1. Introduction
The quasilinear (QL) and generalised quasilinear (GQL) approximations have been proposed recently in various forms for several turbulent flows. Common to all various forms, this approach introduces a decomposition of the given flow into two groups: one in which nonlinear terms are kept, and the other in which all self-interactions are ignored or modelled suitably. The resulting equations for the first group are unchanged from or close to the original, while those for the second become equivalent to a linearisation around the first group, often with an additional model (e.g. stochastic forcing). The early work may be found in Malkus (Reference Malkus1954, Reference Malkus1956) and Herring (Reference Herring1963, Reference Herring1964, Reference Herring1966), where the equations for the second group are closed using the ‘marginal stability’ ansatz. In the last two decades, several variants of the QL framework have been proposed for many different flows with various types of models for the self-interaction term of the second group (e.g. stochastic forcing, eddy viscosity, etc.): for example, stochastic structural stability theory (S3T) (Farrell & Ioannou Reference Farrell and Ioannou2003, Reference Farrell and Ioannou2007, Reference Farrell and Ioannou2012), direct statistical simulation (DSS) (Marston et al. Reference Marston, Conover and Tobias2008; Tobias & Marston Reference Tobias and Marston2013), self-consistent approximation for linearly unstable flows (Mantič-Lugo et al. Reference Mantič-Lugo, Arratia and Gallaire2014; Mantič-Lugo & Gallaire Reference Mantič-Lugo and Gallaire2016), minimal quasilinear approximation augmented with an eddy-viscosity model (Hwang & Eckhardt Reference Hwang and Eckhardt2020; Skouloudis & Hwang Reference Skouloudis and Hwang2021), restricted nonlinear (RNL) model (Thomas et al. Reference Thomas, Lieu, Jovanović, Farrell, Ioannou and Gayme2014, Reference Thomas, Farrell, Ioannou and Gayme2015; Farrell et al. Reference Farrell, Gayme and Ioannou2017; Pausch et al. Reference Pausch, Yang, Hwang and Eckhardt2019; Hernández & Hwang Reference Hernández and Hwang2020) and generalised quasilinear approximations (GQL) (Constantinou Reference Constantinou2015; Marston et al. Reference Marston, Chini and Tobias2016; Tobias & Marston Reference Tobias and Marston2017).
Continuing from Part 1 (Hernández, Yang & Hwang Reference Hernández, Yang and Hwang2022), the starting point of the present study is the RNL model, which we have referred to as the QL model. The important advantage of this model is that it can capture the so-called ‘self-sustaining process’ (Hamilton, Kim & Waleffe Reference Hamilton, Kim and Waleffe1995; Waleffe Reference Waleffe1997) – i.e. the two-way interaction between ‘streamwise elongated’ structure of streamwise velocity (streaks) and its ‘streamwise wavy’ instability involving cross-streamwise velocities (waves and rolls) – in a minimal manner (Thomas et al. Reference Thomas, Lieu, Jovanović, Farrell, Ioannou and Gayme2014, Reference Thomas, Farrell, Ioannou and Gayme2015; Pausch et al. Reference Pausch, Yang, Hwang and Eckhardt2019; Hernández & Hwang Reference Hernández and Hwang2020); in the QL model, the elongated streaks are captured by the nonlinear streamwise mean equations, and the streamwise wavy instability by the linearised equations. The recent examination of this model in uniform shear turbulence (Hernández & Hwang Reference Hernández and Hwang2020) revealed that the inhibition of the streamwise energy cascade in the QL model by construction results in highly elevated spectral energy intensity residing only at the streamwise integral length scale, and consequently, only a small number of streamwise Fourier modes remain active (Thomas et al. Reference Thomas, Lieu, Jovanović, Farrell, Ioannou and Gayme2014, Reference Thomas, Farrell, Ioannou and Gayme2015; Farrell et al. Reference Farrell, Ioannou, Jiménez, Constantinou, Lozano-Durán and Nikolaidis2016; Tobias & Marston Reference Tobias and Marston2017). While the QL model captures the most fundamental physical mechanism in wall-bounded turbulence reasonably well (especially at low Reynolds numbers), it was found that the approximation yields some non-negligible damages to the slow pressure. Consequently, this inhibits the related pressure strain that transfers the turbulent kinetic energy produced at the streamwise component to the cross-streamwise components, explaining the typical anisotropic structures in the QL model throughout the entire wavenumber space.
In Part 1 (Hernández et al. Reference Hernández, Yang and Hwang2022), the QL model was extended by applying the GQL approximation to the streamwise direction in turbulent channel flow at
$Re_\tau \simeq 1700$ to explore a route to improvement of the QL model. The GQL approximation employs the flow decomposition in the QL model in a more flexible manner: the nonlinear equations for the streamwise mean flow are replaced with nonlinear equations for a set of low-wavenumber streamwise Fourier modes (i.e. the low-wavenumber group), and the rest of the flow (i.e. the high-wavenumber group) is described by the linearised equations about the set of low-wavenumber streamwise Fourier modes. This is a direct extension of the previous QL model (Thomas et al. Reference Thomas, Lieu, Jovanović, Farrell, Ioannou and Gayme2014, Reference Thomas, Farrell, Ioannou and Gayme2015; Farrell et al. Reference Farrell, Gayme and Ioannou2017; Pausch et al. Reference Pausch, Yang, Hwang and Eckhardt2019; Hernández & Hwang Reference Hernández and Hwang2020), which offers the dynamics of the energy-containing streamwise waves in a minimal manner. In Part 1 (Hernández et al. Reference Hernández, Yang and Hwang2022), the GQL approximation was found to improve the QL model effectively. By incorporation of only a few more streamwise Fourier modes into the low-wavenumber group, the GQL model was able to recover the self-similar scaling with the distance from the wall of the turbulence statistics and spectra. This was associated with the mean–fluctuation interaction improved by the enhanced energy transfer in the streamwise wavenumber space. Furthermore, it was found that this energy transfer from the low- to the high-wavenumber group in the GQL model depends on the neutrally stable leading Lyapunov spectrum of the linearised equations for the high-wavenumber group. In particular, if the threshold wavenumber distinguishing the two groups is sufficiently high, then this energy transfer is inhibited due to the linear nature of the equations for the high-wavenumber group.
The GQL model employed in Part 1 provides useful physical insights into the modelling of wall-bounded turbulence in the context of the self-sustaining process. However, if the Reynolds number is sufficiently high, then the self-sustaining processes emerge at all (spanwise) integral length scales that vary continuously with the distance from the wall (Flores & Jiménez Reference Flores and Jiménez2010; Hwang & Cossu Reference Hwang and Cossu2010a, Reference Hwang and Cossu2011; Hwang Reference Hwang2015; Hwang & Bengana Reference Hwang and Bengana2016). There have been a number of previous investigations on the interactions between the related energy-containing fluid motions in the near-wall and outer regions (Hutchins & Marusic Reference Hutchins and Marusic2007; Mathis, Hutchins & Marusic Reference Mathis, Hutchins and Marusic2009; Duvvuri & McKeon Reference Duvvuri and McKeon2015; Agostini & Leschziner Reference Agostini and Leschziner2016; Zhang & Chernyshenko Reference Zhang and Chernyshenko2016). In these studies, the velocity field is often decomposed into large and small scales, and their interactions have been modelled and explored in terms of superposition and amplitude modulation in the near-wall region. However, in general, such interactions are expected to exist across almost all wall-normal locations, given the existence of the self-sustaining processes across all integral length scales in the form of the ‘attached eddies’ of Townsend (Reference Townsend1976). Recent investigations on the statistical features in the spectral energy transfer revealed that the interaction dynamics appears to be highly complex (Cho, Hwang & Choi Reference Cho, Hwang and Choi2018; Kawata & Alfredsson Reference Kawata and Alfredsson2018; Lee & Moser Reference Lee and Moser2019), although there was some progress made for an idealised system in which there are only two integral length scales (Doohan et al. Reference Doohan, Bengana, Yang, Willis and Hwang2021a; Doohan, Willis & Hwang Reference Doohan, Willis and Hwang2021b). The newly found scale interactions include mediation of the energy cascade from a given integral scale by larger energy-containing motions (Cho et al. Reference Cho, Hwang and Choi2018), small-scale turbulence production driven by energy cascade from larger scales (Doohan et al. Reference Doohan, Willis and Hwang2021b), the near-wall inverse energy transfer that forms large-scale near-wall inactive motions of Townsend (Reference Townsend1976) (Cho et al. Reference Cho, Hwang and Choi2018; Doohan et al. Reference Doohan, Bengana, Yang, Willis and Hwang2021a,Reference Doohan, Willis and Hwangb), and the mean–fluctuation interactions across all integral scales (Doohan et al. Reference Doohan, Bengana, Yang, Willis and Hwang2021a).
As was discussed briefly in Part 1, the GQL approximation suppresses particular triadic wave interactions between the low- and high-wavenumber groups. This was sketched in figure 1, where the GQL approximations ignore the interactions LH-H (figure 1b), LL-H (figure 1d) and HH-H (figure 1f). Similarly, this feature can be exploited further to study the scale interactions in wall-bounded turbulence: for example, the other interactions such as HH-L can also be suppressed to understand the effect of the inverse energy transfer. From this perspective, a flexible utilisation of the GQL approximation may perhaps provide a unique interventional tool to explore these scale interactions.
Set of triadic interactions for energy transfer to (a–c) low and (d–f) high wavenumbers from interactions (a,d) between low wavenumbers (LL-L, LL-H), (b,e) between low and high wavenumbers (LH-L, LH-H), and (c,f) between high wavenumbers (HH-L, HH-H). In the GQL approximation, three interactions – (a) LL-L, (c) HH-L, (e) LH-H) – are retained out of six in total. The other three triads are discarded (Marston et al. Reference Marston, Chini and Tobias2016).

