Very Large Scale Motions in A Turbulent Boundary Layer

J. Fluid Mech. (2011), vol. 673, pp. 80–120.

c Cambridge University Press 2011
doi:10.1017/S002211201000621X
Very-large-scale motions in a turbulent

boundary layer
J A E H W A L E E and H Y U N G J I N S U N G†
Department of Mechanical Engineering, KAIST, 291 Daehak-ro, Yuseong-Gu, Daejeon 305-701,
Republic of Korea
(Received 3 August 2010; revised 26 October 2010; accepted 29 November 2010;

first published online 17 February 2011)
Direct numerical simulation of a turbulent boundary layer was performed to

investigate the spatially coherent structures associated with very-large-scale motions
(VLSMs). The Reynolds number was varied in the range Re θ = 570–2560. The main
simulation was conducted by using a computational box greater than 50δ o in the
streamwise domain, where δo is the boundary layer thickness at the inlet, and inflow
data was obtained from a separate inflow simulation based on Lund’s method.
Inspection of the three-dimensional instantaneous fields showed that groups of hairpin
vortices are coherently arranged in the streamwise direction and that these groups
create significantly elongated low- and high-momentum regions with large amounts of
Reynolds shear stress. Adjacent packet-type structures combine to form the VLSMs;
this formation process is attributed to continuous stretching of the hairpins coupled
with lifting-up and backward curling of the vortices. The growth of the spanwise scale
of the hairpin packets occurs continuously, so it increases rapidly to double that of the
original width of the packets. We employed the modified feature extraction algorithm
developed by Ganapathisubramani, Longmire & Marusic (J. Fluid Mech., vol. 478,
2003, p. 35) to identify the properties of the VLSMs of hairpin vortices. In the log
layer, patches with the length greater than 3δ–4δ account for more than 40 % of all
the patches and these VLSMs contribute approximately 45 % of the total Reynolds
shear stress included in all the patches. The VLSMs have a statistical streamwise
coherence of the order of ∼6δ; the spatial organization and coherence decrease away
https://doi.org/10.1017/S002211201000621X Published online by Cambridge University Press
from the wall, but the spanwise width increases monotonically with the wall-normal
distance. Finally, the application of linear stochastic estimation demonstrated the
presence of packet organization in the form of a train of packets in the log layer.
Key words: boundary layer structure, turbulent boundary layers, turbulence

simulation
1. Introduction
Turbulent boundary layers (TBLs) are observed in many fluid dynamic engineering
applications, so the characteristics of TBLs have been examined in numerous
experimental and numerical studies. In real engineering applications involving
wall-bounded boundary layer flow, such as automobiles, ships, airplanes and
heat exchangers, the fundamental mechanisms of heat and momentum transfer
are controlled by the dynamics of turbulent structures. Thus, understanding the
† Email address for correspondence: hjsung@kaist.ac.kr

Very-large-scale motions in a turbulent boundary layer 81
fundamental nature of large-scale motions will improve modelling and control in
these important applications.
Many experimental and numerical studies have shown that large-scale organized
structures that tend to align in the streamwise direction into coherent groups in
the logarithmic layer play a crucial role in the transport of turbulent shear flow.
These structures are known as hairpin vortex packets and have streamwise extents
of the order of 2δ–3δ, where δ is the thickness of the boundary layer. Theodorsen
(1952) was the first to propose a hairpin model and visualized a vortex filament
oriented spanwise to the mean flow. Head & Bandyopadhyay (1981) used smoke and
inclined light sheets to observe vortex loops, horseshoes and hairpin structures in a
zero-pressure-gradient TBL at high Reynolds numbers and found that the groups of
hairpin structures are inclined away from the wall at a shallow angle of 15◦ –20◦ . Smith
(1984) confirmed the presence of hairpin loops by using videos of H2 bubble patterns
and demonstrated the formation of successive hairpin structures in an organized
manner in the streamwise direction. Smith et al. (1991) subsequently extended this
research to argue that hairpins can actually regenerate from an existing vortex under
the appropriate conditions. In the direct numerical simulation (DNS) study of a
turbulent channel flow at low Reynolds numbers, Zhou et al. (1997) investigated the
evolution of an initial vortical structure qualitatively similar to a hairpin vortex in a
mean turbulent field and found that when the initial structure is sufficiently strong,
multiple hairpins are created both upstream and downstream of this structure. Zhou
et al. (1999) also found clear evidence that non-symmetric vortices have stronger
growth rates than symmetric vortices and thus their presence in the flow is preferred.
Particle image velocimetry (PIV) measurements of a TBL by Adrian, Meinhart &
Tomkins (2000a, hereafter referred to as AMT) provided strong evidence that hairpin
vortices are common in logarithmic layers and in the wake regions of TBLs at
Re θ = 930, 2370 and 6845. They showed that these vortices are aligned in the
streamwise direction, creating larger-scale hairpin packets, and that long, growing
zones of relatively uniform low momentum are present in the outer layer. Furthermore,
they found that the packets that originate close to the wall frequently occur within
larger, faster packets and that these smaller packets are part of the induced interior
flow of older packets of coherent hairpin vortices that originate upstream and over-
run the younger, more recently generated packets. Christensen & Adrian (2001)
carried out linear stochastic estimations (LSEs) to support the observations of AMT.
They demonstrated that the outer layer of wall turbulence in turbulent channel
flows is populated by spatially coherent groups of vortices. Tomkins & Adrian
(2003) investigated the organization of hairpin packets in the horizontal plane and
showed that additional scale growth occurs as the result of the merging of vortex
packets via a vortex reconnection mechanism. Finally, Ganapathisubramani et al.
(2003) demonstrated the significance of packet motions quantitatively: 25 % of the
Reynolds shear stress is associated with hairpin vortices that occupy less than 4.5 %
of the total area. For more information, refer to the following papers that describe
these packet motions extensively (e.g. Ganapathisubramani et al. 2005; Hutchins,
Hambleton, Marusic 2005; Wu & Christensen 2006).
Recent experimental and numerical results have revealed the existence of a new type
of motion (with sizes in the range 10δ–50δ) with a scale larger than the large-scale
motions (LSMs) corresponding to bulges or hairpin packets (Priymak & Miyazaki
1994; Jimenez 1998; Kim & Adrian 1999; Del Alamo et al. 2004; Guala, Hommema &
Adrian 2006; Balakumar & Adrian 2007). Adrian (2007) and Marusic et al. (2010a)
have provided an excellent review of these interesting structures including the LSMs.
82 J. H. Lee and H. J. Sung
Kim & Adrian (1999) examined pre-multiplied spectra for a turbulent pipe flow at
y + = 132 and interpreted their shapes as indicating a bimodal distribution in which
the wavelengths at which the maxima occur represent VLSMs and LSMs. Guala et al.
(2006) and Balakumar & Adrian (2007) investigated the pre-multiplied power spectra
of velocity fluctuations and determined criteria for distinguishing between VLSMs
and LSMs in turbulent pipe flows, channel flows and boundary layers. They found
that the maximum streamwise extent of LSMs is about 3δ, and that the boundary that
distinguishes VLSMs from LSMs and smaller motions is kx δ = 2 due to the crossover
in the co-spectra of u and v. Recently, several DNS studies of turbulent channel
and pipe flows have been conducted. Jimenez (1998) used pre-multiplied spectra to
identify the beginning and end of the kx−1 range and found eddies with streamwise
lengths of the order of 10δ–20δ in the logarithmic layer. Del Alamo et al. (2004) also
performed DNS of relatively long channel flow boxes (8πδ) at fully turbulent Reynolds
numbers and investigated the super-δ-scale motions in the outer layer. They provided
evidence that different behaviours underlie the LSMs and VLSMs observed in the
pre-multiplied spectra. Numerical simulations of pipe flow by Priymak & Miyazaki
(1994) have shown that in the most energetic mode, the structure has a wavelength of
∼12R, where R is the pipe diameter. However, although they performed DNS with
a streamwise domain length of 50R, their numerical studies were conducted at a low
Reynolds number (Reτ = 150) at which the flows were not fully turbulent.
The VLSMs or superstructures observed in turbulent flows are prominent and have
turbulent kinetic energies far greater than those of the LSMs within the logarithmic
layer (Del Alamo et al. 2004; Guala et al. 2006; Balakumar & Adrian 2007). Guala
et al. (2006) and Balakumar & Adrian (2007) showed that VLSMs typically account
for half of the turbulent kinetic energy of the streamwise component, and that these
motions also account for more than half of the Reynolds shear stress in the turbulent
flows of pipes, channels and boundary layers. In addition, they found that large eddies
are universal phenomena in canonical wall flows with smooth walls. The results of
Del Alamo et al. (2004) at Re τ = 934 also indicated that, at one third of the channel
height h, structures with streamwise extents in excess of 20h still contribute to the
Reynolds shear stress.
However, the general form of VLSMs in turbulent flows is not certain due to
the experimental and computational difficulties in the study of larger wavelengths,
such as the limited spatial region (2δ–3δ) of PIV due to the constraints of the field
of view of the camera, the difficulties in producing extensive light sheets and the
need for long computational boxes to include VLSMs that occur at high Reynolds
numbers. To address these problems, Hutchins & Marusic (2007) and Monty et al.
(2007) performed experiments with a spanwise rake of 10 hot wires in the TBL,
pipe and channel flows. By using Taylor’s approximation, which converts temporal
experimental measurements into a spatial domain, they found evidence of very long
meandering features that had typical streamwise extents of greater than 20δ in
the instantaneous flow fields. By comparing typical PIV images with the very long
meandering rake signal, Hutchins & Marusic (2007) showed that the stripes observed
in the log region of the rake data are similar to those observed previously in PIV
data. They conjectured that these motions in turbulent pipe flow can be explained
by what Kim & Adrian (1999) referred to as VLSMs that arise from the coherent
alignment of LSMs associated with hairpin packets.
Although the VLSMs found by using Taylor’s hypothesis confirm that some large-
scale structures are present in turbulent flows, this technique can lead to incorrect
interpretations, especially for large-scale structures in TBLs (Dennis & Nickels 2008).
Furthermore, because these features are rarely contained by the measurement domains
of these experiments using hot wires (due to meandering) and these experiments are
restricted to single velocity component measurements (Marusic & Hutchins 2008), it is
difficult to determine the precise spatial form and properties of VLSMs with vortical
features. A recent turbulent channel DNS study by Del Alamo & Jimenez (2009)
questioned the application of Taylor’s famous hypothesis to experiments examining
temporal and spatial fluctuations in highly sheared regions near walls because this
application assumes that the convection velocity is independent of both frequency
and wavenumber and that it is simply equal to the local mean velocity. One method
for overcoming these limitations without using Taylor’s hypothesis is high-resolution
DNS, which can provide full volumetric three-dimensional data. Although this method
is confined to lower Reynolds numbers than is the case for experiments, it is very
useful particularly in conjunction with other experimental studies at higher Reynolds
numbers.
In the present study, DNS of a TBL at moderate Reynolds numbers were conducted
to investigate the spatially coherent structures associated with VLSMs (Lee & Sung
2011). Simulating spatially developing TBLs is more difficult than simulating turbulent
channel flows, because non-periodic boundary conditions should be imposed in the
streamwise direction to enable realistic inflow boundary conditions to be generated.
In a turbulent channel flow, the flow does not evolve spatially and spectral algorithms
can be applied in the streamwise and spanwise directions, which enable highly
efficient computation. We employed Lund’s method to impose the inflow boundary
condition, and a large computational domain (length > 50δo ) was used to take into
account the highly elongated VLSMs present in the outer layer. A fully three-
dimensional understanding of these flow structures remains elusive, and several issues
are unresolved: What are the typical structures of VLSMs and their properties?
And how are they created in flow fields? In the following section, these questions are
addressed by considering the time-resolved instantaneous three-dimensional flow fields
as well as statistical fields using the two-point correlations and LSE. In particular,
the modified feature extraction algorithm developed by Ganapathisubramani et al.
(2003) was adopted to identify the properties of the VLSMs, such as their frequency
and contribution to the Reynolds shear stress in the flow fields.
2. Numerical method
For an incompressible flow, the non-dimensional governing equations are
∂ui ∂ui uj ∂p 1 ∂ 2 ui
+ =− + , (2.1)
dt ∂xj ∂xi Re ∂xj ∂xj
∂ui
= 0, (2.2)
∂xi
where xi are the Cartesian coordinates and ui are the corresponding velocity
components. All variables were non-dimensionalized by the free-stream velocity (U∞ )
and the momentum thickness at the inlet (θin ), and Re is the Reynolds number. The
governing equations were integrated in time by using the fractional step method with
the implicit velocity decoupling procedure proposed by Kim, Baek & Sung (2002).
On the basis of a block LU decomposition, both the velocity-pressure decoupling and
the additional decoupling of intermediate velocity components were achieved through
approximate factorization. In this approach, the terms are first discretized in time
+
Re θ Lx /θin Ly /θin Lz /θin Nx , Ny , Nz x + z+ ymin tU∞ /θin
Inflow simulation 570–1600 1000 100 100 2049, 300, 513 13.7 5.5 0.07 0.1
Main simulation 1410–2560 400 40 40 2049, 300, 513 12.3 4.9 0.06 0.1
Table 1. Domain size and mesh resolution.
U∞ Ly
Turbulent boundary layer
y
x
z Lz
Lx
Figure 1. Schematic diagram of the computational domain.
by using the Crank–Nicholson method, and then the coupled velocity components
are solved without iteration. All terms were resolved by using a second-order central
difference scheme in space with a staggered mesh.
Figure 1 shows a schematic diagram of the computational domain used in the
present study. The notational convention is that x, y and z denote the streamwise, wall-
normal and spanwise coordinates, respectively, and u, v and w denote the streamwise,
wall-normal and spanwise velocity components, respectively. The no-slip boundary
condition was imposed at the solid wall, and the boundary conditions on the top
surface of the computational domain were u = U∞ and ∂v/∂y = ∂w/∂y = 0. Periodic
boundary conditions were applied in the spanwise direction. Since the boundary
layer is developing spatially in the downstream direction, it is necessary to use non-
periodic boundary conditions in the streamwise direction. To overcome the difficulties
associated with these boundary conditions, and to avoid the simulation of the laminar
and transitional regions arising near a leading edge, an auxiliary simulation based on
the method of Lund, Wu & Squires (1998) was carried out to acquire time-dependent
turbulent inflow data in the range Re θ = 570–1600 with a long streamwise domain of
x/θin,i = 1000 (=125δo,i ) at the inlet, where θin,i and δo,i are the inlet momentum and
boundary thickness, respectively; the convective boundary condition at the exit was
specified as (∂u/∂t) + c(∂u/∂x) = 0, where c is the local bulk velocity.
The domain size and mesh resolution are summarized in table 1. Note that the
computational box is normalized by the inlet momentum thickness of each simulation.
The ratio of the inlet momentum thickness of the inflow simulation (θin,i ) to that of
the main simulation (θin ) is θin /θin,i = 2.5 and the corresponding Reynolds number
Re z = U∞ Lz /ν is approximately 57 000 in the spanwise direction, which indicates that
the domain sizes are equal in the spanwise direction. The height and width were
chosen to be at least three times 99 % of the largest boundary layer thickness. The
flow at the recycling station that is x/θin,i = 70 (≈8.75δo,i ) downstream from the inlet
was rescaled to that at the inlet station to satisfy the inlet boundary conditions. The
inflow conditions for the main simulation were extracted at the location x/θin,i =
850 (≈106δo,i ) downstream from the inlet. To ascertain the reliability and accuracy
of this inflow numerical simulation, the mean turbulence statistics and the two-point
(a) (b)
3
25 u+rms
U+
Present study (Reθ = 900)
20 2 w+rms
Wu & Moin (2009)
15 Spalart (1988) 1
10
0 v+rms uv+
5
–1
0
10–1 100 101 102 103 0 20 40 60 80
y+ y+
Figure 2. Inflow profiles at Re θ = 670 and 900. (a) Streamwise mean velocity and (b) r.m.s.
of the velocity fluctuations.
(a) 1.0 (b) 1.0
0.5 0.5
Ruu
0 0
–10 –5 0 5 10 –1 0 1
rx/δ rz/δ
Figure 3. (a) Streamwise and (b) spanwise two-point correlations of the streamwise velocity
fluctuations at xref /θin,i = 850 with a wall-normal height of yref /δ = 0.15. The results of
Hutchins & Marusic (2007) are shown by circles for comparison.
correlation of the streamwise velocity fluctuations were compared with the DNS data
of Spalart (1988) and Wu & Moin (2009) and the experimental data of Hutchins &
Marusic (2007), as shown in figures 2 and 3; xref and yref are the reference positions
for the calculations in the streamwise and wall-normal directions, respectively. The
present results for the mean velocity and turbulence stresses are in good agreement
with those of the previous DNS studies. In addition, the one-dimensional two-point
correlation at the location x/θin,i = 850 shows that transitional phenomena are not
present and the phase information was correctly established in the inflow simulation.
The simulation discussed above was used to conduct the main simulation of the
TBL with Re θ = 1410–2560. The computational domain and grid resolution for this
simulation are presented in table 1. The grid distributions in the wall-normal and
spanwise directions are the same as those for the inflow simulation. In order to include
larger-scale coherent groups, a large numerical domain with sufficient grid resolution
was used: specifically, the computational box size was 400 × 40 × 40 based on the
inlet momentum thickness (θin ) of the main simulation, and the total number of grid
points was 315 million. The velocity field was initialized with a mean velocity profile
given by Spalding’s law and random fluctuations were superimposed with a maximum
(a) 28
24
(uτ/U∞)–1
20
16
(b) 1.8
Present study
Spalart (1988)
1.6 Schlatter et al. (2009)
H
1.4
500 1000 1500 2000 2500

