betaFeedback

WorksPhysics and Astronomy

Validation of symmetry-induced high moment velocity and temperature scaling laws in a turbulent channel flow

Abstract

The symmetry-based turbulence theory has been used to derive new scaling laws for the streamwise velocity and temperature moments of arbitrary order. For this, it has been applied to an incompressible turbulent channel flow driven by a pressure gradient with a passive scalar equation coupled in. To derive the scaling laws, symmetries of the classical Navier-Stokes and the thermal energy equations have been used together with statistical symmetries, i.e. the statistical scaling and translation symmetries of the multi-point moment equations. Specifically, the multi-point moments are built on the instantaneous velocity and temperature fields other than in the classical approach, where moments are based on the fluctuations of these fields. With this instantaneous approach, a linear system of multi-point correlation equations has been obtained, which greatly simplifies the symmetry analysis. The scaling laws have been derived in the limit of zero viscosity and heat conduction, i.e. Reτ→∞Re_τ\rightarrow \infty and Pr>1Pr > 1, and apply in the centre of the channel, i.e. they represent a generalization of the deficit law so herewith extending the work of done on the velocity field. The scaling laws are all power laws, with the exponent of the high moments all depending exclusively on those of the first and second moments. To validate the new scaling laws, the data from a large number of DNS for different Reynolds and Prandtl numbers have been used. The results show a very high accuracy of the scaling laws to represent the DNS data. The statistical scaling symmetry of the multi-point moment equations, which characterizes intermittency, has been the key to the new results since it generates a constant in the exponent of the final scaling law. Most important, since this constant is independent of the order of the moments, it clearly indicates anomalous scaling.

Introduction

The open problem in physics with most applications in daily life is probably the behaviour of turbulent flows. Different strategies have been proposed when dealing with predicting turbulence in engineering. From the different approaches known in Computational Fluid Dynamics (CFD), Direct Numerical Simulations (DNS) has proven to be a powerful tool to generate highly reliable data bases for theoretical concepts on the nature of turbulence. In a DNS, no empirical modelling is needed to account for turbulent effects, and the approximations of the solutions of the Navier-Stokes equations are obtained through highly accurate numerical schemes. The main problematic issue of DNSs is their high computational cost since even the smallest scales of turbulence, the Kolmogorov scales, have to be simulated. Hence, this limits DNSs to very simple canonical geometries. However, DNSs have the same validity as experiments, and almost any imaginable quantity can be computed.

It was not until the late 1980s that supercomputers could run the first DNS of turbulent flows. Specifically, in 1987, Kim et al.[2] conducted the first DNS of a turbulent channel flow where a pressure gradient drove the flow at a low Reynolds number. The first DNS of a thermal turbulent channel flow was also performed in 1987 by Kim et al.[3]. The flow was also driven by a pressure gradient at a friction Reynolds number of Reτ=180Re_\tau= 180. Therein and presently, ReτRe_\tau is defined as huτ/νhu_\tau/\nu, where hh is the semi-height of the channel, uτu_\tau is the friction velocity, and ν\nu is the kinematic viscosity of the fluid. The friction velocity is defined as uτ=τw/ρu_\tau= \sqrt{\tau_w/\rho}, where τw\tau_w is the averaged wall shear stress, and ρ\rho is the fluid density. Different Prandtl numbers were used, namely Pr=0.1Pr = 0.1, 0.71 and 2. Here the Prandtl number is defined as the ratio between the momentum diffusivity to the thermal diffusivity, Pr=ν/αPr = \nu/\alpha, where α\alpha is the thermal diffusivity of the fluid. One of the main results from the latter work was the validation of the DNS results comparing several first-order statistics with experimental data.

Since then, the aim of the DNS of thermal channel flows has been to increase the simulated Reynolds number, usually around Pr=0.71Pr = 0.71, which is the Prandtl number of the air. However, reaching higher Reynolds numbers has a computational cost which scales as Lx2LzReτ4Pr3/2L_x^2L_zRe_\tau^4Pr^{3/2}, according to Ref. 4. The largest DNS to date of a thermal channel flow[5] used a friction Reynolds number of 4000, and more recently, for isothermal turbulent channel flows, a friction Reynolds number of Reτ=10000Re_\tau= 10000 was reached in Ref. 6. However, these values are still far below the actual Reynolds numbers of most real-life problems. Therefore, Reynolds number extrapolations of the turbulent behaviour must be made, introducing inevitable errors and uncertainties. For example, the viscous diffusion and dissipation of the streamwise velocity fluctuations present a scaling failure near the wall[7]. An analogous scaling failure was obtained for the temperature variance at moderate Reynolds numbers. However, it was recently found[8] that for high Prandtl numbers and sufficiently high Reynolds numbers, the value of the viscous diffusion and dissipation of the temperature variance presented a much better scaling near the wall. Therefore, it is still an open question whether the streamwise velocity scaling failure will occur at much higher Reynolds numbers.

For all these reasons, turbulence is and will be, for many more years, an open problem without a complete analytical solution. Many researchers have proposed what are called turbulence scaling laws to describe the universal behaviour of turbulent statistics for special flows, though usually limited to the first and second moments. The most well-known scaling law is the universal law of the wall, which describes the profile of the mean streamwise velocity near the wall, consisting of a linear viscous sub-layer, where U‾+=y+\overline{U}^{+}=y^{+}, followed by the buffer layer and the logarithmic region further away from the wall. The overbar here denotes averaged in time, and the superscript ++ refers to dimensionless variables normalized in wall units with uτu_{\tau}, θτ\theta_{\tau} and ν\nu, where θτ=qw/(ρcpuτ)\theta_{\tau}=q_{w}/(\rho c_{p}u_{\tau}) is the friction temperature, and qwq_{w} and cpc_{p} are the normal heat flux to the walls and the specific heat at constant pressure, respectively. Analogously to the law of the wall for the velocity, one can approximate the mean temperature near the wall in a similar form, where, for the first sub-layer, called conductive sub-layer, Θ+=Pr y+\Theta^{+}=Pr\,y^{+}. In particular, the discovery of a logarithmic behaviour of the flow dates back to von Kármán in 1931. However, no connection between the analytical form of the scaling law and the Navier-Stokes equation was made. Despite the fact that a perfectly developed logarithmic region has not been observed in DNS due to the high Reynolds number needed to be simulated, a clear tendency suggests that it will appear for higher Reynolds numbers in different types of flows such as boundary layers, Couette flows, pipe flows,[9, 10, 11, 12, 13, 14, 15, 16] etc. A first derivation of the logarithmic behaviour of the flow, based on first principles, was presented in Ref. [17]. The use of Lie symmetries was the mathematical tool to achieve this.

Lie symmetries are a powerful mathematical theory to develop turbulent flow scaling laws. The origin of the Lie symmetries method dates back to the end of the 19th century when the mathematician Sophus Lie proposed it for obtaining solutions to differential equations and, most importantly, to systems of partial differential equations (PDE), such as the Navier-Stokes equations. The basis of the method consists of finding the symmetries of the system of PDE. Here, symmetry refers to a variable transformation that leads to an identical system of PDE, i.e. the transformed system of PDE has the same solution as the original one. With these symmetries, one can formulate a characteristic system (see §III for more details about the characteristic system), which in turn, leads to what is known as invariant solutions, which in turbulence are also known as scaling laws.

Lie symmetries possess several advantageous properties using ad hoc methods for a concrete application. First, symmetries can be obtained using computer algebra methods such as Maple. Second, symmetries give fundamental insight into the physics of the problem. And third, the scaling laws obtained are solutions to the moment equations and, hence, are based on first principles, not just pure curve fits. For these reasons, Lie symmetries are one of the most powerful tools for obtaining scaling laws of turbulent flows. Most important, it is also applicable to an infinite number of equations such as the moment hierarchy, and in this sense the ubiquitous closure problem of turbulence can be circumvented. This is also the approach that is presently applied.

The method has been widely studied by Oberlack and co-workers in several papers. Starting with Ref. 17, scaling laws for the three regions of wall-parallel shear flows (viscous sub-layer, logarithmic law, and deficit law in the centre of the channel) were obtained. Classical mechanical symmetries of the Navier-Stokes equations were used, but the key to the analysis was to employ the Multi-Point Correlation (MPC) equations. Two additional symmetries, not visible in the Navier-Stokes equations and called statistical symmetries, first discovered in Ref. 18, were used to derive the scaling laws that describe the flow statistics even for high moments. The next section will give more details about the MPC equations (§II). After this successful application of Lie symmetries to turbulent flows, several more works have been done with different geometries or boundary conditions[19, 20, 21, 22, 23, 24].

In this work, Lie symmetries theory will be used to derive new moment scaling laws of velocity and temperature, and mixed moments of arbitrary orders, extending the work in Ref. 1 to include the temperature. To achieve this, symmetries of the energy equation and the MPC equations of the energy and heat fluxes equations are obtained, from which the new scaling laws are formulated. These new scaling laws will be validated using the DNS data obtained by the authors in previous works (see[8, 25, 26]). We will restrict ourselves to moments in the streamwise direction. However, this shows the ability of our method to obtain scaling laws following only strong mathematical arguments.

In the next section, the governing equations of the problem are presented. In the third section, the Lie symmetries method is introduced, together with the application of the method to the governing equations of the problem. Then, in section four, the new scaling laws are developed and validated using DNS data. Finally, the fifth and last section contains conclusions.

Governing equations

The equations that describe the behaviour of a turbulent flow, considering a Newtonian fluid with constant density and viscosity, are the well-known Navier-Stokes equations. For the sake of readability, the temporal and spatial dependencies will be omitted if uniqueness allows doing so. In the most general form, these equations can be written as

C(x)=∂Uk∂xk=0,(1)\mathcal{C}(x) = \frac{\partial U_k}{\partial x_k} = 0, \tag*{(1)}
Mi(x)=∂Ui∂t+Uk∂Ui∂xk+∂P∂xi−ν∂2Ui∂xk∂xk=0,(2)\mathcal{M}_i(x) = \frac{\partial U_i}{\partial t} + U_k\frac{\partial U_i}{\partial x_k} + \frac{\partial P}{\partial x_i} - \nu\frac{\partial^2 U_i}{\partial x_k\partial x_k} = 0, \tag*{(2)}

where t∈R+t \in\mathbb{R}^{+} is time; xix_i and UiU_i are the space coordinate and velocity, i=1,2,3i = 1,2,3; and PP is the pressure divided by the density. The no-slip boundary condition is applied to both walls, periodic boundary conditions are used in the x1x_1 and x3x_3 directions and, to propel the flow, a constant pressure gradient is introduced in the x1x_1 direction so that the mass flux remains constant. The x2x_2 coordinate points in wall normal direction. Additionally, the thermal energy equation is simulated, which for a constant thermal conductivity coefficient, α\alpha, reads

E(x)=∂Θ∂t+Uk∂Θ∂xk−α∂2Θ∂xk∂xk=0,(3)\mathcal{E}(x) = \frac{\partial\Theta}{\partial t} + U_k\frac{\partial\Theta}{\partial x_k} - \alpha\frac{\partial^2 \Theta}{\partial x_k\partial x_k} = 0, \tag*{(3)}

where Θ\Theta is the temperature. It should be noted that a constant heat flux at the wall was assumed in the simulation of the energy equation (3) since only this implies the temperature-scaling laws. This thermal boundary condition is known as the uniform heat flux (UHF) boundary condition. Curious reader is referred to Ref. 8, 25–28 for more information about the UHF boundary condition. This is similar to the constant wall shear stress, which is central to the velocity-scaling laws. With this, since no heat sink was introduced, a constant temperature gradient in the x1x_1 direction is generated. This is removed from the flow by the following transformation to guarantee homogeneity in x1x_1-direction

Θ=⟨Θw⟩x3−Θtr,(4)\Theta= \langle\Theta_w \rangle_{x_3} - \Theta_{tr}, \tag*{(4)}

where ⟨Θw⟩x3\langle\Theta_w \rangle_{x_3} is the temperature at the wall averaged in time and in the x3x_3 direction, and Θtr\Theta_{tr} is the transformed temperature. Therefore, ⟨Θw⟩x3\langle\Theta_w \rangle_{x_3} carries the linear increment of the temperature and only depends on the x1x_1 direction. Then, Θtr\Theta_{\mathrm{tr}} is homogeneous in this streamwise direction. The somewhat unusual choice of the sign in (4) is due to the fact that the temperature at the wall is maximum and therefore the transformed temperature Θtr\Theta_{\mathrm{tr}} remains positive throughout.

