasius boundary layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.4 2D ln(A/A0 ) as function of R for several frequencies plus envelope curve; Blasius boundary
layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.5 Distribution of 2D ln(A/A0 ) with frequency at several Reynolds numbers, and bandwidth of
frequency response as a function of Reynolds number; Blasius boundary layer. . . . . . . . . . 45
6.6 Effect of wave angle on spatial amplification rate at R = 1200 for F × 104 = 0.20, 0.25 and
0.30; Blasius boundary layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.7 Complex group-velocity angle vs. wave angle at R = 1200 for F ×104 = 0.20 and 0.30; Blasius
boundary layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.8 Effect of wave angle on ln(A/A0 ) at several Reynolds numbers for F = 0.20 × 10−4 ; Blasius
boundary layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.9 Eigenfunctions of û amplitude at R = 800, 1200 and 1600 for F = 0.30 × 10−4 ; Blasius
boundary layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.10 Eigenfunctions of û phase at R = 800, 1200 and 1600 for F = 0.30 × 10−4 ; Blasius boundary
layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.11 Energy production term at R = 800, 1200 and 1600 for F = 0.30 × 10−4 ; Blasius boundary
layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.12 2D envelope curves of ln(A/A0 ) for Falkner-Skan family of boundary layers.. . . . . . . . . . . 51
6.13 2D envelope-curve frequencies of Falkner-Skan boundary layers. . . . . . . . . . . . . . . . . . 51
6.14 Frequency bandwidth along 2D envelope curves for Falkner-Skan boundary layers. . . . . . . 52
6.15 Temporal eigenvalue spectrum of Blasius boundary layer for α = 0.179, R = 580. . . . . . . . 53
7.1 Constant-phase lines of wave pattern from harmonic point source in Blasius boundary layer;
F = 0.92 × 10−4 , Rs = 390. [After Gilev et al. (1981)] . . . . . . . . . . . . . . . . . . . . . . 56
7.2 Centerline amplitude distribution behind harmonic point source as calculated by numerical
integration, and comparison with 2D normal mode; F = 0.60 × 10−4 , Rs = 485, Blasius
boundary layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
7.3 Centerline phase distribution behind harmonic point source as calculated by numerical inte-
gration; F = 0.60 × 10−4 , Rs = 485, Blasius boundary layer. . . . . . . . . . . . . . . . . . . . 63
7.4 Comparison of measured and calculated cecnterline amplitude distributions behind harmonic
point source; F = 0.60 × 10−4 , Rs = 485, Blasius boundary layer. . . . . . . . . . . . . . . . . 63
7.5 Spandwise amplitude and phase distribution at R = 700 behind harmonic point source; F =
0.60 × 10−4 , Rs = 485, Blasius boundary layer. . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4
9.1 Phase velocities of 2D neutral inflectional and sonic waves, and of waves for which relative
supersonic region first appears. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . 77
9.2 Multiple wavenumbers of 2D inflectional neutral waves (c = cs ). Insulated wall, wind-tunnel
temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.3 Pressure-fluctuation eigenfunctions of first six modes of 2D inflectional neutral waves (c = cs )
at M1 = 10. Insulated wall, T1∗ = 50 K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.4 Multiple wavenumbers of 2D noninflectional neutral waves (c = 1). Insulated wall, wind-
tunnel temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
9.5 Pressure-fluctuation eigenfunctions of first six modes of 2D noninflectional neutral waves (c =
1) at M1 = 10. Insulated wall, T1∗ = 50 K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
9.6 Effect of Mach number on maximum temporal amplification rate of 2D waves for first four
modes. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . 83
9.7 Effect of Mach number on frequency of most unstable 2D waves for first four modes. Insulated
wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
9.8 Temporal amplification rate of first and second modes vs. frequency for several wave angles
at M1 = 4.5. Insulated wall, T1∗ = 311 K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.9 Temporal amplification rate as a function of wavenumber for 3D waves at M1 = 8.0. Insulated
wall, T1∗ = 50 K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.10 Effect of wave angle on maximum temporal amplification rate of first and second-modes at
M1 = 4.5, 5.8, 8.0 and 10.0. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . 86
9.11 Effect of Mach number on maximum temporal amplification rates of 2D and 3D first-mode
waves. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . 87
9.12 Effect of wall cooling on ratio of maximum temporal amplification rate with respect to both
frequency and wave angle of first and second modes at M1 = 3.0, 4.5, and 5.8 to insulated-wall
maximum amplification rate. Wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . . . 88
9.13 Effect of extreme wall cooling on temporal amplification rates of 2D wave for first four modes
at M1 = 10, T1∗ = 50 K: Solid line, insulated wall; Dashed line, cooled wall, Tw /Tr = 0.05. . . 88
10.1 Comparison of neutral-stability curves of frequency at (a) M1 = 1.6 and (b) M1 = 2.2.
Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
10.2 Effect of Mach number on 2D neutral-stability curves of wavenumber. Insulated wall, wind-
tunnel temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
10.3 Distribution of maximum temporal amplification rate with Reynolds number at (a) M1 = 1.3,
(b) M1 = 1.6, (c) M1 = 2.2 and (d) M1 = 3.0 for 2D and 3D waves. Insulated wall, wind-
tunnel temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
10.4 Distribution of maximum first-mode temporal amplification rates with Reynolds number at
M1 = 4.5, 5.8, 7.0 and 10.0. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . 92
10.5 Neutral-stability curves of wavenumber for 2D first and second-mode waves at (a) M1 = 4.5
and (b) M1 = 4.8. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . 93
10.6 Effect of Reynolds number on maximum second-mode temporal amplification rate at M1 = 4.5,
5.8, 7.0 and 10.0. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . . 94
10.7 Effect of wave angle on second-mode temporal amplification rates at R = 1500 and M1 = 4.5,
5.8, 7.0 and 10.0. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . . 94
10.8 Effect of wall cooling and heating on Reynolds number for constant ln (A/A0 )max at M1 = 0.05. 95
10.9 Effect of wall cooling on 2D neutral-stability curves at M1 = 5.8, T1∗ = 50 K. . . . . . . . . . 96
10.10Effect of Mach number on the maximum temporal amplification rate of first and second-mode
waves at R = 1500. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . 98
10.11Effect of Mach number on the maximum spatial amplification rate of first and second-mode
waves at R = 1500. Insulated wall, wind-tunnel temperatures. . . . . . . . . . . . . . . . . . . 98
11.1 Peak mass-flow fluctuation as a function of Reynolds number for six frequencies. Viscous
forcing theory; M1 = 4.5, ψ = 0◦ , c = 0.65, insulated wall. . . . . . . . . . . . . . . . . . . . . 100
11.2 Ratio of amplitude of reflected wave to amplitude of incoming wave as function of wavenumber
from viscous and inviscid theories; M1 = 4.5, ψ = 0◦ , c = 0.65, insulated wall. T1∗ = 311K. . 101
5
11.3 Ratio of wall pressure fluctuation to pressure fluctuation of incoming wave; M1 = 4.5, ψ = 0◦ ,
c = 0.65, insulated wall. T1∗ = 311K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
11.4 Offset distance of reflected wave as function of frequency at R = 600; M1 = 4.5, ψ = 0◦ , c =
0.65, insulated wall. T1∗ = 311K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6
14.5 Crossflow and streamwise instability at N = 23 (R = 2661). (a) Maximum amplification rate
(with respect to frequency) and frequency as function of wavenumber angle; (b) maximum
amplification rate (with respect to wavenumber) as function of frequency: −·−, incompressible
theory; —, sixth-order compressible theory; 35◦ swept wing. . . . . . . . . . . . . . . . . . . . 133
14.6 Amplification rates of seven zero-frequency wave components in forward instability region of
35◦ swept wing with irrotationality condition applied to wavenumber vector: —, incompress-
ible theory; - - -, sixth-order compressible theory for k1 = 0.35. . . . . . . . . . . . . . . . . . 134
14.7 ln(A/A0 ) of six zero-frequency wave components in forward instability region of 35◦ swept
wing with irrotationality condition applied to wavenumber vector and comparison with SALLY
code; —, incompressible theory; - - -, sixth-order compressible theory for k1 = 0.35; − · −,
eighth-order compressible theory for k1 = 0.35. . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7
List of Tables
3.1 Inviscid eigenvalues of Blasius velocity profile computed with indented contours . . . . . . . . 34
6.1 Effect of ψ on amplification rate and test of transformation rule. F = 0.20 × 10−4 , R = 1200,
ψ = 45◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
10.1 Comparison of temporal amplification rates for 3D waves as computed from sixth-order and
eighth-order systems of equations at several Mach numbers . . . . . . . . . . . . . . . . . . . 96
8
Preface
This document is a reproduction of the original. Effort was made to faithfully reproduce the original
document while incorporating more modern features through LATEX typesetting such as fully hyperlinked
table of contents, list of figures, list of tables, reference citations, and figures references as well as a functional
PDF table of contents. Additional effort was made to double check references and correct them where
necessary. Conversion was performed by Alex Craig ([email protected]) and Jerrod Hofferth,
June 2013. Any errors may be reported. The original may be found under the following DTIC handle:
http://www.dtic.mil/docs/citations/ADP004046.
9
Chapter 1
Introduction
10
experimental test. On the other hand, it is generally accepted that flow parameters such as pressure gradient,
suction and heat transfer qualitatively affect transition in the same manner predicted by the linear theory,
and in particular that a flow predicted to be stable by the theory should remain laminar. This expectation
has often been deceived. Even so, the linear theory, in the form of the e9 , or N-factor, method first proposed
by Smith and Gamberoni (1956) and Van Ingen (1956), is today in routine use in engineering studies of
laminar flow control [see, e.g., Hefner and Bushnell (1979)]. A good introduction to the complexities of
transition and the difficulties involved in trying to arrive at a rational approach to its prediction can be
found in three reports by Morkovin (1968, 1978, 1983), and a review article by Reshotko (1976).
The German investigators were undeterred by the lack of acceptance of the stability theory elsewhere,
and made numerous applications of it to boundary layers with pressure gradients and suction. This work
is summarized by Schlichting (1979). We may make particular mention of the work by Pretsch (1942), as
he provided the only large body of numerical results for exact boundary-layer solutions before the advent of
the computer age by calculating the stability characteristics of the Falkner-Skan family of velocity profiles.
The unconvincing mathematics of the asymptotic theory was put on a more solid foundation by Lin (1945)
and Wasow (1948), and this work has been successfully continued by Reid and his collaborators (Lakin et
al., 1978).
When in about 1960 the digital computer reached a stage of development permitting the direct solution of
the primary differential equations, numerical results were obtained from the linear theory during the next ten
years for many different boundary-layer flows: three-dimensional boundary layers [Brown (1959), following
the important theoretical contribution of of Stuart in Gregory et al. (1955)]; free-convectoin boundary layers
(Kurtz and Crandall, 1962; Nachtsheim, 1963); compressible boundary layers (Brown, 1962; Mack, 1965b,
1969); boundary layers on complient walls (Landahl and Kaplan, 1965); a recomputations of Falkner-Skan
flows (Wazzan et al., 1968a); unsteady boundary layers (Obremski et al., 1969); and heated-wall water
boundary layers (Wazzan et al., 1968b). More recent work has focused on three-dimensional boundary
layers in response to the renewed interest in laminar-flow control for swept wings (Srokowski and Orszag,
1977; Mack, 1977, 1979a,b, 1981; Nayfeh, 1980a,b; Cebeci and Stewartson, 1980a,b; Lekoudis, 1979, 1980).
A notable contribution to linear stability theory that stands somewhat apart from the principal line of
development has been provided by Gaster (1968, 1975, 1978, 1981a,b, 1982a,b) in a series of papers on the
wave packets produced by a pulsed point source in a boundary layer. Gaster’s work on this problem also
includes a major stability experiment (Gaster and Grant, 1975).
There are a number of general references that are helpful to anyone interested in the linear theory. Review
articles are by Schlichting (1959); Shen (1964); Stuart (1963) and Reid (1965). Books are by Lin (1955);
Betchov and Criminale (1967), and Drazin and Reid (1981). Schlichting’s book on boundary-layer theory
(1979) contains two chapters on stability theory and transition, and Monin and Yaglom’s book on tubulence
(1971a; 1971b) contains a lengthy chapter on the same subject, as does the book by White (1974) on viscous
flow theory. Reviews of transition have been given by Dryden (1959); Tani (1969, 1981); Morkovin (1968,
1978, 1983) and Reshotko (1976). An extensive discussion of both stability theory and transition, not all at
high speeds in spite of the title, may be found in the recorded lectures of Mack and Morkovin (1971).
11
Figure 1.1: Typical neutral-stability curves.
At any given distance from the origin of the boundary layer, or better, at any given Reynolds number
Re = U1 x/ν, where ν is the kinematic viscosity, an instability wave of frequency ω will be in one of three
states: damped, neutral, or amplified. The numerical results calculated from the stability theory are often
presented in the form of diagrams of neutral stability which show graphically the boundaries between regions
of stability and instability in (ω, Re) space or (k, Re) space. There are two general kinds of neutral-stability
diagrams to be found, as shown in Fig. 1.1 for a two-dimensional wave in a two-dimensional boundary layer.
In this figure, the dimensionless wavenumber αδ is plotted against Rδ , the Reynolds number based on the
boundary-layer thickness δ. Waves are neutral at those values of αδ and Rδ which lie on the contour marked
neutral; they are amplified inside of the contour, and are damped everywhere else. With a neutral-stability
curve of type (a), all wavenumbers are damped at sufficiently high Reynolds numbers. In this case, the
mean flow is said to have viscous instability. Since decreasing Reynolds number, or increasing viscosity, can
lead to instability, it is apparent that viscosity does not act solely to damp out waves, but can actually
have a destabilizing influence. The incompressible flat-plate (Blasius) boundary layer, and all incompressible
boundary layers with a favorable pressure gradient, are examples of flows which are unstable only through
the action of viscosity. With a neutral-stability curve of type (b), a non-zero neutral wavenumber (αδ)s
exists at Re → ∞, and wavenumbers smaller than (αδ)s are unstable no matter how large the Reynolds
number becomes. A mean flow with a type (b) neutral-stability curve is said to have inviscid instability.
The boundary layer in an adverse pressure gradient is an example of a flow of this kind.
In both cases (a) and (b), all waves with αδ less than the peak value on the neutral-stability curve are
unstable for some range of Reynolds numbers. The Reynolds number Re cr below which no amplification
is possible is called the minimum critical Reynolds number. It is often an objective of stability theory to
determine Re cr , although it must be cautioned that this quantity only tells where instability starts, and
cannot be relied upon to indicate the relative instability of various mean flows further downstream. It is
definitely not proper to identify Re cr with the transition point.
A wave which is introduced into a steady boundary layer with a particular frequency will preserve that
12
frequency as it propagates downstream, while the wavenumber will change. As shown in Fig. 1.1, a wave of
frequency ω which passes through the unstable region will be damped up to (Re)L , the first point of neutral
stability. Between (Re)L and (Re)U , the second neutral point, it will be amplified; downstream of (Re)U it
will be damped again. If the amplitude of a wave becomes large enough before (Re)U is reached, then the
nonlinear processes which eventually lead to transition will take over, and the wave will continue to grow
even though the linear theory says it should damp.
The theory can be used to calculate amplification and damping rates as well as the frequency, wavenumber
and Reynolds number of neutral waves. For example, it is possible to computer the amplification rate as
a function of frequency at a given Re. The neutral-stability curve only identifies the band of unstable
frequencies, but the amplification rate tells how fast each frequency is growing, and which frequency is
growing the fastest. Even more useful than the amplification rate is the amplitude history of a wave of
constant frequency as it travels through the unstable region. In the simplest form of the theory, this result
can be calculated in the form of a ratio of the amplitude to some initial amplitude once the amplification rates
are known. Consequently, it is possible to identify, given some initial disturbance spectrum, the frequency
whose amplitude has increased the most at each Reynolds number. It is presumably one of these frequencies
which, after it reaches some critical amplitude, triggers the whole transition process.
We have divided the following material into three major parts: the incompressible stability theory is in
Part I, the compressible stability theory is in Part II, and three-dimensional stability theory, both incom-
pressible and compressible, is in Part III. The field of laminar instability is a vast one, and many topics that
could well have been included have been left out for lack of space. We have restricted ourselves strictly to
boundary layers, but even here have omitted all flows where gravitational effects are important, low-speed
boundary layers with wall heating or cooling, and the important subject of Görtler instability. Within the
topics that have been included, we give a fairly complete account of what we consider to be the essential ideas,
and of what is needed to understand the published literature and make intelligent use of a computer program
for the solution of boundary-layer stability problems. Attention is concentrated principally on basic ideas,
but also on the formulations which are incorporated into computer codes based on the shooting-method of
solving the stability equations. Only selected numerical results are included, and these have been chosen
for their illustrative value, and not with any pretension to comprehensive coverage. Numerous references
are given, but the list is by no means complete. in particular, a number of USSR references have not been
included because of my unfamiliarity with the Russian language. Much use has been made of a previous
work (Mack, 1969), which is still the most complete source for compressible boundary-layer stability theory.
13
Part I
14
Chapter 2
Formulation of Incompressible
Stability Theory
Ur∗ L∗
R= . (2.2)
ν∗
The dimensionless equations are identical to Eqs. 2.1 except that ν ∗ is replaced by 1/R, and ρ∗ is absorbed
into the pressure scale.
We next divide each flow variable into a steady mean-flow term (denoted by an upper-case letter) and
an unsteady small disturbance term (denoted by a lower-case letter):
When these expressions are substituted into Eqs. 2.1, the mean-flow terms subtracted out, and the terms
which are quadratic in the disturbances dropped, we arrive at the following dimensionless linearized questions
for the disturbance quantities:
∂ui ∂Ui ∂ui ∂p
+ uj + Uj =− + ν∇2 ui , (2.4a)
∂t ∂xj ∂xj ∂xi
∂ui
= 0. (2.4b)
∂xi
15
For a truly parallel mean flow, of which a simple two-dimensional example is a fully-developed channel
flow, the normal velocity V is zero and U and W are functions only of y. The parallel-flow equations, when
written out, are
∂u ∂u ∂u dU ∂p
+U +W +v =− + ν∇2 u, (2.5a)
∂t ∂x ∂z dy ∂x
∂v ∂v ∂v ∂p
+U +W =− + ν∇2 v, (2.5b)
∂t ∂x ∂z ∂y
∂w ∂w ∂w dW ∂p
+U +W +v =− + ν∇2 w, (2.5c)
∂t ∂x ∂z dy ∂z
∂u ∂v ∂w
+ + = 0. (2.5d)
∂x ∂y ∂z
These equations are in separable form, i.e., they permit the normal mode solutions
[u, v, w, p]T = [û(y), v̂(y), ŵ(y), p̂(y)]T exp [i(αx + βz − ωt)] (2.6)
where α and β are the x and z components of the wavenumber vector ~k, ω is the frequency, and û(y), v̂(y),
ŵ(y) and p̂(y) are the complex functions, or eigenfunctions, which gives the mode structure through the
boundary layer, and are to be determined by the ordinary differential equations given below. It is a matter
of convenience to work with the complex normal modes; the physical solutions are the real parts of Eqs. 2.6.
The normal modes are traveling waves in the x, z plane, and in the most general case, α, β, and ω are all
complex. If they are real, the wave is of neutral stability and propagates in the x, z plane with constant
amplitude1 and phase velocity c = ω/k, where k = (α2 + β 2 )1/2 is the magnitude of ~k. The angle of ~k with
respect to the x axis is ψ = tan−1 (β/α). if any of α, β or ω are complex, the amplitude will change as the
wave propagates.
When Eqs. 2.6 are substituted into Eqs. 2.5, we obtain the following ordinary differential equations for
the modal functions:
1 2
D − α2 + β 2 û,
i(αU + βW − ω)û + DU v̂ = −iαp̂ + (2.7a)
R
1 2
D − α2 + β 2 v̂,
i(αU + βW − ω)v̂ + = −Dp̂ + (2.7b)
R
1 2
D − α2 + β 2 ŵ,
i(αU + βW − ω)ŵ + DW v̂ = −iβ p̂ + (2.7c)
R
αû + β ŵ + Dv̂ = 0, (2.7d)
where D = d/dy. For a boundary layer, the boundary conditions are that at the wall the no-slip condition
applies,
û(0) = 0, v̂(0) = 0, ŵ(0) = 0, (2.8a)
and that far from the wall all disturbances go to zero,
û(y) → 0, v̂(y) → 0, ŵ(y) → 0 as y → ∞. (2.8b)
Since the boundary conditions are homogeneous, we have an eigenvalue problem and solutions of Eqs. 2.7
that satisfy the boundary conditions will exist only for particular combinations of α, β, and ω. The relation
for the eigenvalues, usually called the dispersion relation, can be written as
ω = Ω(α, β). (2.9)
There are six real quantities in Eq. 2.9; any two of them can be solved for as eigenvalues of Eqs. 2.7 and 2.8,
and the other four have to be specified. The evaluation of the dispersion relation for a given Reynolds
number and boundary-layer profile (U, W ) is the principal task of stability theory. The eigenvalues, along
with the corresponding eigenfunctions û, v̂, ŵ, and p̂, give a complete specification of the normal modes.
The normal modes, which are the natural modes of oscillation of the boundary layer, are customarily called
Tollmien-Schlichting (TS) waves, or instability waves.
1 The term amplitude will always refer to the peak or rms amplitude, never to the instantaneous amplitude
16
2.2 Non-parallel stability theory
Except for the asymptotic suction boundary layer, most boundary layers grow in the downstream direction,
and even for a wave of constant frequency α, β, û, v̂, ŵ, and p̂ are all functions of x (and z in a general 3D
boundary layer). What we have to deal with is a problem of wave propagation in a nonuniform medium.
Since the complete linearized equations 2.4 are not separable, they do not have the normal modes of Eq. 2.6 as
solutions. The most straightforward approach is to simply set the non-parallel terms to zero on the grounds
that the boundary-layer growth is small over a wavelength, and it is the local boundary-layer profile that
will determine the local wave motion. This approach, called the quasi- or locally-parallel theory, has been
almost universally adopted. It retains the parallel-flow normal modes as local solutions, but is, of course,
an extra approximation beyond linearization and leaves open the question of how important the admittedly
slow growth of the boundary layer really is. It also makes for difficulties in comparisons between theory and
experiment.
The first complete non-parallel theories were developed independently by (in order of journal publication
date) Bouthier (1972, 1973); Gaster (1974) and Saric and Nayfeh (1975). Gaster use the method of successive
approximations; the others used the method of multiple scales. There has been considerable controversy on
this subject, mainly because of the way in which (Saric and Nayfeh, 1975, 1977) chose to present their
numerical results, but it is now generally agreed that the three theories are equivalent. Gaster’s calculations
of neutral-stability curves for the Blasius boundary layer have since been verified to be correct by Van Stijn
and Van De Vooren (1983), and have the additional virtue of being based on quantities that can be measured
experimentally. The calculations show the non-parallel terms to have little effect on local instability except
at very low Reynolds numbers. However, this does not mean that non-parallel effects can be neglected when
dealing with waves over distances of many wavelengths.
In the multiple-scale theory, in addition to the usual “fast” x scale over which the phase changes, there
is a “slow” x scale, x1 = εx, where ε is a small quantity identified with 1/R. The slow scale governs the
boundary-layer growth, the change of the eigenfunctions, and a small additional amplitude modulation. The
disturbances are expressed in the form
Here the mean boundary layer is independent of z, and this is the only kind of boundary layer that we will
consider in this work. Examples are 2D planar boundary layers and boundary layers on a rotating disk, on
a cone at zero incidence and on the infinite-span swept wing.
When Eqs. 2.11 are substituted into Eqs. 2.4 and equal powers of ε collected, the zeroth-order equations
for u(0) , v (0) , w(0) and p(0) are identical to the parallel flow equations 2.5. The normal modes, however, have
the more general form
u(0) (x, y, z, t) = A(x1 )û(0) (x1 , y) exp [iθ(0) (x, z, t)], (2.12)
where the phase function is
Z x
θ(0) (x, z, t) = α(0) (x1 ) dx + β (0) (x1 )z − ω (0) (x1 )t, (2.13)
and A(x1 ) is a complex amplitude modulation function. The dispersion relation also becomes a function of
x1 :
ω (0) = Ω(0) (α(0) , β (0) ; x1 ). (2.14)
The non-parallel theories as developed by Bouthier, Gaster, and Saric and Nayfeh calculate the dispersion
relation only to the zeroth order, just as in the quasi-parallel theory. The next order (ε1 ) enters only as a
solvability condition of the first-order equations. This condition determines the function A(x1 ).
17
We shall use only the quasi-parallel theory in the remainder of this work. Consequently, all of the zeroth-
order quantities are calculated as functions of x in accordance with Eqs. 2.12, 2.13 and 2.14. However, the
quasi-parallel theory cannot determine the quantity A(x1 ) and this is simply set equal to the initial amplitude
A0 . In the non-parallel theory, the product Aû is a unique quantity, independent of the normalization of the
eigenfunction û, that give a precise meaning to the amplitude of the flow variable û as a function of y and
permits direct comparisons of theory and experiment. In the quasi-parallel theory, only the contribution of
the amplitude that comes from the imaginary parts of α, β and ω can be accounted for. The corrections due
to the function A(x1 ) and the x dependence of eigenfunctions are outside of the scope of the theory. This
lack of physical reality in the the quasi-parallel theory introduces an uncertainty in the calculation of wave
amplitude and complicates comparisons with experiment. more on the use of the quasi-parallel theory can
be found in Section 2.6.
and the angle between the direction of the ~k and the x axis is
−1 β
ψ = tan . (2.17)
α
The phase velocity c, which is the velocity with which the constant-phase lines move normal to themselves,
has the magnitude
ωr
c= , (2.18)
k
and is in the direction of ~k. If A represents the magnitude of û at some particular y, say the y for which û
is a maximum, then it follows from Eq. 2.15 that
1 dA
= ωi . (2.19)
A dt
We can identify ωi as the temporal amplification rate. Obviously A could have been chosen at any y, or
for another flow variable besides u, and Eq. 2.19 would be the same. It is this property that enables us to
talk about the “amplitude” of an instability wave in the same manner as the amplitude of a water wave,
even though the true wave amplitude is a function of y and the particular flow variable selected. We may
distinguish three possible cases:
ωi < 0 damped wave,
ωi = 0 neutral wave, (2.20)
ωi > 0 amplified wave.
18
The complex frequency may be written
The real part of c̃ is equal to the phase velocity c, and kc̃i is the temporal amplification rate. The quantity
c̃ appears frequently (as c) in the literature of stability theory. However, it cannot be used in the spatial
theory, and since general wave theory employs only ~k and ω, with the phase velocity being introduced as
necessary, we shall adopt the same procedure.
By analogy with the temporal theory, we may define a real wavenumber vector ~k with magnitude
p
k = αr2 + βr2 . (2.24)
19
2.3.3 Relation between temporal and spatial theories
A laminar boundary layer is a dispersive medium for the propagation of instability waves. That is, different
frequencies propagate with different phase velocities, so that the individual harmonic components of a group
of waves at one time will be dispersed (displaced) from each other at some later time. In a conservative
system, where energy is not exchanged between the waves and the medium, an overall quantity such as the
energy density or amplitude propagates with the group velocity. Furthermore, the group velocity can be
considered a property of the individual waves, and to follow a particular normal mode we use the group
velocity of that mode. Because of damping and amplification, instability waves in a boundary layer do not
constitute a conservative system, and the group velocity is in general complex. However, some of the ideas
of conservative systems are still useful. If we consider an observer moving at the group velocity of a normal
mode, the wave in the moving frame of reference will appear to undergo temporal amplification, while in the
frame at rest, it undergoes spatial amplification. Thus we can write
d d
= cr , (2.31)
dt dxg
where in this argument cr is the magnitude of c~r , the real part of the group velocity vector ~c, and xg is the
coordinate in the direction of ~cr . Therefore, if ωi is the temporal amplification rate, the spatial amplification
rate in the direction parallel to ~cr is immediately given to be
ωi
− (αi )g = . (2.32)
cr
The problem of converting a temporal to a spatial amplification rate was first encountered by Schlichting
(1933a), who used the two-dimensional version of Eq. 2.32 without comment. The same relation was also
used later by Lees (1952), and justified on intuitive grounds, but the first mathematical derivation was given
by Gaster (1962) for the 2D case, and the relation bears his name. Gaster’s derivation is straightforward
and can be generalized to three dimensions with the result given above in Eq. 2.32. It is essential to note
that the Gaster relation is only an approximation that is valid for small amplification rates. Within the
approximation, the frequency and wavenumber of the spatial wave are the same as for the temporal wave.