Reθ
Figure 4. (a) Skin friction and (b) shape factor as functions of the Reynolds number.
amplitude of 20 % of the free-stream velocity. The simulation was run initially for
4000 θin /U∞ in order to eliminate transient processes, and the statistics were sampled
during the last 80 000 (8000 θin /U∞ ) steps with 64 parallel processors (IBM p595) at
the KISTI Supercomputer Center. Details of the numerical method can be found in
Lee & Sung (2011).
3. Mean properties
Since the TBL is developing spatially in the streamwise direction, the characteristics
of the boundary layer change as the flow moves downstream. The mean properties
of the skin friction uτ /U∞ = (Cf /2)1/2 and the shape factor H = δ ∗ /θ, where δ ∗ and
θ are√the displacement and the momentum thickness respectively in the TBL and
uτ = τw /ρ, are shown in figure 4 as functions of the Reynolds number up to
Re θ = 2560. These quantities are essential to establishing quantitative accuracy and
determine how well the full domain simulation accounts for large-scale turbulent
structures. The numerical data of Schlatter et al. (2009) are in excellent agreement
with the results of the present simulation, which indicates that the present TBL can be
assumed to be in equilibrium for an extended range of Reynolds numbers. However,
in the results of Spalart (1988), the skin friction is overpredicted at Re θ = 1410 and
the shape factor is slightly lower when compared with those of the present DNS data.
These discrepancies might be caused by low numerical accuracy at fixed Reynolds
numbers, as discussed by Spalart, Coleman & Johnstone (2008). In the initial part of
the curve in figure 4, the quantities sharply increase and then settle into agreement
with the numerical data. This finding agrees with that of Simens et al. (2009), who
found that the effects of the numerical inflow condition based on Lund’s method,
which are probably comparable with the effects of experimental tripping, induce a
Re θ Re δ θ/θin δ/θin uτ /U∞ y ∗ /θin H /θin

1410 10317 2.473 18.1 0.0445 0.058 1.45
2000 14363 3.509 25.2 0.0426 0.056 1.43
2500 17784 4.386 31.2 0.0412 0.054 1.41
Table 2. Boundary-layer flow parameters.
(a) 25 (b) 25
20 20 Present study (Reθ = 2000)
(U∞–U)/uτ
U+ = 1/0.41 ln y+ +5.0 Present study (Reθ = 1410)
15 15 Spalart (1988)
U+ Khujadze & Oberlack(2004)
10 10 Schlatter et al. (2009)
Degraaff & Eaton (2000)
U+= y+
5 5
0
10–1 100 101 102 103 0 0.2 0.4 0.6 0.8 1.0
y+ y/δ
Figure 5. (a) Mean velocity profiles and (b) mean velocity-defect profiles normalized by the
friction velocity.
divergence of the skin friction in the initial computational domain that then evolves
to equilibrium.
Figure 5 shows the variations for the boundary layer of the mean velocity with
the distance from the wall normalized by the friction velocity. The boundary-layer
flow parameters are listed in table 2. In figure 5(a), the mean velocity profiles are
displayed for Re θ = 1410, 2000 and 2500 in the inner coordinates. The profiles have
the expected logarithmic form given by
1
U+ = ln(y + ) + B, (3.1)
κ
where U + = U/uτ , y + = yuτ /ν, κ = 0.41 and B = 5.0 and the capital U denotes
the time- and spatially averaged quantity. The comparison of the numerical and
experimental data shows that there is good agreement in the wall-normal direction.
In general, experimental studies derive the friction velocity from the best fit to the log
law, so well-fitted log law profiles are always obtained. In the present study, however,
the mean velocity profile was obtained after the friction velocity was derived from the
mean velocity very near the wall, resulting in the good agreement with the log-law
profile shown in figure 5(a). The log law is an effective universal curve for the mean
velocity profile in the inner regions of TBLs. As shown in the inset in figure 5(a),
however, there is a slight downward shift of about 0.3 in U + between the present
simulation and the results of Spalart (1988) at Re θ = 1410, owing to low numerical
accuracy (Spalart et al. 2008). Figure 5(b) shows the mean velocity defect profiles in
the outer coordinates. In the outer region of the boundary layer, the flow is assumed
to be independent of viscosity and depends on the global properties of the flow, such
as δ and U∞ . The profiles for these three Reynolds numbers are in good agreement
with those of Degraaff & Eaton (2000) at Re θ = 2900 throughout the outer region of
the boundary layer despite the different Reynolds numbers. This result suggests that
the flow in the outer layer can be predicted by using the outer scaling variables.
(a) 9 (b)
6 1
u+2
v+2
3
0 0
(c) (d) 1
2
–uv+
w+2
0 0
100 101 102 103 100 101 102 103
y+ y+
Figure 6. (a–d ) Reynolds stresses normalized by the friction velocity in the inner
coordinates. The symbols in this figure are the same as those in figure 5.
The Reynolds stresses normalized by the friction velocity obtained from the DNS
and experimental results are shown in figure 6. The calculated fluctuations are in
good qualitative agreement with the numerical and experimental data. In particular,
the present curves are in good quantitative agreement with the results of the DNS
of Schlatter et al. (2009) at a very similar Reynolds number (Reθ = 2500), which
indicates that the effects of the numerical schemes and inflow generation method
used in the present study are negligible. However, the peak value of Khujadze &
Oberlack (2004) at Re θ = 2240 is slightly lower than that of the present simulation.
They also simulated the TBL by using the fringe method, as did Schlatter et al.
(2009), but their computational box 450δ ∗ |x = 0 was much shorter and might not have
been sufficient for the full resolution of the turbulent motions, in contrast to those
used in the present study (1500δ ∗ |x = 0 ) and by Schlatter et al. (2009) (3000δ ∗ |x = 0 ). The
experimental results of Degraaff & Eaton (2000) are slightly higher than those of the
present simulation owing to the Reynolds number difference. However, the similarity
is satisfactory in the near-wall region when scaled by the friction velocity. For the
streamwise component in the range y + > ∼100, large departures are observed with
a tendency for the intensity to form a second maximum in the outer region, which
becomes more pronounced as Reθ increases (Marusic, Mathis & Hutchins 2010b).
4. Instantaneous analysis
Figure 7 shows visualizations of the very long outer structures in the instantaneous
flow fields for y/δ = 0.19 in the xz plane. The direction of flow is left to right, and the u
fluctuations are normalized by the friction velocity. Since the boundary layer thickness
(δ) increases in the streamwise direction, the representative outer scaling variable (δ)
in each figure was chosen at the relative location rx = 0, resulting in decreases in the
wall-normal heights (y/δ) at which the instantaneous and statistical flow fields are
calculated. However, because the variation of y/δ is small in the streamwise region
u+
–5 –3 –1 1 3 5
1 (a) t+ = 0
rz/δ
0
–1
1 (b) t+ = 0
rz/δ
0
–1
1 (c) t+ = 24.7
rz/δ
0
–1
1 (d) t+ = 49.3
rz/δ
0
–1
–10 –5 0 5 10
rx/δ
Figure 7. (a) Streamwise velocity fluctuations normalized by the friction velocity and (b–d)
the time evolution of the Gaussian-filtered streamwise velocity fluctuations at y/δ = 0.19. The
solid lines are the contours of constant streamwise velocity fluctuations u+ = −0.5 and the
inset in (a) is magnified in figures 12–16.
of interest and does not deviate in the corresponding log and wake regions, this
effect was neglected in the present study. Figure 7(a) shows some of the characteristic
emergent features in the log region. Several very long negative (dark colour) motions
meandering in the downstream direction are apparent. These motions are flanked
by positive u fluctuations (white colour) in the spanwise direction and often extend
more than 10δ in length in the streamwise direction, with characteristic widths of
approximately 0.1δ–0.2δ in the spanwise direction. These patterns are similar to the
‘superstructures’ observed in turbulent channel DNS studies with large numerical
domains, as described by Hutchins & Marusic (2007).
To highlight the characteristics of the very long meandering motions, we applied a
simple Gaussian filter to the instantaneous flow fields. The Gaussian filter is useful
for averaging small-scale features at u+ = 0.5 (filter size = δ/2 × δ/2). Figure 7(b–d)
shows the time evolutions of the filtered streamwise velocity fluctuations; the positive
and negative fluctuations are depicted as white and dark contours, respectively. Note
that figures 7(a) and 7(b) are for the same time. The negative coherent regions along
the centreline clearly extend in the streamwise direction beyond the full field of
view, and this motion exhibits characteristic widths of 0.25δ–0.5δ in the spanwise
Vortex head
0.4 Q4 event
Shear layer
y/δ Q2 event
0.2
0
–0.8 –0.4 0 0.4 0.8
rx/δ
Figure 8. Hairpin vortex signatures in the xy plane. The Reynolds number is approximately
Re θ 2400. The instantaneous velocity vectors are displayed for a frame of reference moving
at Uc = 0.74U∞ and are scaled with outer variables. The vortex heads and inclined shear layers
are indicated schematically.
direction. The time evolution of the instantaneous fields shows that the VLSMs travel
downstream with a small velocity dispersion (less than 7 %) and form coherent flow
structures in the outer layer. Note that a sufficiently long life to catch our eye is
a major prerequisite for a coherent structure, and to be long-lived, the dispersion
must be small (Adrian 2007). These averaged characteristics are in good agreement
with previous experimental observations (Kim & Adrian 1999; Balakumar & Adrian
2007; Hutchins & Marusic 2007). In the present study, in order to determine the
characteristics of the VLSMs, the representative motion shown in the inset in
figure 7(a), which contains a flow feature (indicated by a solid line) with a streamwise
extent of ∼8δ, was inspected throughout the instantaneous and statistical flow fields.
4.1. Hairpin vortex signatures
Galilean decomposition has been used to investigate the flow fields associated with
packets by several researchers (AMT; Tomkins & Adrian 2003; Wu & Christensen
2006). If the convection velocity is subtracted from the flow field, the velocity vectors
with a streamwise component similar to the selected convection velocity have a small
magnitude, so the vortex patterns moving with a velocity equal to the subtracted
convection velocity become highlighted as light regions. Before proceeding further,
therefore, it is necessary to investigate the flow patterns of the hairpin vortex signatures
(HVS) and the convection velocity in the TBL. Figures 8 and 9 show the HVS in
the xy and xz planes. The vortex components induced by hairpin heads and legs
are circled. These HVS were originally suggested by the experimental studies of
AMT and Tomkins & Adrian (2003). AMT defined the HVS as a velocity vector
pattern with a circular vortex and an inclined region of Q2/Q4 vectors beneath it,
which arises when a hairpin vortex cuts through the xy plane of a hairpin vortex
centre. In addition, Tomkins & Adrian (2003) described an idealized HVS in the xz
plane as having three characteristics: (i) two elliptical counter-rotating vortex patterns
created by the intersection of the cut plane with the angled vortex legs, (ii) a low
streamwise velocity fluctuation momentum event created by the backward induction
of the legs and (iii) a stagnation point created at the interface between the induced
low-momentum event and the faster upstream fluid. These signatures in the xy and xz
planes appear frequently in groups in the streamwise direction; thus, the stagnation
point might not be present because the upstream vortex prevents the impingement of
the high-speed fluid onto the low-speed fluid. In reality, hairpins can have one-sided
(a) (b)
Vortex legs 0.4 Low-momentum region
0.2
Stagnation point
rz/δ
0 0
–0.2
–0.4
–0.4 0 0.4 –0.8 –0.4 0 0.4 0.8