The objective of the present study is to explore the scale interactions in wall-bounded turbulence using the GQL approximation. For this purpose, unlike Part 1, the low-wavenumber-mode group will be composed of the plane Fourier modes for
$k_z\leqslant k_{z,c}$ (where
$k_z$ is the spanwise Fourier wavenumber, and
$k_{z,c}$ is the corresponding threshold wavenumber for the decomposition). The spanwise length scale has been understood to represent the size of coherent structures underpinned by the self-sustaining process, as the spanwise Fourier modes well represent the streaks highly elongated in the streamwise directions and the related quasi-streamwise vortical structures (e.g. Hwang Reference Hwang2015; Cho et al. Reference Cho, Hwang and Choi2018). The proposed GQL approximation will then be examined to present its effect in the spanwise direction. Further to this, a set of numerical experiments, which suppress deliberately the terms presented in figure 1, will be performed to understand the scale interaction in wall-bounded turbulence.
The paper is organised as follows. The GQL approximation together with its spectral energy budget is formulated briefly in § 2 for completeness. In § 3, the statistics and spectra of the GQL model with spanwise threshold wavenumber for channel flow are compared to those of a wall-resolved large eddy simulation (LES). The energy-budget spectra are also presented here with a detailed analysis to explain the statistical features of the GQL model in the spanwise direction. Subsequently, comparison of the GQL in the present study is made with that in Part 1, and various numerical experiments suppressing certain interactions are carried out in § 4, highlighting the use of the GQL approximation oriented idea to analyse scale interactions. Finally, the paper concludes in § 5 with some remarks.
2. Problem formulation
2.1. Generalised quasilinear approximation
We consider a pressure-driven plane channel flow, in which the density and kinematic viscosity are given by
$\rho$ and
$\nu$, respectively. The time is denoted by
$t$, and the space by
$\boldsymbol {x}=(x,y,z)$, with
$x$,
$y$ and
$z$ being the streamwise, wall-normal and spanwise directions, respectively. The lower and upper walls of the channel are set to be located at
$y=0,2h$. For the GQL approximation, the velocity
$\boldsymbol {u}$ is decomposed into two groups using a discrete Fourier transform in the wall-parallel directions:
where
\begin{equation} \boldsymbol{U}_l=\sum_{n={-}M_{z,F}}^{M_{z,F}}\sum_{m={-}M_{x,F}}^{M_{x,F}} \hat{\boldsymbol{u}}_{m,n} \exp({{\rm i}(m k_{x,0}x+n k_{z,0}z)}), \end{equation}
and
$\boldsymbol {u}_h$ is given from (2.1a). Here,
$\hat {\boldsymbol {u}}_{m,n}$ is the discrete Fourier mode of the velocity,
$k_{x,0}$ and
$k_{z,0}$ are the fundamental streamwise and spanwise wavenumbers for the given horizontal domain, and
$M_{x,F}$ and
$M_{z,F}$ define the threshold streamwise and spanwise wavenumbers for the decomposition, such that
$k_{x,c}=M_{x,F}k_{x,0}$ and
$k_{z,c}=M_{z,F}k_{z,0}$. Given the scope of the present study, the main parameter of interest is
$k_{z,c}$ (or
$M_{z,F}$), while keeping
$M_{x,F}$ identical to the number of streamwise Fourier modes used in the full simulation (see § 2.3 for further details).
With the decomposition in (2.1), two related projection operators are defined as
with the following properties:
where
$\mathcal {I}[\,\cdot\, ]$ is the identity operator. Using the definition and the properties listed in (2.2a,b) and (2.3), the Navier–Stokes equations are first projected onto the
$\mathcal {P}_l$ (or low-wavenumber) and
$\mathcal {P}_h$ (or high-wavenumber) subspaces. The subsequent linearisation of the equations for
$\boldsymbol {u}_h$ about
$\boldsymbol {U}_l$ leads to the GQL system of interest, i.e.
and
where
$P_l$ and
$p_h$ are given to enforce
$\boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {U}_l=0$ and
$\boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {u}_h=0$, respectively, with
$p=P_l+p_h$. As in Part 1, the terms
$\mathcal {P}_{l}[(\boldsymbol {U}_{l}\boldsymbol {\cdot }\boldsymbol {\nabla })\boldsymbol {u}_{h}]$ and
$\mathcal {P}_{l}[(\boldsymbol {u}_{h}\boldsymbol {\cdot }\boldsymbol {\nabla })\boldsymbol {U}_{l}]$ are neglected in (2.4a). The numerical simulations in this study are performed using an LES. Therefore, the subgrid-scale (SGS) stress tensor, given by
${\boldsymbol {\tau }}_{SGS}= - ( \boldsymbol {u} \boldsymbol {u} - \overline {\boldsymbol {u} \boldsymbol {u}})$, appears in (2.4), where the overbar (
$\bar {\ \cdot\ }$) denotes the application of the grid filter given by the numerical discretisation of the equations (see § 2.3). The SGS tensor for the present LES here employs a mixing-length type model
${\boldsymbol {\tau }}_{SGS}= \nu _t (\boldsymbol {\nabla } \boldsymbol {u}+ \boldsymbol {\nabla } \boldsymbol {u}^{\rm T})$, where the eddy viscosity
$\nu _t$ is determined with the Vreman model (Vreman Reference Vreman2004). Further details are provided in Part 1 (Hernández et al. Reference Hernández, Yang and Hwang2022).
2.2. Spectral energy budget
To analyse the turbulence statistics and spectra of the given GQL and original systems, the Reynolds decomposition of the velocity
$\boldsymbol {u}=(u,v,w)$ is employed, i.e.
in which
$\boldsymbol {U}(\equiv \langle \boldsymbol {u} \rangle _{x,z,t})=(U(y),0,0)$ is the mean velocity, with
$\langle\,\cdot\,\rangle _{x,z,t}$ being an average in
$t$ (time), and the
$x$- and
$z$-directions. Similarly to Part 1, this decomposition is introduced only for the statistical analysis of LES and GQL models, and it should not be confused with the one in (2.1) introduced for the GQL approximations. Using (2.1) and (2.5), the turbulent velocity fluctuation is decomposed further into the low- and high-wavenumber components as in (2.1a):
As in Part 1, the energy balance in the Fourier space for the GQL approximation is given by
\begin{align} & 0 = \underbrace{\left\langle {\rm{Re}} \left\{-{\widehat{u^{\prime}}}^* (k)\,\widehat{v^{\prime}}(k) \, \frac{{\rm d} U}{{\rm d} y} \right\} \right\rangle_{r^\perp,t}}_{\hat{P}(y,k)} +\underbrace{\left\langle - \nu\, \frac{\partial{{\widehat{u_{i}^{\prime}}}^*(k)} }{\partial{x_{j}} }\, \frac{\partial{{\widehat{u_{i}^{\prime}}}(k)}}{\partial{x_{j}} } \right\rangle_{r^\perp,t}}_{\hat{\varepsilon}(y,k)}\nonumber\\ &\quad +\underbrace{\left\langle {\rm{Re}} \left\{-{\widehat{u_{i}^{\prime}}}^*(k) \left(\frac{\partial }{\partial{x_{j}} } \left( \widehat{\tau_{ij,SGS}^{\prime}}(k) \right) \right)\: \right\} \right\rangle_{r^\perp,t}}_{\hat{\varepsilon}_{SGS}(y,k)} +\underbrace{\left\langle {\rm{Re}} \left\{\frac{{\rm d}}{{\rm d} y} \left (\frac{\widehat{p^{\prime}}(k)\,{\widehat{v^{\prime}}}^*(k)}{\rho} \right) \right\} \right\rangle_{r^\perp,t}}_{\hat{T}_{p}(y,k)} \nonumber\\ &\quad + \underbrace{\left\langle {\rm{Re}} \left\{-{\widehat{u_{i}^{\prime}}}^*(k) \left(\frac{\partial }{\partial{x_{j}} } \left( \widehat{u_{i}^{\prime}u_{j}^{\prime} }(k) -\mathcal{P}_h\left[\widehat{u_{h,i}^{\prime}u_{h,j}^{\prime} }(k)\right] -\mathcal{P}_h\left[\widehat{u_{l,i}^{\prime}u_{l,j}^{\prime} }(k)\right] \right. \right. \right. \right.}_{\hat{T}_{turb}\left(y, k\right)} \nonumber\\ &\quad \underbrace{\left. \left. \left. \left. {}-\mathcal{P}_l\left[\widehat{u_{l,i}^{\prime}u_{h,j}^{\prime} }(k)\right] -\mathcal{P}_l\left[\widehat{u_{h,i}^{\prime}u_{l,j}^{\prime} }(k)\right] \right) \vphantom{\frac{\partial }{\partial{x_{j}} }}\right) \right\} \right\rangle_{r^\perp,t}}_{\hat{T}_{turb}\left(y, k\right)} +\underbrace{\left\langle \nu\, \frac{{\rm d}^2{\hat{e}(k)}}{{\rm d}{y}^2 } \right\rangle_{r^\perp,t}}_{\hat{T}_{\nu}(y,k)}, \end{align}
for
$i,j=1,2,3$. Here,
$(x_1,x_2,x_3)=(x,y,z)$,
$(u_1^{\prime },u_2^{\prime },u_3^{\prime })=(u^{\prime },v^{\prime },w^{\prime })$,
$(\,\hat {\cdot }\,)$ denotes the Fourier-transformed coefficient in the
$r$-direction, where
$r(=x\ \mathrm {or}\ z)$ is the streamwise or spanwise coordinate, with
$r^+(=z\ \mathrm {or}\ x)$ being the wall-parallel direction orthogonal to
$r$, and
$k(=k_x\ \mathrm {or}\ k_z)$ being the spatial wavenumber in the
$r$-direction. Also,
$\hat {e}(k)=( |{\widehat {u'}(k)} |^2 + |{\widehat {v'}(k)} |^2 +|{\widehat {w'}(k)} |^2 )/2$, the superscript
$^*$ indicates the complex conjugate, and
${\rm {Re}} \{\,\cdot\, \}$ denotes the real part. The terms on the right-hand side are the rate of turbulence production
$\hat {P}(y,k)$, viscous dissipation
$\hat {\varepsilon }(y,k)$, SGS dissipation
$\hat {\varepsilon }_{SGS}(y,k)$, pressure transport
$\hat {T}_p(y,k)$, turbulent transport
$\hat {T}_{turb}(y,k)$ and viscous transport
$\hat {T}_{\nu }(y,k)$, respectively.
Equation (2.7) can be split further into each component. In this case, the energy transports by pressure strain appear subsequently as
\begin{gather} \left.\begin{gathered} \hat{\varPi}_x(y, k)=\left \langle {\rm{Re}} \left \{ \frac{\widehat{p'} (k)}{\rho}\,\frac{\partial {\widehat{u'}}^* (k)}{\partial x} \right \} \right \rangle_{r^\perp,t}, \quad \hat{\varPi}_y(y, k)= \left \langle {\rm{Re}} \left \{\frac{\widehat{p'} (k)}{\rho}\,\frac{\partial {\widehat{v'}}^*(k)}{\partial y} \right \} \right \rangle_{r^\perp,t} ,\\ \hat{\varPi}_z(y, k)=\left \langle {\rm{Re}} \left \{\frac{\widehat{p'} (k)}{\rho}\,\frac{\partial {\widehat{w'}}^* (k)}{\partial z} \right \} \right \rangle_{r^\perp,t}, \end{gathered}\right\} \end{gather}
where
$\hat {\varPi }_x$,
$\hat {\varPi }_y$ and
$\hat {\varPi }_z$ are one-dimensional spectra of the streamwise, wall-normal and spanwise components of pressure strain, respectively. From the incompressibility of the given fluid, the pressure strain terms satisfy the relation
further discussion on this can be found in previous studies (Mizuno Reference Mizuno2016; Cho et al. Reference Cho, Hwang and Choi2018; Lee & Moser Reference Lee and Moser2019; Hernández & Hwang Reference Hernández and Hwang2020; Hernández et al. Reference Hernández, Yang and Hwang2022).
2.3. Numerical simulations
A large eddy simulation (LES) is first carried out with the numerical solver used in Part 1 (Hernández & Hwang Reference Hernández and Hwang2020). The main difference of the simulation in the present study from that in Part 1 is the computational domain size. In Part 1, the streamwise computational domain was
$L_x/h = {\rm \pi}$. This size was chosen to match the most energetic streak instability length scale of large-scale outer structures in the streamwise direction (de Giovanetti, Sung & Hwang Reference de Giovanetti, Sung and Hwang2017). However, the scope of the GQL in the present study is to understand the scale interactions. For this purpose, the simulations here are carried out by well resolving all the known coherent structures. Since the largest streamwise and spanwise length scales in turbulent channel flow are
$\lambda _x/h \simeq 10$ and
$\lambda _z/h \simeq 1$ (del Alamo & Jiménez Reference del Alamo and Jiménez2003; Hwang Reference Hwang2015), a sufficiently large computational domain is considered in the streamwise and spanwise directions, i.e.
$L_x/h = 8{\rm \pi}$ and
$L_z/h = {\rm \pi}$. We note that this domain size is 16 times bigger than that in Part 1. Table 1 summarises the parameters for the simulations performed in this study. The Reynolds number for all the considered cases is
$Re=U_0 h /\nu =66\,667$. The grid spacings are
$\varDelta _x^+=54.9\unicode{x2013}67.3$ and
$\varDelta _z^+=27.5\unicode{x2013}33.6$. For each case, the number of spanwise Fourier modes used in the GQL approximation is varied from
$M_{z,F}=1$ (GQLZ1) to the maximum number, the latter of which corresponds to that of the given LES.
Simulation parameters in the present study. The domain size in the
$x$-,
$y$- and
$z$-directions is
$L_x/h=8 {\rm \pi}$,
$L_y/h=2$ and
$L_z/h={\rm \pi}$, respectively. Here,
$Re=U_0 h/\nu$ and
$Re_\tau =u_\tau h /\nu$, where
$U_0$ and
$u_\tau$ are the centreline and wall shear velocities, respectively. The grid spacings in the
$x$- and
$z$-directions are
$\varDelta _x^+$ and
$\varDelta _z^+$ (after aliasing);
$\lambda _{z,c}$ is the threshold spanwise wavelength; and
$N_x,N_y,N_z$ are the numbers of grid points in the
$x$-,
$y$- and
$z$-directions, respectively.

3. GQL approximations
3.1. Turbulence statistics and spectra
The first- and second-order turbulence statistics as functions of the wall-normal direction
$y^+$ are plotted in figure 2 (the superscript
$^+$ denotes normalisation by viscous inner scale). Good agreement can be observed between the LES and DNS statistics, plotted with a dashed line from Hoyas & Jiménez (Reference Hoyas and Jiménez2008) at
$Re_\tau =2003$. Similarly to the QL case in Part 1, the statistics of the GQLZ1 case are the most anisotropic in comparison to the reference LES, although the physical nature of the flow decomposition adopted in GQLZ1 is quite different from the QL case (see also §§ 4.1 and 4.2 for a further discussion). In particular, the
$u_{rms}^+$ (figure 2a) and
$U^+$ (figure 2c) values are larger than those of the reference LES, behaving similarly to the QL model in previous studies (Thomas et al. Reference Thomas, Lieu, Jovanović, Farrell, Ioannou and Gayme2014; Farrell et al. Reference Farrell, Ioannou, Jiménez, Constantinou, Lozano-Durán and Nikolaidis2016; Hernández et al. Reference Hernández, Yang and Hwang2022) where the quasilinear approximation is made in the streamwise direction. The behaviour of
$v_{rms}^+$ (figure 2b) and
$\langle u^{\prime } v^{\prime }\rangle _{x,z,t}^+$ depends on the
$y$-location: the values are smaller on and below the log layer, and greater above it. The enrichment of the
$\mathcal {P}_l$-subspace group by incorporating further more spanwise Fourier modes (i.e. GQLZ5 and GQLZ25 cases) leads to the magnitude of the near-wall streamwise velocity peak being reduced considerably, whilst
$v_{rms}^+$,
$w_{rms}^+$ and
$\langle u^{\prime } v^{\prime }\rangle _{x,z,t}^+$ become slightly larger than for the reference LES. For the GQLZ5 case, the inner-scaled mean velocity
$U^+$ now provides a much better approximation to that of the reference LES and the GQLZ1 case. As more spanwise Fourier modes are included in the
$\mathcal {P}_l$-subspace group, all the turbulence statistics converge to those of the full LES (GQLZ5 and GQLZ25 cases in figure 2), similarly to the GQL cases along the streamwise direction investigated in Part 1. Finally, the statistics of the GQLZ25 case give the closest match to those of the reference LES, as expected.
First- and second-order turbulence statistics for the present LES, direct numerical simulations (DNS) at
$Re_\tau =2003$ (Hoyas & Jiménez Reference Hoyas and Jiménez2008), GQLZ1, GQLZ5 and GQLZ25 cases: (a)
$u_{rms}^+(y^+)$, (b)
$v_{rms}^+(y^+)$, (c)
$U^+(y^+)$, and (d)
$\langle u^{\prime } v^{\prime }\rangle _{x,z,t}^+(y^+)$.

Figure 3 compares the premultiplied spanwise wavenumber spectra of streamwise and wall-normal velocities of the reference LES with those of the GQLZ cases. Typical features of the energy-containing motions in turbulent channel flow can be observed in the spectra of LES (e.g. Hwang Reference Hwang2015). The spanwise wavenumber spectra of streamwise velocity appear to be aligned along a linear ridge
$y \approx 0.1 \lambda _z$ throughout the log region. At the bottom end of this ridge (
$y^+\approx 10$ and
$\lambda _z^+\approx 100$), a local maximum with the spanwise spacing of the near-wall streaks (Kline et al. Reference Kline, Reynolds, Schraub and Runstadler1967) is found. At the top end (
$y/h\approx 0.1$ and
$\lambda _z/h\approx 1$), another local maximum can be found, whose spanwise wavelength corresponds to the spanwise length scale of large- and very-large-scale motions (Kovasznay, Kibens & Blackwelder Reference Kovasznay, Kibens and Blackwelder1970; del Alamo & Jiménez Reference del Alamo and Jiménez2003). The spanwise wavenumber spectra of wall-normal velocity are also found to be aligned along a linear ridge
$y =0.5 \lambda _z$ (figures 3a,b).
Premultiplied spanwise wavenumber spectra of (a,c,e,g) streamwise
$k_z^+\, \varPhi _{uu}^+(y^+,\lambda _z^+)$, and (b,d, f,h) wall-normal
$k_z^+\, \varPhi _{vv}^+(y^+,\lambda _z^+)$ velocity, for (a,b) LES, (c,d) GQLZ1, (e, f) GQLZ5, and (g,h) GQLZ25 cases. Here, the vertical line represents the spanwise cut-off wavelength (
$\lambda _{z,c}$) dividing the (a,c,e,g)
$\mathcal {P}_h$-subspace and (b,d, f,h)
$\mathcal {P}_l$-subspace regions. The red dots mark the maxima of the premultiplied spectra.

The GQLZ1 case shows an increased energy in the region
$\lambda _z^+ \approx 500$ (
$\lambda _z/h \approx 0.3$) of the streamwise velocity spectra compared to the LES case, in line with the statistics discussed above. This is in contrast to the reduced energy featured at the other wavelengths (figure 3c). The spanwise wavenumber spectra of wall-normal velocity show increased energy in the region
$\lambda _z^+ \approx 1000$ (
$\lambda _z/h \approx 0.5$), and the spectra span most of the log region but showing little energy for
$\lambda _z^+ \lesssim 200$ (figure 3d). The reduced energy observed for
$\lambda _z^+ \lesssim 200$ explains the lack of Reynolds shear stress in the GQLZ1 case in the near-wall region (figure 2d). By incorporating more spanwise Fourier modes in the
$\mathcal {P}_l$-subspace group (i.e. GQLZ5 and GQLZ25), the peak location in the streamwise velocity spectra gradually moves towards
$(\lambda _z^+,y^+) \approx (100,10)$ along the linear ridge
$y=0.1\lambda _z$, and the outer part of the spectra also becomes more energetic. The wall-normal velocity spectra now span the region close to the wall along the linear ridge
$y=0.5 \lambda _z$. Moreover, the spectra of the GQLZ25 case are found to be fairly similar to those of the reference LES. An additional case GQLZ64 has been computed (see the Appendix) with
$\lambda _{z,c}^+ \approx 97$, resulting in spanwise wavenumber spectra that do not extend below
$\lambda _{z,c}^+$, implying that the velocity field in the
$\mathcal {P}_h$-subspace group yields the trivial solution. This issue, investigated in Part 1 for the streamwise GQL models, was explained in terms of the Lyapunov exponent and the instability of the linearised equations in the
$\mathcal {P}_h$-subspace group.
The premultiplied streamwise wavenumber spectra of streamwise and wall-normal velocities are shown in figure 4 for the LES and GQLZ cases. The energy-containing motions in turbulent channel flow (LES) at a given spanwise length scale are composed of a bimodal structure with two components: a long streaky structure carrying mainly streamwise turbulent kinetic energy, and a short and tall vortex packet carrying turbulent kinetic energy at all the velocity components (Hwang Reference Hwang2015). The corresponding linear ridges
$y \approx 0.01 \lambda _x$ (lower line) and
$y \approx 0.35 \lambda _x$ (upper line), along which these structures contribute to the spectra, are plotted in figure 4. The former corresponds to a long streaky structure, and the streamwise velocity spectra show high energy intensity along it (figure 4a). The latter corresponds to a vortex packet, and the streamwise spectra are less energetic along it, whilst the wall-normal velocity spectra are very well aligned with it. The streamwise velocity spectra of the GQLZ1 case are energetic in the region
$\lambda _x^+ \gtrsim 600$, whilst the wall-normal velocity spectra do not extend below
$\lambda _x^+ \lesssim 200$ (see also Child et al. (Reference Child, Hollerbach, Marston and Tobias2016), where the same was also observed). The former appear to be around a single point along
$y \approx 0.01 \lambda _x$, whilst weak correlation with
$y \approx 0.35 \lambda _x$ is observed in the latter. As more spanwise Fourier modes are incorporated further into the
$\mathcal {P}_l$-subspace group (figures 4(e–h), i.e. GQLZ5 and GQLZ25), the two spectra begin to show more energy at smaller
$\lambda _x$ along the linear ridges. As a consequence, in the GQLZ5 case, the near-wall region is better resolved and the spectra reach out to smaller scales down to
$\lambda _x^+ \approx 150$. The spectra of the GQLZ25 case greatly resemble those of the LES case.
Premultiplied streamwise wavenumber spectra of (a,c,e,g) streamwise
$k_x^+\, \varPhi _{uu}^+(y^+,\lambda _x^+)$ and (b,d, f,h) wall-normal
$k_x^+\, \varPhi _{vv}^+(y^+,\lambda _x^+)$ velocity, for (a,b) LES, (c,d) GQLZ1, (e, f) GQLZ5, and (g,h) GQLZ25 cases.

