2.1. Numerical details

Following previous numerical studies (da Silva & Métais Reference da Silva and Métais2002; Hayashi, Watanabe & Nagata Reference Hayashi, Watanabe and Nagata2021), a hyperbolic-tangent function is employed to describe the vertical distribution of the initial mean streamwise velocity $U_{in}(Y)$, i.e.

(2.1)\begin{equation} U_{in}(Y)=\frac{U_J}{2}+\frac{U_J}{2} \tanh \left[\frac{H_J}{4 \theta_0}\left(1-\frac{2\lvert Y \rvert}{H_J}\right)\right],\end{equation}

where $U_J=1$ and $\theta _0/H_J=35$ with $\theta _0$ and $H_J$ being the initial momentum thickness and the width of the nozzle, respectively. The mean initial velocity in the two other directions is set to zero, i.e. $V_{in}(Y)= 0$ and $W_{in}(Y)= 0$. Throughout the paper, unless otherwise defined, the values of the mean velocity components and the corresponding fluctuating components are presented by uppercase and lowercase letters, respectively. The periodic boundary conditions are adopted in the two quasi-homogeneous directions (i.e. $X$ and $Z$ directions), whereas the free slip boundary condition is used in the vertical direction.

For an efficient transition to the self-similar/self-preserving state, artificially generated 3-D disturbances are superimposed onto the initial mean velocity field ($U_{in}$, $V_{in}$, $W_{in}$) within the vertical range $-1/2 \le Y/H_J \le 1/2$. The generation of the initial disturbances is based on the diffusion procedure proposed by Kempf, Klein & Janicka (Reference Kempf, Klein and Janicka2005). Relatively small-amplitude disturbances (i.e. the root-mean-square (r.m.s.) values of the velocity fluctuations in three directions are only $0.02 U_J$) be superposed onto the initial mean velocity field to allow a natural outward growth of the turbulent plane jet.

The computational details and the corresponding geometric parameters are listed in table 1. Here, the initial Reynolds number $Re_{J}=U_JH_J/\nu$, where $\nu$ is the kinematic viscosity, is set to 4000. This moderate Reynolds number is chosen to ensure a sufficiently fine spatial resolution and, consequently, accurately capture the instantaneous TNTI along with the nearby turbulent flow dynamics. For instance, the first-order and second-order derivatives of vorticity are required to compute the local turbulent entrainment (Holzner & Lüthi Reference Holzner and Lüthi2011). The lengths $L_X$, $L_Y$ and $L_Z$ denote the sizes of the computational domain along the streamwise, vertical and spanwise directions, respectively. Hereafter, $X$, $Y$ and $Z$ refer to the streamwise, vertical and spanwise directions, respectively. The size of the simulation domain is $L_X \times L_Y \times L_Z = 8 H_J \times 12 H_J \times 8 H_J$ and a total number of $N_X \times N_Y \times N_Z = 768 \times 1025 \times 768$ grid mesh points are used for the spatial discretisation. With respect to the finite-difference discretisation, a sixth-order central compact scheme with spectral-like resolution (Lele Reference Lele1992) is employed. A third-order Adams–Bashforth scheme is adopted for time advancement with $\Delta t / (H_{J}/U_{J})=0.0015$. It is worth mentioning that the size of the simulation domain and the corresponding mesh number are comparable to or even better than those used in Hayashi et al. (Reference Hayashi, Watanabe and Nagata2021). To avoid the normal boundary conditions affecting the evolution of the turbulent region, we carefully selected the box size along the normal direction, i.e. $L_Y=12H$. This box size is larger than the normal length used in previous studies (Watanabe et al. Reference Watanabe, da Silva, Nagata and Sakai2017; Silva, Zecchetto & da Silva Reference Silva, Zecchetto and da Silva2018; Hayashi et al. Reference Hayashi, Watanabe and Nagata2021) and is nearly twice that used by Silva et al. (Reference Silva, Zecchetto and da Silva2018). Further validation of the choice of the streamwise and spanwise simulation domains is confirmed by performing an additional DNS with a larger simulation domain, i.e. $L_X \times L_Y \times L_Z = 16 H_J \times 12 H_J \times 16 H_J$ and the corresponding number of grids is $N_X \times N_Y \times N_Z = 1536 \times 1025 \times 1536$ ($\simeq 2.4 \times {10^{ 9}}$). The numeral treatment of the additional DNS is the same as that with a small domain.

Table 1.

Computational and geometric details.

2.2. Simulation validation and spatial resolution

Figure 1 shows the vertical distributions of the normalised one-point statistics (i.e. the mean streamwise velocity $U$ and r.m.s. velocities $U_{rms}$, $V_{rms}$ and $W_{rms}$) at two different time steps ($T/T_{ref}=25.5$ and 36.0, where $T_{ref}=H_J/U_J)$. The corresponding results from previous numerical (i.e. da Silva & Métais Reference da Silva and Métais2002; da Silva & Pereira Reference da Silva and Pereira2008; Watanabe et al. Reference Watanabe, Sakai, Nagata, Ito and Hayase2014b; Hayashi et al. Reference Hayashi, Watanabe and Nagata2021) and experimental (i.e. Gutmark & Wygnanski Reference Gutmark and Wygnanski1976; Thomas & Prakash Reference Thomas and Prakash1991; Stanley, Sarkar & Mellado Reference Stanley, Sarkar and Mellado2002) investigations are included for comparison. It can be seen that our simulation results are in reasonably good accord with those in previous studies. Here, $U_C$ and $b_U$ denote the centreline mean velocity and the jet half-width, respectively. Hereafter, the subscript $C$ denotes a variable along the centreline and the half-width $b_U$ is defined as the vertical distance between the jet centreline and the location where $U(Y)/{U_C} = 0.5$.

Figure 1.

The distributions of the normalised mean flow velocity and r.m.s. velocities. For comparison, results from previous investigations (i.e. Gutmark & Wygnanski Reference Gutmark and Wygnanski1976; Thomas & Prakash Reference Thomas and Prakash1991; da Silva & Métais Reference da Silva and Métais2002; Stanley et al. Reference Stanley, Sarkar and Mellado2002; da Silva & Pereira Reference da Silva and Pereira2008; Watanabe et al. Reference Watanabe, Sakai, Nagata, Ito and Hayase2014b; Hayashi et al. Reference Hayashi, Watanabe and Nagata2021) are included.

To capture small-scale flow dynamics across the TNTI and the small-scale local entrainment velocity $v_n$, high-order derivatives (e.g. the first-order and second-order derivatives of the vorticity) need to be accurately solved. In this study, a high-order compact scheme is adopted, and the global Reynolds number is set to be moderately low. The current spatial resolution follows the suggestion by Laizet, Nedić & Vassilicos (Reference Laizet, Nedić and Vassilicos2015) that a spatial resolution less than $2\eta$ is necessary to have a correct reproduction of the strain-rate and rotation tensors when using the solver Incompact3d. As shown in figure 2(a), the normalised spatial resolution $(\Delta X \Delta Y \Delta Z)^{1/3}/\eta$ is always below 1.2, with $\Delta X$, $\Delta Y$ and $\Delta Z$ being the mesh sizes in the streamwise, vertical and spanwise directions, respectively, and $\eta$ is the Kolmogorov scale. Figure 2(b) shows that the spatial resolution $(\Delta X \Delta Y \Delta Z)^{1/3}/\eta$ at the TNTI detected by a wider range of vorticity magnitude ($1.0\times {10^{-5}}\le {\vert \boldsymbol \omega \vert }_{th} / {\vert \boldsymbol \omega \vert }_{max} \le 1.0\times {10^{-3}}$) is mostly smaller than 1.0 and the peaks of the probability distribution function (PDF) profiles are found around $(\Delta X \Delta Y \Delta Z)^{1/3}/\eta \simeq 0.45$. Further validation of the current spatial resolution is done by assessing the balance between the integral volume flux and the global volume flux, as will be demonstrated below.

Figure 2.

(a) Vertical distributions of the normalised spatial resolution $(\Delta X \Delta Y \Delta Z)^{1/3}/\eta$ at $T/T_{ref}=25.5$ and (b) the PDF distribution of the local resolution at the TNTI for three different thresholds ${\vert \boldsymbol \omega \vert }_{th} / {\vert \boldsymbol \omega \vert }_{max} = 1.0\times {10^{-5}}$ (red solid line), $1.6\times {10^{-4}}$ (cyan solid line) and $1.0\times {10^{-3}}$ (green solid line) with ${\vert \boldsymbol \omega \vert }_{max}$ being the maximum vorticity magnitude.

2.3. Evaluation of the self-similarity property

Figure 1(a) suggests that the vertical distribution of mean velocity is self-similar/self-preserving. To further evaluate the self-similar behaviour of the mean flow, the jet shape factor $F(t)$ (Hickey, Hussain & Wu Reference Hickey, Hussain and Wu2013) is computed (see figure 3). The jet shape factor $F(t)$ is the ratio of the displacement thickness $\delta _{J}=\int U(Y)/U_C \,{\rm d} Y$ to the momentum thickness $\theta _{J}=\int U(Y)^2/U_C^2 \,{\rm d} Y$, that is $F(t) = \delta _{J}(t) / \theta _{J}(t)$. The time evolution of $\delta _{J}$ and $\theta _{J}$ are also included in figure 3. The jet shape factor $F$ remains nearly constant, i.e. $F \simeq 1.37$ after $T/T_{ref}=10.0$.

Figure 3.

Time evolution of the shape factor $F(t)$, the displacement thickness $\delta _{J}(t)$ and the momentum thickness $\theta _{J}(t)$. The horizontal line indicates $F \simeq 1.37$.

The self-similar behaviour of the mean velocity and the corresponding constant jet shape factor $F$ have been confirmed in the previous discussion. The self-similar/self-preserving state can be further evaluated by the second-order statistics such as r.m.s. velocity (Almagro, García-Villalba & Flores Reference Almagro, García-Villalba and Flores2017). Figure 4 shows the vertical distributions of the normalised r.m.s. velocity components and the Reynolds stress as a function of $Y/b_U$. It can be seen that although the profiles are not smooth and exhibit oscillations to some extent, which is probably related to the insufficient data samples for a given time step, a reasonably good profile can still be obtained. A quantitative assessment of self-similarity is determined by calculating the area enclosed within the normalised profiles. Take, for instance, the integral function $S_{\boldsymbol {u}}$ for the normalised profiles of ${\boldsymbol {u}}_{rms}$ is defined by $S_{\boldsymbol {u}} = \int {{{\boldsymbol {u}}_{rms}} \,{\rm d}Y/{\boldsymbol {u}}_{rms}^{max}{b_U}}$. It is evident from figure 5 that the profiles evaluating the self-similar states become roughly constant for $T/T_{ref} >20.0$. The corresponding turbulent Reynolds number $Re_{\lambda }$ based on the Taylor microscale $\lambda$ remains nearly constant at approximately $Re_{\lambda } = {u_{rms}}\lambda /\upsilon \simeq 58.8$ for $T/T_{ref} >20.0$.

Figure 4.

Vertical distributions of the normalised (a) $U_{rms}$, (b) $V_{rms}$ (c) $W_{rms}$ and $\langle u'v' \rangle$ at different time steps, normalised by the peak values of the respective curves.

Figure 5.

Time evolution of the integrals, i.e. $S_{u}(t)$, $S_{v}(t)$, $S_{w}(t)$ and $S_{uv}(t)$.

3.1. Identification and surface area of the TNTI

Figure 6 shows the contours of the magnitude of the vorticity field $\vert \boldsymbol \omega \vert$ in a randomly chosen 2-D $X\unicode{x2013}Y$ plane at $T/T_{ref}=19.5$ and the isocontour lines corresponding to three different vorticity magnitudes (represented by solid lines) are also included. Here, ${\boldsymbol {\omega }}$ stands for the vorticity vector with $\omega _i=\varepsilon _{i j k} \partial u_k / \partial x_j$, where the subscripts $i = 1$, $2$ and $3$ represent $X$, $Y$ and $Z$ directions, respectively. Throughout this paper, the bold letters indicate vectors, and the symbol ‘$|\ |$’ represents the magnitude of a vector. The symbol $\boldsymbol \omega _{max}$ denotes the maximum magnitude of vorticity within the turbulent region for a particular time step. Based on the interface orientation relative to the mean streamwise velocity, the TNTI can generally be classified into three different types as conducted by Watanabe et al. (Reference Watanabe, Sakai, Nagata, Ito and Hayase2014b): leading edge (oriented towards the streamwise velocity), trailing edge (oriented opposite to the streamwise direction) and cross-streamwise edge (aligned with the mean flow or, equivalently, perpendicular to the cross-stream $Y\unicode{x2013}Z$ plane), as shown in figure 6.

Figure 6.

The logarithmic contours of the magnitude of vorticity at $T/T_{ref}=19.5$. The three solid lines correspond to ${\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t)= 1.0 \times {10^{ - 5}}$ (white one), $1.6 \times {10^{ - 4}}$ (black one) and $1.0 \times {10^{ - 3}}$ (blue one), respectively. Only part of the simulation domain is plotted.

As can be seen from figure 6, for a wide range of vorticity magnitudes $1.0\times {10^{-5}}\le {\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t) \le 1.0\times {10^{-3}}$, the turbulent regions surrounded by TNTI rarely change as it should be. Following the previous study by Zhou & Vassilicos (Reference Zhou and Vassilicos2017), the vorticity norm, which is based on the maximum vorticity magnitude $\vert \boldsymbol \omega \vert _{max}(t)$, is used to identify the TNTI. The dependence of the volume fraction on the threshold value and the derivative of the volume fraction with respect to the threshold are plotted in figure 7. A distinct plateau where $-{\rm{d}}{V_T}/{\rm{dlog}}_{10}({\vert \boldsymbol{\omega} \vert _{th}}) \simeq 0$ within the range $1.0\times {10^{-7}} \le {\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t) \le 1.0\times {10^{-3}}$ can be readily identified. In order to accurately identify the TNTI, it is crucial to carefully choose an appropriate vorticity threshold. Based on the joint probability density distribution of the normalised vorticity magnitude ${\vert \boldsymbol \omega \vert }_{th} / {\vert \boldsymbol \omega \vert }_{max}$ and vertical height $Y/b_U$ (not shown herein), we also confirm that the presence of numerical noise (Zhang, Watanabe & Nagata Reference Zhang, Watanabe and Nagata2018) within the range $1.0\times {10^{-7}} \le {\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t) \le 1.0\times {10^{-5}}$, despite the turbulence volume hardly changes with the threshold ${\vert \boldsymbol \omega \vert }_{th} / {\boldsymbol \omega }_{max}$. Therefore, any threshold selection within the range $1.0\times {10^{-5}} \le {\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t) \le 1.0\times {10^{-3}}$ can be employed to detect TNTI and similar results can be obtained. Hereafter, the threshold ${\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t) = 1.6 \times {10^{ - 4}}$ corresponding to the black line in figure 6 is adopted for the identification of the TNTI.

Figure 7.

(a) The volume fraction of the turbulent region with $\vert \boldsymbol \omega \vert \ge {\vert \boldsymbol \omega \vert }_{th}$ as a function of the threshold ${\vert \boldsymbol \omega \vert }_{th} / {\vert \boldsymbol \omega \vert }_{max}$ and (b) the derivative of the volume fraction with respect to the threshold. The vertical dashed line indicates ${\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t)= 1.6 \times {10^{-4}}$ and the shadowed areas correspond to the range $1.0\times {10^{-5}}\le {\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t) \le 1.0\times {10^{-3}}$.

The method introduced by Yurtoglu, Carton & Storti (Reference Yurtoglu, Carton and Storti2018), which employs a grid-based approach for computing implicitly defined surface integrals, is adopted to compute the surface area $A(t)$ and the integral volume flux $Q(t) = \int {{v_n}(t)} \,{\rm d}A(t)$, where ${v_n(t)}$ denotes the local entrainment velocity closely associated with the development of the plane jet. The method uses the divergence theorem along with the characteristics of surface normal vectors to transform the surface integral into a volume integral, i.e.

(3.1)\begin{equation} Q(t) = \int {{v_n}(t)} \,{\rm d}A(t) ={-}\sum_{I, J, K} (v_n)_{I, J, K} {\frac{\boldsymbol{\nabla} \boldsymbol{\chi}(|{\boldsymbol{\omega}} |)_{I, J, K} \boldsymbol{\cdot} \boldsymbol{\nabla} |{\boldsymbol{\omega}} |_{I, J, K}}{\sqrt{\boldsymbol{\nabla} |{\boldsymbol{\omega}} |_{I, J, K} \boldsymbol{\cdot} \boldsymbol{\nabla} |{\boldsymbol{\omega}} |_{I, J, K}}}} \Delta X \Delta Y \Delta Z, \end{equation}

where

(3.2)\begin{equation} \boldsymbol{\chi} (|{\boldsymbol{\omega}} |) = \left\{ \begin{array}{ll} 1, & |{\boldsymbol{\omega}} | \ge |{\boldsymbol{\omega}} {|_{th}},\\ 0, & |{\boldsymbol{\omega}} | < |{\boldsymbol{\omega}} {|_{th}}, \end{array} \right. \end{equation}

represents the occupancy function, which is based on the vorticity threshold $|{\boldsymbol {\omega }} {|_{th}}$. The indices $I$, $J$ and $K$ correspond to the grid nodes along the three directions in the computational coordinate system. The occupancy function $\boldsymbol {\chi }(|{\boldsymbol {\omega }}|)$ is numerically computed using a fourth-order central finite-difference scheme, ensuring conformity with the connection coefficients of Daubechies wavelets for genus 2. When the integrand function $v_n$ equals 1, the value of the surface integral is identical to the surface area $A(t)$, that is, $Q(t) = \int {\rm d}A(t) = A(t)$. The occupancy function $\boldsymbol {\chi } (|{\boldsymbol {\omega }} |)$ exhibits self-adaptation to the highly contorted surface of the TNTI, enabling us to accurately estimate the time evolution of surface area $A(t)$ and integral volume flux $Q(t) = \int {{v_n}(t)} \,{\rm d}A(t)$.

Figure 8 suggests that the time evolution of the normalised surface area $A(t)/(2L_XL_Z)$ and the corresponding time derivative ${\rm d}A(t)/{\rm d}t$. The surface area $A(t)$ of the TNTI closely related to the turbulent entrainment process increases rapidly through wrinkling and deformation for $T/T_{ref} \le 20.0$, which is attributed to the interactions of the multiscale vortex structures (da Silva & dos Reis Reference da Silva and dos Reis2011) during the turbulent transition. It is worth mentioning that the area $A(t)$ remains roughly constant within the whole self-similarity period, i.e. $A(t)/(2L_XL_Z) \simeq 2.12$ with ${\rm d}A(t)/{\rm d}t \simeq 0$ after $T/T_{ref} = 20.0$. The roughly constant surface area at $T/T_{ref} \ge 20.0$ is distinctly different from the slow expansion of the area of the passive scalar isosurface (Blakeley, Olson & Riley Reference Blakeley, Olson and Riley2022).

Figure 8.

Time evolution of the normalised surface area $A(t)$ of the TNTI. The surface area $A(t)$ is normalised by the initial surface area $2L_XL_Z$. The purple solid squares stand for the corresponding results from the simulation with a larger computational domain size. The two horizontal lines indicate $A(t) /(2L_XL_Z) \simeq 2.12$ and ${\rm d}A(t)/{\rm d}t \simeq 0$, respectively.

The surface area calculation method proposed by Yurtoglu et al. (Reference Yurtoglu, Carton and Storti2018) was originally developed for relatively smooth surfaces. Considering the highly distorted nature of the TNTI surface area, we further verified the accuracy of the calculation method using the open-source software ParaView. The verification results indicate that the maximum surface area error is only $0.58\,\%$ after $T/T_{ref} = 20.0$. Furthermore, the method proposed by Yurtoglu et al. (Reference Yurtoglu, Carton and Storti2018) has been successfully utilised by Blakeley et al. (Reference Blakeley, Olson and Riley2022) and Huang, Burridge & van Reeuwijk (Reference Huang, Burridge and van Reeuwijk2023) for the direct calculation of the surface area and entrainment flux. It is worth mentioning that compared with the algorithm embedded in ParaView, this method significantly reduces the computational workload by eliminating the need to remesh the surface grid.

It has been argued that the surface area of the TNTI is adjusted continuously by stretching until the integral volume flux $Q(t)=\int v_n(t) \, {\rm d} A(t)$ balances the integral scale entrainment flux (Holzner & Lüthi Reference Holzner and Lüthi2011; Van Reeuwijk & Holzner Reference Van Reeuwijk and Holzner2014). The current result clearly indicates that this is indeed the case in the developing period with $T/T_{ref} < 20.0$. However, the surface area remains nearly unchanged after $T/T_{ref} = 20.0$. A possible explanation is that the adjustment of the local entrainment velocity $v_n$ near the TNTI may be more closely related to the local mean curvature. The constant surface area for $T/T_{ref} \ge 20.0$ allows us to establish the scaling law of the mean entrainment velocity, as shall be discussed below.

The turbulent motion near the TNTI is highly inhomogeneous due to the influence of large-scale motion (Zecchetto & da Silva Reference Zecchetto and da Silva2021). One may argue that the roughly constant the surface area $A(t)$ may be caused by the effect of size of the computational domain, which inhibits the stretching of the surface area and suppresses the development of the turbulent plane jet. Therefore, to investigate the dependence of the area $A(t)$ on the size of the computational domain, we perform an additional DNS on a larger computational domain in the present study. The purple solid squares in figure 8 depict the temporal evolution of the surface area $A(t)$ after $T/T_{ref} = 20.0$ from the simulation with a larger domain. It can be seen that the corresponding area $A(t)$ is virtually the same, which further enhances the credibility of the obtained results. We confirm that there is a notable effect of the normal boundary condition on the evolution of the TNTI surface area $A(t)$ only after $T/T_{ref} =48$. The current study only considers the range of $0 \le T/T_{ref} \le 40$ to mitigate the effect of the normal boundary condition on the evolution of surface area.

3.2. Turbulent entrainment process of the TNTI

Following Holzner & Lüthi (Reference Holzner and Lüthi2011) and Wolf et al. (Reference Wolf, Lüthi, Holzner, Krug, Kinzelbach and Tsinober2012), the local entrainment velocity $v_n$ can be decomposed into two components with the enstrophy transport equation: contributions from viscous and inviscid effects, namely,

(3.3)\begin{equation} v_n ={-} \frac{{2{\omega _i}{S_{ij}}{\omega _j}}}{{|\boldsymbol{\nabla}| {\boldsymbol{\omega}} |^2 |}} - \frac{{2\upsilon {\omega _i}{\nabla^2}{\omega _i}}}{{|\boldsymbol{\nabla}| {\boldsymbol{\omega}} |^2 |}}= {v_n^{P}}+ {v_n^{vis}}, \end{equation}

where $v_n^{P}$ and $v_n^{vis}$ are related to the inviscid and viscous effects, respectively. Here, ${S_{ij}} = (\partial {u_i}/\partial {X_j} + \partial {u_j}/\partial {X_i})/2$ denotes the flow strain rates tensor.

The PDF distributions of the normalised local entrainment velocity $v_n$ and the two components $v_n^{P}$ and $v_n^{vis}$ for $T/T_{ref}=37.5$ are plotted in figure 9(a). The distribution of the inviscid component $v_n^{P}$ is symmetrical, leading to $\langle v_n^{P} \rangle \simeq 0$. The average value of $\langle v_n \rangle \simeq \langle v_n^{vis} \rangle$ is negative, corresponding to the propagation of the TNTI towards the non-turbulent region, accompanied by the growth of turbulent volume. It is well known that high-order derivatives of velocity can be highly spatially intermittent and the probability distribution is considerably departure from a normal distribution (Davidson Reference Davidson2004). In other words, extreme/rare events are manifest in statistics of high-order derivatives (Sreenivasan & Antonia Reference Sreenivasan and Antonia1997), which is the case for the viscous component $v_n^{vis}$. Figure 9(a) clearly suggests that the local entrainment process is mainly determined by the small-scale viscous effects, as suggested by Holzner & Lüthi (Reference Holzner and Lüthi2011). The small-scale viscous effects dominating the entrainment process are also found in a high-Reynolds-number axisymmetric jet (Mistry, Philip & Dawson Reference Mistry, Philip and Dawson2019).

Figure 9.

PDF distributions of (a) the normalised local entrainment velocity $v_n$ and the two components $v_n^{P}$ and $v_n^{vis}$ and (b) the three components of viscous term, $v_n^{T}$, $v_n^{N}$, and $v_n^{\varepsilon }$ for $T/T_{ref} = 37.5$. The entrainment velocity and its components are normalised by the Kolmogorov velocity ${v_\eta } = {(\varepsilon v)^{1/4}}$ of the jet centreline.

The viscous component $v_n^{vis}$ of the local entrainment velocity $v_n$ can be further decomposed into three parts: the local tangential diffusion term ${v_n^T}$, normal diffusion term ${v_n^N}$ and viscous dissipation term ${v_n^\varepsilon }$ (Holzner & Lüthi Reference Holzner and Lüthi2011; Dopazo et al. Reference Dopazo, Martin, Cifuentes and Hierro2018), namely,

(3.4)\begin{equation} v_n^{vis} ={-} \upsilon ( {\boldsymbol{\nabla} \boldsymbol{\cdot} \boldsymbol{n}} ) - \frac{{\upsilon ({\partial ^2}{|{\boldsymbol{\omega}}|^2}/{\partial ^2}X_n^2)}}{{\partial {|{\boldsymbol{\omega}}|^2}/\partial {X_n}}} + \frac{{2\upsilon \boldsymbol{\nabla} {\omega _i}:\boldsymbol{\nabla} {\omega_i}}}{{|\boldsymbol{\nabla} {|{\boldsymbol{\omega}}|^2}|}}= {v_n^T}+ {v_n^N}+ {v_n^\varepsilon}, \end{equation}

where $\boldsymbol {n} = \boldsymbol {\nabla } |{\boldsymbol {\omega }} |/|\boldsymbol {\nabla } |{\boldsymbol {\omega }} ||$ denotes the unit normal vector and $\partial |{\boldsymbol {\omega }}|^2 / \partial X_n= \boldsymbol {\nabla } |{\boldsymbol {\omega }}|^2 \boldsymbol {\cdot } \boldsymbol {n}$ indicates the gradient along the normal direction of the TNTI. The term $v_n^T$ is also referred to as the curvature term by introducing the local mean curvature $H$, i.e. $v_n^T=2 \upsilon H$ (Van Reeuwijk & Holzner Reference Van Reeuwijk and Holzner2014). This implies that the contribution of curvature to the entrainment velocity $v_n$ is linear, with the slope corresponding to the kinematic viscosity $\upsilon$ of the fluid. Wolf et al. (Reference Wolf, Holzner, Lüthi, Krug, Kinzelbach and Tsinober2013) demonstrated that the local entrainment velocity depends strongly on the geometry of the TNTI surface, with the detrainment process (where $v_n > 0$) being more likely to manifest in a concave shape. An extended introduction of the local mean curvature $H$ along with an in-depth analysis of surface shape shall be presented below. The viscous dissipation term ${v_n^\varepsilon }$ always takes a positive value according to the above definition, which means the effects of ${v_n^\varepsilon }$ correspond to the inwards movement of TNTI towards the turbulent region.

The PDF distributions of the three components are given in figure 9(b). Comparing figures 9(a) and 9(b) shows that the normal diffusion term $v_n^N$ plays a dominant role in the outwards growth of the TNTI. The tangential viscous term $v_n^T$ also contributes to the outwards growth of the TNTI, albeit to a far lesser extent. In contrast, the effects of the viscous dissipation term ${v_n^\varepsilon }$ always suppress the spreading of the TNTI, as it should be. It can be seen from figure 9(b) that when compared with the velocity caused by tangential viscous term ${v_n^T}$ (or, equivalently, the mean curvature $H$), the molecular transport velocity in the normal direction exhibits a much wider range of scales accompanied by intense events. It is worth noting that although only a single threshold at one snapshot $T/T_{ref}=37.5$ is shown in figure 9, we confirm that the reported findings can be applied to other thresholds (e.g. ${\vert \boldsymbol \omega \vert }_{th}(t) / {\vert \boldsymbol \omega \vert }_{max}(t) = 1.0 \times {10^{ - 5}}$) and time steps (e.g. $T/T_{ref}=22.5$). Therefore, our conclusion is expected to be robust, at least for a fully developed temporally evolving turbulent plane jet (not shown herein for economy of space).

The instantaneous entrainment flux $Q(t)$ can be computed directly by integrating the local entrainment velocity $v_{n}$ over the surface of the TNTI (Krug et al. Reference Krug, Holzner, Marusic and van Reeuwijk2017; Zhou & Vassilicos Reference Zhou and Vassilicos2017), which is given by

(3.5)\begin{equation} Q(t)={-}\frac{{{\rm d}V_{J}(t)}}{{{\rm d}t}} = \int {{v_n}(t){\rm d}} A(t) = A(t)\langle {{v_n}(t)} \rangle ,\end{equation}

where ${V_{J}(t)}$ denotes the turbulent volume, and the operator $\langle \ \rangle$ represents a surface-area weighted average over the top and bottom surfaces of the TNTI. The mean entrainment velocity $\langle v_{n}(t) \rangle$ is determined by the local entrainment velocity $v_{n}$ and the surface area $A$.

Following (3.3), we have $Q(t) = \int {{v_n^{P}(t)\,{\rm d}} A(t) + \int v_n^{vis}(t)\,{\rm d}} A(t)=Q^P(t)+Q^{vis}(t)$. Figure 10(a) shows the time evolution of the integral volume flux $Q(t)$ and its two components $Q^P(t)$ and $Q^{vis}(t)$. As expected, the integral volume flux is mainly determined by the viscous effects. The temporal evolution of the integral volume flux $Q(t)$ also indicates a continuous decay of the jet growth rate over time. In figure 10(b), we plot a quantitative description of the balance between the integral volume flux $Q(t)=\int v_n(t) \,{\rm d} A(t)$ and the global volume flux $Q_0(t) = - {\rm d} V_J(t)/{\rm d} t$. The ratio $Q(t)/Q_0(t)$ is approximately 1.0 and for the worst case, the deviation is within 8 %. Note that the entrainment velocity $v_n$, as a small-scale variable being comparable to the Kolmogorov velocity, involves the calculation of both the second- and third-order derivatives of the velocity. The good balance between $Q(t)$ and $Q_0(t)$ further confirms the accuracy of the computation of small-scale variables.

Figure 10.

Time evolution of (a) the integral volume flux $Q(t)$ and its two components $Q^P(t)$ and $Q^{vis}(t)$ and (b) the ratio of the integral volume flux to the global volume flux, $Q(t)/Q_0(t)$ (presented by red solid squares).

3.3. Scaling law of the plane jet flow

The expansion of the turbulent volume can be approximately estimated by the growth rate of jet half-width $b_U$, that is, $V_{J}(t) = 2k{L_X}{L_Z}b_U(t)$, where $k$ denotes a dimensionless constant coefficient. The time independence verification of the dimensionless constant coefficient $k=({\rm d}V_J/{\rm d}t)/(2L_XL_Z\,{\rm d}b_U/{\rm d}t)$ is demonstrated in figure 11. The coefficient $k$ remains roughly constant with a negligibly weak oscillation, which is in reasonably good agreement with the results of Er et al. (Reference Er, Laval and Vassilicos2023). Moreover, they also verified that the value of coefficient $k$ is almost independent of the selected vorticity threshold. This indicates that the turbulent volume can thus be given as

(3.6)\begin{equation} V_{J}(t) \sim {L_X}{L_Z}{b_U}(t).\end{equation}

By applying self-similarity analysis to the Reynolds stress and the average turbulent kinetic energy transport equations, combined with the turbulence dissipation scaling law, the following relations are obtained for the jet half-width $b_{U}$ and the mean velocity along the centreline $U_{C}$:

(3.7)\begin{gather} {b_U}(t) \sim {({H_J}{U_J})^{1/2}}{t^{1/2}}, \end{gather}

(3.8)\begin{gather}{U_C}(t) \sim {({H_J}{U_J})^{1/2}}{t^{{-}1/2}}. \end{gather}

More detailed explanations can be found in the works of Ewing et al. (Reference Ewing, Frohnapfel, George, Pedersen and Westerweel2007) and Er et al. (Reference Er, Laval and Vassilicos2023). One interesting finding is that the scaling law ${b_U}(t) \sim {({H_J}{U_J})^{1/2}}{t^{1/2}}$ can also be found in compressible temporally evolving plane jets (Nagata, Watanabe & Nagata Reference Nagata, Watanabe and Nagata2018). Note that we assume the virtual origin to be located at the coordinate origin, i.e. the scaling laws of self-similar/self-preserving free shear turbulent flow.

Figure 11.

Time evolution of dimensionless constant coefficient $k=({\rm d}V_J/{\rm d}t)/(2L_XL_Z\,{\rm d}b_U/{\rm d}t)$. The horizontal dotted line represents $k \simeq 1.44$.

Figure 12 shows the time evolution of the square of the jet half-width $b_{U}^2$ along with the product $b_U U_C$. Figure 12(a) demonstrates that the square of the jet half-width $b_{U}^2$ obtains a well-defined 1 power-law scale for $T/T_{ref}\ge 20$. The approximately unchanged $b_{U}U_{C}$ shown in figure 12(b) is consistent with the theoretical predictions mentioned above. This will provide a theoretical basis for establishing the scaling of the mean entrainment velocity in the subsequent analysis.

Figure 12.

Time evolution of (a) the square of the jet half-width $b_U^2$ and (b) the product of the centreline velocity $U_C$ and the half-width $b_U$, $b_U U_C$.

Using (3.5) for the surface area of the TNTI with (3.6)–(3.8), the following relation is obtained for the surface area of the TNTI:

(3.9)\begin{equation} A(t) \sim \frac{{{L_X}{L_Z}{H_J}{U_J}}}{{{b_U}(t)\langle {{v_n}(t)} \rangle }}. \end{equation}

Figure 8 suggests the surface area $A(t)$ remains roughly constant, i.e. $A(t) = {\rm const.}$ This implies that the product of ${b_U}(t)$ and $\langle {{v_n}(t)} \rangle$ remains approximately constant, i.e. ${b_U}(t)\langle {{v_n}(t)} \rangle = {\rm const.}$ (see figure 13). By combining (3.7), the scaling relationship for the mean entrainment velocity in a temporally evolving plane jet can be expressed as

(3.10)\begin{equation} \langle {{v_n}(t)} \rangle \sim {( {{H_J}{U_J}} )^{{1}/{2}}}{t^{ - {1}/{2}}}.\end{equation}

This indicates that the mean entrainment velocity $\langle v_n \rangle$ is inversely proportional to the square root of time. It is worth mentioning that the above derivation is not affected by the choice of the dissipation scaling law. In other words, both the classical dissipation scaling law and the non-equilibrium dissipation scaling law lead to the same scaling of the mean entrainment velocity (Vassilicos Reference Vassilicos2015, Reference Vassilicos2023). The constant TNTI surface area and the scaling law for the mean entrainment velocity are also valid for a higher Reynolds number, i.e. $Re_H=8000$ (see the Appendix).

Figure 13.

Time evolution of $b_U \langle v_n \rangle /(H_JU_J)$ for $T/T_{ref}\ge 20$.

Quite recently, the same scaling law of the mean entrainment velocity in a temporally evolving plane jet has been derived by Er et al. (Reference Er, Laval and Vassilicos2023). It is worth mentioning that our analysis framework is based on the direct computation of the surface area $A(t)$, which is distinctly different from the approach in Zhou & Vassilicos (Reference Zhou and Vassilicos2017) and Er et al. (Reference Er, Laval and Vassilicos2023). Their approach involves the fractal or fractal-like properties of the TNTI and the use of the Corrsin length $\eta _I \sim \nu /\langle v_n \rangle$. In this study, our derivation does not require the two assumptions above.

In summary, the present analysis derives from the results of the temporally developing jet based on mass conservation, i.e. $b_U (t) U_C (t)=const$ (see figure 12b). Considering different flow types, such as temporally evolving flows and spatially evolving flows, is expected to result in distinctly different growth behaviours of turbulent shear flows (Nedić et al. Reference Nedić, Vassilicos and Ganapathisubramani2013; Cafiero & Vassilicos Reference Cafiero and Vassilicos2019; Er et al. Reference Er, Laval and Vassilicos2023). This means that applying similar analytical methods to spatially evolving flows that preserve momentum flux conservation (i.e. $b_U (x) U^2_C (x)={\rm const.}$) requires careful consideration.

4.1. Time evolution of the surface area of the TNTI

It still remains to be seen what mechanism is responsible for the approximately constant surface $A(t)$ for $T/T_{ref}\ge 20.0$. The temporal evolution of a non-material infinitesimal element of area $\delta A$, derived by Phillips (Reference Phillips1972), has a great degree of generality and is applicable to the evolution of isosurfaces in many physical fields. In this study, the equation is used to explore the mechanisms responsible for the production and destruction of the TNTI, i.e.

(4.1)\begin{equation} \frac{1}{\delta A(t)} \frac{\mathrm{d} \delta A(t)}{\mathrm{d} t}=(\delta_{i j}-n_i n_j) S_{i j}+ v_n \boldsymbol{\nabla} \boldsymbol{\cdot} \boldsymbol{n}= {S_{{\boldsymbol{\omega}}}}+ {H_{{\boldsymbol{\omega}}}}.\end{equation}

The stretching term ${S_{{\boldsymbol {\omega }}}}= (\delta _{i j}-n_i n_j) S_{i j}$ is associated with flow tangential strain rate. Here, $\delta _{i j}$ is the Kronecker delta, and $n_i$ being the $i$th component of the unit normal vector $\boldsymbol {n}$ of the surface of the TNTI. For incompressible flow, we have $(\delta _{i j}-n_i n_j) S_{i j}+n_i n_j S_{i j}=\partial {u_i}/\partial {X_i} =0$.

The curvature/propagation term ${H_{{\boldsymbol {\omega }}}}=v_n \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ represents the combined influences of the local entrainment velocity $v_n$ and the corresponding surface curvature $H$, which can also be further written as a function of the mean curvature $H$: ${H_{{\boldsymbol {\omega }}}}=v_n \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n} = 2v_n H = v_n(\kappa _1 + \kappa _2)$. The two principal curvatures $\kappa _1$ and $\kappa _2$ are the two non-zero eigenvalues of the curvature tensor $n_{i,j}$, i.e.

(4.2)\begin{equation} n_{i, j}=\frac{1}{\partial |{\boldsymbol{\omega}}| / \partial X_n}(\delta_{i k}-n_i n_k) \frac{\partial^2 |{\boldsymbol{\omega}}|}{\partial X_j \partial X_k }, \end{equation}

where $\partial |{\boldsymbol {\omega }}| / \partial X_n= \boldsymbol {\nabla } |{\boldsymbol {\omega }}| \boldsymbol {\cdot } \boldsymbol {n}$ denotes the gradient along the normal direction of the TNTI. Throughout this study, the local mean curvature $H= (\kappa _1 + \kappa _2)/2$ is negative for bulges with convex shapes and positive for valleys with concave shapes. The mean curvature $H$ equals 0 for a flat surface, resulting in ${H_{{\boldsymbol {\omega }}}} = 0$.

The time evolution of the average mean curvature $\langle {H} \rangle$ of the surface for $T/T_{ref} \ge 20.0$ and the PDF distribution of $H$ for $T/T_{ref} = 37.5$ are shown in figure 14. Here, the average mean curvature is calculated by $\langle H \rangle = (1/2A)\int {({\kappa _1} + {\kappa _2})}\, {\rm d}A$. There is a gradual and consistent decrease in the average value of $H$, corresponding to an increase in the characteristic length scale of the TNTI. Although the decrease of the average mean curvature $\langle {H} \rangle$ suggests that the turbulent motion causes continuous deformation of the surface, the total surface area of the TNTI remains nearly constant. The PDF profile of $H$ is not symmetrical and shows a left-skewed distribution resulting in a positive average value and thus a prevalence of concave shapes. Figure 15 clearly indicates that the 3-D contour of the TNTI is strongly convoluted with multiscale bulges/valleys. The large positive values of $S_{{\boldsymbol {\omega }}}$ and $H_{{\boldsymbol {\omega }}}$ (represented by the red colour) are more likely to be found at the leading edges. The local surface area $\delta A(t)$ of the leading edge is more likely to expand, despite the fact that, as is shown below, the surface area $\langle A(t) \rangle$ remains relatively constant. This observation was also reported by Neamtu-Halic et al. (Reference Neamtu-Halic, Krug, Mollicone, van Reeuwijk, Haller and Holzner2020), somewhat echoing the argument by Watanabe et al. (Reference Watanabe, Sakai, Nagata, Ito and Hayase2014b) that the vortex stretching and compression near the TNTI are significantly influenced by the interface orientations. Another less noted but likely equally important observation is that figure 15 clearly indicates the spatial distribution of $H_{{\boldsymbol {\omega }}}$ appears significantly more intermittent to that of $S_{{\boldsymbol {\omega }}}$. One may conclude that the stretching term ${S_{{\boldsymbol {\omega }}}}$ is related to the large-scale turbulence and ${H_{{\boldsymbol {\omega }}}}$ are more likely to be related to the small-scale motions.

Figure 14.

(a) Time evolution of the average mean curvature $\langle {H} \rangle$ of the surface for $T/T_{ref} \ge 20.0$ and (b) PDF distribution of the mean curvature $H$ for $T/T_{ref} = 37.5$. Note that the left- and right-hand sides of the vertical dotted line represent convex and concave shapes, respectively.

Figure 15.

Top view of the TNTI (i.e. isosurfaces of the vorticity magnitude with $\vert \boldsymbol \omega \vert _{th}(t)/\vert \boldsymbol \omega \vert _{max}(t)= 1.6\times {10^{-4}}$). The colour of the TNTI represent (a) the value of the strain term $S_{{\boldsymbol {\omega }}}$ and (b) the value of the curvature/propagation term $H_{{\boldsymbol {\omega }}}$ for $T/T_{ref} = 37.5$.

Figure 16 shows the PDF distributions of the stretching term ${S_{{\boldsymbol {\omega }}}}$, the curvature/propagation term ${H_{{\boldsymbol {\omega }}}}$ and the unsteady term $((1/\delta A(t)){\rm d}\delta A(t)/{\rm d}t={H_{{\boldsymbol {\omega }}}}+{S_{{\boldsymbol {\omega }}}})$ at $T/T_{ref}=37.5$. The PDF distributions of ${H_{{\boldsymbol {\omega }}}}$ and $S_{{\boldsymbol {\omega }}}$ are distinctly different even though their average influences counterbalance each other (i.e. $\langle S_{{\boldsymbol {\omega }}} \rangle = 0.068$ and $\langle H_{{\boldsymbol {\omega }}} \rangle = -0.074$). The profile of ${H_{{\boldsymbol {\omega }}}}$ exhibits significant fat tails compared to that of ${S_{{\boldsymbol {\omega }}}}$, implying the existence of intense/extreme events in small-scale turbulence (Sebastien & Thierry Reference Sebastien and Thierry1990). The fat tails of local entrainment velocity $v_n$ (see figure 9) imply that the local entrainment velocity $v_n$ exhibits a large range of values but with a mean negative value which is of the order of the Kolmogorov scale. Taking into account the right-skewed PDF of the mean curvature $H$, the intermittent and extensive events of ${H_{{\boldsymbol {\omega }}}}$ are somehow related to the fact. We further confirm that the correlation efficiency between the local mean curvature ${H}$ and the entrainment velocity $v_n$ is considerably small. It is thus not surprising that extreme/intense events are more likely to be related to the curvature/propagation term. The above discussions imply the intrinsic properties of ${H_{{\boldsymbol {\omega }}}}$ and ${S_{{\boldsymbol {\omega }}}}$ are distinctly different from each other.

Figure 16.

PDF distributions of ${H_{{\boldsymbol {\omega }}}}$, ${S_{{\boldsymbol {\omega }}}}$ and the unsteady term ${H_{{\boldsymbol {\omega }}}}$+ ${S_{{\boldsymbol {\omega }}}}$ at $T/T_{ref}=37.5$.

Similar extreme events can also be found in the unsteady term, which is manifest in the fat tails of the distribution of ${H_{{\boldsymbol {\omega }}}}+ {S_{{\boldsymbol {\omega }}}}$. This implies that small-scale areas of the TNTI can experience intense changes, either through production or destruction effects. The rare events involving intense area changes arise from the non-Gaussian distributions of the entrainment velocity $v_n$ and the mean curvature $H$.

When taking an averaging over the TNTI, (4.1) becomes (Neamtu-Halic et al. Reference Neamtu-Halic, Krug, Mollicone, van Reeuwijk, Haller and Holzner2020)

(4.3)\begin{equation} \frac{1}{A(t)} \frac{{\rm d}A(t)}{{\rm d}t}=\langle {{S_{{\boldsymbol{\omega}}} }} \rangle +\langle {{H_{{\boldsymbol{\omega}}} }} \rangle,\end{equation}

where the operator $\langle \ \rangle$ denotes an average over the top and bottom surfaces of the TNTI. It can be seen that the average production/destruction of the TNTI area $A(t)$ stems from the combined effects of $\langle {S_{{\boldsymbol {\omega }}}} \rangle$ and $\langle {H_{{\boldsymbol {\omega }}}} \rangle$. Obviously, if the sum of the average strain and average curvature/propagation terms, i.e. $\langle {S_{{\boldsymbol {\omega }}}} \rangle + \langle {H_{{\boldsymbol {\omega }}}} \rangle$ is positive (negative), the surface area of the TNTI will increase (decrease) with time. One can reasonably expect that for $T/T_{ref} \ge 20$ the average unsteady term ${\rm d}A(t)/(A(t)\,{\rm d}t)$ is relatively small (or equal zero), the contributions of the flow tangential strain rate $S_{{\boldsymbol {\omega }}}$ and the effect of the local mean curvature $H_{{\boldsymbol {\omega }}}$ balance each other.

The time evolution of the three terms in (4.3), i.e. ${\rm d}A/(A\,{\rm d}t)$, $\langle {S_{{\boldsymbol {\omega }}}} \rangle$ and $\langle {H_{{\boldsymbol {\omega }}}} \rangle$ for $T/T_{ref} \ge 20$ are plotted in figure 17. The residual term $\langle \mathcal {R} \rangle$, which is the difference between the left- and right-hand sides of (4.3), that is $\langle \mathcal {R} \rangle =\langle {S_{{\boldsymbol {\omega }}}} \rangle + \langle {H_{{\boldsymbol {\omega }}}} \rangle - {\rm d}A/(A\,{\rm d}t)$ is also included for comparison. The residual term $\langle \mathcal {R} \rangle$ is considerably small, which further implies that the entrainment process across the TNTI and the surface area are captured accurately. As expected, there is an approximately instantaneous balance between $\langle {S_{{\boldsymbol {\omega }}}} \rangle$ and $\langle {H_{{\boldsymbol {\omega }}}} \rangle$ and both magnitudes appears to continuously decay with time, similar to the observations of Neamtu-Halic et al. (Reference Neamtu-Halic, Krug, Mollicone, van Reeuwijk, Haller and Holzner2020) in turbulent flows with or without stable stratification. The positive values of the term $\langle {S_{{\boldsymbol {\omega }}}} \rangle$ indicate that the flow inhomogeneity always contributes to the growth of the surface area $A(t)$. In contrast, the negative $\langle {H_{{\boldsymbol {\omega }}}} \rangle$ term is related to the destruction of the surface area $A(t)$. The decrease in the magnitude of the stretching term $\langle {S_{{\boldsymbol {\omega }}}} \rangle$ can find its roots in the fact that the mean velocity profile becomes flat with time evolution. Moreover, one could reasonably argue that the negative value of $\langle {H_{\boldsymbol {\omega }}} \rangle$ arises due to the overall concave shapes exhibiting $\langle H \rangle > 0$ (see figure 14) and the spreading of the planar jet with $\langle v_n \rangle < 0$ (see figure 9a). Although on average the strain and curvature/propagation terms counterbalance each other, resulting in an approximately constant TNTI area for $T/T_{ref} \ge 20.0$, their intrinsic properties are distinctly different. The local effects of the curvature/propagation term are highly spatially intermittent with small-scale extreme/intense events, whereas the effects of the large-scale stretching term are more continuous.

Figure 17.

Time evolution of the two average terms $\langle {S_{{\boldsymbol {\omega }}}} \rangle$ and $\langle {H_{{\boldsymbol {\omega }}}} \rangle$ involved in the growth of surface area for $T/T_{ref} \ge 20$. For reference, the residual term $\langle \mathcal {R} \rangle =\langle {S_{{\boldsymbol {\omega }}}} \rangle + \langle {H_{{\boldsymbol {\omega }}}} \rangle - {\rm d}A/(A{\rm d}t)$ is also included.

4.2. Contributions of the viscous and inviscid effects and curvature effects

The large-scale motions are associated with the cumulative structure of the surface area, while the curvature/propagation term ${H_{\boldsymbol {\omega }}}$ related to small-scale turbulent motions plays a more significant role in the production/destruction of the surface area, as reported by Catrakis, Aguirre & Ruiz-Plancarte (Reference Catrakis, Aguirre and Ruiz-Plancarte2002). Two of the key quantities effecting the production/destruction of the area $A(t)$ are the local entrainment velocity $v_n$ and the mean curvature $H$ considering the observation $A(t)= {\rm const.}$ The effect of nearby coherent structures on the evolution of the TNTI surface area in a turbulent flow with and without stable stratification was recently studied by Neamtu-Halic et al. (Reference Neamtu-Halic, Krug, Mollicone, van Reeuwijk, Haller and Holzner2020). However, in their study, the entrainment velocity was determined using an interface tracking technique, similar to the method employed by Wolf et al. (Reference Wolf, Lüthi, Holzner, Krug, Kinzelbach and Tsinober2012). The application of the interface tracking technique makes the direct decomposition of the local entrainment velocity impossible, thereby preventing comparisons of viscous and inviscid contributions or exploration of their relationship to the TNTI geometry. It is shown that the local entrainment velocity $v_n$ somewhat depends on the shape of the TNTI. The viscous effects are predominant for convex surfaces, whereas the inviscid effect accompanied by vortex stretching plays a more critical role for concave surfaces (Wolf et al. Reference Wolf, Lüthi, Holzner, Krug, Kinzelbach and Tsinober2012, Reference Wolf, Holzner, Lüthi, Krug, Kinzelbach and Tsinober2013). One may expect that the surface change depends on the shape of the TNTI. To take the observation $\langle {H_{\boldsymbol {\omega }}} \rangle < 0$ one step further, the contributions of the viscous and inviscid effects and curvature effects are explored.

Using (3.3) for the local entrainment velocity $v_n$, the curvature/propagation term $\langle {{H_{{\boldsymbol {\omega }}} }} \rangle$ in (4.3) can be further expressed as

(4.4)\begin{equation} \langle {{H_{{\boldsymbol{\omega}}} }} \rangle= \langle v_n^{P} \boldsymbol{\nabla} \boldsymbol{\cdot} \boldsymbol{n}\rangle+ \langle v_n^{vis} \boldsymbol{\nabla} \boldsymbol{\cdot} \boldsymbol{n}\rangle, \end{equation}

which enables us to distinguish between the contributions of the viscous and inviscid components to the growth of the surface area. By substituting (3.4) into (4.4), we obtain

(4.5)\begin{equation} \langle {{H_{{\boldsymbol{\omega}}} }} \rangle=\langle v_n^P \boldsymbol{\nabla} \boldsymbol{\cdot} \boldsymbol{n}\rangle+\langle v_n^T \boldsymbol{\nabla} \boldsymbol{\cdot} \boldsymbol{n}\rangle+\langle v_n^N \boldsymbol{\nabla} \boldsymbol{\cdot} \boldsymbol{n}\rangle+\langle v_n^{\varepsilon} \boldsymbol{\nabla} \boldsymbol{\cdot} \boldsymbol{n}\rangle,\end{equation}

where $v_n^T\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$, $v_n^N\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and $v_n^\varepsilon \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ represent the contributions of the curvature/tangential diffusion term, the curvature/normal diffusion term together with the curvature/viscous dissipation term to the surface area of TNTI, respectively. It can be seen that the contributions of $\langle v_n^T \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle$ to the surface area are always negative given by $\langle v_n^T \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle = - \upsilon ( {\boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}} )^2= - 4\upsilon H^2$, and $\langle v_n^{\varepsilon } \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle$ to the surface area of TNTI depend entirely on the geometric characteristics of the interface. With respect to dissipation effects, as $v_n^{\varepsilon }$ consistently remains negative ($v_n^{\varepsilon } > 0$), $v_n^\varepsilon \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ is negative (positive) for convex (concave) shapes.

The two components of the average curvature/propagation term $\langle {{H_{{\boldsymbol {\omega }}} }} \rangle$, i.e. the curvature/inviscid term $\langle v_n^{P} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle$ and the curvature/viscous term $\langle v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle$ as well as the time evolution of the stretching term $\langle {S_{{\boldsymbol {\omega }}}} \rangle$ are shown in figure 18. The average term $\langle v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle$ plays a dominant role in the destruction of the surface area of TNTI, i.e. $\langle {{H_{{\boldsymbol {\omega }}} }} \rangle \simeq \langle v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle$. In contrast, the average term $\langle v_n^{P} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n} \rangle$ remains virtually zero, implying $\langle v_n^{P} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n} \rangle$ has a negligibly small effect on the evolution of the surface area. Previous studies (da Silva et al. Reference da Silva, Hunt, Eames and Westerweel2014; Watanabe et al. Reference Watanabe, Sakai, Nagata, Ito and Hayase2014b) have already shown that the viscous diffusion effect near the TNTI dominates the enstrophy transport, and inviscid processes associated with vortex stretching and compression can be neglected.

Figure 18.

Time evolution of the average stretching term $\langle {S_{{\boldsymbol {\omega }}}} \rangle$ and the two components $\langle v_n^{P} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle$ and $\langle v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle$ for $T/T_{ref} \ge 20.0$.

Figure 19(a) shows the PDF distributions of the two components of ${H_{{\boldsymbol {\omega }}}}$, $v_n^P\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and $v_n^{vis}\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ for $T/T_{ref}=37.5$. The PDF distribution of $v_n^P\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ is approximately symmetrical with the peak value being around zero. The profile of $v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ is similar to that of term ${H_{{\boldsymbol {\omega }}}}$, further confirming that the contribution of inviscid effects to the surface change can be somewhat neglected.

Figure 19.

PDF distributions of (a) $v_n^P\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and $v_n^{vis}\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and (b) $v_n^T\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$, $v_n^N\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and $v_n^\varepsilon \boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ for $T/T_{ref} = 37.5$.

To further investigate the viscous processes on the surface change, we decompose the curvature/viscous term $v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ into three components, i.e. the curvature/tangential diffusion term $v_n^{T} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$, the curvature/normal diffusion term $v_n^{N} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ and the curvature/viscous dissipation term $v_n^{\varepsilon } \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$. The PDF distributions of the three components of $v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ are shown in figure 19(b). The always negative values of $v_n^{T} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ are expected considering its definition. The PDF distributions of $v_n^{N} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ and $v_n^{\varepsilon } \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ exhibit significant differences. In terms of the destruction of surface area, $v_n^{N} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ have a higher probability for the large negative values compared with $v_n^{\varepsilon } \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$. The average values of $v_n^{T} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$, $v_n^{N} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ and $v_n^{\varepsilon } \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ are $- 0.45$, $- 0.94$ and $1.32$, respectively. One could also reasonably draw the conclusion that on average $v_n^{T} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ and $v_n^{N} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ lead to the destruction of surface area, whereas $v_n^{\varepsilon } \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$ contributes to the growth of surface area. Meanwhile, the PDF profiles of all three components consistent with ${H_{{\boldsymbol {\omega }}}}$ demonstrate fat tails, implying the existence of intense/extreme events at the TNTI.

In light of the above discussion, the geometric properties of the surface are intimately related to the local production/destruction of the surface area. We further calculate the conditional statistics (in relation to the local mean curvature $H$) involved in the production/destruction of the surface. Here, for any variable $f$, the statistics conditional $\langle \,{f \mid H} \rangle$ are weighted based on the local mean curvature $H$, namely, $\langle \,{f \mid H} \rangle = \int {P(f\mid H)} f\, {\rm d}f,$ where ${P(f\mid H)}$ represents the local probability distribution of the conditional variable $H$. The conditional average statistics on the local shapes are presented below.

Figure 20 shows the conditional averages of $\langle S_{{\boldsymbol {\omega }}}\mid H \rangle$, $\langle H_{{\boldsymbol {\omega }}}\mid H \rangle$ and $\langle {H_{{\boldsymbol {\omega }}}}+{S_{{\boldsymbol {\omega }}}}\mid H \rangle$. It is evident that in terms of conditional averages, the curvature/propagation term $\langle H_{{\boldsymbol {\omega }}} \rangle$ is dominant, albeit $\langle H_{{\boldsymbol {\omega }}} \rangle$ and $\langle S_{{\boldsymbol {\omega }}} \rangle$ counterbalance each other. The stretching effects $\langle S_{{\boldsymbol {\omega }}} \rangle$ produce the surface area for both bulge surfaces with $H<0$ and valley surfaces with $H>0$, albeit with a relatively smaller contribution. In contrast, the curvature/propagation term $\langle H_{{\boldsymbol {\omega }}} \rangle$ is intricately related to the local mean curvature. Figure 20 further indicates that the curved local surfaces with large mean curvatures are usually associated with a strong production/destruction of surface area.

Figure 20.

Conditional averages of (a) the local strain effects $\langle S_{{\boldsymbol {\omega }}}\mid H \rangle$ and $\langle {H_{{\boldsymbol {\omega }}}}+{S_{{\boldsymbol {\omega }}}}\mid H \rangle$; (b) the local curvature/propagation effects $\langle H_{{\boldsymbol {\omega }}}\mid H \rangle$ and $\langle {H_{{\boldsymbol {\omega }}}}+{S_{{\boldsymbol {\omega }}}}\mid H \rangle$ for $T/T_{ref}=37.5$. Note also that the enlarged inset plot clearly shows the dependence of the local strain effects $\langle S_{{\boldsymbol {\omega }}}\mid H \rangle$ on the mean curvature $H$.

The curvature/propagation term $\langle H_{{\boldsymbol {\omega }}} \rangle$ increases the surface area in bulging regions while destroying it in the valleys, ensuring that the total surface area of the TNTI remains roughly unchanged. More specifically, the outwards expansion of the TNTI with $v_n <0$ leads to an increase of the surface area within bulging regions ($H_{{\boldsymbol {\omega }}} =2v_nH > 0$), whereas it destroys the surface area within valley regions ($H_{{\boldsymbol {\omega }}} =2v_nH < 0$). We can also reasonably argue that the continuous decrease of the average mean curvature (see figure 14a) is caused by the destruction of the valley region and the production of the bulging region.

The conditional averages of the two components of the curvature/propagation term ${H_{{\boldsymbol {\omega }}}}$, $v_n^P\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and $v_n^{vis}\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$, are plotted in figure 21(a). The conditional average of $v_n^{vis}\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ exhibits a discernible negative correlation with the local mean curvature $H$. Figure 21(b) suggests that the negative correlation between $\langle v_n^{vis}\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n} \rangle$ and $H$ is caused by the influence of $v_n^N\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$. A pronounced positive correlation between $\langle v_n^\varepsilon \boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}\rangle$ and $H$ is observed. Figure 21(b) also shows that the growth of surface area in bulging regions is attributed to the relatively large positive values of $\langle v_n^{N}\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n} \rangle$, which counteract the destruction of surface area caused by terms $\langle v_n^T\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n} \rangle$ and $\langle v_n^\varepsilon \boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n} \rangle$. In the Appendix, it is further demonstrated that the results are also valid at a higher Reynolds number, i.e. $Re_H = 8000$.

Figure 21.

Conditional averages of (a) $v_n^P\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and $v_n^{vis}\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and (b) $v_n^T\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$, $v_n^N\boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ and $v_n^\varepsilon \boldsymbol {\nabla }\boldsymbol {\cdot }\boldsymbol {n}$ for $T/T_{ref} = 37.5$.

Concerning the outwards movement of the TNTI, the contribution of the tangential diffusion term $v_n^T$ is considerably smaller than the normal diffusion term $v_n^N$ (see figure 9b). However, the conditional average of all three components of the curvature/viscous term $v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}$, are comparable to each other ($\langle v_n^{T} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle =-0.45$, $\langle v_n^{N} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle =-0.94$ and $\langle v_n^{\varepsilon } \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle =1.32$), which results in a relative small sum $\langle v_n^{vis} \boldsymbol {\nabla } \boldsymbol {\cdot } \boldsymbol {n}\rangle = - 0.07$. Therefore, all three components (i.e. $v_n^T$, $v_n^N$ and $v_n^\varepsilon$) of $v_n^{vis}$ are indispensable for the constant surface area at $T/T_{ref} \ge 20.0$.