This allows the use of spectral discretization in the x1x_1 direction. Obviously, the scaling of Θ\Theta and Θtr\Theta_{\mathrm{tr}} in the x2x_2 direction will be the same since only a constant value will differ among them. Furthermore, since the scaling laws are presented as defect laws, for both Θ\Theta and Θtr\Theta_{\mathrm{tr}}, these scaling laws must be the same. For the sake of generality, and without loss of veracity, we will refer to scaling laws of temperature, Θ\Theta, instead of transformed temperature.

Using the Reynolds decomposition, one can separate the instantaneous variables (capital letter) in an average part (capital letter and over-bar) that does not depend on time, and a fluctuation part (lower case), e.g., Ui(x,t)=U‾i(x)+ui(t,x)U_i(x,t)=\overline{U}_i(x)+u_i(t,x) (note that temporal and spatial dependencies have been recovered only to show this example). Therefore, the following properties can be applied: the average in time of a mean quantity, Φ‾\overline{\Phi}, will remain unchanged, i.e., Φ‾‾=Φ‾\overline{\overline{\Phi}}=\overline{\Phi}; and the average in time of a fluctuation quantity is 0, i.e., ϕ‾=0\overline{\phi}=0. In addition, the following simplifications are valid for a developed turbulent channel flow driven by a pressure gradient,

U‾1=U‾1(x2),P‾=P‾(x1,x2),Θ‾=Θ‾(x2),\overline{U}_1=\overline{U}_1(x_2),\qquad\overline{P}=\overline{P}(x_1,x_2),\qquad\overline{\Theta}=\overline{\Theta}(x_2),
U‾2=U‾3=0,uiuj‾=uiuj‾(x2),uiθ‾=uiθ‾(x2).(5)\overline{U}_2=\overline{U}_3=0,\qquad\overline{u_i u_j}=\overline{u_i u_j}(x_2),\qquad\overline{u_i\theta}=\overline{u_i\theta}(x_2). \tag*{(5)}

Introducing the latter into equations (2) and (3) and, in turn, taking the average, the governing equations reduce to

du1u2‾dx2+∂P‾∂x1−νd2U‾1dx22=0,(6)\frac{\mathrm{d}\overline{u_1u_2}}{\mathrm{d}x_2}+\frac{\partial\overline{P}}{\partial x_1}-\nu\frac{\mathrm{d}^2\overline{U}_1}{\mathrm{d}x_2^2}=0, \tag*{(6)}
du2u2‾dx2+∂P‾∂x2=0,(7)\frac{\mathrm{d}\overline{u_2u_2}}{\mathrm{d}x_2}+\frac{\partial\overline{P}}{\partial x_2}=0, \tag*{(7)}
du3u2‾dx2=0,(8)\frac{\mathrm{d}\overline{u_3u_2}}{\mathrm{d}x_2}=0, \tag*{(8)}
dθu2‾dx2−αd2Θ‾dx22=0.(9)\frac{\mathrm{d}\overline{\theta u_2}}{\mathrm{d}x_2}-\alpha\frac{\mathrm{d}^2\overline{\Theta}}{\mathrm{d}x_2^2}=0. \tag*{(9)}

Besides the latter one-point quantities, we may define the two-point correlation functions, or two-point moments, based on the fluctuating velocity,

Rij(x,r)=ui(x)uj(x+r)‾,Rij0(x)=lim⁡r→0Rij(x,r)=ui(x)uj(x)‾.(10)R_{ij}(x,r)=\overline{u_i(x)u_j(x+r)},\qquad R_{ij}^{0}(x)=\lim_{r\to0}R_{ij}(x,r)=\overline{u_i(x)u_j(x)}. \tag*{(10)}

Employing an equivalent definition based on the instantaneous variables reads

Hij(x,r)=Ui(x)Uj(x+r)‾,Hij0(x)=lim⁡r→0Hij(x,r)=Ui(x)Uj(x)‾.(11)H_{ij}(x,r)=\overline{U_i(x)U_j(x+r)},\qquad H_{ij}^{0}(x)=\lim_{r\to0}H_{ij}(x,r)=\overline{U_i(x)U_j(x)}. \tag*{(11)}

and a relation between the two correlation functions RijR_{ij} and HijH_{ij} reads as follows

Rij(x,r)=Hij(x,r)−U‾i(x)U‾j(x+r).(12)R_{ij}(x,r)=H_{ij}(x,r)-\overline{U}_i(x)\overline{U}_j(x+r). \tag*{(12)}

This two-point concept can be extended for any number of points and ultimately forms the basis of the following analysis as well as the resulting scaling laws. Hence, we introduce the MPC equations (see e.g. Ref. 21–24). For high-order moments of velocity and temperature, they give additional information that is not provided in the one-point statistic equations, such as length scales. Also, when deriving a higher-order moment equation, only one unclosed function arises. As observed in equation (10), from the two-point statistics one can obtain every one-point statistic. Finally, regarding Lie symmetries, two extra symmetries are obtained from the MPC equations, which are the key for determining the new scaling laws of the high-order moments, and will be pointed out in section §IV.

Equations (10) and (11) are the basis of the two different approaches that can be used to obtain the MPC equations: the fluctuating approach or the instantaneous approach. On one hand, the fluctuating approach has some advantages such as a straightforward relation to the Reynolds stress tensor or the turbulent heat fluxes. However, as noted in Ref. 18, a non-linear system of equations is obtained. Furthermore, all moment equations are coupled to the mean velocity or temperature, and equations of the third moment or higher, are coupled to the second moment. All this complicates the symmetry analysis that will be done below. On the other hand, the instantaneous approach results in a linear system of MPC equations with an equivalent but much simpler symmetry analysis. For this reason, the instantaneous approach is the one used in this work. It should be noted that the fluctuating approach and the instantaneous approach are bijective, i.e. mathematically physically absolutely equivalent.

Before presenting the MPC equations, some notations must be clarified. The correlation functions for the velocity are defined as

Hi{n}=Hi(1)i(2)…i(n)=U‾i(1)(x(1))U‾i(2)(x(2))…U‾i(n)(x(n)),(13)H_{i_{\{n\}}}=H_{i_{(1)}i_{(2)}\ldots i_{(n)}}=\overline{U}_{i_{(1)}}(x_{(1)})\overline{U}_{i_{(2)}}(x_{(2)})\ldots\overline{U}_{i_{(n)}}(x_{(n)}), \tag*{(13)}

which for n=2n=2, x(1)=xx_{(1)}=x and x(2)=x+rx_{(2)}=x+r yields to (11). Note the i(n)i_{(n)} in U‾i(n)(x(n))\overline{U}_{i_{(n)}}(x_{(n)}), the subscript i(n)i_{(n)} refers to the velocity direction of the nn-th term which is measured at the coordinate x(n)x_{(n)}. The definition of the temperature correlation is

HΘ{m}=HΘ(1)Θ(2)…Θ(m)=Θ‾(x(1))Θ‾(x(2))…Θ‾(x(m)).(14)H_{\Theta_{\{m\}}}=H_{\Theta_{(1)}\Theta_{(2)}\ldots\Theta_{(m)}}=\overline{\Theta}(x_{(1)})\overline{\Theta}(x_{(2)})\ldots\overline{\Theta}(x_{(m)}). \tag*{(14)}

Mixed moments of velocity and temperature, which in the limit of only one temperature and one velocity reduces to the turbulent heat flux, reads

Hi{n}Θ{m}=Hi(1)i(2)⋯i(n)Θ(n+1)Θ(n+2)⋯Θ(n+m)=Ui(1)(x(1))Ui(2)(x(2))⋯Ui(n)(x(n))Θ(x(n+1))Θ(x(n+2))⋯Θ(x(n+m))‾(15)H_{i_{\{n\}}\Theta_{\{m\}}}=H_{i_{(1)}i_{(2)}\cdots i_{(n)}\Theta_{(n+1)}\Theta_{(n+2)}\cdots\Theta_{(n+m)}}= \frac{\overline{U_{i_{(1)}}(\boldsymbol{x}_{(1)})U_{i_{(2)}}(\boldsymbol{x}_{(2)})\cdots U_{i_{(n)}}(\boldsymbol{x}_{(n)})\Theta(\boldsymbol{x}_{(n+1)})\Theta(\boldsymbol{x}_{(n+2)})\cdots\Theta(\boldsymbol{x}_{(n+m)})}}{} \tag*{(15)}

Note that (13) and (14) are just particular cases of (15) for mm and nn equal to 0, respectively, but for the sake of readability, they are presented separately. When pressure is involved in the correlation, the notation, in the general form, is

Ii{n−1}Θ{m}[l]P=Hi(1)⋯i(l−1)Pi(l+1)⋯i(n)Θ(n+1)Θ(n+2)⋯Θ(n+m)=Ui(1)(x(1))⋯P(x(l))⋯Ui(n)(x(n))Θ(x(n+1))⋯Θ(x(n+m))‾(16)I_{i_{\{n-1\}}\Theta_{\{m\}}[l]P}=H_{i_{(1)}\cdots i_{(l-1)}P i_{(l+1)}\cdots i_{(n)}\Theta_{(n+1)}\Theta_{(n+2)}\cdots\Theta_{(n+m)}}= \frac{\overline{U_{i_{(1)}}(\boldsymbol{x}_{(1)})\cdots P(\boldsymbol{x}_{(l)})\cdots U_{i_{(n)}}(\boldsymbol{x}_{(n)})\Theta(\boldsymbol{x}_{(n+1)})\cdots\Theta(\boldsymbol{x}_{(n+m)})}}{} \tag*{(16)}

for 1≤l≤n1 \le l \le n. Finally, the following notation

Hi{n}Θ{m}[i(l)→k](x(l)→x(p))=Ui(1)(x(1))⋯Ui(l−1)(x(l−1))Uk(x(p))Ui(l+1)(x(l+1))⋯Ui(n)(x(n))Θ(x(n+1))⋯Θ(x(n+m))‾(17)H_{i_{\{n\}}\Theta_{\{m\}}^{[i_{(l)}\to k]}(\boldsymbol{x}_{(l)}\to\boldsymbol{x}_{(p)})= \overline{U_{i_{(1)}}(\boldsymbol{x}_{(1)})\cdots U_{i_{(l-1)}}(\boldsymbol{x}_{(l-1)})U_k(\boldsymbol{x}_{(p)})U_{i_{(l+1)}}(\boldsymbol{x}_{(l+1)})\cdots U_{i_{(n)}}(\boldsymbol{x}_{(n)})\Theta(\boldsymbol{x}_{(n+1)})\cdots\Theta(\boldsymbol{x}_{(n+m)})}} \tag*{(17)}

is used to indicate a change in the correlation function of velocity direction, i(l)i_{(l)}, to kk and/or the coordinate where the variable is applied, x(l)\boldsymbol{x}_{(l)}, to x(p)\boldsymbol{x}_{(p)}. With these definitions, the MPC equations of the heat flux moments of order n+mn+m reads (see Appendix A for detailed step-by-step derivation of the MPC equations)

