arXiv:physics/9907041v1 [physics.

flu-dyn] 23 Jul 1999

Drag Redu tion over Dolphin Skin via the

Pondermotive For ing of Vortex Filaments
1
A. G. Lisi

May 4, 1999

1 GarLisi. om
 Abstra t
The skin of Tursiops Trun atus is orrugated with small, quasi-periodi ridges running
ir umferentially about the torso. These ridges extend into the turbulent boundary
layer and ae t the evolution of oherent stru tures. The development and evolution
of oherent stru tures over a surfa e is des ribed by the formation and dynami s of
vortex laments. The dynami s of these laments over a at, non-ridged surfa e is
determined analyti ally, as well as through numeri al simulation, and found to agree
with the observations of oherent stru tures in the turbulent boundary layer. The
al ulation of the linearized dynami s of the vortex lament, su essful for the dy-
nami s of a lament over a at surfa e, is extended and applied to a vortex lament
propagating over a periodi ally ridged surfa e. The surfa e ridges indu e a rapid para-
metri for ing of the vortex lament, and alter the lament dynami s signi antly. A
onsideration of the ontribution of vortex lament indu ed ow to energy transport
indi ates that the behavior of the lament indu ed by the ridges an dire tly redu e
surfa e drag by up to 8%. The size, shape, and distribution of utaneous ridges for
Tursiops Trun atus is found to be optimally ongured to ae t the lament dynam-
i s and redu e surfa e drag for swimming velo ities onsistent with observation.
Contents

1 Introdu tion 7
1.1 Dolphins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Skin pliability and a tive ontrol response . . . . . . . . . . . 7
1.1.2 Cutaneous ridges . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.3 Fluid ow development over dolphin skin . . . . . . . . . . . . 9

2 The boundary layer 11

2.1 The laminar boundary layer . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 The boundary layer TS instability . . . . . . . . . . . . . . . . . . . . 14
2.3 The evolution of boundary layer vorti ity . . . . . . . . . . . . . . . 14
2.3.1 Formation of vortex laments . . . . . . . . . . . . . . . . . . 14
2.3.2 Three-dimensional vortex lament instability . . . . . . . . . . 15
2.4 The turbulent boundary layer . . . . . . . . . . . . . . . . . . . . . . 17
2.4.1 Reynolds de omposition . . . . . . . . . . . . . . . . . . . . . 18
2.4.2 Energy ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.3 Hairpin vortex ontribution to Reynolds stress . . . . . . . . . 21

3 Filament evolution over a at boundary 23

3.1 The vortex lament . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Indu ed ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4 Two dimensional motion . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Hamiltonian formulation . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.6 Self intera tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.7 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.8 Diagonalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.9 Linear instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.10 Numeri al simulation and nonlinear evolution . . . . . . . . . . . . . 34
3.11 Comparison with oherent stru tures in the turbulent boundary layer 36

1
4 Filament evolution over a wavy boundary 37
4.1 The boundary and the vortex lament . . . . . . . . . . . . . . . . . 38
4.2 Orthogonal urvilinear oordinates . . . . . . . . . . . . . . . . . . . 39
4.3 Non- anoni al transformation . . . . . . . . . . . . . . . . . . . . . . 41
4.4 Two dimensional motion . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5 Linearized equations of three dimensional lament motion . . . . . . 44
4.6 The approximate Green fun tion . . . . . . . . . . . . . . . . . . . . 45
4.6.1 Neumann series . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.7 Fourier transform and equations of motion . . . . . . . . . . . . . . . 47
4.8 Floquet analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.9 Numeri al implementation . . . . . . . . . . . . . . . . . . . . . . . . 49
4.10 Stability results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

5 The dolphins' se ret 65

5.1 The ee t of surfa e orrugations . . . . . . . . . . . . . . . . . . . . 65
5.2 Impli ations for turbulent energy transport . . . . . . . . . . . . . . . 66
5.3 Turbulent ow separation . . . . . . . . . . . . . . . . . . . . . . . . 66
5.4 Drag redu tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.5 Appli ation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.5.1 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

2
List of Figures
1.1 A sket h showing the distribution and orientation of utaneous ridges
over the body of a dolphin (From Shoemaker and Ridgway [4℄). . . . 8
1.2 A photo-mi rograph of a ross se tion of dolphin skin, showing the
utaneous ridges and underlying dermal stru ture (From Shoemaker
and Ridgway [4℄). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
ux (y)
2.1 The horizontal velo ity prole, q U
, of the laminar boundary layer.
The verti al s ale is in units of U . The heights of the displa ement
νx

thi kness, δ∗ , and 99% thi kness, δ99 , are shown. . . . . . . . . . . . . 12

2.2 The velo ity eld indu ed by the image vortex, relative to the motion
of a straight vortex lament lo ated at x = 0, y = 1. The image
vortex is lo ated at x = 0, y = −1. The self-indu ed velo ity eld and
ba kground ow eld are not in luded. . . . . . . . . . . . . . . . . . 16
2.3 A plot of the maximally unstable deformation mode of a vortex lament
over a at surfa e. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 The turbulent boundary layer mean velo ity, ux (y), showing the various
regions of the boundary layer. The mean velo ity, along the horizontal
axis, is shown as a fun tion of the height above the surfa e in wall
units, y+ = yy∗ , along the verti al axis. . . . . . . . . . . . . . . . . . . 19
2.5 The Reynolds stress, u′x u′y , as a fun tion of x along the line y = 1,
z = 0 threading a sinusoidal lament of unit ir ulation entered at
y = 1, x = 0 and oriented at 45o. The example lament has amplitude
xa = .75, ya = .75 and a wavenumber of k = .5. The indu ed velo ity
of the image lament is also in luded. . . . . . . . . . . . . . . . . . . 21

3.1 The growth parameter, σ+ , in units of |Γ| y02

, as a fun tion of spanwise
wavenumber, ky0 . The maximum at ky0 ≃ 0.36 is marked. The ore
size was set to be ya0 = 61 to obtain this plot. . . . . . . . . . . . . . . 32

3
3.2 The instability angle, θ+ , as a fun tion of spanwise wavenumber, ky0 .
The maximally unstable wavenumber, ky0 ≃ 0.36, is marked, indi at-
ing the angle of the maximally unstable mode to be θ+ ≃ 43o. The
ore size was set to be ya0 = 61 to obtain this plot. . . . . . . . . . . . 33
3.3 The vortex lament evolving nonlinearly from a small initial Gaussian
perturbation. This gure may be ompared dire tly with Figure 2.3. . 35

4.1 The onformal oordinate mapping of lines of onstant u and v into

the xy plane, with a hosen height to length ratio of Lh = .2. . . . . . 40
4.2 The path of a straight vortex lament at height v0 = 1 propagating
over a wavy boundary of wavelength L = 10 and height to length ratio
h
L
= .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 The growth parameter, σ+ , in units of |Γ| v02
, as a fun tion of spanwise
wavenumber, ky∗ , for a lament over a wavy boundary of length L =
0.61mm, and height to length ratio, Lh = 0.1. The maximum at ky∗ ≃
0.072 is marked. The free stream velo ity is U = 3 ms . The growth
parameter for the at plate is shown for omparison, as a dotted urve. 51
4.4 The instability angle, θ+ , as a fun tion of spanwise wavenumber, ky∗ ,
for a lament over a wavy boundary of length L = 0.61mm, and height
to length ratio, Lh = 0.1. The maximally unstable wavenumber, ky∗ ≃
0.072, is marked, indi ating the angle of the maximally unstable mode,
θ+ ≃ 34o . The instability angle for the at plate is shown as a dotted
urve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.5 The growth parameter, σ+ , in units of |Γ| v02
, as a fun tion of spanwise
wavenumber, ky∗ , for a lament over a wavy boundary of length L =
0.61mm, and height to length ratio, Lh = 0.4. The free stream velo ity
is U = 3 ms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.6 The instability angle, θ+ , as a fun tion of spanwise wavenumber, ky∗ ,
for a lament over a wavy boundary of length L = 0.61mm, and height
to length ratio, Lh = 0.4. . . . . . . . . . . . . . . . . . . . . . . . . . 54
σ+ v02
4.7 A ontour plot of the growth parameter, |Γ|
,
as a fun tion of spanwise
wavenumber, ky∗ , and orrugation wavelength, L, for a lament in
ba kground ow, U = 3 ms , over a wavy boundary of height to length
ratio, Lh = 0.1. The orrugation wavelength for a dolphin, L ≃ .61mm,
is marked by a verti al line. The maximally unstable wavenumber,
k+ y∗ , is plotted over the graph as a fun tion of L. . . . . . . . . . . . 55

4
4.8 A plot of the angle of the maximally unstable mode, θ+ , as a fun tion
of orrugation wavelength, L, for a lament in ba kground ow, U =
3 ms , over a wavy boundary of height to length ratio, Lh = 0.1. The
orrugation wavelength for a dolphin, L ≃ .61mm, is marked by a
verti al line. . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . 56
σ+ v0
4.9 The growth parameter, |Γ| , as a fun tion of spanwise wavenumber,
ky∗ , and ba kground velo ity, U , for a lament at height v0 = 6wu
propagating over a wavy boundary of height to length ratio Lh = 0.2
and length L = .61mm. . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.10 The angle of the maximally unstable mode, θ+ , as a fun tion of ba k-
ground velo ity, U , for a lament at height v0 = 6wu propagating
over a wavy boundary of height to length ratio Lh = 0.2 and length
L = .61mm. . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . 58
σ+ v0
4.11 The growth parameter, |Γ| , as a fun tion of spanwise wavenumber,
ky∗ , and ba kground velo ity, U , for a lament at height v0 = 6wu
propagating over a wavy boundary of height to length ratio Lh = 0.1
and length L = .61mm. . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.12 The angle of the maximally unstable mode, θ+ , as a fun tion of ba k-
ground velo ity, U , for a lament at height v0 = 6wu propagating
over a wavy boundary of height to length ratio Lh = 0.1 and length
L = .61mm. . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . 60
σ+ v0
4.13 The growth parameter, |Γ| , as a fun tion of spanwise wavenumber,
ky∗ , and ba kground velo ity, U , for a lament at height v0 = 6wu
propagating over a wavy boundary of height to length ratio Lh = 0.05
and length L = .61mm. . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.14 The angle of the maximally unstable mode, θ+ , as a fun tion of ba k-
ground velo ity, U , for a lament at height v0 = 6wu propagating
over a wavy boundary of height to length ratio Lh = 0.05 and length
L = .61mm. . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . 62
σ+ v0
4.15 The growth parameter, |Γ| , as a fun tion of spanwise wavenumber,
ky∗ , and ba kground velo ity, U , for a lament at height v0 = 6wu
propagating over a wavy boundary of height to length ratio Lh = 0.025
and length L = .61mm. . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.16 The angle of the maximally unstable mode, θ+ , as a fun tion of ba k-
ground velo ity, U , for a lament at height v0 = 6wu propagating
over a wavy boundary of height to length ratio Lh = 0.025 and length
L = .61mm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

5
5.1 The estimated drag redu tion due to altered lament dynami s, d− ,
plotted as a fun tion of ba kground velo ity, U , for wavy boundaries
of length L = .61mm and height to length ratio Lh = .1, Lh = .06, and
h
L
= .033. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

6
Chapter 1
Introdu tion

1.1 Dolphins
A dolphin gliding through water with gra e and speed presents both a beautiful per-
forman e of nature and a persistent enigma to uid dynami ists. Traditional models of
uid ow applied to a swimming dolphin suggest this mammal is apable of aberrantly
large powers of mus ular exertion. However, studies of dolphin energy expenditure
show that a dolphin has a metabolism similar to that of other mammals [1℄, implying
that these animals, rather then having unusual mus ular apability, may posses some
me hanism that allows them to slip through water with unusually little eort. Sir
John Gray was the rst to hypothesize that dolphins may posses an ability to impede
the development of turbulen e over the skin surfa e and hen e redu e the surfa e
drag of a swimming dolphin [2℄. A method of redu ing boundary layer turbulen e
and surfa e drag would be extremely useful if it ould be implemented e iently, and
the a tualization of su h a method has posed a theoreti al and experimental hallenge
to resear hers.

1.1.1 Skin pliability and a tive ontrol response

Some resear hers have attempted to reprodu e the dolphin's onje tured drag redu -
tion by studying  and attempting to dupli ate  the pliable nature of dolphin skin [3℄.
It was thought that the surfa e drag indu ed by a turbulent boundary layer ould be
redu ed if the surfa e was able to yield to the mi ros opi disturban es produ ed by
turbulent eddies. Unfortunately, this approa h was not signi antly su essful. More
re ently, a large eort has been extended towards exploring the possibility that a tive
ontrol of surfa e shape in response to the for es from the turbulent boundary layer
may be able to redu e turbulent development. The proposal that this sort of dire t
a tive response is implemented by dolphins is ostensibly supported by the existen e

7
Figure 1.1: A sket h showing the distribution and orientation of utaneous ridges
over the body of a dolphin (From Shoemaker and Ridgway [4℄).

of systems of nerves and utaneous mus le that appear apable of a tively ontrol-
ling the skin surfa e. However, detailed measurement of dolphins' skin rea tion to
mi rovibratory stimuli indi ate a minimum response time on the order of 10−2 s [4℄,
whereas the times ale asso iated with the passage of oherent stru tures over the skin
is on the order of 10−4 s [8℄. This indi ates that a dolphin's skin is not able to respond
to the passage of individual oherent stru tures.

