1. Introduction
Understanding coherent structures in wall-bounded turbulent flows has attracted considerable attention over the past half-century. This is because such structures govern the flow dynamics with a certain degree of temporal and spatial coherence, thereby providing valuable insight into the multiscale nature of turbulence (Robinson Reference Robinson1991; Adrian Reference Adrian2007; Jiménez Reference Jiménez2018). Over the past few decades, uniform momentum zones (UMZs) – which are regions of relatively uniform streamwise momentum – have received growing attention, following their initial observation by Meinhart & Adrian (Reference Meinhart and Adrian1995). A remarkable feature of UMZs is that multiple zones appear simultaneously, spanning from the viscous sublayer to the outer region of the flow. They are separated by thin shear layers that exhibit high streamwise velocity gradients, across which a sharp velocity jump occurs. The presence of multiple UMZs is associated with key structural features in wall-bounded turbulence, including hairpin packets (Adrian, Meinhart & Tomkins Reference Adrian, Meinhart and Tomkins2000), vortical fissures (Priyadarshana et al. Reference Priyadarshana, Klewicki, Treat and Foss2007), vortex clusters (del Álamo et al. Reference del Álamo, Jiménez, Zandonade and Moser2006), ejection–sweep pairs (Lozano-Durán et al. Reference Lozano-Durán, Flores and Jiménez2012) and clusters of streamwise velocity fluctuations (Hwang & Sung Reference Hwang and Sung2018). Moreover, the sharp velocity jumps at UMZ edges have been exploited in recent studies to stochastically reconstruct wall-normal profiles of turbulence statistics (de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017; Bautista et al. Reference Bautista, Ebadi, White, Chini and Klewicki2019; Ehsani et al. Reference Ehsani, Heisel, Li, Voller, Hong and Guala2024). These observations underscore the importance of analysing UMZs to gain deeper insight into the structural organisation and statistical behaviour of wall turbulence.
The emergence of multiple UMZs aligns with the hierarchical nature of coherent structures. This zonal-like structural arrangement is not merely a coexistence of several UMZs, but rather a configuration in which distinct streamwise momentum zones are layered in the wall-normal direction, with smaller zones enclosed within larger ones (Adrian et al. Reference Adrian, Meinhart and Tomkins2000; Adrian Reference Adrian2007). As a result, the streamwise velocity deficit increases progressively toward the wall, forming a multiscale momentum structure. Adrian et al. (Reference Adrian, Meinhart and Tomkins2000) described such organisation as a nested hierarchy, explaining that it arises from the superposition of hairpin packets with different convection velocities. Such zonal-like arrangements give rise to discrete velocity jumps in instantaneous flow fields that, upon temporal averaging, contribute to the formation of the logarithmic layer in the mean streamwise velocity profile (Meinhart & Adrian Reference Meinhart and Adrian1995; de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017). This highlights the relevance of UMZs in understanding the coherent structures in the logarithmic layer of wall-bounded turbulence. Moreover, this nested hierarchy of UMZs closely aligns with the self-similar organisation of eddies described in Townsend’s (Reference Townsend1976) attached-eddy hypothesis (AEH). The multiple UMZs can also be interpreted in the context of the mean momentum equation. In particular, the hierarchical distribution and the logarithmic law of the mean velocity can be analytically derived from an invariant form of the mean momentum balance in the inertial sublayer (Fife, Klewicki & Wei Reference Fife, Klewicki and Wei2009; Klewicki et al. Reference Klewicki, Philip, Marusic, Chauhan and Morrill-Winter2014; Klewicki Reference Klewicki2021). Based on this framework, Bautista et al. (Reference Bautista, Ebadi, White, Chini and Klewicki2019) modelled the spatial organisation of UMZs and demonstrated that key turbulence statistics can be well reproduced. This highlights the importance of interpreting UMZs within a hierarchical framework, as it provides a structural basis for understanding self-similar behaviour in wall-bounded turbulence.
The AEH provides a conceptual basis for understanding the multiscale structure of the logarithmic layer, where eddies are assumed to be self-similar. In this framework, each eddy has a characteristic size proportional to its distance from the wall (
$y$
) and a velocity scale given by the friction velocity (
$u_\tau$
). Hereafter, we refer to scaling with
$y$
and
$u_\tau$
as wall scaling. Through the superposition of such eddies across a wide range of scales, the AEH predicts asymptotic behaviours of turbulence statistics – such as the logarithmic variation of streamwise and spanwise turbulence intensity, the constancy of the wall-normal turbulence intensity (Townsend Reference Townsend1976), and the associated spectral behaviours (Perry & Abell Reference Perry and Abell1977; Perry & Chong Reference Perry and Chong1982; Perry, Henbest & Chong Reference Perry, Henbest and Chong1986). Accordingly, considerable efforts have been devoted to validating these statistical predictions (Nickels et al. Reference Nickels, Marusic, Hafez and Chong2005; Hultmark et al. Reference Hultmark, Vallikivi, Bailey and Smits2012; Ahn et al. Reference Ahn, Lee, Jang and Sung2013; Marusic et al. Reference Marusic, Monty, Hultmark and Smits2013; Sillero, Jiménez & Moser Reference Sillero, Jiménez and Moser2013; Lee & Moser Reference Lee and Moser2015; Vallikivi, Ganapathisubramani & Smits Reference Vallikivi, Ganapathisubramani and Smits2015; Örlü et al. Reference Örlü, Fiorini, Segalini, Bellani, Talamelli and Alfredsson2017; Samie et al. Reference Samie, Marusic, Hutchins, Fu, Fan, Hultmark and Smits2018; Pirozzoli et al. Reference Pirozzoli, Romero, Fatica, Verzicco and Orlandi2021; Ono, Furuichi & Tsuji Reference Ono, Furuichi and Tsuji2023), and – most importantly – to identifying coherent structures that are consistent with the AEH (Perry & Marusic Reference Perry and Marusic1995; del Álamo et al. Reference del Álamo, Jiménez, Zandonade and Moser2006; Lozano-Durán et al. Reference Lozano-Durán, Flores and Jiménez2012; Hwang Reference Hwang2015; Hellström et al. Reference Hellström, Marusic and Smits2016; Baars, Hutchins & Marusic Reference Baars, Hutchins and Marusic2017; Hwang & Sung Reference Hwang and Sung2018, Reference Hwang and Sung2019; Baars & Marusic Reference Baars and Marusic2020; Hwang, Lee & Sung Reference Hwang, Lee and Sung2020; Yoon et al. Reference Yoon, Hwang, Yang and Sung2020; Deshpande, Monty & Marusic Reference Deshpande, Monty and Marusic2021; Jae, Sung & Hwang Reference Jae, Sung and Hwang2025); see also the recent review by Marusic & Monty (Reference Marusic and Monty2019).
As mentioned earlier, the presence of multiple UMZs, often forming a nested hierarchy, is closely related to the hierarchical organisation of eddies in the AEH framework. de Silva, Hutchins & Marusic (Reference de Silva, Hutchins and Marusic2016) observed a log–linear increase in the number of UMZs (
$N_{\textit{UMZ}}$
) with increasing friction Reynolds number (
$\textit{Re}_\tau = u_\tau \delta / \nu$
, where
$\delta$
is the outer length scale and
$\nu$
is the kinematic viscosity) in turbulent boundary layers (TBLs) at
$\textit{Re}_\tau \approx 10^3$
–
$10^4$
. This log–linear trend was also reported in synthetic flow fields generated using the attached-eddy model of Marusic (Reference Marusic2001), supporting the interpretation that UMZs exhibit a hierarchical distribution, reminiscent of the AEH. In addition, such a log–linear increase is consistent with the geometrical progression of the hierarchy layers in the inertial sublayer (Klewicki et al. Reference Klewicki, Philip, Marusic, Chauhan and Morrill-Winter2014). Hwang & Sung (Reference Hwang and Sung2018) provided more direct evidence linking UMZs to the AEH by identifying wall-attached structures of streamwise velocity fluctuations (
$u$
) in instantaneous flow fields of a TBL at moderate
$\textit{Re}_\tau \approx 1000$
. These structures were found to be composed of multiple UMZs, and
$N_{\textit{UMZ}}$
within each structure was observed to increase in a log–linear fashion with the structure’s height. Furthermore, the authors demonstrated that the superposition of these wall-attached structures reproduces the logarithmic variation of the streamwise turbulence intensity, consistent with the predictions of the AEH. Subsequently, Hwang & Sung (Reference Hwang and Sung2019) showed that the mean streamwise velocity profile reconstructed from the superposition of wall-attached
$u$
structures exhibits the logarithmic law in turbulent pipe flow (
$\textit{Re}_\tau \approx 1000$
and
$3000$
) even when the total mean velocity profile does not exhibit a clear logarithmic layer. This behaviour may be attributed to the fact that the velocity jumps at UMZ edges scale with the friction velocity
$u_\tau$
(de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017). If this scaling holds and
$N_{\textit{UMZ}}$
is sufficiently large (i.e. locally high Reynolds numbers), the influence of other structures (such as non-self-similar ones) becomes less significant, and the logarithmic law may emerge as a statistical outcome of the UMZ ensemble. Indeed, in high-
$\textit{Re}_\tau$
(
$\sim O(10^4)$
) TBLs (both smooth and rough walls), the logarithmic mean velocity profile was observed through the superposition of UMZs (Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020). These findings suggest that UMZs share structural features with the AEH and could provide a valuable insight into the asymptotic behaviour of turbulence statistics in the logarithmic layer. However, it remains to be examined whether the organisation of UMZs also accounts for other statistical behaviours predicted by the AEH, such as the logarithmic variation of spanwise turbulence intensity and the constancy of wall-normal turbulence intensity.
Although sharp changes in the wall-normal velocity and Reynolds shear stress can occur across UMZ edges, relatively little attention has been given to examining whether the asymptotic behaviours of the wall-normal and spanwise turbulence intensities predicted by the AEH can also be understood within the framework of UMZs. In turbulent channel flows, Kwon et al. (Reference Kwon, Philip, de Silva, Hutchins and Monty2014) observed that, across the boundary of the core region, which corresponds to the UMZ with the highest streamwise momentum, there exist a distinct streamwise velocity jump and a rapid decrease in the Reynolds shear stress. This behaviour is qualitatively similar to that observed at the turbulent/non-turbulent interface (TNTI). Similarly, Eisma et al. (Reference Eisma, Westerweel, Ooms and Elsinga2015) reported the presence of multiple internal shear layers within the turbulent region of a boundary layer, where sharp interfacial changes occur. These internal layers closely resemble UMZ edges in that they are also associated with localised streamwise velocity jumps (de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017; Gul, Elsinga & Westerweel Reference Gul, Elsinga and Westerweel2020). In addition, Laskari et al. (Reference Laskari, de Kat, Hearst and Ganapathisubramani2018) examined the wall-normal velocity and Reynolds shear stress profiles as a function of
$N_{\textit{UMZ}}$
, and found that the magnitudes of both quantities increase with increasing
$N_{\textit{UMZ}}$
, although no clear asymptotic trends were observed. In adverse-pressure-gradient TBLs, Thavamani et al. (Reference Thavamani, Cuvier, Willert, Foucaut, Atkinson and Soria2020) also performed conditional averaging of the streamwise turbulence intensity and Reynolds shear stress based on
$N_{\textit{UMZ}}$
, but similarly did not detect the asymptotic behaviours. Despite the log–linear increase of
$N_{\textit{UMZ}}$
with
$\textit{Re}_\tau$
(de Silva et al. Reference de Silva, Hutchins and Marusic2016), which implies that regions with high
$N_{\textit{UMZ}}$
are representative of locally high-Reynolds-number flows (Hwang & Sung Reference Hwang and Sung2019), the expected asymptotic behaviours (i.e. the logarithmic variation of the wall-parallel turbulence intensities and the approximately constant wall-normal turbulence intensity predicted by the AEH) have not yet been consistently observed. Moreover, while the hairpin packet paradigm (Adrian et al. Reference Adrian, Meinhart and Tomkins2000) suggests that multiple UMZs can accommodate all velocity components, the absence of such asymptotic behaviours in previous studies may be attributed to the fact that not all UMZs within the logarithmic layer scale with
$y$
and
$u_\tau$
(i.e. wall scaling). This raises the question of which UMZs contribute to the constant wall-normal turbulence intensity or the logarithmic variation of the spanwise turbulence intensity predicted by the AEH, and how such wall-scaled UMZs can be systematically identified.
Despite the growing interest in understanding the role of UMZs in the context of the AEH, most studies have primarily focused on TBLs, with relatively few investigations addressing their characteristics in fully developed pipe flows. Yang, Hwang & Sung (Reference Yang, Hwang and Sung2019) analysed the core region in turbulent pipe flow, corresponding to the highest-momentum UMZ observed in channel flows (Kwon et al. Reference Kwon, Philip, de Silva, Hutchins and Monty2014), and examined the influence of wall-attached structures of
$u$
on the turbulence statistics across the core boundary. However, the internal organisation of multiple UMZs within the pipe flow was not explored. Chen, Chung & Wan (Reference Chen, Chung and Wan2020) identified the existence of multiple UMZs in pipe flow and discussed their relation to ejection and sweep motions. Recently, Gunaratne et al. (Reference Gunaratne, Guerrero, Lambert and Chin2025) observed a zonal-like arrangement of UMZs in accelerating turbulent pipe flows. Nevertheless, these studies were limited to a relatively low Reynolds number (
$\textit{Re}_\tau = 550$
), which makes it difficult to examine hierarchical distributions or the asymptotic behaviours. Gul et al. (Reference Gul, Elsinga and Westerweel2020) further investigated the connection between internal shear layers and UMZ edges in pipe flow at
$\textit{Re}_\tau = 340$
–
$1200$
, providing additional evidence for the structural relevance of UMZs. Nonetheless, the Reynolds number remained modest, and the analysis was not performed within the context of the AEH. These limitations highlight the need for a more systematic investigation of UMZs in turbulent pipe flows, especially at higher
$\textit{Re}_\tau$
where hierarchical organisation of UMZs and associated asymptotic behaviours may become more apparent.
Therefore, the objective of the present study is to extract wall-scaled UMZs from pipe flow over a wide Reynolds-number range and to explore all three components of the velocity fluctuations within them in the context of the AEH. We analyse a direct numerical simulation (DNS) database of turbulent pipe flows at
$\textit{Re}_\tau \approx 550$
,
$1000$
,
$3000$
and
$6000$
. By examining the probability density functions (PDFs) of the instantaneous streamwise velocity, we identify UMZs and their edges. We first investigate whether
$N_{\textit{UMZ}}$
increases logarithmically with Reynolds number, as previously observed in TBLs, to assess the existence of hierarchical organisation. Next, we examine the scaling behaviour of the UMZ thickness and the corresponding velocity jump across their edges to extract wall-scaled UMZs. In addition, we reconstruct the mean velocity and the turbulence intensities from the instantaneous velocity components carried by the wall-scaled UMZs to examine the asymptotic behaviours of the turbulence statistics predicted by the AEH. Finally, we assess the self-similarity of the velocity-fluctuation PDFs within the wall-scaled UMZs, in terms of both Reynolds-number independence and wall-normal invariance across the logarithmic region.
The paper is organised as follows. Section 2 describes the numerical details of the present DNS database. The detection methodology for identifying UMZs and characterising their geometrical properties is presented in § 3. In § 4, we examine the Reynolds-number dependence of the UMZ statistics, including the number of UMZs, modal velocities and shear-layer velocities, to assess their hierarchical organisation. The wall-scaled UMZs are identified in § 5, where their thickness is characterised by the wall-normal distance and their velocity jump is found to scale with the friction velocity across
$\textit{Re}_\tau$
. Subsequently, § 6 examines whether the mean velocity and the wall-parallel turbulence intensities obtained by conditional averaging over these wall-scaled UMZs exhibit the logarithmic behaviour predicted by the AEH. Moreover, the wall-normal turbulence intensity and the Reynolds shear stress profiles are analysed. In § 7, we analyse the PDFs of velocity fluctuations within the wall-scaled UMZs to assess the wall-normal invariance. Finally, the summary and conclusions of the present study are provided in § 8.
Throughout this paper, the term ‘geometrically self-similar’ is used to denote structures whose characteristic length scale varies proportionally with the wall-normal distance. The term ‘self-similar’ is used in a statistical sense, referring to the collapse of a given quantity when appropriately normalised. The term ‘wall-scaled’ is used to describe UMZs that are geometrically self-similar and whose velocity scale is given by the friction velocity. When describing UMZs, the terms ‘attached’ and ‘detached’ refer to whether the lower bound of the structure is located within the buffer layer.
2. Computational details
In the present study, we analyse the DNS data of fully developed turbulent pipe flow at
$\textit{Re}_\tau \approx 550$
,
$1000$
,
$3000$
and
$6000$
. Among these, the cases at
$\textit{Re}_\tau \approx 550$
,
$1000$
and
$6000$
were newly computed for this study, while the data at
$\textit{Re}_\tau \approx 3000$
were taken from Ahn et al. (Reference Ahn, Lee, Lee, Kang and Sung2015). The incompressible Navier–Stokes and continuity equations in cylindrical coordinates were solved using the fractional step method of Kim, Baek & Sung (Reference Kim, Baek and Sung2002); further details can be found in Jang, Sung & Krogstad (Reference Jang, Sung and Krogstad2011). The Crank–Nicolson method was employed for time advancement, and a second-order central difference scheme with a staggered grid was used for spatial discretisation. A uniform grid was applied in the streamwise (
$x$
) and azimuthal (
$\theta$
) directions, while a non-uniform grid was adopted in the radial (
$r$
) direction. For
$\textit{Re}_\tau = 550$
,
$1000$
and
$6000$
, a natural stretching function proposed by Pirozzoli & Orlandi (Reference Pirozzoli and Orlandi2021) was applied in the radial direction. The periodic boundary conditions were imposed in the
$x$
and
$\theta$
directions, and the no-slip condition was applied at the wall. At the singular point, the centreline condition for the radial velocity was imposed by averaging the corresponding values across the centreline (Jang et al. Reference Jang, Sung and Krogstad2011). The bulk Reynolds number is defined as
$\textit{Re}_b = 2RU_b/\nu$
, where
$R$
is the pipe radius,
$U_b$
is the bulk velocity and
$\nu$
is the kinematic viscosity. The parameters of the numerical simulations are summarised in table 1. In particular, the case at
$\textit{Re}_\tau \approx 6000$
was computed on the NURION system at the Korea Institute of Science and Technology Information (KISTI) national supercomputing centre, using 24 576 cores with a hybrid OpenMP + MPI parallelisation. For this case, we utilised the open-source library PaScaL_TDMA of Kim et al. (Reference Kim, Kang, Pan and Choi2021) (available at https://github.com/MPMC-Lab/PaScaL_TDMA), which optimises communication overhead and provides excellent scalability on massively parallel architectures. The turbulence statistics were averaged over 10 eddy turnover times (
$tu_\tau /R$
) for all cases, except for
$\textit{Re}_\tau \approx 6000$
, where the averaging duration was approximately 9.7
$tu_\tau /R$
. Although the averaging time for
$\textit{Re}_\tau \approx 6000$
is shorter than that for the other cases, the turbulent statistics were found to be well converged, and their validation shows a good agreement with reference data (see figure 1). The streamwise domain length is set to
$L_x/R = 3\pi$
, which is shorter than that used in the
$\textit{Re}_\tau \approx 3000$
case (Ahn et al. Reference Ahn, Lee, Lee, Kang and Sung2015). However, the present domain length is sufficiently large to capture the characteristic streamwise extent of UMZs (approximately
$1-2\delta$
, where
$\delta$
is the boundary layer thickness), as reported by Adrian et al. (Reference Adrian, Meinhart and Tomkins2000), and is therefore considered adequate for the purposes of the current study. In the present study, the wall-normal direction is defined as
$y = 1 - r$
and the corresponding velocity component is
$V = -u_r$
. The wall-parallel velocity components are
$U = u_x$
and
$W = u_\theta$
, respectively. For convenience, the arc length
$r\theta$
is used to represent the spanwise direction, denoted as
$z = r\theta$
. The ensemble average is represented by angled brackets
$\langle \boldsymbol{\cdot }\rangle$
, and the superscript
$+$
denotes quantities normalised by the viscous scales, namely the viscous length
$\delta _\nu = \nu /u_\tau$
and the friction velocity
$u_\tau$
.
Simulation parameters. Here,
$\textit{Re}_b$
and
$\textit{Re}_{\tau }$
are the bulk Reynolds number and the friction Reynolds number, respectively,
$L_x$
is the domain size in the streamwise direction and
$R$
indicates the pipe radius. Also,
$N_x$
,
$N_r$
and
$N_\theta$
indicate the number of grid points in the streamwise, radial and azimuthal directions, respectively. The streamwise and spanwise grid sizes are denoted by
$\Delta x^+$
and
$\Delta (R\theta )^+$
, respectively. The minimum and maximum radial grid spacings are represented by
$\Delta r_{{min}}^+$
and
$\Delta r_{{max}}^+$
, respectively, and
$\Delta t^+$
is the time step.