∂Hi{n}Θ{m}∂t+∑a=1n(∂Hi{n+1}Θ{m}[i(n+m+1)→k](x(n+m+1)→x(a))∂xk(a)+∂Ii{n−1}Θ{m}[a]P∂xi(a)−ν∂2Hi{n}Θ{m}∂xk(a)∂xk(a))+∑b=n+1n+m(∂Hi{n+1}Θ{m}[i(n+m+1)→k](x(n+m+1)→x(b))∂xk(b)−α∂2Hi{n}Θ{m}∂xk(b)∂xk(b))=0.(18)\frac{\partial H_{i_{\{n\}}\Theta_{\{m\}}}}{\partial t} +\sum_{a=1}^{n}\left( \frac{\partial H_{i_{\{n+1\}}\Theta_{\{m\}}}^{[i_{(n+m+1)}\to k]}(\boldsymbol{x}_{(n+m+1)}\to\boldsymbol{x}_{(a)})}{\partial x_{k_{(a)}}} +\frac{\partial I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}}{\partial x_{i_{(a)}}} -\nu\frac{\partial^2 H_{i_{\{n\}}\Theta_{\{m\}}}}{\partial x_{k_{(a)}}\partial x_{k_{(a)}}} \right) +\sum_{b=n+1}^{n+m}\left( \frac{\partial H_{i_{\{n+1\}}\Theta_{\{m\}}}^{[i_{(n+m+1)}\to k]}(\boldsymbol{x}_{(n+m+1)}\to\boldsymbol{x}_{(b)})}{\partial x_{k_{(b)}}} -\alpha\frac{\partial^2 H_{i_{\{n\}}\Theta_{\{m\}}}}{\partial x_{k_{(b)}}\partial x_{k_{(b)}}} \right)=0. \tag*{(18)}

As was mentioned before in (15), the MPC equations of the velocity and temperature moments are specific cases of (18), which can be obtained by setting, respectively, mm and nn equals to 0. Additionally, the continuity equations read

∂Hi{n}Θ{m}[i(l)→k]∂xk(l)=0forl=1,2,…,n,(19)\frac{\partial H_{i_{\{n\}}\Theta_{\{m\}}}^{[i_{(l)}\to k]}}{\partial x_{k_{(l)}}}=0\quad\text{for}\quad l=1,2,\ldots,n, \tag*{(19)}
∂Ii{n−1}Θ{m}[a]P[i(l)→k]∂xk(l)=0fora,l=1,2,…,n,a≠l,andn≥2.(20)\frac{\partial I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P[i_{(l)}\to k]}}{\partial x_{k_{(l)}}}=0\quad\text{for}\quad a,l=1,2,\ldots,n,\quad a\ne l,\quad\text{and}\quad n\ge2. \tag*{(20)}

Note that pure temperature correlations and heat fluxes correlations with l>nl > n do not admit continuity equations since they would have originated from ∂Θ‾(x)/∂xk\partial\overline{\Theta}(\boldsymbol{x})/\partial x_k, which is not a continuity equation.

As was previously noted, the system of the MPC equations (18), (19), and (20) is linear for any turbulent flow. Moreover, the dependent variables HH and II appear inside spatial or temporal derivatives. As seen in sections §III C and §IV, this is the key to obtaining two important statistical Lie symmetries necessary to derive the scaling laws.

To make the notation easier to understand, the Two-Point Correlation (TPC) equations for the velocity, heat fluxes, and temperature are presented in Appendix B.

Lie symmetries of the MPC equations

In this section, the Lie symmetries method will be presented briefly. After that, the symmetries of the governing equations introduced in the previous section will be given.

Symmetry transformations

Given a system of partial differential equations (PDE)

F(x,y,y(1),y(2),…)=0,(21)F(\boldsymbol{x},\boldsymbol{y},\boldsymbol{y}^{(1)},\boldsymbol{y}^{(2)},\ldots)=0, \tag*{(21)}

where x\boldsymbol{x} are the independent variables, y\boldsymbol{y} are the dependent variables, and y(n)\boldsymbol{y}^{(n)} are the nn-th derivative of the dependent variables with respect to all coordinate combinations of x\boldsymbol{x}. Based on this, a transformation of (21), with the form

x∗=ϕ(x,y),y∗=ψ(x,y),(22)\boldsymbol{x}^{*}=\phi(\boldsymbol{x},\boldsymbol{y}),\quad\boldsymbol{y}^{*}=\psi(\boldsymbol{x},\boldsymbol{y}), \tag*{(22)}

is called a symmetry transformation, or just symmetry, if the following holds:

F(x,y,y(1),y(2),…)=0 ⟺ F(x∗,y∗,y∗(1),y∗(2),…)=0.(23)F(\boldsymbol{x},\boldsymbol{y},\boldsymbol{y}^{(1)},\boldsymbol{y}^{(2)},\ldots)=0 \ \Longleftrightarrow\ F(\boldsymbol{x}^{*},\boldsymbol{y}^{*},\boldsymbol{y}^{*(1)},\boldsymbol{y}^{*(2)},\ldots)=0. \tag*{(23)}

In other words, a symmetry transformation (22) leaves the PDEs (21) invariant and, in addition, maps any solution of (21) into a new solution.

In Lie group analysis, it is the aim to find all possible symmetry transformations of the PDE (21). The notation group refers to the fact that symmetry transformations usually admit group properties. As we are presently dealing with Lie symmetry groups, the group parameter ε∈R\varepsilon\in\mathbb{R} has to be introduced to obtain the so-called one-parameter Lie symmetry group of transformations, with the form

x∗=ϕ(x,y;ε),y∗=ψ(x,y;ε).(24)x^* = \phi(x,y;\varepsilon), \qquad y^* = \psi(x,y;\varepsilon). \tag*{(24)}

Equation (24) provides a continuous group of transformations that allows analytical solutions for the underlying equations. As for the group properties of (24), we may, without loss of generality, assign ε=0\varepsilon= 0 to the identity element, i.e.

x∗=ϕ(x,y;ε=0)=x,y∗=ψ(x,y;ε=0)=y.(25)x^* = \phi(x,y;\varepsilon= 0) = x, \qquad y^* = \psi(x,y;\varepsilon= 0) = y. \tag*{(25)}

Therefore, if a Taylor series at ε=0\varepsilon= 0 is applied to the Lie group of transformation (24), we obtain

x∗=x+∂ϕ(x,y;ε)∂ε∣ε=0ε+O(ε2)=x+ξ(x,y)ε+O(ε2),(26)x^* = x + \left.\frac{\partial\phi(x,y;\varepsilon)}{\partial\varepsilon}\right|_{\varepsilon=0}\varepsilon+ O(\varepsilon^2) = x + \xi(x,y)\varepsilon+ O(\varepsilon^2), \tag*{(26)}
y∗=y+∂ψ(x,y;ε)∂ε∣ε=0ε+O(ε2)=y+η(x,y)ε+O(ε2).(27)y^* = y + \left.\frac{\partial\psi(x,y;\varepsilon)}{\partial\varepsilon}\right|_{\varepsilon=0}\varepsilon+ O(\varepsilon^2) = y + \eta(x,y)\varepsilon+ O(\varepsilon^2). \tag*{(27)}

Equations (26) and (27) are the infinitesimal form of the Lie group of transformation (24), where ξ\xi and η\eta are the so-called infinitesimals. The general form of the transformation (24) and the infinitesimal transformations (26) and (27) are related by Lie’s first theorem (see Ref. 29), i.e. if the infinitesimals of the transformation, ξ\xi and η\eta are known, one can uniquely recover the general form of the symmetry group of transformations (24). In order to obtain all symmetries of a given system of PDEs, one has to invoke the infinitesimal form of the transformations (26) and (27). Based on this, an algebraic algorithm, the Lie algorithm, evolves, which has been implemented into various computer algebra systems. Details on the algorithm may be taken from different textbooks like Ref. 29 or in works such as Ref. 20.

Group invariant solutions

In addition to the fact that symmetries characterize fundamental physical properties of a system, it is the ability to construct solutions that are central to their application, and which will also be applied here. For this purpose, we define the so-called group-invariant solutions, i.e. once the Lie symmetries of the system of PDE are obtained, the next step is to generate invariant solutions, which in turbulence are referred to as scaling laws. We call y=Ψ(x)y = \Psi(x) an invariant solution of a PDE system if and only if

  1. y−Ψ(x)y-\Psi(x) is invariant under XX, where XX is the so-called infinitesimal generator, defined as

X=ξi(x,y)∂∂xi+ηj(x,y)∂∂yj.(28)X=\xi_i(x,y)\frac{\partial}{\partial x_i}+\eta_j(x,y)\frac{\partial}{\partial y_j}. \tag*{(28)}

Hence, we have

X(y−Ψ(x))=0,(29)X(y-\Psi(x))=0, \tag*{(29)}

on y=Ψ(x)y=\Psi(x). Using the operator (28) and differentiating out, we obtain the following hyperbolic system

ξi(x,Ψ)∂Ψj∂ξi=ηj(x,Ψ),i=1,…,k;j=1,…,l,(30)\xi_i(x,\Psi)\frac{\partial\Psi_j}{\partial\xi_i}=\eta_j(x,\Psi),\qquad i=1,\ldots,k;\qquad j=1,\ldots,l, \tag*{(30)}

which generates the invariant solution. The solution of the hyperbolic system (30) can now be determined by the method of characteristics and we obtain the so-called invariant surface condition

dx1ξ1=dx2ξ2=⋯=dxkξk=dy1η1=dy2η2=⋯=dylηl,(31)\frac{dx_1}{\xi_1}=\frac{dx_2}{\xi_2}=\cdots=\frac{dx_k}{\xi_k}=\frac{dy_1}{\eta_1}=\frac{dy_2}{\eta_2}=\cdots=\frac{dy_l}{\eta_l}, \tag*{(31)}

where kk and ll are the respective numbers of the independent and dependent variables.

The integrals of the system (31) are the characteristics of the hyperbolic system (30) but, at the same time, the invariants of the original system of equations. These now form the basis of the invariant solutions and, thus, the new independent and dependent variables - the similarity variables.

  1. Finally, the invariant solution, y=Ψ(x)y=\Psi(x) has to solve the PDE system, which is to be verified by insertion into the original PDE.

Symmetries of the governing equations

In this section, we present the Lie symmetries of the governing equations (1)-(3). For high Reynolds numbers flows, i.e., for Reτ→∞Re_\tau\to\infty, and sufficiently away from the wall, the viscous effects are limited to length-scales of the order of the Kolmogorov length scale. In Oberlack[30], this fact forms the basis for a singular asymptotic expansion similar to boundary layer theory. As a result, two sets of moment equations arise, where the equations for the "outer" solutions is frictionless and acts on length scales larger than the Kolmogorov scale, while an "inner" equation. contains friction terms and operates on the Kolmogorov length. As a result, the frictionless "outer" equation and the corresponding solutions have the symmetries of the Euler equation. The above development again illustrates the fact that although the limit ν→0+\nu\to0^{+} can be considered, this is not identical to ν=0\nu= 0. For the following analyses, this means a focus on the large scales and thus that ν=α=0\nu= \alpha= 0 may be set in equation (18), assuming ν∼α\nu\sim\alpha, i.e. the diffusion terms are of the same order of magnitude. Incidentally, the dissipation therefore results from the "inner" equation, which is not considered presently.