If we use the complex group velocity in teh above derivation, we arrive at the separate transformations for
constant frequency and constant wavenumber obtained by Nayfeh and Padhye (1979) from another point of
view. In this approach, Eq. 2.32 corresponds to a transformation of constant wavenumber.
We can also make use of Eq. 2.32 to arrive at a useful result for spatial waves. The same argument that
led to Eq. 2.32 also applies to a component of the group velocity. Therefore,
ωi
− (αi )ψ̄ = cos(ψ̄ − φr ), (2.33)
cr
where −(αi )ψ̄ is the spatial amplification rate in the arbitrary direction ψ̄. The quantity φr is the real part
of the complex group velocity angle φ defined by
where cx and cz are the complex x and z components of ~c, and c is the complex magnitude of ~c. Eliminating
ωi /cr by Eq. 2.32, we arrive at
(αi )g
(αi )ψ̄ = . (2.35)
cos(ψ̄ − φr )
This relation, which may appear rather obvious, is not a general relation valid for two arbitrary angles. It
is only valid when one of the two angles is φr . When both angles are arbitrary, a more complicated relation
exists and has been derived by Nayfeh and Padhye (1979). There is also a small change in ~k unless the
group-velocity angle is real. We might close this subject by noting that while the various Nayfeh-Padhye
transformation formulae use the complex group velocity, they too are not exact because the group velocity
is considered to be constant in the transformation. We recommend to the interested reader to examine the
instructive numerical examples given by Nayfeh and Padhye.
20
2.4 Reduction to fourth-order system
Equations 2.7 constitute a sixth-order system for the variables û, v̂, ŵ, p̂, Dû, and Dŵ, as can be shown by
rewriting them as six first-order equations. This system may be reduced to fourth order for the determination
of eigenvalues. One approach is to multiply Eq. 2.7a by α and Eq. 2.7c by β and add, and then multiply
Eq. 2.7c by α and Eq. 2.7a by β and subtract, to arrive at the following system of equations for the variables
αû + β ŵ, v̂, αŵ − β û, and β:
1 2
i(αU + βW − ω)(αû + β ŵ) + (αDU + βDW )v̂ = −i(α2 + β 2 )p̂ + [D − (α2 + β 2 )](αû + β ŵ), (2.36a)
R
1 2
i(αU + βW − ω)v̂ = −Dp̂ + [D − (α2 + β 2 )]v̂, (2.36b)
R
1 2
i(αU + βW − ω)(αŵ − β û) + (αDW + βDU )v̂ = [D − (α2 + β 2 )](αŵ − β û), (2.36c)
R
i(αû + β ŵ) + Dv̂ = 0, (2.36d)
where Eqs. 2.7b and 2.7d have been duplicated for convenience as Eqs. 2.36b and 2.36d. The point to note
is that Eqs. 2.36a, 2.36b, and 2.36d are a fourth-order system for the dependent variables αû + β ŵ, v̂, and
p̂. The fourth variable of this system is αDÛ + βDŵ. The dependent variable αŵ − β û appears only in
Eq. 2.36c. Therefore, we may determine the eigenvalues from the fourth-order system, and if subsequently
the eigenfunctions û and ŵ are needed, they are obtained by solving the second order equation 2.36c.
These transformed equations are of the form of Eqs. 2.7 for a two-dimensional wave (β = 0) in a two-
dimensional boundary layer (W = 0) except for the presence of Eq. 2.38c. We may observe from Eq. 2.7c
that even with β = 0, a ŵ velocity component will exist whenever there is a W because of the vorticity
production term DW v̂.
This in a 3D boundary layer with velocity profiles (U, W ) at Reynolds number R, the eigenvalues of an
oblique temporal wave can be obtained from the eigenvalues of a 2D wave of the frequency in a 2D boundary
layer at the same Reynolds number with the velocity profile of the 3D boundary layer in the direction of
the wavenumber vector. The key result that it is the latter velocity profile that governs the instability was
obtained by Stuart (Gregory et al., 1955) in his classic study of the stability of three-dimensional boundary
layers, and by Dunn and Lin (1955) (see also Lin (1955)) in their study of the stability of compressible
boundary layers. We shall refer to this velocity profile as the directional profile.
21
A slightly different transformation was employed by Squire (1933) and bears his name. Squire’s original
transformation was for a 2D boundary layer and the Orr-Sommerfeld equation (see Section 2.5.1),
α = α̃ cos ψ, (2.40)
but there is no justification for applying Eq. 2.40 separately to the real and imaginary parts of a complex α
when α/α̃ is complex. We are able, however, to derive the correct transformation rule from Eq. 2.35. With
ψ̃ = ψ and α̃i = (αi )ψ ,
(−αi )g = −α̃i cos(ψ − φr ), (2.41a)
and with ψ = 0,
(−αi )g
− αi = . (2.41b)
cos φr
Eliminating (−αi )g , we obtain
cos(ψ − φr )
− αi = −α̃i . (2.41c)
cos φr
Consequently, Eq. 2.40 can be used for αi only when the real part of the group-velocity angle is zero. There
is also a small shift in the wavenumber vector whenever φi 6= 0.
An alternative procedure for spatial waves is to use the equations that result from the transformations of
Eq. 2.39, but to not invoke Eq. 2.40 when α/α̃ is complex. The quantities R̃ and ω̃ are complex, as are Ũ and
22
W̃ for a 3D boundary layer, but this causes no difficulty in a numerical solution. Such a procedure, which
amounts to a generalized complex Squire transformation, was incorporated into the JPL viscous stability
code VSTAB/VSP. The approach with Eqs. 2.36, which has the advantage that no transformations are
needed in determining the eigenvalues, is used in the newer JPL stability codes VSTAB/3D, VSTAB/AF,
and SFREQ/EV. It should be noted that even in the spatial theory, the governing real velocity profile is the
profile in the direction of ~k.
[D2 − (α2 + β 2 )]2 v̂ = iR{(αU + βW − ω)[D2 − (α2 + β 2 )] − (αD2 U + βD2 W )}v̂, (2.42)
v̂(0) = 0, Dv̂(0) = 0,
v̂(y) → 0, Dv̂(y) → 0 as y → 0. (2.43)
When W = 0, Eq. 2.42 reduces to the equation for a 2D boundary layer obtained by Squire (1933). When
in addition β = 0,
(D2 − α2 )2 v̂ = iR[(αU − ω)(D2 − α2 ) − αD2 U ]v̂ (2.44)
This is the Orr-Sommerfeld equation and is the basis for most of the work that has been done in incompress-
ible stability theory. it is often derived from the vorticity equation, in which case v̂ is the eigenfunction of
the stream function. The Orr-Sommerfeld equation is valid for a two-dimensional wave in a two-dimensional
boundary layer. However, the generalized Squire transformation, Eq. 2.39, reduces the 3D equation 2.42
to Eq. 2.44 in the tilde coordinates. Consequently, for 3D boundary layers all oblique temporal waves can
be obtained by solving a 2D problem for the renormalized velocity profile in the direction of the wave
number vector, and when the boundary layer is two-dimensional, for the same velocity profile. The 2D
Orr-Sommerfeld equation and the same transformation can also be used for spatial oblique waves, but in
this case R is complex, and for a 3D boundary layer so is U . The inviscid form of the complex Squire
transformation was used by Gaster (1968) for an unbounded 2D shear flow, and the complete viscous form
by Gaster (1975) for a Blasius boundary layer. When one is not trying to make use of previously computer
two-dimensional eigenvalues, it is perhaps simpler to use Eq. 2.42 to calculate 3D eigenvalues as needed, thus
avoiding transformations in R and ω.
23
Eqs. 2.36 can be written as six first-order equations:
DZ1 = Z2 , (2.46a)
DZ2 = α2 + β 2 + iR(αU + βW − ω) Z1 + R(αDU + βDW )Z3 + iR α2 + β 2 Z4 ,
(2.46b)
DZ3 = −iZ1 , (2.46c)
2 2
i α +β
DZ4 = − Z2 − i(αU + βW − ω) + Z3 , (2.46d)
R R
DZ5 = Z6 , (2.46e)
DZ6 = (αDW − βDU )RZ3 + α2 + β 2 + iR(αU + βW − ω) Z5 .
(2.46f)
The fact that the first four Eqs. 2.46 do not contain Z5 or Z6 confirms that eigenvalues can be obtained
from a fourth-order system even though the stability equations constitute a sixth-order system. It is only
the determination of all the eigenfunctions that requires the solution of the full sixth-order system. The
above formulation is applicable when α and β are complex as well as real, and to 3D as well as 2D boundary
layers. Only the transformations of Eq. 2.37b enter in this formulation, and then only in the definitions
of the dependent variables Z1 , Z2 , Z5 and Z6 . No transformations are involved in determination of the
eigenvalues. Another point to note is that only the first derivatives of U and W appear in Eqs. 2.46 instead
of the second derivatives which are present in the Orr-Sommerfeld equation.
where U1 and W1 are the freestream values of U (y) and W (y). Only the upper signs satisfy the boundary
conditions at y → ∞. The components of the characteristic vector A(1) are
(1) 1/2
A1 = −i α2 + β 2 , (2.50a)
(1) 2 2
A2 = i α + β , (2.50b)
(1)
A3 = 1, (2.50c)
(1) αU1 + βW1 − ω
A4 =i 1/2
, (2.50d)
(α2 + β 2 )
(1)
A5 = 0, (2.50e)
(1)
A6 = 0. (2.50f)
24
For real α, β and ω this solution is the linearized potential flow over a wavy wall moving in the direction
of the wavenumber vector with the phase velocity ω/k. It can be called the inviscid solution, although this
designation is valid only in the freestream.
The components of the characteristic vector A(3) are
(3)
A1 = 1, (2.51a)
(3) 1/2
A2 = α2 + β 2 + iR(αU1 + βW1 − ω)
, (2.51b)
(3) −1/2
A3 = α2 + β 2 − iR(αU1 + βW1 − ω)
, (2.51c)
(3)
A4 = 0, (2.51d)
(3)
A5 = 0, (2.51e)
(3)
A6 = 0. (2.51f)
This solution represents a viscous wave and can be called the first viscous solution.
The characteristic vector A(5) is a second viscous solution and its components are
(5)
A1 = 0, (2.52a)
(5)
A2 = 0, (2.52b)
(5)
A3 = 0, (2.52c)
(5)
A4 = 0, (2.52d)
(5)
A5 = 1, (2.52e)
(5) 1/2
= − α2 + β 2 + iR(αU1 + βW1 − ω)
A6 . (2.52f)
The three linearly independent solutions A(1) , A(3) and A(5) are the key to the numerical method that we
will use to obtain the eigenvalues, as they provide the initial conditions for the numerical investigation.
We can observe that the second viscous solution can also be valid in the boundary layer as a pure mode if
Z1 , Z3 and Z4 are all zero. This follows from Eqs. 2.46. In the notation of Eq. 2.37b, the only non-zero flow
variable, Z5 , is α̃w̃, where in the temporal theory w̃ is the eigenfunction of the fluctuation velocity normal to
~k. But since η = ∂w/∂x − ∂u/∂z is the fluctuation vorticity component normal to the wall, Z5 is also −iη̂,
where η̂ is the eigenfunction of η. This interpretation is valid for both the temporal and spatial theories.
The eigensolutions of the second-order equation 2.46f with Z3 = 0 satisfy the boundary condition η̂(0) = 0
and give the vorticity modes in the boundary layer. These modes were first considered by Squire (1933),
and were proven by him to be always stable. Recently it was shown by Herbert (1983a,b) that the Squire
modes provide an important mechanism of subharmonic secondary instability at low, but finite, amplitudes
of a primary 2D instability wave.
Equation 2.54 is the same as Eq. 2.13. We have left β and ω as functions of the slow scale x1 in order
to make it clear that ∂θ/∂x = α, just as for strictly parallel flow. The eigenvalues α, β and ω satisfy
25
the local dispersion relation Eq. 2.14 and the eigenfunction û(y; x) is also a slowly varying function of x.
Consequently, at each x a different eigenvalue problem has to be solved because of the change in the boundary
layer thickness, or velocity profiles, or, as is usually the case, both. The problem we must resolve is how
to “connect” the possible eigenvalues at each x so that they represent a continuous wave train propagating
through the growing boundary layer.
In a steady boundary layer, which is the only kind that we shall consider, the dimensional frequency of
a normal mode is constant. For a 2D wave in a 2D boundary layer, β = 0, and the complex wavenumber α
in the spatial theory, or the real wavenumber α and the imaginary part of the frequency ωi in the temporal
theory, are obtained as eigenvalues for the local boundary-layer profiles. The only problem here is the
relatively minor one of calculating the wave amplitude as a function of x from the amplification rate, adn
we shall discuss this in Section 2.6.2.
26
2.6.2 Some useful formulae
It is worthwhile at this point to list some formulae that will be of use for stability calculations in growing
boundary layers. Only 2D boundary layers are considered here; 3D boundary layers are taken up separately
in Part III. First, we choose as the length scale,
s
∗ ν ∗ x∗
L = , (2.57)
U1∗ (x∗ )
which is the usual length scale of the Falkner-Skan family of boundary layers, and of many nonsimilar
boundary-layer solutions. Other length scales that have been used are the boundary layer thickness, the
displacement thickness, and the inverse unit Reynolds number. The velocity scale is U1∗ (x∗ ), the local velocity
at the edge of the boundary layer. With these choices, the Reynolds number in the stability equation is
√
r
U1∗ (x∗ )L∗ U1∗ x∗
R= ∗
= ∗
= Re. (2.58)
ν ν
The dimensionless coordinate normal to the wall,
y∗
y= R, (2.59)
x∗
in place of ω as the dimensionless frequency. However, F is also a function of x for anything but a flat-plate
boundary layer. For the Falkner-Skan family of velocity profiles, the dimensionless velocity gradient,
x∗ dU1∗
m= , (2.62)
U1∗ dx∗
is constant and related to the usual Hartree parameter βh (the subscript h is used to avoid confusion with
the waveanumber component β), by
2m
βh = . (2.63)
m+1
The variable dimensionless frequency of constant ω ∗ is
4m
m+1
R0
F (R) = F (R0 ) , (2.64)
R
where R0 is the Reynolds number at the initial x station. When a stability code can handle several frequencies
at once, it is more convenient to use some fixed velocity as the reference velocity so that F will remain constant
for each frequency. For the nonsimilar boundary layers on airfoils, the JPL stability codes use the velocity
in the undisturbed freestream.
27
With L∗ a function of x∗ , the irrotationality condition Eq. 2.56 applies to the dimensional spanwise
wavenumber. For the Falkner-Skan family, the dimensionless β for constant β ∗ is given by
1−m
β(R) R 1+m
= . (2.65)
β(R0 ) R0
We note that for a Blasius boundary layer (m = 0), β increases linearly with R. The dimensional wavenumber
αr∗ is almost, but not quite, constant, because there is a small increase in the phase velocity with increasing R.
As a result, the wave angle ψ increases as the wave travels downstream. This increase is at most a few degrees
for a planar a boundary layer. However, on an axisymmetric body, it is the circumferential wavenumber per
radian that is constant. Thus, neglecting the small decrease in αr∗ , tan ψ is inversely proportional to the
radius. For instance, on a cone, where the radius is increasing, an oblique wave is rapidly converted to a
nearly 2D wave as it travels downstream; on a body with decreasing radius, the effect is reversed.
in accord with Eq. 2.27. Here A0 is the amplitude at the intitial station x∗0 , and the integral is evaluated
with constant ω ∗ and β ∗ . if x∗0 is the start of the instability region for the frequency ω ∗ , ln(A/A0 ) is the
N factor that is the basis of the eN method of transition prediction. As discussed in Section 2.2, A may
represent any flow variable at any y location. It may be helpful to think of A as, say, the maximum value
of |û| in the boundary alyer, as this is a quantity that can be determined experimentally. Along with the
amplitude, the time-independent phase relative to the initial phase at x∗0 , z0∗ is
Z x∗
χ(x) − χ(x0 ) = αr∗ dx∗ + βr∗ (z ∗ − z0∗ ) . (2.67)
x∗
0
The phase is a vital quantity in superposition calculations (Chapter 7), but otherwise it is usually not
computed.
For the Falkner-Skan family, the amplitude ratio in terms of R is
Z R
A 2
ln =− αi dR, (2.68)
A0 m + 1 R0
where the integrand αi is calculated as an eigenvalue with the F of Eq. 2.64 and the β of Eq. 2.65. For a
nonsimilar boundary layer, U1∗ (x∗ ) is not an analytical function, and the integration has to be with respect
to x∗ . A formula that is used in the JPL stability codes is
Z xc
αi U1∗
A
ln = −Rc ∗
dxc , (2.69)
A0 (xc )0 R U∞
28
Chapter 3
The system of first-order equations 2.46, or the Orr-Sommerfeld equation in either 2D or 3D form, Eq. 2.42
or 2.44, governs the motion of linear waves at finite Reynolds numbers. With the highest derivative of
v̂ in the Orr-Sommerfeld equation multiplied by 1/R, which is usually a small quantity, it is apparent
that mathematical and numerical methods of some complexity are required to obtain the eigenvalues and
eigenfunctions. On the other hand, if viscosity is considered to act only in the establishment of the mean
flow, but to have a negligible effect on the instability waves, the equations take on a much simpler form. For
example, the 2D Orr-Sommerfeld equation reduces to
(αU − ω) D2 − α2 − αD2 U v̂ = 0.
(3.1)
This is the fundamental equation of the inviscid stability theory, and is usually referred to as the Rayleigh
equation. It is of second order and so only the two boundary conditions
29
3.1 Inflectional instability
3.1.1 Some mathematical results
There are a number of general mathematical results that can be established in the inviscid theory, in contrast
to the viscous theory where few such results are known. We shall give two which demonstrate that no
unstable or neutral temporal waves can exist unless the velocity profile has a point of inflection. The first
result concerns unstable waves. If we multiply Eq. 3.1 by v̂ ∗ , the complex conjugate of v̂, and then subtract
the complex conjugate of the resultant equation, we obtain
2iωi D2 U |v̂|2
D (v̂ ∗ Dv̂ − v̂Dv̂ ∗ ) − = 0. (3.3)
|αU − ω|2
The first term of Eq. 3.3 can be made more meaningful by relating it to the Reynolds stress, which, in
dimensionless form, is
Z 2π/α
α
τ =− uv dx. (3.4)
2π 0
If we recall the necessity of first taking the real parts of u and v before multiplying, and make use of the
continuity equation, we obtain
ωi D2 U hv 2 i
Dτ = , (3.5)
|αU − ω|2
where (1/2)|v̂|2 e2ωi t has been replaced by hv 2 i, the average over a wavelength of the square of the velocity
fluctuation v.
Eq. 3.5 is a special case of a formula derived by Foote and Lin (1950) (see also Lin, 1954, 1955). When
Eq. 3.5 is integrated from y = 0 to infinity, the Reynolds stress at the wall and in the freestream is zero by
the boundary conditions. Therefore, since D2 U = 0 in the freestream,
Z yδ
hv 2 iD2 U
ωi dy = 0, (3.6)
0 |αU − ω|2
where yδ is the dimensionless boundary-layer thickness. It follows from Eq. 3.6 that if ωi 6= 0, D2 U must
change sign somewhere in the interval 0 < y < yδ . Consequently, it has been proven that the velocity profile
must have a point of inflection for there to be an unstable wave. This result was first obtained by Rayleigh.
Later, Fjortoft (1950) strengthened Rayleigh’s necessary condition to D2 U (U − Us ) < 0 somewhere in the
flow, where Us is the mean velocity at the inflection point. This condition is equivalent to requiring that
the modulus of DU have a maximum for there to be an instability. It is always satisfied in a boundary
layer with an inflection point, because DU → 0 as y → ∞ and |DU | cannot have only a minimum. It
was subsequently proven by Tollmien (1935) that for most of the profiles which occur in boundary layers,
including 3D boundary layers, the condition D2 U = 0 is also sufficient. Another result of Rayleigh, for which
the proof will not be given, established that the phase velocity of an unstable wave always lies between the
maximum and minimum values of U . This result was later generalized by Howard (1961) into an elegant
semicircle theorem which relates both ωr /α and ωi /α to the maximum and minimum values of U .
The second result concerns neutral waves. It follows from Eq. 3.5 that with ωi = 0, the Reynolds stress
must be constant everywhere except for a possible discontinuity at the critical layer yc . When Eq. 3.5
is integrated across the boundary layer, the only contribution to the integral comes from the immediate
neighborhood of yc . Hence,
U (yc +0)
D2 U
Z
2 ωi
τ (yc + 0) − τ (yc − 0) = − hv i lim 2 dU. (3.7)
DU c
ωi →0 U (yc −0) (αU − ωr ) + ωi2
The integration variable has been changed from y to U . In the limit of ωi → 0, the integrand of Eq. 3.7 acts
as a delta function, and the integral has a value of π/α. Consequently,
30
D2 U
π
τ (yc + 0) − τ (yc − 0) = hvc2 i. (3.8)
α DU c
Since τ (yc + 0) and τ (yc − 0) are both zero by the boundary condition, D2 Uc must also be zero, and it has
been proven that a wave of neutral stability can exist only when the velocity profile has a point of inflection.
Furthermore, ωr /α = Uc and the phase velocity of a neutral wave is equal to the mean velocity at the
inflection point.
The chief analytical feature of the Rayleigh equation 3.1 is the singularity at αU = ω. Since ω is in
general complex, so is yc . Of course the mean velocity U is real in the physical problem, but it may be
analytically continued onto the complex plane by a power-series expansion of U or by some other method.
Two approaches to obtaining analytical solutions of the inviscid equation are the power spectra series in α2
used by Heisenberg (1924) and Lin (1945), and the method of Frobenius used by Tollmien (1929). The two
solutions obtained by Tollmien are
D2 U
3
1 D U 2
P1 (y − yc ) = 1 + (y − yc ) + + α2 (y − yc ) + . . . ,
2DU c 6 DU c
3 2 (3.10)
D U D U 1 2 2
P2 (y − yc ) = 1 + − + α (y − yc ) + . . . .
2DU c DU 2 c 2
The first solution is regular, but v̂2 is not in general regular near yc because of the logarithmic term. However,
for a neutral wave D2 Uc is zero, and in this one case v̂2 is also regular.
To summarize what we have learned in this section, for a velocity profile without an inflection point, (e.g.,
the Blasius boundary layer), there can be neither unstable nor neutral waves (save for the trivial solution
α = 0, ω = 0). When there is an inflection point, a neutral wave with a phase velocity equal to the mean
velocity at the inflection point can exist, and in boundary layers unstable waves with phase velocities between
0 and 1 can and will exist.
31
quickly for the great variety of velocity profiles encountered in practice. There are at least two methods
available. In the first, which was developed by Conte and Miles (1959), the integration is restricted to the
real axis and is carried past the critical point by the Tollmien solutions. In the second method, which was
developed by Zaat (1958), the solution is produced entirely by numerical integration, and the critical point is
avoided by use of an indented contour in the complex plane. It is as easy to perform the numerical integration
along such a contour as along the real axis, provided the analytic continuation of U away from the real axis
is available. This approach, except for a difference in the method of analytic continuation, which was used
by Mack (1965a) to integrate the compressible inviscid equations. It was later extended to incompressible
flow, and is incorporated into the JPL inviscid stability code ISTAB.
For numerical integration, Eq. 3.1 is replaced by the two first-order equations for v̂ and p̂ which follow
from Eqs. 2.36 when R → ∞:
α
Dv̂ = (DU v̂ + iαp̂) , (3.11a)
αU − ω
Dp̂ = −i (αU − ω) v̂. (3.11b)
v̂ = e−αy , (3.12a)
ω −αy
p̂ = −i 1 − e , (3.12b)
α
where we have chosen the normalization to agree with Eqs. 2.50. These expressions provide the initial values
for the numerical integration to start at some y = y1 > yδ . For chosen values of α and ωr +iωi , the integration
proceeds from y1 to the wall along the real y axis and an indented rectangular contour around the critical
point when necessary. The velocity U is continued on to the indented contour by a power-series expansion in
y − yc . The necessary derivatives of U are obtained from the boundary-layer equations. A Newton-Raphson
search procedure, in which any two of α, ωr , ωi are perturbed, is used to find the eigenvalues, i.e., an α
and ωr + iωi for which the boundary condition v̂(0) = 0 is satisfied. If α is held constant, then the Cauchy-
Riemann equations can be used to eliminate one perturbation because the function Ω(α) in the dispersion
relation is analytic.
32
3.3.2 Amplified and damped solutions as R → ∞ limit of viscous solutions
The clarification of this point is due to Lin (1945), who showed that if the inviscid solutions are regarded
as the infinite Reynolds limit of viscous solutions, a consistent inviscid theory can be constructed in which
damped solutions exist that are not the complex conjugates of amplified solutions. To achieve this result,
integration along the real axis is abandoned for damped waves. Instead, the path of integration is taken
under the singularity just as it is for the inviscid solutions that are used in the asymptotic viscous theory,
and ln(y − yc ) = ln |y − yc | − iπ for y < yc . For damped waves, the effect of viscosity is present even in the
limit R → ∞, and a completely inviscid solution cannot be valid along the entire real axis. Lin’s arguments
were physical and heuristic, but a rigorous justification was given by Wasow (1948).
It is also possible to arrive at Lin’s result from a strictly numerical approach. In Section 3.2, no mention
was made of how to indent the contour of integration. The two possibilities are shown in Fig. 3.1. For an
inviscid neutral solution (ωi = 0), v̂ is purely imaginary and p̂ is real. It makes no difference if the contour is
indented below the real axis, as in Fig. 3.1a, or above, as in Fig. 3.1b. The same eigenvalue α is obtained in
either case. If ωi 6= 0, the integration can be restricted to the real axis. However, unless D2 U = 0 somewhere
in the boundary layer, there are no amplified solutions, or their complex conjugates the damped solutions.
But if we use contour (a) for damped waves, and contour (b) for amplified waves, both solutions exist even
with D2 U 6= 0. Some eigenvalues computed for the Blasius velocity profile are given in Table 3.1, where the
eigenvalues have been made dimensionless by reference to L∗ [Eq. 2.57], which enters the inviscid problem
through the boundary-layer similarity variable y = y ∗ /L∗ . As can be verified from Eqs. 3.11, the solutions
with ωr − iωi and contour (a) are related to the solutions with ωr + iωi and contour (b) by
Figure 3.1: Alternative indented contours for numerical integration of inviscid equations.
Which option do we pick, (a) or (b)? Since the neutral-stability curve for the Blasius profile is of the
type shown in Fig. 1.1a, waves of all wavenumbers are damped in the limit R → ∞. Consequently, if the
inviscid solutions are required to be the R → ∞ limit of viscous solutions, it is evident that contour (a) must
be used, just as in the asymptotic theory and in agreement with Lin. Without an inflection point, there are
no inviscid amplified solutions. For a velocity profile with D2 U = 0 at ys , where the subscript s refers to
the inflection point, both amplified and damped waves exist for each contour, unlike the Blasius case. The
33
Table 3.1: Inviscid eigenvalues of Blasius velocity profile computed with indented contours
Contour α ωr ωi × 103
(a) 0.128 0.0333 -2.33
(b) 0.128 0.0333 2.33
(a) 0.180 0.0580 -6.80
(b) 0.180 0.0580 6.80
neutral wavenumber is αs , and can be obtained with either contour. With contour (a), the wavenumbers of
the amplified waves are located below αs , and the wavenumbers of the damped waves are located above αs ;
contour (b) gives the opposite results. Comparison with the viscous neutral-stability curve, which is of the
type shown in Fig. 1.1b, shows that contour (b) must be rejected in this case also.
The damped solutions with contour (a) do not exist everywhere on the real axis. According to Lin (1955,
p. 136), there is an interval of the real axis in the vicinity of the critical layer where viscosity will always have
an effect even in the limit of R → ∞, and where the inviscid solution is not a valid asymptotic approximation
to the viscous solution. In the final paragraph of his book, Lin remarked that in this interval the viscous
solution has an oscillatory behavior. This remark was confirmed analytically by Tatsumi and Gotch (1971),
and verified numerically by Davey (1981) at an extremely high Reynolds number using the compound matrix
method.