The wall-normal profiles of (
$a$
) the mean streamwise velocity, and (
$b$
–
$d$
) the streamwise, wall-normal and spanwise (i.e. azimuthal) turbulence intensities. The black, green, blue and red lines represent the present data at
$\textit{Re}_{\tau } = 550$
,
$1000$
,
$3000$
and
$6000$
, respectively. The open circle symbols
$(\circ )$
denote the pipe flow data from Pirozzoli et al. (Reference Pirozzoli, Romero, Fatica, Verzicco and Orlandi2021).

To validate the present DNS results, the one-point statistics are presented in figure 1. Figure 1 shows the wall-normal profiles of (
$a$
) the mean streamwise velocity
$\langle U \rangle ^+$
, and (
$b{-}d$
) the turbulence intensities in the streamwise
$\langle u^2 \rangle ^+$
, wall-normal
$\langle v^2 \rangle ^+$
and spanwise (i.e. azimuthal)
$\langle w^2 \rangle ^+$
directions. The solid lines correspond to the present DNS results at
$\textit{Re}_{\tau } = 550$
(black), 1000 (green), 3000 (blue) and 6000 (red), while the open symbols represent the reference data from Pirozzoli et al. (Reference Pirozzoli, Romero, Fatica, Verzicco and Orlandi2021), with colours matched to the respective Reynolds numbers. We can see the present results exhibit excellent agreement with the reference data across all the Reynolds numbers. In particular, despite the use of a relatively short streamwise domain length of
$L_x/R = 3\pi$
, the turbulence intensities remain consistent with the reference data (
$L_x/R = 15$
), indicating that the present computational domain is sufficient for capturing the one-point statistics across the pipe.
It is worth highlighting that the streamwise turbulence intensity does not exhibit a logarithmic variation even at
$\textit{Re}_\tau = 6000$
, whereas the spanwise component shows such behaviour even at lower
$\textit{Re}_\tau$
. For the wall-normal component, a plateau can be observed at
$\textit{Re}_\tau = 6000$
, which is consistent with the behaviour predicted by the AEH. According to the AEH, the logarithmic variations of the wall-parallel turbulence intensities and the constant wall-normal component should both appear within the same wall-normal distance range. However, this is not observed in the overall statistics, since, as mentioned in the Introduction, motions that do not follow wall scaling coexist with the wall-scaled motions (scaled with
$y$
and
$u_\tau$
) in turbulent flows. This issue is further analysed from the perspective of UMZs in the following section.
3. Uniform momentum zone detection method
To investigate the wall-scaling behaviour of UMZs across different
$\textit{Re}_\tau$
, the present section describes the detection method for UMZs. Given that UMZs are instantaneous regions of roughly uniform streamwise velocity, the PDF of the instantaneous streamwise velocity (
$U$
) exhibits distinct local maxima, each corresponding to a distinct UMZ (Adrian et al. Reference Adrian, Meinhart and Tomkins2000). When the PDF is computed from an instantaneous velocity field over a domain whose streamwise length is comparable to or smaller than the extent of UMZs, these local maxima become evident. This feature enables the identification of UMZs using the PDF of
$U$
, and such a histogram-based approach was employed in several studies on TBLs (de Silva et al. Reference de Silva, Hutchins and Marusic2016, Reference de Silva, Philip, Hutchins and Marusic2017; Laskari et al. Reference Laskari, de Kat, Hearst and Ganapathisubramani2018; Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020), channel flows (Kwon et al. Reference Kwon, Philip, de Silva, Hutchins and Monty2014; Yang, Hwang & Sung Reference Yang, Hwang and Sung2016) and pipe flows (Yang et al. Reference Yang, Hwang and Sung2019; Chen et al. Reference Chen, Chung and Wan2020; Gul et al. Reference Gul, Elsinga and Westerweel2020; Gunaratne et al. Reference Gunaratne, Guerrero, Lambert and Chin2025). In the present study, we adopt this histogram-based method for detecting UMZs in turbulent pipe flow. The approach utilises two-dimensional instantaneous velocity fields to construct the velocity PDF. While previous studies, particularly in TBLs, typically employed the streamwise–wall-normal (
$x$
–
$y$
) plane following Adrian et al. (Reference Adrian, Meinhart and Tomkins2000), the present study instead uses the cross-stream plane for PDF analysis. The presence of multiple UMZs gives rise to large-scale structures with heights of
$O(R)$
, which extend into the near-wall region and therefore experience geometric confinement effects induced by pipe curvature (Chung et al. Reference Chung, Marusic, Monty, Vallikivi and Smits2015; Hwang, Lee & Sung Reference Hwang, Lee and Sung2016; Han et al. Reference Han, Hwang, Yoon, Ahn and Sung2019). In addition, Gul et al. (Reference Gul, Elsinga and Westerweel2020) compared the locations of UMZs and internal shear layers in turbulent pipe flow using both the
$x$
–
$y$
and cross-stream planes, and found that the UMZ edges detected in the cross-stream plane aligned more closely with the locations of internal shear layers. Motivated by these findings, the cross-stream plane is employed for UMZ detection in the present study.
An example of UMZ detection in a turbulent pipe flow at
$\textit{Re}_{\tau } = 6000$
using a histogram-based method. (
$a$
) The contour plot of the instantaneous streamwise velocity (
$U/U_{\textit{CL}}$
) in the
$r$
–
$\theta$
plane. The colour contours highlight a subdomain (
$l_z^+ = 2000$
) used for UMZ detection. (
$b$
) The corresponding contour plot in the
$y$
–
$z$
plane. (
$c$
) The PDF of
$U/U_{\textit{CL}}$
within the highlighted subdomain. Here, the local peaks, denoted by red circles, represent the modal velocities
$U_m$
, each corresponding to a distinct UMZ, including the core region. The vertical dashed lines mark the midpoints between adjacent peaks, representing the velocity at the UMZ interfaces (
$U_i$
). The different UMZs are labelled as
$R_i$
, where increasing
$i$
indicates proximity to the wall. The core region is excluded from this indexing. (
$d$
) The wall-normal profile of
$U/U_{\textit{CL}}$
at
$z^+ = 3000$
, corresponding to the vertical dashed line in (
$b$
). The grey line indicates the mean velocity profile
$\langle U \rangle /U_{\textit{CL}}$
.

The method is displayed with an example in figure 2. Figure 2(
$a$
) illustrates the instantaneous streamwise velocity (
$U/U_{\textit{CL}}$
) in the cross-stream plane (i.e. the
$r$
–
$\theta$
plane) at
$\textit{Re}_{\tau } = 6000$
, while the corresponding
$y$
–
$z$
plane of the highlighted coloured contour region is presented in figure 2(
$b$
). Here,
$U_{\textit{CL}}$
denotes the mean centreline velocity and
$z$
(
$= r\theta$
) denotes the spanwise direction (or arc length). A subdomain with a spanwise extent of
$l_z^+ = 2000$
is highlighted for the PDF analysis. The corresponding PDF of
$U$
within the highlighted region is shown in figure 2(
$c$
). Here, only the region above
$y^+ \gt 20$
is considered, since the present study focuses on UMZs located above the buffer layer. As seen, several local maxima appear in the PDF, indicating the presence of multiple UMZs. In this example, six distinct peaks are identified, denoted by red circles. Each peak corresponds to a modal velocity
$U_m$
, representing the characteristic velocity of an individual UMZ. The interface velocity (
$U_i$
) between adjacent UMZs is defined as the midpoint between successive peaks (de Silva et al. Reference de Silva, Hutchins and Marusic2016; Laskari et al. Reference Laskari, de Kat, Hearst and Ganapathisubramani2018; Chen et al. Reference Chen, Chung and Wan2020; Gul et al. Reference Gul, Elsinga and Westerweel2020). Although the interface velocity can also be defined based on the local minima of the PDF (de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017; Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020; Gunaratne et al. Reference Gunaratne, Guerrero, Lambert and Chin2025), such minima are not always clearly distinguishable, particularly when the peaks are broadened or closely spaced. In such cases, the minima can be ambiguous or sensitive to noise, leading to inconsistencies in the detected interfaces. Therefore, the velocity at the midpoint between adjacent peaks is used to define the UMZ interfaces. Note that the choice between the midpoint and local-minimum definitions has only a minor influence on the main results of the present study, and the key statistical trends remain qualitatively unchanged when the wall-scaled UMZs are identified.
The contours corresponding to
$U = U_i$
delineate instantaneous yet intricate UMZ boundaries. However, a given contour of
$U = U_i$
may not only represent UMZ boundaries but also enclose disconnected pockets within the flow field. To remove such isolated regions and retain only the primary elongated interfaces, we apply the detection procedure of Borrell & Jiménez (Reference Borrell and Jiménez2016), who extracted the TNTI by removing disconnected parts of the isoline. The resulting interfaces are shown as the black lines in figure 2(a,b). Throughout the remainder of this paper, the interface is defined as the largest singly connected line in the
$y$
–
$z$
plane.
The UMZ with the highest momentum, whose peak is located around
$U/U_{\textit{CL}} \approx 1$
in figure 2(
$c$
), corresponds to the quiescent core region of internal flows (Kwon et al. Reference Kwon, Philip, de Silva, Hutchins and Monty2014; Yang et al. Reference Yang, Hwang and Sung2016). Since this region is characterised by significantly lower turbulence intensity compared with the surrounding turbulent flow, and its boundary resembles a TNTI, the velocity vectors within the core region are excluded from the present analysis. This treatment is consistent with previous studies on TBLs (de Silva et al. Reference de Silva, Hutchins and Marusic2016, Reference de Silva, Philip, Hutchins and Marusic2017; Laskari et al. Reference Laskari, de Kat, Hearst and Ganapathisubramani2018; Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020), where velocity vectors inside the non-turbulent region were also excluded from the analysis. The velocity at the core boundary is
$U = 0.95 U_{\textit{CL}}$
, consistent with that of Yang et al. (Reference Yang, Hwang and Sung2019) at
$\textit{Re}_\tau = 930$
. Note that this threshold was consistently observed across all Reynolds numbers considered in the present study (not shown here). For the remaining UMZs, we follow the notation of Laskari et al. (Reference Laskari, de Kat, Hearst and Ganapathisubramani2018), where each UMZ is labelled as
$R_i$
based on the ranking of
$U_m$
, with larger
$i$
indicating closer proximity to the wall. In figure 2, excluding the core region, a total of five UMZs are identified (i.e.
$N_{\textit{UMZ}} = 5$
). Here,
$R_1$
denotes the UMZ closest to the core region, while
$R_{{max}} = R_5$
represents the UMZ closest to the wall.
Figure 2(
$d$
) shows the wall-normal profile of the instantaneous streamwise velocity
$U$
(black line) at
$z^+ = 3000$
, corresponding to the vertical dashed line in figure 2(
$b$
). The grey line represents the mean velocity profile
$\langle U \rangle$
. The horizontal dashed lines indicate the wall-normal positions of the UMZ interfaces. We can see several step-like jumps in the instantaneous velocity across these interfaces. Based on the modal velocities (
$U_m$
) and interface velocities (
$U_i$
) identified from the PDF, the geometrical properties (i.e. thickness and central position) of individual UMZs can be defined. These properties are visualised by the overlaid red line. The thickness of each UMZ, denoted as
$H_m$
, is defined by the wall-normal distance between its upper (
$y_U$
) and lower edges (
$y_L$
), while its central position,
$y_m$
, is given by the average of those two locations. This allows the identification of each rank (
$R_i$
) in the instantaneous streamwise velocity profile at a given
$z$
location, along with its corresponding
$H_m$
and
$y_m$
.
It is worth noting that, since the PDF is computed from an instantaneous velocity field, both (i) the PDF construction method and (ii) the peak detection criteria are critical (Adrian et al. Reference Adrian, Meinhart and Tomkins2000; Kwon et al. Reference Kwon, Philip, de Silva, Hutchins and Monty2014; de Silva et al. Reference de Silva, Hutchins and Marusic2016; Laskari et al. Reference Laskari, de Kat, Hearst and Ganapathisubramani2018). First, in constructing the PDF, the domain size and the number of velocity vectors used are particularly important. If too many velocity vectors are included, the PDF becomes overly smoothed, causing the modal velocities to vanish and thereby obscuring the zonal organisation characteristic of the flow. Conversely, if too few vectors are used, the PDF may become noisy, leading to ambiguity in identifying the UMZs. To enable direct comparison with previous studies on TBLs (de Silva et al. Reference de Silva, Hutchins and Marusic2016; Laskari et al. Reference Laskari, de Kat, Hearst and Ganapathisubramani2018), the present study adopts the same number of velocity vectors (approximately 5000 below the core region) and bin size (
$0.5u_\tau$
) for PDF construction. Although the instantaneous flow field is taken from the
$y$
–
$z$
plane in this study – unlike the
$x$
–
$y$
plane used in previous works – the overall subdomain size is kept comparable. Specifically, the subdomain length in earlier studies was
$l_x^+ \approx 2000$
, while that in the present study is
$l_z^+ \approx 2000$
. We also examined the effect of domain size and confirmed that the general trends reported in this study remain qualitatively unchanged (see Appendix A). It is worth noting that
$l_x/\delta = 0.1$
, defined with the outer length scale, was used for a broader
$\textit{Re}_\tau$
range of
$O(10^3$
–
$10^4)$
in Heisel et al. (Reference Heisel, de Silva, Hutchins, Marusic and Guala2020). Second, regarding the peak detection criteria, the work of Laskari et al. (Reference Laskari, de Kat, Hearst and Ganapathisubramani2018) provides a detailed discussion. Following their approach, the present study employs specific parameters for peak identification. These thresholds include the minimum peak height (
$F_h$
), the relative height difference between adjacent peaks (
$F_p$
) and the minimum separation between them (
$F_d$
). Laskari et al. (Reference Laskari, de Kat, Hearst and Ganapathisubramani2018) used
$F_h = 0.3$
,
$F_p = 0.05$
and
$F_d = 2$
to match
$N_{\textit{UMZ}}$
reported by de Silva et al. (Reference de Silva, Hutchins and Marusic2016); the same threshold values are adopted in the present study.
4. Hierarchical distribution of the UMZs
First, we examine the occurrence of multiple UMZs in the instantaneous flow fields across a wide range of the Reynolds number,
$\textit{Re}_\tau \approx 550$
,
$1000$
,
$3000$
and
$6000$
, employing the detection method described in § 3. Figure 3(
$a$
) shows the contours of
$U$
in the
$y$
–
$z$
plane. Here, the blue line contour indicates the core boundary (
$U/U_{\textit{CL}} = 0.95$
), while the black line contours denote the UMZ edges (or interfaces) where
$U = U_i$
as obtained from the corresponding PDFs in figure 3(
$b$
). In figure 3(
$b$
), the vertical dashed lines indicate
$U_i$
and the red circles mark the modal velocities
$U_m$
of each UMZ. As seen, the number of the detected UMZs increases with increasing
$\textit{Re}_\tau$
, implying a hierarchical distribution (de Silva et al. Reference de Silva, Hutchins and Marusic2016). In addition, the interfaces shown in figure 3(
$a$
) exhibit highly convoluted shapes (i.e. folding back and forth), and this trend becomes more pronounced at higher
$\textit{Re}_\tau$
. Figure 3(
$c$
) further reveals a step-like jump in
$U$
across the interfaces, indicating the presence of a strong shear layer near the UMZ boundaries, consistent with previous observations in TBLs (Eisma et al. Reference Eisma, Westerweel, Ooms and Elsinga2015; de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017). Moreover, figure 3(
$c$
) shows the regions of nearly uniform velocity (
$U$
) extending over a finite spanwise extent. These observations confirm that the present detection method is effective for identifying UMZs across
$\textit{Re}_\tau$
in pipe flow.
The instantaneous visualisation of UMZs at
$\textit{Re}_{\tau } = 550$
–
$6000$
. (
$a$
) The instantaneous flow field with the UMZ edges (
$U = U_i$
) denoted by black line contour. The core boundary (
$U / U_{\textit{CL}} = 0.95$
) is indicated by the blue line contour. (
$b$
) The PDFs of
$U$
computed from the subdomains in (
$a$
). The local peaks, marked by red circles, correspond to
$U_m$
of individual UMZs. The vertical dashed line denotes the interface velocity
$U_i$
. (
$c$
) The three-dimensional
$U$
contours over
$\Delta z/R = 0.2$
. Here, the thick black line represents the wall-normal profile of
$U$
at the spanwise location denoted by the vertical dashed lines in (
$a$
).

Figure 3. Long description
Panel A: Four contour plots show the instantaneous flow field with UMZ edges denoted by black line contours and the core boundary indicated by blue line contours. The x-axis is labeled z/R and the y-axis is labeled y/R. The color gradient represents the velocity ratio U/Ucl, ranging from 0.60 to 0.95. Panel B: Four bar histograms display the probability density functions (PDFs) of the velocity ratio U/Ucl. The x-axis is labeled U/Ucl and the y-axis is labeled PDF. Local peaks are marked by red circles, and the vertical dashed line denotes the interface velocity. Panel C: Four 3D surface plots show the three-dimensional contours of the velocity ratio U/Ucl over the normalized height y/R and spanwise location z/R. The thick black line represents the wall-normal profile of U/Ucl at the spanwise location denoted by the vertical dashed lines in Panel A.
The number of UMZs (
$N_{\textit{UMZ}}$
) at varying Reynolds numbers. (
$a$
) The PDFs of
$N_{\textit{UMZ}}$
at
$\textit{Re}_\tau = 550$
–
$6000$
. (
$b$
) The mean number of UMZs,
$\langle N_{\textit{UMZ}} \rangle$
, plotted as a function of
$\textit{Re}_{\tau }$
. The data from TBLs by de Silva et al. (Reference de Silva, Hutchins and Marusic2016) (
$\lozenge$
) and Chen et al. (Reference Chen, Fan, Yue, Cheng, Tang and Jiang2025) (
$\triangle$
) are also shown for comparison. The blue dashed line denotes the prediction of Bautista et al. (Reference Bautista, Ebadi, White, Chini and Klewicki2019).

Figure 4. Long description
Panel A: A line graph shows the probability density functions of the number of UMZs at a specific Reynolds number. The x-axis represents the number of UMZs, and the y-axis represents the probability density function. Panel B: A line graph plots the mean number of UMZs as a function of Reynolds number. The x-axis represents the Reynolds number, and the y-axis represents the mean number of UMZs. Data from TBLs by de Silva et al. and Chen et al. are shown for comparison, along with a blue dashed line denoting the prediction of Bautista et al. Panel C: A scatter plot and line graph show the conditionally averaged wall-normal turbulence intensity and Reynolds shear stress of the wall-scaled UMZs. The closed circles represent the conditional averages over all wall-scaled UMZs, while the open circles denote the corresponding quantities for cases with a specific condition. The grey lines indicate the unconditioned statistics for comparison. The vertical dashed lines denote the peak location of the wall-normal turbulence intensity and the position of a specific value.
To statistically quantify the hierarchical distribution of UMZs, we examine the number of UMZs (
$N_{\textit{UMZ}}$
). Figure 4(
$a$
) presents the PDFs of
$N_{\textit{UMZ}}$
at different
$\textit{Re}_\tau$
. As
$\textit{Re}_{\tau }$
increases from 550 to 6000, the distributions shift toward higher
$N_{\textit{UMZ}}$
values and become broader, indicating that a greater number of UMZs with varying scales coexist at higher Reynolds numbers. This observation, reported here for the first time in pipe flow, highlights the multiscale nature of wall-bounded turbulence. Such a trend was not observed in the previous pipe flow study by Gul et al. (Reference Gul, Elsinga and Westerweel2020), due to the relatively narrow Reynolds-number range (
$\textit{Re}_\tau = 340$
–
$1259$
) considered. These observations reflect the hierarchical organisation of UMZs, which emerges regardless of flow geometry. Figure 4(
$b$
) shows the mean number of UMZs
$\langle N_{\textit{UMZ}} \rangle$
, plotted as a function of
$\textit{Re}_{\tau }$
on a logarithmic scale. The filled circles represent the present DNS results and, for comparison, the data from TBLs by de Silva et al. (Reference de Silva, Hutchins and Marusic2016) and Chen et al. (Reference Chen, Fan, Yue, Cheng, Tang and Jiang2025) are shown as open diamonds and open triangles, respectively. Here, the blue dashed line indicates the predicted variation of the number of UMZs based on the model of Bautista et al. (Reference Bautista, Ebadi, White, Chini and Klewicki2019), which incorporates finite-Reynolds-number corrections by accounting for the partial confinement of the mean velocity variation within the vortical fissures (approximately 75 %). As seen, the present data are consistent with this prediction, with particularly good agreement observed at
$\textit{Re}_\tau = 3000$
and
$6000$
. The present data exhibit a clear logarithmic increase of
$\langle N_{\textit{UMZ}} \rangle$
with Reynolds number (indicated by the dashed line), consistent with the previous TBL results. This logarithmic increase observed in pipe flow further supports the hierarchical nature of the multiple UMZ phenomenon. To the authors’ knowledge, this is the first observation of such a trend in pipe flow. This not only confirms the robustness of the detection method in cylindrical geometries, but also implies that the spatial organisation of UMZs could be a universal feature of wall-bounded turbulence.
The PDFs of (
$a$
) the modal velocities (
$U_m$
) and (
$b$
) the interface velocity (
$U_i$
) at
$\textit{Re}_\tau \approx 550$
–
$6000$
. The dashed line indicates the core-boundary velocity at
$U/U_{\textit{CL}}=0.95$
and the shaded region denotes the core region.

Figure 5 shows the PDFs of (
$a$
) the modal velocities (
$U_m$
) and (
$b$
) the interface velocities (
$U_i$
) at
$\textit{Re}_\tau \approx 550$
–
$6000$
to examine how the distributions vary with
$\textit{Re}_\tau$
. In this figure, the core region, indicated by the shaded area, is also included; however, excluding the core region does not affect the PDF. As seen in figure 5(
$a$
), the distribution of
$U_m$
becomes progressively broader with increasing
$\textit{Re}_\tau$
, indicating the presence of more UMZs characterised by a wider range of different streamwise momenta. This broadening is particularly due to the emergence of UMZs with lower
$U_m/U_{\textit{CL}}$
values at higher
$\textit{Re}_\tau$
, implying that roughly uniform streamwise momentum regions increasingly occupy the near-wall regions over a substantial extent. Interestingly, a comparison between
$\textit{Re}_\tau = 3000$
(blue) and
$6000$
(red) reveals that the PDF becomes relatively flat over a wide range of
$U_m/U_{\textit{CL}}$
, indicating that UMZs with different modal velocities exist with similar probabilities throughout the flow. Given that higher
$U_m$
values are associated with longer-lifetime packets in the context of the hairpin packet paradigm (Adrian et al. Reference Adrian, Meinhart and Tomkins2000), this trend implies that UMZs with different lifetimes are distributed with comparable likelihoods. Although the present study does not address such dynamic aspects, the lifetime and evolution of UMZs, as well as their interactions with surrounding vortical structures (Sayyari, Hwang & Kim Reference Sayyari, Hwang and Kim2022) and the associated wall-shear stress (Park & Hwang Reference Park and Hwang2024), merit further investigation over a wide range of
$\textit{Re}_\tau$
.
Next, we examine the PDFs of the interface velocity,
$U_i/U_{\textit{CL}}$
, as shown in figure 5(
$b$
). At lower Reynolds numbers (
$\textit{Re}_\tau = 550$
and
$1000$
), the PDFs exhibit distinct peaks located at
$U_i/U_{\textit{CL}} \gt 0.8$
, indicating that most UMZs are concentrated in the outer region of the flow. In contrast, at higher Reynolds numbers (
$\textit{Re}_\tau = 3000$
and
$6000$
), the distribution becomes broader and flatter, similar to the trend observed in figure 5(
$a$
). This suggests that the UMZ interfaces become more prevalent at higher
$\textit{Re}_\tau$
. Given that a velocity jump occurs across each interface (Meinhart & Adrian Reference Meinhart and Adrian1995), this suggests that interfaces are distributed over a wide range of wall-normal locations, and such localised velocity jumps occur throughout a substantial portion of the flow domain. Consequently, it is plausible that the superposition of these velocity jumps can accurately reconstruct one-point statistics, as demonstrated in previous studies (de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017; Bautista et al. Reference Bautista, Ebadi, White, Chini and Klewicki2019; Ehsani et al. Reference Ehsani, Heisel, Li, Voller, Hong and Guala2024). A detailed analysis of the velocity jump and the interface thickness in the present pipe flows will be presented in § 5. Furthermore, the results in this section indicate the coexistence of UMZs of various scales, highlighting the multiscale nature of UMZs. As discussed in the introduction, it is necessary to identify, among the detected UMZs, those that satisfy wall scaling (i.e. are characterised by
$y$
and
$u_\tau$
), which becomes increasingly important in high-Reynolds-number flows.
Note that the modal velocity peak consistently occurs at
$U_m/U_{\textit{CL}} \approx 1$
(figure 5
$a$
), while the interface velocity peak occurs at
$U_{i}/U_{\textit{CL}} \approx 0.95$
(figure 5
$b$
), regardless of Reynolds number. These results indicate the presence of a core region with high and uniform momentum, bounded by 0.95
$U_{\textit{CL}}$
in turbulent pipe flows, consistent with the core region found in turbulent channel flows by Kwon et al. (Reference Kwon, Philip, de Silva, Hutchins and Monty2014). As discussed in § 3, however, the core region in internal flows is characterised by a relatively low turbulence level and shares similar features with the non-turbulent region observed in TBLs (Kwon et al. Reference Kwon, Philip, de Silva, Hutchins and Monty2014; Yang et al. Reference Yang, Hwang and Sung2016, Reference Yang, Hwang and Sung2019). For consistency with previous TBL studies (de Silva et al. Reference de Silva, Hutchins and Marusic2016; Laskari et al. Reference Laskari, de Kat, Hearst and Ganapathisubramani2018; Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020), the core region was therefore excluded from the UMZ detection.
5. Identification of wall-scaled UMZs
In § 4, we demonstrated that UMZs of various scales coexist in turbulent pipe flow, and that their number increases with Reynolds number, reflecting the multiscale nature of wall-bounded turbulence. However, not all of these UMZs satisfy wall scaling. In high-Reynolds-number flows, wall-scaled UMZs are of particular interest because they may serve as fundamental building blocks for predicting turbulence statistics in the logarithmic region within the framework of the AEH. In this section, we identify wall-scaled UMZs by quantifying their thickness and the associated velocity jumps across their interfaces.
The joint probability density functions (JPDFs) of the central position (
$y_m/R$
) and the thickness (
$H_m/R$
) of all UMZs; (
$a$
)
$\textit{Re}_{\tau } = 550$
, (
$b$
)
$1000$
, (
$c$
)
$3000$
and (
$d$
)
$6000$
. The red dashed line indicates
$H_m = 2 y_m$
. The insets show the JPDFs of
$y_m^+$
and
$H_m^+$
. The contour levels correspond to
$40\,\%, 60\,\%, 70\,\%$
and
$80\,\%$
of the total probability.

5.1. Geometrically self-similar UMZs
We first examine the UMZ thickness,
$H_m$
. As noted in figure 2, the thickness is defined as the distance between the lower and upper interfaces of a UMZ. Figure 6 presents the joint probability density functions (JPDFs) of the central position (
$y_{m}/R$
) and the thickness (
$H_{m}/R$
) for all detected UMZs (i.e.
$R_1$
to
$R_{{max}}$
). In general, the detected UMZs exhibit a wide range of
$H_m$
, and, in particular, at a given
$y_m$
they display a broad variation in thickness. This indicates the coexistence of UMZs that scale with
$y$
and UMZs that do not follow such scaling. The inserted red dashed line denotes
$H_m = 2 y_m$
, along which UMZs are found to align regardless of
$\textit{Re}_\tau$
. For high
$\textit{Re}_\tau$
(
$= 3000$
and
$6000$
), this tendency occurs over a broader range of
$y_m$
, as is evident in the insets plotted in viscous wall units. Such UMZs have their lower edges located below
$y^+ = 20$
, implying that they extend down to the wall. According to the rank classification, these correspond to
$R_{{max}}$
and can have thicknesses of
$0.3$
–
$0.4R$
or larger. They can therefore be regarded as geometrically self-similar UMZs whose thickness scales with
$y$
.
The remaining UMZs (i.e.
$R_1$
to
$R_{{max}-1}$
) are distributed below the dashed line. The inset shows the distribution in viscous wall units, where it is evident that the spread of these UMZs increases with
$\textit{Re}_\tau$
. Within this population, both geometrically self-similar and non-self-similar UMZs are present. It can also be seen that a large number of UMZs with small thicknesses (
$H_m^+ \lt 50$
) are distributed over a wide range of
$y_m$
. In addition, a noticeable bulge of the contours is observed around
$y_m/R \approx 0.4$
, particularly at lower Reynolds numbers (figures 6
$a$
and 6
$b$
). These thin UMZs and those located within the bulge region can be regarded as non-self-similar, i.e. not scaling with
$y$
, and their population appears to be dominant at lower
$\textit{Re}_\tau$
. To extract wall-scaled UMZs, it is necessary to identify, among the UMZs corresponding to
$R_1$
to
$R_{{max}-1}$
, those that are geometrically self-similar.
(
$a$
) The JPDF of the central position (
$y_m/R$
) and thickness (
$H_m/R$
) of the detected UMZs at
$\textit{Re}_{\tau }=550$
. The colour contours correspond to all UMZs (
$R_1$
to
$R_{\textit{max}}$
), with contour levels at 40 %, 60 %, 70 % and 80 % of the total probability. The line contours indicate the JPDFs for the
$R_1$
and
$R_{\textit{max}}$
, corresponding to 30 % and 95 % of the total probability, respectively. The red dashed line indicates
$H_m = 2y_m$
. (
$b$
) The JPDFs of
$y_m/R$
and
$H_m/R$
for
$R_1$
and
$R_{\textit{max}}$
at
$\textit{Re}_{\tau }=550$
–6000. The line contours for
$R_1$
and
$R_{\textit{max}}$
enclose 30 % and 95 % of the total probability, respectively. The dashed line indicates
$H_m=2y_m$
. (
$c$
) The JPDFs for
$R_2$
to
$R_{{max} - 1}$
, plotted with contour levels enclosing 70 % of the total probability. The open circles indicate
$\langle H_{m}(y_m) \rangle$
. The dashed line denotes
$H_m = 0.75y_m$
, and the dashed–dot line denotes
$H_m/R = 0.75y_m/R - 0.15$
. (
$d$
) The variation of the mean thickness
$\langle H_m \rangle$
of the self-similar UMZs (filled circles), in comparison with that of
$R_2$
to
$R_{{max} - 1}$
in (
$c$
). The open circles correspond to all UMZs in (
$c$
). The inset shows the same result in viscous wall units.

Figure 7. Long description
Panel A: A scatter plot with color contours and line contours. The x-axis is labeled y_m/R and the y-axis is labeled H_m/R. The color contours represent the joint probability density function (JPDF) of the central position and thickness of detected uniform momentum zones (UMZs) at Re_tau = 550, with contour levels at 40 percent, 60 percent, 70 percent, and 80 percent of the total probability. The line contours indicate the JPDFs for R_1 and R_max, corresponding to 30 percent and 95 percent of the total probability, respectively. The red dashed line indicates H_m/R = 2y_m/R. Panel B: A scatter plot with line contours. The x-axis is labeled y_m/R and the y-axis is labeled H_m/R. The line contours represent the JPDFs of H_m/R and y_m/R for Re_tau values of 550, 1000, 3000, and 6000, enclosing 30 percent and 95 percent of the total probability, respectively. The dashed line indicates H_m/R = 2y_m/R. Panel C: A scatter plot with line contours. The x-axis is labeled y_m/R and the y-axis is labeled H_m/R. The line contours represent the JPDFs for Re_tau values of 550, 1000, 3000, and 6000, with contour levels enclosing 70 percent of the total probability. The open circles indicate \langle H_m(y_m)\rangle. The dashed line denotes H_m/R = 0.75y_m/R, and the dashed-dot line denotes H_m/R = 0.75y_m/R - 0.15. Panel D: A scatter plot with filled circles and open circles. The x-axis is labeled y_m/R and the y-axis is labeled \langle H_m/R\rangle. The filled circles represent the mean thickness of the self-similar UMZs, while the open circles correspond to all UMZs. The inset shows the same result in viscous wall units.
To achieve this, we next analyse the relationship between
$y_{m}$
and
$H_{m}$
by grouping the UMZs according to their rank. Figure 7(
$a$
) shows the distribution of all detected UMZs (
$R_1$
–
$R_{{max}}$
) at
$\textit{Re}_{\tau }=550$
. The coloured contours are plotted using the same contour levels as in figure 6, and the line contours of
$R_1$
and
$R_{\textit{max}}$
are superposed. As shown, the bulge observed in figure 6 around
$y_m/R\approx 0.4$
–
$0.5$
is mainly associated with the
$R_1$
UMZs, whereas
$R_{\textit{max}}$
corresponds to the wall-adjacent UMZs aligned with
$H_m=2y_m$
. Figure 7(
$b$
) shows the distributions for
$R_{{max}}$
, with those of
$R_1$
plotted simultaneously. As shown, the
$R_{{max}}$
contours not only collapse across
$\textit{Re}_\tau$
, but also follow the relationship
$H_m = 2y_m$
, indicating that they are geometrically self-similar. Since the JPDFs are plotted using outer scaling, the distributions for the high-
$\textit{Re}_\tau$
cases are less extended compared with those for the low-
$\textit{Re}_\tau$
cases. This is consistent with the earlier result showing that
$N_{\textit{UMZ}}$
increases with
$\textit{Re}_\tau$
(figure 4). As a result, both the thickness and the central position of
$R_{{max}}$
(i.e. the UMZ closest to the wall) decrease.
In figure 7(
$b$
), an examination of the contours of
$R_1$
shows that they collapse reasonably well across Reynolds numbers, except for
$\textit{Re}_\tau = 6000$
. The UMZs corresponding to
$R_1$
, which are located closest to the core region, give rise to the bulge-like distribution shown in figure 6. They are primarily distributed at
$y_m/R \approx 0.4$
–
$0.5$
and appear to scale in outer units. For
$\textit{Re}_\tau = 6000$
, the distribution shifts to slightly lower
$y_m$
values, but, similar to the lower-
$\textit{Re}_\tau$
cases, they remain concentrated at relatively fixed
$y_m$
positions. As a result, they do not exhibit a clear increase in thickness with increasing
$y_m$
. These observations suggest that most
$R_1$
UMZs can be regarded as geometrically non-self-similar. It is also plausible that
$R_1$
contains UMZs influenced by the core region, or even UMZs that should be classified as part of the core region.
Next, we examine the distribution of the remaining ranks, i.e. UMZs corresponding to
$R_2$
to
$R_{{max}-1}$
, in figure 7(
$c$
). The line contours of the JPDFs show a similar distribution pattern regardless of
$\textit{Re}_\tau$
. To examine this in more detail, the mean thickness at each
$y_m$
is plotted as open circles, and we can see that the variation of
$\langle H_m \rangle$
collapses well across all
$\textit{Re}_\tau$
. In addition, the mean thickness aligns along the dashed line (
$\langle H_m \rangle = 0.75 y_m$
) up to
$y_m/R = 0.2$
. This linear relationship is consistent with previous TBL studies (Bautista et al. Reference Bautista, Ebadi, White, Chini and Klewicki2019; Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020). Unlike previous studies, which did not distinguish UMZs by rank, the present analysis reveals that these geometrically self-similar UMZs correspond to
$R_2$
to
$R_{{max}-1}$
.
However, beyond
$y_m/R \gt 0.2$
,
$\langle H_m \rangle$
deviates from the linear relationship due to the increasing presence of relatively thin UMZs. For
$H_m/R \lt 0.15$
, we can see relatively thin UMZs over a wide range of
$y_m$
and become increasingly dominant with
$y_m$
, thereby obscuring the self-similar scaling. If these thin UMZs are excluded, the linear relationship persists even in
$y_m/R \gt 0.2$
. To verify this, we computed
$\langle H_m \rangle$
by conditionally sampling only the UMZs that satisfy
$H_m/R \gt 0.75y_m/R - 0.15$
(denoted by the dashed-dot line), where the slope 0.75 is taken from the mean thickness of UMZs in
$y_m/R \lt 0.2$
, and the offset is set to 0.15. This offset ensures that all UMZs are included for
$y_m/R \lt 0.2$
, while only the thicker UMZs are selected for
$y_m/R \gt 0.2$
. In figure 7(
$d$
), the filled circles indicate the corresponding conditionally sampled UMZs, showing that the relation
$\langle H_m \rangle = 0.75 y_m$
remains valid up to
$y_m/R \approx 0.6$
. This behaviour is further confirmed in viscous wall units in the inset, where the linear scaling extends over a wider range of
$y_m$
as
$\textit{Re}_\tau$
increases. It is worth noting that the choice of the slope and offset could affect the detection of geometrically self-similar UMZs. However, the main purpose of the present work is not to determine the precise criteria, but rather to demonstrate that the linear relationship between
$H_m$
and
$y_m$
persists in the outer region once non-self-similar UMZs are excluded. The influence of the slope and offset is examined in Appendix B, where it is shown that the main findings of this study remain qualitatively unchanged.
According to the above results, we define the geometrically self-similar UMZs as follows. The innermost UMZ (
$R_{{max}}$
), i.e. the UMZ closest to the wall, is included, as it consistently follows the expected self-similar scaling. The outermost UMZ (
$R_1$
), i.e. the UMZ farthest from the wall, is excluded, as it is associated with the quiescent core region. For the remaining UMZs, we apply the following criteria to extract geometrically self-similar structures. First, to exclude the influence of the core region and focus on structures embedded within the logarithmic layer, we restrict the wall-normal position to
$ y_{m} / R \lt 0.35$
. Second, to extract geometrically self-similar UMZs from
$R_2$
to
$R_{{max}-1}$
, we select UMZs whose mean thickness follows the linear relationship
$\langle H_{m} \rangle \approx 0.75 y_{m}$
. Third, to ensure that the selected UMZs contain well-resolved and physically significant shear layers, we impose a minimum thickness threshold of
$H_{m}^{+} \gt 50$
; the mean shear-layer thickness is approximately 50 viscous wall units for
$\textit{Re}_\tau = 6000$
, so thinner UMZs are excluded (see § 5.2). For these self-similar UMZs to exhibit wall scaling, it is essential that the velocity jumps between adjacent UMZs also scale with the friction velocity,
$u_\tau$
. Hence, in the following, we first examine the internal shear layers at UMZ edges, and then analyse the velocity jumps across the shear layers separating adjacent UMZs.
5.2. Internal shear layers
As explained in § 3, the UMZ interface is identified based on the line contours at
$U = U_i$
. The interface location
$y_i$
corresponds to the wall-normal position of the line contour shown in figure 2(
$b$
). To extract a single, continuous contour for a given
$U_i$
, we applied the method of Borrell & Jiménez (Reference Borrell and Jiménez2016) within the subdomain. When the interface was severely folded back, resulting in multiple
$y_i$
at a given
$z$
location, we followed the approach of previous studies that used the outer envelope of each interface to define a single
$y_i$
(Chauhan et al. Reference Chauhan, Philip, de Silva, Hutchins and Marusic2014; Kwon et al. Reference Kwon, Philip, de Silva, Hutchins and Monty2014; de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017). Using this definition of
$y_i$
, the streamwise velocity field in the vicinity of the interface was conditionally averaged, as illustrated schematically in figure 8(
$a$
).
(
$a$
) The schematic of the conditional averaging procedure at UMZ edges. The coloured contours and black line contours denote
$U/U_{\textit{CL}}$
and the UMZ edges, respectively. The red solid line corresponds to the outer envelope of the edges. The red
$\diamond$
symbols indicate the interface coordinates
$y_i$
, while the black
$\diamond$
symbols mark the grid points sampled in the near vicinity of each interface. (
$b$
) The conditionally averaged streamwise velocity profile
$\langle U|_i \rangle$
at
$y_i/R = 0.1$
. The profiles are plotted relative to the local frame of reference at each interface
$y_i$
. Here,
$\Delta U$
,
$\delta _\omega$
and
$\partial \langle U\rangle /\partial y|_{\textit{max}}$
denote the average velocity jump, the shear-layer thickness and the maximum local velocity gradient, respectively. (c,d) The variations of
$\Delta U^{+}$
with wall-normal distance for all UMZs (
$c$
) and for the self-similar UMZs (
$d$
). The insets show the same results in viscous wall units. Note that
$\Delta U^{+}$
in (
$d$
) corresponds only to the UMZs satisfying the self-similarity criteria defined in § 5.1.

Figure 8(
$b$
) shows a sample of the conditionally averaged streamwise velocity,
$\langle U|_i \rangle$
, at
$y_i/R = 0.1$
. Here, the statistics are plotted relative to the local frame of reference at each interface
$y_i$
. As seen, the profiles exhibit a steep gradient in
$U$
confined to a thin region around
$y = y_i$
. Based on these interfacial layer characteristics, the velocity jump (
$\Delta U$
) and shear-layer thickness (
$\delta _\omega$
) are defined following the approach used in previous studies of interfacial layers in boundary layers (Chauhan et al. Reference Chauhan, Philip, de Silva, Hutchins and Marusic2014; Eisma et al. Reference Eisma, Westerweel, Ooms and Elsinga2015; de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017) and in pipe flows (Gul et al. Reference Gul, Elsinga and Westerweel2020). The velocity jump
$\Delta U$
is obtained by projecting linear fits to the regions above and below the sharp velocity change in
$\langle U|_i \rangle$
near the interface location. The shear-layer thickness (
$\delta _\omega$
) is defined as
where
$\partial \langle U|_i \rangle /\partial y|_{\textit{max}}$
denotes the maximum local gradient of the profile (see the illustration in the inset).
To examine whether the velocity jumps of the geometrically self-similar UMZs defined in § 5.1 scale with the friction velocity
$u_\tau$
, we first analyse the velocity jump
$\Delta U$
. Figure 8(c,d) shows the variations of
$\Delta U^+$
with respect to
$y_i$
: panel (
$c$
) presents the results obtained from all UMZs, while panel (
$d$
) corresponds to the geometrically self-similar UMZs. Given that the velocity jump between adjacent UMZs occurs across these shear layers,
$\Delta U$
can be regarded as a characteristic velocity of the UMZs (Bautista et al. Reference Bautista, Ebadi, White, Chini and Klewicki2019; Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020). As seen in figure 8(
$d$
), the self-similar UMZs exhibit
$\Delta U^{+} \approx 1.5$
(indicated by the horizontal dashed line) over the range
$0.05 \lt y_i/R \lt 0.3$
, regardless of
$\textit{Re}_\tau$
. Compared with the results for all UMZs in figure 8(
$c$
), the self-similar UMZs show a much better collapse when scaled with
$u_\tau$
. These results demonstrate that the geometrically self-similar UMZs defined in § 5.1 have a characteristic velocity of
$u_\tau$
, firmly establishing them as wall-scaled UMZs. Note that UMZs that do not satisfy these scaling conditions can be classified as non-wall-scaled UMZs. These structures are distributed over a broad range of the flow and tend to obscure the asymptotic statistical behaviour predicted by the AEH when all UMZs are considered together. A detailed comparison between wall-scaled and non-wall-scaled UMZs, including their appearance in the instantaneous flow field and associated statistics, is provided in Appendix C.
It is worth noting that the mean momentum balance-based framework (Wei et al. Reference Wei, Fife, Klewicki and McMurtry2005; Fife et al. Reference Fife, Klewicki and Wei2009) predicts that the velocity increment associated with each scaling layer is
$O(u_{\tau })$
, and that Klewicki (Reference Klewicki2021) showed that its asymptotic value approaches
$\Delta U^{+}\approx 1.26$
at asymptotically high
$\textit{Re}_\tau$
. In the present study, we find
$\Delta U^{+}\approx 1.5$
, which is slightly higher than this asymptotic value. Nevertheless, the key observation is that the velocity jump associated with the extracted self-similar UMZs exhibits a robust scaling with
$u_\tau$
over a wide wall-normal range. The difference in magnitude may be attributed to finite-Reynolds-number effects and/or differences in the definition of the velocity jump based on the UMZ identification procedure. In particular, the use of an envelope-based method may lead to a slightly overestimated velocity jump. While alternative definitions based on vorticity (e.g. internal shear layers or vortical fissures as in Eisma et al. Reference Eisma, Westerweel, Ooms and Elsinga2015) could also be considered, such an analysis is beyond the scope of the present study.
It is also worth highlighting that the
$u_\tau$
-scaling of the velocity jump across UMZs was reported mainly in TBLs, whereas the present work is the first to demonstrate this behaviour in turbulent pipe flow. In particular, de Silva et al. (Reference de Silva, Philip, Hutchins and Marusic2017) showed that the velocity jump across UMZs can scale with
$u_\tau$
, but their analysis covered only a narrow momentum-deficit region and did not explicitly assess its variation with
$y_i$
. Heisel et al. (Reference Heisel, de Silva, Hutchins, Marusic and Guala2020) further identified UMZs in the logarithmic region of a TBL and reported
$u_\tau$
-scaling of
$\Delta U$
up to
$y/\delta \approx 0.2$
. Gul et al. (Reference Gul, Elsinga and Westerweel2020) also analysed UMZs in pipe flow but observed that
$\Delta U$
decreases with increasing
$y_i$
and did not exhibit a flat region when normalised by
$u_\tau$
– a trend similar to that of de Silva et al. (Reference de Silva, Philip, Hutchins and Marusic2017). In contrast, the present results show
$\Delta U^{+} \approx 1.5$
over an extended wall-normal range, up to
$y_i/R \approx 0.3$
, compared with those reported in TBLs (Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020). This broader extent is attributed to the selective extraction of geometrically self-similar UMZs. In Heisel et al. (Reference Heisel, de Silva, Hutchins, Marusic and Guala2020), only UMZs located within the logarithmic region were considered. As demonstrated in figures 6 and 7, however, the flow contains a mixture of self-similar and non-self-similar UMZs, and the inclusion of the latter tends to obscure the
$u_\tau$
-scaling behaviour. As a result, when only the wall-scaled UMZs are isolated through the present filtering procedure, the
$u_\tau$
-scaling becomes more apparent over a broader range. These wall-scaled UMZs therefore provide a basis for reconstructing turbulence statistics that exhibit the asymptotic behaviours predicted by the AEH, as will be discussed in the next section.
The variations of the shear-layer thickness
$\delta _\omega /\lambda _T$
as a function of
$y_i/R$
. (
$a$
) All UMZs; (
$b$
) wall-scaled UMZs, where
$\lambda _T$
denotes the local Taylor microscale. For the wall-scaled UMZs,
$\delta _\omega$
is further normalised by different characteristic length scales: (
$c$
) the pipe radius
$R$
; (
$d$
)
$\textit{Re}_\tau ^{1/2}$
(associated with the vortical fissure scaling); (
$e$
) the viscous length scale
$\delta _\nu$
; (
$f$
) the local Kolmogorov length scale
$\eta$
.

Figure 9. Long description
The image contains six scatter plots labeled (a) through (f), each depicting the variations of the shear-layer thickness as a function of different parameters. Panel A: A scatter plot titled All UMZs. The x-axis is labeled yi/R and ranges from 0 to 0.6. The y-axis is labeled δω/λT and ranges from 0 to 0.6. Different colored dots represent different Reτ values: black for 550, green for 1000, blue for 3000, and red for 6000. Panel B: A scatter plot titled Wall-scaled UMZs. The x-axis is labeled yi/R and ranges from 0 to 0.6. The y-axis is labeled δω/λT and ranges from 0 to 0.6. Different colored dots represent different Reτ values: black for 550, green for 1000, blue for 3000, and red for 6000. Panel C: A scatter plot titled Wall-scaled UMZs. The x-axis is labeled yi/R and ranges from 0 to 0.6. The y-axis is labeled δω/R and ranges from 0 to 0.05. Different colored dots represent different Reτ values: black for 550, green for 1000, blue for 3000, and red for 6000. Panel D: A scatter plot titled Wall-scaled UMZs. The x-axis is labeled yi/R and ranges from 0 to 0.6. The y-axis is labeled δω+/Reτ^(1/2) and ranges from 0 to 1.5. Different colored dots represent different Reτ values: black for 550, green for 1000, blue for 3000, and red for 6000. Panel E: A scatter plot titled Wall-scaled UMZs. The x-axis is labeled yi/R and ranges from 0 to 0.6. The y-axis is labeled δω+ and ranges from 0 to 100. Different colored dots represent different Reτ values: black for 550, green for 1000, blue for 3000, and red for 6000. Panel F: A scatter plot titled Wall-scaled UMZs. The x-axis is labeled yi/R and ranges from 0 to 0.6. The y-axis is labeled δω/η and ranges from 0 to 15. Different colored dots represent different Reτ values: black for 550, green for 1000, blue for 3000, and red for 6000.
Before proceeding to the next section, we analyse the shear-layer thickness,
$\delta _\omega$
. Given that a velocity jump occurs across the interfaces between adjacent UMZs (i.e. shear layers), which separate large-scale regions of nearly uniform velocity,
$\delta _\omega$
can be regarded as the characteristic length scale of the vortical structures residing within the shear layers that exchange momentum between UMZs (Eisma et al. Reference Eisma, Westerweel, Ooms and Elsinga2015; de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017). In addition, Heisel et al. (Reference Heisel, De Silva, Hutchins, Marusic and Guala2021) analysed the properties of prograde spanwise vortex cores and internal shear layers in TBLs and showed that
$\delta _\omega$
scales with the Taylor microscale, consistent with the previous works (Eisma et al. Reference Eisma, Westerweel, Ooms and Elsinga2015; de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017). They further suggested that the Taylor microscale plays an important role in the dynamical link between spanwise vortices and the formation of UMZs.
To examine this scaling in turbulent pipe flow, figure 9(a,b) shows the variations of
$\delta _\omega /\lambda _T$
as a function of
$y_i/R$
, where
$\lambda _T$
denotes the local Taylor microscale. As seen, the variations of
$\delta _\omega /\lambda _T$
collapse well regardless of
$\textit{Re}_\tau$
. In particular, the
$\delta _\omega$
of the wall-scaled UMZs (figure 9
$b$
) exhibits a better collapse than that of all UMZs (figure 9
$a$
), extending up to
$y_i/R \approx 0.5$
. In contrast, figure 9(c,e) shows that
$\delta _\omega$
does not collapse when normalised by the pipe radius
$R$
or the viscous length scale
$\delta _\nu$
, indicating that
$\lambda _T$
provides the appropriate scaling for the shear-layer thickness. The observation that
$\delta _\omega$
has a magnitude of
$O(\lambda _T)$
, similar to the thickness of the TNTI, was reported in previous studies of boundary layers (Eisma et al. Reference Eisma, Westerweel, Ooms and Elsinga2015; de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017; Heisel et al. Reference Heisel, De Silva, Hutchins, Marusic and Guala2021). However, to the authors’ knowledge, this is the first time such behaviour is observed in pipe flow.
To further examine the scaling behaviour, we consider additional normalisations motivated by the mean momentum balance analysis (Priyadarshana et al. Reference Priyadarshana, Klewicki, Treat and Foss2007; Klewicki Reference Klewicki2013a
). The UMZs are separated by thin shear layers (Meinhart & Adrian Reference Meinhart and Adrian1995), which have also been referred to as vortical fissures, i.e. regions of concentrated vorticity (Priyadarshana et al. Reference Priyadarshana, Klewicki, Treat and Foss2007). The thickness of these vortical fissures is expected to scale with
$\delta _\omega ^+ \sim \textit{Re}_\tau ^{1/2}$
; at relatively low Reynolds numbers, this scaling is comparable to
$\lambda _T$
, whereas the difference becomes more pronounced as the Reynolds number increases (Klewicki Reference Klewicki2013a
). Following this argument, figure 9(
$d$
) shows
$\delta _\omega ^+/\textit{Re}_\tau ^{1/2}$
, corresponding to the vortical fissure scaling. As seen, this normalisation also shows a reasonable collapse, particularly in the outer region. In comparison with figure 9(
$b$
), the results suggest that
$\delta _\omega ^+$
follows a
$\textit{Re}_{\tau }^{1/2}$
scaling in the outer region (
$y_i/R \gt 0.3$
), whereas
$\delta _\omega /\lambda _T$
provides a better collapse closer to the wall. In this sense, the scaling associated with the vortical fissure width appears to be more relevant in the outer region. According to the vortical fissure organisation of UMZs (Klewicki Reference Klewicki2013a
), the scaling with
$\delta _\omega ^+ \sim \textit{Re}_\tau ^{1/2}$
is expected to become more evident at higher Reynolds numbers. In the present Reynolds-number range, this behaviour appears to emerge in regions where the shear-layer thickness becomes relatively small compared with the UMZ thickness, particularly for
$y_i/R \gt 0.3$
. This result implies that viscous action remains important even in the outer region, and that due to the spatial organisation of the UMZs, localised viscous effects can play a significant role in momentum transport within those shear layers in the outer region. Note that figure 9(
$f$
) shows
$\delta _\omega /\eta$
, where
$\eta$
denotes the local Kolmogorov length scale. In this case,
$\delta _\omega /\eta$
does not exhibit a clear collapse and instead increases with
$\textit{Re}_\tau$
, indicating that
$\eta$
is not an appropriate scaling for
$\delta _\omega$
.
Additionally, it is observed that
$\delta _\omega$
increases with increasing
$y_i$
, indicating that the characteristic size of the vortical structures within the shear layers grows with wall-normal distance. As shown in figure 9(
$b$
),
$\delta _\omega /\lambda _T$
approaches a value of approximately 0.4, which corresponds to the typical thickness of the TNTI (da Silva et al. Reference da Silva, Hunt, Eames and Westerweel2014; Eisma et al. Reference Eisma, Westerweel, Ooms and Elsinga2015; Heisel et al. Reference Heisel, De Silva, Hutchins, Marusic and Guala2021). This suggests that the shear-layer characteristics in internal flows are analogous to those observed in TBLs. As seen in figure 9(
$e$
), the mean
$\delta _\omega ^{+}$
for
$\textit{Re}_\tau = 6000$
is approximately
$50$
–
$60$
, with smaller values observed at lower Reynolds numbers. This value provides a reference for the minimum UMZ thickness threshold (
$H_m^{+} \gt 50$
) introduced earlier in § 5.1, ensuring that only UMZs thicker than the representative shear-layer width are included in the analysis.
To further examine the structural organisation of the extracted wall-scaled UMZs, we analyse whether they follow the discrete hierarchy model proposed by Klewicki et al. (Reference Klewicki, Philip, Marusic, Chauhan and Morrill-Winter2014) and Klewicki (Reference Klewicki2021). According to this model, the ratio
$y_L/y_U$
approaches a constant value of approximately 0.618, where
$y_L$
and
$y_U$
denote the lower and upper edges of the UMZs, respectively. Figure 10(
$a$
) shows the relationship between
$y_L$
and
$y_U$
. As seen, the wall-scaled UMZs display a clearer linear relation that is close to
$y_U=1.62\,y_L$
, consistent with the discrete hierarchy model. This behaviour becomes more pronounced with increasing Reynolds number. Figure 10(
$b$
) further shows that the ratio
$y_L/y_U$
as a function of
$y_L$
. The ratio approaches 0.618 and remains nearly constant for
$y_L/R \gtrsim 0.2$
. Moreover, the region where this approximately constant ratio is observed lies in the outer region; in terms of
$y_i$
, this corresponds to
$y_i/R \gtrsim 0.25$
–
$0.3$
, which is consistent with the range identified in figure 9(
$d$
) where
$\delta _\omega ^+$
scales with the vortical fissure width. In contrast, the absence of a constant ratio for
$y_L/R \lesssim 0.2$
is associated with the fact that, in this region,
$\delta _\omega$
is better scaled by the Taylor microscale
$\lambda _T$
(see figure 9
$b$
). This indicates that the velocity-jump layer at the UMZ edge near the wall is more appropriately characterised as a shear layer, while it progressively acquires vortical-fissure-like features farther from the wall. In this sense, the extracted wall-scaled UMZs not only satisfy wall scaling but also remain consistent with a geometrical progression, supporting their interpretation within a hierarchical framework.
(
$a$
) The relationship between the lower edge
$y_{L}$
and the upper edge
$y_{U}$
. The dashed line indicates
$y_{U}=1.62y_L$
. (
$b$
) The ratio
$y_{L}/y_{U}$
as a function of
$y_{L}$
. In (a,b), the grey lines indicate all UMZs, and the coloured symbols indicate the wall-scaled UMZs at each
$\textit{Re}_\tau$
. The inset in (
$b$
) shows the fraction of the wall-scaled UMZs relative to all UMZs,
$N_s/N_{{all}}$
, as a function of
$\textit{Re}_{\tau }$
.

While the wall-scaled UMZs exhibit a clear hierarchical organisation consistent with the discrete model, it is also important to assess their relative significance within the overall UMZ population. To this end, we quantify the fraction of self-similar UMZs,
$N_s/N_{{all}}$
, as shown in the inset of figure 10(
$b$
). Although the fraction is relatively small at low Reynolds numbers, it increases with
$\textit{Re}_\tau$
, approximately following a log–linear trend. At
$\textit{Re}_\tau = 6000$
, the fraction reaches approximately 30 %, indicating that the wall-scaled UMZs constitute a substantial and non-negligible subset of the total population. Nevertheless, non-wall-scaled UMZs remain present in significant numbers, and their inclusion tends to obscure the logarithmic behaviour predicted by the AEH. These results highlight that isolating the wall-scaled UMZs through appropriate filtering is essential for clearly identifying the expected asymptotic behaviour.
6. Turbulence statistics conditioned on wall-scaled UMZs
In this section, we conditionally average the turbulence statistics associated with the wall-scaled UMZs, in order to examine whether the asymptotic behaviours predicted by the AEH can be observed. As discussed in the Introduction, the spatial organisation of UMZs, characterised by the velocity jumps across interfaces separating regions of nearly uniform momentum, provides a physical basis for reconstructing the mean streamwise velocity (de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017), which is shown to contribute to the logarithmic law of the mean profile (de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017; Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020). Moreover, a stochastic modelling of UMZs enables the prediction of second-order statistics, such as the streamwise and wall-normal turbulence intensities and Reynolds shear stress (Ehsani et al. Reference Ehsani, Heisel, Li, Voller, Hong and Guala2024), as well as higher-order moments including the skewness and flatness of the streamwise velocity fluctuations (Bautista et al. Reference Bautista, Ebadi, White, Chini and Klewicki2019). However, as highlighted earlier, although previous studies reconstructed turbulence statistics based on UMZ properties, they did not examine whether these statistics exhibit the asymptotic behaviours predicted by the AEH. This may be partly due to the fact that not all UMZs within the logarithmic layer scale with
$y$
and
$u_\tau$
. Hence, the present study addresses this issue by reconstructing turbulence statistics solely from the wall-scaled UMZs to clarify which of these structures are responsible for the constant wall-normal turbulence intensity and the logarithmic variation of the wall-parallel turbulence intensity predicted by the AEH, along with the presence of the constant Reynolds shear stress region.
To compute the turbulence statistics associated with the wall-scaled UMZs, a conditional sampling procedure is employed. Let
$\varOmega _s$
denote the region occupied by the identified wall-scaled UMZs. An indicator function
$I_s(x,z;y,t)$
is defined as
\begin{equation} I_s(x,z;y,t) = \begin{cases} 1, & \text{if } (x,z;y,t) \in \varOmega _s \ (\text{wall-scaled UMZs}), \\ 0, & \text{otherwise}. \end{cases} \end{equation}
The conditional average of a quantity
$\phi$
is then given by
Accordingly, the conditionally averaged streamwise velocity of the wall-scaled UMZs is defined as
$\langle U_s \rangle$
. This represents the zonal mean velocity associated with the wall-scaled UMZs. The corresponding velocity fluctuations are defined relative to this conditional average as
$u_s = U - \langle U_s \rangle$
. Hence,
$\langle u_s^2 \rangle ^{+}$
represents the streamwise turbulence intensity within the wall-scaled UMZs. This definition of the zonal mean statistics is similar to that used in the works of Kovasznay, Kibens & Blackwelder (Reference Kovasznay, Kibens and Blackwelder1970) and Kwon et al. (Reference Kwon, Philip, de Silva, Hutchins and Monty2014). Similarly,
$\langle v_s^2 \rangle$
,
$\langle w_s^2 \rangle$
and
$\langle u_sv_s \rangle$
are computed in the same manner.
(
$a$
) The conditionally averaged streamwise velocity of the wall-scaled UMZs,
$\langle U_s \rangle ^+$
, at
$\textit{Re}_{\tau } = 550$
,
$1000$
,
$3000$
and
$6000$
. The dashed line indicates the logarithmic law,
$\langle U \rangle ^+ = 1/\kappa \ln (y^+) + B$
, where
$\kappa = 0.39$
and
$B = 4.3$
. (
$b$
) The logarithmic-law indicator function of
$\langle U_s \rangle ^+$
,
$\varXi _1 = y^+ \mathrm{d}\langle U_s \rangle ^+ / \mathrm{d}y^+$
. The inset shows
$\varXi _1$
plotted against
$y/R$
. (
$c$
–
$f$
) The indicator function at each Reynolds number. Here, the vertical lines denote
$y^+ = 3\textit{Re}_\tau ^{1/2}$
and
$0.2R^+$
. In (
$a$
–
$f$
), the grey lines indicate the unconditioned mean streamwise velocity,
$\langle U \rangle ^+$
, for comparison.

6.1. Mean streamwise velocity
First, we examine the conditionally averaged streamwise velocity within the wall-scaled UMZs,
$\langle U_s \rangle ^{+}$
, as shown in figure 11(
$a$
). For comparison, the unconditioned mean streamwise velocity profile,
$\langle U \rangle$
(shown as the grey line), is also plotted for each
$\textit{Re}_\tau$
. Since the wall-scaled UMZs are defined within the range
$50 \le y^{+} \le 0.4R^{+}$
,
$\langle U_s \rangle ^{+}$
is plotted only over this region. As seen,
$\langle U_s \rangle ^{+}$
follows the logarithmic law (dashed line) much more closely than the unconditioned mean profile
$\langle U \rangle$
, irrespective of
$\textit{Re}_\tau$
. Beyond
$y^{+} \gt 100$
,
$\langle U_s \rangle ^{+}$
is slightly lower than
$\langle U \rangle$
, resulting in a longer region where
$\langle U_s \rangle ^{+}$
aligns well with the dashed line. This suggests that the wall-scaled UMZs primarily correspond to low-momentum regions.
To examine the logarithmic law more precisely, the indicator function
$\varXi _1 = y^{+}\mathrm{d}\langle U_s \rangle ^{+}/\mathrm{d}y^{+}$
is evaluated, as shown in figure 11(
$b$
). We can see that
$\varXi _1 = y^{+}\mathrm{d}\langle U_s \rangle ^{+}/\mathrm{d}y^{+}$
exhibits a clear plateau corresponding to
$1/\kappa$
, confirming the presence of the logarithmic region, where
$\kappa$
denotes the von Kármán constant. In particular, a distinct plateau is observed even at lower
$\textit{Re}_\tau$
, indicating that the logarithmic behaviour persists across all Reynolds numbers considered. The von Kármán constant
$\kappa$
is approximately
$0.39$
, consistent with Marusic et al. (Reference Marusic, Monty, Hultmark and Smits2013). These results demonstrate that the wall-scaled UMZs are primarily responsible for the logarithmic mean velocity profile. This finding supports our hypothesis that, once non-wall-scaled UMZs are filtered out, the turbulence statistics could reveal the asymptotic behaviours predicted by the AEH even at low Reynolds numbers.
Next, we examine the upper and lower bounds of the observed logarithmic region. The inset of figure 11(
$b$
) shows the profiles of
$\varXi _1 = y^{+}\mathrm{d}\langle U_s \rangle ^{+}/\mathrm{d}y^{+}$
plotted against
$y/R$
. It can be seen that the upper bound of the logarithmic region is approximately
$0.2R$
(vertical dashed line) for all
$\textit{Re}_\tau$
. This can be more clearly identified in figure 11(c–f). The lower bound of the logarithmic region is observed at approximately
$y^{+} \approx 3\textit{Re}_\tau ^{1/2}$
, consistent with the findings of Marusic et al. (Reference Marusic, Monty, Hultmark and Smits2013). The range observed in the present study,
$3\textit{Re}_\tau ^{1/2} \lt y^{+} \lt 0.2R^{+}$
, also agrees well with the results of Hwang & Sung (Reference Hwang and Sung2019), who extracted self-similar wall-attached structures of the streamwise velocity
$u$
and demonstrated the existence of a logarithmic region through conditional averaging of the velocity carried by those structures. These results further support the view that the wall-attached structures of
$u$
are composed of multiple UMZs (Hwang & Sung Reference Hwang and Sung2018). A detailed investigation of the relationship between wall-attached structures of
$u$
and wall-scaled UMZs would therefore be of particular interest for future work.
While Heisel et al. (Reference Heisel, de Silva, Hutchins, Marusic and Guala2020) demonstrated that UMZs contribute to the logarithmic law of the mean velocity, their analysis was not specifically aimed at identifying the logarithmic region through the indicator function, nor did it explicitly discuss its wall-normal extent. In contrast, the present study provides direct quantitative evidence of the logarithmic region through the indicator function. Furthermore, the identified logarithmic region exhibits consistent scaling, with the lower bound scaling with
$\textit{Re}_\tau ^{1/2}$
and the upper bound scaling with
$R$
. In particular, the observed lower bound is consistent with the mean momentum balance framework, in which the onset of the inertial region is associated with
$y^+=O(\textit{Re}_{\tau }^{1/2})$
, as obtained from matched asymptotic expansions (Afzal Reference Afzal1982) and an invariant form of the mean momentum equation (Wei et al. Reference Wei, Fife, Klewicki and McMurtry2005; Fife et al. Reference Fife, Klewicki and Wei2009; Klewicki Reference Klewicki2013b
). The logarithmic region of
$\langle U_s \rangle ^{+}$
extends to
$y/R \approx 0.2$
, which is slightly higher than the upper bound of the logarithmic region for the mean velocity from mean momentum balance analyses, i.e.
$y/R \approx 0.125$
–
$0.15$
(Klewicki et al. Reference Klewicki, Philip, Marusic, Chauhan and Morrill-Winter2014; Klewicki Reference Klewicki2021). This difference can be attributed to the fact that
$\langle U_s \rangle ^{+}$
represents a conditional average over the wall-scaled UMZs, whereas the corresponding bounds from mean momentum balance analyses are based on unconditional statistics. Given that the shear-layer thickness
$\delta _\omega ^+$
scales with the Taylor microscale
$\lambda _T^+$
or the vortical fissure width
$\textit{Re}_\tau ^{1/2}$
(figure 9), viscous effects remain important within these layers. Since such shear layers are more prevalent near the wall, this may contribute to the
$\textit{Re}_\tau ^{1/2}$
scaling of the lower bound. This consistency across
$\textit{Re}_\tau$
further supports that the wall-scaled UMZs capture the essential features of the self-similar motions embodied in the AEH.
6.2. Streamwise and spanwise turbulence intensities
As discussed earlier, it is essential to examine whether the wall-scaled UMZs exhibit the asymptotic behaviours of the turbulence intensities predicted by the AEH. At sufficiently high
$\textit{Re}_\tau$
, the AEH predicts that both the streamwise and spanwise turbulence intensities follow logarithmic variations in the logarithmic region
where
$A_i$
and
$B_i$
are constants. While the logarithmic variation of the spanwise turbulence intensity was observed even at relatively low
$\textit{Re}_\tau$
(Jiménez & Hoyas Reference Jiménez and Hoyas2008; Sillero et al. Reference Sillero, Jiménez and Moser2013; Lee & Moser Reference Lee and Moser2015; Pirozzoli et al. Reference Pirozzoli, Romero, Fatica, Verzicco and Orlandi2021), that of the streamwise component was only identified at extremely high
$\textit{Re}_\tau$
(Hultmark et al. Reference Hultmark, Vallikivi, Bailey and Smits2012; Marusic et al. Reference Marusic, Monty, Hultmark and Smits2013; Örlü et al. Reference Örlü, Fiorini, Segalini, Bellani, Talamelli and Alfredsson2017; Ono et al. Reference Ono, Furuichi and Tsuji2023). Previous UMZ-based studies (de Silva et al. Reference de Silva, Philip, Hutchins and Marusic2017; Bautista et al. Reference Bautista, Ebadi, White, Chini and Klewicki2019; Heisel et al. Reference Heisel, de Silva, Hutchins, Marusic and Guala2020) showed that UMZs contribute to the logarithmic law of the mean velocity. However, within the UMZ-based framework, the logarithmic behaviour of the streamwise and spanwise turbulence intensities has not been explicitly established. While using a two-state model consisting of an inertial self-similar region and a free-stream state, Krug et al. (Reference Krug, Yang, de Silva, Ostilla-Mónico, Verzicco, Marusic and Lohse2017) showed that the logarithmic behaviour of
$\langle U \rangle$
and
$\langle u^2 \rangle$
can be reproduced through a velocity jump across a stochastic interface separating the two states. However, in their approach, the logarithmic variation within the inertial region is prescribed a priori, rather than emerging from the underlying flow structures. Given that hairpin packets induce UMZs and that the spatial organisation of these UMZs accommodates all velocity components (Adrian et al. Reference Adrian, Meinhart and Tomkins2000), it is of particular interest to investigate whether the wall-scaled UMZs defined in the present study exhibit the logarithmic variation of the wall-parallel turbulence intensity.
(
$a$
) The conditionally averaged streamwise turbulence intensity of the wall-scaled UMZs,
$\langle u_s^2 \rangle ^+$
, at
$\textit{Re}_\tau = 550$
,
$1000$
,
$3000$
and
$6000$
. (
$b$
) The logarithmic-law indicator function
$\varXi _2 = y{\rm d}\langle u_s^2 \rangle ^+ / {\rm d}y$
. (
$c$
–
$f$
) The indicator function for each Reynolds number. Here, the vertical lines represent the range where a plateau (
$\varXi _2 = -0.95 = -A_1$
) occurs. In (
$a$
–
$f$
), the grey lines indicate the unconditioned streamwise turbulence intensity,
$\langle u^2 \rangle ^+$
, for comparison.

Figure 12. Long description
Panel A: A line graph shows the conditionally averaged streamwise turbulence intensity of wall-scaled UMZs at various Reynolds numbers. The x-axis represents y/R on a logarithmic scale, and the y-axis represents the turbulence intensity. Different colored dots represent different Reynolds numbers: black for 550, green for 1000, blue for 3000, and red for 6000. The grey lines indicate the unconditioned streamwise turbulence intensity for comparison. Panel B: A line graph displays the logarithmic-law indicator function. The x-axis represents y/R on a logarithmic scale, and the y-axis represents the indicator function. Different colored lines represent different Reynolds numbers: black for 550, green for 1000, blue for 3000, and red for 6000. Panel C: A line graph shows the indicator function for a Reynolds number of 550. The x-axis represents y/R on a logarithmic scale, and the y-axis represents the indicator function. The black dots represent the data points. Panel D: A line graph shows the indicator function for a Reynolds number of 1000. The x-axis represents y/R on a logarithmic scale, and the y-axis represents the indicator function. The green dots represent the data points. The vertical lines indicate the range where a plateau occurs. Panel E: A line graph shows the indicator function for a Reynolds number of 3000. The x-axis represents y/R on a logarithmic scale, and the y-axis represents the indicator function. The blue dots represent the data points. The vertical lines indicate the range where a plateau occurs. Panel F: A line graph shows the indicator function for a Reynolds number of 6000. The x-axis represents y/R on a logarithmic scale, and the y-axis represents the indicator function. The red dots represent the data points. The vertical lines indicate the range where a plateau occurs.
Figure 12(
$a$
) shows the wall-normal profile of the conditionally averaged streamwise turbulence intensity,
$\langle u_s^2\rangle ^+$
. Although the unconditioned streamwise turbulence intensity does not exhibit a clear logarithmic variation (see the grey lines inserted in figure 12
$a$
), the reconstructed intensity
$\langle u_s^{2} \rangle ^{+}$
reveals a distinct logarithmic region, particularly at
$\textit{Re}_\tau = 3000$
(blue) and
$\textit{Re}_\tau = 6000$
(red). To examine the logarithmic variation, the indicator function
$\varXi _2 = y{\rm d}\langle u_s^2 \rangle ^+ / {\rm d}y$
is plotted in figure 12(
$b$
). As seen, a distinct plateau (denoted by the horizontal dashed line,
$\varXi _2 = -0.95 = -A_1$
) is observed, indicating the presence of a logarithmic variation in
$\langle u_s^{2} \rangle$
.
To further examine the range over which this plateau occurs, the indicator function
$\varXi _2$
at each
$\textit{Re}_\tau$
is plotted in figure 12(
$c$
–
$f$
). At
$\textit{Re}_\tau = 550$
(figure 12
$c$
), no clear plateau is observed, while at
$\textit{Re}_\tau \ge 1000$
, a distinct plateau appears and its extent increases with
$\textit{Re}_\tau$
. Interestingly, the upper bound of the logarithmic variation is located at
$y/R \approx 0.2$
, which coincides with that of
$\langle U_s \rangle ^{+}$
shown in figure 11. The lower bound of
$\langle u_s^{2} \rangle ^{+}$
, on the other hand, scales approximately as
$y^{+} \approx 5\textit{Re}_\tau ^{1/2}$
, which lies higher than that of
$\langle U_s \rangle ^{+}$
. This could be attributed to the fact that, in the outer part of the logarithmic layer, inertial effects become increasingly dominant. Given the inviscid-flow assumption inherent in the AEH (
$\textit{Re}_\tau \rightarrow \infty$
), this region corresponds to where the asymptotic behaviour is expected to emerge. This observation is also consistent with measurements at extremely high Reynolds numbers,
$\textit{Re}_\tau = O(10^4)$
(Vallikivi et al. Reference Vallikivi, Ganapathisubramani and Smits2015; Ono et al. Reference Ono, Furuichi and Tsuji2023). It is worth noting that, despite the relatively low Reynolds numbers considered in the present study, such an asymptotic region is still observed. Moreover, both the lower and upper bounds of the logarithmic region exhibit consistent scaling, highlighting the robustness of the asymptotic behaviours predicted by the AEH even at moderate
$\textit{Re}_\tau$
.
In addition, it is worth highlighting the constant
$A_1 = 0.95$
remains essentially the same across all Reynolds numbers considered, indicating its Reynolds-number independence. The constant
$A_1$
corresponds to the so-called Townsend–Perry constant. In high-Reynolds-number experimental studies based on the unconditioned streamwise turbulence intensity
$\langle u^{2} \rangle$
, reported values include
$A_1 = 1.24$
(Hultmark et al. Reference Hultmark, Vallikivi, Bailey and Smits2012),
$A_1 = 1.30$
(Örlü et al. Reference Örlü, Fiorini, Segalini, Bellani, Talamelli and Alfredsson2017) and
$A_1 = 1.44$
(Ono et al. Reference Ono, Furuichi, Wada, Kurihara and Tsuji2022). However, studies that extracted only wall-scaled structures in the spectral space (Baars & Marusic Reference Baars and Marusic2020; Deshpande et al. Reference Deshpande, Monty and Marusic2021) reported a value of
$A_1 = 0.98$
, which is in close agreement with the present result. It should also be noted that other studies suggested that
$A_1$
may exhibit a weak dependence on Reynolds number (Hwang, Hutchins & Marusic Reference Hwang, Hutchins and Marusic2022), but the variation is minor, with
$A_1$
ranging approximately from
$1.01$
to
$1.09$
over
$\textit{Re}_\tau \approx 6000$
–
$20000$
. Hence, the spatial organisation of UMZs may also play an important role in understanding the spectral overlap argument proposed by Perry & Abell (Reference Perry and Abell1977) and Perry & Chong (Reference Perry and Chong1982).
(
$a$
) The conditionally averaged spanwise turbulence intensity of the wall-scaled UMZs,
$\langle w_s^2 \rangle ^+$
, at
$\textit{Re}_\tau = 550$
,
$1000$
,
$3000$
and
$6000$
. (
$b$
) The logarithmic-law indicator function,
$\varXi _3 = y{\rm d}\langle w_s^2 \rangle ^+ / {\rm d}y$
. (
$c$
–
$f$
) The indicator function for each Reynolds number. Here, the vertical lines represent the range where a plateau (
$\varXi _3 = -0.38 = -A_2$
) occurs. In (
$a$
–
$f$
), the grey lines indicate the unconditioned spanwise turbulence intensity,
$\langle w^2 \rangle ^+$
, for comparison.

Next, we explore the wall-normal profile of the conditionally averaged spanwise (i.e. azimuthal) turbulence intensity,
$\langle w_s^2\rangle ^+$
, in figure 13(
$a$
). Here,
$w_s$
denotes the spanwise velocity fluctuations within the wall-scaled UMZs, defined as
$w_s = W - \langle W_s \rangle$
. As seen,
$\langle w_s^2\rangle ^+$
exhibits a logarithmic variation similar to the unconditioned spanwise turbulence intensity
$\langle w^2 \rangle$
(grey lines), even at low
$\textit{Re}_\tau$
. Such behaviour is also evident in the indicator function
$\varXi _3 = y{\rm d}\langle w_s^2\rangle ^+/{\rm d}y$
shown in figure 13(
$b$
). All the indicator functions exhibit a plateau where
$\varXi _3 = -0.38$
(denoted by the horizontal dashed line), indicating a consistent slope (
$A_2=0.38$
) of
$\langle w_s^2 \rangle$
across a range of
$\textit{Re}_\tau$
. This slope differs among experiments, with reported values of
$A_2 = 0.64$
(Örlü et al. Reference Örlü, Fiorini, Segalini, Bellani, Talamelli and Alfredsson2017),
$A_2 = 0.27$
(Baidya et al. Reference Baidya, Philip, Hutchins, Monty and Marusic2021), and
$A_2 = 0.46$
(Ono et al. Reference Ono, Furuichi and Tsuji2023), but it is similar to the DNS results such as
$A_2 = 0.387$
in channels (Lee & Moser Reference Lee and Moser2015) and
$A_2 = 0.40$
in pipes (Pirozzoli et al. Reference Pirozzoli, Romero, Fatica, Verzicco and Orlandi2021).
The bounds of this logarithmic behaviour are further examined in figure 13(
$c$
–
$f$
), where the upper and lower limits of the logarithmic region are identified for each Reynolds number. As denoted by the vertical lines, the upper bound is found to scale with
$y/R = 0.4$
, while the lower bound scales with
$y^+ = 3\textit{Re}_\tau ^{1/2}$
. Compared with
$\langle u_s^2 \rangle$
, the upper bound extends further and the lower bound shifts closer to the wall, indicating that the logarithmic region of
$\langle w_s^2 \rangle$
spans a wider range in wall-normal distance. This difference in the range of the logarithmic variation may be attributed to the distinct spatial organisation of the
$u$
and
$w$
structures. The two-dimensional spectra of
$u$
and
$w$
in the logarithmic region show that the energetic ridge of the
$w$
spectrum at large scales corresponds to structures that are shorter in the streamwise direction but wider in the spanwise direction compared with those of
$u$
(Jiménez & Hoyas Reference Jiménez and Hoyas2008). Similarly, Baidya et al. (Reference Baidya, Philip, Hutchins, Monty and Marusic2021) observed that the self-similar behaviour of
$w$
structures exists at relatively smaller scales than that of
$u$
structures, which explains the presence of the logarithmic variation of
$\langle w_s^2 \rangle$
closer to the wall. The lower bound reported therein is also consistent with the present results. The upper bound observed at
$y/R = 0.4$
, which extends to a higher position than that of
$u$
, may be associated with the
$\pm 45^\circ$
alignment of the
$w$
structures in the wall-parallel plane, a feature that persists from the outer region down to the logarithmic region (Sillero, Jiménez & Moser Reference Sillero, Jiménez and Moser2014; de Silva et al. Reference de Silva, Kevin, Baidya, Hutchins and Marusic2018). In addition, from the mean momentum balance-based framework, the present range,
$3\textit{Re}_\tau ^{1/2}\lt y^+\lt 0.4R^+$
, is also similar to the self-similar domain proposed by Klewicki (Reference Klewicki2013b
), which begins near
$y^{+}\approx 2.6\sqrt {\delta ^{+}}$
and extends to
$y/\delta \approx 0.45$
. However, a further investigation is required to clarify this behaviour. Although the extent of the logarithmic variation differs between the
$u$
and
$w$
components, the wall-scaled UMZs reveal a common wall-normal region in which all of them exhibit the logarithmic behaviour. Notably, this region coincides with the range where the mean velocity satisfies the logarithmic law.
6.3. Wall-normal turbulence intensity and Reynolds shear stress
In this section, we examine the wall-normal turbulence intensity and the Reynolds shear stress carried by the wall-scaled UMZs. The AEH assumes the presence of a region of constant Reynolds shear stress in the limit of
$\textit{Re}_\tau \rightarrow \infty$
, and accordingly predicts that the wall-normal turbulence intensity becomes constant within this region (Townsend Reference Townsend1976). Therefore, it is of interest to analyse the wall-normal turbulence intensity and Reynolds shear stress carried by the wall-scaled UMZs to assess their consistency with the AEH.
The conditionally averaged wall-normal turbulence intensity and Reynolds shear stress of the wall-scaled UMZs,
$\langle v_s^2 \rangle ^+$
and
$-\langle u_{s}v_{s} \rangle ^+$
, respectively. The closed circles represent the conditional averages over all wall-scaled UMZs, while the open circles denote the corresponding quantities,
$\langle v_s^2 \rangle ^+_\gt$
and
$-\langle u_s v_s \rangle ^+_\gt$
, obtained for cases with
$N_{\textit{UMZ}} \ge \langle N_{\textit{UMZ}} \rangle = 5$
(which corresponds to
$\langle N_{\textit{UMZ}} \rangle$
at
$\textit{Re}_\tau = 6000$
). The grey lines indicate the unconditioned statistics,
$\langle v^2 \rangle ^+$
and
$-\langle uv \rangle ^+$
, for comparison. The vertical dashed lines inserted in each panel denote the peak location of
$-\langle uv \rangle$
(
$y_p$
) and the position of
$y/R = 0.1$
, respectively.

Figure 14 presents the corresponding conditional statistics. Here, the closed circles represent the wall-normal profiles of
$\langle v_s^2 \rangle ^+$
and
$-\langle u_s v_s \rangle ^+$
, while the grey lines indicate the corresponding unconditioned statistics. In general, the conditioned statistics (closed circles) collapse well with the unconditioned statistics (grey lines). This indicates that the spatial organisation of the UMZs reproduces the turbulence statistics accurately, not only for the mean streamwise velocity and streamwise turbulence intensity but also for the wall-normal turbulence intensity and the Reynolds shear stress. This finding is also consistent with the stochastic model of UMZs proposed by Ehsani et al. (Reference Ehsani, Heisel, Li, Voller, Hong and Guala2024), which was shown to capture a substantial part of the Reynolds shear stress and wall-normal turbulence intensity.
A closer examination reveals differences in their consistency with the AEH. For the wall-normal turbulence intensity (closed circles), a distinct constant region is observed at
$\textit{Re}_\tau = 6000$
(figure 14
$d$
). On the other hand, the constant region of
$-\langle u_s v_s \rangle$
(closed circles) is absent at
$\textit{Re}_\tau = 6000$
, even when the wall-scaled UMZs are conditionally sampled. This may be due to the still relatively low value of
$\textit{Re}_\tau$
, raising the question of whether a constant region would emerge at higher Reynolds numbers. However, it is worth noting that, from the invariant form of the mean momentum balance, the logarithmic mean velocity profile can be analytically derived even when the Reynolds shear stress varies with wall-normal position (Morrill-Winter, Philip & Klewicki Reference Morrill-Winter, Philip and Klewicki2017; Klewicki Reference Klewicki2021). Given the AEH, a higher
$N_{\textit{UMZ}}$
could imply a locally higher Reynolds number. To examine this, the conditional statistics were computed for the wall-scaled UMZs with
$N_{\textit{UMZ}} \ge 5$
, corresponding to the mean value
$\langle N_{\textit{UMZ}} \rangle$
at
$\textit{Re}_\tau = 6000$
. The open circles denote the corresponding statistics (
$\langle v_s^2 \rangle _\gt$
and
$-\langle u_s v_s \rangle _\gt$
) conditioned on high
$N_{\textit{UMZ}}$
. As seen, the magnitude of
$\langle v_s^2 \rangle _\gt$
increases slightly but remains almost identical to that of
$\langle v_s^2 \rangle$
(closed circles), and a constant region is still evident. In contrast,
$-\langle u_s v_s \rangle _\gt$
exhibits a larger increase, which is consistent with the results of Laskari et al. (Reference Laskari, de Kat, Hearst and Ganapathisubramani2018), who reported that the Reynolds shear stress becomes significantly larger than the unconditioned statistics when a large number of zones are present in the flow. However,
$-\langle u_s v_s \rangle _\gt$
does not merely increase in magnitude; instead, a constant region appears within the range indicated by the vertical dashed lines, except at
$\textit{Re}_\tau = 550$
. These observations further support that the wall-scaled UMZs extracted in the present study are consistent with the structural framework of the AEH, and suggest that the absence of a constant region is associated with the relatively low Reynolds number. Moreover, the flow within regions of high
$N_{\textit{UMZ}}$
may locally exhibit high-Reynolds-number effects.
The present results confirm that the wall-scaled UMZs successfully reproduce the coexistence of the logarithmic variations in the wall-parallel turbulence intensities over the logarithmic regions of the mean velocity. This is consistent with the prediction of the AEH, indicating that the extracted wall-scaled UMZs retain the essential characteristics of hierarchically organised attached eddies. However, it remains to be determined whether the velocity fluctuations within these UMZs also retain invariant PDFs across
$\textit{Re}_\tau$
and
$y$
. In the following section, this aspect is examined by analysing the PDFs of velocity fluctuations within the wall-scaled UMZs.
7. Probability density functions of velocity fluctuations within wall-scaled UMZs
The results presented in the previous section demonstrate that the velocity fluctuations within the wall-scaled UMZs can reproduce the logarithmic variations of the wall-parallel turbulence intensities. A natural question that follows is whether the velocity fluctuations within these UMZs also exhibit self-similar PDFs. Therefore, in this section, we quantitatively examine whether such invariance is indeed observed.
The self-similar behaviour of the PDFs of velocity fluctuations was examined in several previous studies. For example, Andreopoulos et al. (Reference Andreopoulos, Durst, Zaric and Jovanovic1984) observed that the skewness and flatness factors remain nearly constant in the logarithmic region. In particular, Tsuji & Nakamura (Reference Tsuji and Nakamura1999) demonstrated a connection between the mean velocity profile and the self-similarity of the PDFs, which was later confirmed in high-Reynolds-number TBLs (Lindgren, Johansson & Tsuji Reference Lindgren, Johansson and Tsuji2004; Tsuji, Lindgren & Johansson Reference Tsuji, Lindgren and Johansson2005). Recently, Furuichi, Ono & Tsuji (Reference Furuichi, Ono and Tsuji2025) analysed high-Reynolds-number pipe-flow data and examined the invariance of the PDFs from the AEH viewpoint. Nevertheless, it remains important to investigate such PDF invariance from a coherent-structure perspective. As Tsuji & Nakamura (Reference Tsuji and Nakamura1999) suggested, if coherent structures have the same shape independent of the Reynolds number, the PDFs of the velocity induced by these structures and the corresponding mean velocity may also exhibit a universal form. In the present study, this issue is addressed by analysing the PDFs of velocity fluctuations within the wall-scaled UMZs.
(a) The PDFs of the streamwise velocity fluctuations,
$u'(y) = u(y)/u_{\textit{rms}}(y)$
, at
$y/R = 0.1$
(dashed),
$0.15$
(dashed–dot) and
$0.2$
(solid) for
$\textit{Re}_\tau = 1000$
,
$3000$
and
$6000$
. (
$b$
) The PDFs of the streamwise velocity fluctuations within the wall-scaled UMZs,
$u_s'(y) = u_s(y)/u_{s,{rms}}(y)$
, at the same wall-normal positions and Reynolds numbers. (
$c$
) The wall-normal variation of the skewness
$S(u') = \langle u^3 \rangle /\langle u^2 \rangle ^{3/2}$
(grey line) and
$S(u_s') = \langle u_s^3 \rangle /\langle u_s^2 \rangle ^{3/2}$
(coloured line). The profiles corresponding to each
$\textit{Re}_\tau$
are shifted upward by
$\Delta S$
= 1. (
$d$
) The Kullback–Leibler (KL) divergence between the PDFs of the streamwise velocity fluctuations and the Gaussian distribution. The coloured and grey lines indicate the KL divergence for
$u_s'$
and
$u'$
, respectively. The KL divergence corresponding to each
$\textit{Re}_\tau$
is shifted upward by
$\Delta D_{u'} = 0.01$
.

Figure 15(a,b) compares the PDFs of the streamwise velocity fluctuations at three wall-normal locations (
$y/R = 0.1$
,
$0.15$
and
$0.2$
) over the logarithmic region. Here, the streamwise velocity fluctuations are normalised by the root-mean-square value at each location, and we define the normalised streamwise velocity fluctuations
$u'(y) = u(y)/u_{\textit{rms}}(y)$
for the unconditioned velocity (figure 15
$a$
) and
$u_s'(y) = u_s(y)/u_{s,{rms}}(y)$
for the wall-scaled UMZs (figure 15
$b$
). The bin size is
$\Delta u' = \Delta u_s' = 0.2$
. The case at
$\textit{Re}_\tau = 550$
is excluded from this analysis because its relatively low Reynolds number does not allow a clear logarithmic variation in the streamwise turbulence intensity (figure 12). Both PDFs appear to collapse reasonably well across different Reynolds numbers and wall-normal locations. However, as highlighted by the dashed box, the PDF of
$u_s'$
exhibits a noticeably better collapse than that of
$u'$
. This indicates that the streamwise velocity fluctuations carried by the wall-scaled UMZs exhibit a higher degree of self-similarity. These results further support a close connection between the self-similarity of the PDFs and the logarithmic behaviour of the streamwise turbulence intensity (Furuichi et al. Reference Furuichi, Ono and Tsuji2025), indicating that the wall-scaled UMZs play a central role in this relationship.
Next, we further examine this self-similarity by considering the skewness of the streamwise velocity fluctuations. In the presence of self-similarity, the skewness is expected to be nearly constant and close to zero across the logarithmic region (Andreopoulos et al. Reference Andreopoulos, Durst, Zaric and Jovanovic1984; Tsuji & Nakamura Reference Tsuji and Nakamura1999). Note that Bautista et al. (Reference Bautista, Ebadi, White, Chini and Klewicki2019) modelled the wall-normal profile of
$U$
based on the spatial organisation of UMZs and demonstrated that their model reproduces not only the streamwise turbulence intensity but also the sub-Gaussian behaviour of the skewness and kurtosis observed in the DNS data of Lee, Ahn & Sung (Reference Lee, Ahn and Sung2015). Figure 15(
$c$
) shows the wall-normal variation of the skewness,
$S(u')$
(grey line) and
$S(u_s')$
(coloured line). The skewness of the unconditioned streamwise velocity
$S(u')$
does not attain a constant value and remains negative throughout the logarithmic region (indicated by the vertical lines). In contrast, the skewness associated with the wall-scaled UMZs
$S(u_s')$
exhibits a clear plateau within the logarithmic region, with values remaining close to zero. The near-constant skewness indicates that the degree of asymmetry of the PDF is invariant with respect to the wall-normal distance within the logarithmic region. In particular, skewness values close to zero indicate that the PDF is approximately symmetric with respect to
$u'$
.
To quantify the difference in the PDF quantitatively, we compute the Kullback–Leibler (KL) divergence between the PDFs of the streamwise velocity fluctuations and the Gaussian distribution. From the viewpoint of the AEH, the logarithmic region is populated by wall-attached structures of the same geometric scale, which are assumed to be statistically independent. When the velocity contributions of such structures are sufficiently superposed, the central limit theorem suggests that the PDF of the velocity fluctuations approaches the Gaussian distribution (Meneveau & Marusic Reference Meneveau and Marusic2013). It is therefore instructive to examine how close the PDFs are to the Gaussian shape in order to evaluate whether the self-similar behaviour in the logarithmic region is realised in the present data. For this purpose, a zero-mean, unit-variance Gaussian distribution
$P_G(u') = (2\pi )^{-1/2}\exp (-u'^2/2)$
is adopted as a reference, and the KL divergence is defined as follows:
\begin{equation} D_{u'}(P_y \Vert P_G) = \sum _u P_y(u'(y))\, \log \left [ \frac {P_y\bigl (u'(y)\bigr )}{P_G\bigl (u'(y)\bigr )} \right ]\!. \end{equation}
The KL divergence is used here as a quantitative indicator of the closeness between the PDF of the velocity fluctuations
$P_y$
and the reference Gaussian distribution
$P_G$
(Tsuji & Nakamura Reference Tsuji and Nakamura1999; Tsuji et al. Reference Tsuji, Lindgren and Johansson2005). In other words, smaller values of the KL divergence indicate that the PDF at a given
$y$
is closer to the Gaussian distribution.
Figure 15(
$d$
) presents the wall-normal profiles of the KL divergence for the unconditioned data (
$D_{u'}$
, grey line) and the wall-scaled UMZs (
$D_{u_s'}$
, coloured line). For the unconditioned case,
$D_{u'}$
does not exhibit a clear plateau across the logarithmic region for any of the three Reynolds numbers considered. In particular, for
$\textit{Re}_\tau = 3000$
and
$6000$
, the KL divergence increases from the lower edge of the logarithmic region to the outer region, indicating that the PDF gradually departs from a Gaussian shape as
$y/R$
approaches the outer region. This behaviour reflects the persistent influence of large-scale and detached motions present in the unconditioned field, even within the logarithmic region. In contrast,
$D_{u_s'}$
exhibits an approximately constant value across the logarithmic region regardless of
$\textit{Re}_\tau$
, indicating an invariant PDF shape.
It is worth highlighting that Furuichi et al. (Reference Furuichi, Ono and Tsuji2025) conducted the KL divergence analysis for
$u'$
at
$\textit{Re}_\tau \approx 4000$
–
$20000$
and observed a constant region only for
$\textit{Re}_\tau \gtrsim 10^4$
. However, the present results reveal the emergence of such a constant region even at lower
$\textit{Re}_\tau$
. A similar constant region of the KL divergence is also observed for the wall-normal and spanwise velocity fluctuations, and the extent of the constant region is consistent across all three velocity components (see Appendix D for details). In this regard, the fact that figure 15 exhibits the same wall-normal range as figure 12 suggests that the PDF shape also becomes nearly invariant with wall-normal position over the range where the logarithmic variation of
$\langle u_s^2 \rangle ^+$
is observed. Together with the skewness plateau in figure 15(
$c$
), these results indicate that, under the wall-scaled UMZ condition, the streamwise velocity fluctuations exhibit an approximately Gaussian and height-invariant distribution throughout the logarithmic region.
Based on the PDF analysis presented in figure 15, the PDFs of
$u_s'$
exhibit a high degree of self-similarity in the wall-normal direction. To quantify the statistical differences between the PDFs at different wall-normal locations, we analyse the two-point KL divergence, defined as
\begin{equation} D_{u'}\bigl (P_{y_1}\Vert P_{y_2}\bigr ) = \sum _{u'} P_{y_1}(u'(y_1))\,\log \frac {P_{y_1}(u'(y_1))}{P_{y_2}(u'(y_2))}. \end{equation}
Here,
$D_{u'}$
quantifies the extent to which the PDF at
$y = y_1$
differs from that at another wall-normal location
$y = y_2$
(Tsuji & Nakamura Reference Tsuji and Nakamura1999; Tsuji et al. Reference Tsuji, Lindgren and Johansson2005). A smaller value of
$D_{u'}$
therefore corresponds to a higher degree of similarity between the two PDFs. To ensure statistical convergence, the analysis is restricted to the range
$-6 \lt u' \lt 6$
, and a bin is included in the summation only when samples exist at both wall-normal locations. An analogous definition is adopted for
$u_s'$
, yielding the corresponding KL divergence
$D_{u_s'}$
.
The contours of the two-point KL divergence between the PDFs of the streamwise velocity fluctuations at two wall-normal locations (
$y_1$
and
$y_2$
) for
$\textit{Re}_\tau = 3000$
and
$6000$
. The panels (a,b) show
$D_{u'}(P_{y_1}\Vert P_{y_2})$
for the unconditioned field at
$\textit{Re}_\tau = 3000$
and
$6000$
, respectively. The panels (c,d) show
$D_{u_s'}(P_{y_1}\Vert P_{y_2})$
for the wall-scaled UMZ samples at the same Reynolds numbers. In (
$a$
–
$d$
), the lighter colours indicate smaller values of KL divergence and the solid line indicates the value of 0.001. (e, f) The difference
$\Delta D = D_{u_s'} - D_{u'}$
at
$\textit{Re}_\tau = 3000$
and
$6000$
. Here, the blue contours indicate negative values of
$\Delta D$
, indicating that
$D_{u_s'} \lt D_{u'}$
. The inserted dashed lines mark the lower and upper bounds of the logarithmic region.

Figure 16. Long description
The image contains six contour plots labeled (a) through (f). Each plot shows the two-point KL divergence between the PDFs of streamwise velocity fluctuations at two wall-normal locations (y1 and y2) for different Reynolds numbers (Re_tau). Panel A and Panel B display the unconditioned field at Re_tau values of 3000 and 6000, respectively. Lighter colors indicate smaller values of KL divergence, with a solid line marking the value of 0.001. Panel C and Panel D show the wall-scaled UMZ samples at the same Reynolds numbers. Panel E and Panel F illustrate the difference in KL divergence at Re_tau values of 3000 and 6000, with blue contours indicating negative values, suggesting that the divergence is smaller in the logarithmic region. Dashed lines mark the lower and upper bounds of the logarithmic region.
Figure 16(a,b) shows
$D_{u'}(P_{y_1}\Vert P_{y_2})$
for the unconditioned streamwise velocity fluctuations,
$u'$
, at
$\textit{Re}_\tau = 3000$
and
$6000$
. The case at
$\textit{Re}_\tau = 1000$
is omitted here to avoid repetition. By definition of the two-point KL divergence, the value is zero along the diagonal line
$y_1 = y_2$
. In the off-diagonal region where
$y_1 \neq y_2$
, the contours are generally dark, and the regions of small KL divergence are very limited. This indicates that, except for pairs of nearly identical wall-normal locations, the unconditioned PDFs at different heights do not exhibit similar shapes. Even within the logarithmic region, the KL divergence increases rapidly as
$(y_1, y_2)$
moves away from the diagonal (see the inserted solid contour line for reference). This behaviour demonstrates that the unconditioned PDFs vary strongly with wall-normal position and that self-similarity at the level of the full distribution is not well established.
Figure 16(c,d) presents
$D_{u_s'}(P_{y_1}\Vert P_{y_2})$
for the wall-scaled UMZs. A broad region of lighter colours is observed within the logarithmic region. In particular, comparison with the inserted solid contour line in
$D_{u'}$
reveals that regions of low
$D_{u_s'}$
are distributed over a much wider range of
$(y_1, y_2)$
. This indicates that the PDFs at different wall-normal locations exhibit a higher degree of agreement for the wall-scaled UMZs than for the unconditioned field. Notably, the PDFs near the lower and upper bounds of the logarithmic region remain closely similar. In other words, conditioning on wall-scaled UMZs significantly enhances the self-similarity of the distributions across the logarithmic region.
To further examine this difference, figure 16(e, f) shows the difference
$\Delta D = D_{u_s'} - D_{u'}$
. Here, blue (negative) contours indicate regions where the KL divergence is smaller for the wall-scaled UMZs than for the unconditioned field, i.e. where the similarity between the two PDFs is enhanced by the UMZ conditioning. As seen, the regions of negative
$\Delta D$
appear predominantly in off-diagonal locations corresponding to
$(y_1, y_2)$
spanning the logarithmic region (approximately
$y_1^+ \approx 100$
–
$200$
and
$y_2/R \approx 0.1$
–
$0.2$
). This confirms that the PDFs obtained from the wall-scaled UMZ samples maintain a substantially higher degree of self-similarity between different wall-normal locations than those from the unconditioned field.
8. Summary and conclusions
We analyse DNS data of turbulent pipe flow at
$\textit{Re}_\tau = 550$
,
$1000$
,
$3000$
and
$6000$
to examine the organisation of UMZs, with particular emphasis on identifying wall-scaled UMZs and analysing the associated turbulence statistics within the framework of the AEH. The UMZs and their interfaces are detected using a histogram-based approach. We observe that the number of UMZs,
$N_{\textit{UMZ}}$
, exhibits a log–linear increase with
$\textit{Re}_\tau$
, while the corresponding interface velocities span a wide range, reflecting the hierarchical nature of UMZs in turbulent pipe flows.
By measuring the thickness of each UMZ and the associated velocity jump across its interfaces, we conditionally sample those UMZs whose thickness (
$H_m$
) scales with the wall-normal distance
$y$
and whose velocity jump (
$\Delta U$
) scales with the friction velocity
$u_\tau$
. These UMZs are defined as wall-scaled UMZs and are found to be geometrically self-similar, with a mean thickness scaling as
$\langle H_m \rangle \approx 0.75y$
, and to exhibit a nearly constant velocity jump
$\Delta U \approx 1.5u_\tau$
across the logarithmic region, independent of
$\textit{Re}_\tau$
. These results indicate that the wall-scaled UMZs identified in the present study possess a zonal-like organisation and represent key turbulent structures associated with the AEH. In addition, the thickness of the shear layers for the wall-scaled UMZs is found to scale with the Taylor microscale
$\lambda _T^+$
, while tending towards a
$\textit{Re}_\tau ^{1/2}$
scaling, consistent with the vortical fissure width, in the outer region. This transition suggests that the velocity-jump layers at the UMZ edges exhibit shear-layer characteristics close to the wall, while progressively acquiring vortical-fissure-like features farther from the wall. Consequently, viscous effects remain relevant within the thin shear layers separating adjacent zones, even in the outer region, owing to the spatial organisation of the UMZs. Furthermore, the wall-scaled UMZs are shown to follow a discrete hierarchical organisation, with the ratio between their lower and upper bounds approaching a constant value consistent with a geometrical progression, particularly in the outer region. Taken together, these findings support that the extracted UMZs not only satisfy wall scaling but also exhibit a hierarchical structure consistent with the mean momentum balance framework.
Conditional averaging of the turbulence statistics associated with the wall-scaled UMZs reveals that the mean streamwise velocity and the streamwise and spanwise turbulence intensities exhibit clear logarithmic variation, which is absent in the unconditioned results, particularly for the streamwise component. These logarithmic behaviours are observed over a common wall-normal range,
$5\textit{Re}_\tau ^{1/2} \lt y^+ \lt 0.2R^+$
. The wall-normal turbulence intensity and the Reynolds shear stress exhibit approximately constant regions only when the number of UMZs exceeds its mean value. In addition, the PDFs of the velocity fluctuations within the wall-scaled UMZs are examined in order to test the invariance of the associated statistics. The PDFs of the wall-scaled UMZs collapse over the logarithmic region, regardless of
$\textit{Re}_\tau$
. Moreover, the skewness of all the velocity components exhibits a clear constant region with values close to zero, indicating height-invariant symmetry of the PDFs. Furthermore, a KL divergence analysis quantitatively confirms that the PDFs associated with the wall-scaled UMZs remain invariant across the logarithmic region.
The present results demonstrate that the wall-scaled UMZs successfully reproduce the coexistence of logarithmic variations in the wall-parallel turbulence intensities over the same wall-normal range as the logarithmic region of the mean velocity. This behaviour is consistent with the predictions of the AEH, indicating that the extracted wall-scaled UMZs retain the essential features of hierarchically organised attached eddies. The present findings further show that not all UMZs within the logarithmic layer obey wall scaling with
$y$
and
$u_\tau$
. By reconstructing turbulence statistics exclusively from the wall-scaled UMZs, this study isolates the specific subset of coherent structures responsible for the asymptotic behaviour of turbulence statistics. These results support a close connection between the self-similarity of the velocity-fluctuation PDFs and the logarithmic behaviour of the streamwise turbulence intensity, highlighting the central role of the wall-scaled UMZs in bridging turbulent structure and statistics.
While the present study establishes the statistical and structural significance of wall-scaled UMZs within the framework of the AEH, several open questions remain. Future work will therefore focus on a detailed examination of the relationship between wall-attached and detached structures of
$u$
(Hwang & Sung Reference Hwang and Sung2018; Yoon et al. Reference Yoon, Hwang, Yang and Sung2020) and the wall-scaled UMZs. In addition, an investigation of the lifetime and temporal evolution of UMZs across a wide range of
$\textit{Re}_\tau$
is expected to provide further insight into their dynamical significance, with particular emphasis on the meandering behaviour of wall-coherent structures (Hwang & Lee Reference Hwang and Lee2022) and their role within the attached-eddy framework.
Acknowledgements
The authors acknowledge the support from the Supercomputing Centre of the Korea Institute of Science and Technology Information (KISTI). Constructive criticisms from the anonymous referees are also gratefully acknowledged.
Funding
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (Grant No. RS-2023-00211896).
Declaration of interests
The authors report no conflict of interest.
Appendix A. Subdomain size effect in UMZ detection
The robustness of the UMZ detection procedure is assessed by examining its sensitivity to the spanwise subdomain size
$l_z^+$
. Figure 17 shows the variation of the mean number of detected UMZs,
$\langle N_{\textit{UMZ}} \rangle$
, according to
$l_z^+$
at each
$\textit{Re}_\tau$
. For all cases, the number of velocity vectors in the subdomain is fixed at approximately 5000. It is observed that
$\langle N_{\textit{UMZ}} \rangle$
decreases rapidly with increasing
$l_z^+$
when
$l_z^+$
is small. However, for
$l_z^+ \gtrsim 2000$
,
$\langle N_{\textit{UMZ}} \rangle$
becomes nearly independent of
$l_z^+$
, particularly at higher Reynolds numbers. This behaviour suggests that, beyond this threshold, the detection of UMZs is insensitive to further increases in the spanwise subdomain size.
The variation of the mean number of UMZs,
$\langle N_{\textit{UMZ}} \rangle$
, according to the subdomain size
$l_z^+$
.

The conditionally averaged (a) mean velocity
$\langle U_s \rangle ^+$
and (b) streamwise turbulence intensity
$\langle u_s'^2 \rangle ^+$
from wall-scaled UMZs, extracted using different domain lengths
$l_z^+ = 1000, 2000$
and
$3000$
at
$\textit{Re}_{\tau } = 550$
(black) and
$1000$
(green).

Although the total number of detected UMZs exhibits some sensitivity to the azimuthal sampling length, it is important to assess whether this affects the statistics of the extracted wall-scaled UMZs. To this end, figure 18 compares the conditionally averaged mean velocity,
$\langle U_s \rangle ^+$
, and the streamwise turbulence intensity,
$\langle u_s'^2 \rangle ^+$
, conditionally averaged from wall-scaled UMZs using three different subdomain sizes,
$l_z^+ = 1000$
,
$2000$
and
$3000$
, for
$\textit{Re}_\tau = 550$
and
$1000$
, where the sensitivity of
$N_{\textit{UMZ}}$
to
$l_z$
is most pronounced. Despite the differences in the number of the detected UMZs, the turbulence statistics exhibit excellent agreement across all three cases. The profiles of
$\langle U_s \rangle ^+$
closely follow the logarithmic law irrespective of the subdomain size, while the streamwise turbulence intensities display consistent near-wall peaks and similar decay behaviour in the outer region. These results demonstrate that the statistical properties of the wall-scaled UMZs are robust to variations in the azimuthal sampling length.
Appendix B. Effect of detection criteria on geometrically self-similar UMZs
(a,b) Influence of the detection criteria on
$\langle H_m \rangle$
corresponding to
$R_2$
to
$R_{max -1}$
, obtained by varying (
$a$
) the slope and (
$b$
) the offset. The dashed line denotes
$\langle H_m \rangle = 0.75\,y_m$
. (c,d) Influence of the detection criteria on
$\langle u_s^2 \rangle ^{+}$
of the wall-scaled UMZs, obtained using the same criteria as in (a,b), respectively. The inset in each panel shows the corresponding logarithmic-law indicator function,
$\varXi _2$
. The grey line indicates the corresponding unconditioned result.

Figure 19. Long description
The image contains four panels of scatter plots and line graphs. Panel A: A scatter plot showing the influence of detection criteria on the relationship between Hm/R and ym/R. The x-axis is labeled y_m/R, and the y-axis is labeled H_m/R. Different colored dots represent different criteria for Hm/R and ym/R. A dashed line denotes a specific value. Panel B: A similar scatter plot with different criteria for Hm/R and ym/R. Panel C: A line graph showing the influence of detection criteria on the wall-scaled UMZs. The x-axis is labeled y/R, and the y-axis is labeled u’s^2+. Different colored lines represent different criteria. An inset shows the logarithmic-law indicator function. Panel D: A similar line graph with different criteria for Hm/R and ym/R. The grey line indicates the unconditioned result.
To examine the influence of the detection criteria for geometrically self-similar UMZs introduced in § 5.1, the mean thickness,
$\langle H_m \rangle$
, and the corresponding streamwise turbulence intensity,
$\langle u_s^2 \rangle ^{+}$
, are plotted in figure 19. As discussed in § 5.1, for the UMZs from
$R_2$
to
$R_{{max}-1}$
, both self-similar and non-self-similar UMZs coexist. To extract the self-similar UMZs, we apply the criterion
$H_m/R \gt 0.75 y_m/R - 0.15$
. Figures 19(
$a$
) and 19(
$b$
) show the influence of the slope and offset, respectively. As shown in figure 19(
$a$
), for
$y_m/R \lesssim 0.2$
, all cases collapse onto a linear relationship
$\langle H_m \rangle \approx 0.75 y_m$
, regardless of the slope value. However, beyond
$y_m/R \approx 0.2$
, this collapse is lost unless the slope is close to
$0.75$
. When the slope is smaller than
$0.75$
, relatively thin UMZs are included, whereas larger values of the slope select excessively thick UMZs. In both cases, the resulting structures do not follow a consistent linear scaling with
$y_m$
, indicating that a slope of approximately
$0.75$
characterises the self-similar UMZs. The effect of the offset is shown in figure 19(
$b$
). The offset controls the exclusion of relatively thin UMZs, particularly in the outer region. For all offset values,
$\langle H_m \rangle$
follows the linear relationship up to
$y_m/R \approx 0.2$
. Beyond
$y_m/R \approx 0.2$
, the choice of offset affects the inclusion of thin UMZs, leading to deviations from the linear trend. When the offset is close to 0.15, the linear scaling is preserved over an extended range of
$y_m$
. Although the offset value of 0.15 may appear somewhat arbitrary, it is chosen to effectively exclude thin UMZs that do not follow the self-similar scaling, as discussed in § 5.
We next examine how the slope and offset criteria influence the turbulence statistics. For this purpose, the conditionally averaged streamwise turbulence intensity,
$\langle u_s^2 \rangle ^{+}$
, is plotted in figures 19(
$c$
) and 19(
$d$
), showing the effects of the slope and offset, respectively. Compared with the mean thickness
$\langle H_m \rangle$
, the turbulence intensity exhibits only minor variations across different criteria. However, small differences can be observed in the corresponding indicator function,
$\varXi _2$
, shown in the insets. Although some differences appear in the outer region, the variations are negligible within the range
$y/R \lt 0.2$
, where the logarithmic behaviour discussed in this study is observed. These results demonstrate that the main conclusions remain qualitatively unchanged with respect to the choice of threshold criteria.
Appendix C. Distinction between wall-scaled and non-wall-scaled UMZs
Examples of instantaneous streamwise velocity fields in the
$y$
–
$z$
plane dominated by wall-scaled UMZs (
$a$
) and non-wall-scaled UMZs (
$b$
). In (
$a$
,
$b$
), the horizontal dashed line denotes
$y/R=0.35$
, and each colour represents a different UMZ. (
$c$
,
$d$
) The PDFs of
$U/U_{\textit{CL}}$
computed from the instantaneous fields shown in (
$a$
) and (
$b$
), respectively. The colours in (
$c$
) and (
$d$
) correspond to the UMZs identified in (
$a$
) and (
$b$
), respectively. (
$e$
) The conditionally averaged streamwise turbulence intensity for the wall-scaled UMZs (closed circles) and the non-wall-scaled UMZs (open circles). (
$f$
) The corresponding indicator function,
$\varXi _2=y\,\mathrm{d}\langle u_s^2 \rangle ^{+}/\mathrm{d}y$
. In (
$e$
,
$f$
), the blue and red symbols represent
$\textit{Re}_{\tau }=3000$
and
$6000$
, respectively, and the grey lines indicate the corresponding unconditioned quantities for comparison.

To further illustrate how wall-scaled and non-wall-scaled UMZs appear in the instantaneous flow field, we have added figure 20, which compares representative fields dominated by each type. In figure 20(a,b), the black dashed line denotes
$y/R=0.35$
, and each colour represents a distinct UMZ, with consistent colouring used in the corresponding PDFs in figure 20(c,d). Figure 20(
$a$
) shows a flow field dominated by wall-scaled UMZs (74 %). The detected UMZs are stacked from the near-wall region to the outer region, with their thickness increasing with the wall-normal distance. Most of these UMZs have their centre positions below
$y/R=0.35$
. The corresponding PDF in figure 20 exhibits multiple well-separated peaks over a broad range of
$U/U_{\textit{CL}}$
, indicating the presence of UMZs spanning a wide range of velocities.
In contrast, figure 20(
$b$
) shows a flow field with less than 15 % wall-scaled UMZs, and is therefore dominated by non-wall-scaled UMZs. The detected UMZs are primarily located in the outer region, with most of their centre positions lying above
$y/R=0.35$
. Their thickness is relatively small compared with those observed closer to the wall and does not increase with wall-normal distance. The corresponding PDF in figure 20(
$d$
) indicates that these UMZs are associated with relatively high streamwise velocities (
$U/U_{\textit{CL}} \gtrsim 0.7$
). It also shows a narrower range of modal velocities and smaller separations between adjacent peaks than in figure 20(
$c$
), indicating that the UMZs are confined to a limited range of streamwise velocity.
These differences are also reflected in the associated statistics. Figure 20(
$e$
) shows the conditionally averaged streamwise turbulence intensity for wall-scaled (closed symbols) and non-wall-scaled UMZs (open symbols) at
$\textit{Re}_{\tau }=3000$
and
$6000$
. The intensity associated with non-wall-scaled UMZs is consistently smaller. Moreover, for
$y/R \gtrsim 0.2$
, the profile becomes approximately flat for the wall-scaled UMZs, whereas no such behaviour is observed for the non-wall-scaled UMZs. To examine the logarithmic variation, the corresponding indicator function is plotted in figure 20(
$f$
). For wall-scaled UMZs, the indicator function remains approximately constant, whereas for non-wall-scaled UMZs it varies significantly and does not exhibit a plateau. These results support the conclusion that wall-scaled UMZs are responsible for the logarithmic variation of the streamwise turbulence intensity, whereas non-wall-scaled UMZs are not.
Appendix D. Kullback–Leibler divergence of wall-normal and spanwise velocity fluctuations
The wall-normal variation of the KL divergence between the PDFs of the cross-stream velocity fluctuations and the Gaussian distribution under both unconditioned (grey line) and wall-scaled UMZ conditions (coloured line): (
$a$
) wall-normal velocity fluctuations and (
$b$
) spanwise velocity fluctuations. The KL divergence corresponding to each
$\textit{Re}_\tau$
is shifted upward by
$0.01$
.

The KL-divergence analysis presented in § 7 is extended here to the wall-normal and spanwise velocity fluctuations. Figure 21(a,b) shows the wall-normal variation of the KL divergence for the normalised wall-normal and spanwise velocity fluctuations,
$v'(y) = v(y)/v_{\textit{rms}}(y)$
and
$w'(y) = w(y)/w_{\textit{rms}}(y)$
, respectively. As seen, both
$D_{v_s'}$
and
$D_{w_s'}$
exhibit a clear constant region over the logarithmic layer. For the wall-normal component in particular, the constant region is more clearly observed compared with the unconditioned case
$D_{v'}$
. This result confirms that, under the wall-scaled UMZ condition, the KL divergence for all three velocity components forms a plateau over the same wall-normal range, namely the logarithmic region, in a manner consistent with the high-
$\textit{Re}_\tau$
findings of Furuichi et al. (Reference Furuichi, Ono and Tsuji2025). This finding supports the conclusion of the main text that the PDF shapes within the wall-scaled UMZs are nearly invariant with wall-normal position in the logarithmic layer, not only for the streamwise component but also for the wall-normal and spanwise components.














Reb
Reτ
Lx
R
Nx
Nr
Nθ
Δx+
Δ(Rθ)+
Δrmin+
Δrmax+
Δt+
a
b
d
Reτ=550
1000
3000
6000
(∘)
Reτ=6000
a
U/UCL
r
θ
lz+=2000
b
y
z
c
U/UCL
Um
Ui
Ri
i
d
U/UCL
z+=3000
b
⟨U⟩/UCL
Reτ=550
6000
a
U=Ui
U/UCL=0.95
b
U
a
Um
Ui
c
U
Δz/R=0.2
U
a
NUMZ
a
NUMZ
Reτ=550
6000
b
⟨NUMZ⟩
Reτ
◊
△
a
Um
b
Ui
Reτ≈550
6000
U/UCL=0.95
ym/R
Hm/R
a
Reτ=550
b
1000
c
3000
d
6000
Hm=2ym
ym+
Hm+
40%,60%,70%
80%
a
ym/R
Hm/R
Reτ=550
R1
Rmax
R1
Rmax
Hm=2ym
b
ym/R
Hm/R
R1
Rmax
Reτ=550
R1
Rmax
Hm=2ym
c
R2
Rmax−1
⟨Hm(ym)⟩
Hm=0.75ym
Hm/R=0.75ym/R−0.15
d
⟨Hm⟩
R2
Rmax−1
c
c
a
U/UCL
⋄
yi
⋄
b
⟨U|i⟩
yi/R=0.1
yi
ΔU
δω
∂⟨U⟩/∂y|max
ΔU+
c
d
ΔU+
d
δω/λT
yi/R
a
b
λT
δω
c
R
d
Reτ1/2
e
δν
f
η
a
yL
yU
yU=1.62yL
b
yL/yU
yL
Reτ
b
Ns/Nall
Reτ
a
⟨Us⟩+
Reτ=550
1000
3000
6000
⟨U⟩+=1/κln(y+)+B
κ=0.39
B=4.3
b
⟨Us⟩+
Ξ1=y+d⟨Us⟩+/dy+
Ξ1
y/R
c
f
y+=3Reτ1/2
0.2R+
a
f
⟨U⟩+
a
⟨us2⟩+
Reτ=550
1000
3000
6000
b
Ξ2=yd⟨us2⟩+/dy
c
f
Ξ2=−0.95=−A1
a
f
⟨u2⟩+
a
⟨ws2⟩+
Reτ=550
1000
3000
6000
b
Ξ3=yd⟨ws2⟩+/dy
c
f
Ξ3=−0.38=−A2
a
f
⟨w2⟩+
⟨vs2⟩+
−⟨usvs⟩+
⟨vs2⟩>+
−⟨usvs⟩>+
NUMZ≥⟨NUMZ⟩=5
⟨NUMZ⟩
Reτ=6000
⟨v2⟩+
−⟨uv⟩+
−⟨uv⟩
yp
y/R=0.1
u′(y)=u(y)/urms(y)
y/R=0.1
0.15
0.2
Reτ=1000
3000
6000
b
us′(y)=us(y)/us,rms(y)
c
S(u′)=⟨u3⟩/⟨u2⟩3/2
S(us′)=⟨us3⟩/⟨us2⟩3/2
Reτ
ΔS
d
us′
u′
Reτ
ΔDu′=0.01
y1
y2
Reτ=3000
6000
Du′(Py1‖Py2)
Reτ=3000
6000
Dus′(Py1‖Py2)
a
d
ΔD=Dus′−Du′
Reτ=3000
6000
ΔD
Dus′
⟨NUMZ⟩
lz+
⟨Us⟩+
⟨us′2⟩+
lz+=1000,2000
3000
Reτ=550
1000
⟨Hm⟩
R2
Rmax−1
a
b
⟨Hm⟩=0.75ym
⟨us2⟩+
Ξ2
y
z
a
b
a
b
y/R=0.35
c
d
U/UCL
a
b
c
d
a
b
e
f
Ξ2=yd⟨us2⟩+/dy
e
f
Reτ=3000
6000
a
b
Reτ
0.01