For the case of the Euler equations, a 10-parameter symmetry group of transformation is obtained, where we here only present the scaling groups needed further below,

TSx:t∗=t,x∗=eaSxx,U∗=eaSxU,P∗=e2aSxP,(32)T_{S_x}: \quad t^{*} = t,\quad\boldsymbol{x}^{*} = e^{a_{S_x}}\boldsymbol{x},\quad U^{*} = e^{a_{S_x}}U,\quad P^{*} = e^{2a_{S_x}}P, \tag*{(32)}
TSt:t∗=eaStt,x∗=x,U∗=e−aStU,P∗=e−2aStP.(33)T_{S_t}: \quad t^{*} = e^{a_{S_t}}t,\quad\boldsymbol{x}^{*} = \boldsymbol{x},\quad U^{*} = e^{-a_{S_t}}U,\quad P^{*} = e^{-2a_{S_t}}P. \tag*{(33)}

The coefficients aSxa_{S_x} and aSta_{S_t} are the group parameters of scaling of space and time, respectively.

If the Navier-Stokes equations are considered, i.e., the viscous term is not neglected, the two scaling symmetries, TSxT_{S_x} and TStT_{S_t}, linearly combine into a simple scaling symmetry. This phenomenon, in which a multi-parameter symmetry group of transformations is reduced after a specific condition is applied, is known as symmetry breaking.

An analogous simplification, as the transition from the Navier-Stokes to the Euler equation, applied to the energy equation (3), can be done by neglecting the diffusive term, i.e., Peτ→∞Pe_{\tau} \to\infty, which holds in the center of the channel. Considering this, the energy equation admits the following infinite-dimensional symmetry,

TΘ:t∗=t,x∗=x,U∗=U,P∗=P,Θ∗=f(Θ).(34)T_{\Theta}: \quad t^{*} = t,\quad\boldsymbol{x}^{*} = \boldsymbol{x},\quad U^{*} = U,\quad P^{*} = P,\quad\Theta^{*} = f(\Theta). \tag*{(34)}

For scaling purposes, and in analogy with the scaling symmetries of the Euler equations, we consider the simplification f(Θ)=eaΘΘf(\Theta) = e^{a_{\Theta}}\Theta, so that TΘT_{\Theta} represents a scaling of temperature. It should be noted that the energy equation (3), just like the Navier-Stokes equations (1) and (2), admits the Galilean group.

As noted in Ref. 18, the symmetries obtained for the Navier-Stokes and energy equations transfer to the MPC equations (18). So, in the limit of zero viscosity and diffusion, i.e. Reτ→∞Re_{\tau} \to\infty and

Pr>1Pr > 1, the MPC equations (18) admit the following scaling symmetries:

\begin{aligned} \overline{T}_{Sx}: \quad&t^{*}=t,\quad\boldsymbol{x}^{*}=e^{a_{Sx}}\boldsymbol{x},\quad H_{i_{\{n\}}\Theta_{\{m\}}}^{*}=e^{na_{Sx}}H_{i_{\{n\}}\Theta_{\{m\}}},\\ &I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}^{*}=e^{(n+1)a_{Sx}}I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P. \end{aligned}} \tag*{(35)}\end{aligned}
\begin{aligned} \overline{T}_{St}: \quad&t^{*}=e^{a_{St}}t,\quad\boldsymbol{x}^{*}=\boldsymbol{x},\quad H_{i_{\{n\}}\Theta_{\{m\}}}^{*}=e^{-na_{St}}H_{i_{\{n\}}\Theta_{\{m\}}},\\ &I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}^{*}=e^{-(n+1)a_{St}}I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P. \end{aligned}} \tag*{(36)}\end{aligned}
\begin{aligned} \overline{T}_{S\Theta}: \quad&t^{*}=t,\quad\boldsymbol{x}^{*}=\boldsymbol{x},\quad H_{i_{\{n\}}\Theta_{\{m\}}}^{*}=e^{ma_{\Theta}}H_{i_{\{n\}}\Theta_{\{m\}}},\\ &I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}^{*}=e^{ma_{\Theta}}I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P. \end{aligned}} \tag*{(37)}\end{aligned}

which are immediate consequences of (32), (33) and the scaling version of (34).

In addition to the symmetries induced from the Navier-Stokes/Euler and energy equations, the MPC equations (18) admit an extended set of symmetry transformations. These symmetries are called statistical symmetries and they are the key in the process of deriving scaling laws[21] for high-order moments of the velocity and temperature. These symmetries were discovered in Ref. 18 and detailed information on the physical meaning of the statistical symmetries can be found in Ref. 31.

First, because of the linearity of the MPC equations (18), a scaling symmetry of the dependent variables is admitted

\begin{aligned} \overline{T}_{Ss}: \quad&t^{*}=t,\quad\boldsymbol{x}^{*}=\boldsymbol{x},\quad H_{i_{\{n\}}\Theta_{\{m\}}}^{*}=e^{a_{Ss}}H_{i_{\{n\}}\Theta_{\{m\}}},\\ &I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}^{*}=e^{a_{Ss}}I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P. \end{aligned}} \tag*{(38)}\end{aligned}

This symmetry, as proven in Ref. 31, represents a measure of intermittency. For intermittency, we understand a flow with subsequently changing turbulent and non-turbulent regimes. Moreover, as all dependent variables in (18) appear inside derivatives, a translation symmetry of all moments is also admitted

T‾tra,H:t∗=t,x∗=x,Hi{n}Θ{m}∗=Hi{n}Θ{m}+ai{n}Θ{m}H,Ii{n−1}Θ{m}[a]P∗=Ii{n−1}Θ{m}[a]P+ai{n−1}Θ{m}I.(39)\begin{aligned} \overline{T}_{tra,H}: \quad&t^{*}=t,\quad\boldsymbol{x}^{*}=\boldsymbol{x},\quad H_{i_{\{n\}}\Theta_{\{m\}}}^{*}=H_{i_{\{n\}}\Theta_{\{m\}}}+a^{H}_{i_{\{n\}}\Theta_{\{m\}}},\\ &I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}^{*}=I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}+a^{I}_{i_{\{n-1\}}\Theta_{\{m\}}}. \tag*{(39)} \end{aligned}

Apart from the symmetries presented, we will also include the classical translation in space symmetry, i.e.

\begin{aligned} \overline{T}_{tra,x}: \quad&t^{*}=t,\quad\boldsymbol{x}^{*}=\boldsymbol{x}+\boldsymbol{a}_{x},\quad H_{i_{\{n\}}\Theta_{\{m\}}}^{*}=H_{i_{\{n\}}\Theta_{\{m\}}},\\ &I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}^{*}=I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P. \end{aligned}} \tag*{(40)}\end{aligned}

Note that, in contrast to (38), where aSSa_{SS} is a single group parameter, symmetries (39) and (40) are a condensed way of showing several symmetries. Each component of the vector and tensors a{n}{m}Ha^{H}_{\{n\}\{m\}}, a{n−1}{m}Ia^{I}_{\{n-1\}\{m\}} and ax\boldsymbol{a}_x represent the group parameter of different and independent symmetries. Therefore, infinite symmetries are contained in (39), while (40) contains three symmetries, one for each spatial direction.

In summary, six symmetries have been identified, that will be used to derive the scaling laws of high-order moments of velocity and temperature. One property of the Lie symmetries is that one can combine different one-parameter Lie symmetries into a multi-parameter Lie symmetry. Following this, one can obtain the following multi-parameter Lie symmetry group from the symmetries (35)-(40)

T:t∗=eaStt,x∗=eaSxx+ax,Hi{n}⊗{m}∗=en(aSx−aSt)+maΘ+aSsHi{n}⊗{m}H+ai{n}⊗{m},HIi{n−1}⊗{m}[a]P∗=e(n+1)(aSx−aSt)+maΘ+aSsIi{n−1}⊗{m}[a]P+ai{n−1}⊗{m}I.(41)\begin{aligned} T:\quad&t^*=e^{a_{St}}t,\quad\boldsymbol{x}^*=e^{a_{Sx}}\boldsymbol{x}+\boldsymbol{a}_x,\\ &H^{*}_{i_{\{n\}\otimes\{m\}}}=e^{n(a_{Sx}-a_{St})+ma_{\Theta}+a_{Ss}}H^{H}_{i_{\{n\}\otimes\{m\}}}+a^{H}_{i_{\{n\}\otimes\{m\}},}\\ &I^{*}_{i_{\{n-1\}\otimes\{m\}}[a]P}=e^{(n+1)(a_{Sx}-a_{St})+ma_{\Theta}+a_{Ss}}I_{i_{\{n-1\}\otimes\{m\}}[a]P}+a^{I}_{i_{\{n-1\}\otimes\{m\}}}. \tag*{(41)} \end{aligned}

A different way of writing the symmetry group (41) is using the infinitesimal notation (26) and (27), from which one obtains

ξt=aStt,ξx=aSxx+ax,ηHi{n}⊗{m}=[n(aSx−aSt)+maΘ+aSs]Hi{n}⊗{m}+ai{n}⊗{m},HηIi{n−1}⊗{m}[a]P=[(n+1)(aSx−aSt)+maΘ+aSs]Ii{n−1}⊗{m}[a]P+ai{n−1}⊗{m}I.(42)\begin{aligned} \xi_t&=a_{St}t,\quad\xi_x=a_{Sx}\boldsymbol{x}+\boldsymbol{a}_x,\\ \eta_{H_{i_{\{n\}\otimes\{m\}}}}&=[n(a_{Sx}-a_{St})+ma_{\Theta}+a_{Ss}]H_{i_{\{n\}\otimes\{m\}}}+a^{H}_{i_{\{n\}\otimes\{m\}},}\\ \eta_{I_{i_{\{n-1\}\otimes\{m\}}[a]P}}&=[(n+1)(a_{Sx}-a_{St})+ma_{\Theta}+a_{Ss}]I_{i_{\{n-1\}\otimes\{m\}}[a]P}+a^{I}_{i_{\{n-1\}\otimes\{m\}}}. \tag*{(42)} \end{aligned}

and which will be used below in the next chapter.

High order moment scaling laws and its validation

New velocity, temperature and mixed moment scaling laws

Using the symmetries (41), or rather its infinitesimal form (42), we are now able to compute the invariant solutions of the equations (18), which in turbulence are called the turbulent scaling laws. It should be noted here that in the following we only consider the moments of the instantaneous variables, i.e. the HH approach. Of course, a conversion for each moment into that of the fluctuations is straightforward, but in Oberlack et al.[32] we were able to show that an error accumulation occurs in the calculation from DNS data, which inevitably results from the finite number of available flow fields. This error increases considerably as the order of the moments increases. We have therefore deliberately refrained from displaying the moments from the fluctuations.

