International Journal of Heat and Fluid Flow 112 (2025) 109672

Contents lists available at ScienceDirect

International Journal of Heat and Fluid Flow

journal homepage: www.elsevier.com/locate/ijhff

Aspect-ratio effect on the wake of a wall-mounted square cylinder immersed

in a turbulent boundary layer
Gerardo Zampino a ,∗, Marco Atzori b , Elias Zea c , Evelyn Otero d , Ricardo Vinuesa a
a FLOW Engineering Mechanics, KTH Royal Institute of Technology, Stockholm, Sweden
b
Dipartimento di Scienze e Tecnologie Aerospaziali, Politecnico di Milano, Milan, Italy
c
MWL Engineering Mechanics, KTH Royal Institute of Technology, Stockholm, Sweden
d
CSA Engineering Mechanics, KTH Royal Institute of Technology, Stockholm, Sweden

ARTICLE INFO ABSTRACT

Dataset link: https://www.vinuesalab.com/dat The wake topology behind a wall-mounted square cylinder immersed in a turbulent boundary layer is
abases/ investigated using high-resolution large-eddy simulations (LES). The boundary-layer thickness at the obstacle
Keywords:
location is fixed, with a Reynolds number based on cylinder height ℎ and free-stream velocity 𝑢∞ of 10,000
Wall-mounted square cylinder while the aspect ratio (AR), defined as obstacle height divided by its width, ranges from 1 to 4. The mesh
Turbulent boundary layer resolution is comparable to DNS standards used for similar wall-mounted obstacles, though with relatively
Critical aspect ratio lower Reynolds numbers. The effects of AR on wake structures, turbulence production, and transport are
analyzed via Reynolds stresses, anisotropy-invariant maps (AIM), and the turbulent kinetic energy (TKE)
budget. In particular, the transition from ‘‘dipole’’ to a ‘‘quadrupole’’ wake is extensively examined as AR
increases. With increasing AR, the wake shrinks in both the streamwise and spanwise directions, attributed to
the occurrence of the base vortices (AR = 3 and 4). This change in the flow structure also affects the size of the
positive-production region that extends from the roof and the flank of the obstacle to the wake core. The AIMs
confirm three-dimensional wake features, showing TKE redistribution in all directions (Simonsen and Krogstad,
2005). Stronger turbulence production in AR = 3 and 4 cases highlights the role of tip and base vortices
behind the cylinder. The overall aim is to refine the dipole-to-quadrupole transition as a function of AR and
accounting for the incoming TBL properties. The novelty relies on proposing the momentum-thickness-based
Reynolds number Re𝜃 as a discriminant for assessing TBL effects on turbulent wake structures.

1. Introduction base, and spanwise vortices. The tip vortices are a couple of counter-
rotating vortices developing at the free end of the cylinder roof and
The study of turbulent structures around wall-mounted cylinders producing a strong downwash in the wake. In proximity to the base, a
has gained increased attention from the scientific community due to the second pair of counter-rotating vortices develop with an opposite sense
higher similarities with the urban environment. Predicting turbulent of rotation in respect to the tip vortices. The base vortices are inclined
flows around buildings becomes relevant for environmental purposes upwards and they produce a strong upwash in the lower portion of the
and new technology implementations. In the past decades, both nu- wake. Both tip and base vortices extend in the streamwise direction
merical and experimental studies in the literature have discussed the downstream the cylinder and were used by Wang and Zhou (2009)
behavior of near-wake structures as a function of the aspect ratio (AR), to distinguish between a ‘‘dipole’’, showing only the footprint of the
as the ratio between cylinder height and diameter (Sakamoto and Arie, tip vortices, and ‘‘quadrupole’’ configuration, where both tip and base
1983; Zhang et al., 2017), the Reynolds number (Zhao et al., 2021), the vortices are observed. However, as further discussed by Zhang et al.
turbulence of the incoming flow (Vinuesa et al., 2015), the boundary- (2017), the base vortices weaken downstream and the ‘‘quadrupole’’
layer thickness (Chen et al., 2022; El Hassan et al., 2015) and the in the near-wake region changes into a ‘‘dipole’’ configuration. In
cross-section shape (Kumar and Tiwari, 2019; Rastan et al., 2021), that addition, Wang and Zhou (2009) defined the spanwise vortices as the
makes a comprehensive parametric study of the wake regimes difficult. two vertical turbulent filaments separating at the side of the obstacle,
In recent studies, Wang and Zhou (2009) offered a detailed ex- connected by a horizontal bridge to form an arch-type vortex. Upstream
planation of the time-averaged wake turbulent structures behind a the cylinder a horseshoe vortex has been identified as an elongated
wall-mounted cylinder as a synthesis of three primary structures: tip,

∗ Corresponding author.
E-mail address: [email protected] (G. Zampino).

https://doi.org/10.1016/j.ijheatfluidflow.2024.109672
Received 9 April 2024; Received in revised form 21 November 2024; Accepted 21 November 2024
Available online 10 December 2024
0142-727X/© 2024 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

The ‘‘legs’’ at the sides of the cylinders form a ‘‘C’’ shape in any
Nomenclature
horizontal plane within the wake. Moving downstream, the two ends of
the ‘‘C’’ shape are deflected upstream forming the ‘‘Reverse-C’’ spanwise
Abbreviations vortices (Zhang et al., 2017). These latter turbulent structures transit to
AIM Anisotropy-invariant map the hairpin vortices due to the fragmentation of large pieces of vortices
AR Aspect Ratio because of the interaction between the boundary layer, the shear layer,
and the downwash produced by the free end of the cylinder (Zhang
CR Cross-sectional aspect ratio
et al., 2017).
IR Isolated roughness
Despite differing interpretations and models recently proposed in
SF Skimming flow
the literature, it is consolidated that turbulent structures downstream
TK E Turbulent kinetic energy of a wall-mounted cylinder vary with the AR (Wang and Zhou, 2009;
WI Wake interference Bourgeois et al., 2012; Porteous et al., 2014; Yauwenas et al., 2019).
English symbols For AR below a critical value ARc , the time-averaged wake exhibits
a dipole configuration with a couple of counter-rotating tip vortices.
𝑢′𝑖 𝑢′𝑗 i-jth Reynolds stress term.
The instantaneous wake exhibits a symmetric Kármán vortex shed-
Re𝜏 Friction Reynolds number
ding (Zhang et al., 2017). For increasing AR, above the critical value,
Re𝜃 Momentum-thickness-based Reynolds num-
the base vortices emerges and the instantaneous field becomes anti-
ber
symmetric. The vortex topology plays a crucial role in influencing the
Reℎ Bulk-velocity-based Reynolds number production and distribution of wake turbulence (Atzori et al., 2023)
Re𝛿 ∗ Displacement-thickness-based Reynolds and a comprehensive understanding of this phenomenon is further-
number more necessary. Although the estimated ARc relies between 2 and
𝑑 Cylinder width 6 (Sakamoto and Arie, 1983; Kawamura et al., 1984; Sumner et al.,
ℎ Cylinder height 2004; Wang and Zhou, 2009), universally agreed-upon definition is
𝑘 Turbulent kinetic energy still missing in literature since it depends on many factors, such as
𝐿𝑥 , 𝐿𝑦 , 𝐿𝑧 Length of the numerical domain in the 𝑥, 𝑦 the turbulence characteristics of the incoming flow (Vinuesa et al.,
and 𝑧 directions. 2015), the boundary-layer thickness (Hosseini et al., 2013; El Has-
𝑢𝜏 Friction velocity san et al., 2015; Chen et al., 2022) and the cross-sectional aspect
ratio (CR) (Rastan et al., 2021). Regarding the turbulence proper-
Greek symbols
ties, Goswami and Hemmati (2022) reported a similar transition from
𝛥 Mean element size = (𝛥𝑥𝛥𝑦𝛥𝑧)1∕3 the dipole to a quadrupole wake for relatively low Reynolds number
𝛿∗ Displacement thickness and for a laminar boundary layer. Although the generation mechanism
𝛿99 Boundary-layer thickness of the tip and base vortices is the same for both laminar and turbulent
𝜖 Isotropic dissipation boundary layer, Goswami and Hemmati (2022) motivated the early
𝜂 Kolmogorov scale defined as 𝜂 = (𝜈 3 ∕𝜖)1∕4 transition for a turbulent boundary layer due to the flow instabilities
𝜈 Kinematic viscosity inherent to a turbulent flow. The cross-sectional aspect ratio (CR), de-
fined as the ratio between the cylinder’s longitudinal length and width,
𝜃 Momentum thickness
was observed by Rastan et al. (2021) to have a more significant effect
Other symbols
on wake structures than AR. For CR = 1, the wake is characterized by
⋅+ Inner scaled quantity two streamwise vortices originating from side edges, forming a dipole
⋅ Time-averaged quantity configuration. With increasing CR, the flow reattaches to the top sur-
̃⋅ Filtered quantity face, creating a separation bubble that merges with streamwise vortices
and forms elongated structures called side vortices. Above the cylinder
roof, a weaker pair of tip vortices occurs but rapidly degrades. This
reattachment diminishes the three-dimensionality of the flow, making it
vortical structure that occurs at the junction between the cylinder and more two-dimensional. Accordingly, Rastan et al. (2021) proposed that,
the bottom surface due to the adverse pressure gradient induced by the in the absence of reattachment (low CR), the near wake displays an
obstacle blockage (Sumner et al., 2004). arch-shaped coherent structure with staggered legs at the flanks. In this
However, not all authors concur with this interpretation. For exam- framework, Kumar and Tiwari (2019) conducted a comparative study
ple, Sakamoto and Arie (1983) hypothesized that the flow separating on the wake structures of three wall-mounted cylinders with different
from the roof of the cylinder creates a turbulent filament that bridges cross sections (circular, triangular, and square) using direct numerical
the spanwise vortices from the flank of the cylinder producing an simulations (DNS) at varying Reynolds numbers and as a function of
arch-type vortex. The tip and base vortices are thus its footprint in the shear intensity. Despite the fact that AR was the same for all cases,
a vertical section. More recently Bourgeois et al. (2011) proposed a the transition from dipole to quadrupole wake occurred first for the
fully connected wake interpratation. Two vertical principal cores occur circular cylinder and then for the square and triangular cylinders with
at the lateral edge of the buildings and both sides are connected by changing Reynolds numbers.
a horizontal ‘‘connector strand’’ (Bourgeois et al., 2011, 2012). The To summarize the previous findings, Yauwenas et al. (2019) com-
distinction between ‘‘dipole’’ and ‘‘quadrupole’’ wake configurations bined experiments and LES of a wall-mounted square cylinder in a
depends on the presence of two looped or half-looped vortices oscillat- turbulent boundary layer in order to explore the parametric change of
ing behind the obstacle. Finally, in order of publication, Zhang et al. the critical AR𝑐 as a function of the AR and the BL thickness. By consid-
(2017) observed a third wake configuration known as ‘‘Six-Vortices ering their results with those of other authors such as Wang and Zhou
Type’’, where six poles develop behind the obstacle. Another model (2009), Bourgeois et al. (2011) and Hosseini et al. (2013), Yauwenas
is the hairpin vortex model by Dousset and Pothérat (2010), which et al. (2019) developed a new diagram of the transition from a dipole
describes the instantaneous wake. According to this model, the wake to a quadrupole wake as a function of two parameters: the aspect ratio
behind the obstacle can be distinguished as symmetric vortex shedding and the boundary-layer thickness. Yauwenas et al. (2019) highlighted
for lower AR values, and as an antisymmetric wake for large AR values. the existence of a transition region where both wake configurations