As a numerical example of damped inviscid eigenvalues, Fig. 3.2 gives −ωi , the temporal damping rate,
as a function of α for the Blasius velocity profile. The calculation was performed along an indented contour
of type (a). The inviscid damping rates are, for the most part, much larger than the viscous amplification
rates. That damped inviscid eigenvalues calculated with a type (a) contour are the R → ∞ limit of viscous
eigenvalues was confirmed numerically by Davey in the paper mentioned in the preceding paragraph. For
α = 0.179, the inviscid eigenvalue is ω/α = 0.32126 − 0.03671i; the viscous eigenvalue computed by Davey
at R = 1 × 106 is ω/α = 0.32166 − 0.03629i.
Figure 3.2: Inviscid temporal damping rate vs. wavenumber for Blasius boundary layer.
34
Chapter 4
Numerical Techniques
35
a problem that required a variable step-size integrator. A severe test of any integrator is to calculate the
discrete eigenvalue spectrum, because the higher viscous modes have rapidly oscillating eigenfunctions. The
fixed step-size integrator had little difficulty in calculating a number of additional temporal modes for plane
Poiseuille flow, and its ultimate failure in a portion of the complex ω/α plane for Blasius flow was caused
by a round-off error problem that apparently cannot be cured by any of the usual methods (Mack, 1976, p.
501).
The early applications of shooting methods suffered from the problem of parasitic error growth. This
growth arises because of the presence of a rapidly growing solution in the direction of integration that is
associated with the large characteristic value λ3 in the freestream, which the numerical round-off error will
follow. The parasitic error eventually completely contaminates the less rapidly growing solution, associated
with the characteristic value λ1 in the freestream. The essential advance in coping with this problem, which
had previously limited numerical solutions to moderate Reynolds numbers, was made by Kaplan and Moore
(1964). The Kaplan method “purifies” the contaminated solution by filtering out the parasitic error whenever
it becomes large enough to destroy the linear independence of the solutions. An illuminating presentation
and application of the Kaplan method may be found in Betchov and Criminale (1967). Three recent methods
that cope exceptionally well with the contamination problem are the Riccati method (Davey, 1977, 1979),
the method of compound matrices of Ng and Reid (1979, 1980), and the method of order reduction (Van
Stijn and Van De Vooren, 1982).
Z (3)
S (3) = q , (4.1)
Z (3)∗ Z (3)
where an asterisk refers to a complex conjugate and {} to a scalar product. The metric adopted for the
vector space is the usual Euclidian norm. The scalar product of Z (1) and S (3) is used to form the vector
(1)∗
(1) Z − S (3)∗ Z (1) S (3)
S = q , (4.2)
S (1)∗ S (1)
which replaces Z (1) , and where S refers to the quantity in the numerator.
The numerical integration continues with S (1) and S (3) in place of Z (1) and Z (3) , and when in turn
|S | exceeds the set criterion of, say, 105 with single precision arithmetic and a 36 bit computer word, the
(3)
orthonormalization is repeated. With homogeneous boundary conditions at the wall, it makes no difference
36
in the determination of the eigenvalues whether the Z (i) or S (i) are used. A linear combination of the two
solutions satisfies the β(0) = 0 boundary condition, but the v̂(0) = 0 boundary condition will in general not
be satisfied unless α, β, and ω satisfy the dispersion relation.
Although the orthonormalization procedure has no effect on the method of determining eigenvalues, it
does complicate the calculation of the eigenfunctions. The solution vectors of the numerical integration are
linear combinations of the original solution vectors Z (1) and Z (3) , and it is necessary to “unravel” these
combinations. Two well-known applications of orthonormalization have been given by Conte (1966) and by
Scott and Watts (1977). The latter authors incorporated their method in the general purpose code SUPORT
that has been used in several stability investigations. A different procedure from either of these was worked
out for the JPL stability codes (1971), and is readily applicable to any order of differential equations.
37
Chapter 5
Viscous Instability
38
Z ∞
2
D̄ = ζ dy (5.7b)
0
is the viscous dissipation. A disturbance will amplify, be neutral, or damp depending on whether P is greater
than, equal to, or less than D. Consequently, there can only be instability if τ is sufficiently positive over
enough of the boundary layer so that the production term can outweigh the dissipation term.
39
have been applied.
The additional longitudinal disturbance velocity uv , which is needed to satisfy the no-slip condition,
induces, through the continuity equation, an additional normal disturbance velocity.
Zy
∂uv
vv (y) = − dy, (5.13)
∂y
0
Outside of the viscous region (y → ∞) vv is independent of y and uv is zero. From Eq. 5.14,
αui (0)
vv (∞) = −(1 − i) √ exp[i(αx − ωt)]. (5.15)
2ωR
The consequences of Eq. 5.15 for the Reynolds stress are as follows. For an inviscid neutral disturbance,
u and v are 90◦ out of phase (see Eqs. 2.50a and 2.50c) and τ is zero. However, for any other disturbance u
and v are correlated, and there is a Reynolds stress. Since uv is zero outside of the wall viscous layer, it can
contribute nothing to the τ there. However, vv persists for some distance outside of the wall layer, and since
it is shifted 135◦ with respect to u, it will produce a Reynolds stress. This Reynolds stress must equal the
Reynolds stress set up by the disturbance in the vicinity of the critical layer, and which, in the absence of
viscosity, would extend to the wall. We have already derived a formula for this stress in Section 3.1 (Eq. 3.8).
The formula for the Reynold stress at the edge of the wall viscous region can be derived from Eq. 5.15.
We find r
1 α
τe = −hui vv i = [ui (0)]2 . (5.16)
2 2ωR
If the ratio τe /hvv2 i is formed, we have r
τe 1 2
= . (5.17)
hvv2 i 2α ωR
A general expression for τ in the wall viscous region can be obtained from Eqs. 5.11 and 5.14, and this
expression would give the increase of τ from zero at the wall to the value given by Eq. 5.17 at the edge of the
viscous region. However, Eq. 5.17 establishes the essential result that τ is positive, and thus viscosity acts as
Taylor thought it would, and builds up a Reynolds tress to match the inviscid Reynolds stress, or, in Taylor’s
precise view, permits the momentum of the disturbance to be absorbed at the wall. According to Eq. 5.7a,
with a positive stress energy will be transferred from the mean flow to the disturbance. consequently, the
wall viscous region, which is formed to satisfy the no-slip boundary condition for the disturbance, has the
effect of creating a Reynolds stress which acts to destabilize the flow. This mechanism must be present to
some extent for all disturbance, but whether a particular disturbance is actually amplified or damped will
depend on the magnitude and distribution of the Reynolds stress through the entire boundary layer, and on
the magnitude of the dissipation term.
As a note of caution, it must be recalled that the preceding analysis rests on the neglect of U in the
wall viscous region. Therefore, we can expect the results to be valid only at high values of R, when the wall
viscous region is thin compared to the boundary-layer thickness, and when the critical layer is outside of the
wall viscous region.
40
Chapter 6
In this chapter, we shall present a number of numerical results which have been chosen to illustrate important
aspects of the theory, as well as the give an idea of the numerical magnitudes of the quantities we have been
discussing in the previous chapters.
41
Figure 6.1: Neutral-stability curves for Blasius boundary layer: (a) F vs. R; (b) αr vs. R; (c) c vs. R; − · −,
σmax ; − − −, (A/A0 )max ; both maxima are with respect to frequency at constant R.
42
Figure 6.2: Distribution of 2D spatial amplification rate with frequency in Blasius boundary layer at R = 600
and 1200.
Figure 6.3: Maximum 2D spatial amplification rates σmax and σ̂max as functions of Reynolds number for
Blasius boundary layer.
43
Figure 6.4: 2D ln(A/A0 ) as function of R for several frequencies plus envelope curve; Blasius boundary layer.
The next quantity to examine is the dimensionless spatial amplification rate σ based on L∗ . This am-
plification rate is shown in Fig. 6.2 for 2D waves as a function fo the dimensionless frequency F at the
two Reynolds numbers R = 600 and 1200. From the definition of the amplification rate in Eq. 2.27, the
fractional change in amplitude over a distance equal to one boundary-layer thickness is σyδ . Thus the most
unstable wave of frequency F = 0.33 × 10−4 at R = 1200 grows by 4.0% over a boundary-layer thickness.
The amplification rate based on ν ∗ /U1∗ , σ̂ = σ/R, gives the fractional wave growth over a unit increment in
Re. Thus this same wave grows by 5.6% over an increment in Re of 10,000.
The maximum amplification rates σmax and σ̂max , where the maxima are with respect to frequency (or
wavenumber) at constant Reynolds number, are shown in Fig. 6.3 as functions of Reynolds number. The
amplification rate σ̂, which gives the wave growth per unit of Reynolds number, peaks at the low Reynolds
number of R = 630. The amplification ate σ̂, which is proportional to the wave growth per boundary-
layer thickness, does not peak until until R = 2740 [calculated by Kümmerer (1973)]. The dimensional
amplification rate is proportional to σ̂ for a fixed unit Reynolds number. Figure 6.3 shows that the decline in
the dimensional amplification rate with increasing x-Reynolds number is almost counteracted by the increase
in Reynolds number. However, if the measure of viscous instability is taken to be the wave growth over a
fixed x∗ increment as expressed by σ̂, then by this criterion the maximum viscous instability occurs at low
Reynolds number.
The logarithm of the amplitude ratio, A/A0 , is shown in Fig. 6.4 for 2D waves as a function of R for a
number of frequencies F . The envelope curve, which gives the maximum amplitude ratio possible at any
Reynolds number, is also shown in teh figure along with the corresponding frequencies. It is this type of
diagram that is used in engineering studies of boundary-layer transition. When ln(A/A0 ), which is often
called the N factor, reaches some predetermined value, say nine as suggested by Smith and Gamberoni
(1956), or ten as suggested by Jaffe et al. (1970), transition is considered to take place, or at least to start.
The distribution of the logarithm of the amplitude ratio with frequency is shown in Fig. 6.5 for several
Reynolds numbers. This figure illustrates the filtering action of the boundary layer. The simultaneous
narrowing of the bandwidth of unstable frequencies and the large increase in amplitude ratio as the Reynolds
number increases means that an initial uniform power spectrum of instability waves tends to a spectrum at
high Reynolds number that has a sharp peak at the most amplified frequency. The inset in Fig. 6.5 gives the
bandwidth, defined as the frequency range over which the amplitude ratio is within 1/e of the peak value,
as a function of Reynolds number.
The Squire theorem (section 2.4.1) has told us that it is a 2D wave that first becomes unstable. Fur-
thermore, at any Reynolds number it is a 2D wave that has the maximum amplification rate and also the
maximum amplitude ratio. Thus the envelope curve of amplitude ratio when all oblique waves are considered
as well as 2D waves is still as shown in Fig. 6.4. However, for a given frequency the 2D wave is not necessarily
the most unstable, as is shown in Fig. 6.6. In this figure, teh spatial amplification rate σ, calculated with
44
Figure 6.5: Distribution of 2D ln(A/A0 ) with frequency at several Reynolds numbers, and bandwidth of
frequency response as a function of Reynolds number; Blasius boundary layer.
Figure 6.6: Effect of wave angle on spatial amplification rate at R = 1200 for F × 104 = 0.20, 0.25 and 0.30;
Blasius boundary layer.
45
Figure 6.7: Complex group-velocity angle vs. wave angle at R = 1200 for F × 104 = 0.20 and 0.30; Blasius
boundary layer.
ψ = 0, is plotted against the wave angle ψ for three frequencies at R = 1200. At this Reynolds number,
the maximum amplification rate occurs for F = 0.33 × 10−4 . Above this frequency, 2D waves are certainly
the most unstable. However, below about 0.26 × 10−4 and oblique wave is the most unstable, and the wave
angle of the maximum amplification rate increases with decreasing frequency.
In the calculations for Fig. 6.6, the complex wavenumber was obtained as a function of the spanwise
wavenumber βr with βi = 0 and the frequency real and constant. Thus the complex group-velocity angle φ
can be readily obtained from ∂α/∂βr (= tan φ), and the results are given in Fig. 6.7 for F × 104 = 0.20 and
0.30. The real part of φ is limited to less than 10◦ , and φi can be either plus or minus. It is evident that
at the maximum of σ, where ∂α/∂βr is real, φi must be zero. With the group-velocity angle known, the
accuracy of the simple relation Eq. 2.35 for σ as a function of ψ can be checked. We choose F = 0.20 × 10−4
and ψ = 45◦ in order to have φ real. Table 6.1 gives k, the wavenumber; σ, the amplification rate parallel
to ψ (both of these are calculated as an eigenvalue); σ(ψ), the component of σ in the x direction for the
specified ψ; and σ(0), the amplification rate in the x direction for ψ = 0 as calculated from Eq. 2.41c, the
spatial-theory replacement for the Squire transformation derived from Eq. 2.35, but with ψ replaced by ψ.
In the latter calculation we have used φr = 9.65◦ , the value obtained with ψ = 0. The transformation woks
very well; the small discrepancies from the correct ψ = 0 value are due to φr being a weak function of ψ
instead of constant as assumed in the derivation.
We observe in Table 6.1 that the real Squire transformation, which is the σ(ψ) entry for ψ = ψ, is in
error by 14.4%, whereas the correct transformation is in error by only 0.1%. When the same calculation is
repeated for the other frequency of Fig. 6.7, F = 0.30 × 10−4 , for which φi = −2.48◦ at ψ = 45◦ instead
of 0◦ as for the frequency of the Table, equally good results are obtained for σ(0) from the transformation.
However, k is no longer constant, but increases with ψ; for ψ = 75◦ it is 0.4% larger than at ψ = 0◦ . Nayfeh
and Padhye (1979) provide a formula for this change.
In Fig. 6.8, ln(A/A0 ) is given at several Reynolds numbers for F = 0.20 × 10−4 as calculated with the
irrotationality condition, Eq. 2.64, applied to the wavenumber vector. The abscissa is the initial wave angle
46
htbp]
Table 6.1: Effect of ψ on amplification rate and test of transformation rule. F = 0.20 × 10−4 , R = 1200,
ψ = 45◦ .
Figure 6.8: Effect of wave angle on ln(A/A0 ) at several Reynolds numbers for F = 0.20 × 10−4 ; Blasius
boundary layer.
47
Figure 6.9: Eigenfunctions of û amplitude at R = 800, 1200 and 1600 for F = 0.30 × 10−4 ; Blasius boundary
layer.
at R = 900. The change in the wave angle from R = 900 to 1900 is 1.7◦ for the wave that has an initial
wave angle of 45◦ . This figure shows that the greater amplification rate of oblique waves in the instability
region near the lower branch of the neutral curve translates into an amplitude ratio that is greater than
the 2D value. However, ln(A/A0 ) for an oblique wave is never more than 0.35 greater than the 2D value.
Figure 6.8 also shows that just as the frequency bandwidth narrows with increasing R, so does the bandwidth
in spanwise wavenumber. Although at the lower Reynolds numbers the response extends to the large wave
angles, at R = 1900 the amplitude ratio is down to 1/e of its 2D value at ψ = 37◦ , and on the envelope curve
this angle will be still smaller. For example, the 1/e amplitude for F = 0.60 × 10−4 at the envelope-curve
Reynolds number (R = 900) occurs at ψ = 29◦ ; for F = 0.30 × 10−4 , at ψ = 26◦ . Even so, it is necessary
when thinking about wave amplitudes in the boundary layer to keep in mind that both a frequency band
and spanwise-wavenumber band must be considered, not just a 2D wave.
So far we have only been considering the eigenvalues and not the eigenfunctions. The eigenfunctions give
the possibility of penetrating further into the physics of instability, and we shall take them up briefly at this
point. Eigenfunctions are readily obtained with any of the current numerical methods, but were difficult to
compute with the old asymptotic theory. The first eigenfunctions were obtained by Schlichting (1935), and
the good agreement of the measurements of Schubauer and Skramstad (1947) with these calculations was a
key factor in establishing the validity of the linear stability theory. The problem now is more one of finding
a reasonable way to present the great mass of numerical data that can be computed, and to extract useful
information from this data. Some progress has been made in the latter direction by Hama et al. (1980). For
different amplitudes of 2D waves, these authors calculated streamline patterns, contours of constant total
velocity, Reynolds stress and all terms of the local spatial energy balance.
Figure 6.9 gives the amplitude of the eigenfunction û of the streamwise velocity fluctuation u at R = 800,
1200 and 1600 for the 2D wave of frequency F = 0.30 × 10−4 . The corresponding phases are given in
Fig. 6.10. As may be seen from Fig. 6.1a, these Reynolds numbers are, respectively, just below the lower
branch of the neutral stability curve, near the maximum of σ, and on the envelope curve of the amplitude
ratio. The eigenfunction normalization of Figs. 6.9 and 6.10 is p̂(0) = (2−1/2 , 0). The eigenfunctions have
not been renormalized to, say, a constant peak amplitude as is often done, in order to emphasize that in the
quasi-parallel theory the normalization is completely arbitrary. Nothing can be learned as to the effect of the
variability of the eigenfunction with Reynolds number on the wave amplitude within the framework of this
theory. Attempts have been made to do this, and plausible looking results obtained, but this approach is
without theoretical justification. It has already been pointed out in Section 2.2 that the meaningful quantity
for the amplitude modulation is the product of A(x1 ) adn the eigenfunction, and this product, which has a
fixed value regardless of the normalization of the eigenfunction, can only be calculated from the nonparallel
theory.
For the wave of Fig. 6.9, the critical layer is at about y = 0.15 and varies only slightly with Reynolds
number. Thus the location of the amplitude peak, which is a strong function of R, is only coincidentally
48
Figure 6.10: Eigenfunctions of û phase at R = 800, 1200 and 1600 for F = 0.30 × 10−4 ; Blasius boundary
layer.
Figure 6.11: Energy production term at R = 800, 1200 and 1600 for F = 0.30 × 10−4 ; Blasius boundary
layer.
49
at the critical point. As R increases, the viscous layer near the wall becomes thinner as expected. The
characteristic phase change of approximately 180◦ in the outer part of the boundary layer has nothing to
do with the 180◦ phase change at the critical layer in the inviscid solution (Eq. 3.9b), but is a kinematical
consequence of a wave with zero amplitude at both the wall and at y → ∞. At some ym greater than the
y of maximum amplitude, where viscosity has little influence, the slope of the streamlines relative to the
phase velocity has a maximum. Thus the velocity-streamtube area relation changes sign, and at all y > ym
the u fluctuation from this effect is opposite in sign to the fluctuation that arises from the wavy motion in
a monotonically increasing velocity profile. At some yb > ym , these two effects can exactly balance for a
neutral inviscid wave, and almost balance for nonneutral, viscous waves. For the latter, as shown in Fig. 6.10,
there is a nearly 180◦ shift in the phase of û. The fact that the phase can either advance or retreat in this
region was first noted by Hama et al. (1980), and its significance, if any, is unknown.
It was shown in Section 5.1 that the kinetic energy of a 2D instability wave is produced by the term
τ dU/dy, where τ is the Reynolds stress built up by the action of viscosity. Reynolds stress distributions have
been given by Jordinson (1970) and Kümmerer (1973), among others. The energy production term is shown
in Fig. 6.11 for the frequency and three Reynolds numbers of Figs. 6.9 and 6.10. The peak production does
not occur at the critical layer at any of the three Reynolds numbers. We see that energy production is by
no means limited to the region between the wall and the critical layer, as might be expected from the simple
theory of Chapter 5. At R = 1200, where the amplification rate is near its maximum, there is significant
energy production over about half of the boundary-layer thickness. In these examples, the Reynolds stress
is positive except for the slightly damped wave at R = 800, where there is a small negative contribution over
the outer 70% of the boundary layer. The damping at R = 800 is due to viscous dissipation, not to a negative
production term. Hama et al. (1980) give an example at low Reynolds number where the production term
is negative over the entire boundary layer.
50
Figure 6.12: 2D envelope curves of ln(A/A0 ) for Falkner-Skan family of boundary layers..
51
Figure 6.14: Frequency bandwidth along 2D envelope curves for Falkner-Skan boundary layers.
amplitude is large enough to initiate transition. A measure of this quantity is given in Fig. 6.14, where a
frequency bandwidth of the 3D waves along the envelope curve, expressed as a fraction of the most amplified
frequency, is shown for the Falkner-Skan family. This bandwidth is not identical to the one in the inset of
Fig. 6.5, as it gives only the frequency range less than the most amplified frequency for which the amplitude
ratio is within 1/e of the peak value. The filtering action of the boundary layer is again evident in the
narrowing of the bandwidth with increasing Reynolds number for a given boundary layer, and we see that
the more unstable adverse pressure-gradient boundary layers have the strongest filtering action.
52
Figure 6.15: Temporal eigenvalue spectrum of Blasius boundary layer for α = 0.179, R = 580.
53
examples with finite-width channels in which the upper boundary moved to y → ∞, and with polynomial
velocity profiles of various orders, that both the semi-infinite flow interval and the continuity of the velocity
profile at the edge of the boundary layer, are responsible for the non-existence of the infinite part of the
discrete spectrum of bounded flows. As a finite discrete spectrum is still unable to represent an arbitrary
disturbance, where are the missing eigenvalues?
It is a not uncommon occurrence in eigenvalue problems to have only a finite discrete spectrum. The
remaining part of the spectrum is then a continuous spectrum. An example is the inviscid stability equation,
which has a continuous spectrum associated with the singularity at the critical layer. It was already suggested
by Jordinson (1971) that the discrete viscous spectrum is supplemented by a continuous spectrum along the
cr = 1 axis. The proof by Lin (1961) that a viscous continuous spectrum cannot exist for a bounded flow
does not apply to an unbounded flow. Mack (1976) supported Jordinson’s expectation by means of a few
numerical calculations of continuous-spectrum eigenvalues, and also showed that the continuous spectrum is
always damped because of the restriction ci < −α/R. A more complete and definitive study of the continuous
spectrum was subsequently carried out by Grosch and Salwen (1978), who are responsible for clarifying many
aspects of this problem. Also a paper by Murdock and Stewartson (1977) must be mentioned. Results for
the discrete spatial spectrum of the Blasius boundary layer have been given by Corner et al. (1976).
54
Chapter 7
55
Figure 7.1: Constant-phase lines of wave pattern from harmonic point source in Blasius boundary layer;
F = 0.92 × 10−4 , Rs = 390. [After Gilev et al. (1981)]
spectrum is over frequency; for a pulsed 3D (point) source, the spectrum is over frequency and spanwise
wavenumber; for a harmonic point source, the spectrum is over spanwise wavenumber. It is usually, but not
always, assumed that the spectral densities are uniform (“white noise” spectra).
The solution for a harmonic point source is obtained by evaluating the integral for the complex amplitude
over all possible spanwise wavenumbers. The most straightforward method is to use direct numerical inte-
gration; a second method is to evaluate the integral asymptotically by the method of steepest descent as was
done for parallel flows by Cebeci and Stewartson (1980a,b), and, in more detail, by Nayfeh (1980a,b). Some
numerical results for Blasius flow were cited by Cebeci and Stewartson (1980b), but within the framework
of the eN method of transition prediction. Only the exponential term of the amplitude was evaluated, and
the saddle-point condition was the one for parallel flow.
Experiments on the harmonic point source have been carried out by Gilev et al. (1981), and by Mack and
Kendall (1983). In these experiments, extensive hot-wire measurements of amplitude and phase were made
in the downstream and spanwise directions in a Blasius boundary layer. In Gilev et al. (1981), a Fourier
analysis of the data yielded the oblique normal modes, but no comparisons with theory were made. One
significant result was the mapping out of the lines of constant phase in the x, z plane as shown in Fig. 7.1.
At least three distinct regions can be identified in this figure. Close to the source, the curvature is convex,
and far away it is concave. In an intermediate region, a “dimple” appears at the center line. A region of
concave curvature gradually extends outward to encompass the entire outer portion of the wave pattern,
while the dimple spreads, flattens and finally disappears. All of these features are duplicated in the wave
pattern calculated by numerical integration (Mack and Kendall, 1983).
Figure 7.1 shows that there is a maximum inclination of each constant-phase line that is much less than
the maximum wave angle of unstable normal modes. This feature follows directly from the method of steepest
56
descent, where the saddle-point condition limits the Reynolds-number dependent maximum wave angle to
40◦ –45◦ . This restriction was noted in unpublished calculations by Mack and by Padhye and Nayfeh (private
communication), as well as by Cebeci and Stewartson (1980b).
In the quasi-parallel theory, amplitude is defined as the integral of the spatial amplification rate, and
is not identified with any particular flow variable or distance y from the wall. In the Gaster-Grant (1975)
experiment, amplitude was measured at the outer peak of the amplitude distribution; in Gilev et al. (1981)
at a fixed y ∗ /δ in the boundary layer, and also at a fixed y ∗ just outside of the boundary layer; and in
Mack and Kendall (1983), at the inner peak of the amplitude distribution. A comparison of the calculated
amplitudes with the measurements thus demonstrates whether the amplitude of the quasi-parallel theory
has any relevance to the point-source problems. Exact correspondence can hardly be expected, if for no
other reason than the fact that the disturbance energy is distributed over an ever increasing boundary-layer
thickness as the waves move downstream.
The integral over all spanwise wavenumbers for the dimensional velocity fluctuation ut (the subscript t
denotes time dependence) from a source of frequency ω ∗ located at x∗s , zs∗ is
Z∞
u∗t (x∗ , z ∗ , t∗ ) ∗ ∗
= exp(−iω t ) g ∗ (β ∗ ) exp [iχ (β ∗ ; x∗ , z ∗ )] dβ ∗ , (7.1)
−∞
where g ∗ (β ∗ ) is the (complex) amplitude distribution function of dimensions velocity × length, the frequency
is real,
∗
Zx
χ(x∗ , z ∗ ) = α∗ (x∗ ; β ∗ , ω ∗ )dx∗ + β ∗ (z ∗ − zs∗ ) (7.2)
x∗
s
is the time-dependent part of the phase, and the wavenumber components α∗ and β ∗ are complex. The
eigenfunctions are ignored so that ut is independent of y ∗ , and ut could equally well be considered as any
other flow variable. This integral will be evaluated below by direct numerical integration, and by and
adaptation of citeauthorGaster1981a’s (1981a; 1982b) asymptotic method.
α∗ ν ∗ β∗ν∗
α̂ = , β̂ = ,
U1∗ U1∗
U ∗ x∗ U ∗z∗
x̂ = 1 ∗ , ẑ = 1 ∗ , (7.3)
ν ν
u∗ 2πg ∗
u = ∗, g= ∗ ,
U1 ν
where u∗ is the time-independent part of u∗t , and the reference velocity is the freestream velocity U1∗ We have
chosen the inverse unit Reynolds number ν ∗ /U1∗ as the reference length so that β̂, as well as β ∗ will satisfy
the irrotationality condition in the simplest form, Eq. 2.56. With these choices, the dimensionless x̂ and ẑ
are the usual x and z Reynolds numbers. The reason for the normalization constant 2π in the definition of
g will appear in Section 7.4. With the definitions of Eqs. 7.2, Eq. 7.1 becomes
Z∞
1 h i
u(x̂, ẑ; F ) = g(β̂) exp iχ β̂; x̂, ẑ dβ̂. (7.4)
2π
−∞
57
We take β̂ to be real for convenience, which means that we are going to sum over spatial normal modes of
the type we have been using all along. If we write
χ = χxr + iχi + β̂ ẑ, (7.6a)
where
Zx̂ Zx̂
χxr = α̂r dx̂, χi = α̂i dx̂, (7.6b)
x̂s x̂s
and
Zπ
1
ui (x̂, ẑ) = g β̂ exp (−χi ) sin (χxr ) cos β̂ ẑ dβ̂. (7.7b)
π
0
We have taken advantage of the symmetry in β̂ of g(β̂), χxr and χi to restrict the interval of integration to
the positive β̂ axis. Equations 7.7 are the specific integrals to be evaluated by numerical integration. It is
convenient to present the numerical results in terms of the peak, or envelope, amplitude
q
A (x̂, ẑ) = u2r + u2i , (7.8a)
58
The limit x̂ → ∞ is taken with ẑ/(x̂ − x̂s ) held constant. The condition for the saddle point β̂c is
∂Φ
= 0, (7.11)
∂ β̂
which is equivalent to the two real conditions
Zx̂
∂ α̂
dx̂ = −ẑ, (7.12a)
∂ β̂ r
x̂s
Zx̂
∂ α̂
dx̂ = 0. (7.12b)
∂ β̂ i
x̂s
These integrals are evaluated with the complex β̂ held constant, so that we are dealing with spatial waves
that satisfy the generalized irrotationality condition of kinematic wave theory.