Further, we note that the study focuses now on the streamwise velocity and temperature. For this purpose, we only have to insert the infinitesimals (42) into the invariant surface condition (31) and we get

dx2aSxx2+ax2=dH1{1}[aSx−aSt+aSs]H1{1}+a1{1}H=dHΘ{1}[aΘ+aSs]HΘ{1}+aΘ{1}H=dH1{1}Θ{1}[aSx−aSt+aΘ+aSs]H1{1}Θ{1}+a1{1}Θ{1}H=⋯=dH1{n}Θ{m}[n(aSx−aSt)+maΘ+aSs]H1{n}Θ{m}+a1{n}Θ{m}H(43)\begin{aligned} \frac{dx_2}{a_{Sx}x_2+a_{x2}} &= \frac{dH_{1\{1\}}}{[a_{Sx}-a_{St}+a_{Ss}]H_{1\{1\}}+a^H_{1\{1\}}}\\ &=\frac{dH_{\Theta\{1\}}}{[a_{\Theta}+a_{Ss}]H_{\Theta\{1\}}+a^H_{\Theta\{1\}}}\\ &=\frac{dH_{1\{1\}\Theta\{1\}}}{[a_{Sx}-a_{St}+a_{\Theta}+a_{Ss}]H_{1\{1\}\Theta\{1\}}+a^H_{1\{1\}\Theta\{1\}}}\\ &=\cdots\\ &=\frac{dH_{1\{n\}\Theta\{m\}}}{[n(a_{Sx}-a_{St})+ma_{\Theta}+a_{Ss}]H_{1\{n\}\Theta\{m\}}+a^H_{1\{n\}\Theta\{m\}}} \tag*{(43)} \end{aligned}

Since we consider a shear flow that is fully developed in x1x_1 and x3x_3, all moments in the one-point limit depend only on x2x_2. Furthermore, the dependencies of the other points for the higher-order tensors would have to be formally considered as well, because this would result in further similarity variables. However, from now on, we will focus on one-point statistics, so that every point of application of the variables in (18) will be x(1)=x(2)=⋯=x(n+m)x_{(1)}=x_{(2)}=\cdots=x_{(n+m)}. Integrating (43), notice that we are using the first and the last term because the other terms are just specific moments, we obtain the following invariant solutions for any arbitrary moment

H1{n}Θ{m}=C1{n}Θ{m}′(x2+ax2aSx)n(σ2−σ1)+mσΘ+2σ1−σ2−a1{n}Θ{m}Hn(aSx−aSt)+maΘ+aSs(44)H_{1\{n\}\Theta\{m\}}=C'_{1\{n\}\Theta\{m\}}\left(x_2+\frac{a_{x2}}{a_{Sx}}\right)^{n(\sigma_2-\sigma_1)+m\sigma_{\Theta}+2\sigma_1-\sigma_2}-\frac{a^H_{1\{n\}\Theta\{m\}}}{n(a_{Sx}-a_{St})+ma_{\Theta}+a_{Ss}} \tag*{(44)}

with

C1{n}Θ{m}′=ecnm′[n(aSx−aSt)+maΘ+aSs],(45)C'_{1\{n\}\Theta\{m\}}=e^{c'_{nm}[n(a_{Sx}-a_{St})+ma_{\Theta}+a_{Ss}]}, \tag*{(45)}

where cnm′c'_{nm} denote the constants of integration, σ1=1−aSt/aSx+aSs/aSx\sigma_1=1-a_{St}/a_{Sx}+a_{Ss}/a_{Sx}, σ2=2(1−aSt/aSx)+aSs/aSx\sigma_2=2(1-a_{St}/a_{Sx})+a_{Ss}/a_{Sx} and σΘ=aΘ/aSx\sigma_{\Theta}=a_{\Theta}/a_{Sx}. Similar to Ref. 1, the choice of parameters in the exponent of (44) has been designed so that the high-order moments depend on those of the first and second order. Focusing on the velocity only, as in Ref. 1, i.e. m=0m=0, the exponent for n=1n=1 is σ1\sigma_1, and for n=2n=2, it is σ2\sigma_2, while for pure temperature moments, i.e. n=0n=0, we have for m=1m=1 the exponent σΘ\sigma_{\Theta}. Therefore, σ1\sigma_1 and σ2\sigma_2, are determined from the first two velocity moments, while σΘ\sigma_{\Theta} is determined from the first temperature moment.

At this point, it is important to recall that the invariant solution (44) has been derived in the limit of vanishing viscosity and heat conduction. Therefore, this solution will be only valid in the region where these conditions apply, i.e. the centre of the channel. The invariant solution (44) shows that the moments of velocity, temperature, and higher-order moments scale as power-laws, whose exponents are determined by the parameters σ1\sigma_1, σ2\sigma_2 and σΘ\sigma_\Theta. Note that, in the exponent of the power law, four initial parameters appear (aSxa_{Sx}, aSta_{St}, aΘa_\Theta, and aSsa_{Ss}). However, these parameters appear as ratios, leading to only three free parameters remaining (σ1\sigma_1, σ2\sigma_2, and σΘ\sigma_\Theta).

Analogously to Ref. 17, where the scaling law of the mean velocity of a turbulent shear flow was presented as a deficit law, and significantly extended in Ref. 1 to arbitrary velocity moments, equation (44) can be rewritten to form the final deficit scaling law of velocity, temperature and arbitrarily mixed moments of both as

H1{n}Θ{m}−H1{n}Θ{m}cluτnΘτm=Cnm′(x2h)n(σ2−σ1)+mσΘ+2σ1−σ2(46)\frac{H_{1\{n\}\Theta\{m\}}-H_{1\{n\}\Theta\{m\}cl}}{u_\tau^n\Theta_\tau^m}=C'_{nm}\left(\frac{x_2}{h}\right)^{n(\sigma_2-\sigma_1)+m\sigma_\Theta+2\sigma_1-\sigma_2} \tag*{(46)}

with

Cnm′=α′nβΘ′m,(47)C'_{nm}=\alpha'^{n}\beta_\Theta'^{m}, \tag*{(47)}

where the subscript clcl refers to the value of the moment on the centre line, which comes from the last term on the right-hand side of the equation (44), and Cnm′C'_{nm} are the new exponential scaling factors. Similar to Ref. 1, it has been assumed that cnm′c'_{nm} is independent of nn and mm to derive Cnm′C'_{nm} in (46), and this will indeed be validated below in section §IV B. Also note that, in equation (46), the shift in x2x_2 has been set to 0, as the coordinate is anchored to the centre line.

Validation of the scaling law (46) with DNS data

The new scaling law (46) will be validated by using DNS data of turbulent channel flows driven by a pressure gradient at friction Reynolds numbers of Reτ=500Re_\tau= 500, 1000, 2000 and 5000, and heated by a constant heat flux from both walls with a wide range of values of Prandtl numbers: Pr=0.007Pr = 0.007, 0.01, 0.02, 0.05, 0.1, 0.3, 0.5, 0.71, 1, 2, 4, 7 and 10. Specifically, the combinations of ReτRe_\tau and PrPr used are presented in table I.

The code used to run the simulations is the Liso code, already validated and employed in many other simulations of turbulent channel flows[7, 9, 11, 19, 33, 34]. Detailed information about the code itself and the parameters of the simulation (mesh size, wash-outs run, computational box size,. . . ) can be found in Ref. 8, 25, 26, and 28. As mentioned, the UHF is used as the thermal boundary condition in the DNS cited. This implies that temperature increases linearly in the streamwise direction.

To make the temperature field homogeneous in the x1x_1 direction, the value of the temperature at the wall is removed, obtaining a transformed temperature. Because we want to give a general scaling law for temperature moments, and because the same symmetry, TΘT_{\Theta} (34), is also obtained for the energy equation of the transformed temperature, the general form of the temperature energy equation is used in this work.

The moments calculated in the simulations are limited to order seven for pure moments of velocity and temperature and six for mixed moments of velocity and temperature. The procedure to fit the scaling law (46) to the DNS data has been done by minimizing the infinite norm of the relative error between the fit and the value of the DNS data, i.e.

error⁡=min⁡(∥data⁡(x2)−fit⁡(x2)data⁡(x2)∥∞).(48)\operatorname{error} = \min\left( \left\lVert\frac{\operatorname{data}(x_2)-\operatorname{fit}(x_2)}{\operatorname{data}(x_2)} \right\rVert_{\infty} \right). \tag*{(48)}

The infinite norm is used in this formula since the data takes values across several orders of magnitude. After this fitting is applied to the first and second moments of velocity and the first moment of temperature, σ1\sigma_1, σ2\sigma_2 and σΘ\sigma_{\Theta} from the scaling law (46) are determined and, thus, the exponent for any high order moment is known and only the constants of integration, Cnm′C'_{nm}, must be calculated.

The results of the fits of the velocity moments for Reτ=500Re_{\tau}=500 and Pr=4Pr=4 are depicted in Figure 1a, together with the fits of the velocity moments for Reτ=2000Re_{\tau}=2000 and Pr=7Pr=7, in Figure 1b. Additionally, for Reτ=2000Re_{\tau}=2000 and Pr=7Pr=7, the fits of the temperature moments and mixed moments are presented in Figure 1c and Figure 1d, respectively. Solid lines represent the values from the DNS, while squares are the values obtained from the scaling law (46). Recall that: in all figures below, the wall and centre of the channel are swapped, so the centre line is at x2/h=0x_2/h=0, while the wall is at x2/h=1x_2/h=1. Also, it is important to mention that the range of the centre of the channel where

Four-panel plots of moments of velocity, temperature, and heat fluxes

Figure 1. Moments of velocity, H1{n}H_{1\{n\}}, for (a) Reτ=500Re_\tau= 500 and Pr=4Pr = 4 and (b) Reτ=2000Re_\tau= 2000 and Pr=7Pr = 7. Moments of (c) temperature, HΘ{m}H_{\Theta\{m\}}, and (d) heat fluxes, H1{n}Θ{m}H_{1\{n\}\Theta\{m\}}, for Reτ=2000Re_\tau= 2000 and Pr=7Pr = 7. In (a), (b), and (c), velocity and temperature moments are obtained for nn and m=1,2,…,7m = 1, 2,\ldots, 7, appearing in that order from bottom to top of the plot. For (d), heat fluxes moments are shown for n+m=2,3,…,6n+m = 2, 3,\ldots, 6, appearing in that order from bottom to top of the plot. For heat fluxes moments of the same order, the lower lines are for m=0m = 0, while the upper lines are for n=0n = 0. Solid lines are the values from the DNS, while squares represent the values from the scaling law. The wall and centre of the channel are swapped, so the centre line is at x2/h=0x_2/h = 0, while the wall is at x2/h=1x_2/h = 1. Colours as in Table I.

Reτ\PrRe_{\tau} \backslash Pr0.0070.010.020.050.10.30.50.71124710Colour
500XXXXXXXXXXXXX
1000XXXXXXXXXXXX
2000XXXXXXXXXXXX
5000X

Table I. ReτRe_{\tau} and PrPr numbers used for the validation of the scaling law (46). The last column shows the colours used in the figures to refer to each Reynolds number.

the scaling law has been applied is up to x2/h=0.75x_2/h = 0.75. The most important result of this work is the high accuracy of the scaling law (46) to fit the data of the DNS for all moments, with the highest relative errors of only 0.2% for the higher order moments, calculated with equation (48). Even for the lowest Reynolds numbers of value 500, the accuracy of the fit is almost as good as for Reτ=2000Re_\tau= 2000, as can be seen in Figure 1a and Figure 1b. In the same way, the scaling law is validated with the same accuracy for the temperature and mixed moments as shown in Figure 1c and Figure 1d, respectively.