It is interesting to observe that the overall behaviours of the GQL models in this study are similar to those of the streamwise QL and GQL models in Part 1 (e.g. Thomas et al. Reference Thomas, Lieu, Jovanović, Farrell, Ioannou and Gayme2014, Reference Thomas, Farrell, Ioannou and Gayme2015; Farrell et al. Reference Farrell, Ioannou, Jiménez, Constantinou, Lozano-Durán and Nikolaidis2016; Tobias & Marston Reference Tobias and Marston2017; Hernández & Hwang Reference Hernández and Hwang2020), although the flow decomposition in (2.1) implies a physically different nature of the GQL models here from those in Part 1. The QL and GQL models in Part 1 are designed to offer a minimal description for the self-sustaining processes at all integral length scales, and study the effect of their mutual nonlinear coupling; the full nonlinear equations in the
$\mathcal {P}_l$-subspace are primarily to describe the full nonlinear evolution of the long streaks (or streamwise velocity fluctuations), whereas the linearised equations in the
$\mathcal {P}_h$-subspace are to resolve the streak instability and the subsequent breakdown process of the streaks. On the contrary, in the GQL models considered here, the full nonlinear equations in the
$\mathcal {P}_l$-subspace almost fully resolve every essential element of the self-sustaining process, at least at the integral length scales for
$\lambda _z\geqslant \lambda _{z,c}$ (
$\lambda _{z,c}= 2{\rm \pi} /k_{z,c}$), whereas the linearised equations in the
$\mathcal {P}_h$-subspace do not support its presence for
$\lambda _z<\lambda _{z,c}$. A further detailed discussion on this issue will be given in §§ 4.1 and 4.2.
Finally, the observations made above with figures 3 and 4 suggest that the smallest
$\lambda _{z,c}$, which allows the GQL model here to capture turbulence statistics and spectra reasonably well, would scale in viscous inner units, such that
$\lambda _{z,c}^+ \simeq O(10^2)$ (i.e. GQLZ25 and GQLZ64). In other words, as
$Re_\tau$ is increased,
$\lambda _{z,c}/h$ would need to be decreased linearly with
$1/Re_\tau$. This is also consistent with the attached eddy hypothesis (Townsend Reference Townsend1976), because the requirement
$\lambda _{z,c}^+ \simeq O(10^2)$ implies that most of the energy-containing motions scaling with
$y$ would need to be resolved in the
$\mathcal {P}_h$-subspace in order for turbulence statistics and spectra to be well captured. This implies that
$\lambda _{z,c}$ may need be set to be a function of
$y$ for more efficient implementation of the GQL model, although implementation of this is challenging practically and needs a caution to deal with energy cascade from the energy-containing motions scaling with
$y$.
3.2. Spectral energy transfer
In this subsection, the spectral energetics of the LES and GQLZ cases are considered further. Given that the present study is based an LES, the focus will be on the production and turbulent spectra only. Note that the pressure and viscous transport spectra are negligibly small in the log and outer regions (Cho et al. Reference Cho, Hwang and Choi2018; Lee & Moser Reference Lee and Moser2019). The premultiplied one-dimensional spanwise wavenumber spectra of the production and turbulent transport of each case are plotted in figure 5. The turbulent production spectra in figure 5(a) appear to be almost uniformly distributed along the ridge
$y \approx 0.2 \lambda _z$, especially in the log layer. Owing to the energy conservative nature of the nonlinear terms in the Navier–Stokes equations, the premultiplied turbulent transport spectra in figure 5(b) show positive and negative regions. The negative turbulent transport (blue contours in figure 5b) is almost balanced with the (positive) production in the log and outer layers, and the positive turbulent transport appears along a ridge indicating the Kolmogorov scale (i.e.
$\lambda _{z} \approx 57 \eta$, where
$\eta$ is the Kolmogorov scale). There can also be observed a region of weakly positive turbulent transport spectra in the region very close to the wall (
$y^+<10$) over a wide range of the spanwise wavelength scales (
$200 \delta _\nu \lesssim \lambda _z \lesssim 0.8 h$), due primarily to an inverse energy transfer from small to large scales (Cho et al. Reference Cho, Hwang and Choi2018; Doohan et al. Reference Doohan, Willis and Hwang2021b).
Premultiplied spanwise wavenumber spectra of (a,c,e,g) production
$k_z^+ y^+\,\hat {P}^+(y^+,\lambda _z^+)$ and (b,d, f,h) turbulent transport
$k_z^+ y^+\,\hat {T}_{turb}^+(y^+,\lambda _z^+)$, for (a,b) LES, (c,d) GQLZ1, (e, f) GQLZ5, and (g,h) GQLZ25 cases. Here, the vertical line represents the spanwise cut-off wavelength (
$\lambda _{z,c}$) dividing the (a,c,e,g)
$\mathcal {P}_h$-subspace and (b,d, f,h)
$\mathcal {P}_l$-subspace regions.

The GQLZ1 case shows spectra with increased energy and a production peak centred in
$\lambda _z^+ \approx 400$ (figures 5c,d). The production spectra do not extend for
$\lambda _z^+ \lesssim 200$, and they are damaged at large scales (
$\lambda _z/h \gtrsim 1$), except in the outer region (figure 5c). Moreover, similarly to the velocity spectra (figure 3), the linear scaling of
$\lambda _z^+$ with
$y^+$ is found to have been weakened. The turbulent transport spectra show positive and negative regions that have been reduced over the entire spanwise wavelengths compared to the reference LES (figure 5d). Also, the weakly positive spectra in the region close to the wall are confined to the largest spanwise scales
$\lambda _z^+ \approx 5000$. This is expected somehow, because the linearised part of the GQLZ1 model would not admit a full energy cascade mechanism due to the absence of the self-interacting nonlinear terms in the
$\mathcal {P}_h$-subspace – the spanwise energy cascade should be admitted only through the interaction between
$\boldsymbol {u}_l$ and
$\boldsymbol {u}_h$ (i.e. the ‘scattering’ mechanism proposed by Tobias & Marston Reference Tobias and Marston2017). By enriching the
$\mathcal {P}_l$-subspace group with more spanwise Fourier modes, the resulting spectra start to recover the original range of spanwise wavelengths, the scaling with
$y$, and the original position of the peak in the GQLZ5 and GQLZ25 cases. The recovery of the spectra to those of the full LES, however, appears to be a little delicate, and this is seen in the turbulent transport spectra of GQLZ25 (figure 5h). We note that turbulent transport spectra for
$\lambda _{z}<\lambda _{z,c}$ (i.e. the
$\mathcal {P}_h$-subspace group) in the outer region (say
$y^+\gtrsim 300$) are almost zero, while showing a sharp discontinuity across
$\lambda _{z}=\lambda _{z,c}$. This suggests that in the outer region, the energy cascade from the
$\mathcal {P}_l$-subspace to the
$\mathcal {P}_h$-subspace does not exist in the GQL25 model. This issue will be discussed further in § 4.3.4.
The premultiplied streamwise wavenumber spectra of the energy budget are shown in figure 6. For the full LES, the production spectra and the negative part of the turbulent transport spectra appear to be in the region bounded by two linear ridges, along which the wall-normal and streamwise velocity spectra are highly energetic (i.e.
$y =0.35 \lambda _x$ and
$y =0.01 \lambda _x$ in figure 4). We note that in the log region, the premultiplied production spectra,
$-k_x^+ y^+\,\varPhi _{uv}^+(k_x,y)\,\textrm {d}U^+/{\textrm {d} y}^+$, are similar to those of Reynolds shear-stress co-spectra,
$-k_x^+\,\varPhi _{uv}^+(k_x,y)$, due to
$\textrm {d}U^+/{\textrm {d} y}^+ \sim 1/y^+$ (Cho et al. Reference Cho, Hwang and Choi2018). However, the linear scaling of the production spectra with respect to
$y$ is not very clear, and presumably this is due to the relatively low Reynolds numbers in the present study (for high Reynolds numbers, e.g.
$Re_\tau \approx 5200$, see Lee & Moser (Reference Lee and Moser2019) and Hwang & Lee (Reference Hwang and Lee2020), where the linear scaling is clearly visible.) The turbulent transport spectra in figure 6(b) show a region of positive values along the ridge corresponding to viscous dissipation (
$\lambda _x=57 \eta$), and a region of negative values corresponding to the streamwise production intensity peak in the near-wall region (
$\lambda _x^+ \approx 1000$).
Premultiplied streamwise wavenumber spectra of (a,c,e,g) production
$k_x^+ y^+\,\hat {P}^+(y^+,\lambda _x^+)$ and (b,d, f,h) turbulent transport
$k_x^+ y^+\,\hat {T}_{turb}^+(y^+,\lambda _x^+)$, for (a,b) LES, (c,d) GQLZ1, (e, f) GQLZ5, and (g,h) GQLZ25 cases.

The spectra of GQLZ1 show some resemblance to the LES case but still a very damaged energy cascade compared to that case (figure 6c). In particular, the production spectra exhibit an intense peak highly localised around
$\lambda _x^+ \simeq 5000$ and
$y^+ \simeq 25$, while a secondary peak is also observed in the outer region. The production spectra show no energy intensity for wavelengths below
$\lambda _x^+ = 700$ (figure 6c). The positive region of the transport spectra appear at approximately the same locations for larger streamwise length scales compared to LES (peaks at
$\lambda _x^+ \approx 1000$ and
$\lambda _x^+ \approx 500$, respectively), while the negative region is displaced to higher streamwise wavelengths and reduced to the
$y^+ \approx 30$ and
$y^+ \approx 500$ locations. As more spanwise modes are incorporated in the
$\mathcal {P}_l$-subspace group, the scalings of both spectra with
$y$ and
$\eta$ are greatly recovered, i.e. the GQLZ5 and GQLZ25 cases (figures 6e–h).
3.3. Componentwise energy transport and pressure strain
Finally, the pressure strain spectra are explored to understand the mechanism of componentwise turbulent kinetic energy distribution in the GQL model. Figure 7 shows the spanwise wavenumber spectra of the pressure strain terms for LES and the GQL models. For all cases, a negative
$\hat {\varPi }_x$ and mainly positive
$\hat {\varPi }_y$ and
$\hat {\varPi }_z$ (combined into
$\hat {\varPi }_{yz}=\hat {\varPi }_y + \hat {\varPi }_z$) can be observed in the spanwise wavenumber space. This means that the pressure strain redistributes the energy produced at the streamwise component of the turbulent kinetic energy through the production term to the wall-normal and spanwise components. The spectra appear to be distributed approximately along the linear ridge
$y=0.2 \lambda _z$. The GQLZ1 case underestimates the spectral energy intensity of all pressure strain terms, and their spanwise wavenumber spectra do not extend to length scales below
$\lambda _z^+ \approx 200$. The two features are improved as
$\lambda _{z,c}$ is decreased, and the spectra reach smaller spanwise scales in the GQLZ5 case (
$\lambda _z^+ \approx 100$). Both spectra of the
$\mathcal {P}_h$-subspace group of GQLZ5 and GQLZ25 reproduce with good agreement the spectra of the LES. Figure 8 shows the streamwise wavenumber spectra of the pressure strain terms. The pressure strain intensity appears to be concentrated for wavelengths
$\lambda _x^+ \gtrsim 300$ in the GQLZ1 case, but the spectra quickly extend to smaller
$\lambda _x$ as the number of spanwise modes in the
$\mathcal {P}_l$-subspace group is increased. The streamwise spectra of GQLZ5 and GQLZ25 show excellent agreement with those of the LES.
Premultiplied spanwise wavenumber spectra of (a,c,e,g)
$k_z^+ y^+\, \hat {\varPi }_x^+(y^+,\lambda _z^+)$ and (b,d, f,h)
$k_z^+ y^+\, \hat {\varPi }_{yz}^+(y^+,\lambda _z^+)$, for (a,b) LES, (c,d) GQLZ1, (e, f) GQLZ5, and (g,h) GQLZ25. Here, the vertical line represents the spanwise cut-off wavelength (
$\lambda _{z,c}$) dividing the (a,c,e,g)
$\mathcal {P}_h$-subspace and (b,d, f,h)
$\mathcal {P}_l$-subspace regions. The dashed lines indicate the ridge
$y=0.5 \lambda _z$.

Premultiplied streamwise wavenumber spectra of (a,c,e,g)
$k_x^+ y^+\, \hat {\varPi }_x^+(y^+,\lambda _x^+)$ and (b,d, f,h)
$k_x^+ y^+\, \hat {\varPi }_{yz}^+(y^+,\lambda _x^+)$, for (a,b) LES, (c,d) GQLZ1, (e, f) GQLZ5, and (g,h) GQLZ25. The dashed lines indicate the ridge
$y=0.35 \lambda _x$.

As was shown in Part 1, the GQL approximations change only the equations for the ‘slow pressure’,
$p_l^S$, given by
in the
$\mathcal {P}_l$-subspace, and
in the
$\mathcal {P}_h$-subspace. In the GQL models, the second term on the right-hand side of (3.1a), and the first and last terms on the right-hand side of (3.1b), are absent. Therefore, similarly to the GQL approximations considered in the streamwise direction in Part 1, the GQL models here should also experience the lack of pressure strain transport mediated by such slow pressure, explaining why the GQL models show highly anisotropic turbulent fluctuations biased to the streamwise component, especially when
$\lambda _{z,c}$ is large (see figure 2).
4. Discussion
4.1. Comparison with GQL approximations in the streamwise direction
Thus far, we have investigated a set of GQL approximations and compared its turbulence statistics, spectra and energy transfer to those of a full LES at
$Re_\tau \simeq 1700$ in turbulent channel flow. For the GQL approximations in this study, the flow is decomposed into a group of low spanwise wavenumber Fourier modes (
$\mathcal {P}_l$-subspace) and the rest (
$\mathcal {P}_h$-subspace); see (2.1). The former group is solved considering the full nonlinear equations, whereas the latter is obtained by solving the linearised equations about the former. The decomposition employed for the GQL models in the present study evidently has different physical implications to that in Part 1. The QL and GQL models in Part 1 (or RNL from Thomas et al. Reference Thomas, Lieu, Jovanović, Farrell, Ioannou and Gayme2014, Reference Thomas, Farrell, Ioannou and Gayme2015; Farrell et al. Reference Farrell, Gayme and Ioannou2017) place a minimal model for the self-sustaining process at each of the (spanwise) integral length scales, and examine how their nonlinear coupling dynamics and the related streamwise energy cascade play a role in the description of full turbulent dynamics (i.e. the GQL models in Part 1). On the contrary, in the present study, the flow field in the
$\mathcal {P}_l$-subspace is supposed to fully incorporate the self-sustaining process at least within the given integral length scales, while that in the
$\mathcal {P}_h$-subspace is not expected to support the self-sustaining process (see § 4.2). Despite the significant difference in the physical implications between the GQL approximations considered in Part 1 and in this study, it is worth pointing out that the two different types of GQL models exhibit some notable similarities, which would originate from the general nature of the GQL approximation.
First, the GQL models with large cut-off wavelengths (i.e.
$\lambda _{x,c}(\equiv 2{\rm \pi} /k_{x,c})$ in Part 1 and
$\lambda _{z,c}$ in this study) exhibit a considerably reduced multi-scale behaviour. Indeed, the QL model in Part 1 and the GQLZ1 case in this study show spanwise wavenumber spectra, where the energy density is focused on a particular length scale (figures 3c,d in Part 1 and figures 3c,d here) instead of being spread over a wide range of the spanwise wavenumbers. As discussed in detail in § 4.1 of Part 1, the blockage of nonlinear energy transfer in the GQL approximations with a large cut-off wavelength appears to elevate energy and Reynolds shear stress at a particular integral length scale. We note that the dimensionless streamwise mean-momentum equation in pressure-driven channel flow is given by
Here,
$\varPhi _{uv}(y;k_x,k_z)$ is the two-dimensional Reynolds shear-stress co-spectrum, which is mostly negative, and
$\tau _w$ is the mean wall shear stress. Equation (4.1) implies that the elevation of the Reynolds shear-stress spectrum magnitude around particular
$k_x$ and
$k_z$ would lead to a significantly reduced mean shear at the related wall-normal location and its nearby locations. In particular, the change of the mean shear in the nearby wall-normal locations appears to affect negatively the turbulence production at other wavenumbers, resulting in a reduced multi-scale behaviour. This has been observed in both Part 1 and here, indicating that it is rather common regardless of the nature of the flow decomposition at least in the present channel flow, especially for large cut-off wavelength. This implies that a key to the successful prediction of low-order turbulence statistics using the QL/GQL framework would lie in an accurate description of the balanced mean–fluctuation interactions across the wide range of integral length scales. Indeed, much simpler quasilinear models (Hwang & Eckhardt Reference Hwang and Eckhardt2020; Skouloudis & Hwang Reference Skouloudis and Hwang2021), which are more controllable theoretically and computationally, were found to be reasonably successful in the description of general statistical behaviours of the same flow primarily by focusing on the modelling of the mean–fluctuation interactions.
Second, when
$\lambda _{z,c}$ or
$\lambda _{x,c}$ is large, the GQL models in Part 1 as well as in the present study exhibit a highly anisotropic turbulent fluctuation that contains significantly elevated energy in the streamwise component. Since the turbulence production in parallel wall-bounded shear flows appears only in the streamwise component, this implies that, in general, GQL approximations prevent the energy distribution mechanism into the other velocity components. As depicted in Part 1 and § 3.3, the pressure fluctuation may be decomposed as
$p=p^F+p^S$, where
$p^F$ is the ‘fast pressure’ related to the linear part of the advection term in the velocity fluctuation equations, and
$p^S$ is its nonlinear part. Using this decomposition, the pressure strain transport, satisfying (2.9), is written as
where the superscripts
$^F$ and
$^S$ indicate the pressure strain transport by the fast and slow pressures, respectively. Given that any QL/GQL approximations ignore certain nonlinear fluctuation–fluctuation interactions, they would damage and/or disrupt the slow pressure originating from the nonlinear fluctuation–fluctuation interactions. The turbulent kinetic energy transfer between the different velocity components is mediated largely by the pressure strain transport through (4.2), which originates essentially from the continuity equation (Cho et al. Reference Cho, Hwang and Choi2018). In particular, the streamwise pressure strains,
$\hat {\varPi }_{x}^F$ and
$\hat {\varPi }_{x}^S$, are negative, while the rest components are positive, indicating that the pressure strain transfers the energy produced at the streamwise component to the other two components. It is also this process that converts highly anisotropic energy-containing eddies at integral length scales into isotropic small scales during the energy cascade of the flow. This explains why, commonly, the QL/GQL models in Part 1 and in the present study exhibit more anisotropic turbulent fluctuations across all scales – the pressure strain transport by slow pressure through (4.2) is damaged in the QL/GQL models. This observation would be important if one wished to improve or develop a QL model simpler than the GQL models considered here, as the role of the pressure strain transport mediated by slow pressure needs to be taken into account for the correct prediction of anisotropy of the given flow.
4.2. Statistical description of near-wall turbulence and the self-sustaining process
The near-wall turbulence approximately below
$y^+ \approx 50 \unicode{x2013} 70$ has been understood to be largely independent of the motions in the log and outer regions. It has been shown consistently that the absence and/or disruption of the log- and outer-region motions in turbulence does not affect the near-wall dynamics, which attains the self-sustaining process at the viscous inner length scale (e.g. Jiménez & Moin Reference Jiménez and Moin1991; Hamilton et al. Reference Hamilton, Kim and Waleffe1995; Jiménez & Pinelli Reference Jiménez and Pinelli1999; Hwang Reference Hwang2013; Doohan, Willis & Hwang Reference Doohan, Willis and Hwang2019). Here, the ‘self-sustaining process’ refers to the particular nonlinear mechanism that drives turbulence utilising the energy from linearly stable mean flow
$\boldsymbol {U}$ in wall-bounded turbulent shear flows (e.g. Hamilton et al. Reference Hamilton, Kim and Waleffe1995; Waleffe Reference Waleffe1997), i.e. the two-way interaction between ‘streamwise elongated’ structures of streamwise velocity (streaks) and their ‘streamwise wavy’ instability involving cross-streamwise velocities (waves and rolls). This early observation extends to the log and outer regions, where similar self-sustaining processes have been found to exist at larger scales in the form of Townsend's attached eddies (Flores & Jiménez Reference Flores and Jiménez2010; Hwang & Cossu Reference Hwang and Cossu2010b, Reference Hwang and Cossu2011; Hwang Reference Hwang2015; Hwang & Bengana Reference Hwang and Bengana2016).
Now, to examine the relevance of the self-sustaining process in the
$\mathcal {P}_h$-subspace of the GQL model, we compare the equations of motion in the
$\mathcal {P}_h$-subspace from the Navier–Stokes simulation with those in the
$\mathcal {P}_h$-subspace from the GQL model. For a mathematical description of the self-sustaining process in the
$\mathcal {P}_h$-subspace of a full simulation (or the full LES), let us consider the ‘full’ Navier–Stokes equations (without the GQL approximation), while retaining each of the
$\mathcal {P}_l$- and
$\mathcal {P}_h$-subspaces defined by the given non-zero spanwise threshold wavenumber
$k_{z,c}$ in (2.1). With the two
$\mathcal {P}_l$- and
$\mathcal {P}_h$-subspaces defined by
$k_{z,c}$ (or
$\lambda _{z,c}$), we will now assume a situation where
$\boldsymbol {u}_l=\boldsymbol {0}$, i.e.
$\hat {\boldsymbol {u}}_{m,n}=0$ for every
$m$ and
$|n|\leqslant M_{z,F}$ in (2.1), except for
$(m,n)=(0,0)$. Employing the velocity decomposition used in Part 1 for
$\boldsymbol {u}_h$ leads to
where
$\boldsymbol {u}_{h,0}=\langle \boldsymbol {u}_{h} \rangle _x$. The equations for
$\boldsymbol {u}_{h,0}$ and
$\boldsymbol {u}_{h,r}$ are then given by
\begin{align} \frac{\partial
\boldsymbol{u}_{h,0}}{\partial t}+({\boldsymbol{u}}_{h,0}
\boldsymbol{\cdot}\boldsymbol{\nabla})\boldsymbol{U}&={-}\frac{1}{\rho}\,\boldsymbol{\nabla}
p_{h,0}+ \nu\,{\nabla}^2
\boldsymbol{u}_{h,0}
-\mathcal{P}_{h}[({\boldsymbol{u}}_{h,0}\boldsymbol{\cdot}\boldsymbol{\nabla})
{\boldsymbol{u}}_{h,0}]\nonumber\\ &\quad
-\langle\mathcal{P}_{h}
[({\boldsymbol{u}}_{h,r}\boldsymbol{\cdot}\boldsymbol{\nabla})
{\boldsymbol{u}}_{h,r}]\rangle_{x}
\end{align}
and
\begin{align} &\frac{\partial
\boldsymbol{u}_{h,r}}{\partial
t}+(\boldsymbol{U}\boldsymbol{\cdot}\boldsymbol{\nabla})
{\boldsymbol{u}}_{h,r}+({\boldsymbol{u}}_{h,r}\boldsymbol{\cdot}\boldsymbol{\nabla})
\boldsymbol{U}\nonumber\\ &\quad
={-}\frac{1}{\rho}\,\boldsymbol{\nabla} p_{h,r}+
\nu\,{\nabla}^2 \boldsymbol{u}_{h,r}
-\mathcal{P}_{h}[({\boldsymbol{u}}_{h,0}\boldsymbol{\cdot}\boldsymbol{\nabla})
\boldsymbol{u}_{h,r}]-\mathcal{P}_{h}[(\boldsymbol{u}_{h,r}
\boldsymbol{\cdot}\boldsymbol{\nabla}){\boldsymbol{u}}_{h,0}]
\nonumber\\ &\qquad
-\mathcal{P}_{h}[({\boldsymbol{u}}_{h,r}\boldsymbol{\cdot}\boldsymbol{\nabla})
{\boldsymbol{u}}_{h,r}]+\langle\mathcal{P}_{h}[(
{\boldsymbol{u}}_{h,r}\boldsymbol{\cdot}\boldsymbol{\nabla}){\boldsymbol{u}}_{h,r}]\rangle_x,
\end{align}
where
$p_{h,0}$ and
$p_{h,r}$ are defined to enforce
$\boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {u}_{h,0}=0$ and
$\boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {u}_{h,r}=0$, respectively, with
$p_h=p_{h,0}+p_{h,r}$, and the eddy viscosity term for LES is ignored for simplicity. Here,
$\boldsymbol {u}_{h,0}$ in (4.4a) describes the evolution of the streamwise elongated (or constant) streaks and rolls, whereas
$\boldsymbol {u}_{h,r}$ in (4.4b) depicts the streak instability (i.e. instability of a streaky flow,
$\boldsymbol {U}+\boldsymbol {u}_{h,0}$) – note that (4.4b) in the absence of its third line containing self-interacting nonlinear terms gives the linearised equations about
$\boldsymbol {U}+\boldsymbol {u}_{h,0}$ in the
$\mathcal {P}_h$-subspace (i.e. the QL model in Part 1). From (4.4), it is seen that
$\boldsymbol {u}_{h,0}$ is driven by the presence of
$\boldsymbol {u}_{h,r}$ (i.e. the last term in (4.4a)), whereas
$\boldsymbol {u}_{h,r}$ originates physically from the inflectional instability, transient growth and/or parametric mechanism depicted primarily by the second term in the second line of (4.4b) (Hamilton et al. Reference Hamilton, Kim and Waleffe1995; Schoppa & Hussain Reference Schoppa and Hussain2002; Park, Hwang & Cossu Reference Park, Hwang and Cossu2011; Farrell & Ioannou Reference Farrell and Ioannou2012; de Giovanetti, Hwang & Choi Reference de Giovanetti, Hwang and Choi2016; de Giovanetti et al. Reference de Giovanetti, Sung and Hwang2017; Lozano-Durán et al. Reference Lozano-Durán, Constantinou, Nikolaidis and Karp2021). This is a two-way interaction between
$\boldsymbol {u}_{h,0}$ (streaks) and
$\boldsymbol {u}_{h,r}$ (streamwise instability wave/quasi-streamwise vortices), which sustains turbulence in the absence of
$\boldsymbol {u}_{l}$.
Now we consider the equations of motions for
$\boldsymbol {u}_h$ from the GQL models given by (2.2a,b) for the purpose of comparing with (4.4) from the full equations of motion. Unlike the case described in (4.4), here we do not assume
$\boldsymbol {u}_l = \boldsymbol {0}$. Further to the velocity decomposition in (4.3),
$\boldsymbol {u}_l$ is also decomposed such that
where
$\boldsymbol {u}_{l,0}=\langle \boldsymbol {u}_{l} \rangle _x$. For the GQL models in this study, the equations for
$\boldsymbol {u}_{h,0}$ and
$\boldsymbol {u}_{h,r}$ are given by
\begin{align} &\frac{\partial
\boldsymbol{u}_{h,0}}{\partial
t}+(\boldsymbol{U}\boldsymbol{\cdot}\boldsymbol{\nabla})
{\boldsymbol{u}}_{h,0}+({\boldsymbol{u}}_{h,0}\boldsymbol{\cdot}\boldsymbol{\nabla})\boldsymbol{U}\nonumber\\
&\quad={-}\frac{1}{\rho}\,\boldsymbol{\nabla} p_{h,0}+
\nu\,{\nabla}^2 \boldsymbol{u}_{h,0}
-\mathcal{P}_{h}[(\boldsymbol{u}_{h,0}\boldsymbol{\cdot}\boldsymbol{\nabla})
\boldsymbol{u}_{l,0}]-\mathcal{P}_{h}
[(\boldsymbol{u}_{l,0}\boldsymbol{\cdot}\boldsymbol{\nabla})\boldsymbol{u}_{h,0}]
\nonumber\\ &\qquad
-\langle\mathcal{P}_{h}[(\boldsymbol{u}_{h,r}\boldsymbol{\cdot}\boldsymbol{\nabla}
)\boldsymbol{u}_{l}]\rangle_{x}-\langle\mathcal{P}_{h}[
(\boldsymbol{u}_{l}\boldsymbol{\cdot}\boldsymbol{\nabla})\boldsymbol{u}_{h,r}]\rangle_x
\end{align}
and
\begin{align} &\frac{\partial
\boldsymbol{u}_{h,r}}{\partial
t}+(\boldsymbol{U}\boldsymbol{\cdot}\boldsymbol{\nabla})
\boldsymbol{u}_{h,r}+(\boldsymbol{u}_{h,r}\boldsymbol{\cdot}\boldsymbol{\nabla}
)\boldsymbol{U}\nonumber\\ &\quad
={-}\frac{1}{\rho}\,\boldsymbol{\nabla}p_{h,r}+ \nu\,
{\nabla}^2
\boldsymbol{u}_{h,r}-\mathcal{P}_{h}[(\boldsymbol{u}_{h,r}\boldsymbol{\cdot}\boldsymbol{\nabla}
)\boldsymbol{u}_{l}]-\mathcal{P}_{h}[(\boldsymbol{u}_{l}
\boldsymbol{\cdot}\boldsymbol{\nabla})\boldsymbol{u}_{h,r}]
\nonumber\\ &\qquad
+\langle\mathcal{P}_{h}[(\boldsymbol{u}_{h,r}
\boldsymbol{\cdot}\boldsymbol{\nabla})\boldsymbol{u}_{l}]
\rangle_{x}+\langle\mathcal{P}_{h}[(\boldsymbol{u}_{l}
\boldsymbol{\cdot}\boldsymbol{\nabla})\boldsymbol{u}_{h,r}]\rangle_x.
\end{align}
Comparing (4.4) with (4.6), the differences between the turbulent fluctuations in the
$\mathcal {P}_h$-subspace from the self-sustaining process and the GQL models in the present study are now evident. If (4.6a) admits a non-trivial
$\boldsymbol {u}_{h,0}$ for
$\boldsymbol {U}$ that does not yield any instability, then it should be from the interaction between
$\boldsymbol {u}_l$ and
$\boldsymbol {u}_h$ (the last two lines in (4.6a)). In such a case, if
$\boldsymbol {u}_l=\boldsymbol {0}$, then there is no way for (4.6a) and (4.6b) to have a non-trivial statistically stationary solution. Furthermore, the term that depicts an instability and/or transient growth mechanism from a streaky flow (i.e. the second term in the second line of (4.4b)) does not exist in (4.6b). In this case, possibly an instability may exist from the two second terms, respectively, in the second and third lines of (4.6b), especially when
$\boldsymbol {u}_l$ is non-zero. Importantly, if
$\boldsymbol {u}_l$ is a chaotic velocity field, then the non-trivial solution
$\boldsymbol {u}_h$ in (4.6b) emerges in the form of the leading neutrally stable Lyapunov vector of (4.6), as was discussed in Farrell et al. (Reference Farrell, Gayme and Ioannou2017) and in Part 1.
Given the discussion above, it is crucial to have non-zero
$\boldsymbol {u}_l$ for (4.6) (i.e. the equations in the
$\mathcal {P}_h$-subspace in the present study), whereas (4.4) does not need
$\boldsymbol {u}_l$ to create its own dynamics. The equations for the
$\mathcal {P}_l$-subspace have been omitted here; we note that when the cut-off wavelength is large, the equations do support the self-sustaining process by themselves, and eventually
$\boldsymbol {u}_h$ becomes irrelevant (e.g. GQLZ64 case). Despite this important difference between turbulent fluctuations from the self-sustaining process (4.4) and from the GQL model (4.6), it is remarkable to observe that the GQL model admits velocity spectra fairly similar to those of the full LES (GQLZ25, in particular; compare figures 4a,b with 4g,h). We note that both (4.4) and (4.6) share the same Navier–Stokes operator linearised about the mean velocity
$\boldsymbol {U}$. In this respect, they can also be viewed as the linearised Navier–Stokes operator driven by chaotic fluctuations, the idea resonating with the previous and ongoing linearised Navier–Stokes-based modelling efforts for turbulent flows (e.g. Hwang & Cossu Reference Hwang and Cossu2010a; McKeon & Sharma Reference McKeon and Sharma2010; Moarref et al. Reference Moarref, Sharma, Tropp and McKeon2013; Zare, Jovanović & Georgiou Reference Zare, Jovanović and Georgiou2017; McKeon Reference McKeon2017; Hwang & Eckhardt Reference Hwang and Eckhardt2020; Skouloudis & Hwang Reference Skouloudis and Hwang2021; Jovanović Reference Jovanović2021). Although such approaches would not be able to incorporate fully some key nonlinear processes described in (4.4) and (4.6) without additional modelling efforts, the observations here might offer an explanation for why the linearised Navier–Stokes-based modelling efforts have been useful to study coherent structures in wall-bounded turbulence. One must, however, be cautious about this interpretation, because the linearised approaches often make the assumption of state-independent randomness in the excitation (e.g. stochastic forcing in time or frequency domain) of the linearised operator about the prescribed
$\boldsymbol {U}$ obtained from DNS. However, this is not the case for
$\boldsymbol {u}_h$ in the GQL model, as seen in (4.6) where the nonlinear terms depend on
$\boldsymbol {u}_l$. Furthermore, such approaches have to consider linearisation about the ‘correct mean’ prescribed from DNS data. However, this is no longer needed in the QL/GQL models, where the mean and fluctuations are solved together in a self-consistent manner.
4.3. Scale interactions
As discussed in § 1, the GQL approximation can be utilised as an interventional tool to examine scale interactions, given that it is capable of suppressing particular triadic interactions in a controlled manner (see figure 1; see also table 2). Here, we consider the GQLZ25 case as a baseline, where
$\lambda _{z,c}^+ \simeq 246$ (
$\lambda _{z,c}/h\simeq 0.13$) is given for the decomposition of velocity in (2.1). Subsequently, we carry out four additional numerical experiments, in which certain types of triadic interactions between
$\mathcal {P}_l$- and
$\mathcal {P}_h$-subspace groups are suppressed artificially. In this respect, it is worth mentioning how important each of the triadic interactions in figure 1 is expected to be. The most crucial triadic interaction would be LL-L (figure 1a), regardless of the value of
$\lambda _{z,c}$, as it is involved directly in energy exchange with mean shear. The interaction LH-L (figure 1b) appears to be not so important, as this is often ignored in the GQL approximation – this was tested in Part 1 and here as well (not shown). All the other triadic interactions in figure 1 are expected to be important especially if
$\lambda _{z,c}^+ \gg 100$: the HH-L interaction (figure 1c) describes the energy transfer from small to large scale (including the mean) (Cho et al. Reference Cho, Hwang and Choi2018); the LL-H interaction (figure 1d) is the main mechanism of energy cascade (Cho et al. Reference Cho, Hwang and Choi2018); the HL-H interaction (figure 1e) depicts the scattering mechanism that would essentially make the GQL approximation work; the HH-H interaction (figure 1f) would be important especially when
$\lambda _{z,c}$ is sufficiently large, as it would depict all the nonlinear interactions, including the energy cascade, within the
$\mathcal {P}_h$-subspace group. In principle, the full details of the importance of each triadic interaction can be investigated by performing repeatedly the same numerical experiments for different values of
$\lambda _{z,c}$. However, this requires a large number of numerical experiments, which depends on the choice of
$\lambda _{z,c}$, if one wishes to unveil the full triadic interaction dynamics. This is beyond the scope of the present study. Therefore, in the present study, we will focus on the GQLZ25 case for the purpose of conceptual demonstration.
Triadic interactions in the present simulation cases. The tickmark (✓) denotes inclusion, whereas the crossmark (✗) stands for removal of the corresponding triadic interaction. Here, the types of the triadic interactions are described in figure 1.

The four additional numerical experiments are listed in table 2. Their first- and second-order turbulence statistics (mean and velocity fluctuations) are reported in figure 9. The spanwise and streamwise wavenumber spectra of streamwise and spanwise velocities are shown in figures 10 and 11, and those of production and turbulent transport spectra are given in figures 12 and 13. We note that the QL/GQL approximations were introduced to model the dynamics using appropriate statistical frameworks by linearising the
$\mathcal {P}_h$-subspace equations: for example, the QL approximation has been thought to be equivalent to the cumulant expansion of the Navier–Stokes equations, while ignoring any third-order cumulant contribution. However, all the four cases considered here do not lead to linear
$\mathcal {P}_h$-subspace equations, and they are not useful for modelling the statistical state dynamics.
First- and second-order turbulence statistics for the present LES, DNS at
$Re_\tau =2003$ (Hoyas & Jiménez Reference Hoyas and Jiménez2008), TRIAZ25, TRIBZ25, TRICZ25 and TRIDZ25 cases: (a)
$u_{rms}^+(y^+)$, (b)
$v_{rms}^+(y^+)$, (c)
$U^+(y^+)$, and (d)
$-\langle u^{\prime } v^{\prime }\rangle _{x,z,t}^+(y^+)$.

Premultiplied spanwise wavenumber spectra of (a,c,e,g) streamwise
$k_z^+\, \varPhi _{uu}^+(y^+,\lambda _z^+)$ and (b,d, f,h) wall-normal
$k_z^+\, \varPhi _{vv}^+(y^+,\lambda _z^+)$ velocity, for (a,b) TRIAZ25, (c,d) TRIBZ25, (e, f) TRICZ25, and (g,h) TRIDZ25 cases.

Premultiplied streamwise wavenumber spectra of (a,c,e,g) streamwise
$k_x^+\, \varPhi _{uu}^+(y^+,\lambda _x^+)$ and (b,d, f,h) wall-normal
$k_x^+\, \varPhi _{vv}^+(y^+,\lambda _x^+)$ velocity, for (a,b) TRIAZ25, (c,d) TRIBZ25, (e, f) TRICZ25, and (g,h) TRIDZ25 cases. Here, the vertical line represents the spanwise cut-off wavelength (
$\lambda _{z,c}$) dividing the (a,c,e,g)
$\mathcal {P}_h$-subspace and (b,d, f,h)
$\mathcal {P}_l$-subspace regions.

Premultiplied spanwise wavenumber spectra of (a,c,e,g) production
$k_z^+ y^+\,\hat {P}^+(y^+,\lambda _z^+)$ and (b,d, f,h) turbulent transport
$k_z^+ y^+\,\hat {T}_{turb}^+(y^+,\lambda _z^+)$, for (a,b) TRIAZ25, (c,d) TRIBZ25, (e, f) TRICZ25, and (g,h) TRIDZ25 cases.

Premultiplied streamwise wavenumber spectra of (a,c,e,g) production
$k_x^+ y^+\,\hat {P}^+(y^+,\lambda _x^+)$ and (b,d, f,h) turbulent transport
$k_x^+ y^+\,\hat {T}_{turb}^+(y^+,\lambda _x^+)$, for (a,b) TRIAZ25, (c,d) TRIBZ25, (e, f) TRICZ25, and (g,h) TRIDZ25 cases. Here, the vertical line represents the spanwise cut-off wavelength (
$\lambda _{z,c}$) dividing the (a,c,e,g)
$\mathcal {P}_h$-subspace and (b,d, f,h)
$\mathcal {P}_l$-subspace regions.

4.3.1. TRIAZ25: self-interactions within the
$\mathcal {P}_h$-subspace
The TRIAZ25 case explores the contributions of the self-interacting nonlinear terms in the
$\mathcal {P}_h$-subspace to its own subspace (i.e.
$\mathcal {P}_h$-subspace), and this is represented by the triadic interaction HH-H in figure 1. In the TRIAZ25 case, this interaction term is activated further to the GQL25 model, and it has been designed to understand the importance of such an interaction in the GQL approximation. In fact, figure 9 shows that the statistics of TRIAZ25 resembles that of GQLZ25 the most, compared to other cases, showing that the effect of this triadic interaction on the low-order statistics of the GQL approximation is moderate. The spanwise wavenumber velocity spectra of the TRIAZ25 case (figures 10a,b and 11a,b) are also almost identical to those of the GQLZ25 case (figures 3g,h and 4g,h). The same holds true for most of the energy spectra (figures 12a,b and 13a,b), although some minor differences can be observed in the near-wall region.
It should be mentioned that the addition of the triadic interaction HH-H in figure 1 is associated directly with the activation of the nonlinear terms associated with the self-sustaining process in the
$\mathcal {P}_h$-subspace. This becomes evident from (4.4), where the contributions of the self-interacting nonlinear terms in the
$\mathcal {P}_h$-subspace to its own subspace (i.e.
$\mathcal {P}_h$-subspace) were shown to be directly associated with some key sub-processes of the self-sustaining process. A possible explanation for this is that there may exist competition and/or cooperation between the self-sustaining process and the scattering mechanism in driving the near-wall turbulent fluctuations, and in the presence of all scales of turbulent flow, the scattering mechanism may also play an important role.
4.3.2. TRIBZ25: inverse energy transfer
The TRIBZ25 case eliminates the contributions of the self-interacting nonlinear terms in the
$\mathcal {P}_h$-subspace to the
$\mathcal {P}_l$-subspace group (i.e. the inverse energy transfer from the
$\mathcal {P}_h$-subspace to the
$\mathcal {P}_l$-subspace), but activates all triadic interactions in the
$\mathcal {P}_h$-subspace group. In this case, it is found that its turbulence statistics shows major differences with respect to GQLZ25 and the other cases (figure 9). First,
$-\langle u^{\prime } v^{\prime }\rangle _{x,z,t}^+$ is greater than 1, and
$u_{rms}^+$ and
$v_{rms}^+$ are also much greater than their values for GQLZ25. This strong overestimation of turbulent fluctuations of TRIBZ25 can be explained with the following dimensionless streamwise mean-momentum equation:
In the TRIBZ25 case, the third term in (4.7) is omitted artificially, thus the mean is balanced only with
$\langle u_l v_l \rangle _{x,z,t}$. We note that the present simulation imposes a constant mass flow rate and that
$\langle u_h v_h \rangle _{x,z,t}$ is not taken into account in (4.7) in the TRIBZ25 case. Therefore, no matter how strong
$-{\langle u_h v_h \rangle _{x,z,t}}$ is, the mean shear has no direct control in reducing its strength, explaining the breakdown of (4.7) in the TRIBZ25 case.
In fact, the spanwise wavenumber spectra of velocity and energy budget in figures 10(c,d) and 12(c,d) show large elevation of the fluctuation energy in the
$\mathcal {P}_h$-subspace in the TRIBZ25 case. This effect is also seen in the streamwise wavenumber spectra in figures 11(c,d) and 13(c,d), where spectral intensity is quite high in the region close to the wall where the spectral intensity of the
$\mathcal {P}_h$-subspace group would be important. This behaviour might have given the impression that the range of spanwise wavenumbers covered by the spectra is reduced in the TRIBZ25 case. However, it is important to note that turbulent velocity fluctuations in the outer regions remain more or less the same as those of GQLZ25 and LES (see figure 9), indicating that the spectral intensity of the outer region in the
$\mathcal {P}_l$-subspace is actually not affected significantly. The same behaviour is also observed in all the other spectra in figures 11(c,d), 12(c,d) and 13(c,d). This confirms the outer-similarity hypothesis of Townsend (Reference Townsend1976), who conjectured that turbulence in the outer region remains largely independent of the near-wall dynamics, consistent with many previous studies conducted with surface roughness (e.g. Flores, Jiménez & Del Álamo Reference Flores, Jiménez and Del Álamo2007).
Finally, it should be mentioned that the spanwise wavenumber spectra of turbulent transport in the TRIBZ25 case exhibit a strong relative suppression of the near-wall positive turbulent transport in the
$\mathcal {P}_h$-subspace (figure 12d) in comparison to LES (figure 5b) and GQLZ25 (figure 5h). The weakly positive turbulent transport observed in GQL25 and LES was proposed originally to be the main consequence of inverse energy cascade by Cho et al. (Reference Cho, Hwang and Choi2018), who visualised the full triadic nonlinear interactions in a turbulent channel flow. However, some recent studies have suggested that it is also generated by the self-sustaining process at each scale (Lee & Moser Reference Lee and Moser2019; Kawata & Tsukahara Reference Kawata and Tsukahara2021). The TRIBZ25 simulation here shows that the positive turbulent transport in the region close to the wall is weakened significantly by the inhibition of the energy transfer from the
$\mathcal {P}_h$- to the
$\mathcal {P}_l$-subspace, favouring the proposition by Cho et al. (Reference Cho, Hwang and Choi2018) on its primary origin.
4.3.3. TRICZ25: the scattering mechanism in GQL approximations
The TRICZ25 case enables the contributions of the self-interacting nonlinear terms in both the
$\mathcal {P}_l$- and
$\mathcal {P}_h$-subspaces to the
$\mathcal {P}_h$-subspace group (previously eliminated by the GQL approximation), while it excludes the low/high wavenumber interaction terms in the
$\mathcal {P}_h$-subspace group, i.e. the ‘scattering mechanism’ in the GQL models (e.g. Tobias & Marston Reference Tobias and Marston2017). The statistics of TRICZ25 shows some subtle differences from that of GQLZ25 (see figure 9). In particular,
$u_{rms}^+$ (figure 9a) appears to be larger, while
$\langle u^{\prime } v^{\prime }\rangle _{x,z,t}^+$ depends on the
$y$-location: they are greater on the log layer and smaller below and above it.
The spectra of the TRICZ25 case show a more precise picture in the subtle differences (figures 10e, f, 11e, f, 12e, f and 13e, f). In particular, the spanwise wavenumber spectra of velocity and energy budget in figures 10(e, f) and 12(e, f) are found to extend over a narrower range of spanwise wavenumbers. Furthermore, the spectral intensity in the
$\mathcal {P}_h$-subspace (for
$\lambda _z<\lambda _{z,c}$) is weakened significantly, not covering wavenumbers below
$\lambda _z^+ \approx 100$. In particular, the spanwise wavenumber spectra of turbulence production are suppressed completely (figure 12e). This suggests that the low/high wavenumber interaction terms in the
$\mathcal {P}_h$-subspace group, incorporated in the GQL approximations, are probably the most crucial triadic interactions, because their suppression weakens significantly the energy-containing region of the spanwise wavenumber spectra of velocity.
It is interesting to note that this observation favours that the scattering mechanism is more important than the self-sustaining process for the generation of near-wall turbulence, when all the other scales are present in the flow – we note that all the previous numerical experiments demonstrating the self-sustaining process were without the motions larger than the given spanwise length scale (e.g. Jiménez & Moin Reference Jiménez and Moin1991; Hamilton et al. Reference Hamilton, Kim and Waleffe1995; Hwang Reference Hwang2013). However, care needs to be taken in understanding this behaviour, because the TRICZ25 case also suppresses all the production terms in (4.6) (i.e. the third terms in (4.6a) and (4.6b)), shared by the self-sustaining process (see (4.4)). In this respect, the issue on the relative importance between the self-sustaining process and the scattering mechanism in the generation of near-wall turbulence remains an open question, although the numerical experiment carried out with the TRIAZ25 case (e.g. see figures 10a,b) appears to support that perhaps the scattering mechanism may be more crucial than the self-sustaining process in the presence of all the scales. Since the existence of the self-sustaining processes and their roles have been well established (Jiménez & Moin Reference Jiménez and Moin1991; Hamilton et al. Reference Hamilton, Kim and Waleffe1995; Hwang Reference Hwang2013), a more careful examination with additional numerical experiments would certainly be required for the precise understanding on this new observation, and we will leave this for future investigation.
4.3.4. TRIDZ25: direct driving from the
$\mathcal {P}_l$-subspace to the
$\mathcal {P}_h$-subspace
Finally, the TRIDZ25 case incorporates the contribution of the self-interacting nonlinear term in the
$\mathcal {P}_l$-subspace to the
$\mathcal {P}_h$-subspace group, and its objective is to analyse the influence of this direct driving mechanism from the
$\mathcal {P}_l$-subspace to the
$\mathcal {P}_h$-subspace group. The statistics of TRIDZ25 is found to approximate that of GQLZ25 fairly well (figure 9), with a little overestimation of
$u_{rms}^+$,
$v_{rms}^+$ and
$\langle u^{\prime } v^{\prime }\rangle _{x,z,t}^+$, showing that the inclusion of the triadic interaction LL-H has a moderate effect on the statistics.
The spectra of the TRIDZ25 case offer a more detailed picture on its effect (figures 10g,h, 11g,h, 12g,h and 13g,h). The spanwise wavenumber velocity spectra in figures 10(g,h) appear to be similar qualitatively to those of the GQLZ25 case. However, the streamwise and wall-normal velocity spectra in the
$\mathcal {P}_h$-subspace (for
$\lambda _z<\lambda _{z,c}$) are intensified compared to those of the GQLZ25 case (figures 3g,h), especially with a large elevation of the energy in the relatively outer region. This can be explained by the spanwise wavenumber spectra of turbulent transport in figure 12(h), where a strong positive energy transport now appears in the outer region of the
$\mathcal {P}_h$-subspace. This is in contrast to that of the GQLZ25 case (figure 5h), where such a transport barely exists in the relative outer region of the
$\mathcal {P}_h$-subspace (i.e.
$y^+\gtrsim 300$).
In Part 1, it was proposed that the
$\mathcal {P}_h$-subspace group of a GQL model can admit the trivial solution if the cut-off wavelength for the decomposition of the flow (i.e.
$\lambda _{z,c}$) is sufficiently small (see § 4.2 in Hernández et al. Reference Hernández, Yang and Hwang2022). This was also explained in terms of the Lyapunov exponent of the
$\mathcal {P}_h$-subspace group, which can be interpreted alternatively as the activation process of the scattering mechanism proposed by Tobias & Marston (Reference Tobias and Marston2017). In the present study, such a case also appears for
$\lambda _{z,c}^+ \approx 97$ (GQLZ64 in table 1 and in the Appendix), the typical spanwise spacing of the near-wall streaks (Kline et al. Reference Kline, Reynolds, Schraub and Runstadler1967). In this case, the simulation result of the TRIDZ25 case suggests that the contribution of the self-interacting nonlinear term in the
$\mathcal {P}_l$-subspace to the
$\mathcal {P}_h$-subspace group becomes important when the scattering mechanism discussed in § 4.3.3 plays little role in generating the fluctuation in the
$\mathcal {P}_h$-subspace (i.e. when
$\lambda _{z,c}$ is sufficiently small). In such a case, the dynamics in the
$\mathcal {P}_h$ subspace remains essentially passive to the
$\mathcal {P}_l$-subspace in the sense that the
$\mathcal {P}_h$-subspace would admit a non-trivial solution only in the presence of the direct driving from the
$\mathcal {P}_l$-subspace. In this respect, finally, it is worth mentioning that the smallest spanwise computational domain retaining turbulence is
$L_z^+ \approx 100$ (Jiménez & Moin Reference Jiménez and Moin1991; Hwang Reference Hwang2013) and the critical spanwise wavelength, below which the GQL model in this study yields the trivial solution in the
$\mathcal {P}_h$-subspace, is
$\lambda _{z,c}^+ \approx 97$, showing an interesting analogue.
5. Concluding remarks
In the present study, a generalised quasilinear (GQL) approximation has been applied to turbulent channel flow at
$Re_\tau \approx 1700$, continuing the work of Part 1 (Hernández et al. Reference Hernández, Yang and Hwang2022). The flow is decomposed into two groups, the former of which contains a set of low-wavenumber spanwise Fourier modes, and the latter composed of the rest, high-wavenumber modes. The former group is then solved by considering the full nonlinear equations, while the latter group is obtained from the linearised equations about the former. In Part 1, the flow decomposition was based on the streamwise Fourier modes. By doing so, the corresponding GQL models placed a minimal model for a self-sustaining process at all integral length scales varying from viscous inner units to outer units, and the nonlinear coupling between the self-sustaining processes from different scales and the related nonlinear energy transfer were assessed. However, in this study, the flow decomposition gives the low wavenumber group in which the self-sustaining process is fully resolved at least within the given integral length scales, while the linearised high wavenumber group is not able to support the self-sustaining process. Some key conclusions are summarised as follows.
(i) Despite the important physical difference of the GQL approximation here from that in Part 1, they share some important similarities. First, due to the interference of nonlinear energy transfer given by GQL approximations, they tend to exhibit significantly reduced multi-scale behaviours, especially when the cut-off wavelength for the flow decomposition is sufficiently large. Furthermore, turbulent fluctuations have been found to be strongly anisotropic for large cut-off wavelengths, and this is due to damaged slow pressure originating from the fluctuation–fluctuation interactions, some of which are omitted in the GQL approximations.
(ii) While the high wavenumber group of the GQL models in the present study cannot support the self-sustaining process proposed e.g. by Hamilton et al. (Reference Hamilton, Kim and Waleffe1995) and Waleffe (Reference Waleffe1997), they can generate some key statistical features at integral scales through the scattering mechanism to a certain extent; in other words, a different mechanism from the self-sustaining process (i.e. the two-way interaction between ‘streamwise elongated’ structures of streamwise velocity (streaks) and their ‘streamwise wavy’ instability involving cross-streamwise velocities (waves and rolls)) acts to sustain turbulence in the GQL cases studied here. Although it is not clear exactly how this is possible, this observation could have some potential links to the success of the previous linearised Navier–Stokes operator-based models (e.g. Hwang & Cossu Reference Hwang and Cossu2010a; McKeon & Sharma Reference McKeon and Sharma2010; Moarref et al. Reference Moarref, Sharma, Tropp and McKeon2013; McKeon Reference McKeon2017; Zare et al. Reference Zare, Jovanović and Georgiou2017; Hwang & Eckhardt Reference Hwang and Eckhardt2020; Jovanović Reference Jovanović2021; Skouloudis & Hwang Reference Skouloudis and Hwang2021).
(iii) Finally, a set of numerical experiments has been performed to explore the roles of ignored triadic interaction terms in the GQL approximations focusing on scale interactions. These numerical experiments have clarified the role of: (1) self-interactions within the high-wavenumber group (TRIAZ25); (2) inverse energy transfer in the region close to the wall (TRIBZ25); (3) the scattering mechanism (TRICZ25); (4) a direct driving mechanism to the high-wavenumber group (TRIDZ25). The experiment TRIAZ25 has allowed the activation of the nonlinear terms associated with the self-sustaining process (HH-H triadic interaction) in the
$\mathcal {P}_h$-subspace (4.4). Its statistics and spectra resemble those of GQLZ25 the most. The TRIBZ25 case eliminates the contributions of the HH-L interaction (i.e. the inverse energy transfer from the
$\mathcal {P}_h$-subspace to the
$\mathcal {P}_l$-subspace), but activates all triadic interactions in the
$\mathcal {P}_h$-subspace group. The positive turbulent transport in the region close to the wall is shown to be weakened significantly by the inhibition of the energy transfer from the
$\mathcal {P}_h$-subspace to the
$\mathcal {P}_l$-subspace. The TRICZ25 case enables the LL-H and HH-H interactions (previously eliminated by the GQL approximation), while it excludes the LH-H interaction, i.e. the ‘scattering mechanism’ in the GQL models (e.g. Tobias & Marston Reference Tobias and Marston2017). The weakened spectral intensity of the production appears to suggests the significance of this mechanism, when all the other scales are present in the flow. Finally, the TRIDZ25 case incorporates the LL-H interaction. The strong positive energy transport found in the outer region of the
$\mathcal {P}_h$-subspace in contrast to that of the GQLZ25 case suggests that the contribution of this interaction becomes important, when the scattering mechanism plays little part in generating the fluctuation in the
$\mathcal {P}_h$-subspace (i.e. when
$\lambda _{z,c}$ is sufficiently small).
The GQL approximations were introduced originally to improve simpler QL models. Here, we have utilised the GQL approximations to explore the coupling dynamics between self-sustaining processes at different length scales, scale interactions and nonlinear energy transfer for wall-bounded turbulence. The knowledge gained here would be useful to develop a simpler and more robust self-consistent description of turbulence using the concepts of the ‘statistical-state dynamics’. Our ongoing efforts have been made along this direction, and some recent work can be found in Hwang & Eckhardt (Reference Hwang and Eckhardt2020) and Skouloudis & Hwang (Reference Skouloudis and Hwang2021).
Funding
C.G.H. and Y.H. acknowledge the support of the Leverhulme Trust (RPG-123-2019). Q.Y. acknowledges the Natural Science Foundation of China (11802322). Y.H. is also supported by the Engineering and Physical Sciences Research Council (EPSRC; EP/T009365/1) in the UK.
Declaration of interests
The authors report no conflict of interest.
Appendix. Velocity and energy-balance spectra of GQLZ64
Figure 14 shows the premultiplied spanwise wavenumber spectra of the streamwise velocity (figure 14a), the wall-normal velocity (figure 14b), the production (figure 14c) and the turbulent transport (figure 14d) of the GQLZ64 case. All spectra greatly resemble those of the reference LES in the
$\mathcal {P}_l$-subspace. The trivial solution
$\boldsymbol {u}_h=0$ is, however, observed in the
$\mathcal {P}_h$-subspace group, for a cut-off wavelength
$\lambda _{z,c}^+ \approx 97$. This phenomenon, investigated in Part 1 for the streamwise GQL models, has been explained in terms of the Lyapunov exponent and the instability of the linearised equations in the
$\mathcal {P}_h$-subspace group (Hernández et al. Reference Hernández, Yang and Hwang2022).
Premultiplied spanwise wavenumber spectra of (a) streamwise
$k_z^+\,\varPhi _{uu}^+(y^+,\lambda _z^+)$ and (b) wall-normal
$k_z^+\,\varPhi _{vv}^+(y^+,\lambda _z^+)$ velocity, (c) production
$k_z^+ y^+\,\hat {P}^+(y^+,\lambda _z^+)$, and (d) turbulent transport
$k_z^+ y^+\,\hat {T}_{turb}^+(y^+,\lambda _z^+)$, for GQLZ64. Here, the vertical line represents the spanwise cut-off wavelength (
$\lambda _{z,c}$) dividing the (a,c)
$\mathcal {P}_h$-subspace and (b,d)
$\mathcal {P}_l$-subspace regions.














































