1.1.2 Cutaneous ridges

It is only re ently that Shoemaker and Ridgway do umented a feature of odonto ete
skin that had been virtually unknown previous to their work; namely, the existen e
of utaneous ridges overing most of the body surfa e [5℄. These orrugations, with
furrows running ir umferentially around the whales, are reported to be h ≃ 10 −
60µm in height when the animal is at rest, with a separation of L ≃ 0.4 − 1.7mm,
and vary slightly in s ale over the dolphin's body. The ridges are absent from the
dolphin's snout and from mu h of the head  regions asso iated with laminar ow,
as well as from the ippers and tail, (Figure 1.1). The fa t that the orrugations
exist only in regions of steady turbulent ow is a strong indi ation that these ridges
are asso iated with turbulent development. The absen e of the orrugations from the
ippers and tail indi ates that the ridges may not be useful in these regions whi h
are exposed to large velo ity and pressure variations.
Mi ros opi studies of se tions of skin over the dolphin's body show the apillary
and mus ular stru ture to be orrelated with utaneous ridge spa ing. This anatom-
i al analysis is onsistent with the possibility that a dolphin may a tively tune the

8
Figure 1.2: A photo-mi rograph of a ross se tion of dolphin skin, showing the uta-
neous ridges and underlying dermal stru ture (From Shoemaker and Ridgway [4℄).

amplitude of surfa e orrugations to orrespond with swimming velo ity. Although

the measured surfa e ridge height is small for a dolphin at rest, produ ing a ridge
height to length ratio of Lh ≃ 0.025, a photo-mi rograph of a se tion of dolphin skin,
(Figure 1.2), displays a mu h larger ratio, Lh ≃ 0.2. This indi ates that a live dolphin
may be redu ing the ridge height to the smaller value while at rest by using uta-
neous mus le, while the relaxed ridge height ratio, indi ative of the available range
of ridge heights, is mu h larger. This hypothesis is further supported by reports from
trainers at Sea World that the surfa e ridges be ome more pronoun ed with greater
swimming speed [5℄.
The existen e of regular orrugations with tunable amplitude overing the bodies
of these e ient swimmers, in regions asso iated with steady turbulent boundary layer
ow, presents a hallenging new mystery. This hallenge is met in this dissertation,
whi h presents a theoreti al model for and an analysis of the ee t of su h a wavy
boundary on turbulent development.

1.1.3 Fluid ow development over dolphin skin

Allthough utaneous ridges were observed and measured by Shoemaker and Ridgway
in seven spe ies of toothed whales (odonto etes), ranging in body length from two to
eight meters [5℄, no orrelation between body size and orrugation s ale is visible in
the data. It may be possible, through future study, to establish a orrelation between
orrugation s ale and swimming speed. However, detailed information regarding the
swimming speeds of these various spe ies is not urrently available. Hen e, a or-

9
relation between swimming speed and orrugation size is not presently possible to
establish. The best observational data urrently exists for the bottlenosed dolphin
(Tursiops Trun atus ), and this spe ies is taken as representative for this analysis.
Adult Tursiops vary in size from two to four meters, mass approximately 200kg ,
and have been observed to be apable of bursts of swimming speed up to 50 km hr
, though
onstant ruising speed is reported to be roughly U ≃ 10 hr [6℄.
km

Sin e the boundary layer thi kness at the lo ation where the transition to turbu-
len e o urs is small, approximately
2
ν 10−6 ms
δ∗c ≃ 420 = 420 = .15mm (1.1)
U 10 km
hr

at ruising speed, the ir umferential urvature of the dolphin's body is not signi ant.
Thus, a laterally at dolphin may be assumed for the purpose of studying boundary
layer turbulen e. The transition to turbulen e o urs for ow over a at plate at a
al ulated position of
2
ν 10−6 ms
c
x ≃ 6 × 10 = 6 × 104
4
= 2.2cm (1.2)
U 10 km
hr

from the edge of the plate. However, transition to turbulen e in ow over a dolphin
is signi antly delayed by the pressure gradient indu ed by the longitudinal urva-
ture of the dolphin's head, and is observed via biolumines ent ex itation to begin
approximately 40cm from the snout [7℄.
Allthough transition to turbulen e in ow over a dolphin's urved head is delayed
by the favorable pressure gradient, on e initiated, turbulent development and evolu-
tion is assumed to be of the same hara ter as turbulent development over a at plate
with no pressure gradient.

10
Chapter 2
The boundary layer
A onsideration of the ee t of surfa e orrugations on turbulent ow development
starts with understanding the evolution of the boundary layer over a at surfa e.
The ow of water, with density ρ ≃ 103 m3 and vis osity µ ≃ 10
kg −3 kg
ms
, is des ribed
by the Navier-Stokes equation,
−
→ → 
1−→
dt −
→
u = ∂t −
→
u + −
→
u ·∇ −
u = − ∇p + ν∇2 −
→
u (2.1)
ρ
2
in whi h ν = µρ ≃ 10−6 ms is the kinemati vis osity, dt is the adve tive derivative,
and the uid velo ity eld, −
→
u = ux xb + uy yb + uz zb, is restri ted to be in ompressible,
−
→ −
∇ ·→u =0 (2.2)
In ompressibility may be assumed, provided uid velo ities remain mu h less than
the speed of sound in the uid. All velo ities are des ribed with respe t to the rest
frame of the boundary surfa e. Therefore, the skin of a dolphin swimming at speed U
through stationary uid is modeled as a stationary at plate immersed in uid with
ba kground (free stream) velo ity of limy→∞ ux = U in the xb (streamwise) dire tion.
The plate is oriented su h that yb is normal to the plate surfa e, whi h lies in the
plane y = 0, with the leading edge of the plate lo ated at x = 0. The z oordinate
runs along the span of the plate, with arbitrary origin.

2.1 The laminar boundary layer

Vis osity plays the primary role in boundary layer evolution as the irrotational uid
rst en ounters the leading edge of the at plate. The uid at the plate surfa e
is atta hed to the surfa e via mole ular intera tions and must satisfy the no slip
boundary ondition,
−
→u (x > 0, y = 0) = 0 (2.3)

11
 7

5
δ 99

4
y

2
δ
*
1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
u x /U

Figure 2.1: The horizontal velo ity prole, uxU(y) , of the laminar boundary layer. The
q
verti al s ale is in units of νx
U
. The heights of the displa ement thi kness, δ∗ , and
99% thi kness, δ99 , are shown.

The equations of motion, (2.1), for the ow may be solved numeri ally or approx-
imately to obtain the laminar boundary layer prole, ux (x, y), (Figure 2.1). The
prole may be hara terized by several des riptors of boundary layer thi kness. The
99% thi kness, δ99 , is the most ommon des riptor, and is dened arbitrarily as the
height at whi h ux (y = δ99 ) = .99U . Another less arbitrary des riptor, the displa e-
ment thi kness, δ∗ , is dened to be equal to the distan e the wall would have to be
displa ed into the uid su h that the mass ux of uid over the boundary would be
the same as that for uid owing over the wall free slip with the free stream velo ity
U, Z ∞ 
ux
δ∗ = 1− dy
0 U
and may be onsidered the middle of the boundary layer.

12
 The boundary layer uid indu es a shear stress of
∂ux
τ0 = µ
∂y y=0

on the wall in the xb dire tion. This stress produ es the drag for e per unit area on a
body traveling through the uid. The shear stress at the surfa e provides a natural
s ale for velo ities and distan es near the wall for both laminar and turbulent ow.
The fri tion velo ity is dened as
s
τ0
u∗ =
ρ
and the length s ale, one wall unit, is dened as
ν
y∗ = 1wu = (2.4)
u∗
The boundary layer thi kness grows as the uid progresses over the body. For
ow over a at plate, the laminar boundary layer displa ement thi kness grows as
r
νx
δ∗ ≃ 1.72
U
The orresponding boundary layer Reynolds number, Re∗ = Uνδ∗ , in reases with the
boundary layer thi kness until it rea hes a riti al value of Rec∗ ≃ 420, at whi h
point the laminar ow be omes unstable. This o urs at the orresponding riti al
boundary layer thi kness of
ν
δ∗c ≃ 420
U
and, for a at plate, at the position
ν
xc ≃ 6 × 104
U
when the boundary layer be omes unstable to the growth of TS waves. The Reynolds
c
number des ribing the transition point is also often give as Recx = Uνx , orresponding
to the distan e from the plate edge at whi h transition o urs.
Be ause the shear stress, τ0 , de reases as the boundary layer thi kness, δ99 , grows,
the wall units of the laminar boundary layer do not grow as qui kly in x as the
boundary layer thi kness. The wall units for the laminar prole grow as
 1  1
ν νx 4 ν 2
y∗ = ≃ 1.74
u∗ U U
At the transition point these wall units are y∗c ≃ 27 Uν .

13
2.2 The boundary layer TS instability
The linear stability analysis of laminar boundary layer ow over a at plate was rst
al ulated by Tollmien in 1929, S hli hting in 1933, and onrmed experimentally by
S hubauer and Skramstadt in 1947. The stability analysis pro eeds by onsidering
an innitesimal, z independent perturbation, − →
u1 , to the laminar velo ity eld, −
→
u0 , of
the form
−
→u =−→
u0 + −
→
u1 (y)ei(kx−ωt)
and solving the resulting eigenvalue problem for the wavenumber, k , frequen y, ω ,
and perturbing prole, − →
u1 (y). For a laminar velo ity prole, − →
u0 , determined from a
solution of the Navier-Stokes equations, the imaginary omponent of ω rst be omes
positive when the boundary layer has grown to the thi kness δ∗c ≃ 420 Uν , implying
the onset of instability. The frequen y and wavelength of the unstable TS waves at
this onset of instability are found to be ℜ(ω c) ≃ .15 δU∗c and λcx = 2πkc
≃ 17δ∗c . The TS
waves grow very rapidly beyond the transition point, and the boundary layer takes on
a qualitatively dierent form. The new ow, now with large s ale periodi variations
in x, be omes unstable to a three-dimensional instability with periodi variations in
z . The three-dimensional instability, best understood through a study of vorti ity
dynami s, evolves dire tly to turbulent ow.

2.3 The evolution of boundary layer vorti ity

The evolution of the boundary layer may be more deeply understood via a des ription
of the dynami s of vorti ity. The vorti ity eld of the uid is dened as the url of
the velo ity,
−
→ −
→ →
ω = ∇ ×− u
Taking the url of the Navier-Stokes equation, (2.1), and using a ve tor identity
produ es the vorti ity equation for the in ompressible uid,
−
→ →  −
→ →  
dt −
→
ω = ∂t −
→
ω + −
→
u ·∇ −
ω = −
→
ω ·∇ −
u + ν∇2 −
→
ω (2.5)

The laminar uid ow over the at plate is des ribed simply as the reation of on-
entrated vorti ity at the leading edge and its diusion, adve tion, and stret hing
downstream.

2.3.1 Formation of vortex laments

The adve ted vorti ity initially oats the at plate with a smooth layer of vorti ity.
The entire laminar boundary layer may be onsidered a vortex sheet of onstant

14
strength, with sheet thi kness in reasing with x. From this point of view, the TS
instability of the boundary layer is seen as similar to the Kelvin-Helmholtz instability
of a shear layer.
With the onset of the TS instability, the smooth vorti ity of the sheet ondenses
into areas of higher vorti ity separated periodi ally in x. These on entrations of
vorti ity then ontinue via nonlinear evolution to wrap and ondense into spanwise
oriented vortex laments, evolution similar to the nonlinear development of a shear
layer [9℄. This pro ess of nonlinear development into vortex laments is highly inu-
en ed by small irregularities in the initial uid velo ity eld whi h produ e signi ant
variations from perfe tly uniform vortex laments. Although the shape and distri-
bution may deviate signi antly from the ideal, these laments and their evolution
represent the dominant stru ture of the boundary layer after the transition point, and
play the entral role in further boundary layer development.

2.3.2 Three-dimensional vortex lament instability

The initial vortex laments, evolving from the vortex sheet after the TS instability,
are separated by the distan e λcx ≃ 17δ∗c in x, and are lo ated at the initial height y0
above the surfa e. A al ulation of the three dimensional instability will show that
this height must be equal to y0 ≃ 6wu. Sin e λcx ≫ y0 , the ee t of initial neighboring
laments on evolution is small and may be ignored [10℄.
A straight vortex lament in the presen e of a boundary is unstable due to the
velo ity eld indu ed by its intera tion with the surfa e. The lament velo ity eld
is al ulated by onsidering it to arise from an image vortex beneath the at surfa e.
A plot of the image indu ed velo ity eld surrounding the lament, relative to the
motion of the lament, (Figure 2.2), shows the nature and dire tion of the stret hing
ow.
The lament is linearly unstable to deformations in this stret hing plane, oriented
at an angle of approximately 45o to the boundary surfa e. This three dimensional
vortex lament instability is treated in detail in Chapter 3, and the ee ts of a wavy
boundary are onsidered in Chapter 4. The lament over a at boundary is found to
be maximally unstable to deformations of wavelength λcz ≃ 17yo, and is sket hed in
Figure 2.3.
A numeri al investigation onrms that a lament with a small initial perturbation
rapidly deforms into a sinusoid of wavelength λcz in lined at an angle of approximately
43o . The expanding sinusoidal lament then leaves the realm of linear evolution as
the bottom of the lament nears the surfa e and adheres to it, forming streamwise
oriented vortex legs, while the top of the lament ontinues to rise with linear rather
then exponential growth as it adve ts downstream. In this manner a horseshoe or
hairpin vortex is formed. Although the existen e of hairpin vorti es was postulated

15
 2

1.75

1.5

1.25

1
y

0.75

0.5

0.25

0
-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1
x

Figure 2.2: The velo ity eld indu ed by the image vortex, relative to the motion
of a straight vortex lament lo ated at x = 0, y = 1. The image vortex is lo ated
at x = 0, y = −1. The self-indu ed velo ity eld and ba kground ow eld are not
in luded.

16
 U
Γ
o
43
y

x
z y0

Figure 2.3: A plot of the maximally unstable deformation mode of a vortex lament
over a at surfa e.

by Theodorsen in 1952 [18℄, and has been examined in great detail in experiments by
Head and Bandyopadhyay [17℄ and others [19℄, the vortex lament instability me ha-
nism, the Crow instability [11℄, leading to the formation of these hairpin vorti es, has
only now been established [10℄. The linear Crow instability of a vortex lament pair
was initially al ulated, and onrmed numeri ally [12℄, for the ase of two air raft
trailing vorti es. The same analysis applies for the near wall lament, and is des ribed
in Chapter 3, with the se ond of the vortex pair repla ed by the image lament.
Re ent experimental investigations indi ate that autogeneration o urs in the fur-
ther evolution of these vortex laments [13℄. In this pro ess, the legs of the hairpin
vortex pin h together and form another lament, whi h then grows via the same in-
stability, to align itself slightly behind the original. This pro ess repeats and produ es
a long hain of hairpin vorti es in line behind ea h original hairpin. These trains of
tightly pa ked hairpin vorti es, hairpin pa kets, are the dominant stru tures in the
turbulent boundary layer.

2.4 The turbulent boundary layer

The fully turbulent boundary layer over a at plate is a tumultuous olle tion of
hairpin vortex pa kets and lament remnants. Sin e boundary layer evolution is
highly nonlinear and sensitively dependent on initial onditions, any simple model,
su h as the one des ribed with the idealized vortex lament pi ture, an only represent

17
the dominant pro esses of the dynami s and annot apture the ne stru ture of small
s ale motions. Rather, a statisti al analysis of uid ow is employed to understand
the pro esses of energy transport and indu ed surfa e drag. The ee ts of oherent
stru tures, su h as olle tions of hairpin vorti es, may then be onsidered on the basis
of how they ee t the mean properties of the ow.

2.4.1 Reynolds de omposition

The mean velo ity eld in the turbulent boundary layer is found by taking the innite
time average of the velo ity eld
Z
−
→ 1 t0
−
→
u = lim u dt
t0 →∞ t0 0

The Reynolds de omposition is performed by de omposing the variables into the time
−
→
average quantity and the remaining u tuating quantity, u′ ,
−
→ −
→
u =−
→
u + u′

This de omposition allows the steady, time independent, mean ow to be onsidered
separately from the more omplex underlying turbulent ow. The mean equations of
motion are obtained by taking the time average of the in ompressibility ondition,
(2.2), and Navier-Stokes equation, (2.1), to get, in omponent notation,

∂i ui = 0 (2.6)
1  
dt ui = uj ∂j ui = − ∂i p + ν∂j ∂j ui − ∂j u′i u′j (2.7)
ρ
 
The last term, −∂j u′i u′j , the Reynolds stress term, represents the ee t of the
turbulent eddies on the mean ow.
The mean ow of the boundary layer develops a prole similar to the laminar
ow prole. A standard model of the mean velo ity prole for turbulent boundary
layer ow has emerged from many de ades of experimental resear h. The prole,
ux , as a fun tion of y , breaks into several regions. The inner region extends to
approximately 5wu, and is tted by the urve ux (y < 5wu) ≃ u∗ yy∗ , with yy∗ equal
to the non-dimensionalized distan e from the surfa e in wall units as dened by
(2.4). The
 log layer
 begins at approximately 30wu, and is tted by the urve
ux ≃ u∗ .4 ln( y∗ ) + 5 . These urves are mat hed to ea h other in the buer layer
1 y

between 5 and 30wu. The log layer extends out into the outer region, where it
asymptoti ally approa hes the free stream ow, U , (Figure 2.4). The mean boundary
layer thi kness grows linearly with x as δ99 ≃ .4 uU∗ x, faster then the square root
dependen e of the laminar boundary layer.

18
 80
75
70 outer region
65
60
55
50
45
log layer
+ 40
y

35
30
25
20
buffer layer
15
10
5
inner region
0
0 5 10 15 20
u x /u *

Figure 2.4: The turbulent boundary layer mean velo ity, ux (y), showing the various
regions of the boundary layer. The mean velo ity, along the horizontal axis, is shown
as a fun tion of the height above the surfa e in wall units, y+ = yy∗ , along the verti al
axis.

19
 Unlike the laminar ow ase, in whi h the primary energy loss is due to vis osity,
the mean ow of the turbulent boundary layer is inuen ed more strongly by the
energy loss to the underlying turbulent ow due to the Reynolds stress. In regions
far from the boundary the ee ts of vis osity on the mean equations of motion are
negligible, and the vis osity term may be ignored.

2.4.2 Energy ow

The mean energy transport equation is obtained by multiplying (2.7) by ui to get
  !
1 1
dt uiui = ∂i − pui + 2νuj Eij − u′iu′j uj − 2νEij Eij + u′iu′j ∂j ui
2 ρ

in whi h the mean strain rate is Eij = 21 (∂i uj + ∂j ui ). This equation is best under-
stood by onsidering it a des ription of the mean energy balan e of an innitesimal
uid element. The hange in kineti energy of the element, dt 12 ui ui , is equal to
the energy transported a ross its boundary (the divergen e term), plus the loss to
vis osity, −2νEij Eij , plus the loss to turbulen e, u′i u′j ∂j ui .
The most signi ant mean strain in the boundary layer is ∂y ux . Hen e, the dom-
inant energy transport from mean to turbulent motion is due to the term u′x u′y ∂y ux ,
and large energy losses to turbulen e are asso iated with u′x u′y < 0 events in a region
of large mean shear. This may be understood physi ally as the reation of turbulent
ow by the lifting of low speed uid near the boundary layer up and against the
higher speed uid in the upper boundary layer, a low speed streak and turbulent
burst, and by the onverse, a sweep, in whi h high speed uid des ends into the
lower boundary layer [14℄. This intermittent turbulent bursting and sweeping phe-
nomenon is experimentally observed in the turbulent boundary layer, and measured
to a ount for the majority, 60 − 80% [15℄[16℄, of the turbulent energy transfer. These
phenomenon are onsistent with the hairpin vortex model of turbulent stru tures,
and the large energy transfer may be onsidered as owing from the mean shear into
the stret hing of the in lined hairpin vortex laments in the boundary layer.
The energy imparted to the turbulent development of the uid, that is, the growth
of hairpin pa kets, must ultimately arise from the surfa e via surfa e drag. The mean
surfa e stress is τ0 = µ ∂u x
∂y y=0
. The mean velo ity prole, ux (x, y), of turbulent
boundary layer ow is strongly ae ted by the Reynolds stress. The addition of
the Reynolds stress term with u′x u′y < 0 in the mean momentum equation, (2.7),
generates a mu h steeper prole than for laminar ow, ∂y ux ≫ ∂y uxlaminar , and hen e
mu h greater surfa e drag.

20
 0.02

0.01

u’x u’y

-0.01

-0.02
-3 -2 -1 0 1 2 3
x

Figure 2.5: The Reynolds stress, u′x u′y , as a fun tion of x along the line y = 1,
z = 0 threading a sinusoidal lament of unit ir ulation entered at y = 1, x = 0
and oriented at 45o . The example lament has amplitude xa = .75, ya = .75 and a
wavenumber of k = .5. The indu ed velo ity of the image lament is also in luded.

2.4.3 Hairpin vortex ontribution to Reynolds stress

It is important to note that the naturally o urring hairpin vorti es over a at sur-
fa e, in lined at an approximately 45o angle, produ e the largest ontribution to the
Reynolds stress. The legs of su h a vortex, oriented in the ± (xb + yb) dire tion, have
large orresponding uid velo ities in the ± (xb − yb) dire tion, produ ing u′x u′y < 0.
A plot of u′x u′y along a line threading an example vortex, (Figure 2.5), shows this
expli itly. The legs of a hairpin vortex, in lined at an an angle θ, will produ e a
Reynolds stress,
1
u′x u′y ∼ (− sin(θ)) (cos(θ)) = − sin(2θ) (2.8)
2
proportional to sin(2θ).
The velo ity eld around an adve ting hairpin vortex, or vortex pa ket, dire tly
explains the observation of streaks, bursts, and sweeps. The streak and burst events
are des ribed by the upwards and ba kwards ow beneath the vortex lament heads,
while the sweep events are asso iated with the passing of the vortex legs near the
boundary. Furthermore, the well do umented spanwise spa ing of the streaks, λ ≃

21
100wu [17℄, is dire tly explained as the wavelength of the maximally unstable mode
of the vortex lament instability that arises from a lament born at y0 ≃ 6wu, the
top of the high shear inner region of the boundary layer prole.
A oherent model of turbulent motions over a at surfa e is formed by the on-
sideration of the birth and development of hairpin vorti es. The spanwise vorti es
ondense during the TS instability, and rapidly develop into hairpins, via the three
dimensional instability, to onstitute a turbulent boundary layer. Further spanwise
vorti es ondense in the high shear region, y0 ≃ 6wu, via autogeneration, and evolve,
on e again via the three dimensional instability, to form hairpin pa kets and ll the
evolving boundary layer.

22
Chapter 3
Filament evolution over a at
boundary
The dominant stru tures in the turbulent boundary layer, the hairpin vorti es and
hairpin pa kets, develop via the three dimensional instability of on entrated pat hes
of spanwise oriented vorti ity. The vortex lament model of the developing turbulent
boundary layer provides an elegant framework in whi h to al ulate the spe i s of
this instability.
The mean ambient vorti ity at the high shear, inner region of the turbulent bound-
ary layer,
−→ u∗
ω a ≃ −zb∂y ux ≃ −zb
y∗
a umulates into the thin, spanwise laments entered at the lament height, y0 ,
above the wall. The vorti ity of ea h lament is onsidered to ondense from a
square pat h of side y0 entered around the lament enter, produ ing a vortex with
ir ulation
u∗
Γ ≃ − y02
y∗
whi h propagates in the surrounding, lo ally onstant, mean ba kground ow, ub ≃
ux (y0 ) ≃ u∗ yy∗0 , in the xb dire tion.
In the region of the turbulent boundary layer dominated by vorti al stru tures,
above several wall units, the ee ts of vis osity are small ompared to the dynami al
intera tion of the vorti es with the wall and with themselves. The simplifying assump-
tion is thus made that the lament propagates in an invis id ba kground, µ = 0, with
the no slip ondition at the wall relaxed to the free slip ondition, uy (y = 0) = 0, or,
more generally,
b ·−
n →
u |B = 0 (3.1)
where n
b is the normal at the boundary, B .

23
3.1 The vortex lament
Consider a single, nearly straight, innite vortex lament, parallel to the ~z -axis,
lo ated in spa e above the at surfa e at the position
−
→
R (z, t) = (X (z, t) , Y (z, t) , z) (3.2)

This time, t, and spanwise oordinate, z , dependent des ription of the lament posi-
tion is adequate for the linear stability analysis in whi h the lament is not allowed
to loop ba k on itself in the z dire tion.
Ideally, the lament is represented as an innitely on entrated region of vorti ity,

−
→f Z ∞
−
→
−
→ −
→ ′ dR
ω =Γ ′ 3
dz δ ( x − R (z , t)) ′ (3.3)
−∞ dz
The a tual vorti ity distribution is not singular, but is hara terized by a nite vortex
ore size. This must be onsidered in al ulating the self-intera tion of the lament.
However, the above approximation is valid in the ase of small vortex ore sizes, and
the results of the stability al ulation will display only a weak dependen e on ore
size.

3.2 Equations of motion

The motion of the lament is obtained by using (3.3) in the vorti ity equation, (2.5),
and dropping the vis ous term to get
−
→ −
→ −
→ →
→ ∂z R = −
dt R = ∂t R + uz |− u |−
→ (3.4)
R R
−
→ − →
in whi h the velo ity eld, −
→u = uf + ub , whi h in ludes the ba kground ow and the
lament indu ed ow ompatible with the vorti ity distribution, is being evaluated
−
→
at the vortex lament position, R (z, t). Inserting (3.2) into (3.4) gives the equations
of motion for the X(z, t) and Y (z, t) oordinates of the lament,

∂t X = −uz ∂z X + ux
(3.5)
∂t Y = −uz ∂z Y + uy

The ow indu ed by the vortex lament, −

→
u , must be al ulated at ea h moment from
the geometry of the lament and boundary.

24
3.3 Indu ed ow
−
→ −
→ − → − → −→
The lament ompatible velo ity eld, uf , must satisfy ω f = ∇ × uf , with ω f given
by (3.3), as well as satisfying the free-slip boundary onditions, (3.1). Sin e the
uid is onsidered in ompressible, the velo ity may be written as the url of a stream
−
→ − → → −
− → − →
fun tion, −
→u = ∇ × Ψ , with the stream fun tion , Ψ = Ψf + Ψb , onsisting of terms for
the lament indu ed ow and the ba kground ow, and ba kground streamfun tion
−
→
Ψb = ub y zb for the urrent ase. This gives, via a ve tor identity, the relationship
between the vorti ity and stream fun tion,

−
→f −→ − → − →f −
→ −
2 f → −
→ − →f 
ω = ∇ × ∇ × Ψ = −∇ Ψ + ∇ ∇ · Ψ

→ −
− → −
→
The gauge hoi e ∇ · Ψf = 0 may be made without ae ting the ow, uf . Finding the
stream fun tion, and hen e the indu ed velo ity, is then redu ed to the problem of
−
→ −
→
solving the ve tor Poisson's equation, ∇2 Ψf = −ω f , with the appropriate boundary
onditions. This solution may be written as
Z
→
− −
→
Ψfi = d3 x′ Gij (−
→
x , x′ )ωj ( x′ )
Z
−
→ dRj
= Γ dz ′ Gij (−
→
x , R (z ′ )) ′ (3.6)
dz
by using the (dyadi ) Green fun tion satisfying
−
→
∇2 Gij = −δ 3 (−
→
x − x′ )δij (3.7)

and the boundary ondition

ǫijk nj Gkl |B = 0 ∀ {i, l} (3.8)

−
→
equivalent to −
→
n × G = 0.
B
Green's fun tion satisfying (3.7) and (3.8) for a at boundary at y = 0 is
 
1 0 0

GF ((x, y, z), (x′ , y ′ , z ′ )) = √ 1
 0 1 0 

4π (x−x′ )2 +(y−y ′ )2 +(z−z ′ )2
0 0 1 
 (3.9)
1 0 0

− √ 1
 0 −1 0 
4π (x−x′ )2 +(y+y ′ )2 +(z−z ′ )2
0 0 1

The onstru tion of this Green fun tion may be understood via the method of images.
For ea h vorti al element that a ts as a ve tor potential sour e through the rst term

25
in (3.9), there is a mirror image sour e beneath the surfa e that a ts through the
se ond term in (3.9). The minus sign in GFyy arises be ause the mirrored element
points in the opposite dire tion in yb. The image ensures that the boundary onditions,
−→
(3.8), are satised, sin e it an els out the omponents of GF parallel to the surfa e.
The velo ity eld indu ed by the vortex lament and its image, as well as the
ba kground ow, is found by taking the url of the stream fun tion to get
  

  → 
− −
→ → 
− →I 
− 
Z  −
 →x − R (z ′ ) −
→ x − RI (z ′ ) 
−
→ Γ dR dR 
u = ubxb − dz ′
−
→ × ′ − × (3.10)
4π 
 −
→ 3
dz −
→ −
→ 3
dz ′ 


 x − R (z ′ ) x − RI (z ′ ) 

whi h is re ognizeable as the Bio-Savart law for the vorti ity/ urrent sour e, with the
−
→
image vortex position dened as RI (z) = (X(z), −Y (z), z).

3.4 Two dimensional motion

The above equations simplify dramati ally for the ase of a straight lament. In this
ase, the lament position is independent of z , and an be written as
−
→
R2 (t) = (x0 (t) , y0 (t))

The equations of motion, (3.5), be ome simply

∂t x0 = ux
∂t y0 = uy

and the indu ed velo ity may be found by integrating (3.10) or solving Poisson's
equation in two dimensions, and dierentiating, to get
Γ (y−y0 ) Γ (y+y0 )
ux = ub − 2π (x−x0 )2 +(y−y0 )2
+ 2π (x−x0 )2 +(y+y0 )2
Γ (x−x0 ) Γ (x−x0 )
uy = 2π (x−x0 )2 +(y−y0 )2
− 2π (x−x0 )2 +(y+y0 )2

Evaluating this indu ed velo ity at the lament in the equations of motion gives
Γ 1
∂t x0 = ub + 2π 2y0 (3.11)
∂t y0 = 0

whi h implies that a straight lament travels over the at surfa e at a onstant height,
with a onstant velo ity of ub + 2πΓ 1
2y0
in the xb dire tion.

26
3.5 Hamiltonian formulation
It is interesting and very useful that the dynami s of the vortex lament motion,
(3.5), have a Hamiltonian formulation [21℄. Taking the dynami al variables to be
X(z, t) and Y (z, t) and the anoni al, instantaneous, fundamental Poisson bra ket to
be
1
{X(z), Y (z ′ )} = δ(z − z ′ ) (3.12)
Γ
the equations of motion are obtained as
∂t X = {X, H} = Γ1 δH = −uz ∂z X + ux
δY
1 δH (3.13)
∂t Y = {Y, H} = − Γ δX = −uz ∂z Y + uy
in whi h the Hamiltonian is the kineti energy of the lament indu ed ow as a
fun tional of X and Y ,

1 R 3 −
→ 2 1 R 3
−
→f − →f −
→f − →b 
H = 2 d x | u | = 2 d x u · u + 2u · u
 !
1 R 3 → −
− →f 2 → −
− →f  − → − →b 
= 2 d x ∇ ×Ψ +2 ∇ ×Ψ · ∇ ×Ψ
 (3.14)
1 R 3
−→f − →f −
→f − →b 
= 2 d x ω · Ψ + 2ω · Ψ
2 R R
−
→ → ′  ∂Rj
− R −
→ − →b
dR
= Γ2 dzdz ′ ∂R
∂z
i
G ij R (z), R (z ) ∂z ′ + Γ dz dz
· Ψ

Note the ommon abuse of notation by whi h a variational derivative su h as δH

δX(z)
is
dened impli itly su h that
Z Z
δH δH
δH = dz δX(z) + dz δY (z)
δX(z) δY (z)
The existen e of Hamiltonian dynami s for the motion of the lament will fa ilitate
the linear stability analysis as well as aid in the al ulation of the equations of motion
in new variables via anoni al transformations.

3.6 Self intera tion

The equations for the three dimensional motion of the lament, (3.4), are presently ill
dened be ause the indu ed velo ity, (3.10), diverges at the lament. This is remedied
by using a lament with a nite ore size. Rather then make a smoothing orre tion
to the vorti ity distribution, (3.3), with the asso iated ompli ation of dealing with
ore dynami s [22℄[23℄, a uto is imposed on the self-indu tion term for the stream
fun tion [9℄
Z
−
→ dRj
f
Ψi = Γ dz ′ Gij (−
→
x , R (z ′ )) ′
[δ] dz

27
where the [δ] stands for the region z ′ < z − aδ and z ′ > z + aδ , a is the vortex ore
radius, and δ is a numeri al fa tor. The approximate value of δ is determined by
establishing onsisten y with the known self indu ed velo ity of a solid ored vortex
1
ring, and is found to be δ ≃ 12 e 4 .
For small ore sizes this uto method is equivalent to the modi ation of the
Green fun tion to [9℄
 
1 0 0

GF = q 1
3
 0 1 0 

4π (x−x′ )2 +(y−y ′ )2 +(z−z ′ )2 +e− 2 a2 0 0 1
 (3.15)
1 0 0

− √ 1
 0 −1 0 

4π (x−x′ )2 +(y+y ′ )2 +(z−z ′ )2
0 0 1

The use of this approximate, non-singular Green fun tion is superior to the uto
method for al ulations. It is used in (3.14) to form the new, non-singular Hamilto-
nian.
The radius of the vortex ores, a, for the initial spanwise laments must be es-
tablished by dire t experimental observation. An inspe tion of slides of turbulent
boundary layer ow indi ates ore sizes on the order of a ≃ 1wu [17℄. For a lament
lo ated at an initial height of y0 ≃ 6wu this orresponds to a relative ore radius of
a 1
= (3.16)
y0 6
Although this quantity is only roughly estimated, the lament instability of inter-
est, the instability indu ed by the lament intera tion with the wall, is only weakly
dependent on ore size.

3.7 Linearization
The equations of motion may be linearized for small perturbations from the straight
lament solution. An analysis of the lament behavior for small perturbations will
display the nature of the three dimensional instability, in luding its angle and growth
as a fun tion of spanwise wavelength. This linear analysis will later be onrmed and
extended by a fully nonlinear numeri al simulation in Se tion 3.10.
The equations of motion, (3.5), are linearized around the two dimensional lament
motion to get,
X (z, t) = x0 (t) + εxe (z, t)
(3.17)
Y (z, t) = y0 (t) + εye (z, t)

28
where ε is a bookkeeping variable used to keep tra k of the order in the small per-
turbative variables, xe and ye. The analysis of linearized motion is fa ilitated by the
Hamiltonian formulation. The Hamiltonian, (3.14), for the lament over the at
boundary is expanded order by order in ε to obtain

H F = H F (0) + εH F (1) + ε2 H F (2) + O(ε3) (3.18)

in whi h H F (2) is the part of the Hamiltonian quadrati in xe and ye, and H F (1) is zero.
The equations of motion, (3.13), are then similarly expanded order by order in ε. The
zeroth order equations are the two dimensional equations of motion, (3.11). The rst
order equations produ e the linearized equations of motion for the lament,
1 δH F (2)
∂t xe = Γ δe y
F (2) (3.19)
∂t ye = − Γ1 δHδex

whi h depend only on the quadrati part of the Hamiltonian,

Z Z !
Γ2 g
∂R g
α F (0) ∂ Rβ
H F (2)
= dzdz ′
Gαβ + GFzz(2) (3.20)
2 ∂z ∂z ′

−
→
f F (0)
in whi h R = (xe, ye), Greek indi es {α, β} range only over {x, y}, and Gαβ and GFzz(2)
are the modied Green fun tion, (3.15), omponents of zeroth and quadrati order in
ε,
F (0) −
→ −
→
Gαβ = GFαβ ( R (z), R (z ′ ))
∂2 F − → → ε=0
−
GFzz(0) = 12 ∂ε ′
2 Gzz ( R (z), R (z ))
ε=0

3.8 Diagonalization
The linearized equations of motion, (3.19), along with the quadrati Hamiltonian,
(3.20), omprise a set of linear, integro-partial dierential equations for the dynami al
variables xe(z, t) and ye(z, t). This set of equations is redu ed to a set of linear, ordinary
dierential equations by applying a Fourier transform in z to obtain the new omplex
dynami al variables, xe (k, t), dened as

1 Z∞
xe(k, t) = dz e−ikz xe(z, t)
2π −∞
with Z ∞
xe(z, t) = dk eikz xe(k, t)
−∞

29
and ye(k, t) dened similarly. This is a anoni al transformation. The new fundamen-
tal Poisson bra ket be omes
n o
1 R
dz δδexex(k) δe
− δδyeex(z)
y ∗ (k ′ ) ∗ (k ′ ) δe
x(k)
{xe(k), ye∗ (k ′ )} = Γ
1
(z) δe
′
y (z) δe
y (z) (3.21)
= 2πΓ
δ(k − k )
The next step is to show how the quadrati Hamiltonian, (3.20), is diagonalized
by this transformation, and al ulate the resulting equations of motion. Sin e the
Green fun tion, (3.15), within the Hamiltonian is expli itly dependent only on z − z ′
and not on z + z ′ , it is useful to write it as a fun tion of ξ ≡ z − z ′ and perform the
Fourier transform in this variable to obtain
Z
′ ′
F
G (x, y, ξ, x , y ) = dk eikξ GF (x, y, k, x′ , y ′)

in whi h the Fourier transformed Green fun tion is al ulated as

1 R
GF (x, y, k, x′, y ′ ) = 2π
dξ e−ikξ GF (x, y, ξ, x′, y ′)  
q 1 0 0
− 23 2 
1 ′ 2 ′ 2
= (2π)2 K0 (k (x − x ) + (y − y ) + e a )  0 1 0 

0 0 1
q 1 0 0
1
− (2π) K (k (x − x′ )2 + (y + y ′ )2 ) 
 0 −1 0


2 0
0 0 1
with K0 the modied Bessel fun tion. The quadrati Hamiltonian is then written as

Γ2 RRR fα F (0)
∂R f
∂R
H F (2) = 2
dzdz ′ dk eikξ ∂z
Gαβ (X(z), Y (z), k, X(z ′ ), Y (z ′ )) ∂zβ′
o
+GFzz(2) (X(z), Y (z), k, X(z ′ ), Y (z ′ ))

The GF (0) and GF (2) are omputed, the Fourier integrals are substituted for xe(z) and
ye(z), and the resulting quintiple integral vanishes in a poof of delta fun tions to give
the diagonalized Hamiltonian in the Fourier variables,
Γ2 Z
H F (2) = dk k 2 {A(k) xe∗ (k, t) xe (k, t) + B(k) ye∗ (k, t) ye(k, t)} (3.22)
2
with n o
1 1
A = S(ka) + 2 − K0 (2ky0 ) − 2ky K1 (2ky0 )
n (2ky0 )
1 1
o
0
(3.23)
B = S(ka) + − (2ky )2 − 2ky0 K1 (2ky0 )
0

and the vortex lament self intera tion term

3 3
1 1
S(ka) = −  3
2 + K0 (e− 4 ka) + 3 K1 (e− 4 ka)
e− 4 ka
(3.24)
−4
e ka

≃ 0.183 − 12 ln(ka) + O((ka)2 )

30
The linear equations of motion for ea h mode are then easily omputed from (3.22)
and (3.21) to be
1 δH F (2) Γk 2
∂t xe(k, t) = 2πΓ δe
= B(k) ye
y∗
1 δH F (2)
2π
2 (3.25)
∂t ye(k, t) = − 2πΓ δex∗ = − Γk
2π
A(k) xe

3.9 Linear instability

The set of linear ordinary dierential equations of motion for ea h mode, (3.25), are
redu ed to an eigenvalue problem by assuming a solution of the form

xe = xf1 eσt
ye = yf1 eσt

to get " # " #" #

Γk 2
xf1 0 2π
B(k) xf1
σ = Γk 2
yf1 − 2π A(k) 0 yf1
The two resulting eigenvalues are

Γk 2 √
σ± (k) = ± −AB (3.26)
2π

with the angle, up from xb, of the orresponding eigenve tors equal to
√
yf1 Γ −AB
θ± = arctan( ) = arctan(± )
xf1 |Γ| B

The positive eigenvalue, or growth parameter, σ+ , is plotted in Figure 3.1. All

small wavenumber perturbations of the lament are unstable up to ky0 ≃ .53, with
the maximum instability o urring at ky0 ≃ .36. For wavenumbers above .53 the
eigenvalues, (3.26), be ome imaginary, orresponding to lament perturbations that
rotate rather then grow. The unstable eigenmodes orresponding to σ+ grow at an
angle, θ+ , plotted in Figure 3.2. All unstable modes grow in planes with angles of 0o
to 90o up from xb, with the maximally unstable mode growing at an angle of θ+ ≃ 43o .
The wavenumber of the maximally unstable mode is dependent on the hoi e of ore
size, (3.16), for the al ulation. However, the instability is only weakly dependent on
this hoi e. For example, a ore of only one tenth the urrent estimated size of ya0 = 16
produ es a similar instability plot with the maximum at ky0 ≃ 0.28. Also, the angle
of the maximally unstable mode remains θ+ ≃ 43o over several orders of magnitude
of ore size.

31
 0.04

0.03

0.02

0.01

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
ky0

Figure 3.1: The growth parameter, σ+ , in units of |Γ|

y02
, as a fun tion of spanwise
wavenumber, ky0 . The maximum at ky0 ≃ 0.36 is marked. The ore size was set to
be ya0 = 61 to obtain this plot.

32
 90
85
80
75
70
65
60
θ 55
50
45
40
35
30
25
20
15
10
5
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
ky0

Figure 3.2: The instability angle, θ+ , as a fun tion of spanwise wavenumber, ky0 .
The maximally unstable wavenumber, ky0 ≃ 0.36, is marked, indi ating the angle of
the maximally unstable mode to be θ+ ≃ 43o . The ore size was set to be ya0 = 61 to
obtain this plot.

33
 For any small initial perturbation of the straight lament, the high wavenumber
modes rotate about the ore while the low wavenumber modes expand in the unstable
plane, θ+ , and ontra t in the stable plane, θ− . The maximally unstable mode, ky0 ≃
0.36, expands exponentially faster then the others, and rapidly be omes the dominant
deformation of the lament. The lament evolves into a sinusoidal shape in lined at
approximately 43o , with a wavelength of λ = 2π k
≃ 17y0, as plotted in Figure 2.3.
Although the linear analysis gives a pre ise des ription of vortex lament motion for
small perturbations, the full equations of motion, (3.5), are highly nonlinear; and it
is expe ted that nonlinear me hanisms will determine the dynami s after the initial
perturbations grow to signi ant size.

3.10 Numeri al simulation and nonlinear evolution

Numeri al simulation provides an ee tive means of exploring nonlinear behavior,
as well as onrming linear analysis. The fully nonlinear equations of motion for a
vortex lament lend themselves readily to numeri al analysis. However, the equation
for vortex position, (3.2), must be modied to a fully parametri representation to
allow for the possibility of the lament doubling ba k in the zb dire tion,
−
→
R (s, t) = (X(s, t), Y (s, t), Z(s, t))

in whi h s is an arbitrary parameter along the lament. The equation of motion for
the lament is then simply
−
→
∂t R (s, t) = −
→
u |−
→
R
with the velo ity eld given by the Bio-Savart integral, (3.10), with a nite ore size.
The full set of nonlinear integro-partial dierential equations is


 −
→ −
→ 

 R (s,t)− R (s′ ,t)
−
→ Γ R −
→
∂t R (s, t) = ub xb − 4π
ds′   32 × ∂s′ R (s′ , t)

 −
→ −
→ 2


3
R (s,t)− R (s′ ,t) +e− 2 a2
 −
→  
−
→ 
R (s,t)−RI (s′ ,t) →I ′ 
− 
− −
→ −
→ 3 × ∂s ′ R (s , t)

R (s,t)−RI (s′ ,t) 


−
→
with the image position RI (s, t) = (X(s, t), −Y (s, t), Z(s, t)). These equations are
then dis retized in s and integrated forward in time. Care is taken to insert new
dis retization points along the lament as it is stret hed, to preserve the a ura y of
the s heme.

34
 U

x
z y0

Figure 3.3: The vortex lament evolving nonlinearly from a small initial Gaussian
perturbation. This gure may be ompared dire tly with Figure 2.3.

An example of the resulting lament motion is shown in Figure 3.3. For this ase,
the initial lament
 was taken to have a small Gaussian perturbation in the xb dire -

−
→ s 2
tion, R (s, 0) = .1e−( 4 ) , 1, s , to simulate the introdu tion of a small disturban e
to the ow. This Gaussian has a small proje tion onto the maximally unstable mode
of wavelength λ ≃ 17y0 , whi h rapidly grows to dominate the lament geometry, on-
rming the a ura y of the linear analysis. The nonlinear ee ts begin to ontribute
signi antly to the evolution as the legs of the lament approa h the surfa e. The
legs are drawn out in the −xb dire tion, produ ing ounter rotating streamwise vor-
ti es near the wall. And the initially planar sinusoidal urve of the lament be omes
slightly bowed as it expands.
Although this numeri al simulation gives a satisfa tory des ription of lament mo-
tion for short times, it is insu ient to address long time evolution questions su h as
the details of lament autogeneration and multiple lament intera tion. A simulation
of autogeneration along these lines would require more detailed ore dynami s as well
as the use of vortex lament surgery. Alternatively, a three dimensional simulation
of the Eulerian vorti ity eld of an autogenerating hairpin produ es ex ellent results
[13℄. However, the present numeri al model serves as a su essful onrmation of the
linear result, displaying the dominant evolution of θ+ ≃ 43o modes with wavelength
λ ≃ 17y0, and provides a onsistent pi ture of the evolution of single hairpin vortex
features.

35
3.11 Comparison with oherent stru tures in the tur-
bulent boundary layer
The qualitative agreement between the analyti results and the reported experimen-
tal observations of hairpin vorti es leaves little doubt that this lament instability is
responsible for the development of hairpin vorti es in the turbulent boundary layer.
The observed 45o in lination is given a solid mathemati al foundation, and the nu-
meri al simulation of the evolving lament, 3.3, produ es a geometry that onforms
ex ellently with the reports of observers and on eptual models as reviewed by Robin-
son [19℄, and more re ently elu idated by Zhong and others [24℄, and by Delo and
Smits [25℄.
The analyti al ulation of the existen e of the maximally unstable wavenumber,
ky0 ≃ .36, together with the experimental observation of spanwise wavelengths of
100wu for these stru tures in the boundary layer, demands that the average initial
height of the vortex laments be equal to y0 ≃ 6wu. This is reasonably lo ated in a
region of high mean shear in the turbulent boundary layer, and agrees well with the
measured lo ation of maximal turbulent energy produ tion [26℄.
However, the true power of the newly established mathemati al model for vortex
lament dynami s in the boundary layer lies in its potential for des ribing and pre-
di ting new phenomenon. In the next hapter the existant mathemati al framework
will be extended to the ase of a orrugated boundary surfa e, and new predi tions
will be extra ted regarding the subsequent dynami al behavior of the vortex lament.

36
Chapter 4
Filament evolution over a wavy
boundary
Several de ades of intense a ademi study have provided an extensive
knowledge base of the simple anoni al ase. The immediate need is to
learn to utilize this store of information in the ontext of boundary-layer
modeling and ontrol methodologies, with the eventual goal of pra ti al
appli ation to engineering problems involving real-world, non anoni al
boundary layers.  Stephen K. Robinson [19℄

In the last paragraph of his ex ellent review arti le on oherent motions in the turbu-
lent boundary layer, Stephen K. Robinson urged resear hers to build upon the foun-
dation of knowledge in vortex lament behavior and to forge ahead into the un harted
territory of ontrol methodologies. This hapter addresses one su h methodology: the
use of a stati , wavy boundary to ontrol vortex lament evolution.
The study of this ontrol methodology is motivated by the existen e of su h sur-
fa e orrugations on the bodies of dolphins, with the physi al parameters of ba k-
ground ow speed and surfa e ridge geometry taken dire tly from measurements of
these animals. A study of vortex lament dynami s above su h a boundary, using
the tools developed in Chapter 3, elu idates the ee ts on turbulent boundary layer
development and the potential for drag redu tion.
A vortex lament propagating over a wavy surfa e is subje ted to rapid periodi
for ing due to its intera tion with the surfa e. The lament is near the surfa e over
the ridge peaks and far from the surfa e over the ridge troughs. Also, the unstable
manifold, shown in Figure 2.2, os illates in dire tion and magnitude as the surfa e
beneath the lament is in lined to the horizontal. This rapid parametri for ing is
similar to the pondermotive for ing of a harged parti le in an ele tromagneti wave,
and dynami ally similar to the parametri for ing of an inverted pendulum [27℄,[28℄.

37
The pondermotive for ing has a dramati ee t on the instability, as it does in the
ase of the pendulum.

4.1 The boundary and the vortex lament

−
→
The wavy boundary is taken to be an innite surfa e, B , beneath the lament,
uniform in the ~z dire tion, and periodi in the xb dire tion. Hen e,
−
→
B (x, z) = (x, B(x), z)

with B(x) a smooth, periodi fun tion with period L and trough to peak ridge height
h. The surfa e normal is
1
b = √
n ′2
(−B ′ (x)xb + yb)
1+B
and the free slip boundary ondition, n b·− → → = 0, is assumed.
u |−
B
The utaneous ridges of Tursiops Trun atus, (Figure 1.2), are near sinusoidal,
with slightly steeper ridges then troughs. The ridges of a live dolphin at rest are
measured, 25cm behind the blowhole [5℄, to have an average height of h ≃ .015mm
and an average spa ing of L ≃ .61mm, produ ing a height to length ratio of Lh ≃ .025.
It is expe ted that a dolphin will in rease or de rease the ridge height, using utaneous
mus le, to a ommodate hanges in swimming speed, with the ridge height to length
ratio possibly ranging up to Lh ≃ 0.2.
Dolphins swim at a ruising speed of 10 km hr
≃ 3 ms , with a top speed of 50 km
hr
≃
14 ms . For a dolphin swimming at U = 3 ms , the fri tion velo ity at, and slightly
beyond, the transition point is u∗ ≃ 27 1
U ≃ .11 ms . A wall unit at this point is thus
y∗ = u∗ ≃ .009mm, and the vortex laments are assumed to form at a height of
ν

y0 ≃ 6wu ≃ .054mm. The vorti es at this height propagate in a ba kground ow

eld of velo ity
y0 m
ub ≃ u∗ ≃ .66
y∗ s
and have a ir ulation of
!2
u∗ y0 m2
Γ ≃ − y02 = −ν ≃ −3.6 × 10−5
y∗ y∗ s
Although these initial values for the laments are al ulated for a at plate, they will
be taken as the approximate values for laments evolving over a wavy boundary.
The lament position as a fun tion of z and t is, for the ase of a wavy boundary,
still written as
−
→
R (z, t) = (X (z, t) , Y (z, t) , z) (4.1)

38
with the dynami al variables X(z, t) and Y (z, t) satisfying the equations of motion,
(3.5),
∂t X = −uz ∂z X + ux
(4.2)
∂t Y = −uz ∂z Y + uy
However, the indu ed velo ity, as well as the ba kground velo ity eld, diers from
the eld for a at plate, (3.10), be ause of the wavy boundary.

4.2 Orthogonal urvilinear oordinates

The existen e of a wavy boundary signi antly ompli ates the al ulation of the
indu ed velo ity eld over the surfa e. However, the problem is simplied by the
introdu tion of orthogonal urvilinear oordinates that mat h the boundary.
A oordinate transformation is made from x and y to new oordinates, u and v ,
su h that
    
L
x(u, v) = π
arctan tan πu tanh Lπ (v + vB ) + 2Lp
  L  
y(u, v) = L
2π
ln 4 cos 2 πu
cosh 2 π
(v + v B ) (4.3)
  L  L 
2 π
+4 sin2 πu L
sinh L
(v + vB ) − vB
 
in whi h the term with p, equal to the integer part of 12 Lu + 1 , is ne essary to give
a smoothly in reasing x for u > L, and vB is the ridge height parameter, related to
the ridge height, h, by !
L πh
vB = − ln tanh( )
2π 2L
The oordinate lines of onstant u and v , mapped into the xy plane, are shown
in Figure 4.1. This oordinate transformation approximates the wavy boundary of
dolphin skin,
B (x(u, v))|v=0 = y(u, v)|v=0
with peaks slightly sharper then troughs.
The transformation, (4.3), is a onformal transformation derived from the omplex
analyti mapping,
  
iL π
ζ(w = u + iv) = x + iy = ln 2 cos (u + iv + ivB ) + 2Lp − ivB (4.4)
π L
and thus preserves the orthogonality of the oordinates. The great advantage of using
a onformal transformation is the ease of al ulating the two dimensional Lapla e

39
 1

0.8

0.6
y/L
0.4

0.2

0.5 1 1.5 2
x/L

Figure 4.1: The onformal oordinate mapping of lines of onstant u and v into the
xy plane, with a hosen height to length ratio of Lh = .2.

operator, and solutions to the Lapla e equation, in the new oordinates. Be ause the
analyti fun tion, (4.4), satises the Cau hy-Riemann equations,

∂u x = ∂v y
∂v x = −∂u y

the two dimensional Lapla e operator, in the new oordinates, be omes

1  
∇2 = ∂x2 + ∂y2 = ∂ 2
+ ∂ 2
s2 (u, v) u v

in whi h the fun tion s2 (u, v) is the s aling fa tor, or Ja obian, of the onformal
transformation  a measure of the area hange of the new urvilinear oordinate grid
 and is equal to

s2 (u, v) = |ζ ′(w)|2 = (∂u x) (∂v y) − (∂u y) (∂v x)

2 cosh( 2π (v+vB ))
= cosh 2π (v+vL ) +cos 2πu − 1
(L B ) ( L )

for the parti ular oordinate hange, (4.3).

Using the new oordinates, the ba kground potential ow satisfying the free-slip
boundary onditions over the wavy surfa e may be obtained from the simple ve tor
−
→
potential, Ψb (u, v) = ub v zb. This ve tor potential produ es an irrotational ba kground

40
 −
→ − → − → → −
− →
velo ity eld, ub = ∇ × Ψb , whi h satises the in ompressibility ondition, ∇ · ub = 0,
−
→
as well as the free-slip ondition, n → = 0. The lines of onstant v in the new
b · ub −
B
oordinate system orrespond to streamlines of this ba kground ow over the wavy
surfa e.
Although it is possible to ontinue to onsider the motion of the lament in arte-
sian oordinates above the wavy boundary, using the urvilinear oordinates only for
the purpose of velo ity eld omputation, the analysis is greatly simplied by trans-
forming the dynami al variables of the lament, X and Y , into the new oordinate
system as well. This obviates the need to onvert ba k and forth between urvilinear
and artesian oordinates.

4.3 Non- anoni al transformation

A non- anoni al transformation is made from the old dynami al lament variables,
X and Y , to the new variables, U(X, Y ) and V (X, Y ), via the inverse of the oordi-
nate transformation, (4.3). The fundamental Poisson bra ket in the new oordinates
be omes
RR n ′
o
δU (z) δV (z ) δV (z ′ ) δU (z) 1
{U(z), V (z ′ )} = dz ′′ dz ′′′ δX(z ′′ ) δY (z ′′′ ) − δ(z ′′ − z ′′′ )
1 ′
δX(z ′′ ) δY (z ′′′ ) Γ (4.5)
= Γs2 (U,V )
δ(z − z )

with the s ale fa tor, s2 , appearing as a dire t result of the ompression and expansion
of uid as it is arried over the wavy surfa e. The position of the vortex lament in
uvz oordinates is now given by
−
→
R (z, t) = (U (z, t) , V (z, t) , z)uvz (4.6)

and the Hamiltonian, (3.14), in the new oordinates is written as

Γ2 Z Z →
′ ∂Ri W − → ′  ∂Rj
−
Z
dRi b
H W
= dzdz Gij R (z), R (z ) + Γ dz Ψ (4.7)
2 ∂z ∂z ′ dz i
with the indi es now ranging over {u, v, z}, and GW the new Green fun tion for the
wavy boundary.
The boundary onditions for the wavy boundary Green fun tion,

ǫijk nj GW
kl = 0 ∀ {i, l} (4.8)
B

take a parti ularly simple form in the urvilinear oordinates, in whi h n b, with
b =v
(4.8) simplifying to
GWul = GW
zl = 0 ∀ {l} (4.9)
v=0 v=0

41
 −−→
whi h is vb × GW = 0. The wavy Green fun tion must also satisfy Poisson's
v=0
equation,
∇2ik GW → −
3 − →′
(4.10)
kj = −δ ( x − x )δij

with the three dimensional Lapla e operator in urvilinear oordinates taking the
form
−→ − → − →
∇2 = ∇· − ∇ × ∇×
 
1 0 0
1 2 2 2 2  
= s2 (∂u + ∂v + s ∂z )  0 1 0 
0 0 1  
  1 0 0 (4.11)
 
+ s13 (∂u2 s) − 2s (∂u s)2 + (∂v2 s) − 2s (∂v s)2  0 1 0 
 
0 0 0
0 −1 0
 
+ s23 ((∂u s) ∂v − (∂v s) ∂u )  1 0 0 
0 0 0

Although the solution to (4.10) is simple for the ase of two dimensional motion,
be ause the z dependen e is integrated out, the full three dimensional solution is
ompli ated by the position dependent Lapla ian, and an approximation to this so-
lution will be made in Se tion 4.6.

4.4 Two dimensional motion

The goal of this analysis is to obtain a des ription of the behavior of a vortex lament
over a wavy surfa e. The dynami s of a straight lament must be determined, and
then extended via a small perturbation, to predi t the lament behavior. The be-
havior of the straight lament is determined by onsidering the motion in the proper
urvilinear oordinates. A two dimensional Hamiltonian formulation, whi h arises
as the limit of the full three dimensional Hamiltonian formulation, fa ilitates this
analysis.
The position of a straight lament, parallel to the orrugations of the surfa e, is
des ribed by
−−→
R2W (t) = (u0 (t) , v0 (t))uv
The dynami s of the lament are des ribed via Hamiltonian dynami s, with the
fundamental Poisson bra ket, arising from (4.5), equal to
1
{u0 , v0 } =
Γs2 (u 0 , v0 )

42
and the Hamiltonian, from (4.7), equal to

Γ2 W
−→
−−→ −
H2W = 2
G2 R2W , R2W + ΓΨbz
Γ2 1
= 2 2π
ln(2v0 ) + ΓUv0
in whi h the two dimensional Green fun tion for the wavy surfa e,
 ′ q
−−→ −−→
GW
2 R W
2 , R W
2 = − 1
ln (u0 − u′0 )2 + (v0 − v0′ )2
2π
q (4.12)
1
+ 2π ln (u0 − u′0 )2 + (v0 + v0′ )2

is the solution to
 
1 ∂ ∂2 2
∇2 G W
2 = s2 (u
0 ,v0 )
+ ∂v GW
∂u20
2 2
−
→ −→′
0
(4.13)
= −δ 2 (R2 − R2 )
= − s2 (u10 ,v0 ) δ(u0 − u′0 )δ(v0 − v0′ )

and satises the boundary onditions, GW

2 = 0. (Note the appearan e of the s ale
v0 =0
fa tor,1
s2
,
when the delta fun tion in (4.13) is onverted from a distribution over at
to urvilinear oordinates.) The resulting equations of motion for the straight lament
propagating over the wavy boundary are
n o  
1 Γ
∂t u0 = u0 , H2W = s2 (u
ub +
n o 0 ,v0 ) 2π(2v0 )
(4.14)
∂t v0 = v0 , H2W = 0

The lament travels along a streamline of onstant v0 , with u0 (t) steadily in reas-
ing with period T in time, u0 (nT +t) = nL+u0 (t). The rst order, nonlinear ordinary
dierential equation for u0 (t), (4.14), is integrated numeri ally to obtain the lament
−−→
position as a fun tion of time. The straight lament, R2W (t) = (u0 (t) , v0 )uv , moving
along the streamlines of onstant v0 with varying velo ity, (Figure 4.2), is used as the
basis for small three dimensional perturbations about this solution.
It is also possible to integrate the equations of motion, (4.14), analyti ally, rather
then numeri ally, to obtain an equation in losed form for the lament postion, u0 ,
as a fun tion of time. Carrying out the integration gives
1
t =  Γ
 (−u0
ub + 2π(2v0 )
2L      (4.15)
π π
+ π
arctan tan u
L 0
tanh (v0 + vB ) )
(
tanh 2π
L
(v0 +vB ) ) L

whi h may be solved for u0 (t). However, solving this trans endental equation for u0 (t)
is no less di ult then integrating the equations of motion numeri ally, and equation

43
 y U
3

2 Γ
1

5 10 15
-1
h 20 L
25
x
30

Figure 4.2: The path of a straight vortex lament at height v0 = 1 propagating over
a wavy boundary of wavelength L = 10 and height to length ratio Lh = .2.

(4.15) is hen e used only to al ulate the period of lament motion over the surfa e,
 
L 2
T =  −1 +  
Γ 2π
ub + 2π(2v0 )
tanh L
(v0 + vB )

whi h is then used to set up the numeri al integration of (4.14) over one period.

4.5 Linearized equations of three dimensional la-

ment motion
The Hamiltonian, (4.7), with the Poisson bra kets, (4.5), generates the dynami s of
the vortex lament above the orrugated surfa e,
n o
1 δH W
∂t U(z, t) = U, H W = Γs2 (U,V ) δV
n o
1 δH W
(4.16)
∂t V (z, t) = V, H W = − Γs2 (U,V ) δU

These equations of motion are linearized about the straight lament solution, as was
done in Se tion 3.7, with the breakup of U and V into two dimensional and small
three dimensional terms,
U (z, t) = u0 (t) + εue (z, t)
(4.17)
V (z, t) = v0 + εve (z, t)
The Hamiltonian is expanded order by order in ε to obtain
H W = H W (0) + εH W (1) + ε2 H W (2) + O(ε3 ) (4.18)
The equations of motion, (4.16), are then similarly expanded order by order in ε,
with the rst order equations produ ing the linearized equations of motion for the
lament,  2 
∂s2 e W (2)
∂t ue = − ∂tsu2 o ∂s u
e + v + Γs1 2 δHδev
∂u
W (2)
∂v
(4.19)
∂t ve = − Γs1 2 δHδeu

44
in whi h s2 and its derivatives are evaluated at the lament enter, (u0 , v0 ). The
quadrati part of the Hamiltonian is equal to
Z Z !
Γ2 g
∂R g
α W (0) ∂ Rβ
H W (2)
= dzdz ′
Gαβ + GW
zz
(2)
(4.20)
2 ∂z ∂z ′
−
→
f W (0)
in whi h R = (ue, ve), Greek indi es {α, β} range only over {u, v}, and Gαβ and
GW zz
(2)
are the modied Green fun tion omponents of zeroth and quadrati order in
ε,
W (0) −
→ −
→ ′
Gαβ = GW αβ ( R (z), R (z )) ε=0
∂2 W − → −
→ ′
GW
zz
(0)
= 21 ∂ε 2 Gzz ( R (z), R (z ))
ε=0

4.6 The approximate Green fun tion

The al ulation of the linear equations of motion for the vortex lament over the
wavy boundary, (4.19), requires the use of the Green fun tion solution to the singular
Poisson equation, (4.10), satisfying the boundary onditions, (4.9). This Green fun -
tion may not be easily found in losed form be ause of the non-trivial nature of the
position dependent Lapla e operator in the wavy oordinates, (4.11). However, it is
possible to approximate this Green fun tion, and to improve upon this approximation
by employing a Neumann series.
The approximate Green fun tion is required to have the following properties: it
must redu e to the at boundary Green fun tion as Lh → 0, it must integrate to
the orre t two dimensional Green fun tion, (4.12), and it must satisfy the boundary
onditions at the surfa e, (4.9). The at boundary Green fun tion, (3.15), with
urvilinear oordinates as arguments,
GW ≃ GF ((u, v, z), (u′, v ′ , z ′ ))  
1 0 0

= q 1
3
 0 1 0 

4π (u−u′ )2 +(v−v′ )2 +(z−z ′ )2 +e− 2 a2

0 0 1

(4.21)
1 0 0
√ 1  
−  0 −1 0 
4π (u−u′ )2 +(v+v′ )2 +(z−z ′ )2
0 0 1
satises these onditions. It is a good approximation to the exa t Green fun tion for
the wavy boundary when applied in the ase of small amplitude lament perturbations
and small Lh .
Although the approximate Green fun tion, (4.21), is not an exa t solution to the
Poisson equation, it is an approximate solution, in that it is a solution to
1 2F F 1
2
∇ G ij ((u, v, z), (u ′ ′ ′
, v , z )) = − 2
δ(u − u′ )δ(v − v ′ )δ(z − z ′ )δij (4.22)
s s
45
with the at Lapla e operator, ∇2F = ∂u2 + ∂v2 + ∂z2 . Equation (4.22) approximates
the Poisson equation, (4.10), whi h may be written as
 
1
∇2ik GW ′ ′ ′ 2F
kj ((u, v, z), (u , v , z )) = s2 ∇ δik + ∇ik
2W
GW kj
= − s12 δ(u − u′ )δ(v − v ′ )δ(z − z ′ )δij (4.23)
−
→
= −δ 3 (−→
x − x′ )δij

by breaking the Lapla e operator in urvilinear oordinates, (4.11), into a at part,
∇2F , and wavy part,

∇2W = s2 ∇2 − ∇2F 
1 0 0
= (s − 1) ∂z  0 1 0 
2 2

0 0 1  
  1 0 0
  (4.24)
+ 1s (∂u2 s) − 2s (∂u s)2 + (∂v2 s) − 2s (∂v s)2  0 1 0 
 
0 0 0
0 −1 0
 
+ 2s ((∂u s) ∂v − (∂v s) ∂u )  1 0 0 
0 0 0

whi h goes to 0 as h
L
→ 0.

4.6.1 Neumann series

If ne essary, the approximate Green fun tion, (4.21), may be improved by adding
terms from a Neumann series expansion. By rewriting (4.23) as

∇2F GW ′ ′ ′ 2W W
ij = −δ(u − u )δ(v − v )δ(z − z )δij − ∇ik Gkj = −ρij (u, v, z) (4.25)

onsidering ρ to be the sour e for a at Poisson equation, and using the at
Green fun tion, GF , to solve for the GW on the left hand side of (4.25), a Fredholm
equation of the se ond kind for GW is obtained,
RRR
GW
ij = du′′dv ′′ dz ′′ GFik ((u, v, z), (u′′, v ′′ , z ′′ )) ρkj (u′′ , v ′′ , z ′′ )
RRR
= GFij + du′′ dv ′′ dz ′′ GFik ∇2W
kl Glj
W

This integral equation may be solved for GW by re ursive iteration, produ ing a
Neumann series,
Z Z Z
GW F
ij = Gij + du′′ dv ′′ dz ′′ GFik ∇2W F
kl Glj + ... (4.26)

46
whi h will onverge for small ∇2W , and hen e for small Lh .
In pra ti e, the integrals in (4.26) annot be written in losed form, and must
be omputed numeri ally to determine the orre tions to the Green fun tion. This
pro edure is not ne essary to the al ulation of the linear instability of the lament
over a surfa e of small Lh , whi h uses only the simple approximate Green fun tion,
GW ≃ GF ((u, v, z), (u′, v ′ , z ′ )).

4.7 Fourier transform and equations of motion

The diagonalization of (4.20) is arried out via a Fourier transform in the z oordinate,
as in Se tion 3.8, with the small dynami al variables of the perturbation be oming
the Fourier transformed pair, ue(k, t) and ve(k, t). The use of the simple approximate
Green fun tion produ es an approximate quadrati Hamiltonian identi al to (3.22),

Γ2 Z n o
H W (2) ≃ dk k 2 AW (k) ue∗ (k, t) ue (k, t) + B W (k) ve∗ (k, t) ve(k, t)
2
with A and B given, as before, by
n o
1 1
AW = S(ka) + 2 − K0 (2kv0 ) − 2kv K1 (2kv0 )
n (2kv0 ) o
0
1 1
B W = S(ka) + − (2kv )2 − 2kv0 K1 (2kv0 )
0

with the self intera tion term given by (3.24).

The linear equations of motion, (4.19), for ea h mode then be ome, in matrix
form, " # " #" #
2 ∂s2 Γk 2 W
ue 1 − (∂t uo) ∂s − (∂ t u o ) + B u
e
∂t = 2 2
∂u ∂v 2π (4.27)
ve s − Γk
2π
AW
0 v
e

with s2 and its derivatives evaluated at (u0 (t), v0 ). The ee t of the wavy boundary
is apparent in the dieren e of (4.27) from the equations of motion in the presen e
of a at boundary, (3.25). The most important dieren e is the appearan e of time
dependent terms in the matrix oe ients, due to the periodi motion of the straight
vortex lament over the wavy surfa e, u0 (t). This ae ts the lament primarily
through an overall time periodi amplitude stret hing, s2 (u01(t),v0 ) , due to the expansion
and ontra tion of the ow streamlines over the surfa e. The perturbative mode
amplitudes, ue and ve, are also ae ted independently by the hanges in the ow
velo ity eld surrounding the propagating lament  a result of the velo ity eld,
and hen e the uv frame, ro king rapidly ba k and forth as the enter of the lament
travels over the undulations.
Although the method of images is not a urate for a wavy surfa e, it is a good
approximation when the lament height is small ompared to the period, Lh ≪ 1,

47
and provides an alternative view of the ee ts of the wavy surfa e on the lament
instability. As the lament enter travels over the surfa e, following a streamline,
(Figure 4.2), it travels periodi ally loser to and farther from the boundary, with
its image lament drawing orrespondingly nearer and farther. This rapid, time
periodi , hange in the most fundamental parameter, the vortex lament height from
the surfa e, has a signi ant ee t on the evolution of the vortex lament perturbation
amplitudes.

4.8 Floquet analysis

The linear equations of motion, (4.27), governing the evolution of the vortex lament
perturbation for ea h wavenumber, k , ontain oe ients that are periodi in time,
with period T . Floquet's theorem implies that the general solution to these equations
may be written as
" # " # " #
ue(t) f1 (t) g1 (t)
= ae σ1 t
+ be σ2 t
(4.28)
ve(k) f2 (t) g2 (t)

with omplex Floquet exponents, {σ1 , σ2 }, time periodi fun tions, {f1 , f2 , g1 , g2 },
and oe ients , {a, b}, determined by the initial onditions.
An inspe tion of the time reversal symmetries of the oe ients in the equations
of motion provides further information about the solution, allowing the se ond Flo-
quet solution set, {g1 , g2 } and σ2 , to be related to the rst. The symmetries of the
oe ients in (4.27) are
s2 (−t) = s2 (t)
∂t u0 |−t = ∂t u0 |t
∂s2 2
∂u
= − ∂s ∂u
−t t
∂s2 ∂s2
∂v −t
= ∂v t

Hen e, the existen e of a solution with the time dependen e

" # " #
ue σ1 t f1 (t)
=e
ve f2 (t)

implies  through time reversal of the equations of motion  the existen e of the
solution " # " #
ũ −σ1 t f1 (−t)
=e
ṽ −f2 (−t)
Sin e the general solution, (4.28), onsists of only two independent terms, the se ond

48
term must a ommodate this time mirrored solution, and hen e

g1 (t) = f1 (−t)
g2 (t) = −f2 (−t) (4.29)
σ2 = −σ1

It is not surprising that the Floquet exponents are inverses, sin e the motion is due
to Hamiltonian ow and to oordinate frame rotation.
The stability of the lament is now ompletely des ribed by the Floquet solution.
Sin e the exponents are inverses, the perturbation has either neutral stability, with
ℜ(σ1 ) = 0, or a pair of unstable and stable modes, with exponents σ± ≡ ± |σ1 |and
ℑ(σ1 ) = 0. A omplete des ription of the linear behavior of the perturbation to the
vortex lament requires only the al ulation of f1 (t), f2 (t), and σ1 .

4.9 Numeri al implementation

Although analyti methods exist to al ulate the Floquet exponent, σ1 , they are not
pra ti al for the urrent ase be ause of the ompli ated time dependent oe ients
in the equations of motion, (4.27). Rather, a numeri al solution over one period, T ,
of the motion is used to determine the Floquet fun tions, f1 and f2 , and Floquet
exponent, σ1 , as a fun tion of wave number, k , lament height, v0 , ba kground ow,
U , and surfa e geometry, h and L.
Of primary interest are the size of the growth parameter, σ+ , if non-zero, and the
angle, θ = arctan( ff12 ), at whi h the perturbation grows over one period. To determine
these quantities, the equations of motion, (4.27), are numeri ally integrated over one
period of motion from the initial onditions,

ue(t = 0) = 1
ve(t = 0) = 0

to obtain the resulting perturbation amplitudes at the end of one period of motion
over the surfa e, ue(T ) and ve(T ). These quantities are ompared with the general
Floquet solution, (4.28), to obtain the equations,

1 = af1 + bg1
0 = af2 + bg2
(4.30)
ue(T ) = aλ1 f1 + bλ2 g1
ve(T ) = aλ1 f2 + bλ2 g2

in whi h λ1 = eσ1 T and λ2 = eσ2 T are the Floquet multipliers, and the Floquet
fun tions are evaluated at t = 0 (or t = T ). The time symmetry relations, (4.29), are

49
used to eliminate g1 , g2 , and λ2 in (4.30), and the resulting equations are solved for
λ1 and ff21 ,
q
λ1,2 = ue(T ) ± ue(T )2 − 1
f2
f
= √ ev(T )
1 e
u(T )2 −1

The lament is linearly unstable to perturbations if and only if |ue(T )| > 1, with the
growth parameter,  
1 q
2
σ+ = ln |ue(T )| + ue(T ) − 1
T
If ue(T ) > 1, the angle of this unstable mode, after ea h period, T , is
 
ve(T )
θ+ = arctan  q 
ue(T )2 − 1

and if ue(T ) < −1,  

−ve(T )
θ+ = arctan  q 
ue(T )2 − 1
If |ue(T )| < 1, the Floquet exponents are pure imaginary and the lament has neutral
stability, with perturbations at this wavenumber os illating, but not growing, about
the lament enter as it propagates over the surfa e.

4.10 Stability results

The wavy boundary signi antly ae ts the evolution of the vortex lament. As a
representative example, a plot of the growth parameter for a lament propagating
over a boundary with height to length ratio Lh = .1 is shown in Figure 4.3. Previously
stable wavenumbers be ome unstable, and the wavenumber and growth parameter
of the maximally unstable mode in rease. This indi ates that the hairpin vorti es
evolving over a wavy boundary will be more tightly pa ked, having hara teristi
spanwise wavelengths less then 100wu. The wavy boundary also has a signi ant
ee t on the growth angle of the hairpin vorti es. The angle of the unstable modes,
orresponding to the plot in Figure 4.3, is shown in Figure 4.4. The maximally
unstable mode over this wavy boundary now grows at an angle of approximately 34o ,
a signi ant hange to the 43o growth over a at plate.
For very large values of Lh , the strong for ing is su ient to indu e new instability
bands at previously stable wavenumbers, (Figure 4.5). Although this new instability
region has a smaller growth parameter than the primary instability, and will hen e
not signi antly ee t the evolution of a randomly perturbed vortex lament, it is

50
 0.05

0.04

σ
0.03

0.02

0.01

0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
ky
*

Figure 4.3: The growth parameter, σ+ , in units of |Γ|

v02
, as a fun tion of spanwise
wavenumber, ky∗ , for a lament over a wavy boundary of length L = 0.61mm, and
height to length ratio, Lh = 0.1. The maximum at ky∗ ≃ 0.072 is marked. The free
stream velo ity is U = 3 ms . The growth parameter for the at plate is shown for
omparison, as a dotted urve.

51
 90
85
80
75
70
65
60
θ 55
50
45
40
35
30
25
20
15
10
5
0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
ky
*

Figure 4.4: The instability angle, θ+ , as a fun tion of spanwise wavenumber, ky∗ , for
a lament over a wavy boundary of length L = 0.61mm, and height to length ratio,
h
L
= 0.1. The maximally unstable wavenumber, ky∗ ≃ 0.072, is marked, indi ating
the angle of the maximally unstable mode, θ+ ≃ 34o . The instability angle for the
at plate is shown as a dotted urve.

52
 0.13

0.1
σ
0.08

0.05

0.03

0
0 0.1 0.2 0.3 0.4
ky
*

Figure 4.5: The growth parameter, σ+ , in units of |Γ| v02

, as a fun tion of spanwise
wavenumber, ky∗ , for a lament over a wavy boundary of length L = 0.61mm, and
height to length ratio, Lh = 0.4. The free stream velo ity is U = 3 ms .

worth noting that the instability angle of the rst of these new instability bands
has the opposite sign of the previous instability, (Figure 4.6). A lament perturbed
at this wavenumber grows in the previously stable dire tion, between −90o and 0o ,
illustrating the signi ant dynami al ee t of the surfa e ridges.
A ontour plot of the growth parameter for a range of surfa e orrugation wave-
lengths, L, is shown in Figure 4.7, and a plot of the maximal growth angle is shown
in Figure 4.8. If the orrugations are small ompared to the lament height, L ≪ v0 ,
the surfa e is ee tively smooth and does not ae t the lament dynami s. Alter-
natively, if the orrugation length is very large, L ≫ v0 , the parametri os illations
are too slow to ae t the lament dynami s signi antly. However, when the orru-
gation length properly mat hes the lament height, L ≃ 20v0 , the dynami s of the
lament are maximally ae ted, with an in rease in maximally unstable wavenumber
and a de rease in instability angle for the vortex lament. The surfa e orrugation
wavelength of the dolphin, at L ≃ .61mm, is well tuned to the orresponding height
of vortex lament formation for a large range of swimming velo ities.
If the ridge length, L, and initial lament height in wall units, v0 , are taken as xed
properties of the boundary and developing ow, the instability may be al ulated for
a variety of ridge heights, Lh , over a range of ba kground velo ities, U . This gives

53
 90
80
70
60
50
40
30
θ 20
10
0
-10
-20
-30
-40
-50
-60
-70
-80
-90
0 0.1 0.2 0.3 0.4
ky
*

Figure 4.6: The instability angle, θ+ , as a fun tion of spanwise wavenumber, ky∗ , for
a lament over a wavy boundary of length L = 0.61mm, and height to length ratio,
h
L
= 0.4.

54
 σ v2
Figure 4.7: A ontour plot of the growth parameter, +|Γ|0 , as a fun tion of spanwise
wavenumber, ky∗ , and orrugation wavelength, L, for a lament in ba kground ow,
U = 3 ms , over a wavy boundary of height to length ratio, Lh = 0.1. The orrugation
wavelength for a dolphin, L ≃ .61mm, is marked by a verti al line. The maximally
unstable wavenumber, k+ y∗ , is plotted over the graph as a fun tion of L.

55
 45

40

θ+

35

30
0 0.25 0.5 0.75 1 1.25 1.5
L(mm)

Figure 4.8: A plot of the angle of the maximally unstable mode, θ+ , as a fun tion
of orrugation wavelength, L, for a lament in ba kground ow, U = 3 ms , over a
wavy boundary of height to length ratio, Lh = 0.1. The orrugation wavelength for a
dolphin, L ≃ .61mm, is marked by a verti al line.

56
 σ v2
Figure 4.9: The growth parameter, |Γ| + 0
, as a fun tion of spanwise wavenumber,
ky∗ , and ba kground velo ity, U , for a lament at height v0 = 6wu propagating over
a wavy boundary of height to length ratio Lh = 0.2 and length L = .61mm.

an indi ation of the ee t of xed length surfa e ridges for a range of swimming
velo ities. For delphin ridge spa ing of L = .61mm and an initial lament height
of v0 = 6wu, the instability and maximum instability angle is plotted, for ridges
of height Lh = .2, .1, .05, .025, over a large velo ity range, (Figures 4.9-4.16). The
maximum instability angle has a minimum at U ≃ 5 ms  a velo ity that is roughly
independent of ridge height. The maximum instability angle at U ≃ 5 ms ranges from
27o for large ridges, Lh = .2, to 400 for very small ridges, Lh = .025.

57
 45
44
43
42
41
40
39
38
θ+ 37
36
35
34
33
32
31
30
29
28
27
26
25
2 4 6 8 10 12 14
U (m/s)

Figure 4.10: The angle of the maximally unstable mode, θ+ , as a fun tion of ba k-
ground velo ity, U , for a lament at height v0 = 6wu propagating over a wavy
boundary of height to length ratio Lh = 0.2 and length L = .61mm.

58
 σ v2
Figure 4.11: The growth parameter, |Γ|
+ 0
, as a fun tion of spanwise wavenumber, ky∗ ,
and ba kground velo ity, U , for a lament at height v0 = 6wu propagating over a
wavy boundary of height to length ratio Lh = 0.1 and length L = .61mm.

59
 45
44
43
42
41
40
θ+ 39
38
37
36
35
34
33
32
31
30
2 4 6 8 10 12 14
U (m/s)

Figure 4.12: The angle of the maximally unstable mode, θ+ , as a fun tion of ba k-
ground velo ity, U , for a lament at height v0 = 6wu propagating over a wavy
boundary of height to length ratio Lh = 0.1 and length L = .61mm.

60
 σ v2
Figure 4.13: The growth parameter, |Γ|
+ 0
, as a fun tion of spanwise wavenumber, ky∗ ,
and ba kground velo ity, U , for a lament at height v0 = 6wu propagating over a
wavy boundary of height to length ratio Lh = 0.05 and length L = .61mm.

61
 45

44

43

42

θ+ 41

40

39

38

37

36

35
2 4 6 8 10 12 14
U (m/s)

Figure 4.14: The angle of the maximally unstable mode, θ+ , as a fun tion of ba k-
ground velo ity, U , for a lament at height v0 = 6wu propagating over a wavy
boundary of height to length ratio Lh = 0.05 and length L = .61mm.

62
 σ v2
Figure 4.15: The growth parameter, |Γ|
+ 0
, as a fun tion of spanwise wavenumber, ky∗ ,
and ba kground velo ity, U , for a lament at height v0 = 6wu propagating over a
wavy boundary of height to length ratio Lh = 0.025 and length L = .61mm.

63
 45

44

43

42

θ+ 41

40

39

38

37

36

35
2 4 6 8 10 12 14
U (m/s)

Figure 4.16: The angle of the maximally unstable mode, θ+ , as a fun tion of ba k-
ground velo ity, U , for a lament at height v0 = 6wu propagating over a wavy
boundary of height to length ratio Lh = 0.025 and length L = .61mm.

64
Chapter 5
The dolphins' se ret
By altering the dynami s of vortex laments, dolphin skin redu es the rate of energy
transport to turbulen e in the evolving turbulent boundary layer. This results in
a net redu tion of the drag indu ed by skin fri tion. The ee tiveness of this drag
redu tion te hnique is limited by the ountering inuen e of in reased skin fri tion
s
due to the form drag of the surfa e ridges. Above a riti al ridge height, Lh , the
turbulent ow over the ridges separates, forming large eddies behind ea h ridge, and
generates a large drag in rease. The drag redu tion at a given velo ity will be optimal
at ridge heights just below this riti al value.

5.1 The ee t of surfa e orrugations

The pondermotive for ing indu ed by the wavy boundary auses the vortex laments
over the ridges to stret h at a maximally unstable angle, θ+ , that is signi antly
less then the maximally unstable angle for laments over a at boundary. This
dynami al ee t a ts over a range of velo ities entered about an optimum velo ity
determined by the wavelength of the surfa e orrugations. For a dolphin with surfa e
ridges of wavelength L ≃ .61mm, the optimal swimming speed for this dynami
ee t is U ≃ 5 ms . Although most ee tive at this velo ity, surfa e ridges of this
wavelength produ e a signi ant redu tion in the maximally unstable angle for a
range of velo ities between 2 − 15 ms , (Figures 4.9-4.16). This result is in ex ellent
agreement with the reported swimming speed for this spe ies of dolphin.
The boundary layer vortex laments developing over su h ridges grow into hairpin
vorti es in lined at the redu ed angle. Pa kets of these hairpin vorti es, now in lined
at an angle lower than the 43o hara teristi of hairpins over a at surfa e, produ e
a signi antly lower Reynolds stress within the turbulent boundary layer.

65
5.2 Impli ations for turbulent energy transport
The Reynolds stress ontribution from hairpin vorti es in lined at an angle θ+ , (2.8),
is proportional to sin(2θ+ ). Any deviation in hairpin in lination away from 45o pro-
du es a redu tion of Reynolds stress, with a orresponding redu tion in the rate of
turbulent energy transport. Sin e the dominant ontribution to turbulent energy
transport omes from the hairpin vorti es, and the rate of energy transport is di-
re tly proportional to the drag for e on the surfa e, the per entage drag redu tion
due to the altered lament dynami s, d− , an be roughly equated with the per entage
redu tion in Reynolds stress,
sin (2 × 430 ) − sin (2θ+ )
d− ≃ 100
sin (2 × 430 )
The drag redu tion obtained using surfa e ridges is dire tly proportional to the degree
to whi h the maximally unstable angle of the vortex lament may be lowered. This
angle depends upon the swimming velo ity and orrugation wavelength, as well as on
the ridge height, Lh . However, for ridge heights that are very large, the turbulent ow
over the ridges will separate, and form large stationary eddies behind ea h ridge that
ee tively lower the ridge height and signi antly in rease drag.

5.3 Turbulent ow separation

The drag redu tion is limited by the maximal ridge height allowed before turbulent
ow separation is indu ed. A linear model of turbulent boundary layer ow over a
s
sinusoidal surfa e gives an estimate for the riti al ridge height, Lh , at whi h sepa-
ration o urs. This estimate is al ulated, and onrmed by dire t measurement, in
papers by Zilker and Hanratty [29℄,[30℄. For low fri tion velo ities, ReL∗ = u∗νL < 630,
and hen e for low ba kground velo ities, the riti al ridge height is al ulated to be
hs
L
≃ .033. However, the riti al ridge height in reases with velo ity, be ause of the
tenden y of turbulent ow to inhibit separation. For example, at ReL∗ ≃ 6300, the
s
riti al ridge height be omes Lh ≃ .1. Also, be ause the shape of utaneous ridges
over dolphin skin are not exa tly sinusoidal, but have slightly steeper peaks and wider
troughs, the riti al ridge height for separation to o ur over these ridges may vary
from the value for perfe tly sinusoidal ridges. It is likely that the shape and distribu-
tion of the dolphins' utaneous ridges is adapted to maximize the ridge height, and
the dynami al ee t on the vortex laments, without indu ing ow separation. Given
the stru ture of utaneous mus le beneath the ridges, it is also very likely that the
dolphins adjust the height of their utaneous ridges to mat h the swimming velo ity
 keeping the ridge height slightly below the riti al value in order to maximize drag
redu tion.

66
 8

(%) 6

|
4
d

2 4 6 8 10 12 14
U (m/s)

Figure 5.1: The estimated drag redu tion due to altered lament dynami s, d− ,
plotted as a fun tion of ba kground velo ity, U , for wavy boundaries of length L =
.61mm and height to length ratio Lh = .1, Lh = .06, and Lh = .033.

5.4 Drag redu tion

The per ent drag redu tion due to the redu ed in lination angle of the vortex instabil-
ity for a variety of surfa e ridge heights is plotted in Figure 5.1. The drag redu tion at
the optimum swimming velo ity of 5 ms varies from 2% for very small ridges of height
h
L
= .033 to 8% for ridges of height Lh = .1. Sin e it is likely that a dolphin in reases
the ridge height with swimming velo ity, the per ent drag redu tion a hieved is likely
to in rease with swimming velo ity as a dolphin adjusts the ridge height to adapt to
the higher riti al ridge height.

5.5 Appli ation

Man has long looked to the animal world for inspiration in e ient engineering de-
sign. The dis overy and analysis of utaneous ridges in odonto etes presents another
engineering feat of mother nature that may be emulated and put to human use.
The drag redu tion obtained through the use of surfa e orrugations will only be
signi ant if properly mat hed to the sele ted appli ation. The ridge spa ing, L, must

67
be tuned for a spe i velo ity range, and the ridge height must not be so great as
to indu e separated ow between the ridges. Although a dolphin may vary the ridge
height to a ommodate greater speeds, it is unlikely that an arti ial implementation
of variable ridge height would be ost ee tive to manufa ture and employ. Therefore,
utaneous ridges may nd pra ti al appli ation only when surfa es are expe ted to
move through a uid at a steady velo ity.
The optimal ridge spa ing for uid owing over a at plate is al ulated to be
L ≃ 120y∗ . This spa ing depends only on the near boundary stru ture of the turbulent
boundary layer, parti ularly the length of one wall unit, y∗ . The length of a wall unit
over a smooth boundary may be determined experimentally by measuring the mean
surfa e stress, τ0 , by observing the spanwise wavelength of the oherent stru tures
present, λz ≃ 100y∗, or, alternatively, by approximating a wall unit as y∗ ≃ 27 Uν . For
example, an air raft whi h regularly travels at a speed of 150 ms may wish to employ
surfa e orrugations of wavelength
 2


1.5 × 10−5 ms 
L ≃ 120 27 ≃ .3mm
150 ms

The ridge height should be set just below the riti al value, to avoid separation.
A ridge height ratio of Lh ≃ .03 will always be below the riti al value, and produ e a
surfa e drag redu tion of 2%. However, the riti al value of the ridge height should be
determined experimentally, so that the maximal ridge height may be employed and
the greatest drag redu tion a hieved. It is possible that ridge heights of ratio greater
than .1 may be allowed, produ ing an 8% or higher redu tion in surfa e drag.

5.5.1 Experiment
Although several resear hers have investigated the development of a turbulent bound-
ary layer over a wavy wall, through experiment and numeri al simulation [29℄[30℄[31℄[32℄,
none have fo used on the parameter range relevant to the ow of water over dolphin
skin, or given attention to the development and evolution of oherent stru tures.
However, it should be straightforward to extend these investigations to the relevant
parameter range and thus determine the validity of the theoreti al predi tions.
The optimal ridge height for drag redu tion, the riti al height at whi h separation
o urs, should be determined experimentally. It would also be very usefull to deter-
mine if a swimming dolphin does in rease the utaneous ridge height with swimming
velo ity, and to what degree; although it is not lear how su h an experiment would
be arried out. The dolphins have had sixty million years to optimize the redu tion
of drag through the ontrol of vortex lament dynami s, and there is still a great deal
to learn from these fas inating reatures.

68
Bibliography
[1℄ T. Williams, The evolution of ost e ient swimming in marine mammals: limits
to energeti optimization, Phil Trans R So Lond B 354, 193 (1999).

[2℄ J. Gray, Studies in animal lo omotion, J Exp Biol 13, 192 (1936).
[3℄ M. Kramer, Boundary layer ontrol by `arti ial dolphin oating', Nav Engnrs
J O t, 41 (1977).

[4℄ S. Ridgway and D. Carder, Features of dolphin skin with potential hydrodynami
importan e, IEEE Engineering in Medi ine and Biology 12, 83 (1993).

[5℄ P. Shoemaker and S. Ridgway, Cutaneous ridges in odonto etes, Marine Mam-
mal S ien e 7, 66 (1991).

[6℄ Dolphin Resear h In , http://www.dolphinresear h.org.au/

[7℄ J. Rohr, A novel ow visualization te hnique using biolumines ent marine plank-
ton, IEEE Journal of O eani Engineering 20, 147 (1995).

[8℄ H. Abarbanel et al, Nonlinear analysis of high-Reynolds number ows over a

buoyant axisymetri body, Phys Rev E 49, 4003 (1994).

[9℄ P. Saman, Vortex Dynami s, Cambridge University Press (1992).

[10℄ H. Abarbanel et al, Vortex lament stability and boundary layer dynami s,
Phys Rev E 50, 1206 (1994).

[11℄ S. Crow, Stability theory for a pair of trailing vorti es, AIAA Journal 8,2172
(1970).

[12℄ D. Moore, Finite amplitude waves on air raft trailing vorti es, Aeronauti al
Quarterly 23, 307 (1972).

[13℄ J. Zhou, R. Adrian, and S. Bala handar, Autogeneration of near-wall vorti al

stru tures in hannel ow, Physi s of Fluids 8, 288 (1996).

69
[14℄ S. Kline et al, The stru ture of turbulent boundary layers, Journal of Fluid
Me hani s 30, 741 (1967).

[15℄ A. Grass, Stru tural features of turbulent ow over smooth and rough bound-
aries, Journal of Fluid Me hani s 50, 233 (1971).

[16℄ H. Kim, S. Kline, and W. Reynolds, The produ tion of turbulen e near a smooth
wall in a turbulent boundary layer, Journal of Fluid Me hani s 50, 133 (1971).

[17℄ M. Head and P. Banyopadhyay, New aspe ts of turbulent boundary-layer stru -

ture, J Fluid Me h 107, 297 (1981).

[18℄ T. Theodorsen, Me hanisms of turbulen e, Pro 2nd Midwestern Conf. of Fluid
Me h, Ohio State University (1952).

[19℄ S. Robinson, Coherent motions in the turbulent boundary layer, Annu Rev
Fuid Me h 23, 601 (1991).

[20℄ H. Aref and E. Flin hem, Dynami s of a vortex lament in a shear ow, J Fluid
Me h 148, 477 (1984).

[21℄ A. Rouhi and J. Wright, Hamiltonian formulation for the motion of vorti es in
the presen e of a free surfa e for ideal ow, Phys Rev E 48,1850 (1993).

[22℄ R. Klein and O. Knio, Asymptototi vorti ity stru ture and numeri al simula-
tion of slender vortex laments, J Fluid Me h 284, 275 (1995).

[23℄ A. Qi, Numeri al study of wave propagation on vortex laments, Journal of

Computational Physi s 104, 185 (1993).

[24℄ J. Zhong, T. Huang, and R. Adrian, Extra ting 3D vorti es in turbulent uid
ow, IEEE Transa tions on Pattern Analysis and Ma hine intelligen e 20, 193
(1998).

[25℄ C. Delo and A. Smits, Volumetri Visualization of Coherent

Stru ture in a Low Reynolds Number Turbulent Boundary Layer,
http://www.prin eton.edu/gasdyn/Carl_IJFD/DeloSmits.html

[26℄ J. Balint, J. Walla e, and P. Vukoslav evi , The velo ity and vorti ity ve tor
elds of a turbulent boundary layer, J Fluid Me h 228, 53 (1991).

[27℄ M. Levi and W. We kesser, Stabilization of the inverted linearized pendulum

by high frequen y vibrations, SIAM Review 37, 219 (1995).

70
[28℄ J. Bla kburn, H. Smith, and N. Gronbe h-Jensen, Stability and Hopf bifur a-
tions in an inverted pendulum, Am J. Phys 60, 903 (1992).

[29℄ D. Zilker, G. Cook, and T. Hanratty, Inuen e of the amplitude of a solid wavy
wall on a turbulent ow. Part 1. Non-separated ows, J Fluid Me h 82, 29 (1977).

[30℄ D. Zilker and T. Hanratty, Inuen e of the amplitude of a solid wavy wall on a
turbulent ow. Part 2. Separated ows, J Fluid Me h 90,257 (1979).

[31℄ J. Hudson, L. Dykhno, and T. Hanratty, Turbulen e produ tion in ow over a
wavy wall, Experiments in Fluids 20, 257 (1996).

[32℄ V. De Angelis, P. Lombardi, and S. Banerjee, Dire t numeri al simulation of

turbulent ow over a wavy wall, Phys. Fluids 9, 2429 (1997).

71