To analyse the influence of the Prandtl number, we compare the scaling between a high Prandtl number of 4 and a very low one of 0.01 in Figure 2. In the case of the scaling of the temperature moments, in Figure 2a and Figure 2b, for cases Pr=4Pr = 4 and 0.01, respectively, there is a noticeable difference. While for Pr=4Pr = 4 the scaling (46) represents the DNS data with high accuracy, errors are lower than 0.01%, for the case of Pr=0.01Pr = 0.01, the deviation is high and clearly visible in Figure 2b. The reason for this error comes from the assumption in section §III of zero heat conduction in the symmetry analysis, which, obviously, for Pr=0.01Pr = 0.01 is not true. The high diffusivity for very low Prandtl numbers affects the temperature field in a deeper region away from the wall, and the temperature moments are no longer parallel for x2/hx_2/h approximately greater than 0.2. In the same manner, the scaling of mixed moments is no longer correct for very low Prandtl numbers. While in Figure 2c, the scaling is again very accurate for Pr=4Pr = 4, in Figure 2d, a similar failure, as obtained in Figure 2b, appears in the scaling for Pr=0.01Pr = 0.01. Note that in Figure 2d, only moments for n+m=2n+m = 2, 4, and 6 are plotted for clarity of the figure.

Moments of temperature and heat fluxes for different Reynolds and Prandtl numbers

Figure 2. Moments of temperature, HΘ{m}H_{\Theta\{m\}}, for (a) Reτ=500Re_\tau= 500 and Pr=4Pr = 4 and (b) Reτ=500Re_\tau= 500 and Pr=0.01Pr = 0.01. Moments of heat fluxes, H1{n}Θ{m}H_{1\{n\}\Theta\{m\}}, for (c) Reτ=500Re_\tau= 500 and Pr=4Pr = 4 and (d) Reτ=500Re_\tau= 500 and Pr=0.01Pr = 0.01. In (a) and (b), temperature moments are obtained for m=1,2,…,7m = 1, 2,\ldots, 7, appearing in that order from the bottom to the top of the plot. For (c), heat fluxes moments are shown for n+m=2,3,…,6n+m = 2, 3,\ldots, 6, appearing in that order from bottom to top of the plot. For heat fluxes moments of the same order, the lower lines are for m=0m = 0, while the upper lines are for n=0n = 0. For (d), heat fluxes moments are shown for n+m=2,3n+m = 2, 3, and 66, appearing in that order from bottom to top of the plot. For heat fluxes moments of the same order, the lower lines are for n=0n = 0, while the upper lines are for m=0m = 0. Solid lines are the values from the DNS, while squares represent the values from the scaling law. The wall and centre of the channel are swapped, so the centre line is at x2/h=0x_2/h = 0, while the wall is at x2/h=1x_2/h = 1. Colours as in table I.

One important point to note is that the functional structure of the invariant solution (44) that leads to the scaling law (46) does not change for vanishing or not vanishing viscosity/diffusivity, in the sense that the same number of parameter appears in the exponent of the scaling law. Therefore, one should expect the scaling law to be correct also for these cases. This is true for a limited region of the centre of the channel. In the case of Figure 2b and Figure 2d, if one tries to use the scaling law only for in the region of y∗<0.1y^* < 0.1 away from the centre, instead of y∗<0.75y^* < 0.75, as was done in Figure 2, then a perfect matching with the DNS data will be obtained, even for the lowest Prandtl number cases. As mentioned above, this happens because the viscous/diffusive effects appear so deep away from the wall. Effectively, as shown in Ref. 25, for such low Reynolds and Prandtl numbers, one cannot even see the emergence of a logarithmic layer, so a bad scaling is also expected. Also, remark that the coefficient σ1\sigma_1 and σ2\sigma_2 approach to a value of 2 as the region where the fitting is done is reduced.

Note here that although a Reτ=500Re_{\tau}=500 is a low Reynolds number and one may expect the assumption of vanishing viscosity to fail, the velocity field is still turbulent and in the centre of the channel, the mentioned assumption is true. However, Pr=0.01Pr=0.01 produces a much less turbulent temperature field, even also laminar[25], and for this reason, the assumption of vanishing heat conduction is not true for such a low Prandtl number, or at least is only true in a very central region of the channel (less than 10%), where the temperature moments are parallel in Figure 2b. In addition, the key parameter is actually not the Prandtl number by itself, but the friction Péclet number, defined as Peτ=ReτPrPe_{\tau}=Re_{\tau}Pr. Therefore, in our plots, we are comparing a Reτ=500Re_{\tau}=500 with Pr=4Pr=4 and 0.01, which means Peτ=2000Pe_{\tau}=2000 compared with Peτ=5Pe_{\tau}=5, explaining the errors in the scaling for such a low Prandtl or friction Péclet numbers.

It is important to analyse the values of the different exponential parameters of the scaling law (46) and see the influence of each symmetry on the final scaling. As it was shown in equation (46), the exponent of the power-law is formed by a constant term, 2σ1−σ22\sigma_{1}-\sigma_{2}, that comes from the statistical scaling symmetry of the moments, T‾Ss\overline{T}_{\mathrm{Ss}} (38). A second term that scales with nn, i.e. n(σ2−σ1)n(\sigma_{2}-\sigma_{1}), emerged from the classical scaling symmetries of space and time, T‾Sx\overline{T}_{\mathrm{Sx}} (35) and T‾St\overline{T}_{\mathrm{St}} (36), respectively. A third term that scales with mm, i.e. mσΘm\sigma_{\Theta}, has its roots in the scaling symmetry of the temperature, T‾SΘ\overline{T}_{\mathrm{S\Theta}} (37). However, as can be clearly seen in Figure 1, all moments have a more or less constant slope in the log-log plot, which translates into a very weak dependence on nn and mm. In other words, the values of σ1\sigma_{1} and σ2\sigma_{2} are very similar, and σΘ\sigma_{\Theta} is small compared with the value of 2σ1−σ22\sigma_{1}-\sigma_{2}. This, in turn, implies that scaling of space and time has almost no influence in the centre of the channel, and the statistical scaling of moments is dominant, which makes sense, since, as it was said before, it is a measure of intermittency. Scaling independent of the dimensions of space and time is called anomalous scaling and has its origin in the intermittency symmetry.

Figure 3a presents the values of σ1\sigma_{1} and σ2\sigma_{2} for all the DNS simulations, while the values of σΘ\sigma_{\Theta} are shown in Figure 3b. The values of σ1\sigma_{1} are almost the same as σ2\sigma_{2} (note that the left and right axes are shifted for better visualization), which confirms that the symmetries of scaling in space and time have almost no influence in the centre of the channel. Similarly, the scaling symmetry of temperature has barely any influence, since σΘ≪2σ1−σ2\sigma_{\Theta}\ll2\sigma_{1}-\sigma_{2}. This last term, 2σ1−σ22\sigma_{1}-\sigma_{2}, is indeed the only dominant term in the exponent of the scaling law (46), with a value slightly below 2, confirming that the symmetry of scaling of moments, T‾Ss\overline{T}_{\mathrm{Ss}} (38), is dominant in the centre of the channel. Further, we observe that the parameters, σ1\sigma_{1}, σ2\sigma_{2} and σΘ\sigma_{\Theta}, in the investigated ranges, are largely independent of ReτRe_{\tau} and PrPr. Small differences are due to small numerical errors or noise in the fitting.

Values of σ₁, σ₂ and σΘ

Figure 3. Values of (a) σ1\sigma_1, solid lines left axis, and σ2\sigma_2, dashed lines right axis, and (b) σΘ\sigma_{\Theta}. Colours as in table I. Note that black circles at Pr=0.71Pr = 0.71 represent the value for the single simulation at Reτ=5000Re_\tau= 5000.

The second important part of the scaling law (46) is the prefactor Cnm′C^\prime_{nm}. Figure 4a shows the values of Cnm′C^\prime_{nm} for Reτ=500Re_\tau= 500 and Pr=4Pr = 4. An almost perfect plane is observed in the vertical log-scaling plot, which confirms that Cnm′C^\prime_{nm} is an exponential function in nn and mm, and further verifies that the constants of integration cnm′c^\prime_{nm} in equation (45) are independent of nn and mm.

Values of $C_{1_{\langle n\rangle}\Theta_{\langle m\rangle}}$ and parameters $\alpha'$, $\beta'$, and $\beta'_{\Theta}$

Figure 4. (a) Values of C1⟨n⟩Θ⟨m⟩C_{1_{\langle n\rangle}\Theta_{\langle m\rangle}} for Reτ=500Re_{\tau}=500 and Pr=4Pr=4. Parameters from equation (47): (b) α′\alpha', (c) β′\beta' and (d) βΘ′\beta'_{\Theta}. Colours as in table I. Note that black points at Pr=0.71Pr=0.71 in (b), (c), and (d) represent the values for the single simulation at Reτ=5000Re_{\tau}=5000.

In figures 4b, 4c and 4d, the values of α′\alpha^\prime, β′\beta^\prime and βΘ′\beta^\prime_{\Theta} for all the present simulations are shown, respectively. To calculate them, only the constants of integration of the moments up to order 2 have been used, i.e., C10′C^\prime_{10}, C20′C^\prime_{20}, C01′C^\prime_{01}, C02′C^\prime_{02} and C11′C^\prime_{11}. With these five values, a fit of the parameters α′\alpha^\prime, β′\beta^\prime, and βΘ′\beta^\prime_{\Theta} have been done minimizing again as in equation (48).

Interestingly enough, and other than the parameters in the exponent σ1\sigma_1, σ2\sigma_2 and σΘ\sigma_{\Theta}, the coefficients α′\alpha^\prime, β′\beta^\prime and βΘ′\beta^\prime_{\Theta} are not independent of ReτRe_\tau and PrPr. We want to point out that for high friction Péclet numbers, the values of α′\alpha^\prime, β′\beta^\prime and βΘ′\beta^\prime_{\Theta} seem to be independent of the friction Reynolds number, which may be related to a more realistic assumption of the zero viscosity and heat conduction. However, this is just a point to be investigated. From the theory developed in section §III, the dependency on ReτRe_\tau or PrPr is not apparent but goes beyond the scope of the theory in its present form. Here we limited the study to confirm that the scaling law (46) can represent the behaviour of the arbitrary moments obtained from the DNS data, including exponential scaling of Cnm′C^\prime_{nm} with nn and mm.

Conclusions

A new set of turbulent scaling laws for arbitrary moments of the streamwise velocity, temperature, and high order moments of both in a turbulent channel flow has been obtained using the symmetry-based turbulence theory. These scaling laws apply to incompressible flows driven by a pressure difference and with a passive scalar. For the derivation of the scaling laws, we had to assume vanishing viscosity and diffusion, i.e., Reτ→∞Re_{\tau}\to\infty and Pr>1Pr>1, which holds in the central region of the channel, and they are finally cast as deficit laws.

The deficit form of the arbitrary moments in the wall-normal direction can be represented as power functions, where the exponent is determined by the order of the moments and three different parameters that emerged from four different scaling symmetries (σ1\sigma_{1}, σ2\sigma_{2} and σΘ\sigma_{\Theta}). Besides the classical symmetries of the Navier–Stokes and energy equations, we employed statistical symmetries of the multi-point correlation equations, which were the key to obtaining a constant exponent of the power-law scaling function that can accurately represent the DNS data. Instead of the usual fluctuation approach as the basis for the MPC equations, which yields a non-linear system of equations, we presently employ the instantaneous approach, which results in a linear system of equations. The statistical symmetries are trivially displayed in this representation as scaling and translation of moments. This statistical scaling of moments represents a measure of intermittency. It appears as the dominant term in the exponent of all moments as the constant 2σ1−σ22\sigma_{1}-\sigma_{2}, which is independent of the moment order.