2
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

are equally possible, but without providing an analytical expression for by Atzori et al. (2023), the near-wall resolution follows the guidelines
critical AR𝑐 . 𝛥𝑥+ < 18, 𝛥𝑦+ < 0.5, and 𝛥𝑧+ < 9. Moving away from the wall, the
In conclusion, we provide an accurate analysis of the turbulent mesh resolution satisfies the condition 𝛥∕𝜂 < 9, where 𝛥 = (𝛥𝑥𝛥𝑦𝛥𝑧)1∕3 ,
flows around finite, wall-mounted square cylinders immersed in a 𝜂 = (𝜈 3 ∕𝜖)1∕4 is the Kolmogorov scale and 𝜖 is the local isotropic
turbulent boundary layer with data support from the TKE budget, dissipation. Here, the overall resolution refers to the fine resolution
the Reynolds stresses and anisotropy-invariant maps (AIM), and using with considering the spectral-element nodes. This makes the present
a high-resolution LES simulation at a relatively high Reynolds num- resolution close to the that of a DNS (Negi et al., 2018). For the sake of
ber. The overall aim is to study the transition from a ‘‘dipole’’ to a completeness, the mesh for the case with aspect ratio AR=3 is reported
‘‘quadrupole’’ as a function of the aspect ratio (AR), investigated in the in Fig. 1 (panel a). In the same figure, we also report the ratio 𝛥∕𝜂 at
past but still an open question when considering the role played at high 𝑧 = 0. The high-resolution LES is based on the time relaxation described
Reynolds number by the properties of the oncoming turbulent boundary by Negi et al. (2018) for the approximate deconvolution model. The
layer. While Yauwenas et al. (2019) focused on the transition from governing equations include a forcing term defined as the following
‘‘dipole’’ to ‘‘quadrupole’’ wake as effect of only the BL thickness and high-passed filtered velocity:
AR, we believe that this classification, based on recent findings, needs ∑
𝑁

to consider additional geometric and turbulent parameters (Vinuesa (̃

𝑢𝑖 ) = 𝜒 𝛾𝑘 𝛼𝑘 𝑃𝑘 , (3)
𝑘=0
et al., 2015). Consequently, we propose a brand new conceptual and
comprehensive definition of the critical AR𝑐 as a function of Re𝜃 . where 𝑢̃ is the filtered velocity, 𝜒 is the filter weight, while 𝑃𝑘 and
The methodology and numerical setup are extensively discussed and 𝛼𝑘 denote the base polynomials and the corresponding coefficients,
motivated in Section 2. The influence of the aspect ratio (AR) on the respectively. Hereinafter, the symbol (̃⋅) denotes the filtered quantities.
wake turbulence is detailed in Section 3, supported by the analysis of The filter transfer function is:
anisotropy-invariant maps in Section 3.3, and Reynolds stresses and ⎧ 0 𝑘𝑠 ≤ 𝑘𝑠𝑐 ,
the turbulent kinetic energy (TKE) budget are discussed in Section 3.4 ⎪( )2
𝛾𝑘 = ⎨ 𝑘𝑠 − 𝑘𝑠 (4)
𝑐
and Section 3.5, respectively. Conclusions and final considerations are ⎪ 𝑘𝑠 > 𝑘𝑠𝑐 ,
provided in Section 4. ⎩ 𝑁 − 𝑘𝑠𝑐
where 𝑘𝑠 is the wave number and 𝑘𝑠𝑐 is the cut-off wave number. The
2. Computational setup governing equations are non-dimensionalized with ℎ and 𝑢∞ .The forc-
ing term defined in this way is purely dissipative. The Navier–Stokes
A wall-mounted square cylinder immersed in an incompressible equations for the filtered velocity ̃
𝑢𝑗 then become:
turbulent boundary layer is investigated using high-resolution LES. 𝜕̃𝑢𝑖 𝜕̃𝑢𝑖 𝜕 𝑝̃ 1 𝜕 ̃
2𝑢
𝑖
The instantaneous streamwise 𝑢, vertical 𝑣 and spanwise 𝑤 velocities +̃
𝑢𝑗 =− + − (̃𝑢𝑖 ), (5)
𝜕𝑡 𝜕 𝑥𝑗 𝜕 𝑥𝑖 Reℎ 𝜕 𝑥𝑗 𝜕 𝑥𝑗
along the 𝑥, 𝑦, and 𝑧 axes, respectively, are Reynolds averaged and
𝜕̃𝑢𝑖
decomposed into a mean quantity in time (indicated by ⋅) 𝐮 and a = 0, (6)
𝜕 𝑥𝑖
turbulent fluctuation 𝐮′ . The Reynolds stresses are hence denoted by
𝑢′𝑖 𝑢′𝑗 . The coordinate system and the governing equations are dimen- Following the resolution criteria reported above, this methodology has
sionless and scaled by the height of the cylinder ℎ and the free-stream been shown to yield excellent agreement for the mean velocity, the
velocity 𝑢∞ of the incoming flow. Thus, the dimensionless time is Reynolds-stress tensor and the turbulent-kinetic-energy budget. The
divided by 𝑢∞ ∕ℎ. Given these definitions, the Reynolds number is Reℎ = domain size is 𝐿𝑥1 = 10ℎ, 𝐿𝑥2 = 6ℎ, 𝐿𝑦 = 3ℎ and 𝐿𝑧 = 4ℎ in the 𝑥,
𝑢∞ ℎ∕𝜈, where 𝜈 is the kinematic viscosity of the flow, and the √
friction 𝑦, 𝑧 directions, respectively (see sketch in Fig. 2). It is worth noting
Reynolds number is defined as Re𝜏 = 𝑢𝜏 𝛿99 ∕𝜈, where 𝑢𝜏 = 𝜏𝑤 ∕𝜌, that, in the present analysis, the domain dimensions are considerably
𝜏𝑤 the wall-shear stress and 𝛿99 is the boundary-layer thickness. The smaller compared to those reported in previous studies (Saeedi et al.,
momentum-thickness-based Reynolds number is Re𝜃 = 𝑢∞ 𝜃∕𝜈, where 2014; Vinuesa et al., 2015; Zhang et al., 2017; Rastan et al., 2019). This
the momentum thickness is reduction in the domain size can be justified by the outflow boundary
∞ ( ) conditions imposed in the present analysis (see the following).
𝑢 𝑢
𝜃= 1− d𝑦. (1) At the inlet, an incoming Blasius boundary layer profile with a
∫ 0 𝑢∞ 𝑢∞
displacement-thickness-based Reynolds number Re𝛿 ∗ = 450 is imposed.
Furthermore, the displacement-thickness-based Reynolds number The transition from laminar to turbulent flow is forced by a numerical
Re𝛿∗ = 𝑢∞ 𝛿 ∗ ∕𝜈 is expressed as function of tripping consisting of a numerical random volume forcing to model
∞( )
𝑢 the presence of an experimental trip. The tripping forcing is applied
𝛿∗ = 1− d𝑦. (2)
∫0 𝑢∞ along a transversal line at 𝑥 = −9. The position and the parameters of
the tripping have been calibrated to enable fully-developed turbulent
The high-resolution LES are carried out with the open-source code boundary layer before reaching the obstacle. For more details about the
Nek5000, which was developed by Fischer et al. (2008) using the tripping function, see Hosseini et al. (2016). At the top boundary we
spectral-element method (SEM) as described by Patera (1984). The prescribed a uniform freestream velocity and an outflow, leading to a
governing equations are discretized using the 𝑃𝑛 − 𝑃𝑛−2 formulation zero-pressure-gradient (ZPG) boundary layer. In the spanwise direction
of the Galerkin projection method, where 𝑛 represents the maximum we impose periodic boundary conditions, and at the outflow we set
polynomial order of the trial function for velocity, while 𝑛 − 2 cor- the stabilized condition by Dong et al. (2014). While the downstream
responds to the polynomial order for the pressure trial function. The domain size (𝐿𝑥2 ) may appear relatively small for the purposes of this
total number of elements is between 210,000 to 240,000 for the cases study, it has been chosen carefully in the light of the Dong outflow
under studies, with 101 elements in 𝑥, 41 elements in 𝑦 and 24 elements condition, which introduces a novel approach for highly truncated
in 𝑧, while in proximity of the obstacle the mesh resolution increases. unbounded domains. The dimensions of the domain have been metic-
For each element, a 7th-order polynomial following Gauss–Lobatto– ulously calibrated in both the streamwise and spanwise directions to
Legendre (GLL) quadrature has been considered to properly simulate strike a balance between computational efficiency and result accuracy.
the multi-scale character of turbulence. The resolution is increased by At the upper domain boundary (𝑦 = 𝐿𝑦 ) a zero-stress condition in the
reducing the size of the elements close to the wall. The mesh follows normal direction is imposed for the streamwise velocity component in
the criteria proposed by Negi et al. (2018) for the same high-resolution order to simulate irrotational flow in the outer boundary layer. Both
LES implementation. In agreement with the simulations carried out vertical 𝑣 and spanwise 𝑤 velocity components are otherwise set to

3
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 1. Three-dimensional view of the mesh at AR=3 (panel a) and 𝛥∕𝜂 at 𝑧 = 0 (panel b).

of the obstacle (see Atzori et al. (2023) for additional details). The
boundary-layer thickness is computed following the approach proposed
by Vinuesa et al. (2016) for APG TBL. In fact, upstream the cylinder,
the obstacle generates a small adverse pressure gradient that needs to
be considered in the definition of boundary layer thickness. This APG
due to the obstacle blockage makes the boundary layer ‘‘thick’’ as 𝛿99 ∕ℎ
exceeds the threshold value of 0.3 distinguishing ‘‘thin’’ from ‘‘thick’’
configuration. In this study at the obstacle location, 𝛿99 ∕ℎ and shape
factor 𝐻 are approximately 0.38 and 1.62, respectively, for all cases
considered.
In the present paper, four main configurations are studied at differ-
ent aspect ratios AR = 1, 2, 3, and 4 which are obtained by fixing the
height of the cylinders and changing their width. Thus, the incoming
turbulent properties are unaltered as well as the boundary-layer thick-
ness with respect to the obstacle height. The cases will be denoted as
Fig. 2. Sketch of the domain and the obstacle. ARx, where 𝑥 is the corresponding value of the aspect ratio.

2.1. Oncoming TBL properties

zero (Fischer et al., 2008). In the present analysis we report the results
for a wall-mounted square cylinder in a turbulent boundary layer with Before focusing on the effect of the aspect ratio on the mean wake
Reℎ = 10,000, corresponding to Re𝜏 ≈ 180 and Re𝜃 ≈ 460 at the location configurations, it is pivotal to observe the development of the oncoming

4
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

+
Fig. 3. Inner-scaled mean streamwise velocity 𝑈 + (panel a) and velocity fluctuation 𝑢′ 𝑢′ (panel b) profiles at 𝑥 = −4 and for cases AR2-AR4. The friction Reynolds number
considered is Re𝜏 = 168, corresponding to Re𝜏 . The production (green), transport (black), viscous diffusion (blue), dissipation (magenta) and pressure correlation (red) terms of the
TKE budget are reported in panel (c) only for the case AR2 and compared with the LES from Eitel-Amor et al. (2014) for matching Re𝜏 . (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

turbulent boundary layer in front of the obstacle. The mean inner- recirculation region behind the obstacle. This figure shows that the
scaled streamwise velocity 𝑈 + profile at 𝑥 = −4 is reported in panel more evident effect of AR is on the size of the wake because of the
(a) of Fig. 3 for the cases AR2, AR3 and AR4 and compared with the increasing influence of the tip and the base vortices. The occurrence
profile from Eitel-Amor et al. (2014) at matching Re𝜏 = 168. The present of both tip and base vortices can be outlined by the mean streamwise
results display a good agreement with the reference velocity profile for and vertical velocity components. The tip vortices are associated with
all cases considered. It is worth noting that blockage of the obstacle a downwash movement in the wake: stronger tip vortices imply a
slightly affects the velocity profiles in the outer layer and this explains stronger negative vertical velocity, a shorter wake, and hence a higher
the differences across the cases AR2-AR4 and with the reference zero- streamwise velocity. Similarly, the base vortices are associated with an
pressure-gradient boundary layer (ZPG BL) (Eitel-Amor et al., 2014). upwash movement, starting from the edge between the obstacle and
The good agreement with the reference profile not only demonstrates the bottom surface and extending for a short portion of the wake. For
the incoming TBL is properly simulated but also the effectiveness of the case AR1, the tip vortices are weak, as evidenced by the fact that
the numerical tripping formulation by Schlatter and Örlü (2012). The the negative velocity in the panel (b) is weaker than the other cases.
+
second-order statistics 𝑢′ 𝑢′ are also reported in panel (b) where the Additionally, base vortices do not emerge because there is not a positive
present results show higher peaks attributed to the small APG in front vertical velocity in the bottom part of the wake. Thus, the wake is
of the obstacles. It is interesting to observe that the APG affects the stronger and exhibits a single recirculation bubble that occupies the
TKE budget (panel c) for the case AR2, in particular for the pressure entire wake length until 𝑥 = 2, where the impingement point occurs at
correlation term that depends on the pressure field (not zero of the the bottom wall. The impingement point, defined by Zhang et al. (2017)
APG), but this effect is minor. This confirms that the assumption of as the saddle point where both upwash and downwash coexist, for the
ZPG BL upstream the obstacle is reasonable. case AR1 (panel a) is located at the bottom wall and slightly moves
upwards for higher AR. For increasing AR, the wake shrinks in both
3. Effect of AR on the wake structure 𝑥 and 𝑦 due to the stronger tip vortices associated with a downwelling
motion (see panels f and h). The recirculation bubble is reduced in size,
Although the effect of AR on the near wake is well-documented and a second recirculation zone occurs in the lower wake region. The
in the literature (Sakamoto and Arie, 1983; Wang and Zhou, 2009; reduction of the wake length can be correlated with the streamwise
Dousset and Pothérat, 2010; Rastan et al., 2017; Zhao et al., 2021), location of the impingement point from 𝑥 = 2 for AR1 to 𝑥 = 1.5 for
the majority of previous simulations have been conducted at low AR2 and 𝑥 = 1 for both AR3 and AR4. The change in the wake structure,
Reynolds numbers or employed significantly lower resolution at a from a dipole to a quadrupole wake, can also be described by observing
higher Reynolds number. In the present study, we aim to evaluate how the 𝑦 coordinate of the impingement point that moves from the
the influence of AR on a fully developed turbulent boundary layer wall (panel a) to the center of the wake (panels e and g). As observed
without wall modeling. High-resolution LES is employed to better by Zhang et al. (2017), the vertical shift of the impingement point is
describe the statistical behavior of the wake and to investigate in detail due to the upwash movement produced by the separation of the base
the various terms of the production, transportation, and distribution vortices from the bottom edge of the cylinder. In this context, a similar
of turbulence energy throughout the analysis of the TKE budget. It behavior in the change of the wake configuration has been observed
is worth noting that, despite the large amount of experimental and by Zhao et al. (2021), who carried out a series of simulations of square
numerical data (Wang and Zhou, 2009; Yauwenas et al., 2019; Zhang wall-mounted cylinders in a turbulent boundary layer at increasing
et al., 2017), a straightforward comparison is challenging due to the Reℎ while its AR is fixed to 4. For increasing Reynolds number, the
higher Reynolds number of the present study. wake of the wall-mounted cylinder changes, and the impingement point
moves from the center of the wake (for Reℎ = 100) to the wall (for
3.1. Mean and instantaneous flow field Reℎ = 500). This suggests that both the turbulence characteristics of
the incoming flow and the AR are important. For case AR1, the flow
The mean streamwise and vertical velocity components in the ver- detaches at the upper edge and reattaches on the roof, forming a small
tical symmetry plane are displayed in Fig. 4. Streamlines are included and weak recirculation region. For AR2–AR4, the flow does not reattach
in the same figure to help the reader with the visualization of the on the roof because of the shorter width of the cylinders. In front of the

5
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 4. Mean streamwise (left) and vertical (right) velocity components for the cases AR1 (a, b), AR2 (c, d), AR3 (e, f) and AR4 (g, h). The streamlines in the left panels illustrate
the recirculation bubbles inside the wake.

obstacle, a horseshoe vortex emerges. Its size is bigger for AR1 than b), but its intensity is weaker than that of cases AR3 (panel f) and AR4
AR4, where the position of the horseshoe vortex moves downstream (panel h). In addition, for the cases AR3 and AR4, in panels (f) and (g)
until it becomes not discernible from the base vortices. The horseshoe respectively, a positive velocity is barely visible at the bottom part of
vortex surrounds the sides of the obstacle, causing a downward flow the wake; note that this is a clear effect of the base vortices.
within the vertical tails on either side of the obstacle (Hussein and The time-averaged streamwise vorticity 𝛺𝑥 = 𝜕 𝑤∕𝜕 𝑦 − 𝜕 𝑣∕𝜕 𝑧 at
Martinuzzi, 1996). 𝑥 = 0 (left panels) 0.25 (central-left panels) 0.5 (central-right panels)
In the right panels of Fig. 4, the vertical velocity is broadly similar and 0.75 (right panels) is reported in Fig. 5 for the cases AR1 to AR4,
for all cases studied since the separation region remains constant across arranged from the top to the bottom row. The streamwise vorticity is
all the cases. The maximum positive velocity occurs at the front edge here utilized to depict the size and the strength of the mean turbulent
of the roof where the flow detaches, while the outer part of the wake structures. In particular, the cases AR1 (first row) AR2 (second row)
displays a negative 𝑣 as a result of the tip vortices (see panels f and only the tip vortices are observed while at the bottom the footprints
h). Due to this negative velocity, the wake captures the high-energetic of two pairs of horseshoe vortices are discernible in agreement with
flow at the outer layer and transports it to the core of the wake. the conclusion by Zhang et al. (2017). It is worth noting that Zhang
The consequences of this mechanism will be clear when analyzing the et al. (2017) also discerned into two types of tip vortices as primary and
production term of the TKE budget (see Section 3.5 for more details). secondary vortices produced by the leading and trailing edge, respec-
The magnitude of the negative vertical velocity is larger for AR1 (panel tively. For increasing AR the horseshoe vortices become progressively

6
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 5. Contours of the Streamwise vorticity 𝛺𝑥 at 𝑥 = 0 (left) 𝑥 = 0.25 (center-left) 𝑥 = 0.5 (center-right) and 𝑥 = 0.75 (right) for the cases AR1-AR4 moving from the top to the
bottom. 𝛺𝑥 varies from −6 to 6. Horseshoe, arch, tip and base vortices are properly indicated in the figure.

smaller and move closer to the bottom surface, flattening out. As largely hairpin vortices (Dousset and Pothérat, 2010; Zhang et al., 2017) (panel
anticipated in the previous section, only the cases AR3 and AR4 display c) agglomerated into a complex turbulent structure.
the base vortices, more evident at 𝑥 = 0.5 (panels m and n) and 𝑥 = 0.75 In this context, Dousset and Pothérat (2010) performed direct nu-
(panels q and r). Although at the first sight the horseshoe and base merical simulations of a wall-mounted square cylinder, immersed in a
vortices appear similar and indistinguishable, a more detailed analysis turbulent flow delimited by a rectangular duct, to investigate the forma-
reveals that the base vortices occur away form the bottom surface tion of the hairpin vortex for varying Reynolds numbers. Dousset and
and at a quarter of the wake height (i.e. panel m and panel q). In Pothérat (2010) distinguished between a symmetric shedding (Re𝑑 =
addition, the horsewhoe vortices developing in the closest proximity 𝑢∞ 𝑑∕𝜈 = 200), where a single row of the hairpin vortices is placed in the
of the obstacles have the same rotational direction as the tip vortices streamwise direction, and antisymmetric vortex shedding (Re𝑑 > 250),
from the same side of the cylinder. where the upwash induced by the base vortices forces the hairpin
vortices to aggregate in a complex turbulent structure where their
The effect of the downstream distance from the obstacle is also
parts are barely discernible. This latter wake theory was expanded
evident. At 𝑥 = 0 (center of the obstacle) the tip vortices are still in
by Zhang et al. (2017), who observed that for both the symmetric
development and the flow field surrounds the obstacle forming the
and antisymmetric configurations, the hairpin vortices are the results
well-known arch vortex. Moving downstream (from left to the right
of the fragmentation of the Reverse-C spanwise vortices due to the
columns) the arch-type vortex disrupts, the tip and base vortices emerge
interaction between the downwash, the boundary layer and the shear
more clearly. These changes in the time-averaged structures are also
layer. Although the hairpin vortices have been already described for
justified by the difference in the instantaneous wake field that changes low Reynolds numbers, the wake structures observed in Fig. 6 for
from a symmetric to an antisymmetric vortex shedding. higher Reynolds numbers are not related to any of the aforementioned
The instantaneous vortical motions are identified using the 𝜆2 - structures. The wake is more chaotic and the only common pattern is
criterion (Jeong and Hussain, 1995) in Fig. 6. The wake is symmetric the arch-type vortex separating at the cylinder edges.
for AR1 (panel a) and AR2 (panel b), when AR < ARc , and the Describing the wake complexity is not straightforwards. However,
wake is defined as a dipole. For AR3 and AR4, the wake is antisym- the comparison between the instantaneous field and the averaged
metric (Kármán vortex shedding). In panels (c) and (d), the wake is quantities can demonstrate that studying the changes in the mean flow
a quadrupole, and the base vortices are discernible. Downstream of field can provide info about the unsteady characteristics of the flow
the obstacle, these turbulent structures produce structures resembling field around the high aspect-ratio buildings.

7
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 6. Instantaneous vortical motions identified by 𝜆2 = −40 for the cases AR1 (a), AR2 (b), AR3 (c), AR4 (d). The colormap is the instantaneous streamwise velocity 𝑢, ranging
from −1.1 (dark blue) to 1.75 (dark red). The arrow indicates the flow direction in each case. (For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)

flow parameters, such as the momentum thickness, the boundary-layer

thickness and the turbulence properties of the incoming boundary
layer. The novelty proposed in this study consists of a qualitative
description of the transition between dipole and quadrupole wakes,
say the ARc , as a function of Re𝜃 = 𝑢∞ 𝜃∕𝜈, based on the momentum
thickness 𝜃. Here, the momentum thickness is preferred because it
accounts not only for the boundary-layer thickness, as proposed by
many authors in literature (see Yauwenas et al., 2019 and Cao et al.,
2022 among others) but also for the turbulent intensity and properties
of the incoming boundary layer as reported in the available data in the
literature (Hosseini et al., 2013; Bourgeois et al., 2012; Sattari et al.,
2012; Vinuesa et al., 2015; Uffinger et al., 2013; Wang et al., 2006).
For experiments regarding a turbulent boundary layer developing on
a flat plate from Wang et al. (2006), when 𝜃 is not available, Re𝜃 is
calculated by considering a naturally developing turbulent boundary
layer. The data of wall-mounted square cylinder immersed in a turbu-
lent boundary layer are reported in Fig. 7 where closed symbols denote
dipole configurations while open symbols are used for a quadrupole
configuration. We distinguish a dipole-wake region (blue area) and
a quadrupole-wake region (red area) that overlaps in the transition
region (Yauwenas et al., 2019). Yauwenas et al. (2019) observed a
transition region where both dipole and quadrupole configurations
are possible. Note that Yauwenas et al. (2019) considered only the
Fig. 7. Comparison of the wake configuration as a function of AR and Re𝜃 (see figure
dependence on the boundary-layer thickness, and not the turbulence
legend for further details). In the figure, closed symbols denote dipole wakes, while properties of the flow. However, it is not enough to characterize the
open symbols refer to quadrupole wakes. transition between the symmetric and antisymmetric wakes (Yauwenas
et al., 2019): as discussed by Vinuesa et al. (2015), the turbulence
characteristics of the incoming flow can alter the wake configura-
3.2. Critical aspect ratio (ARc ) tions and this discrepancy is sufficient to explain the difference in the
wake configuration between the present data and the results produced
The present analysis suggests a critical aspect ratio (ARc ) between by Yauwenas et al. (2019). However, the lack of data within the tran-
2 and 3 where the wake loses its symmetry and Kármán-like vortex sition region makes it difficult to establish the exact extension of this
shedding is observed. The critical aspect ratio ARc depends on various area. The transition threshold between the dipole and the quadrupole

8
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 8. Anisotropy-invariant maps of cases AR1–AR4 (see legend for further details). The velocity profiles are extracted on the symmetry plane (𝑧 = 0) and along the vertical
direction between 𝑦 = 0 (bottom surface) and 𝑦 = 1.0 (obstacle height) and different locations downstream the cylinder at 𝑥 = 0.5 (panel a), 𝑥 = 0.75 (panel b) and 𝑥 = 1.0 (panel
c). The black box delimits the region 𝑦 in [0, 0.5]. The black dashed lines are the theoretical limits of the II-III which definitions are reported in Eq. (8) (Lumley and Newman,
1977).

wake configuration changes as a function of Re𝜃 and the AR critical, 3.3. Anisotropy-invariant maps
between Re𝜃 = 460 and 1000, increases with the Reynolds number.
The dependence of the ARc on the Reynolds number is well known The literature widely demonstrated that the wake of a finite wall-
in the literature and largely investigated by other authors, Wang and mounted cylinder is three dimensional. These characteristics add com-
Zhou (2009), Zhang et al. (2017) among others. The Fig. 7 summarizes plexity to the flow field, making it important to quantify the three-
this dependence on the Re𝜃 . Within a moderate range of Re𝜃 between dimensionality. Therefore, the anisotropy-invariant maps (AIM) allow
480 and 800, the ARc increases and its value is between 2–6, in us to visualize the changes in the turbulence distribution across the
agreement with the literature. However, for lower Re𝜃 between 100 𝑥, 𝑦, and 𝑧 directions. The anisotropy tensor was originally introduced
and 400, the critical value rapidly decreases and the dipole wake has by Lumley and Newman (1977) as:
been observed experimentally also for very large 𝐴𝑅 (Uffinger et al., 𝑢′𝑖 𝑢′𝑗 𝛿𝑖𝑗
𝑎𝑖𝑗 = . − (7)
2013; Hosseini et al., 2013). On the other hand, ARc peaks at 5.5 2𝑘 3
for Re𝜃 = 809 and a further increase of Re𝜏 produces a reduction of Lumley and Newman (1977) identified three invariants, namely:
ARc that needs to be evaluated. Although the literature regarding wall-
𝐼 = 𝑎𝑖𝑖 , 𝐼 𝐼 = 𝑎𝑖𝑗 𝑎𝑗 𝑖 , 𝐼 𝐼 𝐼 = 𝑎𝑖𝑗 𝑎𝑖𝑛 𝑎𝑗 𝑛 . (8)
mounted square cylinders immersed in turbulent boundary layers under
different conditions is reasonably rich, many authors did not report The anisotropy-invariant map (AIM) graphically displays the develop-
the momentum thickness, a fact that makes it difficult to provide a ment of the dynamics of turbulence in the II-III space (Lumley and
more accurate and general expression for ARc at the moment based Newman, 1977; Banerjee et al., 2007). The space of the possible II-
on Fig. 7. Additional data will be needed to accurately depict this III values is delimited by the two-component limit (II = 2/9 + 2III)
transition region. It is worth noting that the present study does not that connects the one-component limit (II = 2/3, III = 2/9), where
investigate the effect of the cross-sectional ratio because the CR is the turbulent energy is distributed along a single direction, and the
constant and equal to 1 for all cases considered in order to have a two-component axisymmetric limit (II = 1/6, III = −1/36), where the
square section. Although many authors (Rastan et al., 2021; Zargar turbulence energy is distributed along two directions. The isotropic
et al., 2022) have suggested that the CR parameter is important for limit is given by the conditions II = III = 0, which are connected
describing the transition from the dipole to the quadrupole wake. with the 2D-isotropic case by the axisymmetric contraction limit (II
Based on the present findings, this transition can be attributed to the = 3/2(4/3|III|)2∕3 ) and to the 1D limit by the axisymmetric expan-
sion limit (II = 3/2(4/3|III|)2∕3 ). Using the interpretation proposed
change in the 𝛿99 ∕𝑤 ratio. Nevertheless, a comprehensive study of the
by Simonsen and Krogstad (2005), the nature of the anisotropy can
combination effect of both the cylinder height and width changes is still
be represented as an ellipsoid whose radius and axis are proportional
missing in literature. Extending these theoretical considerations while
to the Reynolds stress tensor and its eigenvalue. In agreement with
maintaining a general and speculative viewpoint, we can hypothesize
this interpretation, the 1D turbulence corresponds to a line, the two-
that the transition from dipole to quadrupole wake is proportional to
component limit is represented as a plane elliptical disk that becomes
the parameter 𝛿99 ℎ∕(𝑏𝑑), where 𝑏 is the cylinder spanwise width. In fact,
a circular disk for the 2D axisymmetric limit, while a prolate (or
as observed by previous papers (Wang and Zhou (2009) among others) oblate) spheroid represents the axisymmetric contraction (or expan-
ℎ and high Reynolds numbers or 𝛿 (Wang et al., 2006) promote the sion) limit (Simonsen and Krogstad, 2005). The limits of the II-III space
emergence of the base vortices while 𝑑 and 𝑏 tend to stabilize the dipole are reported in panel (a) of Fig. 8 for the sake of clarification.
wake (Rastan et al., 2021). In conclusion, this study emphasizes the The AIMs for the cases AR1–AR4 at the longitudinal symmetric
necessity of standardizing the parameters used to describe the effect of plane (𝑧 = 0) and for the velocity profiles at 𝑥 = 0.5 (panel a),
turbulence intensity in the oncoming TBL on the transition between the 0.75 (panel b), 1 (panel c) are reported in the same figure. The AIMs
‘‘dipole’’ and the ‘‘quadrupole’’ wakes. We propose that the Reynolds display the invariants II and III moving from the two-component limits
number based on momentum thickness (Re𝜃 ) and aspect ratio (AR) are at the base wall (denoted by the circle) to a 3D isotropic limit at
sufficient to accurately characterize this transition. the upper region of the wake (denoted by a square marker). This

9
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 9. Comparison between the anisotropy invariant maps for two square cylinders in a row with AR=2 and placed at different gaps indicated as skimming flow (SF) in panel
(a), Wake interference (WI) in panel (b) and Isolated Roughness (IR) in panel (c). The curves have been obtained at 𝑥 = 0.5 (blue lines) to 𝑥 = 0.75 (orange lines) and 𝑥 = 1
(yellow lines). The data of AR2 (light blue lines) for 𝑥 = 0.5 (solid) to 𝑥 = 0.75 (dot-dashed) and 𝑥 = 1 (dashed) are superimposed. Data extracted from Atzori et al. (2023). The
black dashed lines are the theoretical limits of the II-III (Lumley and Newman, 1977). (For interpretation of the references to color in this figure legend, the reader is referred to
the web version of this article.)

Fig. 10. Turbulent kinetic energy 𝑘 at the symmetry plane 𝑧 = 0 for the cases AR1 (panel a), AR2 (panel b), AR3 (panel c) and AR4 (panel d). The contour lines delimit the
regions of high 𝑢′ 𝑢′ (yellow) 𝑣′ 𝑣′ (red) and 𝑤′ 𝑤′ (black). The lines correspond to a third of the maximum stress. (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)

trend is consistent for all positions considered downstream of the compared with the corresponding maps for an isolated square cylinder
obstacle. Except for the first point at the wall, the wake is highly (light blue lines). For IR, the AIM closely aligns with the AR2 case at
three-dimensional. However, for the cases AR2–AR4, the turbulence both 𝑥 = 0.75 and 𝑥 = 1, while closer to the obstacle (𝑥 = 0.5) the
distribution follows the axisymmetric expansion limit. For 𝑥 = 0.75 two curves differ. The difference between the AIMs increases for WI
(panel b), a different trend is observed for the case AR1 where the and SF. In this latter case, the turbulence structures observed between
curve follows the 2D contraction limit. It is worth noting that the the two cylinders and behind the isolated cylinder are very different.
case AR2 is the only case displaying an increase in peak magnitude In particular, the AR2 case displays a stronger three-dimensionality.
(panels b and d). In the framework of turbulent structures in the urban The curve at 𝑥 = 0.5 is interesting because it follows the asymptotic
environment, the effect of the downstream obstacle is more evident limit and shows a highly two-dimensional energy distribution. Moving
when comparing the present analysis with the simulations published downstream, the turbulence becomes three-dimensional, but the peak
by Atzori et al. (2023) for two obstacles in tandem and varying gaps as of the AIM is lower than the AR2 case for all positions considered.
displayed in Fig. 9. The AIMs from Atzori et al. (2023) for the velocity
profiles in the wake of the upfront square cylinder with AR=2 in a 3.4. Reynolds stresses
two-tandem configuration are included for comparison. Specifically,
three configurations have been studied: skimming flow (SF) is shown in From the previous analysis, the wake of a wall-mounted square
panel (a), wake interference (WI) in panel (b), and isolated roughness cylinder is three-dimensional and the turbulent energy is not equally
(IR) in panel (c) at different locations from 𝑥 = 0.5 (blue lines) to distributed across the three reference directions (see Fig. 8). The tur-
𝑥 = 0.75 (orange lines) and 𝑥 = 1 (yellow lines). These curves are bulent kinetic energy 𝑘 = 1∕2(𝑢′ 𝑢′ + 𝑣′ 𝑣′ + 𝑤′ 𝑤′ ) in the longitudinal

10
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 11. Contour plots of the shear stresses 𝑢′ 𝑣′ at 𝑧 = 0 (left column) and 𝑢′ 𝑤′ at the horizontal plane 𝑦 = 0.5 (right column). The high turbulent kinetic energy 𝑘 is delimited
by the white contour corresponding to 1/3 of the maximum value in the domain for all cases.

symmetric plane (𝑧 = 0) is first plotted in Fig. 10. The regions of high components being negligible or zero. Within the wake, a third, weaker
𝑢′ 𝑢′ , 𝑣′ 𝑣′ and 𝑤′ 𝑤′ are also reported in the same figure in order to peak is observed in the upper part. This peak aligns with the peak
highlight the correlations between these quantities and the turbulence region of 𝑢′ 𝑢′ (yellow line), 𝑣′ 𝑣′ (red line), and 𝑤′ 𝑤′ (black line). For
peak 𝑘. In front of the cylinder, the region of higher turbulent kinetic AR3 and AR4, a fourth peak is situated inside the bottom half of the
energy corresponds to the footprint of the horseshoe vortex which size wake, closer to the cylinder base, attributed to the base vortices, with
and position depends on the AR. For the cases AR1 (panel a) and AR2 the primary contributors being 𝑢′ 𝑢′ and 𝑤′ 𝑤′ .
(panel b), the horseshoe vortex is larger because of the stronger adverse This difference in the turbulence energy can be attributed to the
pressure region upfront the cylinder induced by the building blockage. different wake configurations. For AR1 (panel a) and AR2 (panel b)
For increasing AR (panel c and d), the building blockage decreases, the the wake is a dipole, the dominant structure in the wake is the so-called
separation line moves towards the leeward surface of the building and arch-type vortex and the source of the turbulence is located above the
the horseshoe vortex reduces in size. For AR3 (panel c) and AR4 (panel cylinder. For the AR3 (panel c) and AR4 (panel d), the peak moves
d) the horseshoe vortex is negligible and it does not affect the wake inside the wake region where the most important turbulent structures
strength. are the hairpin vortices. The wake shrinks in the spanwise direction,
Apart from the peak of 𝑘 corresponding to the horseshoe vortex, the maximum of 𝑘 tends to occupy the region behind the obstacle and
a second peak emerges above the cylinder roof within the separa- the maximum contribution is given by 𝑤′ 𝑤′ . The streamwise normal
tion region induced by the roof edge. The primary contributor is the stresses peak above the obstacle for AR1 and AR2 where the flow
streamwise turbulence fluctuation 𝑢′ 𝑢′ (yellow line), with the other shows a high-shear region. This is probably due to the interaction

11
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 12. Production 𝑃 𝑘 (left column) and transport 𝑇 𝑘 (right column) terms of the turbulent kinetic energy budget for 𝑧 = 0. The contour lines delimit the peak region of 𝑢′ 𝑢′
(yellow) and 𝑘 (black). The cases AR1-AR4 are plotted from the top to the bottom panels. (For interpretation of the references to color in this figure legend, the reader is referred
to the web version of this article.)

between the low-momentum flow at the roof of the obstacle and the The shear stresses 𝑢′ 𝑣′ (left column) and 𝑢′ 𝑤′ (right column) for
high-momentum flow in the upper part of the boundary layer. For the cases from AR1 (top panels) to AR4 (bottom panels) are reported
higher aspect ratios, cases AR3 and AR4, the regions of high 𝑢′ 𝑢′ above in Fig. 11. The white contour delimits the region of high TKE. The
the obstacle are reduced, i.e. the wake is affected by a very strong shear-stress 𝑢′ 𝑣′ is very negative in the upper part of the wake. This
downwash due to the tip vortices which transports the high-energy proves the tendency of the tip vortices to transport downwards the high
flow in the outer part of the boundary layer towards to the low- momentum flow behind the obstacle. The positive 𝑢′ 𝑣′ is displayed at
momentum flow in the wake. This creates a region in the wake where the base of the cylinder and is less intense than the negative region.
the streamwise momentum 𝑢′ 𝑢′ is maximum. Inside the wake, 𝑣′ 𝑣′ is Thus, the flow is driven towards the wake center but the intensity is
weak and negligible for the case AR1 while it becomes important for quite weak because of the wall. The positive region is stronger for
higher aspect ratios. The component 𝑤′ 𝑤′ is equally important inside AR4 than AR1 because of the effect of the base vortices that drive
the wake across all the cases considered in this paper. However, the the flow from the base away from the wall. Similarly, the horizontal
only difference observed between the cases AR1 (low aspect ratio) and stress 𝑢′ 𝑤′ displays one negative (left-hand side of the obstacle) and
AR4 (high aspect ratio) consists of the extended of the peak region. one positive (right-hand side of the obstacle) region that transport the
This region is larger for AR1 because of the wider wake caused by the high-energetic flow towards the wake center. The strongest transport
larger cylinder width, while for AR4 the normal stress 𝑤′ 𝑤′ displays a term has been predicted for the case AR4. Although the second-order
secondary peak above the obstacle. statistics have been carried out with a sample of at least 150 time units,

12
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 13. Profiles of the TKE budget for the cases AR1 (a), AR2 (b), AR3 (c) and AR4 (d) at 𝑥 = −1. In particular, the figure displays the production term 𝑃 𝑘 (green), transport
term 𝑇 𝑘 (black), viscous diffusion term 𝐷𝑘 (magenta) and convective term 𝐶 𝑘 (blue). (For interpretation of the references to color in this figure legend, the reader is referred to
the web version of this article.)

the 𝑢′ 𝑤′ lacks of symmetry. The time sample is otherwise enough for the 𝑢′ 𝑢′ (yellow line) and 𝑘 (black line) peak regions are also shown.
the statistical convergence of the TKE budget (third-order statistical From the figure, both 𝑃 𝑘 and 𝑇 𝑘 are negligible in the free stream and
moments). However, the high computational cost to significantly in- assume a nonzero value in proximity to the obstacle and in its wake.
crease the statistical convergence for all cases is not justified by the In particular, the production term exhibits four strong regions, their
purposes of the present paper. size and intensity depend on the aspect ratio (AR): (i) at the roof of
the cylinder in the flow separation region, (ii) at the upstream edge
3.5. TKE budget of the roof, (iii) upstream of the obstacle in correspondence with the
horseshoe vortex, and (iv) at the flank of the cylinders (see Fig. 14).
The turbulent kinetic energy (TKE) budget is reported to display the For the case AR1 (panel a), the peak region of 𝑃 𝑘 extends to the roof of
effect of AR on the regions of high turbulence production, transport and the cylinder but rapidly degrades in the wake, where it assumes a low
diffusion. Note that the transport equation of the Reynolds-stress tensor positive value. For increasing aspect ratio (left column), the positive
is: production term is distorted, and it is deflected downwards in the wake
𝜕 𝑢𝑗 𝜕 𝑢′ 𝜕 𝑢′𝑗 as a side effect of the tip vortices. Meanwhile, the horseshoe moves
𝜕 ′ ′ 𝜕𝑢 𝜕2 ′ ′ 𝜕 ′ ′ ′
𝑢𝑖 𝑢𝑗 = − 𝑢′𝑖 𝑢′𝑘 − 𝑢′𝑗 𝑢′𝑘 𝑖 − 2𝜈 𝑖 +𝜈 𝑢𝑢 − 𝑢𝑢𝑢 + towards the upfront face of the cylinder, and its contribution to 𝑃 𝑘 is
𝜕𝑡 𝜕 𝑥𝑘 𝜕 𝑥𝑘 𝜕 𝑥𝑘 𝜕 𝑥𝑘 𝜕 𝑥2𝑘 𝑖 𝑗 𝜕 𝑥𝑘 𝑖 𝑗 𝑘 confined to a very small region at the wall. For the cases AR3 (panel e)
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏟⏞⏞⏞⏟ ⏟⏞⏞⏟⏞⏞⏟ ⏟⏞⏞⏞⏟⏞⏞⏞⏟
𝑃𝑖𝑗 𝜖𝑖𝑗 𝐷𝑖𝑗 𝑇𝑖𝑗 and AR4 (panel g), a weak positive region occurs at the bottom wall due
to the base vortices. As observed in Fig. 4, the upper portion of the wake
⎛ ′ ⎞
𝜕 𝑢′𝑗 ( )
𝜕 ′ ′ 1 𝜕𝑢 ⎟− 1 𝜕 𝜕 captures the high-energetic flow of the outer boundary layer that is
+ 𝑢𝑘 𝑢𝑖 𝑢𝑗 + − ⎜𝑝 𝑖 + 𝑝 𝑝𝑢′𝑖 + 𝑝𝑢′𝑗 , (9)
𝜕 𝑥𝑘 𝜌 ⎜ 𝜕 𝑥𝑗 𝜕 𝑥𝑖 ⎟ 𝜌 𝜕 𝑥𝑗 𝜕 𝑥𝑖 transported to the center of the wake. Similarly, at the bottom surface,
⏟⏞⏞⏞⏟⏞⏞⏞⏟ ⎝ ⎠ ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞
⏟
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ due to the occurrence of the base vortices for the cases AR3 and AR4,
𝐶 𝑖𝑗 𝑡
𝛱𝑖𝑗
𝛱𝑖𝑗𝑠 the flow is pushed upward towards the wake core. As a consequence,
the wake becomes the major region of turbulence production. Note
where 𝑃𝑖𝑗 is the production term, 𝜖𝑖𝑗 is the pseudo dissipation term,
that 𝑃 𝑘 is slightly altered and is negative in front of the cylinder and
𝐷𝑖𝑗 refers to the viscous diffusion, 𝑇𝑖𝑗 refers to the turbulent transport
above the roof of the obstacles. It is worth noting that the maximum
while 𝐶𝑖𝑗 is the convection term. The pressure strain and the pressure
of 𝑃 𝑘 broadly coincides with the peak region of 𝑢′ 𝑢′ (yellow line). The
transport are denoted b 𝛱𝑖𝑗𝑠 and 𝛱𝑖𝑗𝑡 . Given the definition of the
( ) transport term 𝑇 𝑘 is also shown in Fig. 12, and it exhibits a negative–
1 ′ ′
turbulent kinetic energy (TKE) as 𝑘 = 𝑢 𝑢 + 𝑣′ 𝑣′ + 𝑤′ 𝑤′ , the TKE positive pattern above the obstacles while it is negligible in the wake.
2
equation becomes: Comparing the left panels, the increasing aspect ratio (AR) produces a
𝜕𝑘
= 0 = 𝑃 𝑘 + 𝜖 𝑘 + 𝐷𝑘 + 𝑇 𝑘 − 𝛱 𝑘 − 𝐶 𝑘 , (10) deformation of the region of negative 𝑇 𝑘 that extends into the wake
𝜕𝑡 behind the obstacle. The horseshoe vortex is barely visible only for the
where each term is given by the trace of the corresponding tensor on the case AR1, while the transport of turbulence induced by the horseshoe
right-hand side of Eq. (9). The production 𝑃 𝑘 and transport 𝑇 𝑘 terms vortex in front of the obstacle is negligible for the remaining cases. For
are shown in Fig. 12 at the vertical symmetry plane for AR value from a deeper understanding of the AR effect we report the profiles of the
1 (top panels) to 4 (bottom panels). The remaining terms of the TKE TKE budget at 𝑥 = −1 in Fig. 13 for all the cases. At this downstream
budget are not reported for brevity of discussion. The contour lines of location, the peaks of the production 𝑃 𝑘 and the convective 𝐶 𝑘 terms

13
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Fig. 14. Production 𝑃 𝑘 (left column) and transport 𝑇 𝑘 (right column) terms of the turbulent kinetic energy budget for 𝑦 = 0.5. The contour lines delimit the peak region of 𝑢′ 𝑢′
(yellow) and 𝑘 (black). The cases AR1-AR4 are plotted from the upper to the bottom panels. (For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)

for the case AR1 (panel a) are situated slightly above the obstacle. AR4 (panel g), the positive regions of 𝑃 𝑘 are deflected from the flank
However, for increasing AR, the peak of the turbulence production towards the wake center. Similarly, the term 𝑇 𝑘 has two peak regions
moves downwards and it becomes slightly stronger. A secondary peak at the flank of the obstacles, while for increasing aspect ratio, the
occurs for the cases AR3 (panel c) and AR4 (panel d), balanced by wake displays a negative transport term. The present findings indicate
the transport term 𝑇 𝑘 . This quantitatively proves the importance of that the turbulence production is associated with the streamwise shear
the turbulence production within the wake for larger AR. It is also stress regardless the aspect ratio, suggesting that, from the trajectory
worth noting that, for the case AR1 (panel a) the viscous diffusion 𝐷𝑘 optimization perspective, it is convenient to avoid these regions. The
is slightly negative all along 𝑦 direction. For the remaining cases, 𝐷𝑘 case AR1 is the only one that does not display positive production
is constant and negative, that is more prominent and playing a more in the wake because it is the only case where the flow reattaches at
significant role than the convective phenomenon. the roof, the tip vortices are less intense, and their influence on the
The 𝑃 𝑘 and 𝑇 𝑘 in the horizontal plane at 𝑧 = 0.5 are finally reported wake is weak. On the contrary, for the cases AR3 and AR4, the tip and
in Fig. 14. The figure displays a region of high turbulence production base vortices are stronger and they mainly impact on the turbulence
at the flank of the obstacles. The turbulence produced within this generation in the wake. From a physical standpoint, the examination
region is then transported inside the wake by the transport term that of the TKE budget corroborates the idea that the turbulence is initially
shows a positive/negative pattern. For the cases AR2 (panel c) to generated in the flow-separation region and is subsequently transported

14
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

into the wake. Here, both the tip and base vortices are strongly related CRediT authorship contribution statement
to the production of turbulence.
Gerardo Zampino: Writing – original draft, Resources, Methodol-
ogy, Investigation, Data curation. Marco Atzori: Formal analysis, Data
4. Conclusions
curation, Conceptualization. Elias Zea: Writing – review & editing,
Supervision. Evelyn Otero: Writing – review & editing, Visualization.
The present study investigates the influence of aspect ratio (AR) Ricardo Vinuesa: Supervision, Project administration, Investigation,
on the wake structure of a wall-mounted square cylinder immersed Funding acquisition, Conceptualization.
in a turbulent boundary layer with Reℎ = 10,000, employing high-
resolution LES simulations with the open-source software Nek5000. Declaration of competing interest
Four configurations are studied at different AR values (1, 2, 3, and
4) to gain insights into the turbulence wake behavior. The change in The authors declare the following financial interests/personal rela-
AR has been achieved by altering the width of the square cylinder to tionships which may be considered as potential competing interests:
maintain the ratio of the incoming boundary-layer thickness and the Gerardo Zampino reports financial support was provided by KTH Royal
obstacle height constant. This approach differs from Yauwenas et al. Institute of Technology. If there are other authors, they declare that
(2019), in which the change in the AR is achieved by changing the they have no known competing financial interests or personal relation-
cylinder height. However, the change in the cylinder height causes ships that could have appeared to influence the work reported in this
a different degree of interaction between the incoming BL and the paper.
obstacle, and consequently, at similar AR values, the time-averaged
turbulent structures can be significantly different. This effect is not Acknowledgments
investigated in the present paper.
Our findings align with the empirical interpretation by Wang and This Project has received funding from the European Union’s HORI-
Zhou (2009) regarding the roles of tip, base, and spanwise vortices ZON Research and Innovation Programme, project REFMAP, under
in the wake. A key contribution of this work is the identification of Grant Agreement number 101096698. The computations were car-
a critical aspect ratio ARc between 2 and 3, where the wake regime ried out at the supercomputer Dardel at PDC, KTH, and the com-
changes from a dipole to a quadrupole configuration. puter time was provided by the National Academic Infrastructure for
In this study, we introduce a novel framework to describe the Supercomputing in Sweden (NAISS).
dipole-to-quadrupole transition in terms of Re𝜃 and ARc , which incor-
porates the effects of both boundary-layer thickness and turbulence Data availability
intensity, which has not been formally described in prior research. This
framework combines the present results with data from the literature All the data will be available in the following open-access repository
and delineates the regions corresponding to dipole, quadrupole, and after the article is published:
transitional wake configurations as a function of Re𝜃 , addressing a https://www.vinuesalab.com/databases/.
gap in previous research that focused mainly on the boundary-layer
thickness alone. While this study does not investigate the effect of
the cross-sectional ratio (CR), it highlights the importance of cylinder References
dimensions in modulating the wake configuration. We thus suggest that
Atzori, M., Torres, P., Vidal, A., Le Clainche, S., Hoyas, S., Vinuesa, R., 2023. High-
the transition could be related to the 𝛿99 ℎ∕(𝑏𝑑) ratio, despite further resolution simulations of a turbulent boundary layer impacting two obstacles in
data collection is still necessary. This theoretical dissertation opens up tandem. Phys. Rev. Fluids 8, 063801.
new directions for future research into the combined effects of cylinder Banerjee, S., Krahl, R., F., D., Zenger, C., 2007. Presentation of anisotropy properties
height and width on the wake dynamics. of turbulence, invariants versus eigenvalue approaches. J. Turb. 8, N32.
Bourgeois, J.A., Sattari, P., Martinuzzi, R.J., 2011. Alternating half-loop shedding in the
Additionally, we offer new insights into turbulence anisotropy and turbulent wake of a finite surface-mounted square cylinder with a thin boundary
energy dynamics within the wake. The Reynolds stresses reveal the layer. Phys. Fluids 23, 095101.
anisotropic nature of turbulence, particularly near the wake edges, Bourgeois, J.A., Sattari, P., Martinuzzi, R.J., 2012. Coherent vortical and straining
structures in the finite wall-mounted square cylinder wake. Int. J. Heat Fluid Flow
where strong three-dimensionality is observed. This is confirmed by
35, 130–140.
anisotropy-invariant maps (AIM), which indicate that the outer wake Cao, Y., Tamura, T., Zhou, D., Bao, Y., Han, Z., 2022. Topological description of near-
region consistently exhibits high turbulence intensity and isotropic wall flows around a surface-mounted square cylinder at high Reynolds numbers.
characteristics. Analysis of the TKE budget further demonstrates that J. Fluid Mech. 933, A39.
Chen, G., Li, X.-B., Sun, B., Liang, X.-F., 2022. Effect of incoming boundary layer
turbulence production and transport are especially pronounced near the
thickness on the flow dynamics of a square finite wall-mounted cylinder. Phys.
roof and wake regions for AR3 and AR4 configurations, emphasizing Fluids 34, 015105.
the contributions of tip and base vortices to turbulence generation. Dong, S., Karniadakis, G., Chryssostomidis, C., 2014. A robust and accurate outflow
These findings underscore the role of AR in modulating TKE distribu- boundary condition for incompressible flow simulations on severely-truncated
unbounded domains. J. Comput. Phys. 261, 83–105.
tion, notably in enhancing turbulence production and transport in the
Dousset, V., Pothérat, A., 2010. Formation mechanism of hairpin vortices in the wake
wake. of a truncated square cylinder in a duct. J. Fluid Mech. 653, 519–536.
Extending the simulations to higher Reynolds numbers, closer to Eitel-Amor, G., Örlü, R., Schlatter, P., 2014. Simulation and validation of a spatially
realistic urban conditions, could yield a more comprehensive char- evolving turbulent boundary layer up to Re𝜃=8300. Int. J. Heat Fluid Flow 47,
57–69.
acterization of urban-flow turbulence but would demand significant
El Hassan, M., Bourgeois, J., Martinuzzi, R., 2015. Boundary layer effect on the vortex
computational resources and make it unfeasible to maintain the same shedding of wall-mounted rectangular cylinder. Exp. Fluids 56, 33.
accuracy level. Thus, the high-resolution LES setup becomes imprac- Fischer, P., Lottes, J., Kerkemeier, S., 2008. NEK5000: Open source spectral element
tical. Unanswered questions, involving the exploration of additional CFD solver. https://nek5000.mcs.anl.gov/.
Goswami, S., Hemmati, A., 2022. Mechanisms of wake asymmetry and secondary
geometrical parameters such as cross-section aspect ratio (CR) or the
structures behind low aspect-ratio wall-mounted prisms. J. Fluid Mech. 950, A31.
boundary-layer thickness, and the correlation between the TKE budget Hosseini, Z., Bourgeois, J.A., Martinuzzi, R.J., 2013. Large-scale structures in dipole
and the trajectory optimization of drones will be addressed in future and quadrupole wakes of a wall-mounted finite rectangular cylinder. Exp. Fluids
works. 54, 1595.

15
G. Zampino et al. International Journal of Heat and Fluid Flow 112 (2025) 109672

Hosseini, S., Vinuesa, R., Schlatter, P., Hanifi, A., Henningson, D., 2016. Direct Sattari, P., Bourgeois, J.A., Martinuzzi, R.J., 2012. On the vortex dynamics in the wake
numerical simulation of the flow around a wing section at moderate Reynolds of a finite surface-mounted square cylinder. Exp. Fluids 52 (5), 1149–1167.
number. Int. J. Heat Fluid Flow 61, 117–128. Schlatter, P., Örlü, R., 2012. Turbulent boundary layers at moderate Reynolds numbers:
Hussein, H.J., Martinuzzi, R.J., 1996. Energy balance for turbulent flow around a inflow length and tripping effects. J. Fluid Mech. 710, 5–34.
surface mounted cube placed in a channel. Phys. Fluids 8, 764–780. Simonsen, A.J., Krogstad, P., 2005. Turbulent stress invariant analysis : Clarification of
Jeong, J., Hussain, F., 1995. On the identification of a vortex. J. Fluid Mech. 285, existing terminology. Phys. Fluids 17, 088103.
69–94. Sumner, D., Heseltine, J., Dansereau, O., 2004. Wake structure of a finite circular
Kawamura, T., Hiwada, M., Hibino, T., Mabuchi, I., Kumada, M., 1984. Flow around cylinder of small aspect ratio. Exp. Fluids 37, 720–730.
a finite circular cylinder on a flat plate : Cylinder height greater than turbulent Uffinger, T., Ali, I., Becker, S., 2013. Experimental and numerical investigation of the
boundary layer thickness. Bull. JSME 27 (232), 2142–2151. flow around three different wall-mounted cylinder geometries of finite length. J.
Kumar, P., Tiwari, S., 2019. Effect of incoming shear on unsteady wake in flow past Wind. Eng. Ind. Aerodyn. 119, 13–17.
surface mounted polygonal prism. Phys. Fluids 31, 113607. Vinuesa, R., Bobke, A., Örlü, R., Schlatter, P., 2016. On determining characteristic
Lumley, J.L., Newman, G., 1977. The return to isotropy of homogeneous turbulence. length scales in pressure-gradient turbulent boundary layers. Phys. Fluids 28 (5),
J. Fluid Mech. 82, 161. 055101.
Negi, P., Vinuesa, R., Hanifi, A., Schlatter, P., Henningson, D., 2018. Unsteady Vinuesa, R., Schlatter, P., Malm, J., Mavriplis, C., Henningson, D.S., 2015. Direct
aerodynamic effects in small-amplitude pitch oscillations of an airfoil. Int. J. Heat numerical simulation of the flow around a wall-mounted square cylinder under
Fluid Flow 71, 378–391. various inflow conditions. J. Turb. 16, 555–587.
Patera, A.T., 1984. A spectral element method for fluid dynamics: Laminar flow in a Wang, H.F., Zhou, Y., 2009. The finite-length square cylinder near wake. J. Fluid Mech.
channel expansion. J. Comput. Phys. 54, 468. 638, 453–490.
Porteous, R., Moreau, D.J., Doolan, C.J., 2014. A review of flow-induced noise from Wang, H.F., Zhou, Y., Chan, C.K., Lam, K.S., 2006. Effect of initial conditions on
finite wall-mounted cylinders. J. Fluids Struct. 51, 240–254. interaction between a boundary layer and a wall-mounted finite-length-cylinder
Rastan, M., Shahbazi, H., Sohankar, A., Md., M.A., Zhou, Y., 2021. The wake of a wall- wake. Phys. Fluids 18, 065106.
mounted rectangular cylinder: Cross-sectional aspect ratio effect. J. Wind. Eng. Ind. Yauwenas, Y., Porteous, R., Moreau, D.J., Doolan, C.J., 2019. The effect of aspect ratio
Aerodyn. 213, 104615. on the wake structure of finite wall-mounted square cylinders. J. Fluid Mech. 875,
Rastan, M.R., Sohankar, A., Alam, M.M., 2017. Low-Reynolds-number flow around a 929–960.
wall-mounted square cylinder: Flow structures and onset of vortex shedding. Phys. Zargar, A., Tarokh, A., Hemmati, A., 2022. The unsteady wake transition behind a
Fluids 29, 103601. wall-mounted large-depth-ratio prism. J. Fluid Mech. 952, A12.
Rastan, M.R., Sohankar, A., Doolan, C., Moreau, D., Shirani, E., Alam, M.M., 2019. Zhang, D., Cheng, L., An, H., Zhao, M., 2017. Direct numerical simulation of flow
Controlled flow over a finite square cylinder using suction and blowing. Int. J. around a surface-mounted finite square cylinder at low Reynolds numbers. Phys.
Mech. Sci. 156, 410–434. Fluids 29 (4), 045101.
Saeedi, M., LePoudre, P.P., Wang, B.C., 2014. Direct numerical simulation of turbulent Zhao, M., Mamoon, A., Wu, H., 2021. Numerical study of the flow past two wall-
wake behind a surface-mounted square cylinder. J. Fluids Struct. 51, 20–39. mounted finite-length square cylinders in tandem arrangement. Phys. Fluids 33
Sakamoto, H., Arie, M., 1983. Vortex shedding from a rectangular prism and a circular (9), 093603.
cylinder placed vertically in a turbulent boundary layer. J. Fluid Mech. 126,
147–165.

16