The saddle-point conditions of Eq. 7.12 are of the same type as introduced by Gaster (1981a, 1982b) for a
2D wave packet in a growing boundary layer. Usually the saddle-point method is applied to problems where
the wave-propagation medium (here the boundary layer) is independent of x̂, but Gaster demonstrated the
correctness of the present procedure when the medium is a function of x̂. In a strictly parallel flow, the
boundary layer meets the more restricted requirement of x̂ independence, and the saddle-point conditions
simplify to
∂ α̂ ẑ
=− , (7.13a)
∂ β̂ r (x̂ − x̂s )
∂ α̂
= 0. (7.13b)
∂ β̂ i
For a constant-frequency wave,
∂ω
∂ α̂ ∂ β̂
=− = − tan φ (7.14)
∂ β̂ ∂ω
∂ α̂
where φ is the complex angle of the group-velocity vector, and we see that the parallel-flow saddle-point
condition is equivalent to requiring the group-velocity angle to be real. Consequently, the observed wave
pattern in a parallel flow consists of waves of constant complex spanwise wavenumber β̂c moving along
group-velocity trajectories in the real x̂, ẑ plane. This saddle-point condition has been applied to a browing
boundary layer by Cebeci and Stewartson (1980a,b) and by Nayfeh (1980a,b). This procedure can yield
satisfactory results in a restricted region of the x̂, ẑ plane, but cannot be valid everywhere as the correct
asymptotic representation of Eq. 7.9 is in terms of Eq. 7.12 saddle points rather than Eq. 7.13 saddle points.
The “rays” defined by Eq. 7.12 are not physical rays in the usual sense. For a complex β̂c that satisfies
Eq. 7.12, ẑ is complex at all x̂ > x̂s except at the final, or observation point. The trajectory that is traced
out in the x̂, ẑ plane by satisfying Eq. 7.12 at successive x̂ > x̂s for the same (β̂c )r has a different (β̂c )i at
each point. In a parallel flow, a single normal mode defines an entire ray; here a single normal mode defines
only a single point.
With Φ expanded in a power series in β̂ − β̂c , and with only hte first nonzero term retained (assuming it
is the second derivative), Eq. 7.9 becomes
" #
1 h i Z β̂c ∂ 2 Φ
u= exp (x̂ − x̂s ) Φ β̂c exp (x̂ − x̂s ) β̂ − β̂c dβ̂. (7.15)
2π 2 ∂ β̂ 2
C
We write
∂2Φ
β̂c = D̂ exp (iθd ) , (7.16a)
∂ β̂ 2
β̂ − β̂c = ±is exp (iθs ) , (7.16b)
59
where s is the path length measured from teh saddle point, and θs is its inclination. With the contour C
selected to pass through β̂c from left to right at the constant angle θs = −θd /2, the final result is
1/2
1 h i π θd
u(x̂, ẑ) = (x̂ − x̂s ) D̂ exp (x̂ − x̂s ) Φ β̂c exp i − . (7.17)
2π 4 2
Replacing D̂, x̂, α̂ and β̂ by D, R, α and β, where the reference length is L∗ of Eq. 2.57, we obtain
r
2
u(R, ẑ) = exp (−χi + iχr ) , (7.18)
πD
where R
2
Z
∂ α
D = 2 R2 2 dR ,
(7.19)
∂β
Rs
ZR
(βc )r ẑ π θd
χr = 2 αr (R; βc ) dR + + − , (7.20a)
R 4 2
Rs
ZR
(βc )i ẑ
χi = 2 αi (R; βc ) dR + , (7.20b)
R
Rs
and θd is the argument of the complex integral in Eq. 7.19. We continue to use ẑ for the z-Reynolds number.
In these variables, the saddle-point conditions are
ZR
∂α
2 R dR = −ẑ, (7.21a)
∂β r
Rs
ZR
∂α
R dR = 0. (7.21b)
∂β i
Rs
With the parallel-flow saddle-point conditions of Eqs. 7.13, Eq. 7.18 is still valid, but D and θd have different
meanings. With ∂ 2 α̂/∂ β̂ 2 constant, s
∂2α
D= R (R2 − Rss ) , (7.22)
∂β 2
and θd is the argument of ∂ 2 α̂/∂ β̂ 2 rather than of its integral.
For a given R and ẑ, a double iteration procedure is needed to find the complex β̂c that satisfies Eq. 7.12.
As each iteration involves the recalculation of eigenvalues and ∂ 2 α̂/∂ β̂ 2 from Rs to R, the computational
requirements are large. If only R is given, then an iteration of β̂i for a sequence of β̂r will produce the wave
pattern at that R with much less computation, but the specific ẑ at which the amplitude and phase are
calculated will not be known in advance. Or, both β̂r and β̂i can be specified, and R advanced until the
integral in Eq. 7.12b changes sign. This will not always happen, but when it does, a saddle point and its
location in the R, ẑ plane are obtained without iteration.
Because of the iteration requirement, the saddle-point method is less suited than numerical integration
to the detailed calculation of the entire wave pattern, but it can more readily produce results at just a few
locations. Its greatest advantage, however, is that along the centerline (z = 0) the amplitude and phase can
be obtained at a specified R without iteration, and a single integration pass from Rs to R produces results
at all intermediate R at which eigenvalues are calculated. This is possible because the saddle point is at
β̂ = 0 all along the centerline, and only Eq. 7.19 has to be used, and not Eqs. 7.12. We can also note that
there is no real saving by using the approximate Eq. 7.22 in place of Eq. 7.19, because ∂ 2 α/∂β 2 has to be
calculated in any case, and only the numerical integration of this derivative is eliminated.
60
7.4 Superposition of point sources
We can imagine sources of instability waves to occur not just as single point sources, but as multiple point
sources and as distributed sources. For several discrete sources, the formulas of the preceding Section apply,
and we just have to add the contributions from various sources. We can use this same approach for distributed
sources: The distributed source is represented by discrete, closely spaced, infinitesimal point sources. In this
Section, we apply this idea to line sources.
We replace the function g ∗ (β ∗ ) in Eq. 7.1 with a more general function
1 ∗ ∗ ∗
g ∗ (β ∗ , x∗s , zs∗ ) = us (xs , zs ) ∆ξs∗ g β̂ , (7.23)
2π
where u∗s , the source strength, has the same dimensions as u∗t , and ξs∗ is the arc length along the source. We
substitute Eq. 7.23 into Eq. 7.1 without the time factor, use the definitions of Eq. 7.3, adn arrive at
Z∞
1
∆u(x̂, ẑ) = us ∆ξs g β̂ exp(iχ)dβ̂ (7.24a)
2π
−∞
for the contribution to u at x̂, ẑ of an infinitesimal line source at x̂s , ẑs . In Eq. 7.24a, us = u∗s /U1∗ , ξs =
ξs∗ U1∗ /ν ∗ , and
Zx̂
χ (β; x̂, ẑ) = α̂ dx̂ + β̂ (ẑ − ẑs ) . (7.24b)
x̂s
A finite-length source which extends from s1 = (x̂s , ẑs )1 to s2 = (x̂s , ẑs )2 will produce at x̂, ẑ the velocity
Zs2 Z∞
1
u(x̂, ẑ) = us dξs g β̂ exp(iχ) dβ̂, (7.25)
2π
s1 −∞
The integral over ẑs must converge because the β̂ integral is just the point-source solution Eq. 7.3. A physical
interpretation of Eq. 7.26 is that Eq. 7.3 can be regarded as either the distribution of u with respect to ẑ
at the observation station x̂ due to a single source at x̂s , 0, or as the variation of u at the single observation
point x̂, 0 as the point source at x̂s moves from ẑs → −∞ to ẑs → ∞. Consequently, if the point-source
solution is weighted by us and integrated with respect to ẑs , the resultant amplitude and phase must be that
produced by an infinite-length spanwise line source.
At x̂ = x̂s , the phase function χ reduces to β̂(ẑ − ẑs and Eq. 7.26 becomes
Z∞ Z∞
1 h i
u (x̂s , ẑ) = us dẑs cos β̂ (ẑ − ẑs ) dβ̂. (7.27)
2π
−∞ −∞
Therefore, u(x̂, ẑ) = us as it should, and we see the reason for the factor 2π in the definition of the function
g in Eqs. 7.3 and 7.23. Thus when applied to an infinite-length line source of constant amplitude As and
61
Figure 7.2: Centerline amplitude distribution behind harmonic point source as calculated by numerical
integration, and comparison with 2D normal mode; F = 0.60 × 10−4 , Rs = 485, Blasius boundary layer.
of constant phase, Eq. 7.25 must yield the amplitude ratio A/As of a 2D normal mode. This property of
the point-source solution offers a convenient check on numerical results. Furthermore, if us = As sin(β̂s ẑs )
(standing wave) or As exp(iβ̂s ẑs ) (traveling wave), Eq. 7.25 will give the amplitude ratio of an oblique normal
mode of spanwise wavenumber β̂s . Applications of Eq. 7.25 to finite-length 2D and oblique line sources have
been given by Mack (1984a).
62
Figure 7.3: Centerline phase distribution behind harmonic point source as calculated by numerical integra-
tion; F = 0.60 × 10−4 , Rs = 485, Blasius boundary layer.
Figure 7.4: Comparison of measured and calculated cecnterline amplitude distributions behind harmonic
point source; F = 0.60 × 10−4 , Rs = 485, Blasius boundary layer.
63
Figure 7.5: Spandwise amplitude and phase distribution at R = 700 behind harmonic point source; F =
0.60 × 10−4 , Rs = 485, Blasius boundary layer.
The centerline amplitude distribution has also been calculated from Eq. 7.18 of the extended saddle-point
method. Starting at about R = 650, the saddle-point results are virtually identical with those obtained from
numerical integration in both amplitude and phase. Even the parallel-flow saddle-point method gives a good
result to about the region of maximum amplitude, after which there is a slight departure. Consequently,
Eq. 7.18 gives us a way to obtain the centerline amplitude accurately everywhere except quite close to the
source with only a little more calculation than is needed to obtain the normal-mode A/A0 .
The important question now is whether or not the amplitude distribution of Fig. 7.2 has anything to do
with an experimentally determined amplitude. The answer is given in Fig. 7.4 (Mack and Kendall, 1983). For
the same conditions as the calculations, a hot-wire anemometer was moved downstream in a Blasius boundary
layer. At each Reynolds number station, the maximum fluctuation amplitude in the boundary layer was
determined by a vertical traverse of the hot wires. The source strength was well within the range for which
the response at the hot wire varied linearly with the source amplitude. The amplitude in Fig. 7.4 is the actual
measured amplitude expressed as a fraction of the freestream velocity. The level of the calculated amplitude
has been adjusted accordingly. The calculated amplitude increases more rapidly than in the experiment,
but the Gaster correction for boundary-layer growth makes the two amplitude distributions identical up to
about R = 890, where the measurements depart abruptly from the theory. This disagreement was traced
to a favorable pressure gradient on the flat plate that started precisely at the point of departure. The good
agreement in this one example of the calculation with the Gaster growth correction and the measurement in
the zero pressure-gradient region, while hardly conclusive, does suggest that when dealing with wave motion
over many wavelengths, the growth at the boundary layer cannot be neglected.
The off-centerline wave pattern is of considerable complexity, as shown by Gilev et al. (1981). The peak
amplitude occurs initially off centerline, and it is only well downstream of the source that it is found on
the centerline. A typical calculated spanwise amplitude and phase distribution is shown in Fig. 7.5. The
complex evolution of the phase that appears in Fig. 7.1 is reproduced quite closely by Eq. 7.7, but the
calculated off-centerline amplitude is less exact. Indeed, the saddle-point method, even in its extended form,
fails to give off-centerline amplitude peaks of sufficient magnitude, and only agrees well with the numerical-
integration results after these peaks have disappeared. The parallel-flow saddle-point method fails badly in
calculating the off-centerline wave pattern. The difficulty of correctly computing the amplitude with the
present methods is probably related to the complicated nature of the eigenfunctions. In order for amplitude
calculations to agree as well with experiment as do the phase calculations, it will be necessary to include
the eigenfunctions in the calculations. However, even with this limitation, the numerical-integration method
does remarkably well in reproducing the measured wave pattern, and provides another example of the utility
of linear stability theory in dealing with point source problems.
64
Part II
65
Chapter 8
66
and phase velocity, whereas in the compressible theory there is an infinite sequence of wavenumbers for each
phase velocity whenever the mean flow relative to the phase velocity is supersonic (Mack, 1963, 1964, 1965b,
1969; Gill, 1965b). These additional solutions are called the higher modes. They are of practical importance
for boundary layers because it is the first of the additional solutions, the second mode, that is the most
unstable according to the inviscid theory. Above about M1 = 4, it is also the most unstable at almost all
finite Reynolds numbers.
Subsequent to the work of Lees and Lin, a report of Lees (1947) presented neutral-stability curves for
insulated-wall flat plate boundary layers up to M1 = 1.3, and for cooled-wall boundary layers at M1 = 0.7.
This report also included the famous prediction that cooling the wall acts to stabilize the boundary layer.
However, this prediction must be considerably modified because of the existence of the higher modes. These
modes require for their existence only a region of supersonic relative flow, and thus cannot be eliminated by
cooling the wall. Indeed, the are actually destabilized by cooling (Mack, 1965b, 1969).
where
!
1 ∂ ū∗i ∂ ū∗j
ē∗ij = + , (8.2a)
2 ∂x∗j ∂x∗i
∗ ∗ ∗ 2 ∗ ∗
∗ ∗
τ̄ij = 2µ̄ ēij + λ̄ − µ̄ ēkk − p̄ δij . (8.2b)
3
Again asterisks denote dimensional quantities, overbars time-dependent quantities, and the summation con-
vention has been adopted as in Chapter 2. The equations are, respectively, of momentum, continuity, energy
and state. The quantities which did not appear in the incompressible equations are T̄ ∗ , the temperature; κ̄∗ ,
the coefficient of thermal conductivity; R∗ , the gas constant; c∗v , the specific heat at constant volume, which
will be assumed constant; and λ̄∗ , the coefficient of second viscosity (= 1.5× bulk viscosity coefficient).
The stability equations are obtained from the Navier-Stokes equations by the same procedure that we
used for incompressible flow in Section 2.1. First, all quantities are divided into mean flow and fluctuation
terms. With primes used to denote fluctuations of the transport coefficients,
ū∗ = U ∗ + u∗ , p̄∗ = P ∗ + p∗ ,
T ∗ = T ∗ + θ∗ , ρ̄∗ = ρ∗ + %∗ , (8.3)
∗ ∗ 0∗ ∗ ∗ 0∗ ∗ ∗ 0∗
µ̄ = µ + µ , κ̄ = κ + κ , λ̄ = λ + λ ,
where the first variable on each RHS is a steady mean-flow quantity, and the second is an unsteady fluctuation.
Next, the equations are linearized, the mean-flow terms are subtracted out, and, finally, the parallel-flow
assumption is made. The resulting equations are then made dimensionless with respect to the local freestream
velocity U1∗ , a reference length L∗ , and the freestream values of all state variables (including pressure). Both
viscosity coefficients are referred to µ∗1 and κ∗ is referred to c∗p µ∗1 , where c∗p is the specific heat at constant
67
pressure. The transport coefficients are functions only of temperature, so that their fluctuations can be
written
dµ dκ dλ
µ0 = θ, κ0 = θ, λ0 = θ. (8.4)
dT dT dT
Therefore, µ, κ, and λ in the following equations, along with ρ, are mean-flow quantities, not fluctuations.
The dimensionless, linearized x-momentum equation is
∂u ∂u dU ∂u
ρ + U +v +W =
∂t ∂x dy ∂z
∂2u
2
∂ u ∂2u ∂2v ∂2w
1 ∂p 1
− + 2µ 2 + µ + 2 + +
γM12 ∂x R ∂x ∂y 2 ∂z ∂x∂y ∂x∂z
2
∂2v ∂2w
2 ∂ u dµ dT ∂u ∂v
+ (λ − µ) + + + +
3 ∂x2 ∂x∂y ∂x∂z dT dy ∂y ∂x
2
d2 µ dT dU
dµ d U dU ∂θ
+ θ+ + θ . (8.5a)
dT dy 2 dy ∂y dT 2 dy dy
68
The energy equation is
∂θ ∂θ dT ∂θ
ρ + U +v +W =
∂t ∂x dy ∂z
γµ ∂ 2 θ ∂2θ ∂2θ 1 dκ d2 T
∂u ∂v ∂w
− (γ − 1) + + + + + + θ
∂x ∂y ∂z Pr R ∂x2 ∂y 2 ∂z 2 κ dT dy 2
2 #
1 d2 κ dT
2 dκ dT ∂θ 2 1 dU ∂u ∂v
+ + θ + γ(γ − 1)M 1 2µ +
κ dT dy ∂y κ dT 2 dy R dy ∂y ∂x
2 2 #
dW ∂v ∂w dµ dU dµ dW
+2µ + + θ+ θ .. (8.5e)
dy ∂y ∂y dT dy dT dy
% θ
p= + . (8.5f)
ρ T
Previously undefined quantities which appear in these equations are M1 , the local Mach number at the edge
of the boundary layer; γ, the ratio of specific heats; and Pr = c∗p µ∗ /κ∗ , the Prandtl number, which is a
function of temperature. Equations 8.5 are the compressible counterparts of the incompressible stability
equations 2.5, and are valid for a 3D disturbance in a 3D mean flow. It should be noted that unlike most
compressible stability analyses, Eq. 8.5e, the energy equation, is valid for a variable Prandtl number. The
constant Prandtl number form is recovered by replacing κ with µ in the three terms in which it occurs.
The boundary conditions at y = 0 are
The boundary conditions on the velocity fluctuations are the usual no-slip conditions, and the boundary
condition on the temperature fluctuation is suitable for a gas flowing over a solid wall. For almost any
frequency, it is not possible for the wall to do anything other than to remain at its mean temperature. The
only exception is for a stationary, or near stationary, crossflow disturbance, when θ(0) = 0 is replaced by
Dθ(0) = 0. The boundary conditions at y → ∞ are
This boundary condition is less restrictive than requiring all disturbances to be zero at infinity, but in
supersonic flow waves may propagate to infinity and we wish to include those that do so with constant
amplitude.
where we have adopted the quasi-parallel form of the complex phase function. The normal modes may grow
either temporally or spatially or both, depending on whether ω or ~k, or both, are complex. The discussion
in Section 2.3 applies to the compressible theory just as well as to the incompressible theory.
When Eqs. 8.7 are substituted into Eqs. 8.5, and the same linear combinations of x and z momentum
equations are formed as in Section 2.4 for the variables
69
we obtain a system of equations which are the compressible counterparts of Eqs. 2.36. The momentum
equation in the direction parallel to the wavenumber vector ~k is
iρ(αU + βW − ω)v̂ =
Dp̂ µ 2 2λ−µ
2D v̂ + iα̃Dũ − α2 + β 2 v̂ + D2 v̂ + iα̃ũ
− +
γM12 R 3 R
1 dµ dµ 2 dµ
+ i (αDU + βDW )θ̂ + 2 DT Dv̂ + (λ − µ)DT (Dv̂ + iα̃ũ) . (8.9b)
R dT dT 3 dT
%̂ θ̂
p̂ = + . (8.9f)
ρ T
To reiterate, in these equations the eigenfunctions of the fluctuations are functions only of y and are denoted
by a carent or a tilde; the mean-flow velocities U and W are also functions of y, as are the other mean-flow
quantities: density ρ (= 1/T ), temperature T , viscosity coefficients µ and λ, thermal conductivity coefficient
κ, and Prandtl number. The specific heats are constant. The reference velocity for U and W is the same as
for R and M1 , and the reference length for y is the same as in R.
70
8.4 First-order equations
8.4.1 Eighth-order system
Equations 8.9 are the basic equations of the compressible stability theory, but are not yet in a form suitable
for numerical computation. For this purpose we need a system of first-order equations as in Section 2.5.2.
With the dependent variables defined by
and the fact that this reduction is possible proves that Eqs. 8.9 constitute an eight-order system. The lengthy
equations for the matrix elements are listed in Appendix A.
The boundary conditions are
71
complexity of these equations compared to those for incompressible flow, we are still able to arrive at
analytical solutions. The lengthy derivation is given in Appendix B (Mack, 1965a). The exact freestream
solutions are the ones to use to calculate the initial values for a numerical integration of Eqs. 8.11, but they
do not lend themselves to a ready physical interpretation. For this purpose, we examine the limit of large
Reynolds number. The characteristic values simplify to
h i1/2
2
λ1,2 = ∓ α2 + β 2 − M12 (αU1 + βW1 − ω) , (8.13a)
1/2
λ3,4 = ∓ [iR (αU1 + βW1 − ω)] , (8.13b)
1/2
λ5,6 = ∓ [iPr R (αU1 + βW1 − ω)] , (8.13c)
λ7,8 = λ3,4 . (8.13d)
We can identify our solutions as, in order, the inviscid solution, the first viscous velocity solution, a viscous
temperature solution, which is new and does not appear in the incompressible theory, and the second viscous
velocity solution. We shall only use the upper signs in what follows, as these are the solutions which enter
the eigenvalue problem.
The components of the characteristic vector of the inviscid solution are
(1) 1/2
A1 = −i α2 + β 2 , (8.14a)
1/2
2
α2 + β 2 − M12 (αU1 + βW1 − ω)
(1)
A3 = 1/2
, (8.14b)
(α2 + β 2 )
(1) i (αU1 + βW1 − ω)
A4 = 1/2
, (8.14c)
(α2 + β 2 )
(1) i(γ − 1)M12 (αU1 + βW1 − ω)
A5 = 1/2
. (8.14d)
(α2 + β 2 )
The normalization has been changed to correspond to the incompressible solutions of Eq. 2.50. It can be
noted that these expressions are correct when we set M1 = 0.
The components of the characteristic vector corresponding to the first viscous velocity solution are
(3)
A1 = 1, (8.15a)
(3) i
A3 = 1/2
, (8.15b)
[iR (αU1 + βW1 − ω)]
(3) (3)
A4 = 0, A5 = 0. (8.15c)
This solution is identical to the λ3 incompressible solution only in the limit of large Reynolds numbers.
The components of the characteristic vector corresponding to the viscous temperature solution are
(5)
A1 = 0, (8.16a)
1/2
(5) i (αU1 + βW1 − ω)
A3 = − , (8.16b)
(iPr R)1/2
(5)
A4 = 0, (8.16c)
(5)
A5 = 1. (8.16d)
The components of the characteristic vector corresponding to the second viscous velocity solution are
(7) (7) (7) (7)
A1 = 0, A3 = 0, A4 = 0, A5 = 0, (8.17a)
(7)
A7 = 1, (8.17b)
(7) 1/2
= − α2 + β 2 + iR (αU1 + βW1 − ω)
A8 . (8.17c)
72
This solution is exact and is the same spanwise viscous wave solution as in incompressible flow.
We may observe that the viscous velocity solutions have only fluctuations of velocity, not of pressure or
temperature. The velocity fluctuations in the x, z plane are in the direction of ~k for the first solution, and
are normal to ~k for the second solution which is periodic only in time. The viscous temperature solution
has no velocity fluctuations in the x, z plane, or pressure fluctuations. We may regard these solutions as the
responses to sources of ũ, w̃ and θ̂, and to emphasize this fact the respective solutions have been normalized
to make these quantities unity. The second viscous velocity solution still has the interpretation of a normal
vorticity wave, as in the incompressible flow, but this wave cannot exist as a pure mode in the boundary
layer (Squire mode) because of the a68 dissipation term that couples the latter two of Eqs. 8.11 to the first
six equations.
73
Chapter 9
%̂ θ̂
p̂ = + . (9.1f)
ρ T
We note that the w̃ momentum equation, Eq. 9.1c, and the energy equation, Eq. 9.1e, are decoupled from
the other equations. Therefore, we can eliminate α̃eu and %̂ from the latter to arrive at the following two
first-order equations for v̂ and p̂:
2 2 2 αU + βW − ω p̂
(αU + βW − ω) Dv̂ = (αDU + βDW ) v̂ + i α + β T − M1 , (9.2a)
α2 + β 2 γM12
p̂
D = −iρ (αU + βW − ω) v̂. (9.2b)
γM12
These equations are the 3D compressible counterparts of Eqs. 3.12. The boundary conditions are
The inviscid equations can be written in a simplified form if we introduce the Mach number
(αU + βW − ω) M1
M= 1/2
. (9.4)
(α2 + β 2 ) T 1/2
For a temporal neutral wave, M is real and is the local Mach number of the mean flow in the direction of
the wavenumber vector ~k relative to the phase velocity ωr /k. In all other cases, M is complex, but even so
74
we shall refer to it as the relative Mach number. In terms of M , Eqs. 9.2 simplify to
v̂ p̂
= i 1 − M2
D , (9.5a)
αU + βW − ω γM 2
2 α2 + β 2 v̂
Dp̂ = −iγM . (9.5b)
αU + βW − ω
We observe that these equations are identical to two-dimensional equations (β = 0) when written in the tilde
variables of Eq. 2.37. Therefore, inviscid instability is governed by the mean flow in the direction of ~k, just
as for incompressible flow. Either Eqs. 9.5 or 9.2 can be used for numerical integration, but the latter have
the advantage that v̂ is a better behaved function near the critical point than is v̂/(αU + βW − ω).
Equation 9.5a is the familiar linearized pressure-area relation of one-dimensional flow. The quantity
v̂/(αU + βW − ω) is the amplitude function of the streamtube area change. The other flow variables can be
written in a similar manner as
v̂ 1 e 1 v̂
U = i DU
e e + (α̃U − ω) D , (9.6a)
e − ω α̃
α̃U 1 − M2 α̃Ue −ω
v̂ 1 v̂
ω̂ = i DT − (γ − 1)T 2
D , (9.6b)
α̃U − ω
e 1−M α̃U − ω
e
M2
v̂ v̂
%̂ = i Dρ −ρ D , (9.6c)
α̃U e −ω 1 − M2 e −ω
α̃U
v̂
w̃ = iDW̃ . (9.6d)
α̃U − ω
e
where we have used the tilde variables for simplicity. When the second terms of these equations are written
with p̂ in place of v̂/(α̃U
e ), they can be readily recognized as the linearized momentum equation, the isentropic
temperature-pressure relation, and the isentropic density-pressure relation, respectively. The first terms are
in the nature of source terms, and arise from the combination of a vertical fluctuation velocity and a mean
shear. Because Eq. 9.6d is an equation for the vertical vorticity component α̃w̃, only the source term is
present.
A manipulation of Eqs. 9.1 leads to a single second-order equation for v̂:
" #
e − ω)Dv̂ − α̃DU v̂
(α̃U
D − (α2 + β 2 )(α̃U
e − ω)v̂ = 0. (9.7)
1 − M2
This equation, which in 2D form was used by Lees and Lin (1946), is the 3D compressible counterpart of
e − ω) follows directly from Eq. 9.5:
the Rayleigh equation. A second-order equation for v̂/(α̃U
M2
v̂ v̂ v̂
D2 + D ln 2
D − α̃2 (1 − M 2 ) = 0. (9.8)
e −ω
α̃U 1−M e −ω
α̃U e −ω
α̃U
The corresponding equation for p̂ is
D2 p̂ − D ln M 2 Dp̂ − α2 + β 2 1 − M 2 p̂ = 0.
(9.9)
D2 p̂ − α2 + β 2 1 − M 21 p̂ = 0.
(9.10)
75
which agrees with Eq. 8.14c. Equations 9.11 and 8.14b provide the initial values for the numerical integration.
The freestream solutions may be classified into three groups: subsonic waves with M 21 < 1; sonic waves
with M 21 = 1; and supersonic waves with M 21 > 1. Neutral supersonic waves are Mach waves of the relative
flow, and can exist as either outgoing on incoming waves. True instability waves, which must satisfy the
boundary condition at y = 0 as well as infinity are almost all subsonic, but eigenmodes which are supersonic
waves of the outgoing family in the freestream have been found for highly cooled boundary layers (Mack,
1969). A combination of incoming and outgoing waves permits the boundary condition at y = 0 to be
satisfied for any combination of α, β, and ω, as pointed out by Lees and Lin (1946). It is when only one
family of waves is present that we have an eigenvalue problem. The combination of both families is the basis
of the forcing theory presented in Chapter 11.
D (ρDU ) = 0, (9.12)
and yo is the point at which U = 1 − 1/M1 . The phase velocity of the neutral wave is cs , the mean velocity
at ys . This necessary condition is the generalization of Rayleigh’s condition for incompressible flow that
there must be a point of inflection in the velocity profile for a neutral wave to exist. The point ys , which
plays the same role in the compressible theory as the inflection point in the incompressible theory, is called
the generalized inflection point. The proof of sufficiency given by Lees and Lin requires M to be everywhere
subsonic.
(ii) A sufficient condition for the existence of an unstable wave is the presence of a generalized inflection
point at some y > yo , where yo is the point at which U = 1 − 1/M1 . The proof of this condition also requires
M to be subsonic.
(iii) There is a neutral sonic wave with the eigenvalues α = 0, c = co = 1 − 1/M1 .
(iv) If M 2 < 1 everywhere in the boundary layer, there is a unique wavenumber αs corresponding to cs
for the neutral subsonic wave.
Lees and Lin obtained these results by a direct extension of the methods of proof used for incompressible
flow. The necessary condition for a neutral subsonic wave was derived from the discontinuity of the Reynolds
stress τ = −huvi at the critical point yc . As in incompressible flow, τ is constant for a neutral inviscid wave
except possibly at the critical point. For ωi = 0,
π D (ρDU )
τ (yc + 0) − τ (yc − 0) = hvc2 i. (9.13)
α DU c
Equation 9.13 is the same as Eq. 3.9 in the incompressible theory except that D(ρDU ) appears in place of
D2 U . Since τ is zero at the wall and in the freestream by the boundary conditions for a subsonic wave, it
follows that D(ρDU ) must be zero at yc . We may also note that for a neutral supersonic wave, where c < c0
and τ (yc + 0) = (α2 /2)(M 2 − 1)1/2 from the freestream solutions, the discontinuity at the critical point must
equal this value of τ and the phase velocity must be other than Us .
At this point we can examine the numerical consequences of the finding that neutral and unstable waves
depend on the existence of a generalized inflection point. For the Blasius boundary layer, D2 U is negative
everywhere except at y = 0. However, for a compressible boundary layer on an insulated flat plate, D(ρDU )
is always zero somewhere in the boundary layer. Consequently, all such boundary layers are unstable to
inviscid waves. Figure 9.1 shows that cs , the mean velocity at the generalized inflection point and thus
the phase velocity of the neutral subsonic wave, increases with increasing freestream Mach number M1 in
accordance with the outward movement of the generalized inflection point. If we recall from Section 6 that
inviscid instability increases for the adverse pressure-gradient Falkner-Skan profiles as the inflection point
76
moves away from the wall, we can expect in this instance that inviscid instability will increase with increasing
Mach number. Figure 9.1 also includes both c0 , the phase velocity of a neutral sonic wave, and the phase
velocity for which M = −1 at the wall. In the exact numerical solutions of the boundary-layer equations
which were used for Fig. 9.1, the wall is insulated and the freestream temperature T1∗ is characteristic of
wind-tunnel conditions. The stagnation temperature is held constant at 311 K until, with increasing M1 ,
T1∗ drops to 50 K. For higher Mach numbers, T1∗ is held constant at 50 K.
Figure 9.1: Phase velocities of 2D neutral inflectional and sonic waves, and of waves for which relative
supersonic region first appears. Insulated wall, wind-tunnel temperatures.
For a wave to be subsonic relative to the freestream, and hence have vanishing amplitude at y → ∞ even
when neutral, c must be greater than c0 . It is often said that only subsonic waves are considered in stability
theory, but this statement is not entirely correct. It is true that the neutral subsonic wave with eigenvalues
αs , cs can only exist when cs > 1 − 1/M1 . However, this does not rule out amplified and damped waves with
c < 1 − 1/M1 , or even neutral supersonic waves with a c different from cs . Examples of such waves have
been found, all of which satisfy the boundary conditions at infinity and so are solutions of the eigenvalue
problem. For ωi 6= 0, the amplitudes of outgoing amplified and incoming damped waves vanish at infinity
regardless of the value of c; for neutral waves, the amplitude will only be bounded at infinity when c < c0 .
What does turn out to be true is that the most unstable waves are always subsonic. Furthermore, for one
class of waves, the amplified first-mode waves, the phase velocity is always between c0 and cs . This result
has important consequences.
77
solutions have been worked out in more detail by Reshotko (1960). Both v̂ and û have the same analytical
behavior as in compressible flow. What is new here is the temperature fluctuation, which, according to
Reshotko, has the behavior
1 T
θ̂ ≈ + [D (ρDU )]c ln (y − yc ) + · · · . (9.15)
y − yc DU c
Hence, even for a neutral subsonic wave, where [D(ρDU )]c = 0 and v̂ and û are both regular, θ̂ has a
singularity at yc .
Two methods have been devised for the numerical integration of the inviscid stability equations. The first
method (Lees and Reshotko, 1962) transforms the second-order linear equation into a first-order nonlinear
equation of the Riccati type. This equation is solved by numerical integration except for the region around
the critical point, where the power series in y − yc are used. The second method (Mack, 1965a) is a
generalization to compressible flow of Zaat’s (1958) method. This method has already been described in
Section 3.2. For neutral and damped solutions, the contour of integration is indented under the singularity,
just as for compressible flow.
where Eq. 9.17b follows from the boundary condition v̂(0) = 0. We have written α̃s as α̃sn . The subscript s
denotes a neutral subsonic solution as before; the subscript n refers to the multiple solutions. The constant
in Eq. 9.17b is chosen as −i to make p̂ real and positive for y > ya . Either sign is possible for y > ya . Since p̂
e − ω)], from Eq. 9.5a, must go to zero as y → ya as does M 2 − 1.
is continuous and finite at y = ya , D[v̂/(α̃U
The derivative of v̂/(α̃U − ω) gives a factor (M 2 − 1)1/2 , and the required factor of (M 2 − 1)1/2 can only
e
come from the cosine having a zero at ya . Consequently,
78
Z ya 1/2
2
cos α̃sn M −1 dy = 0, (9.18a)
0
and Z ya
1/2 1
α̃sn M2 − 1 dy = n− , n = 1, 2, 3, . . . . (9.18b)
0 2
Equation 9.18b is the final result, and demonstrates that there is an infinite sequence of discrete neutral
wavenumbers with the phase velocity cs . The difference between adjacent values of α̃sn is
Z ya −1
1/2
α̃s(n+1) − α̃sn = π M2 − 1 dy . (9.19)
0
We may also observe that according to Eq. 9.18b, the sequence of values of 2α̃sn is 1, 3, 5, 7, . . .. This result
was first noted and given a physical explanation by Morkovin [private communication (1982)]. Because
Eq. 9.16 is only approximate, the magnitude of α̃sn , the difference formula, and the ratio sequence are not
expected to be numerically correct. However, as we shall see below, with an important exception they are
either correct, or approximately correct.
When the numerical integration of Eqs. 9.2 is carried out for 2D waves with c = cs and ωi = 0 for the
insulated-wall flat-plate boundary layers described in Section 9.3, the αsn which are found by the eigenvalues
search procedure are shown in Fig. 9.2. The solution for each n will be referred to as a mode: n = 1 is the
first mode, n = 2 the second mode, etc. The wavenumbers of the first mode were first computed by Lees and
Reshotko (1962). With c = cs , a ya where M 2 = 1 occurs first at M1 = 2.2 (ya = 0). With increasing M1 ,
the relative sonic point ya moves out into the boundary layer, and αsn varies in inverse proportion to ya as
required by Eq. 9.18b. No higher modes with c = cs could be found numerically for M1 < 2.2, in agreement
with the theory given above.
Figure 9.2: Multiple wavenumbers of 2D inflectional neutral waves (c = cs ). Insulated wall, wind-tunnel
temperatures.
A prominent feature of Fig. 9.2 is that the upward sloping portion of the first-mode curve between M1 = 2
and 4.5 is in a sense continuous through the other modes, i.e., there is a Mach number range for each mode
where the αsn vs. M1 curve has a positive slope. The end point of this region for one mode is close to the
starting point of a similar region for the next higher mode. The approach becomes closer as M1 increases.
The significance of these intervals of positive slope is that they provide the exceptions to the correctness, or
approximate correctness, of the results given by, or deduced from, Eqs. 9.18b. Indeed we could well identify
these modes as the “exceptional” modes.
With the wavenumbers of the multiple neutral waves established, the next step is to examine the eigen-
functions. For this purpose, the eigenfunction p̂/(γM12 ) is shown in Fig. 9.3 for the first six modes at
79
M1 = 10. The first thing to note is that the number of zeroes in p̂ is one less than the mode number n.
For example, the second mode has one zero, and p̂(0) is 180◦ out of phase with p̂(δ); the third mode has
two zeroes and p̂(0) is in phase with p̂(δ). The number of zeroes in p̂(y) is the surest identification of the
mode under consideration. By keeping track of the phase difference between p̂(0) and p̂(δ), it is possible to
determine when there is a change from one mode to another.
Figure 9.3: Pressure-fluctuation eigenfunctions of first six modes of 2D inflectional neutral waves (c = cs ) at
M1 = 10. Insulated wall, T1∗ = 50 K.
The appearance of the eigenfunctions in Fig. 9.3 confirms the simple theory given above: there is an
infinite sequence of periodic solutions in the supersonic relative flow region which can satisfy the boundary
conditions. The magnitude of p̂(0) is a minimum for the fourth mode [p̂(δ) is the same for all modes]. Since
the fourth mode at M1 = 10 is on the upward sloping portion of the eigenvalue curve in Fig. 9.2, this
is another indication of the special nature of such neutral solutions. For other modes, p̂(0)/p̂(δ) tends to
become large away from n = 4, and tends to infinity as n → ∞.
There is one important difference between the simple theory and Fig. 9.3. According to the theory, p̂(δ)
is positive for all modes; there are no zeroes in the interval y > ya , and the number of zeroes in y < ya
increases by one for each successive mode. We see from Fig. 9.3 that p̂(δ) is negative for n > 4, and the
number of zeroes in y < ya is the same for n = 5 as for n = 4. The total number of zeroes increases by one
from n = 4 to n = 5 only because of the zero in y > ya . However, we note that the “exceptional” mode is
extraneous to the simple theory, and preserves something of a first-mode character which probably betrays
a different physical origin from the other modes. Indeed, the other higher modes are nothing more than
sound waves which reflect back and forth between the wall and the sonic line of the relative flow at y = ya as
first suggested by Lees and Gold (1964). Morkovin’s theory is based on this idea, and its duplication of the
wavenumber ratio sequence 1,3,5,... attests to its correctness. The “exceptional” modes are not part of this
theory; they are perhaps vorticity waves associated with the generalized inflection point as are incompressible
and low Mach number first-mode waves. In this view, the modes which have been identified in Figs. 9.2
and 9.3 as first-mode waves for M1 > 5 are not first-mode waves at all; this distinction is reserved for the
modes whose wavenumbers increase monotonically with increasing M1 . However, we shall continue to refer
to n = 1 as the first mode.
80
These waves are characterized by having phase velocities in the range 1 ≤ c ≤ 1 + 1/M1 . For each phase
velocity there is an infinite sequence of wavenumbers, just as for the inflectional neutral waves. A wave with
c = 1 is at rest with respect to the freestream; a wave with c = c02 = 1 + 1/M1 propagates downstream
relative to U1∗ with the freestream speed of sound. The Lees-Lin neutral sonic wave propagation upstream
relative to U1∗ with the freestream speed of sound.
All of the 1 ≤ c ≤ 1 + 1/M1 waves are subsonic waves, and, because D(ρDU ) = 0 in the freestream,
there is no discontinuity in the Reynolds stress and the necessary condition for the existence of a subsonic
neutral wave is satisfied. Unlike the inflectional neutral waves, D(ρDU ) does not have to be zero in the
boundary layer, and the 1 ≤ c ≤ 1+1/M1 waves exist for any boundary layer subject only to the requirement
that M 2 > 1 somewhere. The importance of the c = 1 neutral waves is that in the absence of an interior
generalized inflection point they are accompanied by a neighboring family of unstable waves with c < 1.
Consequently, a compressible boundary layer is unstable to inviscid waves whenever M 2 > 1, regardless of
any other feature of the velocity and temperature profiles.
Figure 9.4: Multiple wavenumbers of 2D noninflectional neutral waves (c = 1). Insulated wall, wind-tunnel
temperatures.
If we examine the inviscid equations 9.2, we see that when c > 1 they are no longer singular; i.e., there
is no critical layer. Even when c = 1, and the critical layer is in a sense the entire freestream, Eq. 9.2a is
still not singular because DU/(U − 1) and p̂(y)/(U − 1) both have finite limits as y → yδ . We call this class
of solutions the noninflectional neutral waves. These waves persist to low subsonic Mach numbers, because,
except at M1 = 0, it is always possible to find a c large enough so that M = −1 somewhere in the boundary
layer.
The approximate theory of the preceding section applies to the noninflectional neutral waves just as well
as to the inflectional neutral waves provided the initialization is changed for c = 1 to make v̂/(αU − ω) finite
in the freestream. This change is needed because with c = 1 the wave motion is confined to the boundary
layer and v̂ must be zero for y > yδ . An infinite sequence of wavenumbers is obtained with the spacing given
by Eq. 9.19, but since c is different from cs the numerical values are not the same as for the inflectional waves.
The wavenumbers obtained from the numerical integration with c = 1 are shown in Fig. 9.4 as functions of
Mach number. These wavenumbers are denoted by α1n , where the first subscript refers to c = 1, and the
second is the mode number. There is now no portion of any wavenumber curve with a positive slope, and
the spacing agrees reasonably well with the approximate formula. The discrepancy is about 10% for the first
two modes, and decreases to about 1% for the fifth and sixth modes.
The eigenfunctions p̂(y) of the first six modes of the noninflectional neutral waves with c = 1 at M1 = 10
are shown in Fig. 9.5. Here the ratio p̂(y)/p̂(0) is plotted, rather than p̂(y) with p̂(δ) fixed as in Fig. 9.3.
The appearance of these eigenfunctions is in complete accord with the simple theory, unlike the inflectional
neutral waves where the modes on the upward sloping portions of the wavenumber curves interrupt the
orderly sequence, and where an outer zero appears in the eigenfunctions for n > 4.
81
Figure 9.5: Pressure-fluctuation eigenfunctions of first six modes of 2D noninflectional neutral waves (c = 1)
at M1 = 10. Insulated wall, T1∗ = 50 K.
82
The numerical results for 1 < c < 1 + 1/M1 are similar to those presented for c = 1. Since these waves
have no neighboring unstable or damped waves, they are of less importance in the inviscid theory than the
other neutral waves. Consequently, these waves will not be considered further, and the term noninflectional
neutral wave will refer only to a c = 1 wave. However, we might mention that the viscous counterparts of
the c > 1 waves, which are damped rather than neutral, do have a role to play in certain cases.
Figure 9.6: Effect of Mach number on maximum temporal amplification rate of 2D waves for first four modes.
Insulated wall, wind-tunnel temperatures.
83
Figure 9.7: Effect of Mach number on frequency of most unstable 2D waves for first four modes. Insulated
wall, wind-tunnel temperatures.
inflection point has moved out to Us = 0.38. The inflection point is a fixed feature of the boundary-layer
profile, and so is independent of the wave orientation. The phase velocity cs of a 3D wave is Us cos ψ, and
the phase velocity c0 is (1 − 1/M f1 ) cos ψ, where Mf1 = M1 cos ψ. Thus as the wave angle ψ increases from
zero, c0 decreases more than by cos ψ, and the difference cs − c0 increases. Consequently, we can expect the
first mode to become more unstable. At the same time the thickness of the supersonic relative flow region,
where one exists, will decrease along with M f1 and we shall not be surprised to find that the higher modes
become more stable.
Figure 9.8 shows the temporal amplification rate ωi of the first and second modes at M1 = 4.5 as a
function of the frequency ωr for several wave angles. Three-dimensional first-mode waves are indeed more
unstable than 2D waves, and second-mode 3D waves are more stable than the corresponding 2D waves. The
latter result also holds for all of the higher modes. The most unstable first-mode wave is at an angle of
close to 60◦ , with an amplification rate about twice the maximum 2D rate and with a frequency a little over
one-half of the frequency of the most unstable 2D wave.
At M1 = 4.5, the unstable regions of the first two modes are separated by a damped region for all wave
angles. However, at M1 = 8.0, Fig. 9.9 shows that for 2D waves the first three modes are merged into a single
unstable region. If we look at Fig. 9.2 we see that at this Mach number the exceptional mode is the third
mode. Thus we can note another feature of the neutral wavenumbers αsn of these modes: They serve as the
“end points” of the merged unstable regions of the modes lying below them. As the wave angle increases
from zero and M f1 decreases, the merging is still in general accord with Fig. 9.2 for M
f1 , as is confirmed by
the calculation of αesn . The same pattern of upward sloping exceptional wave numbers is found for oblique
waves as for 2D waves (Mack, 1969). For ψ = 60◦ , the second-mode is stable; for ψ = 56◦ , there are still
second-mode unstable waves, as can be verified by examining the phase change across the boundary layer of
the pressure fluctuations, even though no peak is visible on the curve of Fig. 9.9.
In Fig. 9.10, the maximum temporal amplification rate with respect to frequency is plotted against the
four Mach numbers 4.5, 5.8, 8.0 and 10.0. At all of these Mach numbers the most unstable first-mode wave
is at an angle of between 50◦ and 60◦ , and has a maximum amplification rate that is roughly double the
most unstable 2D wave. The effect of Mach number on the maximum first-mode amplification rate with
respect to both frequency and wave angle is shown in Fig. 9.11. The wave angle of the most unstable wave
is noted on the figure to within 5◦ , and the maximum 2D amplification rates are shown for comparison. An
interesting change in the relationship between the 2D and 3D amplification rates takes place for M1 < 4. The
3D maximum amplification rate is no longer only double the 2D rate as at higher Mach numbers; instead, at
M1 = 3.0 the ratio of the 3D rate to the 2D rate is 5.8, at M1 = 2.2 it is 33, and at M1 = 1.8 it is 130. We
recall from Fig. 9.1 that it is near M1 = 1.6 that the difference cs − c0 is the smallest. Therefore, the sonic
limit acts as a severe constraint on the amplification of 2D waves at low Mach numbers. When this constraint
84
Figure 9.8: Temporal amplification rate of first and second modes vs. frequency for several wave angles at
M1 = 4.5. Insulated wall, T1∗ = 311 K.
Figure 9.9: Temporal amplification rate as a function of wavenumber for 3D waves at M1 = 8.0. Insulated
wall, T1∗ = 50 K.
85
is removed, as it is for 3D waves, the amplification rates increase sharply. We may consider the 3D maximum
amplification rate as the one that properly reflects the inherent instability of a given boundary-layer profile.
Figure 9.10: Effect of wave angle on maximum temporal amplification rate of first and second-modes at
M1 = 4.5, 5.8, 8.0 and 10.0. Insulated wall, wind-tunnel temperatures.
86
Figure 9.11: Effect of Mach number on maximum temporal amplification rates of 2D and 3D first-mode
waves. Insulated wall, wind-tunnel temperatures.
shown, just as originally predicted by Lees (1947). However, the second mode is not only not stabilized, it is
actually destabilized, although if the amplification rate is based on the boundary-layer thickness, the increase
in ωi is just about compensated for by the reduction in yδ and ωi yδ is virtually unchanged by cooling.
As a final result on the effect of cooling, we give Fig. 9.13 which shows the temporal amplification rate
at M1 = 10 as a function of wavenumber for an insulated wall and a highly-cooled wall (Tw /Tr = 0.05).
For the former, the first four modes are merged to form a single unstable region, and the limiting upper
wavenumber is the exceptional wavenumber of Fig. 9.2. For the latter, the unstable regions of the four modes
are separate, as is true at lower Mach numbers for an insulated wall, and the maximum amplification rate
of each mode is about double the uncooled value.
87
Figure 9.12: Effect of wall cooling on ratio of maximum temporal amplification rate with respect to both
frequency and wave angle of first and second modes at M1 = 3.0, 4.5, and 5.8 to insulated-wall maximum
amplification rate. Wind-tunnel temperatures.
Figure 9.13: Effect of extreme wall cooling on temporal amplification rates of 2D wave for first four modes
at M1 = 10, T1∗ = 50 K: Solid line, insulated wall; Dashed line, cooled wall, Tw /Tr = 0.05.
88
Chapter 10
The early theoretical work on the viscous stability theory of compressible boundary layers was based on the
asymptotic methods that had proven to be successful for incompressible flow. However, these theories, which
were developed by Lees and Lin (1946), Dunn and Lin (1955), and Lees and Reshotko (1962), turned out to
be valid only up to low supersonic Mach numbers. Some results for insulated-wall flat-plate boundary layers
obtained with the asymptotic method are given in Fig. 10.1, and compared with direct numerical solutions of
the eigenvalue problem. All numerical results in this Section are for the same family of flat-plate boundary
layers used in Section 9. In Fig. 10.1 neutral-stability curves of frequency at M1 = 1.6 and 2.2 as computed
from the Dunn-Lin 1955 theory by Mack (1960) are compared with results obtained by numerical integration
using both the sixth-order simplified equations of Dunn and Lin, and the sixth-order constant Prandtl number
version of the complete stability equations of Appendix A. At M1 = 1.6, the three calculations are in good
agreement for R > 700, but at M1 = 2.2, the agreement between the Dunn-Lin theory and the numerical
solution with the complete equations is poor at all Reynolds numbers. The asymptotic theory is supposed to
solve the simplified equations with an error no larger than the error involved in dropping the missing viscous
terms. It is evident from the numerical solutions of the Dunn-Lin equations in Fig. 10.1, that the equations
are better than the method used to solve them, but even so at M1 = 2.2 the differences compared to the
complete equations are too large to permit their use. However, there is little reason in any case to use these
equations in numerical work, because they are of the same order as the complete 2D equations, and for 3D
waves the sixth-order approximation given in this Section is more accurate.
89
Figure 10.1: Comparison of neutral-stability curves of frequency at (a) M1 = 1.6 and (b) M1 = 2.2. Insulated
wall, wind-tunnel temperatures.
90
above M1 = 3 are more like those of a free shear layer than of a low-speed zero pressure-gradient boundary
layer.
Figure 10.2: Effect of Mach number on 2D neutral-stability curves of wavenumber. Insulated wall, wind-
tunnel temperatures.
We have learned in Section 9 that 2D amplification rates above M1 = 1 are strongly influenced by the
constraint of the sonic limit on the phase velocity, and do not represent the true instability of a boundary-
layer profile. Therefore, to get a complete view of the influence of Mach number on viscous instability we
must turn to 3D waves. The instability of 2D and 3D waves up to M1 = 3.0 is summarized in Fig. 10.3, where
the maximum temporal amplification rate is given at M1 = 1.3, 1.6, 2.2, and 3.0 as a function of Reynolds
number up to R = 2000. The most unstable wave angles (to within 5◦ ) of the 3D waves are shown in the
figure. It is apparent that these angles differ little from the inviscid values except near the critical Reynolds
number at M1 = 1.3. We see that viscous instability, which at M1 = 1.3 is totally responsible for both 2D
and 3D instability at the Reynolds numbers of the figure, decreases with increasing M1 for 3D as well as
for 2D waves. However, there is little change in the maximum 3D amplification rate with increasing Mach
number, in contrast to the large decrease in the maximum 2D amplification rate. At M1 = 3.0, viscosity acts
only to maintain the maximum amplification rate at about the same level down to low Reynolds numbers,
rather than as the main source of instability as at lower Mach numbers.
There are unfortunately no calculations available between M1 = 3.0 and 4.5, but the distribution with
Reynolds number of the maximum temporal amplification rate is given in Fig. 10.4 at M1 = 4.5, 5.8, and
7.0 for wave angles that are approximately the most unstable. All of these waves are first-mode waves. At
M1 = 10 it is difficult to assign a maximum in the first-mode region as the single peak in the ωi vs α curves
for ψ > 50◦ occurs near the transition from the first to the second mode, and 55◦ has been rather arbitrarily
selected as the most unstable angle. In any case, it is clear from Fig. 10.4 that in this Mach number range
there is no viscous instability and the influence of viscosity is only stabilizing.
91
Figure 10.3: Distribution of maximum temporal amplification rate with Reynolds number at (a) M1 = 1.3,
(b) M1 = 1.6, (c) M1 = 2.2 and (d) M1 = 3.0 for 2D and 3D waves. Insulated wall, wind-tunnel temperatures.
Figure 10.4: Distribution of maximum first-mode temporal amplification rates with Reynolds number at
M1 = 4.5, 5.8, 7.0 and 10.0. Insulated wall, wind-tunnel temperatures.
92
10.2 Second mode
The lowest Mach number at which the unstable second mode region has been located at finite Reynolds
numbers is M1 = 3.0, where the minimum critical Reynolds number Rcr is 13,900 (Mack, 1984b). As
the Mach number increases, the inviscid second-mode maximum amplification rate increases, as shown in
Fig. 9.6, and the unstable second-mode region moves rapidly to lower Reynolds numbers. At M1 = 3.8,
Rcr is 827; at M1 = 4.2 it is 355; and at M1 = 4.5 it is 235. Furthermore, the first and higher-mode
unstable regions go through the same process of successive mergers as they do in the inviscid theory. The
first merger, between the first and second-mode unstable regions, takes place at about M1 = 4.6. Examples
of neutral-stability curves of wavenumber just before merger (M1 = 4.5), and just after merger (M1 = 4.8),
are shown in Fig. 10.5. The shapes of the neutral-stability curves, both before and after merger, are such
as to suggest that viscosity is only stabilizing for all higher modes, and this is confirmed for the 2D second
mode by Fig. 10.6, where the distribution of (ωi )max with Reynolds number is shown for M1 = 4.5, 5.8, 7.0,
and 10.0.
Figure 10.5: Neutral-stability curves of wavenumber for 2D first and second-mode waves at (a) M1 = 4.5
and (b) M1 = 4.8. Insulated wall, wind-tunnel temperatures.
The effect of wave angle on second-mode amplification rates is shown in Fig. 10.7, where (ωi )max is
plotted against wave angle for the same Mach numbers as in Fig. 10.6. This figure is to be compared with
the comparable inviscid results in Fig. 9.10. In both instances, increasing Mach number brings a reduction
in the rapidity with which the maximum amplification falls off with increasing wave angle.
93
Figure 10.6: Effect of Reynolds number on maximum second-mode temporal amplification rate at M1 = 4.5,
5.8, 7.0 and 10.0. Insulated wall, wind-tunnel temperatures.
Figure 10.7: Effect of wave angle on second-mode temporal amplification rates at R = 1500 and M1 = 4.5,
5.8, 7.0 and 10.0. Insulated wall, wind-tunnel temperatures.
94
layer (M1 = 0.05). The x-Reynolds numbers of 2D normal modes for three constant values of the N factor,
ln (A/A0 )max , are plotted against the wall temperature ratio Tw /Tr . We see that cooling has a strong
stabilizing effect, and that heating has a strong destabilizing effect. The frequencies that correspond to the
N factors are also strongly affected by the wall temperatures. For example, at Tw /Tr = 0.90, the frequency
for N = 9 is F = 0.157 × 10−4 ; at Tw /Tr = 1.15, it is F = 0.445 × 10−4 .
Figure 10.8: Effect of wall cooling and heating on Reynolds number for constant ln (A/A0 )max at M1 = 0.05.
As an example of the effect of wall cooling at hypersonic speeds, Fig. 10.9 shows 2D neutral curves at
M1 = 5.8 for Tw /Tr = 1.0, 0.65, 0.25, and 0.05. The freestream temperature is 50 K except for the lowest
wall temperature where it is 125 K. When the wall is cooled to Tw /Tr = 0.65, a noticeable stabilization
takes place for the first-mode, but only a narrowing of the unstable wavenumber band can be detected in the
second-mode region. At the other two temperature ratios, there is no unstable first-mode region. The lowest
temperature ratio is of interest because there is no generalized inflection point in the boundary layer, and thus
no αs2 to serve as the limit of the upper branch of the neutral curve. We may observe that the wavenumbers
at the critical Reynolds numbers of the three cooled cases are in the inverse proportion 1.0:0.71:0.48, and
the corresponding boundary-layer thicknesses are in the proportion 1.0:0.69:0.53. Consequently, the length
scale is the controlling factor in the location of the second-mode unstable region in terms of wavenumber.
95
Figure 10.9: Effect of wall cooling on 2D neutral-stability curves at M1 = 5.8, T1∗ = 50 K.
Table 10.1: Comparison of temporal amplification rates for 3D waves as computed from sixth-order and
eighth-order systems of equations at several Mach numbers
96
at these two Mach numbers than with only the coupling term zero. It is not known how general this result
is, but experience with the Dunn-Lin equations indicates that it is limited to waves with ψ well away from
zero.
The small effect of the dissipation terms on the amplification rates of the 3D waves in the above-mentioned
calculations is in distinct contrast to what happens when the Dunn-Lin equations are used for 2D waves.
The sixth-order system with only the coupling term zero is exact for ψ = 0, unlike the Dunn-Lin equations
where all of the dissipation terms are neglected along with a number of other terms that are supposed to
be of the same order. The differences between the neutral-stability curves in Fig. 10.1 computed directly
from the Dunn-Lin equations and those computed from the complete equations testify to the importance
of the neglected terms. A calculation at M1 = 2.2 and R = 600 for α = 0.045 gave the result that the
maximum 2D amplification rate from the Dunn-Lin equations is 63% larger than when computed from the
complete equations. A more favorable result is obtained at this Mach number for a 60◦ wave with α = 0.045
at R = 1000, where the Dunn-Lin equations give an amplification rate that is 15% too high. This is an
improvement over the 2D results, but still not as good as the result obtained when only the coupling term is
neglected. At M1 = 4.5 and R = 1500, the amplification rate of the most unstable first-mode wave computed
from the Dunn-Lin equations is in error by 23%; the error for the most unstable (2D) second-mode wave is
14%. The conclusion to be drawn is that the Dunn-Lin approximation is too severe, and the equations are
unsuitable for numerical work above about M1 = 1.6. On the contrary, the sixth-order system with only the
coupling term neglected can be used for numerical computations where high accuracy is not important, and
they offer a substantial saving in computer time and expense.
97
Figure 10.10: Effect of Mach number on the maximum temporal amplification rate of first and second-mode
waves at R = 1500. Insulated wall, wind-tunnel temperatures.
Figure 10.11: Effect of Mach number on the maximum spatial amplification rate of first and second-mode
waves at R = 1500. Insulated wall, wind-tunnel temperatures.
98
Chapter 11
Forcing Theory
99
The most important result of the forcing theory is shown in Fig. 11.1, where mp /mi from the viscous
theory is plotted against Reynolds number for waves of six dimensionless frequencies in an insulated-wall,
flat-plate boundary layer at M1 = 4.5. The waves are 2D, and the phase velocity has been assumed to be
c = 0.65. We see that the amplitude of each wave starts to grow at the leading edge, reaches a peak at
a Reynolds number that varies inversely with frequency, and then declines. The lower the frequency, the
higher the maximum value of mp /mi . This is the principal result of the forcing theory, and has been found
to be true for all boundary layers and all waves regardless of the wave angle and the phase velocity (provided
only that M 1 > 1). As a consequence of this behavior, the forcing mechanism provides boundary-layer waves
with amplitudes from 6-14 times as large as freestream sound waves without any instability amplification.
In the inviscid theory, once c and ψ have been specified the only remaining parameter is α. When
the mass-flow fluctuation amplitude ratio is plotted against α for a 2D wave with c = 0.65 and the same
boundary layer as in Fig. 11.1, the inviscid theory gives a result that is significantly different from the viscous
theory. Since F = αc/R, a wave of given dimensionless frequency F travelling downstream at a constant c
will have its dimensionless wavenumber increase linearly with R. Consequently, the α axis is equivalent to
the R axis in Fig. 11.1. What we find from the inviscid theory is that inviscid waves decrease in amplitude
for α > 0.0075. All of the amplitude peaks in Fig. 11.1 occur at an α larger than this except for the
lowest frequency. Consequently, the initial growth of Fig. 11.1, which is just what is found in experiments
in supersonic and hypersonic wind tunnels with turbulent boundary layers on the tunnel walls, is purely a
viscous phenomenon. However, when the viscous response curves from Fig. 11.1 are also plotted against α,
they show that the decrease in amplitude which follows the region of growth in Fig. 11.1 is described closely
by the inviscid theory. This result is in contrast to stability theory, where the inviscid amplification and
damping rates are only approached by the viscous theory in the limit R → ∞. The higher the frequency,
the lower the Reynolds number at which the viscous curve joins the inviscid curve.
Figure 11.1: Peak mass-flow fluctuation as a function of Reynolds number for six frequencies. Viscous forcing
theory; M1 = 4.5, ψ = 0◦ , c = 0.65, insulated wall.
100
The major difficulty in using the forcing theory as a solution of the receptivity problem is that forced
waves are distinct from free waves, and the process by which the former become the latter is unknown.
An experiment by Kendall (1971) showed that, as measured by the phase velocity, a forced wave near the
leading edge evolves into a free instability wave farther downstream. In the paper from which a portion of
the text of this Section has been adapted (Mack, 1975), we assumed that the forcing theory applies up to
the neutral-stability point of the particular frequency under consideration, and that stability theory applies
downstream of that point. The conversion from one wave to the other would seem most likely to occur if the
amplitude distribution through the boundary layer at the neutral-stability point matched the eigenfunction
of the instability wave of the same frequency and wavelength. A limited number of calculations at M1 = 4.5
show that the two distributions are indeed close together for the same F , α, and R. With the only mismatch
between the two waves a phase-velocity difference of 20%, conversion of forced into free waves can be expected
to take place quickly.
Consequently, with the approach just outlined the forcing theory can be used to calculate A0 /A1 , the
ratio of the instability wave amplitude at the neutral point to the amplitude of the sound waves radiated
by the turbulent boundary layer on the wind-tunnel wall. The subsequent ratio of the instability-wave
amplitude to A1 is found by multiplying A0 /A1 by the usual amplitude ratio A/A0 calculated from stability
theory. Thus, with the forcing theory we can replace the previously unknown constant A0 with a known
frequency-dependent A0 .
Figure 11.2: Ratio of amplitude of reflected wave to amplitude of incoming wave as function of wavenumber
from viscous and inviscid theories; M1 = 4.5, ψ = 0◦ , c = 0.65, insulated wall. T1∗ = 311K.
According to inviscid theory, when α = 0, Ar /Ai = 1.0 and β(0)/βi (0) = 2.0; when α → ∞, Ar /Ai = 1.0
and β(0)/βi (0) = 0. Thus for α = 0, the boundary layer effectively has zero thickness and the sound wave
reflects as from a solid surface in the absence of a boundary layer. The reflected wave has the same amplitude
and phase at y = 0 as the incoming wave so that the wall pressure fluctuation is twice βi (0). A the other
limit, α → ∞, the boundary layer is infinitely thick compared to the wavelength, and the reflection is the
same as from a constant-pressure surface. The amplitude of the reflected wave is again equal to that of the
101
Figure 11.3: Ratio of wall pressure fluctuation to pressure fluctuation of incoming wave; M1 = 4.5, ψ = 0◦ ,
c = 0.65, insulated wall. T1∗ = 311K.
incoming wave, but its phase at y = 0 differs by 180◦ from the incoming wave. Thus the pressure fluctuation
at the wall is zero. Between these two limits, the amplitude of the reflected wave is always greater than the
amplitude of the incoming wave.
The viscous results are quite different. For small α, Ar is always less than Ai . Furthermore, a minimum
exists in Ar for each frequency. A similar minimum exists in β(0), but it is located at a larger α than is
the Ar minimum. If the Ar minimum were to reach zero, that particular α would constitute an instability
eigenvalue for the family of incoming waves. However, in stability theory, this type of wave has not been
encountered, either as a supersonic wave with c < 1 − 1/M1 as in the present example, or as a subsonic wave
with c > 1 − 1/M1 where the amplitude increases exponentially with increasing y. Figure 11.2 indicates that
if such an eigenvalue exists it would be at such a low Reynolds number to make the use of the quasi-parallel
theory invalid.
When the incoming Mach waves of the external travelling sound field reflect from a solid surface in the
absence of a boundary layer, there is no phase shift at the wall. Compression waves reflect as compression
waves, and the reflected waves originate at the points where the corresponding incoming waves intersect
the surface. However, when a boundary layer is present, there is a phase shift at the wall. Consequently,
a reflected Mach wave of the same phase appears to originate at a distance ∆ away from the point of
intersection. This offset distance, expressed as a ratio to the boundary-layer thickness, is given by
c
∆∗ /δ = [θi (0) − θr (0)] , (11.2)
F Ryδ
where θi (0) is the phase (in radians) of the pressure fluctuation of the incoming wave at the wall, and θr (0)
is the same quantity for the reflected wave. When the phase of the reflected wave lags the phase of the
incoming wave, the reflected wave originates at a point downstream of the intersection with the incoming
wave. When the phase difference is an integer multiple of π, the incoming wave reflects as a wave of the
opposite sign at the point of intersection.
In Fig. 11.4, the ratio ∆∗ /δ is given at R = 600 as a function of frequency for the same conditions as
in Figs. 11.2 and 11.3. At only one frequency, F = 0.975 × 10−4 , is the offset distance zero. For all smaller
102
frequencies, the phase of the reflected wave lags behind the phase of the incoming wave, and ∆∗ is positive
with a maximum of 4.5δ at F = 0.08 × 10−4 . Because of the long wavelength at this frequency, this offset
is only 0.077λ∗ , or 28◦ in phase. Offsets have been observed experimentally in unpublished measurements
of Kendall. The measurements were made with a broad-band hot-wire signal, so no direct comparison with
the single-frequency calculations is possible.
Figure 11.4: Offset distance of reflected wave as function of frequency at R = 600; M1 = 4.5, ψ = 0◦ , c =
0.65, insulated wall. T1∗ = 311K.
103
Table 11.1: Dimensionless boundary-layer thickness (U = 0.999), displacement thickness and momentum
thickness of insulated-wall, flat-plate boundary layers. (Wind-tunnel temperature conditions.)
M1 yδ yδ∗ yθ
0 6.0 1.72 0.664
0.7 6.2 1.92 0.660
1.0 6.4 2.13 0.656
1.6 7.0 2.77 0.648
2.0 7.6 3.37 0.644
2.2 8.0 3.72 0.643
3.0 9.8 5.48 0.642
3.8 12.1 7.83 0.644
4.2 13.5 9.22 0.646
4.5 14.6 10.34 0.646
4.8 15.8 11.55 0.646
5.8 20.0 15.73 0.636
6.2 21.7 17.49 0.629
7.0 25.4 21.19 0.616
7.5 27.8 23.62 0.607
8.0 30.3 26.13 0.598
8.5 32.9 28.72 0.590
9.0 35.5 31.38 0.581
9.5 38.2 34.10 0.573
10.0 41.0 36.88 0.565
104
Part III
105
Chapter 12
Up to this point we have been concerned in the numerical examples exclusively with two-dimensional bound-
ary layers, although the formulations of Chapters 2 and 8 are also valid for three-dimensional boundary
layers. In the final three chapters we shall take advantage of this fact to present a number of results for 3D
boundary layers. A fundamental difference between the stability of 3D and 2D boundary layers is that a 3D
boundary layer is subject to crossflow instability. This type of instability, which cannot occur in a 2D bound-
ary layer, is responsible for early transition on sweptback wings. Its essential features can best be introduced
by studying the simple boundary layer on a rotating disk. This self-similar boundary layer of constant
thickness was first used for this purpose by Gregory et al. (1955) in their classic paper on three-dimensional
boundary-layer instability.
106
Figure 12.1: Rotating-disk boundary-layer velocity profiles.
The dimensionless azimuthal and radial velocity profiles in the coordinate system rotating with the disk
are shown in Fig. 12.1 The azimuthal, or circumferential, profile is of the same type as in a 2D boundary layer
with the velocity increasing monotonically from the surface to the outer flow, and it will be referred to as the
streamwise profile. With the disk rotating in the direction of positive θ (counterclockwise), the outer flow
relative to the disk is in the negative (clockwise) direction. The radial profile is of a type that cannot occur in
a 2D boundary layer. The velocity, directed outward from the disk center, is zero both at the wall and in the
outer flow, so that there is of necessity and inflection point, which is located at ζ = 1.812, where U = 0.133
and V = −0.760. The radial velocity, being normal to the streamwise flow, is by definition the crossflow
velocity. The maximum radial velocity of Umax = 0.181 is located at ζ = 0.934, where V = −0.496.u
107
located well away from the disk surface, we can expect there to be a strong instability.
In addition to the inflectional radial profile, there is a whole family of profiles in directions close to the
radial which also have inflection points. Stuart noted that for the velocity profile at an angle of ε = 13.2◦ ,
where ε is measured from the radius in the positive θ direction, the inflection point is located where the
magnitude of the velocity is zero. Consequently, according to the Rayleigh theorem, which was shown to
still be valid for this type of profile, a stationary neutral normal mode (phase velocity c = 0) can exist with
a wave angle equal to ε. Stuart also showed by calculating streamlines in the plane of ζ and the wavenumber
vector for the rotating disk with large suction that the stationary inviscid disturbance consists of a system
of vortices close to the surface, all rotating in the same direction (clockwise, looking along the spiral towards
the disk center) and spaced one wavelength apart, and a second system of vortices farther from the surface.
Brown (1960) repeated this calculation for the rotating disk without suction using the viscous equations,
and confirmed the vortices near the surface, but not those farther out. The vortices near the surface were in
accord with conjectures made earlier. Thus the streaks and the spiral angle were explained as manifestations
of inflectional instability associated with the crossflow, and the whole phenomenon was named crossflow
instability.
This explanation, while very suggestive, left many questions unanswered. The azimuthal wavelength
calculated by Stuart for the inviscid neutral wave gave the result that there should be 113 vortices around
the circumference at R = 443, whereas in the experiments only about 30 were observed. This discrepancy
was attributed to the neglect of viscosity. Another reason for the discrepancy, not mentioned at the time, is
that the theory dealt with neutral waves, while the waves that form in the china clay were unstable spatial
waves, i.e., they were amplifying in the outward radial direction. Brown (1960) calculated a neutral-stability
curve from the Orr-Sommerfeld equation for the velocity profile in the direction 11.5◦ [said to be measured
from the photograph in Gregory et al. (1955)], and also determined the locus in α − R space of unstable
stationary temporal waves with this wave angle. According to Brown’s calculation, the number of vortices
at R = 433 is 23.6, and at R = 540 is 31.5. These numbers are more in accord with experiment, but no
explanation was given as to why these particular waves should be observed.
108
Figure 12.2: Spatial amplification rate vs. azimuthal wavenumber at seven Reynolds numbers for zero-
frequency waves; sixth-order system.
Figure 12.3: Wave angle vs. azimuthal wavenumber at three Reynolds numbers for zero-frequency waves;
sixth-order system.
109
Figure 12.4: ln(A/A0 ) vs. azimuthal wavenumber at four Reynolds numbers for zero-frequency waves and
wave angle at peak amplitude ratio; sixth-order system.
necessary to mention that the normal-mode solution represented in Figs. 12.2 and 12.3 is not unique. There
is a second solution with larger wave angles that is completely damped for R ≤ 500. At R = 500, the
minimum wave angle of this solution is 18.3◦ at βθ = 23.5, and the minimum damping is 1.8 × 10−3 at
βθ = 22.2. At a Reynolds number somewhere above 500, the two solutions exchange identities for certain
βθ , with consequences that have not yet been worked out.
The logarithm of the amplitude ration A/A0 obtained by integrating σ along the radius is given in
Fig. 12.4 at R = 35, 400, 450 and 500. The reference amplitude A0 is at R = 250, rather than at the lower-
branch neutral point of each Fourier component. The wave angle at the maximum amplitude of each R is
noted in the figure. These numerical results differ from those of Malik and Orszag (1981) because here the
irrotationality condition, Eq. 2.55c, has been applied to the wavenumber vector of each Fourier component.
For the disk, this condition is that the azimuthal wavenumber βθ , or number of vortices, is constant. That
is, in Fig. 12.2 the path of integration is parallel to the ordinate. In Fig. 12.4, ln(A/A0 ) is given as a function
of both βθ and (βr )0 , the value of β at the reference Reynolds number of 250. We observe that although
the bandwidth of βθ for which A is greater than A0 increases with increasing R, the bandwidth for which
A/A0 is within 1/e of the maximum amplitude ratio decreases slightly. The values of ln(A/A0 ) in this figure
contrast with the much higher values obtained by Cebeci and Stewartson (1980b) from the fourth-order
system and the parallel-flow saddle point criterion. Transition is usually observed to start at a Reynolds
number in the vicinity of 500, so that the N factors of Fig. 12.4 are of the magnitude customarily associated
with transition in 2D boundary layers. Thus we see that crossflow instability in the rotating-disk boundary
layer is powerful enough to lead to transition at lower-than-normal Reynolds numbers where the streamwise
profile is completely stable.
110
Figure 12.5: Ensemble-averaged normalized velocity fluctuations of zero-frequency waves at ζ = 1.87 on
rotating disk of radius rd = 22.9 cm. Roughness element at Rs = 249, θs = 173◦ . [After Fig. 18 of
Wilkinson and Malik (1983)]
111
Figure 12.6: Normalized wave forms and constant-phase lines of calculated wave pattern produced by zero-
frequency point source at Rs = 250 in rotating-disk boundary layer.
angle. The shift of hte wave pattern to the right in Figs. 12.5 and 12.6 with respect to the constant-phase
lines is because amplitude propagates essentially along group-velocity trajectories. The agreement between
Figs. 12.5 and 12.6 conclusively demonstrates that the observed stationary waves on a rotating disk are the
result of the superposition of the entire spectrum of normal modes, both amplified and damped.
The calculated amplitudes along the constant-phase lines are given in Fig. 12.7. Vortex No. 11 is the one
that comes from the point source, and it is the only one with an amplitude minimum, which, it should be
noted, is well beyond the critical Reynolds number of 273. The reference amplitude of this vortex was selected
to fit the minimum amplitude of the experiment, adn then used for all the other vortices. A comparison
is given in Fig. 12.8 of the calculated and experimental envelope amplitude distributions at R = 400 and
466. In this figure, the experimental amplitudes have been normalized to the arbitrary theoretical maximum
amplitude at R = 400. At R = 400, the agreement is excellent except at the right-hand edge of the wave
pattern, where a second wave pattern was present in the experiment. At R = 466, the influence of the
second wave pattern has spread almost to the center of the principal wave pattern, and is the reason for the
disagreement between theory and experiment in Fig. 12.8 to the right of the maximum amplitude.
112
Figure 12.7: Calculated amplitudes along constant-phase lines of wave pattern behind zero-frequency point
source at Rs = 250 in rotating-disk boundary layer.
Figure 12.8: Comparison of calculated envelope amplitudes at R = 400 and 466 in wave pattern produced by
zero-frequency point source at Rs = 250 in rotating-disk boundary layer, and comparison with measurements
of Wilkinson and Malik (1983)(
, R = 397; , R = 466).
113
Chapter 13
where the wedge angle is (π/2)βh and βh = 2m/(m + 1) as in Eq. 2.62. We shall refer to this velocity as the
chordwise velocity. The velocity parallel to the leading edge, or spanwise velocity, is
The subscript 1 refers to the local freestream. For this inviscid flow, the boundary-layer equations in the xc
direction, as shown by Cooke (1950), reduce to
m+1
f 000 + f f 00 + βh − f 02 = 0. (13.3)
2
This equation is the usual Falkner-Skan equation for a two-dimensional boundary layer, and is indepen-
dent of the spanwise flow. The dependent variable f (y) is related to the dimensionless chordwise velocity
by
U∗
2
Uc = ∗c = f 0 (y), (13.4)
Uc1 m+1
and the independent variable is the similarity variable
s
∗
Uc1
y = y∗ . (13.5)
ν ∗ x∗c
Once f (y) is known, the flow in the spanwise direction zs∗ is obtained from
g 00 + f g 0 = 0, (13.6)
where
Ws∗
Ws = ∗ = g(y). (13.7)
Ws1
114
Figure 13.1: Coordinate systems for Falkner-Skan-Cooke boundary layers.
Both f 0 (y) and g(y) are zero at y = 0 and approach unity as y → ∞. Tabulated values of g(y) for a few
values of βh may be found in Rosenhead (1963, p. 470).
The final step is to use f 0 (y) and g(y) to construct the streamwise and crossflow velocity components
needed for the stability equations. A flow geometry appropriate to a swept back wing is shown in Fig. 13.1.
There is no undisturbed freestream for a Falkner-Skan flow, but such a direction is assumed and a yaw, or
sweep, angle ψsw is defined with respected to it. the local freestream, or potential flow, is at an angle ψp
with respect to the undisturbed freestream. It is the potential flow that defines the x, z coordinates of the
stability equations. The angle of the potential flow with respect to the chord is
∗
−1 Ws1
θ = tan ∗ , (13.8)
Uc1
and θ is related to ψsw and ψp by
θ = ψsw + ψp . (13.9)
p
With local potential velocity, U1∗ = Uc1
∗2 + W ∗2 , as the reference velocity, the dimensionless streamwise
s1
and crossflow velocity components are
These velocity profiles are defined by βh , which fixes f 0 (y) and g(y), and the angle θ. We note from
Eq. 13.10b that for a given pressure gradient all crossflow profiles have the same shape; only the magnitude
of the crossflow velocity changes with the flow direction. In contrast, according to Eq. 13.10a streamwise
profiles change shape as θ varies. for θ = 0, U (y) = f 0 (y); for θ = 90◦ , U (y) = g(y); for θ = 45◦ , the two
functions make an equal contribution.
When the Eq. 13.10 velocity profiles are used directly in the stability equations, the velocity and length
scales of the equations must be the same as in Eq. 13.10. This identifies the velocity scale as U1∗ , the length
scale as s
∗ ν ∗ x∗c
L = ∗ , (13.11)
Uc1
and the Reynolds number U1∗ L∗ /ν ∗ as
Rc
R= . (13.12)
cos θ
115
Figure 13.2: Falkner-Skan-Cooke crossflow velocity profiles for βh = 1.0, 0.2, -0.1 and SEP (separation,
-0.1988377); INF, location of inflection point; MAX, location of maximum crossflow velocity.
∗ ∗
where Rc = (Uc1 xc /ν ∗ )1/2 is the square root of the Reynolds number along the chord. For positive pressure
gradients (m > 0), θ = 90◦ at xc = 0 and θ → 0 as x → ∞; for adverse pressure gradients (m < 0), θ = 0◦
at xc = 0 and θ → 90◦ as xc → ∞. The Reynolds number Rc is zero at x = 0 for all pressure gradients, as
is R with one important exception. The exception is where m = 1 (βh = 1). For a 2D planar flow, βh = 1 is
the stagnation-point solution; here it is the attachment-line solution. In the vicinity of xc = 0, the chordwise
velocity is ∗
∗ dUc1
Uc1 = x∗c . (13.13)
dx∗c x=0
∗
The potential velocity along the attachment line is Ws1 , and the Reynolds number is
∗
Ws1
Rx=0 = ∗
, (13.14)
dUc1
ν∗
dx∗c x=0
a non-zero value.
For our purposes in this chapter, we may regard θ as a free parameter, and use the velocity profiles of
Eq. 13.10 at any Reynolds number. However, for the flow over a given wedge, θ can be set arbitrarily at
only one Reynolds number. if θref is θ at Rc = (Rc )ref , the θ at any other Rc is given by
m
m+1
(Rc )ref
tan θ = tan θref . (13.15)
Rc
For m 1, the dependence on Rc is so weak that θ is constant almost everywhere. One way of choosing
(Rc )ref within the present context is to make it the chord Reynolds number where ψp = 0; i.e., the local
potential flow is in the direction of the undisturbed freestream. Then θref is equal to the yaw angle ψsw .
Figure 13.2 shows the crossflow velocity profiles for θ = 45◦ and four values of βh . The inflection point
and point of maximum crossflow velocity (Wmax ) are also noted on the figure. In Fig. 13.3, Wmax for θ = 45◦
is given as a function of βh from near separation to βh = 1. The crossflow velocity for any other flow angle
is obtained by multiplying the Wmax of the figure by cos θ sin θ. The maximum crossflow velocity of 0.133
116
Figure 13.3: Effect of pressure gradient on maximum crossflow velocity; Falkner-Skan-Cooke boundary layers.
is generated by the separation profile rather than by the stagnation profile, where Wmax = 0.120. However,
Wmax varies rapidly with βh in the neighborhood of separation, as do all other boundary-layer parameters,
and for βh = −0.190, Wmax is only 0.102.
The function g(y) is only weakly dependent on βh , and, unlike f 0 (y), never has an inflection point even
for an adverse pressure gradient. Indeed it remains close to the Blasius profile in shape, as underlined by a
shape factor H (ratio of displacement to momentum thickness) that only changes from 2.703 to 2.539 as βh
goes from -0.1988377 (separation) to 1.0 (stagnation). The weak dependence of g(y) on βh has been made
the basis of an approximate method for calculating boundary layers on yawed cylinders. For our purposes,
it allows some of the results of the stability calculations to be anticipated. For waves with the wavenumber
vector aligned with the local potential flow, we can expect the amplification rate to vary smoothly form its
value for a two-dimensional Falkner-Skan flow to a value not too far from Blasius as θ goes from zero to 90◦ .
The stability results will be presented in terms of the Reynolds number R and the similarity length
scale L∗ . In order that the results may be converted to the length scales of hte boundary-layer thickness,
displacement thickness or momentum thickness, Table 13.1 lists the dimensionless quantities yδ = δ/L∗ ,
yδ∗ = δ ∗ /L∗ and the shape factor H of the streamwiseR profile for several combinations of βh and θ. Also
listed are Wmax , the average crossflow velocity W = ( W dy)/y; yinf , the y of the inflection point of the
crossflow velocity profile; and εinf , the deflection angle of the streamline at y = yinf . The quantity yδ is
defined as the point where U = 0.999.
117
Table 13.1: Properties of three-dimensional Falkner-Skan-Cooke boundary layers.
Figure 13.4: Effect of flow angle on maximum amplification rate with respect to frequency of ψ = 0◦ waves
at R = 1000 and 2000 in Falkner-Skan-Cooke boundary layers with βh = ±0.02.
118
Figure 13.5: Effect of pressure gradient on minimum critical Reynolds number: —–, zero-frequency crossflow
instability waves in Falkner-Skan-Cooke boundary layers with θ = 45◦ ; - - -, 2D Falkner-Skan boundary layers
[from Wazzan et al. (1968a)].
a ratio to the Balsius value (σb )max . It will be recalled that with βh = 0, g(y) = f 0 (y), and the velocity
profile remains the Blasius function for all flow angles. The effect of a non-zero flow angle with βh 6= 0 is
destabilizing for a favorable pressure gradient, and stabilizing for an adverse pressure gradient. Consequently,
it reduces the pressure-gradient effect of 2D Falkner-Skan boundary layers. The reason for this result is easy
to understand by reference to Eq. 13.10. We have already pointed out in Section 13.1 that the spanwise
velocity profile g(y) is always close to the Blasius function. Thus as the flow angle increases from zero, the
amplification rate must change from the two-dimensional Falkner-Skan value at θ = 0◦ to a value not far
from Blasius at θ = 90◦ .
As discussed previously, the only physically meaningful flow with θ = 90◦ and a non-zero Reynolds
number is the attachment-line flow (βh = 1). For all other values of βh , R at this flow angle must be either
zero (βh > 0) or infinite (βh < 0). With βh = 1.0 and R = 1000 (Rθ = 404.2, where Rθ is the momentum-
thickness Reynolds number), σmax /(σb )max = 0.766. The minimum critical Reynolds number of this profile
is (Rθ )cr = 268 (the parallel-flow Blasius value is 201), yet turbulent bursts have been observed as low as
Rθ = 250 for small disturbances by Poll (1977).
We must still show that the waves with ψ = 0◦ properly represent the maximum instability of three-
dimensional profiles with small crossflow. For this purpose a calculation was made of σ as a function of ψ
for βh = −0.02, θ = 45◦ , R = 1000 and F = 0.4256 × 10−4 , the most unstable frequency for ψ = 0◦ at this
Reynolds number. It was found that the crossflow indeed introduces an asymmetry into the distribution of
σ with ψ, and the maximum of σ is located at ψ = −6.2◦ rather than at 0◦ . However, this maximum value
differs from the σmax of Fig. 13.4 by only 0.7%.
119
Figure 13.6: Effect of flow angle on minimum critical Reynolds number of zero-frequency crossflow waves for
βh = 1.0 and -0.1988377 Falkner-Skan-Cooke boundary layers.
instability to also be a maximum near this angle. Figure 13.5 shows the minimum critical Reynolds number
Rcr at θ = 45◦ for the zero-frequency crossflow disturbances as a function of βh . For comparison, Rcr for
Tollmien-Schlichting waves in 2-D Falkner-Skan crossflow boundary layers, as computed by Wazzan et al.
(1968a), is also given. For adverse pressure gadients, the steady crossflow disturbances become unstable
at Reynolds numbers well above the Rcr for the 2D profiles. On the contrary, for βh > 0.07 the reverse
is true, and for most pressure gradients in this range the steady disturbances become unstable at much
lower Reynolds numbers than the 2D Rcr (for βh = 1.0, the 2D Rcr is 19,280 compared to Rcr = 212 for
zero-frequency crossflow instability).
The distribution of Rcr with θ is shown in Fig. 13.6 for βh = 1.0 over the complete range of θ, and for
the separation profiles (βh = −0.1988377) over the range 0◦ < θ < 50◦ . Near θ = 0◦ and 90◦ , Rcr is very
sensitive to θ; near, but not precisely at, θ = 45◦ Rcr has a minimum. This minimum occurs close to the
maximum of the streamline deflection angle at y = yinf , εinf (see Table 13.1), which, unlike Wmax , is not
symmetrical about θ = 45◦ . Table 13.2 lists the critical wave parameters for a few combinations of βh and
θ. The extensive computations needed to fix these parameters precisely were not carried out in most cases,
and so the values in the table are not exact. It can be noted that the relation
βh
ψ= (90 − |ε|inf (13.16)
|βh |
gives ψcr to within a degree for the separation profiles, and to within 0.1◦ for the other profiles of Tables 13.1
and 13.2. This result holds in general for the most unstable wave angle.
As an example of a boundary layer which is unstable at low Reynolds number only as a result of crossflow
instability, we select βh = 1.0 and θ = 45◦ , and present results for the complete range of unstable frequencies.
Although this pressure gradient can only occur at an attachment line, Fig. 13.5 leads us to expect that all
profiles with a strong favorable pressure gradient will have similar results. For this type of profile, the
120
Table 13.2: Wave parameters at minimum critical Reynolds number of zero-frequency disturbances.
Figure 13.7: Instability characteristics of βh = 1.0, θ = 45◦ Falkner-Skan-Cooke boundary layers at R = 400:
(a) maximum amplification rate with respect to wavenumber, and unstable ψ − F region; (b) unstable k − F
region.
121
Figure 13.8: Effect of wave angle on amplification rate, wavenumber, and group-velocity angle for F =
2.2 × 10−4 at R = 276; βh = −0.10, θ = 45◦ Falkner-Skan-Cooke boundary layer.
minimum critical Reynolds number of the least stable frequency is very close to the Rcr of Fig. 13.5. We
therefore choose R = 400, which is well above Rcr and where the instability is fully developed, and present
a summary of the instability characteristics in Fig. 13.7.
Figure 13.7a gives σmax as a function of the dimensionless frequency F , and also shows the portion of the
ψ − F plane for which there is instability. The unstable region is enclosed between the curves marked ψU and
ψL . These curves represent either neutral stability points or extrema of ψ. The corresponding wavenumber
magnitudes are shown in Fig. 13.7b. The negative frequencies signify that with ψ taken to be continuous
through F = 0, the phase velocity changes sign. If we choose ψ so that the wavenumber and phase velocity
are both positive, then it is ψ that changes sign at F = 0. Consequently, there are two groups of positive
unstable frequencies with quite different wave angles. The first group, which includes the peak amplification
rate, is oriented anywhere from 5◦ to 31◦ (clockwise) from the direction opposite to the crossflow direction.
The second group is oriented close to the crossflow direction itself.
122
Figure 13.9: Instability characteristics of βh = −0.10, θ = 45◦ Falkner-Skan-Cooke boundary layer at
R = 555: (a) maximum amplification rate with respect to wavenumber, and unstable k − F region; (b)
unstable ψ − F region.
123
at R = 555. Figure 13.9a shows σmax as a function of F (here, as in Fig. 13.7, σmax is the maximum with
respect to k), as well as the unstable region of the k − F plane; the unstable region of the ψ − F plane
appears in Fig. 13.9b. These two unstable regions are quite different from those of Fig. 13.7 where there is
only crossflow instability. The negative frequencies do resemble those of Fig. 13.7 in that the unstable range
of ψ is small, the unstable range of k is large, and with ψ redefined so that F > 0, the orientations are close to
the crossflow direction. However, for the higher frequencies, which are by far the most unstable, the unstable
regions of Fig. 13.9 bear more of a resemblance to those of a 2D boundary layer than to Fig. 13.7. The main
differences from the 2D case are the asymmetry about ψ = 0◦ already noted in Fig. 13.8, the one-sidedness
of ψmax , and, for F < 0.4 × 10−4 , the replacement of a lower cutoff frequency for instability by a rapid shift
with decreasing frequency to waves oriented opposite to the crossflow direction and which are unstable down
to zero frequency. The instability shown in Fig. 13.9 represents primarily an evolution of the small crossflow
boundary layers of Fig. 13.4 to larger crossflow. Only the lower frequencies, say F < 0.2 × 10−4 , have to do
with the pure crossflow instability of Fig. 13.7. For frequencies near 0.4 × 10−4 , ψ varies little with k in one
part of the unstable region, as with crossflow instability; in the other part, as with streamwise instability,
the opposite is true. This behavior becomes more pronounced at high Reynolds numbers.
124
Chapter 14
The 3D boundary layers that have received the most attention in aeronautical practices are those on transonic
swept wings. The desirability of maintaining laminar flow on the wings of large transonic aircraft has led to
the study of the instability of such boundary layers as a means of estimating the occurrence of transition and
the effectiveness of various methods of laminar-flow control. The basic phenomenon of crossflow instability
was encountered and its origin explained by the early investigators, as we have learned in Chapter 12, and
means of coping with its adverse consequences were developed. However, interest in laminar-flow control
was waning by the time computer-aided stability analysis became commonplace in the 1960’s, and nothing
more was done on the subject of 3D boundary-layer stability following Brown’s work (1959; 1960; 1961)
until the energy crisis of the mid-1970’s. In response to the sudden need for an analysis tool, Srokowski and
Orszag (1977) brought out the SALLY code. In spite of using the incompressible stability theory and a non-
physical method of computing wave amplitude, this code has been widely used. It has since been superseded
by COSAL, a compressible version of SALLY (Malik, 1982). Work that was directed at developing more
fundamental methods of stability analysis for swept wing boundary layers was carried out by Cebeci and
Stewartson (1980a,b), Lekoudis (1979, 1980), Mack (1979a, 1981), and Nayfeh (1980a,b).
Attention has so far been restricted to infinite-span swept wings. Even with this simplification, the
nonsimilarity of the boundary layers has made it necessary to proceed on the basis of specific examples,
and to try and glean a general understanding of the instability of this type of boundary layer on the basis
of extensive numerical calculations. We shall follow this same practice in this chapter. Detailed numerical
results for a single example that were obtained by an application of methods already presented in this
document (Mack, 1979a) are given in the hope that a careful study will yield some understanding of the
instabilities that arise and the procedures to follow in analyzing them.
is two-dimensional, and is non-zero at the attachment line. The Reynolds number Re s is U1∗ s∗ /ν1∗ . The
velocity derivative which defines m and thus the Hartree βh was evaluated by the numerical differentiation
of Uc1 as calculated from the pressure coefficient. The very large −βh near the trailing edge have been
omitted from the table.
Table 14.2 lists some properties of the boundary-layer solution calculated for the potential flow of Ta-
125
Figure 14.1: Coordinate systems used for infinite-span swept wing.
126
ble 14.1 and the suction distribution CQ (s∗ ) given in the last column. For comparison, the profile parameters
of an insulated flat-plate boundary layer with no suction at M1 = 1.2 are yδ = 5.85, yδ∗ = 2.31 and H = 3.54.
The maximum crossflow is Wmax = −0.115, and it occurs at station N = 4 (s∗ /c∗ = 0.0059). This value
is virtually identical to the maximum possible crossflow generated by the Falkner-Skan-Cooke profiles of
Chapter 13, where Wmax = −0.119 for βh = 1.0 and ψsw + ψp = 45◦ . Although the pressure gradient first
becomes adverse at N = 14, Wmax does not change sign until N = 21. For N = 15 to 26 (S ∗ /c∗ = 0.128 to
0.625), there are two inflection points in the crossflow velocity profile. Up to N = 20 (s∗ /c∗ = 0.364), W
has the same sign at both inflection points; for N > 20, W has opposite signs. There is a reverse crossflow
from N = 17 to 27 (s∗ /c∗ = 0.211 to 0.661). The angle εi is the angle of the velocity vector at the y of the
inflection point of the crossflow velocity profile. When there are two inflection points, the listed εi is for the
outer point when Wmax < 0, and for the inner point when Wmax > 0.
127
Table 14.2: Properties of mean boundary layer
128
Figure 14.2: Amplification rate, wave angle, and group-velocity angle as functions of wavenumber at N = 4
(R = 301) for F = 0: − · −, incompressible theory; —,sixth-order compressible theory; 35◦ swept wing.
129
Figure 14.3: Unstable frequency range at N = 4 (R = 301) for k = 0.520: − · −, incompressible theory; —,
sixth-order compressible theory; 35◦ swept wing.
The real and imaginary parts of φ are also shown in Fig. 14.2. The real part, φr , has the same sign as
Wmax , and φr at kmax , the wavenumber for maximum amplification rate, is proportional to Wmax . At kmax ,
ψ − φr = 88.6◦ . Further back on the wing, this difference approaches 90◦ as the crossflow diminishes. The
imaginary part of φ, φi , reverses sign at the point of maximum amplification rate, a behavior that is true at
all stations.
Although crossflow instability has been defined as the existence of unstable steady disturbances, a whole
band of frequencies becomes unstable at about the same critical Reynolds number. Figure 14.3 gives σ as
a function of F at N = 4 as calculated from both the incompressible and sixth-order compressible theories
for k = 0.520, and the kmax of F = 0. It is apparent that the effect of compressibility is about the same for
all frequencies as for F = 0. The values of σmax × 103 for k = 0.520 are: incompressible, 8.91; sixth-order
compressible, 8.00; eighth-order compressible, 7.90.
There are both positive and negative unstable frequencies in Fig. 14.3. The negative frequencies simply
mean that with the direction of ~k defined by the values of ψ shown in the figure, the phase velocity is
negative. If, instead, the direction of ~k had been defined in the usual manner to be the direction of the
phase velocity, there would be two groups of positive unstable frequencies. For the positive frequencies of
Fig. 14.3, ~k is within 25◦ of the direction opposite to the crossflow; for the negative frequencies, ~k is within
5◦ of the direction of the crossflow. The sign convention of Fig. 14.3 has been adopted so that the maximum
instability will always be associated with a positive frequency, and this nonstandard definition of ~k is used
here, as in Section 13.3, to make it easier to plot the numerical results.
There is a wide band of unstable frequencies in Fig. 14.3. The dimensional frequency corresponding
to F = 2.0 × 104 is 57.8 kHz, and the most unstable frequency is about 17 kHz. The unstable bandwidth
becomes much narrower further downstream as both the crossflow and maximum amplificatoin rate decrease.
It was found in Mack (1979b) that for a boundary layer with crossflow instability only, kmax does not vary
much with frequency for F > 0. For F < 0, kmax does change with frequency. It is estimated from Chap. 13
that the unstable region of negative frequencies at N = 4 extends to F = −1.5 × 10−4 with ψ = 92◦ , and
that the maximum amplification rate of the wave with ~k in the crossflow direction is about one-half of that
for the steady disturbance.
As crpssflow instability is an example of inflectional instability, it is possible to deduce something about
the stability characteristics from the location of the inflection point of the relevant velocity profile. It is not
necessary for this purpose to consider the generalized inflection point of the compressible theory which is
little different from the true inflection point at transonic Mach numbers. The relevant velocity profile is the
one in the direction of ~k. For the most unstable stationary wave (ψ = 84.8◦ ), this profile has reverse flow,
130
and the inflection point occurs almost at the zero velocity point in accordance with the theory of Stuart
(Gregory et al., 1955). Inflectional profiles exist for ψ both greater and less than 84.8◦ . The sign of the
mean velocity at the inflection point determines the sign of the frequency (except very near F = 0 because
of the finite Reynolds number). The profiles with ψ > 84.8◦ give the negative frequencies; the profiles with
ψ < 84.8◦ the positive frequencies.
131
Figure 14.4: Crossflow and streamwise instability at N = 15 (R = 1323); (a) maximum amplification rate
(with respect to frequency) and frequency as functions of wave angle; (b) maximum amplification rate (with
respect to wavenumber) as function of frequency: − · −, incompressible theory; —, sixth-order compressible
theory; 35◦ swept wing.
132
Figure 14.5: Crossflow and streamwise instability at N = 23 (R = 2661). (a) Maximum amplification rate
(with respect to frequency) and frequency as function of wavenumber angle; (b) maximum amplification
rate (with respect to wavenumber) as function of frequency: − · −, incompressible theory; —, sixth-order
compressible theory; 35◦ swept wing.
133
Figure 14.6: Amplification rates of seven zero-frequency wave components in forward instability region of
35◦ swept wing with irrotationality condition applied to wavenumber vector: —, incompressible theory; - - -,
sixth-order compressible theory for k1 = 0.35.
134
Figure 14.7: ln(A/A0 ) of six zero-frequency wave components in forward instability region of 35◦ swept
wing with irrotationality condition applied to wavenumber vector and comparison with SALLY code; —,
incompressible theory; - - -, sixth-order compressible theory for k1 = 0.35; − · −, eighth-order compressible
theory for k1 = 0.35.
Reynolds numbers. it is this pattern that prevails in the entire rear crossflow instability region. There the
wave component with ki = 0.35 at N = 21 (R = 2400, s∗ /c∗ = 0.419) is the most unstable from R = 2600
to at least R = 3570 (s∗ /c∗ = 0.906). Consequently, in this region the procedure we are using here gives the
same result for the amplitude ratio as does the SALLY code.
The ln(A/A0 ) values that correspond to the amplification rates of Fig. 14.6 are shown in Fig 14.7 for six
Fourier components along with the result given by the SALLY code (computed by Dr. A. Srokowski). The
present method gives a peak in the envelope curve, ln(A/A0 )max vs. R, at about R = 1400 (s∗ /c∗ = 0.128).
In contrast, the curve from the SALLY code continues to rise to a value of ln(A/A0 ) = 11.2 at R = 1880
(s∗ /c∗ = 0.259). The peak with the irrotationality condition is a consequence of following Fourier components
from a more unstable region to a less unstable region, and can also be encountered in 2D boundary layers
with laminar-flow control.
Two additional curves included in Fig. 14.7 give ln(A/A0 ) for ki = 0.35 as computed from the sixth- and
eighth-order compressible equations. The peak ln(A/A0 ) of the latter is about 6.9 compared to 7.8 from
the incompressible theory and 11.2 from the SALLY program. Consequently, the method of integrating the
maximum amplification rate overestimates the peak amplitude ratio by over 70 times.
As both Figs. 14.3 and 14.4b show that a non-zero frequency has the maximum amplification rate for
crossflow instability, it is also a non-zero frequency that gives the maximum amplitude ratio. The possi-
ble importance of these frequencies is, however, counteracted by the narrowing of the unstable frequency
bandwidth in the downstream direction. The results is that at N = 15 the frequency with the maximum
amplitude ratio is the low frequency F = 0.05 × 104 (1.4 kHz), and the peak ln(A/A0 ) of this frequency is
only 2% larger than for zero frequency. Of course, larger differences than this exist upstream of N = 15
where higher frequencies are still unstable.
at station N = 35 in the rear crossflow instability region, the amplitude ratio of the most unstable
zero-frequency wave component, ki = 0.35, is 6.54 according to the incompressible theory, a result almost
135
identical to the SALLY value of n = 6.46. However, compressibility cannot be neglected in this region as it
was in the forward instability region. The sixth-order compressible theory gives ln(A/A0 ) = 5.25 at N = 35;
thus the incompressible theory overestimates ln(A/A0 ) by 25%.
Streamwise instability is limited to the region from N = 15 to N = 25 (s∗ /c∗ = 0.588) and leads to
smaller amplitude ratios than does crossflow instability. As these waves travel downstream, their wave angle
ψ remains very close (within about 1◦ ) to its initial value, in contrast to the crossflow disturbances which
are required by the dispersion relation to keep their angles within the narrow band set by the profile angle
εi . According to Fig. 14.7b, F = 0.375 × 10−4 is the most unstable frequency for streamwise instability
at N = 15. However, this and the neighboring unstable frequencies damp out quickly in the downstream
direction. The frequencies which give the largest amplitude ratios are those which are unstable at N = 23
(s∗ /c∗ = 0.513), where the largest amplification rates of streamwise instability occur. For example, F =
0.09 × 10−4 becomes unstable at N = 21 and has a peak ln(A/A0 ) of 2.3 at N = 25 for an initial wave angle
of −70◦ ; F = 0.15 × 10−4 becomes unstable at N = 20 and has a smaller peak at N = 24. Consequently,
the maximum amplitude growth of streamwise-instability waves is only about 1% of that of the crossflow
disturbances. Examples of amplification rates for a wing without suction may be found elsewhere (Mack,
1981).
136
Appendix A
There are 30 non-zero elements of the coefficient matrix ai,j (y) of Eq. 8.11. The Z1 equation has only one
non-zero coefficient:
a12 = 1. (A.1)
The Z2 equation has six non-zero coefficients:
iR
a21 = (αU + βW − ω) + α2 + β 2 , (A.2a)
µT
1 dµ
a22 = − DT, (A.2b)
µ dT
R 1 dµ
a23 = (αDU + βDW ) − i α2 + β 2 DT
µT µ dT
1 DT
− i (1 + 2d) α2 + β 2 , (A.2c)
3 T
iR 2 1
α + β 2 − (1 + 2d) α2 + β 2 γM12 (αU + βW − ω) ,
a24 = (A.2d)
µ 3
11 1 dµ
(1 + 2d) α2 + β 2 (αU + βW − ω) − αD2 U + βD2 W
a25 =
3T µ dT
1 d2 µ
− DT (αDU + βDW ) , (A.2e)
µ dT 2
1 dµ
a26 = − (αDU + βDW ) . (A.2f)
µ dT
The Z3 equation has four non-zero coefficients:
137
R 2
E= + i (2 + d) γM12 (αU + βW − ω) , (A.4)
µ 3
the six non-zero coefficients are
i 2 dµ 2 DT
a41 = − DT + (2 + d) , (A.5a)
E µ dT 3 T
i
a42 = − , (A.5b)
E
1 2 DT 2 1 dµ
a43 = − α2 + β 2 + (2 + d)
E 3 T µ dT
2
2 D T iR
+ (2 + d) − (αU + βW − ω) , (A.5c)
3 T µT
i 2 1 dµ
a44 = − (2 + d) γM12 (αU + βW − ω) DT +
E3 µ dT
DT
+αDU + βDW + (αU + βW − ω) , (A.5d)
T
i 1 dµ
a45 = (αDU + βDW )
E µ dT
2 1 dµ DT (αDU + βDW )
+ (2 + d) (αU + βW − ω) + , (A.5e)
3 µ dT T T
i 2
a46 = (2 + d) (αU + βW − ω) . (A.5f)
E3
The Z5 equation has only one non-zero coefficient:
a56 = 1. (A.6)
The Z6 equation has six non-zero coefficients:
(αDU + βDW )
a62 = −2σ (γ − 1) M12 , (A.7a)
(α2 + β 2 )
Rσ DT
a63 = − i2σ (γ − 1) M12 (αDU + βDW ) , (A.7b)
µ T
Rσ
a64 = −i (γ − 1) M12 (αU + βW − ω) , (A.7c)
µ
Rσ D2 T dκ
a65 = i (αU + βW − ω) + α2 + β 2 −
µT κ dT
2 2
DT d κ 1 dµ
− σ (γ − 1) M12 DU 2 + DW 2 ,
− 2
(A.7d)
κ dT µ dT
2 dκ
a66 = − DT, (A.7e)
κ dT
(αDW − βDU )
a68 = −2σ (γ − 1) M12 . (A.7f)
(α2 + β 2 )
a78 = 1. (A.8)
The Z8 equation has five non-zero coefficients:
138
R
a83 = (αDW − βDU ) , (A.9a)
µT
1 dµ 1 d2 µ
a85 =− αD2 W − βD2 U − DT (αDW − βDU ) , (A.9b)
µ dT µ dT 2
1 dµ
a86 =− (αDW − βDU ) , (A.9c)
µ dT
iR
a87 = (αU + βW − ω) + α2 + β 2 , (A.9d)
µT
1 dµ
a88 =− DT. (A.9e)
µ dT
In these equations, the ratio of the second to the first viscosity coefficient
λ
d= (A.10)
µ
is taken to be a constant and equal to 1.2. (Stokes’ assumption corresponds to λ = 0).
In the numerical computations, we use
T ∗3/2
µ∗ × 105 = 1.458 , T ∗ ≥ 110.4K
T ∗ + 110.4 (A.11)
= 0.0693873T ∗ , T ∗ < 110.4K
for the viscosity coefficient in cgs units, and
T ∗1/2
κ∗ = 0.6325 (A.12)
1 + (245.4/T ∗ ) 10−12/T ∗
for the thermal conductivity coefficient in cgs units. The Prandtl number σ = c∗p µ∗ /κ∗ is computed as a
function of temperature from µ∗ , κ∗ , and a constant specific heat of c∗p = 0.24.
139
Appendix B
where
V 1 = Zz , V2 = Z4 , V3 = Z5 . (B.2)
The three coefficients of Eq. B.1a are
where E1 is Eq. A.4 evaluated in the freestream. The two coefficients of Eq. B.1c are
where the V (i) are the six three-component solution vectors, the λi are the six characteristic values, and
the B (i) are the six three-component characteristic vectors. Upon substituting Eq. B.6 into Eqs. B.1, the
140
characteristic values are found to be
( 1/2 )1/2
1 1 2
λ1,2 = ∓ (b22 + b33 ) − (b22 − b33 ) + b23 b32 , (B.7a)
2 4
1/2
λ3,4 = ∓ (b11 ) , (B.7b)
( 1/2 )1/2
1 1 2
λ5,6 = ∓ (b22 + b33 ) + (b22 − b33 ) + b23 b32 , (B.7c)
2 4
(B.7d)
where the numbering has been arranged so that the first two of these will correspond to the first two of
Eq. 2.49.
The last two of Eqs. 8.11 give a fourth uncoupled second-order equation
D2 Z7 = α2 + β 2 + iR (αU1 + βW1 − ω) Z7 ,
(B.8)
141
References
Bellman, R.E. and Kalaba, R.E. 1965 Quasilinearization and Boundary-Value Problems, American Elsevier,
N.Y.
Betchov, R. and Criminale, W.O. 1967 Stability of Parallel Flows, Academic Press, N.Y.
Bouthier, M. 1972 Stabilité linéaire des écoulements presque paralléles, J. de Mécanique, Vol. 11, pp. 599–621.
Bouthier, M. 1973 Stabilité linéaire des écoulements presque paralléles, Partie II, J. de Mécanique, Vol. 12,
pp. 76–95.
Brown, W.B. 1959 Numerical Calculation of the Stability of Cross Flow Profiles in Laminar Boundary Layers
on a Rotating Disc and on a Swept Wing and an Exact Calculation of the Stability of the Blasius Velocity
Profile, Report N. NAI 59-5, Northrop Aircraft Inc., Hawthorne, CA.
Brown, W.B. 1960 Exact Numerical Solution of the Stability Equations of the Laminar Boundary Layer,
Paper presented at International Congress of Applied Mechanics, Stresa, Italy.
Brown, W.B. 1961 A Stability Criterion for Three-Dimensional Laminar Boundary Layers, in “Boundary
Layer and Flow Control” (G. V. Lachmann, ed.), Vol. 2, pp. 913–923, Pergamon Press, London.
Brown, W.B. 1962 Exact Numerical Solutions of the Complete Linearized Equations for the Stability of
Compressible Boundary Layers, Norair Report No. NOR-62-15, Northrop Aircraft Inc., Hawthorne, CA.
Brown, W.B. and Sayre, P.H. 1954 An Exact SOlution of the Orr-Sommerfeld Stability Equation for Low
Reynolds Numbers, Report No. BLC-43, Northrop Aircraft Inc., Hawthorne, CA.
Cebeci, T. and Stewartson, K. 1980a Stability and Transition in Three-Dimensional Flows, AIAA J., Vol.
18, pp. 398–405
Cebeci, T. and Stewartson, K. 1980b On the Prediction of Transition in Three-Dimensional Flows, in
“Laminar-Turbulent Transition” (R. Eppler and H Fasel, eds.), Springer Verlag, Berlin.
Cochran, W.G. 1934 The Flow due to a Rotating Disk, Proc. Cambridge Phil. Soc., Vol. 30, pp. 165–375.
Conte, S.D. and Miles, J.W. 1959 On the Numerical Integration of the Orr-Sommerfeld Equation, SIAM J.
Vol. 7, pp. 361–366.
Conte, S.D. 1966 The numerical solution of linear boundary value problems, SIAM Review. Vol. 8, pp.
309–321.
Cooke, J.C. 1950 The Boundary layer of a Class of Infinite Yawed Cylinders, Proc. Cambridge Phil. Soc.,
Vol. 46, pp. 645–648.
Corner, D., Houston, D.T.R. and Ross, M.A.S. 1976 Higher Eigenstates in Boundary-Layer Stability Theory,
J. Fluid Mech., Vol. 77, pp. 81–103.
Criminale, W.O. and Kovasnay, L.S.G. 1962 The Growth of Localized Disturbances in a Laminar Boundary
Layer, J. Fluid Mech., Vol. 14, pp. 59–80.
142
Davey, A. 1973 A Simple Numerical Method for Solving Orr-Sommerfeld Problems, Quart. J. Mech. Appl.
Math., Vol. 26, pp. 401–411.
Davey, A. 1977 On the Numerical Solution of Difficult Eigenvalue Problems, J. Comp. Phys., Vol. 24, pp.
331–338.
Davey, A. 1979 On the Removal of the Singularities from the Riccati Method, J. Comp. Phys., Vol. 30, pp.
137–144.
Davey, A. 1981 A Difficult Numerical Calculation Concerning the Stability of the Blasius Boundary Layer,
in Proceedings of Numbrecht Conf.
Demetriades, A. 1977 Laminar Boundary Layer Stability Measurements at Mach 7 Including Wall Temper-
ature Effects, AFOSR TR-77-1311.
Dougherty, N.S. and Fisher, D.F. 1980 Boundary Layer Transition on a 10◦ Cone: Wind Tunnel/Flight Data
Correlations, AIAA Paper No. 80-0154 ; also see NASA TP 1971 1982.
Dryden, H. 1959 Transition from Laminar to Turbulent Flow, in “Turbulent Flows and Heat Transfer” (C.
C. Lin, ed.), pp. 1–74, Princeton Univ. Press, Princeton
Drazin, P.G. and Howard, L.N. 1966 Hydrodynamic Stability of Parallel Flow of Inviscid Fluid, in “Advances
in pplied Mechanics” (G. Kuerti, ed.), Vol. 9, pp. 1–89, Academic Press, N.Y.
Drazin, P.G. and Reid, W.H. 1981 Hydrodynamic Stability, Cambridge Univ. Press, Cambridge.
Dunn, D.W. and Lin, C.C. 1955 On the Stability of the Laminar Boundary Layer in a Compressible Fluid,
J. Aero. Sci., Vol. 22, pp 455–577.
El-Hady, N.M. and Nayfeh, A.H. 1979 Nonparallel Stability of Compressible Boundary-layer Flows, Report
VPI-E-79-13, Virginia Polytechnic Institute and State Univ., Blacksburg, VA.
Fjortoft, R. 1950 Application of Integral Theorems in Deriving Criteria of Stability for Laminar Flows and
for the Baroclinic Circular Vortex, Geofys. Publ., Vol. 17.
Foote, J.R. and Lin, C.C. 1950 Some Recent Investigations in the Theory of Hydrodynamic Stability, J.
Fluid Mech., Vol. 8, pp. 165–280.
Gallagher, A.P. and Mercier, A.M. 1962 On the Behavior of Small Disturbances in Plane Couette Flow, Part
I, J. Fluid Mech., Vol. 13, pp. 91–100.
Gaster, M. 1962 A Note on a Relation between Temporally Increasing and Spatially Increasing Disturbances
in Hydrodynamic Stability, J. Fluid Mech., Vol. 14, pp. 222–224.
Gaster, M. 1963 On the Stability of Parallel Flows and the Behavior of Separation Bubbles, Ph.D. Thesis,
Univ. of London.
Gaster, M. 1965a On the Generation of Spatially Growing Waves in a Boundary Layer, J. Fluid Mech., Vol.
22, pp. 433–441.
Gaster, M. 1965b The Role of Spatially Growing Waves in the Theory of Hydrodynamic Stability, in “Progress
in Aeronautical Sciences” (D. Küchemann and L. H. G. Sterne, eds.), Vol. 6, Pergamon Press, Oxford
Gaster, M. 1968 The Development of Three-Dimensional Wave Packets in a Boundary Layer, J. Fluid Mech.,
Vol. 32, pp. 173–184.
Gaster, M. 1974 On the Effect of Boundary Layer Growth in Flow Stability, J. Fluid Mech., Vol. 66, pp.
465–480.
Gaster, M. 1975 A Theoretical Model of a Wave Packet in the Boundary Layer on a Flat Plate, Proc. Roy.
Soc. A, Vol. 347, pp. 271–289.
143
Gaster, M. 1978 On the Application of Ray Mathematics to Non-Conservative Systems, Proceedings of NSF
Regional Conference on Geophysical Fluid Dynamical Wave Mathematics, pp. 61–66, Univ. of Washington,
Seattle, Wash.
Gaster, M. 1981a On Transition to Turbulence in Boundary Layers, in “Transition to Turbulence” (R. E.
Meyer, ed.), Proceedings of Symposium, Univ. of Wisconsin 1980, Academic Press, N.Y.
Gaster, M. 1981b Propagation of Linear Wave Packets in Laminar Boundary Layers, AIAA J., Vol. 19, pp.
419–423.
Gaster, M. 1982a Estimates of Errors Incurred in Various Asymptotic Representations of Wave Packets, J.
Fluid Mech., Vol. 121, pp. 365–377.
Gaster, M. 1982b The Development of a Two-Dimensional Wave Packet in a Growing Boundary Layer, Proc.
Roy. Soc. A, Vol. 384, pp. 317–332.
Gaster, M. and Davey, A. 1968 The Development of Three-Dimensional Wave Packets in Inviscid Parallel
Flows, J. Fluid Mech., Vol. 32, pp. 801–808.
Gaster, M. and Grant, I. 1975 An Experimental Investigation of the Formation and Development of a Wave
Packet in a Laminar Boundary Layer, Proc. Roy. Soc. A, Vol. 347, pp. 253–269.
Gilev, V.M., Kachanov, Y.S. and Kozlov, V.V. 1981 Development of Spatial Wave Packets in a Boundary
Layer, Preprint No. 34-81, Inst. of Theoretical and Appl. Math., USSR Acad. Sci., Novosibirsk.
Gill, A.E. 1965a A Mechanism for Instability of Plane Couette Flow and of Poiseuille Flow in a Pipe, J.
Fluid Mech., Vol. 21, pp. 505–511.
Gill, A.E. 1965b Instabilities of ”Top-Hat” Jets and Wakes in Compressible Fluids, Phys. Fluids, Vol. 8, pp.
1428–1430.
Godunov, S.K. 1961 On the Numerical Solution of boundary value Problems for Systems of Ordinary Dif-
ferential Equations, Uspekhi Mat. Nauk, Vol. 16, pp. 171–174.
Gray, W.E. 1952 The effect of wing sweep on laminar flow, Royal Aircraft Establishment.
Gregory, N., Stuart, J.T. and Walker, W.S. 1955 On the Stability of Three-Dimensional Boundary Layers
with Application to the Flow Due to a Rotating Disk, Phil. Trans. Roy. Soc. A, Vol. 248, pp. 155–199.
Grosch, C.E. and Salwen, H. 1968 The Stability of Steady and Time-Dependent Plane Poiseuille Flow, J.
Fluid Mech., Vol. 34, pp. 177–205.
Grosch, C.E. and Salwen, H. 1978 Continuous Spectrum of the Orr-Sommerfelf Equation. Part I. The
Spectrum and the Eigenfunctions, J. Fluid Mech., Vol. 87, pp. 333–354.
Hama, F.R., Williams, D.R. and Fasel, H. 1980 Flow Field and Energy Balance According to the Spatial
Linear Stability Theory of the Blasius Boundary Layer, in “Laminar-Turbulent Transition” (R. Eppler
and H. Fasel, eds.), pp. 73–85, Springer, Berlin.
Hefner, J.N., and Bushnell, D.M. 1979 Application of Stability Theory to Laminar Flow Control, AIAA
Paper No. 79-1493.
Heisenberg, W. 1924 Über Stabilität und Turbulenz von Flüssigkeitsströmen, Ann. Phys.-Lpz. (4), Vol. 79,
pp. 577–627 (also NACA TM No. 1291).
Herbert, T. 1983a On perturbation methods in nonlinear stability theory, J. Fluid Mech., Vol. 126, pp.
167–186.
Herbert, T. 1983b Secondary instability of plane channel flow to subharmonic three-dimensional disturbances,
Phys. Fluids, Vol. 26, pp. 871–874.
144
Howard, L.N. 1961 Note on a paper of John W. Miles, J. Fluid Mech., Vol. 10, pp. 509–512.
Jaffe, N.A., Okamura, T.T. and Smith, A.M.O. 1970 Determination of Spatial Amplification Factors and
Their Application to Prediction Transition, AIAA J., Vol. 8, pp. 301–308.
Jordinson, R. 1970 The Flat Plate Boundary Layer. Part I. Numerical Integration of the Orr-Sommerfeld
Equation, J. Fluid Mech., Vol. 43, pp. 801–811.
Jordinson, R. 1971 Spectrum of Eigenvalues of the Orr-Sommerfeld Equation for Blasius Flow, Phys. Fluids,
Vol. 14, pp. 2535–2537.
Joseph, D.D. 1976 Stability of Fluid Motions, Springer Tracts on Natural Philosophy, Vol. 27 and 28,
Springer, Berlin.
Kaplan, B. and Moore, A.H. 1964 Implicit Finite Difference Techniques for Solving Transient Heat Conduc-
tion Problems, Air Force Inst. Tech., Wright-Patterson AFB, OH.
Kármán, Th. von 1921 Über laminare und turbulente Reibung, ZAMM, Vol. 1, pp. 233–252. (also NACA
TM No. 1092, 1946)
Kelvin, Lord 1880 On a Disturbing Infinity in Lord Rayleigh’s Solution for Waves in a Plane Vortex Stratum,
Math. and Phys. Papers, Vol. 8, pp. 186–187, Cambridge Univ. Press, Cambridge.
Kendall, J.M. 1967 Supersonic Boundary-Layer Stability Experiments, in Proceedings of Boundary Layer
Transition Study Group Meeting (W. D. McMauley, ed.), Aerospace Corp. Report No. BSD-TR-67-213,
Vol. II.
Kendall, J.M. 1971 JPL Experimental Investigations, in Proceedings of the Boundary-Layer Transition
Workshop, Vol. IV, Aerospace Corp., San Bernardino, CA.
Kendall, J.M. 1975 Wind Tunnel Experiments Relating to Supersonic and Hypersonic Boundary-Layer Tran-
sition, AIAA J., Vol. 13, pp. 240–299.
Küchemann, D. 1938 Störungsbewegungen in einger Gasströmung mit Grenzschicht, ZAMM, Vol. 18, pp.
207–222.
Kümmerer, H. 1973 Numerische Untersuchungen zur Stabilität ebener laminarer Grenzschichtströmungen,
Dissertation, Universität Stuttgart.
Kurtz, E.F., Jr. 1961 Study of the Stability of Parallel Flows, Ph.D. Thesis, Massachusetts Institute of
Technology, Cambridge, MA.
Kurtz, E.F., Jr. and Crandall, S.H. 1962 Computer-aided Analysis of Hydrodynamic Stability, J. Math.
Phys., Vol. 41, pp. 264–279.
Lakin, W., Ng, B. and Reid, W.H. 1978 Approximation to the Eigenvalue Relation for the Orr-Sommerfeld
Equation, Phil. Trans. Roy. Soc. A, Vol. 289, pp. 147–371.
Landahl, M.T. and Kaplan, R.E. 1965 Effect of Compliant Walls on Boundary Layer Stability and Transition,
AGARDograph 97, Part 1, pp. 363–394, NATO, Paris.
Laufer, J. and Vrebalovich, T. 1960 Stability and Transition on a Supersonic Laminar Boundary Layer on
an Insulated Flat Plate, J. Fluid Mech., Vol. 9, pp. 257–299.
Lees, L. 1947 The Stability of the Laminar Boundary Layer in a Compressible Fluid, NACA Tech. Report
No. 876
Lees, L. 1952 Instability of Laminar Flows and Transition to Turbulence, Consolidated Aircraft Corp., San
Diego, CA, Report No. ZA-7-006.
145
Lees, L. and Gold, H. 1964 Stability of Laminar Boundary Layers and Wakes at Hypersonic Speeds: Part I.
Stability of Laminar Wakes, Inter. Symp. on Fundamental Phenom. in Hypersonic Flow, GALCIT Memo.,
California Institute of Technology, Pasadena, CA.
Lees, L. and Lin, C.C. 1946 Investigation of the Stability of the Laminar Boundary Layer in a Compressible
Fluid, NACA Tech. Note No. 1115.
Lees, L. and Reshotko, E. 1962 Stability of the Compressible Laminar Boundary Layer, J. Fluid Mech., Vol.
12, pp. 555–590.
Lekoudis, S. 1979 Stability of Three-Dimensional Boundary Layers over Wings with Suction, AIAA Paper
No. 79-0265.
Lekoudis, S. 1980 Stability of the Boundary Layer on a Swept Wing with Wall Cooling. AIAA J., Vol. 18,
pp. 1029–1035.
Lighthill, M.J. 1963 Introduction. Boundary Layer Theory, in “Laminar Boundary Layers” (L. Rosenhead.
ed.), pp. 46–113.
Lin, C.C. 1945 On the Stability of Two-Dimensional Parallel Flows, Parts I, II, III, J. Fluid Mech., Vol. 3,
pp. 117–142, 218–234, 277–301.
Lin, C.C. 1954 Some Physical Aspects of the Stability of Parallel Flows, Proc. Nat. Acad. Sci., Vol. 40, pp.
741–747.
Lin, C.C. 1955 The Theory of Hydrodynamic Stability, Cambridge Univ. Press, Cambridge.
Lin, C.C. 1961 Some Mathematical Problems in the Theory of the Stability of Parallel Flows, J. Fluid Mech.,
Vol. 10, pp. 430–438.
Mack, L.M. 1960 Numerical Calculation of the Stability of the Compressible Laminar Boundary Layer, Jet
Propulsion Laboratory, Report No. 20-122, Pasadena, CA.
Mack, L.M. 1963 The Inviscid Stability of the Compressible Laminar Boundary Layer, in “Space Programs
Summary,” No. 37-23, p. 297, Jet Propulsion Laboratory, Pasadena, CA.
Mack, L.M. 1964 The Inviscid Stability of the Compressible Laminar Boundary Layer: Part II, in “Space
Programs Summary,” No. 37-26, Vol. IV, p. 165, Jet Propulsion Laboratory, Pasadena, CA.
Mack, L.M. 1965a Computation of the Stability of the Laminar Boundary Layer, in “Methods in Compu-
tational Physics” (B. Alder, S. Fernbach and M. Rotenberg, eds.), Vol. 4, pp. 247–299, Academic Press,
N.Y.
Mack, L.M. 1965b Stability of the Compressible Laminar Boundary Layer According to a Direct Numerical
Solution, in AGARDograph 97, Part I, pp. 329–362.
Mack, L.M. 1969 Boundary-Layer Stability Theory, Jet Propulsion Laboratory, Pasadena, CA, Document
No. 900-277, Rev. A.
Mack, L.M. 1971 Progress in Compressible Boundary Layer Stability Computations, in Proceedings of the
Boundary Layer Transition Workshop, Vol. 1, pp. 1-1 to 1-19, Aerospace Corp., San Bernardino, CA.
Mack, L.M. 1975 Linear Stability and the Problem of Supersonic Boundary-Layer Transition, AIAA J., Vol.
13, pp. 278–289.
Mack, L.M. 1976 A Numerical Study of the Temporal Eigenvalue Spectrum of the Blasius Boundary Layer,
J. Fluid Mech., Vol. 79, pp. 497–520.
Mack, L.M. 1977 Transition Prediction and Linear Stability Theory, in AGARD Conference Proceedings No.
224, pp. 1-1 to 1-22, NATO, Paris.
146
Mack, L.M. 1979a On the Stability of the Boundary Layer on a Transonic Swept Wing, AIAA Paper No.
79-0264.
Mack, L.M. 1979b Three-Dimensional Effects in Boundary-Layer Stability, in Proceedings of Twelfth Sym-
posium on Naval Hydrodynamics, pp. 63–70, National Academy of Sciences, Washington, D.C.
Mack, L.M. 1981 Compressible Boundary-Layer Stability Calculations for Sweptback Wings with Suction,
AIAA J., Vol. 20, pp. 363–369.
Mack, L.M. 1984a Line Sources of Instability Waves in a Blasius Boundary Layer, AIAA Paper No. 84-0168.
Mack, L.M. 1984b Remarks on Disputed Results in Compressible Boundary-Layer Stability Theory, Phys.
Fluids, Vol. 27, pp. 342–347.
Mack, L.M. 1985 The Wave Pattern Produced by a Point Source on a Rotating Disk, AIAA Paper No.
85-0490.
Mack, L.M. and Kendall, J.M. 1983 Wave Pattern Produced by a Localized Harmonic Source in a Blasius
Boundary Layer, AIAA Paper No. 83-0046.
Mack, L.M. and Morkovin, M.V. 1971 High-Speed Boundary-Layer Stability and Transition, Recorded Lec-
ture Series, AIAA, N.Y.
Malik, M.R. 1982 COSAL - A Black-Box Compressible Stability Analysis Code for Transition Prediction in
Three-Dimensional BOundary Layers, NASA CR-165925
Malik, M.R. and Orszag, S.A. 1981 Efficient Computation of the Stability of Three-Dimensional Boundary
Layers, AIAA Paper No. 81-1277.
Malik, M.R., Wilkinson, S.P. and Orszag, S.A. 1981 Instability and Transition in Rotating Disk Flow, AIAA
J. Vol. 19, pp. 1131–1138.
Schlichting, H. 1971a Statistical Fluid Mechanics, Volume I: Mechanics of Turbulence, M.I.T. Press, Cam-
bridge, MA.
Schlichting, H. 1971b Statistical Fluid Mechanics, Volume II: Mechanics of Turbulence, M.I.T. Press, Cam-
bridge, MA.
Morkovin, M.V. 1968 Critical Evaluation of Transition from Laminar to Turbulent Shear Layers with Em-
phasis on Hypersonically Travelling Bodies, Report AFFDL-TR-68-149, Wright-Patterson Air Force Base,
Ohio.
Morkovin, M.V. 1978 Instability, Transition to Turbulence, and Predictability, AGARDograph No. 236,
NATO, Paris.
Morkovin, M.V. 1983 Understanding Transition to Turbulence in Shear Layers, Report AFOSR FR-83,
Illinois Institute of Technology, Chicago, IL.
Morkovin, M.V. 1985 Guide to Experiments in Instability and the Laminar Turbulent Transition in Shear
Layers, in Notes from AIAA Professional Study Seminar: Instabilities and Transition to Turbulence,
Cincinnati, Ohio.
Murdock, J.W. and Stewartson, K. 1977 Spectra of the Orr-Sommerfeld Equation, Phys. Fluids, Vol. 20, pp.
1404–1416.
Nachtsheim, P. 1963 Stability of Free-Convection Boundary Layer Flows, NASA Tech. Note No. D-2089.
Nachtsheim, P. 1964 An Initial Value Method for the Numerical Treatment of the Orr-Sommerfeld Equation
for the Case of Plane Poiseuille Flow, NASA Tech. Note No. D-2414.
Nayfeh, A.H. 1980a Stability of Three-Dimensional Boundary Layers, AIAA J., Vol. 18, pp. 406–416.
147
Nayfeh, A.H. 1980b Three-Dimensional Stability of Growing Boundary Layers, in “Laminar-Turbulent Tran-
sition” (R. Eppler and H. Fasel, eds.), pp. 201–217, Springer, Berlin.
Nayfeh, A.H. and Padhye, A. 1979 The Relation between Temporal and Spatial Stability in Three-
Dimensional Flows, AIAA J., Vol. 17, pp. 1084–1090.
Ng, B. and Reid, W.H. 1979 An Initial Value Method for Eigenvalue Problems Using Compound Matrices,
J. Comp. Phys., Vol. 30, pp. 135–136.
Ng, B. and Reid, W.H. 1980 On the Numerical Solution of the Orr-Sommerfeld Equation: Asymptotic Initial
Conditions for Shooting Method, J. Comp. Phys., Vol. 38, pp. 275–293.
Noether, F. 1921 Das Turbulenzproblem, ZAMM, Vol. 1, pp. 125–138.
Obremski, H.T., Morkovin, M.V. and Landahl, M.T. 1969 A Portfolio of Stability Characteristics of Incom-
pressible Boundary Layers, AGARDograph No. 134, NATO, Paris.
Orszag, S.A. 1971 Accurate Solution of the Orr-Sommerfeld Stability Equation, J. Fluid Mech., Vol. 50, pp.
689–703.
Osborne, M.R. 1967 Numerical Methods for Hydrodynamic Stability Problems, SIAM J. Appl. Math., pp.
539–557.
Pfenninger, W. 1977 Laminar Flow Control-Laminarization, in “Special Course on Concepts for Drag Re-
duction,” AGARD Report No. 654, pp. 3-1 to 3-75, NATO, Paris.
Phillips, O. 1960 On the Generation of Sound by Supersonic Shear Layers, J. Fluid Mech., Vol. 9, pp. 1–28.
Poll, D.I.A. 1977 Leading Edge Transition on Swept Wings, in “Laminar-Turbulent Transition,” AGARD
Conference Proceedings No. 224, pp. 21-1 to 21-11, NATO, Paris.
Prandtl, L. 1921 Bemerkungen über die Enstehung der Turbulenz, ZAMM, Vol. 1, pp. 431–436.
Pretsch, J. 1942 Anfachung instabler Störungen in einer laminaren Reigungsschicht, Jb. deutsch. Luftfahrt-
for., pp. 154–171.
Radbill, J.R. and Van Driest, E.R. 1966 A New Method for Prediction of Stability of Laminar Boundary
Layers, AFOSR Report No. 66-0702.
Rayleigh, Lord 1880 On the Stability or Instability of Certain Fluid Motions, in “Scientific Papers,” Vol. 1,
pp. 474–487, Cambridge Univ. Press, Cambridge
Rayleigh, Lord 1887 On the Stability or Instability of Certain Fluid Motions. II, in “Scientific Papers,” Vol.
3, pp. 17–23, Cambridge Univ. Press, Cambridge
Rayleigh, Lord 1892 On the Question of the Stability of the Flow of Fluids, in “Scientific Papers,” Vol. 3,
pp. 575–584, Cambridge Univ. Press, Cambridge
Rayleigh, Lord 1895 On the Stability or Instability of Certain Fluid Motions. III, in “Scientific Papers,” Vol.
4, pp. 203–219, Cambridge Univ. Press, Cambridge
Rayleigh, Lord 1913 On the Stability of the Laminar Motion of an Inviscid Fluid, in “Scientific Papers,” Vol.
6, pp. 197–204, Cambridge Univ. Press, Cambridge
Reid, W.H. 1965 The Stability of Parallel Flows, in “Basic Developments in Fluid Dynamics” (M. Hold,
ed.), Vol. 1, pp. 249–307, Academic Press, N.Y.
Reshotko, E. 1960 Stability of the Compressible Laminar Boundary Layer, GALCIT Memo. No. 52, California
Institute of Technology, Pasadena, CA.
Reshotko, E. 1962 Stability of Three-Dimensional Compressible Boundary Layers, NASA Tech. Note D-1220.
148
Reshotko, E. 1976 Boundary Layer Stability and Transition, Annu. Rev. Fluid Mech., Vol. 8, pp. 311–349.
Reynolds, O. 1895 On the Dynamical Theory of Incompressible Viscous Fluids and the Determination of the
Criterion, Phil. Trans. Roy. Soc. A, Vol. 186, pp. 123–164.
Rosenhead, L. 1963 Laminar Boundary Layers, Oxford Univ. Press, Oxford.
Saric, W.S. and Nayfeh, A.H. 1975 Non-parallel Stability of Boundary Layer Flows, Phys. Fluids, Vol. 18,
pp. 945–950.
Saric, W.S. and Nayfeh, A.H. 1977 Nonparallel Stability of Boundary Layers with Pressure Gradients and
Suction, in AGARD Conference Proceedings No. 224, NATO, Paris.
Schlichting, H. 1933a Zur Entstehung der Turbulenz bei der Plattenströmung, Nachr. Ges. Wiss. Göttingen,
Math.-phys. Kasse, pp. 181–208.
Schlichting, H. 1933b Berechnung der Anfachung kleiner Störungen bei der Plattenströmung, ZAMM, Vol.
13, pp. 171–174.
Schlichting, H. 1935 Amplitudenverteilung und Energiebilanz der kleinen Störungen bei der Plattengren-
zschicht, Nachr. Ges. Wiss. Göttingen, Math.-phys. Kasse, Vol. 1, pp. 47–78.
Schlichting, H. 1940 Über die theoretische Berechnung der kritischen Reynoldschen Zahl einer Reibun-
gaschicht in beschleunigter ind verzögerter Stromung, J. deutsch. Luftfahrtfor. I, p. 97.
Schlichting, H. 1959 Entstehung der Turbulenz, in “Handbuch der Physik” (S. Flugge, ed.), Vol. VIII/I, pp.
351–450, Springer, Berlin.
Schlichting, H. 1979 Boundary Layer Theory (7th Ed.), McGraw-Hill, N.Y.
Schubauer, G.B. and Skramstad, H.K. 1947 Laminar Boundary Layer Oscillations and Transitions on a Flat
Plate, J. Aero. Sci., Vol. 14, pp. 69–76; also NACA Report 909, 1948.
Scott, M.R. and Watts, H.A. 1977 Computational Solution of Linear Two-Point Boundary Value Problems
and Orthonormalization, SIAM J. Numer. Anal., Vol. 14, pp. 40–70.
Shen, S.F. 1964 Stability of Laminar Flows, in “Theory of Laminar Flows” (F.K. Moore, ed.), pp. 719–853,
Princeton Univ. Press, Princeton.
Smith, A.M.O. and Gamberoni, N. 1956 Transition, Pressure Gradient and Stability Theory, Report No.
26388, Douglas Aircraft Co., Inc., El Segundo, CA.
Squire, H.B. 1933 On the Stability of Three-Dimensional Disturbances of Viscous Flow between Parallel
Walls, Proc. Roy. Soc. A, Vol. 142, pp. 621–628.
Srokowski, A. and Orszag, S.A. 1977 Mass Flow Requirements for LFC Wing Design, AIAA Paper No.
77-1222.
Stetson, K.F., Thompson, E.R., Donaldson, J.C. and Siler, L.G. 1983 Laminar Boundary Layer Stability
Experiments on a Cone at Mach 8. Part 1: Sharp Cone, AIAA Paper No. 83-1761.
Stetson, E.R., Donaldson, J.C. and Thompson, E.R. 1984 Laminar Boundary Layer Stability Experiments
on a Cone at Mach 8. Part 2: Blunt Cone, AIAA Paper No. 84-0006.
Stuart, J.T. 1963 Hydrodynamic Stability, in “Laminar Boundary Layers” (L. Rosenhead, ed.), pp. 629–670,
Oxford Univ. Press, Oxford.
Tani, I. 1969 Boundary-Layer Transition, Annu. Rev. Fluid Mech., Vol. 1, pp. 169–196.
Tani, I. 1981 Three-Dimensional Aspects of Boundary-Layer Transition, Proc. Indian Acad. Sci. (Engg. Br.),
Vol. 4, Pt. 2, pp. 219–238.
149
Tatsumi, T. and Gotch, K. 1971 in Instability of Continuous Systems (H. Leipholz, ed.), p. 368.
Taylor, G.I. 1915 Eddy Motion in the Atmosphere, Phil. Trans. Roy. Soc. A, Vol. 215, pp. 1–26.
Taylor, G.I. 1938 Some Recent Developments in the Study of Turbulence, Proceedings of 5th Inter. Congr.
Appl. Mech., Cambridge, MA, pp. 294–310.
Thomas, L.H. 1953 The Stability of Plane Poiseuille Flow, Phys. Rev., Vol. 91, pp. 780–783.
Tollmien, W. 1929 Über die Entstehung der Turbulenz, Nachr. Ges. Wiss. Göttingen, Math.-phys. Klasse,
pp. 21–44.
Tollmien, W. 1935 Ein allgemeines Kriterium der Instabilität laminarer Geschwindigkeitsverteilungen, Nachr.
Ges. Wiss. Göttingen, Math.-phys. Klasse, Vol. 50, pp. 79–114.
Van Ingen, J.L. 1956 A Suggested Semi-Empirical Method for the Calculation of the Boundary Layer Tran-
sition Region, Univ. of Tech., Dept. Aero. Eng., Report VTH-74, Delft, Holland.
Van Stijn, T.L. and Van De Vooren, A.I. 1982 An Accurate Method for Solving the Orr-Sommerfeld Equation,
J. Eng. Math., Vol. 14, pp. 17–26.
Van Stijn, T.L. and Van De Vooren, A.I. 1983 On the Stability of Almost Parallel Boundary-Layer Flows,
Computers and Fluids, Vol. 10, pp. 223–241.
Wasow, W. 1948 The Complex Asymptotic Theory of a Fourth Order Differential Equation of Hydrodynam-
ics, Ann. Math. (2), Vol. 49, pp. 852–871.
Wazzan, A.R., Okamura, T.T. and Smith, A.M.O. 1968a Spatial and Temporal Stability Charts for the
Falkner-Skan Boundary-layer Profiles, Report No. DAC-67086, McDonnell-Douglas Aircraft Co., Long
Beach, CA.
Wazzan, A.R., Okamura, T.T. and Smith, A.M.O. 1968b The Stability of Water Flow over Heated and
Cooled Flat Plates, J. Heat Transfer, Vol. 90, pp. 109–114.
Wazzan, A.R., Taghavi, H. and Keltner, G. 1984 The Effect of Mach Number on the Spatial Stability of
Adiabatic Flat Plate Flow to Oblique Disturbances, Phys. Fluids, Vol. 27, pp. 331–341.
White, F.M. 1974 Viscous Fluid Flow, McGraw-Hill, N.Y.
Wilkinson, S.P. and Malik, M.R. 1983 Stability Experiments in Rotating-Disk Flow, AIAA Paper No. 83-
1760.
Zaat, J.A. 1958 Numerische Beiträge zur Stabilitätstheorie der Grenzschichten, Grenzschichtforschung
Symp., IUTAM, pp. 127–138, Springer, Berlin.
Acknowledgement
This paper represents the results of research carried out at the Jet Propulsion Laboratory, California Institute
of Technology under NASA Contract No. NAS7-918.
Support from the Fluid and Thermal Physics Branch of the Aerospace Research Division, Office of
Aeronautics and Space Technology, NASA, for the preparation of this document is gratefully acknowledged.
150