The scaling laws have been validated with data from different DNS at different Reynolds and Prandtl numbers. The accuracy of the scaling laws to represent the data is remarkable, especially for high Péclet numbers. For cases with low Péclet numbers, the centre of the channel gets influenced by viscosity and heat conduction, and the assumption of Reτ→∞Re_{\tau} \to\infty and Pr>1Pr > 1 no longer holds, which entails a significant deviation from the theoretical scaling of the moments in the centre of the channel.

The exponential prefactor in nn and mm in equation (47) has been obtained by the observation that the constants of integration cnm′c_{nm}^{\prime} in 45 are indeed constant and independent of nn and mm. So far, no justification based on first principles can be given for this, though we speculate that the Probability Density Function (PDF) contains deeper information on this. Therefore, we presently follow the idea of deriving invariant solutions to the PDF equations.

One point of this theory to be explored is the derivation of scaling laws for other important statistics such as fluctuating quantities, cross velocities, the wall-normal and spanwise heat fluxes, and their high-order moments. So far, with the symmetries obtained in this work, it was impossible to properly describe these statistics. However, as mentioned before, further symmetries can be obtained from the MPC equations or from other equations that describe turbulence, such as the PDF equations. This is left as future work which is already being investigated, but we would like to point out that our method is able to obtain scaling laws following only strong mathematical arguments, removing lucky curve fitting.

Acknowledgments

This work was supported by PID2021-128676OB-I00 of MINECO/FEDER. FAA is partially funded by GVA/FEDER project ACIF2018. The computations of the new simulations were made possible by a generous grant of computing time from the Barcelona Supercomputing Centre, reference AECT-2020-1-0024. MO expresses his gratitude for the partial support of the German Research Foundation (DFG) within the project OB 96/48-1. We are grateful to Mr Jonathan Laux for providing us with the scripts to do the fittings of the scaling laws with the data. Declaration of Interests. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A

In this appendix, we present a step-by-step derivation of the MPC equation of the mixed moments. The derivation of the MPC equations of velocity and thermal energy are just two specific cases of the general MPC equations.

Starting with equations (2) and (3) we perform the following operation to obtain the MPC equation of order n+mn+m of the mixed moments

M‾i(1)(x(1))U‾i(2)(x(2))⋯U‾i(n)(x(n))Θ(x(n+1))⋯Θ(x(n+m))+U‾i(1)(x(1))Mi(2)(x(2))U‾i(3)(x(3))⋯U‾i(n)(x(n))Θ(x(n+1))⋯Θ(x(n+m))+⋯+U‾i(1)(x(1))⋯U‾i(n−1)(x(n−1))Mi(n)(x(n))Θ(x(n+1))⋯Θ(x(n+m))+U‾i(1)(x(1))⋯U‾i(n)(x(n))E(x(n+1))Θ(x(n+2))⋯Θ(x(n+m))+U‾i(1)(x(1))⋯U‾i(n)(x(n))Θ(x(n+1))E(x(n+2))Θ(x(n+3))⋯Θ(x(n+m))+⋯+U‾i(1)(x(1))⋯U‾i(n)(x(n))Θ(x(n+1))⋯Θ(x(n+m−1))E(x(n+m))=∑a=1nMi(a)(x(a))∏c=1c≠anU‾i(c)(x(c))∏d=n+1n+mΘ(x(d))+∑b=n+1n+mE(x(b))∏c=1nU‾i(c)(x(c))∏d=n+1d≠cn+mΘ(x(d))=0,(49)\begin{aligned} &\overline{\mathcal{M}}_{i(1)}\left(\boldsymbol{x}_{(1)}\right)\overline{U}_{i(2)}\left(\boldsymbol{x}_{(2)}\right)\cdots\overline{U}_{i(n)}\left(\boldsymbol{x}_{(n)}\right)\Theta\left(\boldsymbol{x}_{(n+1)}\right)\cdots\Theta\left(\boldsymbol{x}_{(n+m)}\right)\\ &+\overline{U}_{i(1)}\left(\boldsymbol{x}_{(1)}\right)\mathcal{M}_{i(2)}\left(\boldsymbol{x}_{(2)}\right)\overline{U}_{i(3)}\left(\boldsymbol{x}_{(3)}\right)\cdots\overline{U}_{i(n)}\left(\boldsymbol{x}_{(n)}\right)\Theta\left(\boldsymbol{x}_{(n+1)}\right)\cdots\Theta\left(\boldsymbol{x}_{(n+m)}\right)\\ &+\cdots\\ &+\overline{U}_{i(1)}\left(\boldsymbol{x}_{(1)}\right)\cdots\overline{U}_{i(n-1)}\left(\boldsymbol{x}_{(n-1)}\right)\mathcal{M}_{i(n)}\left(\boldsymbol{x}_{(n)}\right)\Theta\left(\boldsymbol{x}_{(n+1)}\right)\cdots\Theta\left(\boldsymbol{x}_{(n+m)}\right)\\ &+\overline{U}_{i(1)}\left(\boldsymbol{x}_{(1)}\right)\cdots\overline{U}_{i(n)}\left(\boldsymbol{x}_{(n)}\right)\mathcal{E}\left(\boldsymbol{x}_{(n+1)}\right)\Theta\left(\boldsymbol{x}_{(n+2)}\right)\cdots\Theta\left(\boldsymbol{x}_{(n+m)}\right)\\ &+\overline{U}_{i(1)}\left(\boldsymbol{x}_{(1)}\right)\cdots\overline{U}_{i(n)}\left(\boldsymbol{x}_{(n)}\right)\Theta\left(\boldsymbol{x}_{(n+1)}\right)\mathcal{E}\left(\boldsymbol{x}_{(n+2)}\right)\Theta\left(\boldsymbol{x}_{(n+3)}\right)\cdots\Theta\left(\boldsymbol{x}_{(n+m)}\right)\\ &+\cdots\\ &+\overline{U}_{i(1)}\left(\boldsymbol{x}_{(1)}\right)\cdots\overline{U}_{i(n)}\left(\boldsymbol{x}_{(n)}\right)\Theta\left(\boldsymbol{x}_{(n+1)}\right)\cdots\Theta\left(\boldsymbol{x}_{(n+m-1)}\right)\mathcal{E}\left(\boldsymbol{x}_{(n+m)}\right)\\ &=\sum_{a=1}^{n}\mathcal{M}_{i(a)}\left(\boldsymbol{x}_{(a)}\right)\prod_{\substack{c=1\\c\ne a}}^{n}\overline{U}_{i(c)}\left(\boldsymbol{x}_{(c)}\right)\prod_{d=n+1}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)\\ &\quad+\sum_{b=n+1}^{n+m}\mathcal{E}\left(\boldsymbol{x}_{(b)}\right)\prod_{c=1}^{n}\overline{U}_{i(c)}\left(\boldsymbol{x}_{(c)}\right)\prod_{\substack{d=n+1\\d\ne c}}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)=0, \tag*{(49)} \end{aligned}

where Ui(l)U_{i(l)} and x(l)x_{(l)} are the velocity and the different points where the equations and the variables are applied, for i(l)=1,2,3i_{(l)}=1,2,3; and l=1,2,…,n+ml=1,2,\ldots,n+m (ll can be aa or bb). Introducing the momentum and energy equations, (2) and (3), into (A1), we obtain

∑a=1n∂Ui(a)(x(a))∂t∏c=1c≠anUi(c)(x(c))∏d=n+1n+mΘ(x(d))+∑a=1nUk(x(a))∂Ui(a)(x(a))∂xk(a)∏c=1c≠anUi(c)(x(c))∏d=n+1n+mΘ(x(d))+∑a=1n∂P(x(a))∂xi(a)∏c=1c≠anUi(c)(x(c))∏d=n+1n+mΘ(x(d))−1Reτ∑a=1n∂2Ui(a)(x(a))∂xk(a)∂xk(a)∏c=1c≠anUi(c)(x(c))∏d=n+1n+mΘ(x(d))+∑b=n+1n+m∂Θ(x(b))∂t∏c=1nUi(c)(x(c))∏d=n+1d≠cn+mΘ(x(d))+∑b=n+1n+mUk(x(b))∂Θ(x(b))∂xk(b)∏c=1nUi(c)(x(c))∏d=n+1d≠cn+mΘ(x(d))−1Peτ∑b=n+1n+m∂2Θ(x(b))∂xk(b)∂xk(b)∏c=1nUi(c)(x(c))∏d=n+1d≠cn+mΘ(x(d))=0.(50)\begin{aligned} &\sum_{a=1}^{n}\frac{\partial U_{i_{(a)}}\left(\boldsymbol{x}_{(a)}\right)}{\partial t}\prod_{\substack{c=1\\c\ne a}}^{n}U_{i_{(c)}}\left(\boldsymbol{x}_{(c)}\right)\prod_{d=n+1}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)\\ &+\sum_{a=1}^{n}U_k\left(\boldsymbol{x}_{(a)}\right)\frac{\partial U_{i_{(a)}}\left(\boldsymbol{x}_{(a)}\right)}{\partial x_{k_{(a)}}}\prod_{\substack{c=1\\c\ne a}}^{n}U_{i_{(c)}}\left(\boldsymbol{x}_{(c)}\right)\prod_{d=n+1}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)\\ &+\sum_{a=1}^{n}\frac{\partial P\left(\boldsymbol{x}_{(a)}\right)}{\partial x_{i_{(a)}}}\prod_{\substack{c=1\\c\ne a}}^{n}U_{i_{(c)}}\left(\boldsymbol{x}_{(c)}\right)\prod_{d=n+1}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)\\ &-\frac{1}{Re_\tau}\sum_{a=1}^{n}\frac{\partial^2U_{i_{(a)}}\left(\boldsymbol{x}_{(a)}\right)}{\partial x_{k_{(a)}}\partial x_{k_{(a)}}}\prod_{\substack{c=1\\c\ne a}}^{n}U_{i_{(c)}}\left(\boldsymbol{x}_{(c)}\right)\prod_{d=n+1}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)\\ &+\sum_{b=n+1}^{n+m}\frac{\partial\Theta\left(\boldsymbol{x}_{(b)}\right)}{\partial t}\prod_{c=1}^{n}U_{i_{(c)}}\left(\boldsymbol{x}_{(c)}\right)\prod_{\substack{d=n+1\\d\ne c}}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)\\ &+\sum_{b=n+1}^{n+m}U_k\left(\boldsymbol{x}_{(b)}\right)\frac{\partial\Theta\left(\boldsymbol{x}_{(b)}\right)}{\partial x_{k_{(b)}}}\prod_{c=1}^{n}U_{i_{(c)}}\left(\boldsymbol{x}_{(c)}\right)\prod_{\substack{d=n+1\\d\ne c}}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)\\ &-\frac{1}{Pe_\tau}\sum_{b=n+1}^{n+m}\frac{\partial^2\Theta\left(\boldsymbol{x}_{(b)}\right)}{\partial x_{k_{(b)}}\partial x_{k_{(b)}}}\prod_{c=1}^{n}U_{i_{(c)}}\left(\boldsymbol{x}_{(c)}\right)\prod_{\substack{d=n+1\\d\ne c}}^{n+m}\Theta\left(\boldsymbol{x}_{(d)}\right)=0. \tag*{(50)} \end{aligned}