rx/δ rx/δ
Figure 9. Hairpin vortex signatures for the xz planes at y/δ = 0.19. The Reynolds numbers
are approximately Re θ 2400. The instantaneous velocity vectors are extracted at the different
instants and are displayed for a frame of reference moving at Uc = 0.74U∞ with outer variables.
The vortex legs, stagnation point and low-momentum region are indicated in (a) schematically,
but in (b) hairpin vortices appear in groups in the streamwise direction, thus the stagnation
point might not be present.
or cane-like shapes, or two-sided asymmetric shapes with varying size, age and aspect
ratio at different stages of their evolution; hence, the HVS is not applicable to all
hairpin vortices in the raw field. Here, we used the HVS defined above to selectively
discriminate hairpin vortices in the instantaneous flow fields.
4.2. Convection velocity
The convection velocity of a perturbation in a turbulent flow has been of fundamental
interest for several decades, because Taylor’s hypothesis uses the convection velocity
to relate the spatial and temporal distributions of flow variables by assuming frozen
turbulence. This hypothesis can be formulated as (Townsend 1976)
U (x, t) = U (x − Uc t, t + t), (4.1)
for a turbulence intensity that is small compared to the mean velocity and a value
of t (the time delay) that is not too large, where Uc is the assumed convection
velocity. The convection velocity determines the distance downstream at which the
perturbation can be located for a given time delay. It is often assumed to be equal to
the mean velocity at that height in the boundary layer, as in AMT, Tomkins & Adrian
(2003) and and Hutchins & Marusic (2007) for example. However, it is necessary to
check that this assumption is appropriate. We examined the validity of Taylor’s
hypothesis by plotting the spatio-temporal correlation defined as
Ruu (r, t) = u(x, t)u(x + r, t + t) . (4.2)
In general, the convection velocity is given by xmax /t, where xmax is the streamwise
separation for which the correlation is maximum at a given t (Kim & Hussain 1993).
If Taylor’s hypothesis is valid, the contours should be symmetric with respect to the
horizontal and wall-normal axes.
Figure 10 shows the contours of the spatio-temporal correlations of the streamwise
velocity fluctuations normalized by the outer scale at yref /δ = 0.066 and 0.18. Although
the present results were obtained at the fixed streamwise position xref /θin = 330
(Reθ = 2400), the overall trends are the same in the streamwise direction. In the
near-wall region (figure 10a), the inclinations of the quantities are less than that of
the symmetric line, which indicates that the convection velocity close to the wall
(a) 4 (b)
2
0.
2
0 .2
4
6
0.
0.
2
rtU/δ
0.
2
0 0.8.4 0.
0
4
0.4
0.
6 .6
0. 0
2
0.
–2 Convection velocity
Mean velocity
2
0.
–4 –2 0 2 4 –4 –2 0 2 4
rx/δ rx/δ
Figure 10. Spatio-temporal correlations of the streamwise velocity fluctuations normalized

by the outer scale: (a) yref /δ = 0.066 (y + = 30); (b) yref /δ = 0.18 (y + = 100). The solid lines are
the convection velocities and the dashed lines are symmetric rt U/δ = rx /δ lines.
(a) (b) 1.0

0 .2
2
50
0.
0.8
Mean velocity
Convection velocity
0 .6
0.6
rtU∞ /θin
yref /δ
0 t
0.4
0 .2
0 .2
x
0.2
–50
0
–50 0 50 0.6 0.8 1.0
rx/θin Uc/U∞
Figure 11. (a) Spatio-temporal correlation of the streamwise velocity fluctuations and (b) the
corresponding convection velocity in the wall-normal direction at xref /θin = 330 (Re θ = 2400)
downstream from the inlet.
is larger than the local mean velocity. In the log layer, however, the convection
velocity is almost equal to the mean velocity. This conclusion suggests that Taylor’s
hypothesis is a good approximation throughout the boundary layer except in the wall
region. These results are consistent with the LES data for a turbulent channel flow
obtained by Chung & McKeon (2010). The variations of the convection velocity in
the wall-normal direction are shown in figure 11 along with the mean streamwise
velocity. Except for the slight deviation in the near-wall region, the curve for the
mean velocity is in good agreement with the data shown as circles (the convection
velocity), so the local mean velocity is a suitable approximation for the convection
velocity in the outer region. Since the TBL develops spatially in the streamwise
direction (inhomogeneous), it was impossible in the present study to compute the
Fourier transform for the streamwise spectra related to the scale-dependent convection
velocity. Although turbulent structures with different wavenumbers move downstream
with different convection velocities, this variation mainly affects the small scales, so
this does not contradict the assumption that the VLSMs travel at a given convection
velocity in the streamwise direction (Krogstad, Kaspersen & Rimestad 1997).
4.3. Instantaneous fields
To examine the long streamwise-aligned motions related to the VLSMs, we applied
the Galilean decomposition technique to the instantaneous velocity fields with the
convection velocities Uc = 0.68 and 0.78U∞ in the logarithmic and wake regions.
This method has been discussed in detail in AMT and Adrian, Christensen & Liu
(2000b). In addition, vortical structures were identified by positive values of the
swirling strength λci , which is defined as the imaginary part of the complex conjugate
eigenvalue of the local velocity gradient tensor. Here, λci was calculated from the two-
dimensional gradient tensor D 2−D to detect wall-normal (angled vortex legs or neck)
and spanwise swirling motions (vortex head) in the xz and xy planes, respectively
(Adrian et al. 2000b). Here, D 2−D can be expressed as
⎡ ∂u ∂u ⎤ ⎡ ⎤
∂u ∂u
⎢ ∂x ∂z ⎥ ⎢ ∂x ∂y ⎥
D 2−D = ⎣ ⎦ or ⎢ ⎥
⎣ ∂v ∂v ⎦ . (4.3)
∂w ∂w
∂x ∂z ∂x ∂y
The swirling strength is frame-independent and can be used to extract compact
vortical cores within regions of shear (Zhou et al. 1999). In the present study,
the swirling strength was multiplied by the sign of the vorticity to distinguish the
direction of the rotation in each plane (λci ωy /|ωy | and λci ωz /|ωz |). By using this
method, Tomkins & Adrian (2003) identified the vortices with clockwise (positive)
and anticlockwise (negative) rotations in the xz plane, and Wu & Christensen (2006)
used a modified swirling strength to identify the direction of rotation in the xy plane.
A negative λci ωz /|ωz | corresponds to a prograde spanwise vortex with a rotation
in the same direction as the mean shear and the heads of the hairpin vortices.
In contrast, a positive λci ωz /|ωz | corresponds to a retrograde spanwise vortex with
rotation against that of the mean shear. These terminologies are autonomous concepts
used to distinguish the orientations of rotation in the two orthogonal planes.
Figure 12 shows the velocity vectors in the range Re θ = 2300–2500, which results
from subtracting the convection velocity Uc = 0.68U∞ from the flow field at y/δ = 0.18
in the xz plane. The blue and red contours represent the positive and negative patches
of λci ωy /|ωy |, respectively, with contours indicating 30 % of the maximum positive
and minimum negative values. The swirling strength is invariant with respect to the
convection velocity, and therefore it provides additional information that is useful for
the extraction of all of the vortical structures. The contours in figure 12(b) indicate the
Reynolds shear stresses associated with the Q2 and Q4 events, in which slow-moving
fluid is ejected away from the wall and high-moving fluid is swept into the wall.
The solid lines indicate the constant streamwise velocity fluctuations u+ = −1.0. The
contour and vector plots shown are derived from the inset in figure 7(a); these patterns
are typical of those found in many instantaneous fields with VLSMs. If we employ
the HVS discussed above, it can be seen that the log layer is densely populated with
velocity fields associated with streamwise-aligned hairpin vortices. In figure 12(a),
one- and two-legged hairpin vortices occur frequently in the streamwise direction and
these induce highly elongated regions of momentum deficit: anticlockwise vortices
94
(a)
1
B
rz /δ
J. H. Lee and H. J. Sung

C
–1
(b) 1
–uv+
29
25
21
rz /δ
0
17
13
9
–1 5
–2 0 2 1
rx /δ
Figure 12. (Colour online at journals.cambridge.org/FLM) Instantaneous field at y/δ = 0.18 (y + = 100) with the convection velocity Uc = 0.68U∞
subtracted. The contour and vector plots are derived from the inset in figure 7(a). Velocity vectors with (a) patches of λci ωy /|ωy |; the blue and red
contours represent the positive and negative values and (b) second- and fourth-quadrant components of the Reynolds shear stress.
appear on the right of the low-speed region and clockwise vortices appear on the left.
The VLSM with a streamwise length that exceeds the entire domain at the horizontal
line rz /δ = −0.51 (C) is relatively well aligned in the streamwise direction with the
spanwise meandering motions induced by the asymmetry and imperfect alignment of
the vortices. For the line B (rz /δ = 0.19), the upstream VLSM is streamwise-aligned
with slight spanwise meandering, while the downstream component deviates from line
B, but still extends throughout the entire domain. These very long low-momentum
regions are flanked by high-speed regions in the spanwise direction and are created
by the alignment of hairpin vortices. Along line A (rz /δ = 0.66), on the other hand,
the vortices that make up the VLSM are not coherently aligned and there are many
small-scale regions of low- and high momentum. Tomkins & Adrian (2003) reported
approximately 15 vortices in their relatively small domain of 1.5δ × 0.7δ (see figure 8
in their paper). Their results are consistent with our results, although the vector fields
were adjusted here to improve the quality of the figure by depicting only every second
and third vectors in the streamwise and spanwise directions, respectively, because
it is nearly impossible to represent all vortex patterns related to the hairpins, given
their size of approximately 30 viscous units in the large domain (AMT). However,
the magnified view with all velocity vectors in the inset of figure 12(a) shows that the
patterns of circular streamlines coincide closely with the contours of swirling strength
in the flow field.
Moreover, the Reynolds shear stress distributions shown in figure 12(b) confirm
that these long positive and negative streamwise-aligned meandering motions are
associated with the Reynolds shear stress, which shows that strong Reynolds shear
stress occurs between the regions of swirl motions. The maximum value of the
second- and fourth-quadrant Reynolds shear stress is approximately 30 in the aligned
regions and these zones of Reynolds shear stress often reach values 35 times larger
than the mean Reynolds shear stress of 0.89. Throughout the whole domain, these
results are qualitatively similar to the packet motions previously reported in PIV data
by Ganapathisubramani et al. (2003) in an examination of the spanwise structures
present in TBLs. They found that in the logarithmic layer, large-scale regions with
lengths of ∼2δ dominate and are consistently bordered by vortices with a significant
amount of the Reynolds shear stress in the streamwise direction. Our findings show
that the LSMs of the hairpin packets cooperatively induce highly elongated low-
momentum regions of VLSMs sandwiched between cores of positive and negative
swirling strength and that these sandwiched regions are associated with the Reynolds
shear stress in the logarithmic region.
Figure 13 shows the velocity vector fields that are present further from the wall.
Note that the convection velocity (Uc = 0.78U∞ ) at y/δ = 0.35 is from figure 11(b) and
the instantaneous flow field was obtained simultaneously with the results shown in
figure 12. Compared with the patterns that are apparent in the logarithmic layer, the
elongated patterns in the wake region with Reynolds shear stresses and streamwise-
aligned counter-rotating swirling motions are less organized, but large spanwise
regions of low-speed motions are present. In addition, there are similarities between
the coherent structures in the logarithmic and wake regions. This result means that
most hairpin vortices consisting of VLSMs do not extend further from the wall,
whereas larger hairpins that extend into the wake region occur frequently, as shown
in figure 13. These results are consistent with those of the previous study of packets by
Longmire, Ganapathisubramani & Marusic (2001). They found that the streamwise
spatial coherence and organization of hairpin packets are broken beyond the log
layer, and these motions contain larger individual vortex cores and spanwise strips of
96
(a)
1
rz /δ

–1
(b) 1
–uv+
29
25
21
rz /δ
0
17
13
9
–1 5
–2 0 2 1
rx /δ
Figure 13. (Colour online) The same as figure 12, but at y/δ = 0.35 (y + = 200) with the convection velocity Uc = 0.78U∞ subtracted.
wall-normal velocity with alternating signs, possibly cross-sections of individual
hairpin heads. Recently, Hutchins & Marusic (2007) found that large log scale
structures maintain a footprint in regions near the walls, as determined from
instantaneous DNS data. Although not shown here, similar results were found in
the present study, i.e. the very-large-scale structures in the log region simultaneously
influence the structures near the wall and in the wake regions.
To obtain additional information about the very long streamwise-aligned structures,
the instantaneous fields in the xy planes with the convection velocity Uc = 0.7U∞ were
investigated and are shown in figure 14. The vector fields were extracted at the same
point in time in the streamwise direction for the fixed lines A, B and C (rz /δ = 0.66,
0.19 and −0.51) depicted in figure 12. The blue and red contours represent positive and
negative patches respectively of λci ωz /|ωz | associated with the hairpin vortex heads,
and their magnitudes are 30 % of the maximum and minimum values, respectively.
The solid lines are the streamwise velocities U = 0.5, 0.7 and 0.9U∞ . For clarity, i.e.
to remove the ambiguities in the vortex patterns, the intervals of the vectors were
adjusted along the streamwise and wall-normal directions: every third and seventh
vectors are shown for y/δ < 0.2, every second and fifth vectors for 0.2 6 y/δ < 0.3
and every second and third vectors for y/δ > 0.3. However, the enlarged view with
high resolution in the inset of figure 14(c) shows the apparent vortex patterns.
It is clear that the three vertical planes are densely populated with prograde and
retrograde spanwise vortices and that prograde vortices consistent with the heads of
hairpin vortices are more frequently observed throughout the boundary layer (Wu &
Christensen 2006).
When we subtracted the constant convection velocity Uc = 0.7U∞ from the flow
field, it became possible to examine the patterns of nearly circular streamlines with
the patches of λci ωz /|ωz | along the constant streamwise velocity U = 0.7U∞ contour,
and found that the spatial coherence of the successive streamwise-aligned hairpin
vortices leads to strongly retarded low-momentum zones, as observed by Meinhart &
Adrian (1995). In their study, large, irregularly shaped regions of flow have relatively
uniform values of the streamwise momentum that are separated by thin regions of
large ∂u/∂y. In addition, the smaller and younger packets (U = 0.5U∞ ) arise within
the interior of the envelope (AMT). These results are consistent with those shown
for the xz plane in figures 12 and 13. In figure 14, highly elongated packet-type
structures are present along lines B and C based on the contour U = 0.7U∞ . If we put
horizontal lines at y/δ = 0.18 over the three planes, it becomes obvious that the low-
and high-momentum regions coincide closely with those over the xz plane in figure 12
at the corresponding streamwise locations. The flow induced by each vortex creates
a streamwise-aligned packet with a region of low momentum, and in this way these
packet motions induce VLSMs that are significantly longer than any single hairpin.
These features satisfy the criteria for HVS in packet motions reported by AMT.
Figure 15 shows the variations of the streamwise, wall-normal velocities and the
Reynolds shear stress along the fixed lines A, B and C shown in figure 12. The solid and
dashed lines depict the streamwise and wall-normal velocity fluctuations normalized
by the friction velocity and the bold solid line is the local Reynolds shear stress. A
careful inspection of the HVS in figures 8 and 9 indicates the flow patterns of the
hairpin vortices in the streamwise direction, which are consistent with a VITA event
(Robinson 1991): (i) The streamwise velocity exhibits the peak of high-momentum
region with negative wall-normal velocity, corresponding to Q4 event of upstream
fluids. (ii) The regions along the x-axis at which the streamwise and wall-normal
velocities change sign correspond to the stagnation points on the shear layers and (iii)
98
(a) A: rz /δ = 0.66
z
y/δ 0.5
0
(b) B: rz /δ = 0.19

z
y/δ 0.5
(c) C: rz /δ = –0.51
y/δ 0.5
0
–2 0 2
rx /δ
Figure 14. (Colour online) Instantaneous velocity vector fields with the convection velocity Uc = 0.7U∞ subtracted along the lines (a) A and
(b) B and (c) C shown in figure 12: the solid lines are the contours of the constant streamwise velocities U = 0.5, 0.7 and 0.9U∞ and the patches
of λci ωz /|ωz | associated with hairpin heads are shown for 30 % of the maximum and minimum values, respectively; the blue and red contours
represent the positive and negative values. A magnified view of the inset in (c) is included to show the apparent vector field.
5 (a) A: rz /δ = 0.66 u+
v+ 10
u/uτ, v/uτ
–uv+
–uv/u2τ
0 0
–5 –10
5 (b) B: rz /δ = 0.19 10
u/uτ, v/uτ
–uv/u2τ
0 0
–5 –10
5 (c) C: rz /δ = –0.51 10
u/uτ, v/uτ
–uv/u2τ
0 0
–5 –10
–4 –2 0 2 4
rx /δ
Figure 15. Variations of the streamwise (u+ ) and wall-normal (v + ) velocity fluctuations and
Reynolds shear stress (−uv + ) at y/δ = 0.18 along the lines (a) A, (b) B and (c) C. Lines: solid
line, u+ ; dashed line, v + ; bold solid line, −uv + .
a Q2 event (low-momentum region and positive wall-normal velocity) occurs behind

them with a large Reynolds shear stress due to vortex induction from the hairpin
legs and head. The streamwise histories in figure 15 show that there are many hairpin
signatures in the flow fields and these are significantly associated with high values of
the Reynolds shear stress. In figure 15(a), the signs of the velocity components for
the Q4 event (u > 0, v < 0) frequently change to those for the Q2 event in several
streamwise locations and these are closely related to the hairpin signatures with the
local Reynolds shear stress. However, these individual signatures are not aligned
coherently in the streamwise direction, i.e. these structures are single hairpin vortices.
On the other hand, figures 15(b) and 15(c) contain velocity signatures consistent with
highly elongated packets (AMT; Ganapathisubramani et al. 2003; Tomkins & Adrian
2003). These figures also contain several long low-speed regions and a positive wall-
normal velocity with a strong Q2 event, which indicates that the streamwise alignment
of the packets creates the VLSM when there is substantial Reynolds shear stress in
the TBL, especially in figure 15(c).
Figure 16 shows a visualization of the three-dimensional vortical structures that
uses an iso-contour in λci of 50 % of the maximum value. The vertical slices in
figure 14 are included at 20 % translucency for reference in figure 16(a) and the
three-dimensional low-momentum regions (u+ = −2) are depicted in figure 16(b) to
highlight the VLSMs visible in figure 12. Several highly elongated low-speed regions
are present in the outer region and each of them lies beneath many hairpin-type
vortices of asymmetric cane- and horseshoe-like hairpins including hairpin heads or
arches in the flow field. These flow patterns consist of three-dimensional vortical
structures with correlated spatial relationships that form much longer structures that
(a)
Flow direction 4
0
rx /δ
C
B –2
–1 A
0 Flow direction
rz /δ –4
1
(b)
–1
rz /δ 0
B
A
1
–4 –2 0 2 4
rx /δ
Figure 16. Instantaneous vortical structure visualized with an iso-surface of swirling strength
(50 % of maximum λci ). The xy plane data of figure 14 are included at 20 % translucency
for reference. (a) Perspective view with the vertical slices of figure 14; (b) top view of a
low-momentum region with u+ = −2.
erupt from the wall and grow towards the wake region and are consistent with those
in the two-dimensional planes.
4.4. Mechanisms for scale growth
4.4.1. Evidence for streamwise merging
The preceding results provide evidence for the presence of VLSMs and their
association with low-momentum regions and Reynolds shear stress throughout the
logarithmic layer. In this section, we address the question of the streamwise scale of
the packets that create VLSMs in the flow field. The idea that small scales organize
coherently to create larger scales is interesting and has received increasing support
over the past few decades. The hairpin packet model for the effects of hairpin vortices
offers an explanation for the most fundamental features of the logarithmic layer, such
as bursting processes and quasi-streamwise vortices. AMT described an idealized
conceptual model based on the hairpin packet paradigm, and showed how it explains
most features of the coherent structures in the outer region. In contrast, Kim &
2
(a) t+ = 0
rz /δ
0
–2
2
(b) t+ = 101.8
rz /δ
0
–2
2
(c) t+ = 203.5
rz /δ
0
–2
2
(d ) t+ = 305.3
rz /δ
0
–2
2
(e) t+ = 407.0
rz /δ
0
–2
2
( f ) t+ = 508.8
rz /δ
0
–2
2
(g) t+ = 610.5
rz /δ
0
–2
2
(h) t+ = 711.3
rz /δ
0
–2
–20 –10 0 10 20
rx /δ
Figure 17. (a–h) Time evolution of the instantaneous fields at y/δ = 0.21 in the xz plane. The
black and grey levels indicate the −1.0 and −0.5 u+ contours. The dashed line is added to
show the growth of the VLSMs and the box in (g) is included to depict the range of figure 7.
Adrian (1999) suggested a simple hypothesis that does not require the recognition
of VLSMs as a new type of turbulent motion. They conjectured that VLSMs may
be alignments of spatial coherence between hairpin packets; turbulent flows are
abundantly populated with packets. Unfortunately, however, little is known about the
formation of VLSMs and how they are created in the flow field. The objective of the
present research was to obtain spatial information about the formation of VLSMs
from packets by examining the time-resolved instantaneous flow fields.
The autogeneration mechanisms mean that the hairpins combine to create packets
that are long in the streamwise direction by continually spawning new hairpins. As
they grow, the packets become longer and larger in the streamwise and spanwise
directions, and are eventually probably broken up into smaller scale structures or
merge with adjacent packets. In the present paper, as a representative case of the
merging of vortex packets, and hence the merging associated with low-momentum
regions, the VLSM shown in figure 7 is discussed in detail. Figure 17 shows
the time evolution of the entire flow field in the xz plane with a time interval
t + = 101.8. The black and grey levels depict the −1.0 and −0.5 streamwise velocity
fluctuations, respectively, normalized by the friction velocity and the wall-normal
height is y/δ = 0.21. Since the real turbulence flow pattern is complex and it is hard
to discern the obvious features of scale growth in the full domain, the Gaussian
filter discussed in the preceding section was employed. Note that the instantaneous
rz /δ 1 (a) t+ = 0 2 (d) t+ = 40.7

0 0
–1 –1
1 (b) t+ = 13.6 1 (e) t+ = 54.3

rz /δ
0 0
–1 –1
1 (c) t+ = 27.1 1 ( f ) t+ = 67.8

rz /δ
0 0
–1 –1
–14 –12 –10 –8 –6 –14 –12 –10 –8 –6
rx /δ rx /δ
Figure 18. Evidence of the streamwise scale growth in the xz plane. The black and grey levels
indicate the −1.0 and −0.5 u+ contours. The dashed lines are depicted for the xy plane in
figure 19.
field (at t + = 0) shown in figure 7(a) is approximately from t + = 610.5 in figure 17,
as depicted by a black box. The dashed line was added to show the growth of the
VLSM. It is clear that the VLSM at t + = 711.3 in the range 8 < rx /δ < 24 originates
in several packet motions present at t + = 0. These packets that are still growing in
the streamwise direction interact with adjacent packets in the streamwise direction
and the merging packets result in the streamwise-aligned VLSM at t + = 101.8. The
VLSM is also growing continuously both in the streamwise and spanwise directions.
After t + = 407.0, however, although the streamwise growth of the VLSM occurs, the
spanwise length becomes thinner.
To show the streamwise scale growth, magnified views of figure 17 with small
temporal variation were examined through the xz and xy planes in figures 18 and 19.
Figure 18(a) shows two adjacent low-momentum regions induced by the hairpin
packets in the range −14.7 < rx /δ < −11.7 and −11.2 < rx /δ < −6.5. These regions are
travelling with similar convection velocities in the xz plane. As they move downstream,
the low-momentum regions merge in the streamwise direction, resulting in a longer
structure on the streamwise scale. Figure 19 shows the associated flow features in the
xy plane. The contours of the streamwise velocities U = 0.6, 0.7 and 0.9U∞ are shown
and the data were extracted along the streamwise direction at rz /δ = 0.2 from figure
18, as depicted by the dashed lines in figure 18. There are zones of multiple uniform
momenta that are created by successive hairpin vortices and larger, presumably
older packets, including the small presumably young packets lying close to the wall,
although the swirling strengths and velocity vectors are omitted from the figure to
show the dynamics of the hairpin packets. Based on the contour U = 0.7U∞ , there
are two packets of hairpins containing uniform momentum zones labelled IA and
IB at t + = 13.6 in the regions −14 < rx /δ < −11.5 and −11.5 < rx /δ < −8.5. These are
inclined at angles of 10◦ –15◦ in an almost linear fashion. Note that the flow patterns
of these aligned packets are consistent with the observations of AMT; see e.g. figure
19. If we search for low-momentum regions along the streamwise slice at y/δ = 0.21
(dashed lines), results consistent with those in the xz plane are found.
1
(a) t+ = 13.6
y/δ IA IB
0
1
(b) t+ = 40.7
y/δ
0
1
(c) t+ = 67.8
y/δ
0
–14 –12 –10 –8 –6
rx /δ
Figure 19. Evidence of the streamwise scale growth in the xy plane. The contours are the
streamwise velocities, U = 0.6, 0.7 and 0.9U∞ and the data are extracted along the streamwise
direction at rz /δ = 0.2 in figures 17 and 18. The dashed lines are depicted for the xz plane in
figure 18.
As can be seen in this figure, these packets are growing continuously in the
downstream direction and the part of the upstream packet further away from the
wall (the hairpin head or arch) experiences the higher mean flow velocity. This part
flows downstream faster than the lower-lying parts, which causes the vortex to lift
away from the wall with a still higher mean velocity, and results in still greater
stretching and rolling-up in the upper part of the upstream packet. Consequently,
this packet merges with the adjacent packet in the downstream to form a VLSM
with a shallow angle to the streamwise direction at t + = 67.8. These processes are
consistent with the hairpin structure dynamics proposed by Zhou et al. (1999). They
found that quasi-streamwise vortices lift up away from the wall due to the vortex
leg and the hairpin vortices are influenced by the mean shear, which stretches and
intensifies the vortices in the downstream direction. These findings indicate that one
consequence of a balance between the lifting-up and backward curling of vortices
due to the vortex-induced motion of the hairpin legs and the vortex stretching due
to the mean shear is the merging of adjacent hairpin packets and thus the formation
of a VLSM in the logarithmic layer. However, this merging process does not occur
frequently, so VLSMs are found less often than packets in flow fields.
It makes sense to explore the origin of adjacent hairpin packets at this point.
One possibility is that nearby hairpin vortices create individual hairpin packets
through the independent autogeneration mechanism suggested by Zhou et al. (1999).
A short three-dimensional disturbance grows into a hairpin, and then creates new
hairpins upstream and downstream if a threshold value is exceeded. Finally, an
interaction between the upstream and downstream packets occurs and the merging of
these packets is then possible. This mechanism appears reasonable, because hairpin
vortices are abundant in turbulent flows and the autogeneration mechanism is realistic.
Another possibility is that a new hairpin that creates a new adjacent packet develops
during the autogeneration process. In general, the primary hairpin continues to grow
in all directions, and two new hairpin vortices form downstream: a secondary hairpin
vortex and a tertiary hairpin vortex, etc. (Zhou et al. 1999). However, the tertiary
vortex occasionally forms at a location between the primary and secondary hairpins,
instead of at a location upstream of the secondary hairpin (Adrian 2007). As suggested
by Adrian (2007), this vortex could be the source of a new adjacent hairpin vortex
packet created by older, larger hairpins and this combination of the new packet with
an older packet results in VLSM formation. However, there is as yet no definitive
evidence for the processes in which packets trigger the formation of new packets or
the alignment of packets. In reality, an enormous range of relative flow configurations
is possible during merging, and many of these configurations could lead to merging
scenarios. Since the present DNS data contain very complex vortical structures with
large numbers of small-scale vortices, it is difficult to recognize all the vortex dynamics
involved in the formation of a VLSM by eye or by systematic image analysis. The
present observations describe one possible mechanism for the formation of VLSMs,
and hence a more thorough analysis is needed to disclose the general mechanism of
VLSM formation in the future.
4.4.2. Evidence for spanwise merging
As discussed in the previous section, hairpins can spawn new downstream hairpins
and create packets that are long in the streamwise direction. As they grow, the
packets simultaneously become larger in the wall-normal and spanwise directions.
This growth is associated with the growth of individual hairpins, but streamwise
growth also depends on the coherent alignment of successive hairpins (Zhou et al.
1999; AMT). The growth of hairpin packets moving in the streamwise direction can
be explained as a consequence of a balance between the mean shear stretching of the
hairpin head in the streamwise direction and the effects of self-induction, which result
in the curling up of quasi-streamwise vortices and the lifting of the head away from the
wall (Zhou et al. 1999). These processes are consistent with Townsend’s attached-eddy
hypothesis, which suggests that the sizes of eddies are proportional to the distance
from the wall. This scale growth occurs continuously in the wall-normal and spanwise
directions, which implies that there is likely to be an interaction between hairpin
vortices in the spanwise direction as they move downstream. Tomkins & Adrian
(2003) analysed the merging of vortex packets in a mechanism of spanwise scale
growth that is similar to those of Wark & Nagib (1989) and Adrian, Balachander &
Liu (2001).
Figure 20 shows the spanwise merging signatures at a final stage of the merging
process for the different two wall-normal locations to introduce the merging
mechanism, as suggested by Tomkins & Adrian (2003). The flow direction is from
left to right. The upstream, unmerged packets (rx /δ < 0.4) appear as two adjacent
spanwise packets and create low–high–low fluid patterns. The downstream low-
momentum region at approximately rx /δ > 0.4 is merged, or is in the process of
merging. The vector fields with the low-momentum regions show that the strength
of the backflow is relatively weaker in the vicinity of the merged structure than in
the upstream unmerged structure due to their double spanwise size, according to
the Biot–Savart law. In addition, the stagnation points are present at the interface
between the upstream and downstream packets. The upstream structures contain
many swirling motions due to hairpin legs and these are annihilated as the vortices
interact. These merging signatures are consistent with those of Tomkins & Adrian
(2003). Through the visual investigation for 4000 instantaneous vector fields, this type
of merging is frequently observed in the flow field and the 20 % instantaneous fields in
the log layer contain the merging signatures. In the present study, the time evolution
of the instantaneous flow fields was analysed to provide evidence for the spanwise
scale growth.
(a)
0.3
0
rz/δ
–0.3
–0.6
–1 0 1
(b)
0.5
0
rz/δ
–0.5
–1 0 1
rx/δ
Figure 20. (Colour online) Evidence of the spanwise scale growth through packet merging in
the xz planes. Velocity vectors and patches of λci ωy /|ωy | (30 % of the maximum and minimum
values): (a) Uc = 0.7U∞ , y/δ = 0.21; (b) Uc = 0.69U∞ , y/δ = 0.19.
The spanwise scale growth occurs between VLSMs, as shown in figure 21. The
vector fields with patches of λci ωy /|ωy | are depicted in figure 21(a) for the xz plane
and the time is set to t + = 0 for convenience. At t + = 0, two elongated low-momentum
regions induced by the alignment of the hairpin vortices are travelling parallel to the
streamwise direction. These motions appear to have comparable size in the spanwise
direction and create low–high–low fluid patterns. When the interaction of the VLSMs
occurs at a certain downstream position where the distance between the vortices
becomes small, the low-momentum regions are gradually merging in the spanwise
direction and eventually combined in the range 1 < rx /δ < 2 at t + = 122.1. The merged
structure has roughly double the spanwise scale, which induces a weaker backward
flow according to the Biot–Savart law. In addition, a stagnation point is present at
the interface between the upstream and downstream packets. Figure 22 illustrates
the flow features in the yz plane at the four streamwise locations rx /δ = −1.8, −0.9,
0.5 and 1.2 shown in figure 21. Note that these data were extracted at the same time.
There are two low-momentum regions elongated in the wall-normal direction and
these were created by the induction of the legs of the hairpin packets, as shown in
figure 22(a) at t + = 0. The inner sides of these momentum zones gradually merge
due to the interaction between them at t + = 81.4, which indicates that the inner legs of
(a)
0.5
rz /δ
–0.5
(b)
0.5
rz /δ
–0.5
(c)
0.5
rz /δ
–0.5
(d)
0.5
rz /δ
0
–0.5
(e)
0.5
rz /δ
–0.5
–2 –1 0 1 2
rx /δ
Figure 21. (Colour online) Evidence of the spanwise scale growth through packet merging in
the xz plane. Velocity vectors and patches of λci ωy /|ωy | at (a) t + = 0; (b) t + = 40.7; (c) t + = 81.4;
(d) t + = 122.1; (e) t + = 162.8. Here, Uc = 0.7U∞ , y/δ = 0.19. The dashed lines are depicted for
the yz and xy planes in figures 22 and 23.
(a) (c)
0.8
y/δ
0.4
0
(b) (d)
0.8
y/δ
0.4
0
–0.5 0 0.5 –0.5 0 0.5
rz/δ rz/δ
Figure 22. Time evolution of the merging structure in the yz plane at the streamwise locations
−1.8, −0.9, 0.5 and 1.2δ shown in figure 21. Velocity vectors with streamwise velocities, U = 0.3,
0.5, 0.7 and 0.9U∞ at (a) t + = 0; (b) t + = 40.7; (c) t + = 81.4; (d) t + = 122.1. The dashed lines
are depicted for the xz and xy planes in figures 21 and 23.
the hairpins that have the opposite swirl are annihilated and only the outer
legs of the larger hairpins eventually survive along the wall-normal direction.
However, the size in the wall-normal direction is comparable with that of the initial
vortex. These conclusions are similar to the features of the proposals of Wark &
Nagib (1989) and Tomkins & Adrian (2003). They assumed a complete annihilation
of the legs of hairpin-type structures, which results in a new hairpin of approximately
double the width of each of the original individual hairpins.

After merging, the new hairpin is expected to have an effect on the evolution of the
other structures. Figure 23 shows the characteristics of the merged structure in the xy
plane. The data were extracted along the streamwise direction at rz /δ = 0 in figures 21
and 22. The merged structure at t + = 122.1 is larger in the wall-normal direction than
the adjacent streamwise motions lying on the constant streamwise velocity contour of
0.7U∞ , but its height is similar to that in the spanwise direction shown in figure 22.
As the packet becomes larger, it induces a weaker backward flow owing to the greater
mean distance from the vortex elements in the hairpin cores to the centre of the
hairpin loop (Adrian 2007) and is stretched by the difference between the streamwise
velocities of its legs and head. Therefore, larger hairpin packets travel downstream
faster and become longer in the downstream direction. In addition, a new upstream
hairpin is spawned approximately in the range of 5.5 < rx /δ < 6.3 at t + = 325.5 and
this induces the much longer structure in the downstream at t + = 393.3. These results
from the time sequence of the instantaneous fields in the xz, yz and xy planes
demonstrate that the spanwise scale growth due to merging occurs continuously in
time and the spanwise scale increases rapidly to twice the original width of the
packets, which indicates that this mechanism for spanwise merging is an essential
process in turbulent flows.
Merged structure
1 (a) t+ = 122.1
y/δ
0
1 (b) t+ = 189.9
y/δ
0
1
(c) t+ = 257.7
y/δ
0
1
(d) t+ = 325.5
y/δ
0
1
(e) t+ = 393.3
y/δ
0 2 4 6 8 10
rx /δ
Figure 23. (a–d ) Time evolution of the merged structure in the xy plane. The contours are the
streamwise velocities, U = 0.5, 0.7 and 0.9U∞ and the data were extracted along the streamwise
direction at rz /δ = 0 from figures 21 and 22.
5. Properties of the very-large-scale motions

5.1. Feature extraction algorithm
The properties of the VLSMs discussed in the following section were estimated
statistically from the instantaneous fields. As shown above, very long meandering
structures travel with coherence in the downstream direction; these regions are
enclosed with multiple pairs of vortex cores and are associated with large values
of the Reynolds shear stress. We employed the feature extraction algorithm originally
developed by Ganapathisubramani et al. (2003) to identify the regions in which the
events related to VLSMs, LSMs and smaller scale motions occur. However, it is
difficult to identify the long features of packets with this algorithm, so we modified
it to detect large- and very-large-scale motions in the xz plane. We classified the
Reynolds shear stress into the second- and fourth-quadrant components that are
associated with very long negative and positive packets respectively and examined
their properties. The algorithm is as follows. (i) Zones of the second quadrant of the
+ +
Reynolds shear stress > 2σ−uv are identified, where σ−uv is the r.m.s. of the Reynolds
shear stress. (ii) If the zones found in (i) are observed between positive and negative
swirling strength > |1.5σλ+ci |, where σλ+ci is the r.m.s. of the swirling strength, one of the
points in the centre regions are defined as seed point at each location. Note that since
all points found in each zone are included in one packet motion, it is necessary to
select a representative point, although it does not matter which point is chosen. (iii)
On the basis of the seed point, we identified the connected low-momentum regions
(negative streamwise velocity fluctuations) in the flow fields. Since there are many
types of hairpins, for example, asymmetric hairpins with one dominant leg, hairpins
with a different stage of growth and hairpins with relatively weak Reynolds shear
stress, we only used the presence of low-momentum regions as the criterion for the
growth of large- and very-large-scale structures from the seed point. Actually, when
we considered all features of the hairpins described above (strong swirl and Reynolds
(a) (b)
0 5000 10 000 0 100 200 300 400 500
30 20
frequency|Q2(%)
y/δ = 0.35 y/δ = 0.18

20
10
y/δ = 0.35
10 y/δ = 0.18
0
0 5 10 15 20 0.2 0.4 0.6 0.8 1.0
Length Width
Figure 24. (a) Length and (b) width statistics for the patches found by the feature extraction
algorithm (Q2 event) at y/δ = 0.18 and 0.35. The length is normalized by the inner (top) and
outer (bottom) units.
shear stress and low-momentum region, etc.), it was difficult to detect the VLSMs
even for short packets.
After the calculations for the Q2 event, we applied the same procedures for the
Q4 event with high-momentum regions (positive streamwise velocity fluctuations).
However, since the estimated results for the Q4 event are almost the same as those for
the Q2 event, only the results for the Q2 event are shown in the present study. Such
results can be expected, because the inclination of hairpin vortices with respect to
the wall is such that the region outside of the legs/neck should induce the Q4 events
and the counter-rotating hairpin vortices inducing the low-speed flow also create the
high-speed flows on either side of the low-momentum regions, as shown in the xz
plane (AMT; Ganapathisubramani et al. 2003).
5.2. Frequency
Figure 24 shows the length and width distributions of the identified structures,
which were obtained with the feature extraction algorithm at y/δ = 0.18 and 0.35.
The ordinate in figure 24 is the ratio of the number of the patches found in each
length (width) to the total number of detected patches. Note that the length for the
VLSMs is defined as greater than 3δ–4δ in the present study, based on the criteria
of Guala et al. (2006) and Balakumar & Adrian (2007). In the log layer, the peak
occurs at less than approximately 0.25δ (∼150 in wall units), which indicates that the
predominant structure is a single hairpin vortex. This conclusion is consistent with
that of Ganapathisubramani et al. (2003). They showed that the peak occurs near
100 wall units. As the length increases, the frequency of the occurrence of patches
decreases monotonically. The long tail of the curve is due to the presence of elongated
streamwise-aligned structures in the log layer, although the number of VLSMs with
lengths larger than 10δ is small. The results show that the alignment of successive
hairpin vortices into packets occurs more frequently than the alignment of hairpin
packets to form VLSMs. However, the VLSMs exceeding 3δ–4δ in length account for
more than 40 % of all the patches identified by the extraction scheme. The maximum
length of the VLSMs found by the identification algorithm is approximately ∼18δ
in the streamwise direction, a result that is comparable with that of Hutchins &
Marusic (2007). They found regions of very long meandering positive and negative
streamwise velocity fluctuations exceeding 20δ. Simple calculations show that the
mean streamwise length of the patches with various length scales is 5.87δ; this value
(a) (b)
0 5000 10 000 0 5000 10 000
40 15
y/δ = 0.35 (×102)

30
–uv+|Q2(%)
10
y/δ = 0.18
20
y/δ = 0.18
5
10
y/δ = 0.35
0 5 10 15 20 0 5 10 15 20
Length Unit length
Figure 25. The contribution to the total second-quadrant Reynolds shear stress identified by
the feature extraction algorithm (Q2 event) as functions of (a) the length and (b) the unit
length of patches. The length is normalized by the inner (top) and outer (bottom) units.
is similar to that indicated by the two-point correlations in the latter. Individual

hairpin packets (less than 2δ–3δ) are more frequently observed in the wake region
than in the log layer, and the highly elongated streamwise coherence of the structures
found in the logarithmic layer decreases far from the wall.
The width distribution in figure 24(b) contains peaks at 0.1δ–0.2δ (60–115 in wall
units) in both the logarithmic layer and the wake region. There are no signs of
long tails in the spanwise direction, whereas the identified structures are significantly
elongated in the streamwise direction. The width statistics are also consistent with
those of Ganapathisubramani et al. (2003). They reported that the peaks are located
from 50 to 80 wall units in the width distributions and the width of the packets
increases with distance from the wall. This result confirms the growth of the spanwise
scale of VLSMs throughout the log and wake regions.
5.3. Contributions to the Reynolds shear stress
The length contributions to the total Reynolds shear stress are shown in figure 25.
The values in figure 25(a) were computed as the ratio of the total second-quadrant
Reynolds shear stress associated with all patches included in a certain length to the
sum of the second-quadrant Reynolds shear stress for all patches identified by the
feature extraction algorithm. The maximum contribution to the Reynolds shear stress
arises at 1δ–2δ in the log and wake regions (Ganapathisubramani et al. 2003), and the
LSMs and VLSMs with lengths greater than ∼1δ contribute more significantly to the
Reynolds shear stress than the smaller single hairpin vortices. Although the frequency
is maximum at approximately ∼0.5δ, the streamwise-aligned packets contribute most
of the Reynolds shear stress. The VLSMs (>3δ–4δ) related to the Q2 events in the log
layer contribute more than approximately 45 % of all the second-quadrant Reynolds
shear stress identified by the feature extraction algorithm and since the results for
the Q4 event are the same as for the Q2 event, it can be inferred that the VLSMs in
the log layer contribute more than 45 % of the Reynolds shear stress identified by the
identification algorithm. These findings demonstrate the importance of the VLSMs to
the Reynolds shear stress in wall-bounded turbulent flows. When the total Reynolds
shear stress identified by the extraction algorithm is divided by the global sum of all
Reynolds shear stress in all vector fields, we find that the contribution of the log layer
is approximately 20 %. This value is similar to that reported by Ganapathisubramani
et al. (2003).
(a) (b)
5
.0
–0
0.1
rz /δ
0.2
0.25
0 0.05 0.05 0.4
.05
–1 –0
–2 0 2 –2 0 2
rx/δ rx/δ
Figure 26. Two-point correlations of the streamwise velocity fluctuations at (a) yref /δ = 0.18
and (b) yref /δ = 0.35 in the xz plane.
Figure 25(b) shows the contribution to the Reynolds shear stress of the patches
per unit length. As expected, linear growth occurs in the log and wake regions, which
indicates that the contribution to the Reynolds shear stress is almost proportional to
the length of the motions through the boundary layer. Furthermore, the strength of
the Reynolds shear stress is larger in the log layer than in the wake region because the
larger and older packets create weaker back-induced flow in the wake region (AMT).
6. Statistical analysis
The instantaneous flow fields in the horizontal and vertical planes demonstrate
that elongated meandering motions are present in the log layer and are created as
a result of the alignment of pre-existing packet structures. If these features of the
instantaneous images are not intermittent but instead common features in the TBL,
the presence of these structures should affect the statistics of the flow. Hence, to
further examine the coherent structures associated with the VLSMs, we now examine
the two-point correlations and the conditionally averaged flow field.
6.1. Two-point correlations of the velocity fluctuations
To investigate the streamwise organization of the VLSMs in the logarithmic layer,

we examined the two-point correlations of the velocity fluctuations. The two-point
correlation coefficient is defined as
A(x)B(x + r)
RAB (r) = , (6.1)
Arms Brms
where r denotes the spatial separation vector and Arms and Brms are the r.m.s. of A
and B, respectively. The two-point correlations of the streamwise velocity fluctuations
at yref /δ = 0.18 and 0.35 are displayed in figure 26 to confirm the previous qualitative
observations of the instantaneous flow fields in the xz plane. The contour level spans
the range from −0.05 to 1.0, with intervals of 0.05. Figure 26 shows the presence of
long streamwise coherence with anti-correlated behaviour in the spanwise direction,
which is due to spanwise-adjacent low- and high-momentum regions that correspond
to elongated low-momentum regions with spanwise periodicity. In the logarithmic
layer, the correlation falls to a value of 0.05 at a distance of approximately 2.5δ,
which indicates that there is significant streamwise spatial coherence. Note that the
one-dimensional correlations shown in figure 3 cross from positive to negative values
at approximately 3δ, which implies that the statistical flow structure with a length of
∼6δ fully occupies the streamwise domain. These results are consistent with those of
(a) (b) (c)
1.0
0.8
0.6 Hutchins et al. (2005)

yref /δ
0.4
0.2
0
2 4 6 0 0.2 0.4 0.6 5 10 15 20
lx/δ lz/δ lx/lz
Figure 27. Variations with yref /δ of the characteristic (a) streamwise and (b) spanwise length
scales based on the streamwise two-point correlations. (c) Ratio of the streamwise and spanwise
length scales. The threshold was chosen as 0.05. For comparison, the experimental data of
Hutchins et al. (2005) for the 135◦ inclined plane are included in (b).
Hutchins & Marusic (2007), namely that the streamwise length scale inferred from
the peak of kx Φuu is of the order of λx ≈ 6δ. The results in figures 3 and 11 show that
VLSMs scale by the outer length scale (δ) regardless of the Reynolds number and
statistically they have streamwise lengths of ∼6δ in the flow field, although there are
a variety of length scales in the instantaneous flow fields shown in figure 24. In figure
26(b), the streamwise coherence is reduced above the log layer, but the spanwise width
increases with distance from the wall.
The variations in the streamwise and spanwise length scales were obtained from
the two-point correlation functions of the streamwise velocity fluctuations, and then
examined by extracting length scales with a nominal threshold (th = 0.05) for the
flow shown in figure 26; this process was repeated for several wall-normal locations.
Figure 27 shows the streamwise and spanwise length scales normalized by the outer
variable. The streamwise length scale lx /δ increases as far as yref /δ = 0.2, then sharply
decreases in the wake region. Such reduced streamwise coherence is apparent in the
instantaneous flow field shown in figure 13, which contains larger individual vortex
cores. However, the rate of decrease is not constant but stabilizes at a moderate
rate beyond y/δ = 0.5. This result implies that only larger or older packet structures
are present beyond the log region, which reduces the coherence in the streamwise
direction. This behaviour is distinct from the behaviour of pipe and channel flows,
in which the length of the correlations increases with the wall-normal distance.
Ganapathisubramani et al. (2005) attributed this behaviour to additional structures
originating from the opposite wall, which can influence the measurement regions.
In addition, this behaviour is strongly influenced by the selection of the threshold.
For example, when we chose a larger value of the threshold, th = 0.3, the streamwise
length scale was found to increase throughout the boundary layer, although these
results are not shown here.
In contrast to the streamwise length scale, the spanwise extent of the feature shown
in figure 27(b) increases along the full wall-normal distance, which indicates that
typical flow structures increase in width with increases in the wall-normal distance.
This result is consistent with Townsend’s attached-eddy hypothesis. When scaled by
the boundary layer thickness, the spanwise length of the extended features exhibits
(a) (b)
Ruu | u > 0 Ruu | u < 0

1
-0 .0 5
rz /δ
0.05 0.05
0 0 .30.55 0 .1
0 .2 5 5 0.15 0.2 5
0.1 0 .1
–0 .0 5
–1
–2 0 2 –2 0 2
rx/δ rx/δ
Figure 28. Conditional two-point correlations of (a) the positive and (b) the negative
streamwise velocity fluctuations at yref /δ = 0.18 in the xz plane.
a hyperbolic-type increase as a function of the wall-normal height: the increase is

more pronounced near the wall, and the rate of increase gradually decreases in the
wall-normal direction. These results are similar to those obtained by Tomkins &
Adrian (2003), Hutchins et al. (2005) and Ganapathisubramani et al. (2005) in their
investigations based on Ruu of the spanwise length scale. Hutchins et al. (2005)
reported that the spanwise length scale based on the 135◦ inclined plane exhibits an
approximately linear increase that is more pronounced in the log region (see figure
27b). The threshold used in their study is the same as that used here. In the present
study, the near-wall behaviour was found to be different from that above the log
region. This conclusion is similar to that for wall-normal length scales of Hutchins
et al. (2005). They found that there is a clear height from the wall at which the
inclination of the contour for the wall-normal length scales based on the 45◦ and
135◦ inclined planes seems to change; they tentatively described these two regimes as
the attached and detached regions. Scale growth in the spanwise direction might be
coupled with scale growth in the vertical direction, because continuous scale growth of
packets occurs in the wall-normal and spanwise directions simultaneously. The ratio
of the streamwise and spanwise length scales is shown in figure 27(c). This quantity
passes through a maximum value near the wall and decreases along the wall-normal
direction, which indicates that the structures are more elongated close to the wall
with a maximum aspect ratio of ∼20 and are less elongated further from the wall
with an aspect ratio of ∼4.
Previous investigations have shown that the long positive and negative meandering
motions in the instantaneous flow travel with coherence, and statistically these motions
are on the length scale of ∼6δ in the streamwise direction. The influence of the positive
and negative u fluctuations on the VLSMs is of particular interest. These structures
could result either from long zones of negative streamwise velocity fluctuations as
observed in the packet structures or from long zones of positive fluctuations, which
are sometimes observed on either side of the packets. The fluctuations related to
the long meandering motions were identified by calculating the conditional two-
point correlations of the streamwise velocity fluctuations at yref /δ = 0.18, as shown in
figure 28. The conditional correlations are defined as
u(x)u(x + r) | u(x) > 0
Ruu (r | u(x) > 0) = for high-speed streaks, (6.2)
u(x)u(x) | u(x) > 0
u(x)u(x + r) | u(x) < 0
Ruu (r | u(x) < 0) = for low-speed streaks. (6.3)
u(x)u(x) | u(x) < 0
(a)
0.6
0.4
y/δ
5
02
0.2
25
0.
0
0
-0 .
-0 .
0 50
0
0
(b)
0.6
0
0.4
0
y/δ
-0 .0 4
0
0 .0 4
0.2 0
-0 .
0 .0 8
04
0
0
0 0
–1.0 –0.5 0 0.5 1.0
rx/δ
Figure 29. Two-point correlations between the swirling strength and the velocity fluctuations
for yref /δ = 0.18 in the xy plane. (a) Rλu : contour levels are from −0.05 to 0.1 with increments
of 0.025 and (b) Rλv : contour levels are from −0.12 to 0.12 with increments of 0.02.
The resulting conditional two-dimensional correlations associated with low- and

high-speed streaks are displayed in figure 28. The streamwise extents of the coherences
for the positive and negative streamwise fluctuations are nearly identical to those in
figure 26 in the streamwise direction with the anti-correlated behaviour in the spanwise
direction, except that the positive fluctuations are biased in the downstream direction,
whereas the negative fluctuations are biased in the upstream direction. These biased
behaviours in the streamwise direction might be due to the strong Q4 events (positive
fluctuations) upstream of the inclined packets and the Q2 events downstream of the
packets. This means that the long streamwise coherence observed in the two-point
correlations (figure 26a), which have a statistical streamwise length of ∼6δ, results
from long zones of individual positive and negative streamwise fluctuations; these
findings are consistent with those from the previous extraction algorithm.
6.2. Two-point correlations between the swirling strength and the velocity
The instantaneous flow fields in figure 12 indicate the presence of long hairpin packet
motions moving downstream together. To further investigate the packet organizations
of these VLSMs, the two-point correlations between the swirling strength and the
velocity fluctuations are depicted in figure 29. The dashed lines indicate negative values
and the contour levels are from −0.05 to 0.1 with increments of 0.025 (Rλv ) and from
−0.12 to 0.12 with increments of 0.02 (Rλv ). Christensen & Adrian (2001) investigated
the two-point correlations of the swirling strength and the velocity fluctuations to
search for hairpin vortices aligned in a coherent manner. They measured the spatial
velocities by using PIV but the field of view of the measurements in the streamwise
direction was of the order of δ, which is not large enough to monitor a train of
packet motions. The same procedures were used in the present study, except that the
E F
0.6 D
C
y/δ
0.4
B
A
0.2
0
0 0.5 1.0 1.5 2.0 2.5
rx/δ
Figure 30. Linear estimate of the conditional velocity fields based on a spanwise swirl at
yref /δ = 0.18. The locations of the vortex cores are indicated by the swirling strength contours.
streamwise domain size was sufficient to account for coherently propagating hairpin
packets, one behind the other. The correlation contours show the average velocity
field associated with a vortex core located at the reference point. Since the swirling
strength always has a positive or zero value, Rλu and Rλv have the same signs as the
streamwise and wall-normal velocities, respectively. Thus, the statistically dominant
direction of rotation around the vortex core can be determined from the two-point
correlations between the swirling strength and the velocity fluctuations. In the xy
plane, Rλu is positive above the vortex core and negative below the vortex core.
Likewise, Rλv is positive on the left of the reference position and negative on the
right, i.e. the rotation in the spanwise vortices is clockwise. These results imply that
spanwise vortices with clockwise rotation are detected in this region more often than
those with anticlockwise rotation, which is consistent with the instantaneous fields
shown in figure 14. The inclined interface depicted by the zero contour is an estimate
of the average inclination angle of the outer vortex organization, which is induced by
the uniform low-momentum regions created by the several hairpin vortices moving
together in the flow direction. A similar trend has been observed in the two-point
correlation, Rλv (Christensen & Adrian 2001).
6.3. Conditionally averaged flow field

To determine whether a train of hairpin packet motions is present in the outer layer,
we examined the conditionally averaged flow field. Through an appropriate choice
of event (or condition), a conditionally averaged flow field can be obtained. We
approximated the conditionally averaged flow field by using LSE (Adrian 1996). The
conditional event was chosen to be a positive value of λci . The LSE ûi (x + r) of this
conditional average ui (x + r) |λci (x) is given by (Christensen & Adrian 2001)
ui (x + r) = Li λci (x), (6.4)
where i = 1, 2, 3, and
λci (x)ui (x + r)
Li = . (6.5)
λci (x)2
It can be seen from (6.5) that a stochastic estimate can be obtained from the
two-point correlations between the swirling strength and the velocity fluctuations, as
computed in the previous section.
The LSEs of the conditionally averaged velocity fields around the spanwise vortex
head were performed at yref /δ = 0.18. The flow structure induced by the LSE is
displayed in figure 30. Since the estimated velocity strength is strongest around the
event location, the length of the velocity vector was set to the unit vector to represent
the relatively weak motions away from the event location more clearly (Christensen &
Adrian 2001). The locations of the vortex cores are visualized by using iso-contours
of swirling strength equal to 10 % of its maximum value. In figure 30, the upward
motion of low-speed fluid (a Q2 event) can be seen upstream of and below the vortex
core at the event location and a downward motion of high-speed fluid (a Q4 event)
is present just behind the vortex core. Spanwise-oriented vortex structures are located
not only at the event location but also along the dashed line that corresponds to the
heads of the hairpin-type vortices that make up the hairpin packets in the logarithmic
layer. This line of swirling strength is inclined away from the wall at an angle of
13◦ –14◦ . These results imply that the vortices are arranged in the streamwise direction
with considerable coherence and are consistent with the conclusions of Zhou et al.
(1999) and Christensen & Adrian (2001).
It is clear that the streamwise separation of the vortex cores varies with the distance
from the event location. The streamwise distances from the vortex cores A to B and
B to C are approximately 0.3δ (x + = 160) in the outer variable. This separation
is similar to the streamwise separation between hairpin heads with 100–200 viscous
wall units (AMT) and a spacing of approximately 0.3δ–0.4δ (Christensen & Adrian
2001), but is slightly smaller than the 165–220 viscous scale range found by Zhou
et al. (1997, 1999) for asymmetric hairpin vortex packets in channel flows. AMT
reported that the vortex spacing is governed by inner scales but increases as hairpins
grow as they proceed downstream. However, Christensen & Adrian (2001) found
that the average streamwise spacing can be normalized by the outer length scale
δ, regardless of the Reynolds number. Away from the event location, the vortex
cores C and D are separated by approximately 1.1δ in the streamwise direction.
This spacing is approximately 3–4 times larger than the upstream vortex spacing.
The increased spacing between the vortex cores found in the present data might be
due to a train of packets in the streamwise direction (Christensen & Adrian 2001).
The streamwise spacings between D and E and between E and F are approximately
0.4δ and 0.23δ, respectively, which are similar to the distances between A and B
and between B and C, respectively. These results indicate that the vortex separation
in a packet is governed by both inner and outer variables, whereas the distance
between the packets is governed by the large scales associated with the streamwise
dimensions of packets. The present results demonstrate the following features of
packet organization in the logarithmic layer: hairpins concatenate to form packets
and these packets align coherently to form VLSMs. These conclusions are consistent
with the previous conjecture of Kim & Adrian (1999) that the large streamwise extent
of VLSMs is a consequence of the large-scale motions associated with the coherent
alignment of packets of hairpin eddies.
7. Summary and conclusions

We carried out the DNS of a TBL to investigate the spatial features and properties
of its VLSMs. The Reynolds number based on the momentum thickness (θin ) and
the free-stream velocity (U∞ ) was varied in the range Re θ = 570–2560. The simulation
of a spatially developing TBL is more difficult than that of a turbulent channel
flow because a non-periodic boundary condition must be imposed in the streamwise
direction to enable realistic inflow boundary conditions to be generated. We employed
Lund’s method with a sufficiently long streamwise domain and obtained realistic
inflow data that are in excellent agreement in both mean turbulence statistics and
phase information with the results of previous studies. The main simulation with a
length greater than 50δ o was then conducted independently by using the inflow data.
The statistical properties of the first and second moments of the velocity fluctuations
were compared with those of previous data, and good agreement was found; thus,
these results provide a useful database for the turbulence statistics of TBLs, which
are the topic of the current debate.
In order to investigate the flow patterns associated with VLSMs, we examined the
HVS and the convection velocity in the TBL. The present DNS data set contains HVS
consisting of circular vortex patterns created by hairpin heads with the inclined region
of the Q2/Q4 vectors in the xy plane and elliptical counter-rotating vortex patterns
induced by the hairpin legs with zones of low momentum in the xz plane. In addition,
the results extracted from the spatio-temporal correlations of the streamwise velocity
fluctuations showed that the convection velocity is almost equal to the mean velocity
throughout the boundary layer, except in the near-wall region, which indicates that
the local streamwise mean velocity provides an adequate estimate of the convection
velocity in the outer region and that Taylor’s hypothesis is a good approximation.
When the convection velocity was subtracted from the instantaneous fields in the xz
plane, we found that the hairpin vortices are streamwise-aligned coherently and that
they induce the significantly elongated high- and low-momentum regions of VLSMs
in the logarithmic layer. These zones are associated with large values of the Reynolds
shear stress, often 35 times larger than the mean Reynolds shear stress. Strongly
retarded uniform regions are present beneath the successive hairpin vortices aligned
along the constant streamwise velocity contour and these motions induce the VLSMs.
In the wake region, in contrast, the elongated regions with large Reynolds shear
stress and counter-rotating vortex pairs are less organized in the streamwise direction,
because only larger hairpin vortices extend into the wake region.
We also inspected the time-resolved instantaneous flow fields to investigate the
streamwise and spanwise scale growth of the hairpin vortex packets. As the packets
grow in the streamwise direction, they interact with adjacent packets and then merge
with them, resulting in longer structures in the streamwise direction. Two adjacent
packets of hairpins interacting with each other were observed. These packets grew
continuously in the downstream and upstream directions, and packets further away
from the wall (hairpin heads or arches) exhibited higher mean velocities. This upstream
packet convected downstream faster than its lower-lying parts, which caused the
vortex to lift away from the wall at a still higher mean velocity and resulted in
still greater stretching and rolling of the hairpin head. As a result, this packet
merged with the upstream part of the adjacent downstream packet to form a VLSM
with a shallow angle with respect to the streamwise direction. These observations
show how large streamwise scales of packets merge to create VLSMs in flow fields.
However, although this process occurs in many instantaneous fields, in reality an
enormous range of relative flow configurations not examined in the present study
can result in merging. These findings describe only one of the possible mechanisms
for VLSM formation through merging, and hence a more thorough analysis of other
merging mechanisms is required. In addition, to provide additional information about
the mechanism of spanwise scale growth, the three-dimensional characteristics of the
merging process were determined by examining the time evolution of the instantaneous
fields. Unmerged packets with comparable size in the spanwise direction were found
upstream; these two adjacent packets moved in parallel before merging and then these
packets combined to create the downstream, merged packet with weaker backflows
and double the spanwise width. The inner sides of the two elongated low-momentum
regions in the wall-normal direction, which feature swirling motions with opposite
signs, are annihilated and eventually only the outer sides survive as part of the larger
structure in the spanwise direction. These results support the mechanism for spanwise
scale growth of merging.
The frequency and the contribution to the Reynolds shear stress of the VLSMs were
estimated via a feature extraction algorithm that was developed to search for regions
of uniform momentum enveloped by cores of swirling motions of opposite sign with
significant Reynolds shear stress. The identified patches with lengths greater than
3δ–4δ account for more than 40 % of all the patches and these VLSMs contribute
approximately 45 % of the total Reynolds shear stress due to all patches, which
indicates the importance of VLSMs to the Reynolds shear stress as well as to
turbulence production and the transport of momentum in the logarithmic layer.
Our examination of the two-point correlations of the streamwise velocity
fluctuations found highly elongated streamwise coherence of the structure associated
with the very long positive and negative momentum regions residing in the logarithmic
layer of the TBL. This coherence is of the order of ∼6δ, which is the statistical
estimate of the mean streamwise length for VLSMs, LSMs and smaller motions,
and is consistent with that obtained from the feature extraction algorithm. Although
the streamwise length scale is higher in the logarithmic layer and lower above the
wake region, the spanwise length scale increases continuously along the wall-normal
direction. In particular, this increase is prominent in the near-wall region, although the
rate of increase gradually decreases. The LSE of the conditionally averaged flow field
based on the spanwise swirl demonstrated the following packet organization: hairpins
amalgamate to form packets and the packets align coherently to form the VLSM
in the logarithmic layer. The present study supports the hypothesis that packets or
bulges concatenate to form VLSMs, as conjectured by Kim & Adrian (1999).
This work was supported by the Creative Research Initiatives of NRF/MEST of

Korea (No. 2011-0000423).
REFERENCES
Adrian, R. J. 1996 Stochastic estimation of the structure of turbulent fields. In Eddy Structure
Identification (ed. J. P. Bonnet), pp. 145–196. Springer.
Adrian, R. J. 2007 Hairpin vortex organization in wall turbulence. Phys. Fluids 19, 041301.
Adrian, R. J., Balachander, S. & Liu, Z.-C. 2001 Spanwise growth of vortex structure in wall
turbulence. KSME Intl J. 15, 1741–1749.
Adrian, R. J., Christensen, K. T. & Liu, Z.-C. 2000b Analysis and interpretation of instantaneous
turbulent velocity fields. Exp. Fluids 29, 275–290.
Adrian, R. J., Meinhart, C. D. & Tomkins, C. D. 2000a Vortex organization in the outer region
of the turbulent boundary layer. J. Fluid Mech. 422, 1–54.
Balakumar, B. J. & Adrian, R. J. 2007 Large- and very-large-scale motions in channel and
boundary-layer flows. Phil. Trans. R. Soc. Lond. A 365, 665–681.
Christensen, K. T. & Adrian, R. J. 2001 Statistical evidence of hairpin vortex packets in wall
turbulence. J. Fluid Mech. 431, 433–443.
Chung, D. & Mckeon, B. J. 2010 Large-eddy simulation of large-scale structures in long channel
flow. J. Fluid Mech. 661, 341–364.
Degraaff, D. B. & Eaton, J. K. 2000 Reynolds-number scaling of the flat-plate turbulent boundary
layer. J. Fluid Mech. 422, 319–346.
Del Alamo, J. C. & Jimenez, J. 2009 Estimation of turbulent convection velocities and corrections
to Taylor’s approximation. J. Fluid Mech. 640, 5–26.
Del Alamo, J. C., Jimenez, J., Zandonade, P. & Moser, R. D. 2004 Scaling of the energy spectra
of turbulent channels. J. Fluid Mech. 500, 135–144.
Dennis, D. J. C. & Nickels, T. B. 2008 On the limitations of Taylor’s hypothesis in constructing
long structures in a turbulent boundary layer. J. Fluid Mech. 614, 197–206.
Ganapathisubramani, B., Hutchins, N., Hambleton, W. T. & Marusic, I. 2005 Investigation of
large-scale coherence in a turbulent boundary layer using two-point correlations. J. Fluid
Mech. 524, 57–80.
Ganapathisubramani, B., Longmire, E. K. & Marusic, I. 2003 Characteristics of vortex packets
in turbulent boundary layers. J. Fluid Mech. 478, 35–46.
Guala, M., Hommema, S. E. & Adrian, R. J. 2006 Large-scale and very-large-scale motions in
turbulent pipe flow. J. Fluid Mech. 554, 521–542.
Head, M. R. & Bbndyopadhyay, P. 1981 New aspects of turbulent boundary-layer structure.
J. Fluid Mech. 107, 297–338.
Hutchins, N., Hambleton, W. T. & Marusic, I. 2005 Inclined cross-stream stereo particle image
velocimetry measurements in turbulent boundary layers. J. Fluid Mech. 541, 21–54.
Hutchins, N. & Marusic, I. 2007 Evidence of very long meandering features in the logarithmic
region of turbulent boundary layers. J. Fluid Mech. 579, 1–28.
Jimenez, J. 1998. The largest scales of turbulent wall flows. Center for Turbulence Research, Annual
Research Briefs, pp. 137–154. Stanford University.
Khujadze, G. & Oberlack, M. 2004 DNS and scaling laws from new symmetry groups of ZPG
turbulent boundary layer flow. Theor. Comput. Fluid Dyn. 18, 391–411.
Kim, J. & Hussain, F. 1993 Propagation velocity of perturbations in turbulent channel flow. Phys.
Fluids 5, 695–706.
Kim, K., Baek, S.-J. & Sung, H. J. 2002 An implicit velocity decoupling procedure for the
incompressible Navier–Stokes equations. Intl J. Numer. Meth. Fluids 38, 125–138.
Kim, K. C. & Adrian, R. J. 1999 Very large-scale motion in the outer layer. Phys. Fluids 11,
417–422.
Krogstad, P. A., Kaspersen, J. H. & Rimestad, S. 1997 Propagation velocity of perturbations in
turbulent channel flow. Phys. Fluids A 10, 949–957.
Lee, J. H. & Sung, H. J. 2011 Direct numerical simulation of a turbulent boundary layer up to
Re θ = 2500. Intl J. Heat Fluid Flow 32, 1–10.
Longmire, E. K., Ganapathisubramani, B. & Marusic, I. 2001 Structure identification and
analysis in turbulent boundary layers by stereo PIV. In Proc. 4th Intl Symp. on Particle
Image Velocimetry, 17–19 Sept., Gottingen, Germany.
Lund, T. S., Wu, X. & Squires, K. D. 1998 Generation of turbulent inflow data for spatially-
developing boundary layer simulation. J. Comput. Phys. 140, 233–258.
Marusic, I. & Hutchins, N. 2008 Study of the log-layer structure in wall turbulence over a very
large range of Reynolds number. Flow Turbul. Combust. 81, 115–130.
Marusic, I., Mathis, R. & Hutchins, N. 2010b High Reynolds number effects in wall turbulence.
Intl J. Heat Fluid Flow 31, 418–428.
Marusic, I., Mckeon, B. J., Monkewitz, P. A., Nagib, H. M., Smits, A. J. & Sreenivasan, K.
R. 2010a Wall-bounded turbulent flows at high Reynolds numbers: recent advances and key
issues. Phys. Fluids 22, 065103.
Meinhart, C. D. & Adrian, R. J. 1995 On the existence of uniform momentum zones in a turbulent
boundary layer. Phys. Fluids 7, 694–696.
Monty, J. P., Stewart, J. A., Williams, R. C. & Chong, M. S. 2007 Large-scale features in turbulent
pipe and channel flows. J. Fluid Mech. 589, 147–156.
Priymak, V. G. & Miyazaki, T. 1994 Long-wave motions in turbulent shear flows. Phys. Fluids 6,
3454–3464.
Robinson, S. K. 1991 Coherent motion in the turbulent boundary layer. Annu. Rev. Fluid Mech. 23,
601–639.
¨ ü, R., Li, Q., Brethouwer, G., Fransson, J. H. M., Johansson, A. V.,
Schlatter, P., Orl
Alfredsson, P. H. & Henningson, D. S. 2009 Turbulent boundary layer up to Re θ = 2560
studied through simulation and experiment. Phys. Fluids 21, 051702.
Simens, M. P., Jimenez, J., Hoyas, S. & Mizuno, Y. 2009 A high-resolution code for turbulent
boundary layers. J. Comput. Phys. 228, 4218–4231.
Smith, C. R. 1984 A synthesized model of the near-wall behavior in turbulent boundary layers.
In Proc. 8th Symp. on Turbulence (ed. J. Zakin & G. Patterson), pp. 299–325. University of
Missouri-Rolla.
Smith, C. R., Walker, J. D. A., Haidari, A. H. & Sobrun, U. 1991 On the dynamics of near-wall
turbulence. Phil. Trans. R. Soc. Lond. A 336, 131–175.
Spalart, P. R. 1988 Direct simulation of a turbulent boundary layer up to Rθ = 1410. J. Fluid Mech.
187, 61–98.
Spalart, P. R., Coleman, G. N. & Johnstone, R. 2008 Retraction: ‘Direct numerical simulation
of the Ekman layer: a step in Reynolds number, and cautious support for a log law with a
shifted origin’. Phys. Fluids 21, 109901.
Theodorsen, T. 1952 Mechanism of turbulence. In Proc. 2nd Midwestern Conf. on Fluid Mech.,
pp. 1–19. Ohio State University, Columbus, Ohio.
Tomkins, C. D. & Adrian, R. J. 2003 Spanwise structure and scale growth in turbulent boundary
layers. J. Fluid Mech. 490, 37–74.
Townsend, A. A. 1976 The Structure of Turbulent Shear Flow, 2nd edn. Cambridge University Press.
Wark, C. E. & Nagib, H. M. 1989 Relation between outer structures and wall-layer events in
boundary layers with and without manipulation. In Proc. 2nd IUTAM Symp. on Structure of
Turbulence and Drag Reduction (ed. A. Gyr), pp. 467–474. Springer.
Wu, X. & Moin, P. 2009 Direct numerical simulation of turbulence in a nominally zero-pressure-
gradient flat-plate boundary layer. J. Fluid Mech. 630, 5–41.
Wu, Y. & Christensen, K. T. 2006 Population trends of spanwise vortices in wall turbulence.
J. Fluid Mech. 568, 55–76.
Zhou, J., Adrian, R. J., Balachandar, S. & Kendall, T. M. 1999 Mechanisms for generating
coherent packets of hairpin vortices. J. Fluid Mech. 387, 353–396.
Zhou, J., Meinhart, C. D., Balachandar, S. & Adrian, R. J. 1997 Formation of coherent
hairpin packets in wall turbulence. In Self-Sustaining Mechanisms of Wall Turbulence (ed. R.
L. Panton), pp. 109–134. Computational Mechanics, Southampton, UK.

Very Large Scale Motions in A Turbulent Boundary Layer

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Very Large Scale Motions in A Turbulent Boundary Layer

Uploaded by

Copyright:

Available Formats

J. Fluid Mech. (2011), vol. 673, pp. 80–120.

Very-large-scale motions in a turbulent

(Received 3 August 2010; revised 26 October 2010; accepted 29 November 2010;

Direct numerical simulation of a turbulent boundary layer was performed to

Key words: boundary layer structure, turbulent boundary layers, turbulence

† Email address for correspondence: hjsung@kaist.ac.kr

and contribution to the Reynolds shear stress in the ﬂow ﬁelds.

Figure 1. Schematic diagram of the computational domain.

(a) 1.0 (b) 1.0

500 1000 1500 2000 2500

Re θ Re δ θ/θin δ/θin uτ /U∞ y ∗ /θin H /θin

–0.4 0 0.4 –0.8 –0.4 0 0.4 0.8

Figure 10. Spatio-temporal correlations of the streamwise velocity ﬂuctuations normalized

(a) (b) 1.0

J. H. Lee and H. J. Sung

J. H. Lee and H. J. Sung

J. H. Lee and H. J. Sung

a Q2 event (low-momentum region and positive wall-normal velocity) occurs behind

rz /δ 1 (a) t+ = 0 2 (d) t+ = 40.7

1 (b) t+ = 13.6 1 (e) t+ = 54.3

1 (c) t+ = 27.1 1 ( f ) t+ = 67.8

double the width of each of the original individual hairpins.

5. Properties of the very-large-scale motions

y/δ = 0.35 y/δ = 0.18

y/δ = 0.35 (×102)

is similar to that indicated by the two-point correlations in the latter. Individual

To investigate the streamwise organization of the VLSMs in the logarithmic layer,

0.6 Hutchins et al. (2005)

Ruu | u > 0 Ruu | u < 0

a hyperbolic-type increase as a function of the wall-normal height: the increase is

The resulting conditional two-dimensional correlations associated with low- and

6.3. Conditionally averaged ﬂow ﬁeld

7. Summary and conclusions

This work was supported by the Creative Research Initiatives of NRF/MEST of

You might also like