At this point, the continuity equation (1) should be applied to introduce the terms Uk(x(l))U_k(\boldsymbol{x}_{(l)}) inside the derivatives with respect to xk(l)x_{k_{(l)}} in the second and sixth lines of equation (A2). Also, the product terms can be introduced in the derivatives with respect to the spatial coordinates, since the points x(a)\boldsymbol{x}_{(a)} and x(b)\boldsymbol{x}_{(b)} are excluded from the product series. Regarding the temporal derivatives in the first and fifth lines of (A2), the chain rule is applied to reduce it to a single term. Finally, using definitions (15), (16) and (17) one can obtain the MPC equation for all mixed moments written in the following way

\begin{aligned} &\frac{\partial H_{i_{\{n\}}\Theta_{\{m\}}}}{\partial t} +\sum_{a=1}^{n}\left(\frac{\partial H_{i_{\{n+1\}}\Theta_{\{m\}}[i_{(n+m+1)}\to k]}\left(\boldsymbol{x}_{(n+m+1)}\to\boldsymbol{x}_{(a)}\right)}{\partial x_{k_{(a)}}} +\frac{\partial I_{i_{\{n-1\}}\Theta_{\{m\}}[a]P}}{\partial x_{i_{(a)}}} -\frac{1}{Re_\tau}\frac{\partial^2H_{i_{\{n\}}\Theta_{\{m\}}}}{\partial x_{k_{(a)}}\partial x_{k_{(a)}}}\right)\\ &+\sum_{b=n+1}^{n+m}\left(\frac{\partial H_{i_{\{n+1\}}\Theta_{\{m\}}[i_{(n+m+1)}\to k]\left(\boldsymbol{x}_{(n+m+1)}\to\boldsymbol{x}_{(b)}\right)}{\partial x_{k_{(b)}}} -\frac{1}{Pe_\tau}\frac{\partial^2H_{i_{\{n\}}\Theta_{\{m\}}}}{\partial x_{k_{(b)}}\partial x_{k_{(b)}}}\right)=0. \end{aligned}} \tag*{(51)}\end{aligned}

As mentioned before, the MPC equations of the velocity arises if m=0m=0 in (A3). Similarly, one can obtain the MPC equations of the temperature by setting n=0n=0 in (A3).

Appendix B

In this appendix, the Two-Point Correlation (TPC) equations for the velocity, heat fluxes and temperature are given as examples of the MPC equation (18), in order to make the notation clearer. These equations can be obtained by setting in equation (18) n=2n=2, 11, 00 and m=0m=0, 11, 22, respectively.

∂Hi(1)i(2)(x(1),x(2))∂t+∂Hi(1)i(2)k(x(1),x(2),x(1))∂xk(1)+∂Hi(1)i(2)k(x(1),x(2),x(2))∂xk(2)+∂IPi(2)(x(1),x(2))∂xi(1)+∂Ii(1)P(x(1),x(2))∂xi(2)−ν∂2Hi(1)i(2)(x(1),x(2))∂xk(1)∂xk(1)−ν∂2Hi(1)i(2)(x(1),x(2))∂xk(2)∂xk(2)=0,(52)\begin{aligned} \frac{\partial H_{i_{(1)}i_{(2)}}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial t} &+\frac{\partial H_{i_{(1)}i_{(2)}k}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)},\boldsymbol{x}_{(1)})}{\partial x_{k_{(1)}}} +\frac{\partial H_{i_{(1)}i_{(2)}k}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(2)}}}\\ &+\frac{\partial IP_{i_{(2)}}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{i_{(1)}}} +\frac{\partial I_{i_{(1)}}P(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{i_{(2)}}}\\ &-\nu\frac{\partial^2 H_{i_{(1)}i_{(2)}}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(1)}}\partial x_{k_{(1)}}} -\nu\frac{\partial^2 H_{i_{(1)}i_{(2)}}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(2)}}\partial x_{k_{(2)}}}=0, \tag*{(52)} \end{aligned}
∂Hi(1)Θ(x(1),x(2))∂t+∂Hi(1)Θk(x(1),x(2),x(1))∂xk(1)+∂Hi(1)Θk(x(1),x(2),x(2))∂xk(2)+∂IPΘ(x(1),x(2))∂xi(1)−ν∂2Hi(1)Θ(x(1),x(2))∂xk(1)∂xk(1)−α∂2Hi(1)Θ(x(1),x(2))∂xk(2)∂xk(2)=0,(53)\begin{aligned} \frac{\partial H_{i_{(1)}\Theta}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial t} &+\frac{\partial H_{i_{(1)}\Theta k}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)},\boldsymbol{x}_{(1)})}{\partial x_{k_{(1)}}} +\frac{\partial H_{i_{(1)}\Theta k}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(2)}}}\\ &+\frac{\partial IP_{\Theta}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{i_{(1)}}}\\ &-\nu\frac{\partial^2 H_{i_{(1)}\Theta}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(1)}}\partial x_{k_{(1)}}} -\alpha\frac{\partial^2 H_{i_{(1)}\Theta}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(2)}}\partial x_{k_{(2)}}}=0, \tag*{(53)} \end{aligned}
∂HΘΘ(x(1),x(2))∂t+∂HΘΘk(x(1),x(2),x(1))∂xk(1)+∂HΘΘk(x(1),x(2),x(2))∂xk(2)−α∂2HΘΘ(x(1),x(2))∂xk(1)∂xk(1)−α∂2HΘΘ(x(1),x(2))∂xk(2)∂xk(2)=0.(54)\begin{aligned} \frac{\partial H_{\Theta\Theta}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial t} &+\frac{\partial H_{\Theta\Theta k}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)},\boldsymbol{x}_{(1)})}{\partial x_{k_{(1)}}} +\frac{\partial H_{\Theta\Theta k}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(2)}}}\\ &-\alpha\frac{\partial^2 H_{\Theta\Theta}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(1)}}\partial x_{k_{(1)}}} -\alpha\frac{\partial^2 H_{\Theta\Theta}(\boldsymbol{x}_{(1)},\boldsymbol{x}_{(2)})}{\partial x_{k_{(2)}}\partial x_{k_{(2)}}}=0. \tag*{(54)} \end{aligned}

References

References

  1. [2]J. Kim, P. Moin, and R. Moser, “Turbulence statistics in fully developed channels flows at low Reynolds numbers,” Journal of Fluid Mechanics 177, 133–166 (1987).
  2. [3]J. Kim and P. Moin, “Transport of passive scalars in a turbulent channel flow,” NASA 1, 1–14 (1987).
  3. [5]⁵S. Pirozzoli, M. Bernardini, and P. Orlandi, “Passive scalars in turbulent channel flow at high Reynolds number,” Journal of Fluid Mechanics 788, 614–639 (2016).
  4. [7]⁷S. Hoyas and J. Jiménez, “Reynolds number effects on the Reynolds-stress budgets in turbulent channels,” Physics of Fluids 20, 101511 (2008).DOI
  5. [8]⁸F. Alcántara-Ávila, S. Hoyas, and M. Pérez-Quiles, “Direct numerical simulation of thermal channel flow for Reτ = 5000 and Pr = 0.71,” Journal of Fluid Mechanics 916, A29 (2021).
  6. [9]⁹S. Hoyas and J. Jiménez, “Scaling of the velocity fluctuations in turbulent channels up to Reτ = 2003,” Physics of Fluids 18, 011702 (2006).DOI
  7. [10]¹⁰J. Jiménez and S. Hoyas, “Turbulent fluctuations above the buffer layer of wall-bounded flows,” Journal of Fluid Mechanics 611, 215–236 (2008).
  8. [11]¹¹V. Avsarkisov, S. Hoyas, M. Oberlack, and J. García-Galache, “Turbulent plane Couette flow at moderately high reynolds number,” Journal of Fluid Mechanics 751, R1 (2014).
  9. [12]¹²S. Pirozzoli, M. Bernardini, and P. Orlandi, “Turbulence statistics in Couette flow at high Reynolds number,” Journal of Fluid Mechanics 758, 323–343 (2014).
  10. [13]¹³M. Lee and R. Moser, “Direct numerical simulation of turbulent channel flow up to Reτ ≈ 5200,” Journal of Fluid Mechanics 774, 395–415 (2015).
  11. [14]¹⁴A. Lozano-Durán and J. Jiménez, “Effect of the computational domain on direct simulations of turbulent channels up to Reτ = 4200,” Physics of Fluids 26, 011702 (2014).
  12. [15]¹⁵S. Kraheberger, S. Hoyas, and M. Oberlack, “Dns of a turbulent couette flow at constant wall transpiration up to Reτ = 1000,” Journal of Fluid Mechanics 835, 421–443 (2018).
  13. [16]¹⁶S. Pirozzoli, “An explicit representation for mean profiles and fluxes in forced passive scalar convection,” Journal of Fluid Mechanics 968, R1 (2023).DOI
  14. [17]¹⁷M. Oberlack, “A unified approach for symmetries in plane parallel turbulent shear flows,” Journal of Fluid Mechanics 427, 299–328 (2001).DOI
  15. [19]¹⁹V. Avsarkisov, M. Oberlack, and S. Hoyas, “New scaling laws for turbulent Poiseuille flow with wall transpiration,” Journal of Fluid Mechanics 746, 99–122 (2014).
  16. [20]²⁰A. M. Rosteck, Scaling laws in turbulence – a theoretical approach using Lie-point symmetries, Ph.D. thesis, Darmstadt, Technische Universität (2014).DOI
  17. [21]M. Oberlack, M. Wacławczyk, A. Rosteck, and V. Avsarkisov, “Symmetries and their importance for statistical turbulence theory,” Mechanical Engineering Reviews 2, 15–00157 (2015).DOI
  18. [22]H. Sadeghi, M. Oberlack, and M. Gauding, “On new scaling laws in a temporally evolving turbulent plane jet using lie symmetry analysis and direct numerical simulation,” Journal of Fluid Mechanics 854, 233–260 (2018).
  19. [23]H. Sadeghi and M. Oberlack, “New scaling laws of passive scalar with a constant mean gradient in decaying isotropic turbulence,” Journal of Fluid Mechanics 775, A10 (2020).
  20. [24]H. Sadeghi, M. Oberlack, and M. Gauding, “On new scaling laws in a temporally evolving turbulent plane jet using lie symmetry analysis and direct numerical simulation-corrigendum,” Journal of Fluid Mechanics 885, E1 (2020).DOI
  21. [25]F. Alcántara-Ávila, S. Hoyas, and M. Pérez-Quiles, “DNS of thermal channel flow up to Reτ = 2000 for medium to low Prandtl numbers,” International Journal of Heat and Mass Transfer 127, 349–361 (2018).
  22. [26]Alcántara-Ávila, F. and Hoyas, S., “Direct Numerical Simulation of thermal channel flow for medium–high Prandtl numbers up to Reτ = 2000,” International Journal of Heat and Mass Transfer 176, 121412 (2021).
  23. [30]M. Oberlack, “Asymptotic expansion, symmetry groups, and invariant solutions of laminar and turbulent wall-bounded flows,” Zeitschrift für Angewandte Mathematik und Mechanik 80, 791–800 (2000).
  24. [32]M. Oberlack, S. Hoyas, J. Laux, and D. Klingenberg, “Oberlack et al. reply:,” Physical Review Letters 130, 069403 (2023).
  25. [33]S. Gandía-Barberá, S. Hoyas, M. Oberlack, and S. Kraheberger, “The link between the Reynolds shear stress and the large structures of turbulent Couette-Poiseuille flow,” Physics of Fluids 30, 041702 (2018), https://doi.org/10.1063/1.5028324.
  26. [34]34F. Lluesma-Rodríguez, S. Hoyas, and M. Peréz-Quiles, “Influence of the computational domain on DNS of turbulent heat transfer up to \(Re_\tau = 2000\) for \(Pr = 0.71\),” International Journal of Heat and Mass Transfer 122, 983–992 (2018).

Paper details

vlnc:W_bmrdc65d
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents