2.1. Previous representation of scalar diffusivity in the single-component fluid

The Osborn (Reference Osborn1980) formulation of the mixing-efficiency problem was derived on the basis of the following simplified equation for the conservation of turbulent kinetic energy:

(2.1a)\begin{equation} \mathcal{P}=\mathcal{H}+\varepsilon, \end{equation}

in which the shear production of the background flow is $\mathcal {P}$, the turbulent buoyancy flux is $\mathcal {H}$ and the viscous dissipation is $\varepsilon$, all of which are defined as follows:

(2.2a)$$\begin{gather} \mathcal{P}={-\left\langle \overline{u'w'} \frac{{\rm d}\bar{u}}{{\rm d} z} \right\rangle}, \end{gather}$$

(2.2b)$$\begin{gather}\mathcal {H}=\frac{g}{\rho_0} \langle \overline{\rho'w'}\rangle, \end{gather}$$

(2.2c)$$\begin{gather}\varepsilon=2\nu \overline{\langle s_{ij}s_{ij} \rangle}. \end{gather}$$

In above equations, the overbar on a variable $\bar {f}$ represents the horizontal average of the field $f$, the bracket $\langle\, \cdot\, \rangle$ represents the vertical average, $\boldsymbol {u}=(u,v,w)$ is the velocity field that is further separated into the horizontally averaged fields $\bar {\boldsymbol {u}}$ and the perturbated field $\boldsymbol {u}'$ to it; $\rho _0$ is the reference density and $\rho '=\rho -\rho _0$ is the density perturbation; $s_{ij}=(\partial u_i/\partial x_j+\partial u_j/\partial x_i)/2$ is the strain rate tensor.

By employing the definition of the flux Richardson number $R_f=\mathcal {H}/\mathcal {P}$, Osborn (Reference Osborn1980) wrote the diapycnal diffusivity $K^{Osb}_\rho$ in the form of

(2.3a)$$\begin{gather} K^{Osb}_\rho=\frac{\mathcal{H}}{N^2}=\nu \frac{\mathcal{H}}{\varepsilon} \frac{\varepsilon}{\nu N^2}=\nu \varGamma^{Osb} Re_b, \end{gather}$$

(2.3b)$$\begin{gather}\varGamma^{Osb}=\frac{\mathcal{H}}{\varepsilon}=\frac{R_f}{1-R_f}, \end{gather}$$

in which $\varGamma ^{Osb}$ is usually referred to as the flux coefficient and the value 0.2 was estimated to be the upper bound for $\varGamma ^{Osb}$ in the original work of Osborn (Reference Osborn1980). In the subsequent practical application of this formulation of the mixing problem, $\varGamma ^{Osb}$ has always been assumed to be equal to the constant value 0.2 when applied to the understanding of oceanographic measurements. This is in spite of the fact that there exists significant evidence from simulations demonstrating that the value of $\varGamma =0.2$ may not be accurate (see the recent review of Gregg et al. (Reference Gregg, D'Asaro, Riley and Kunze2018) concerning its application in the field of oceanography).

However, as pointed out by Winters et al. (Reference Winters, Lombard, Riley and D'Asaro1995) and Peltier & Caulfield (Reference Peltier and Caulfield2003), the buoyancy flux defined in (2.1) contains the influence of both irreversible and reversible mixing processes whereas only the irreversible component should contribute to mixing when this is represented by a diapycnal diffusivity. In order to differentiate true irreversible mixing from adiabatic stirring, a background potential energy $BPE$ is defined by ‘sorting’ the three-dimensional density field into a vertical profile $\rho _\ast (z,t)$ with a decreasing upwards density $BPE=g/\rho _0 \overline {\langle \rho _*(z,t) z \rangle }$. The energy stored in this background potential energy reservoir is the minimum potential energy which cannot be transformed into kinetic energy. On the other hand, the differences between this $BPE$ and the total potential energy $PE=g/\rho _0 \overline {\langle \rho z \rangle }$ is defined as the available potential energy ($APE$), as this part of the potential energy is ‘available’ to be transferred back to macroscopic motion. In a closed domain (no body force, no boundary flux), the time derivative of the $BPE$ can be shown (Winters & D'Asaro Reference Winters and D'Asaro1996) to be

(2.4a)$$\begin{gather} \frac{{\rm d}}{{\rm d} t}BPE=\mathcal{M}+\mathcal{D}_p, \end{gather}$$

(2.4b)$$\begin{gather}\mathcal{M}+\mathcal{D}_p= \frac{\kappa g}{\rho_0 V}\int_V -\frac{{\rm d} z}{{\rm d} \rho_*} |\boldsymbol{\nabla} \rho|^2 \,{\rm d} V. \end{gather}$$

In above equations, $\kappa$ is the density diffusivity in the single-component system. Since ${\rm d} BPE/{\rm d} t$ is always positive, $BPE$ is a monotonically increasing function in time. Here, $\mathcal {M}$ is the irreversible buoyancy flux that characterizes the instantaneous mixing strength across the pycnocline that is generated due to the macroscopic fluid motion and $\mathcal{D}_p$ characterizes the part of mixing that would occur even in a completely motionless flow. In fact, $\mathcal{D}_p$ is always negligible if any form of turbulence is developed in a system that is not incredibly small so that the second equation in (2.4) can also be treated as the definition for $\mathcal {M}$.

Based on these definitions, Salehipour & Peltier (Reference Salehipour and Peltier2015) derived a modified expression for the diapycnal diffusivities in which the flux Richardson number $R_f$ in (2.3) is replaced by the irreversible mixing efficiency $\mathcal {E}$ as

(2.5a)$$\begin{gather} K^{irr}_\rho=\nu \frac{\mathcal{M}}{\varepsilon}\frac{\varepsilon}{\nu N_*^2} =\nu \varGamma^{irr} Re_{b*}, \end{gather}$$

(2.5b)$$\begin{gather}\varGamma^{irr}= \frac{\mathcal{M}}{\varepsilon} =\frac{\mathcal{E}}{1-\mathcal{E}}, \end{gather}$$

in which $N_*^2$ is the squared buoyancy frequency in the sorted profile but is always identical to the traditional definition of $N^2$ (see Salehipour & Peltier (Reference Salehipour and Peltier2015), so that $Re_{b*}=Re_b$). Equations (2.5) have the same form as (2.3), except that the irreversible versions of physical quantities in (2.5) are employed in place of Osborn's original expressions. Through these modifications, the formulae now correctly define the diapycnal diffusivities in terms of quantities involving irreversible mixing processes.

2.2. Scalar diffusivities in the presence of two diffusing species

We will here proceed to extend (2.5) to a diffusive–convection system, following similar approaches that were applied in Ma & Peltier (Reference Ma and Peltier2021) to the understanding of diapycnal diffusivities in salt-fingering turbulence. The existence of the unstably stratified scalar field of temperature in the Arctic Ocean region allows potential energy to kinetic energy conversion and thereby the creation of macroscopic motion, which was unavailable in the single-component case in which the background stratification of density was stably stratified. Thus, an energy budget analysis will be needed in order for a correct characterization of the diapycnal diffusivities for both scalars to be defined.

The total kinetic energy per unit mass may be represented as $\mathcal {K}=\overline {|\boldsymbol {u}^2|}/2$. Based on the assumption of the linear equation of state $\rho = \rho _0(1 -\alpha (\varTheta -\varTheta _0) + \beta (S-S_0))$ (thermal expansion rate $\alpha$ and haline contraction rate $\beta$ are both assumed to be constant), we define the averaged potential energy per unit mass and decompose it into a temperature reservoir $PE_\varTheta$ and a salinity reservoir $PE_S$ as follows:

(2.6)\begin{align} PE &= \frac{g}{\rho_0}\overline{ \langle \rho z\rangle} , \nonumber\\ &={-}g\alpha \overline{\langle \varTheta z\rangle}+g\beta \overline{\langle S z\rangle} + { g\overline{\langle z\rangle}}, \nonumber\\ &= PE_\varTheta+PE_S+ PE_0. \end{align}

Here, $PE_0$ is a constant term that will be ignored in what follows.

The time derivatives of $\mathcal {K}$, $PE$, $PE_\varTheta$, $PE_S$ can be derived straightforwardly by assuming that the two fluid components obey the Boussinesq governing equations which leads to the system

(2.7a)\begin{align} \frac{{\rm d} \mathcal{K}}{{\rm d} t} &={-}\mathcal{H}_\varTheta-\mathcal{H}_S-\varepsilon, \end{align}

(2.7b)\begin{align} \frac{{\rm d} PE_\varTheta}{{\rm d} t} &= \mathcal{H}_\varTheta+\mathcal{D}_{p\varTheta}, \end{align}

(2.7c)\begin{align} \frac{{\rm d} PE_S}{{\rm d} t} &= \mathcal{H}_S+\mathcal{D}_{pS}, \end{align}

(2.7d)\begin{align} \frac{{\rm d} PE}{{\rm d} t} &= \frac{{\rm d} PE_\varTheta}{{\rm d} t}+\frac{{\rm d} PE_S}{{\rm d} t} ,\nonumber\\ &= \mathcal{H}_\varTheta+\mathcal{H}_S+\mathcal{D}_{p\varTheta}+\mathcal{D}_{pS}, \nonumber\\&=\mathcal{H}+\mathcal{D}_{p}, \end{align}

where

(2.8a)$$\begin{gather} \mathcal{H}_\varTheta ={-}g\alpha \overline{\langle \varTheta' w' \rangle}, \end{gather}$$

(2.8b)$$\begin{gather}\mathcal{H}_S = g\beta \overline{\langle S'w' \rangle}, \end{gather}$$

(2.8c)$$\begin{gather}\mathcal{D}_{p\varTheta} = g\alpha \kappa_\theta \overline{\left\langle \frac{\partial \varTheta}{\partial z} \right\rangle}, \end{gather}$$

(2.8d)$$\begin{gather}\mathcal{D}_{pS} ={-} g\beta \kappa_s \overline{ \left\langle \frac{\partial S}{\partial z} \right\rangle}. \end{gather}$$

Just as in the single-component case, the buoyancy fluxes $\mathcal {H}_S$ and $\mathcal {H}_\varTheta$ contain the contributions from both reversible processes and irreversible processes. The reversible fluxes capture the energy transfer between the kinetic energy reservoir and the available potential energy reservoirs $APE_S$ and $APE_\varTheta$, while the irreversible fluxes transfer energy between $APE_S$ and $APE_\varTheta$ and the background potential energies $BPE_S$ and $BPE_\varTheta$. Specifically, the background potential energies $BPE_\varTheta$ and $BPE_S$ are defined as the part of the potential energy that is associated with adiabatic re-arrangements of the temperature and salinity profiles to monotonically decreasing profiles $\varTheta (z_{\theta \ast })$ and $S(z_{s\ast })$ and $APE_\varTheta$ and $APE_S$ describes the differences between total energies and background potential energies, namely

(2.9a)$$\begin{gather} BPE_\varTheta={-}g\alpha \overline{\langle \varTheta(z_{\theta\ast},t) z_{\theta\ast} \rangle}, \end{gather}$$

(2.9b)$$\begin{gather}BPE_S=g \beta \overline{\langle S(z_{s\ast},t) z_{s\ast} \rangle}, \end{gather}$$

(2.9c)$$\begin{gather}APE_\varTheta=PE_\varTheta-BPE_\varTheta, \end{gather}$$

(2.9d)$$\begin{gather}APE_S=PE_S-BPE_S. \end{gather}$$

The irreversible buoyancy fluxes for heat ($\mathcal {M}_\varTheta$) and salt ($\mathcal {M}_S$) again characterize the time derivative of $BPE_\varTheta$ and $BPE_S$ in a closed system as

(2.10a) \begin{align} \frac{{\rm d}}{{\rm d} t}BPE_\varTheta &= g\alpha \kappa_\theta \left\langle\frac{{\rm d} z_{\theta\ast}}{{\rm d}\varTheta}|\boldsymbol{\nabla} \varTheta|^2\right\rangle, \nonumber\\ &= \mathcal{M}_{\varTheta}+D_{p \varTheta},\end{align}

(2.10b)\begin{align} \frac{{\rm d}}{{\rm d} t}BPE_S &={-}g\beta \kappa_s \left\langle \frac{{\rm d} z_{s\ast}}{{\rm d} S}|\boldsymbol{\nabla} S|^2\right\rangle, \nonumber\\ &= \mathcal{M}_{S}+D_{pS}, \end{align}

(2.10c)\begin{align} \frac{{\rm d}}{{\rm d} t}BPE &= \frac{{\rm d}}{{\rm d} t}BPE_\varTheta+\frac{{\rm d}}{{\rm d} t}BPE_S, \nonumber\\ &= \mathcal{M}_{\varTheta}+\mathcal{M}_{S}+D_{p \varTheta}+D_{p S}, \nonumber\\ &\equiv \mathcal{M}+D_{p}. \end{align}

The above sets of equations imply simply that, while $BPE_S$ is a monotonical increasing function with time as in the traditional definition of background potential energy for a single-component fluid, $BPE_\varTheta$ is a monotonically decreasing function which irreversibly releases energy to $APE_\varTheta$ which can then be transported to the kinetic energy reservoir. The total background potential energy $BPE$, however, can either increase or decrease with time, depending upon the relative strengths of the negative $\mathcal {M}_{\varTheta }$ and positive $\mathcal {M}_{S}$ in the system. The energy exchanges described above can also be visualized in the simplified diagram shown in figure 1. It should be noticed that the $APE_\varTheta$ has a slightly different meaning from the traditional implication of available potential energy; while available potential energy usually refers to the amount of potential energy stored by the reversible process that is available to be released to the kinetic energy reservoir in the single-component case (also applies for $APE_S$), here, $APE_\varTheta$ (having a negative value through its definition) represents the amount of energy that has already been transported into the kinetic energy reservoir. However, this part of energy is lost through a reversible process so that it could possibly be transported back through convection in the future evolution of the flow field. Combining $APE_\varTheta$ and $APE_S$, the $APE$ reservoir represents the part of the potential energy that can be exchanged with the kinetic energy reservoir through reversible processes.

Figure 1.

Graphical demonstration of energy budgets in the diffusive–convection environment. The direction of the energy flow of the positive/negative transportation is clarified using arrows.

Given the definition of the irreversible buoyancy fluxes $\mathcal{M}_\varTheta$ and $\mathcal{M}_S$ above, we can derive the irreversible diapycnal diffusivities for heat and salt as follows:

(2.11a)\begin{align} K^{irr}_\varTheta &= \frac{\mathcal{M}_\varTheta}{ g \alpha \left\langle \dfrac{{\rm d} \varTheta}{{\rm d} z_{\theta \ast}} \right\rangle }, \end{align}

(2.11b)\begin{align} &= \nu \frac{\mathcal{M}_\varTheta}{\varepsilon} \frac{ N^2 }{ g \alpha \left\langle \dfrac{{\rm d} \varTheta}{{\rm d} z_{\theta \ast}} \right\rangle} \frac{\varepsilon}{\nu N^2 }, \end{align}

(2.11c)\begin{align} &= \nu \frac{\mathcal{M}_\varTheta}{\varepsilon} \frac{ R_{\rho \ast}-1 }{-1} \frac{\varepsilon}{\nu N^2 }, \end{align}

(2.11d)\begin{align} &= \nu \varGamma^{irr}_{\varTheta} Re_b, \end{align}

(2.11e)\begin{align} K^{irr}_S &={-}\frac{\mathcal{M}_S}{ g \beta \left\langle \dfrac{{\rm d} S}{{\rm d} z_{s \ast}} \right\rangle }, \end{align}

(2.11f)\begin{align} &={-}\nu \frac{\mathcal{M}_S}{\varepsilon} \frac{ N^2 }{ g \beta \left\langle \dfrac{{\rm d} S}{{\rm d} z_{s \ast}}\right\rangle} \frac{\varepsilon}{\nu N^2 }, \end{align}

(2.11g)\begin{align} &= \nu \frac{\mathcal{M}_S}{\varepsilon} \frac{ R_{\rho \ast}-1 }{R_{\rho \ast}} \frac{\varepsilon}{\nu N^2 }, \end{align}

(2.11h)\begin{align} &= \nu \varGamma^{irr}_{S} Re_b, \end{align}

where

(2.12a)$$\begin{gather} R_{\rho \ast}\equiv\frac{\beta \left\langle \dfrac{{\rm d} S}{{\rm d} z_{s \ast}} \right\rangle}{ \alpha \left\langle \dfrac{{\rm d} \varTheta}{{\rm d} z_{\theta \ast}} \right\rangle}, \end{gather}$$

(2.12b)$$\begin{gather}\varGamma^{irr}_\varTheta \equiv \frac{- (R_{\rho \ast}-1) \mathcal{M}_\varTheta} {\varepsilon}, \end{gather}$$

(2.12c)$$\begin{gather}\varGamma^{irr}_{S} \equiv \frac{(R_{\rho \ast}-1)\mathcal{M}_S }{\varepsilon R_{\rho \ast}}. \end{gather}$$

In above equations, $R_{\rho \ast }$ is always identical to the traditional $R_\rho$ (due to the same reason that $N_\ast ^2$ is identical to $N^2$, which we have mentioned above) so that we will not differentiate them in what follows. Also, $\varGamma ^{irr}_{\varTheta }$ and $\varGamma ^{irr}_S$ are defined as the flux coefficients for temperature and salinity separately ($\varGamma ^{irr}_{\varTheta }$ has also been previously introduced as ‘the dissipation ratio’ in the literature e.g. Laurent & Schmitt Reference St. Laurent and Schmitt1999). Since the overall stratification is stable we have $R_\rho >1$, this leads to the fact that both $\varGamma ^{irr}_{\varTheta }$ and $\varGamma ^{irr}_S$ are positive, guaranteeing that the diapycnal diffusivities for both scalars $K^{irr}_\varTheta$, $K^{irr}_S$ are positive.

Meanwhile, the diapycnal diffusivity for density can be derived in the form of the density flux coefficient as

(2.13a)\begin{align} K^{irr}_\rho&=\frac{\mathcal{M}}{ N^2 }, \end{align}

(2.13b)\begin{align} &=\nu \frac{\mathcal{M}}{\varepsilon} \frac{\varepsilon}{\nu N^2 }, \end{align}

(2.13c)\begin{align} &=\nu \varGamma^{irr}_\rho Re_b. \end{align}

By employing the buoyancy flux $\mathcal {M}$ as a summation of $\mathcal {M}_\theta$ and $\mathcal {M}_s$, it is straightforward to show that $\varGamma ^{irr}_{\rho }$ (or $K^{irr}_\rho$) can be determined by $\varGamma ^{irr}_{\varTheta }$ (or $K^{irr}_\varTheta$) and $\varGamma ^{irr}_{S}$ (or $K^{irr}_S$) from

(2.14a)$$\begin{gather} \varGamma^{irr}_\rho=\frac{R_\rho}{R_\rho-1} \varGamma^{irr}_S -\frac{1}{R_\rho-1} \varGamma^{irr}_\varTheta, \end{gather}$$

(2.14b)$$\begin{gather}K^{irr}_\rho=\frac{R_\rho}{R_\rho-1} K^{irr}_S -\frac{1}{R_\rho-1} K^{irr}_\varTheta. \end{gather}$$

Although $K^{irr}_\varTheta$ and $K^{irr}_S$ are both positive, as has been demonstrated above, (2.14) shows that $K^{irr}_\rho$ can be negative if the temperature term dominates. As we will see below, this situation might occur in the early and late evolution stage of KH instability growth in the strongly stratified case, in which situation the strength of the turbulence is weak enough and the temperature mixes more efficiently than salinity.

As in the single-component case, the irreversible flux coefficient $\varGamma ^{irr}_\rho$ can be written in the form of instantaneous mixing efficiency as

(2.15a)$$\begin{gather} \varGamma^{irr}_\rho=\frac{\mathcal{E}}{1-\mathcal{E}}, \end{gather}$$

(2.15b)$$\begin{gather}\mathcal{E}=\frac{\mathcal{M}}{\mathcal{M}+\varepsilon}= \frac{\mathcal{M}_\varTheta+\mathcal{M}_S}{\mathcal{M}_\varTheta+\mathcal{M}_S+\varepsilon}. \end{gather}$$

In the single-component case $\mathcal {E}$ always remains in the range $0<\mathcal {E}<1$ and clearly represents the amount of irreversible mixing relative to the viscous dissipation. However, in the diffusive–convection environment $\mathcal {E}$ can take both negative values and values that are much larger than 1, in the cases of $\mathcal {M}<0$ following its definition in (2.15). Therefore, $\mathcal {E}$ no longer carries the meaning of ‘efficiency’ in the doubly diffusive system and we will employ the flux-coefficient form of the diffusivities in (2.11) rather than the mixing-efficiency form in our analyses in what follows.

Another important physical quantity is the ratio of (irreversible) diapycnal diffusivity for salinity to that for temperature, namely

(2.16)\begin{equation} d=\frac{K^{irr}_S}{K^{irr}_\varTheta}. \end{equation}

The ratio of diapycnal diffusivities $d$ has been widely used in the literature (e.g. Gargett, Merryfield & Holloway Reference Gargett, Merryfield and Holloway2003; Merryfield Reference Merryfield2005; Smyth et al. Reference Smyth, Nash and Moum2005; Jackson & Rehmann Reference Jackson and Rehmann2009) to characterize the degree of differential diffusivity in the system where both temperature and salinity fields are stably stratified. These analyses demonstrate that $d$ is close to unity in the strong turbulence limit, but decreases rapidly as turbulence intensity decreases or stratification strengthens, see Gregg et al. (Reference Gregg, D'Asaro, Riley and Kunze2018) for further discussion.

The above formulae provide us with the theoretical basis required for calibration of the irreversible components of diapycnal diffusivities in studies of doubly diffusive turbulence. Using DNSs that we will introduce in the next section, we will investigate quantitively how energy is transferred between the different energy reservoirs and how the irreversible diapycnal diffusivities evolve in a typical KH life cycle.

3.1. Theoretical preliminaries

In order to study the mixing induced by vertical shear in a system stratified in both temperature and salinity, we apply the idealized initial vertical profiles for horizontal velocity, temperature and salinity as follows:

(3.1a)$$\begin{gather} u(x,y,z,t=0)=U_0 \tanh \left(\frac{z}{h} \right), \end{gather}$$

(3.1b)$$\begin{gather}\varTheta(x,y,z,t=0)={-}\Delta \varTheta \tanh \left(\frac{z}{h} \right), \end{gather}$$

(3.1c)$$\begin{gather}S(x,y,z,t=0)={-}\Delta S \tanh \left(\frac{z}{h} \right), \end{gather}$$

where ($x,y,z$) are the streamwise, spanwise and vertical directions (positive $z$ direction is set to be antiparallel with gravity), respectively, and ($u,v,w$) represents the velocity component in each of these directions; $h$ is half the thickness for the shear layer (which is also half the thickness of the salinity and temperature interfaces in the model system to be employed), $\Delta \varTheta$, $\Delta S$ and $U_0$ are half the variations of initial temperature, salinity and horizontal velocity profiles across the interface, as shown in the sketch of these initial profiles in figure 2. Both $\varTheta (z)$ and $S(z)$ will contribute to the density through an idealized linear equation of state $\rho = \rho _0(1 -\alpha \varTheta + \beta S)$. To mimic the stratification in the Arctic region, we have relatively colder and fresher water above warmer and saltier water while keeping the density profile gravitationally stable. This requires that the stably stratified salinity contributes more to density than the unstably stratified temperature profile, namely $\Delta \rho =\beta \Delta S-\alpha \Delta \varTheta >0$, as illustrated in figure 2.

Figure 2.

Sketch of initial condition for the scalar fields (a) and streamwise velocity field (b) in our KH simulation. In (a) the dashed, dotted and sold lines represent $S(z)$, $\varTheta (z)$ and $\rho (z)$, respectively.

The flows of interest to us will be described by the (non-dimensional) Boussinesq approximation with the system

(3.2a)$$\begin{gather} \frac{\partial \boldsymbol{u}}{\partial t}+\boldsymbol{u}\boldsymbol{\cdot}\boldsymbol{\nabla} \boldsymbol{u}={-}\boldsymbol\nabla p -J \left(\frac{R_\rho}{R_\rho-1}S -\frac{1}{R_\rho-1} \varTheta\right) \boldsymbol{e_z}+\frac{1}{Re}\nabla^2\boldsymbol{u}, \end{gather}$$

(3.2b)$$\begin{gather}\boldsymbol{\nabla}\boldsymbol{\cdot} \boldsymbol{u} = 0, \end{gather}$$

(3.2c)$$\begin{gather}\frac{\partial \varTheta}{\partial t}+\boldsymbol{u}\boldsymbol{\cdot}\boldsymbol{\nabla} \varTheta =\frac{1}{Re\,Pr}\nabla^2\varTheta, \end{gather}$$

(3.2d)$$\begin{gather}\frac{\partial S}{\partial t}+ \boldsymbol{u}\boldsymbol{\cdot}\boldsymbol{\nabla} S= \frac{1}{Re\,Sc} \nabla^2S, \end{gather}$$

in which the non-dimensionalization has employed $h$ as the length scale and $\Delta \varTheta$, $\Delta S$ and $U_0$ as the temperature, salinity and velocity scales, respectively. The five non-dimensional control parameters in this set of field equations are the Reynolds number $Re$, the bulk Richardson number $J$, the density ratio $R_\rho$, the Prandtl number $Pr$ as well as the Schmidt number $Sc$, which are defined as follows:

(3.3a)$$\begin{gather} Re=\frac{U_0 h}{\nu}, \end{gather}$$

(3.3b)$$\begin{gather}J=\frac{g\Delta \rho h}{\rho_0 U_0^2}=\frac{g(\beta \Delta S-\alpha \Delta \varTheta) h}{\rho_0U_0^2}, \end{gather}$$

(3.3c)$$\begin{gather}R_\rho=\frac{\beta \Delta S}{\alpha \Delta \varTheta}, \end{gather}$$

(3.3d)$$\begin{gather}Pr=\frac{\nu}{\kappa_\theta}, \end{gather}$$

(3.3e)$$\begin{gather}Sc=\frac{\nu}{\kappa_s}. \end{gather}$$

Compared with the single-component fluid upon which most studies of KH instability to date have focused, we have introduced the Schmitt number $Sc$ and the density ratio $R_\rho$ into the parameter space; $Sc$ represents the ratio of momentum diffusivity to the salinity diffusivity in the ocean. It is usually much higher than $Pr$ due to much lower diffusivity of salinity compared with that of heat. The density ratio $R_\rho$ characterizes the relative importance of salinity and temperature to the stratification of density, a parameter which lies in the range of $1< R_\rho <\infty$ in the system which is our intention to study. In the limit of $R_\rho \to \infty$, the unstably stratified temperature field $\varTheta (x,y,z,t)$ is decoupled from the momentum equation in (3.2a), so that the system described by (3.1) and (3.2) essentially returns to that for a single-component fluid whose stratification is entirely determined by salinity. On the other hand, if $R_\rho$ is close to 1, the unstably stratified component in the system becomes so strong that the system will also be susceptible to the buoyancy induced oscillatory diffusive–convection instability. In this scenario, the system is difficult to investigate numerically since both shear-driven instability and buoyancy-driven instability are involved and the widely separated length scales are activated simultaneously. More importantly, this small density ratio region of parameter space has seldom been observed in the Arctic ocean (Shibley et al. Reference Shibley, Timmermans, Carpenter and Toole2017). For this reason we will open our discussion in this paper to a much wider range of density ratio $R_\rho \ge 2$ that is more representative of observed conditions in the Arctic Ocean.

3.2. Detailed design characteristics of the ensemble of DNSs

Governing equations (3.2) are integrated in a hexahedron of size $(L_x,L_y,L_z)$ using the open-source computational fluid dynamics solver Nek5000 (Paul, James & Kerkemeier Reference Paul, James and Kerkemeier2008). Nek5000 was originally developed at Argonne National Laboratory based on the spectral element method in such a way as to support a user-defined complex geometry (see Fischer (Reference Fischer1997) and Fischer, Kruse & Loth (Reference Fischer, Kruse and Loth2002) for example). It is well suited to simulating highly turbulent flows (see Salehipour et al. (Reference Salehipour, Peltier and Mashayek2015) and Ma & Peltier (Reference Ma and Peltier2021) for example) since it allows users to economically design the computational mesh in such a way as to contain higher resolution in more strongly turbulent regions and lower resolution elsewhere.

The detailed information for each of our numerical simulations that are to be discussed in this paper is summarized in table 1. We integrate the doubly diffusive systems with different initial bulk Richardson numbers $J$ and different density ratios $R_\rho$ to investigate their influences on the evolution of the KH lifecycle. We furthermore perform control simulations of the single-component KH billow (simulation numbers 4 and 7) to illustrate in detail how the introduction of another diffusing species will influence the evolution of KH billows. For most of the simulations performed in this paper, we set the streamwise extent of our domain $L_x$ to contain one wavelength of the fastest growing mode of linear instability, except in simulation number 5, in which we select the domain length to contain twice the fastest growing wavelength in order to investigate the secondary pairing instability that we will describe in the next section. The spanwise extent of the domain $L_y$ is set to be 5$h$ and a slightly smaller domain of 3$h$ has been selected for the high-resolution simulation numbers 5 and 6, both of which have been shown to be large enough to ensure that the fastest growing modes of secondary cross-stream instabilities are adequately resolved (Mashayek & Peltier Reference Mashayek and Peltier2011). The value of $L_z$ is set to 20$h$ in all these simulations.

Table 1.

Governing parameters for the DNSs discussed in this paper.

It is notoriously difficult to perform DNSs that involve the evolution of the salinity field: the low haline diffusivity requires an extremely high resolution so that the Batchelor scale for salinity $L_B=(\nu \kappa ^2_s/ \varepsilon )^{1/4}$ can be resolved in our DNS grids. To this end we employ compromise values of $Sc=70$ and $Pr=7$, a condition which has relatively mild mesh requirements while keeping an order of magnitude difference between the salinity and temperature diffusivities. Meanwhile, the small Batchelor scale that needs to be resolved in DNS also exerts a constraint on the Reynolds number: a value of $Re=600$ provides the $L_B$ value that is required for our current simulations. As will be demonstrated in what follows, this intermediate value of the Reynolds number will lead to values of the buoyancy Reynolds number in the turbulent phase of billow evolution of $O(10)$, which is in the range of moderate turbulent intensity observed to characterize Arctic Ocean turbulence, as discussed in Dosser et al. (Reference Dosser, Chanona, Waterman, Shibley and Timmermans2021). To design the most efficient mesh for each of these simulations we have employed a series of low-resolution simulations to calibrate $L_B$, according to which the mesh resolution for the high-resolution simulations has been selected so that the depth-dependent mesh size is always smaller than $3L_B$ within the entire lifecycle of the KH turbulence (the pre-determination of mesh grids are described in Appendix A).

In these simulations, the initial condition (3.1) is seeded with a small-amplitude two-dimensional structure equal to that of the fastest growing mode (the non-dimensional horizontal velocity amplitude is set to 0.005) in the linear stability analysis of the Taylor–Goldstein equation. A further component of the initial conditions consisting of white noise of magnitude $0.0005(\Delta \varTheta, \Delta S)$ is included to seed the growth of the secondary instabilities. We choose periodic boundary conditions for salinity and temperature as well as velocity fields in the streamwise and spanwise directions. Meanwhile, on the top and bottom surfaces of the domain, we assume free-slip and impermeable boundary conditions for velocity and insulated boundary conditions for the temperature and salinity fields.

4.1. Different phases of evolution of KH instability with two oppositely stratified species

In order to aid our analysis of the KH instability and its subsequent nonlinear evolution, we decompose the velocity field into the horizontally averaged mean field $\bar {\boldsymbol {u}}$, the spanwise-averaged component $\boldsymbol {u_{2d}}$ associated with the primary KH wave as well as an inherently three-dimensional component $\boldsymbol {u_{3d}}$ that is associated with the secondary instability arising from the primary KH billow, namely

(4.1)\begin{equation} \boldsymbol{u}=\bar{\boldsymbol{u}}+\boldsymbol{u_{2d}}+\boldsymbol{u_{3d}}, \end{equation}

the individual components of these vector fields are defined as

(4.2a)$$\begin{gather} (\bar{u},0,0)= \overline{(u,v,w)}, \end{gather}$$

(4.2b)$$\begin{gather}(u_{2d},0,w_{2d})=\langle (u-\bar{u},v,w) \rangle_{y}, \end{gather}$$

(4.2c)$$\begin{gather}(u_{3d},v_{3d},w_{3d})=(u-\bar{u}-u_{2d},v,w-w_{2d}). \end{gather}$$

In the above equations, the symbol $\langle\, \cdot\, \rangle _{y}$ represents averaging of the field over the spanwise direction. The total kinetic energy $\mathcal {K}$ of the flow can then be decomposed as $\mathcal {K}=\bar {\mathcal {K}}+\mathcal {K}_{2d}+\mathcal {K}_{3d}$ and the values of $\mathcal {K}_{2d}$ and $\mathcal {K}_{3d}$ represent the growth of the primary KH billow and the growth of three-dimensional turbulence, respectively. Here, we illustrate the evolution of $\mathcal {K}, \mathcal {K}_{2d}$ and $\mathcal {K}_{3d}$, normalized by the initial kinetic energy $\mathcal {K}_0$ in figures 3(a) and 3(b) for simulation number 2 ($J=0.12$, $R_\rho =5$). Following Peltier & Caulfield (Reference Peltier and Caulfield2003), this compartmentalization allows us to define four different characteristic times $t_{2dmax}, t_d, t_{3dmax}, t_{end}$ to divide the system into four different phases of evolution. The first phase represents the growth of the initially two-dimensional primary KH billow, it begins at $t=0$ and ends at $t=t_{2dmax}$, which is defined as the time when the two-dimensional KH billow saturates (the time that $\mathcal {K}_{2d}$ reaches its maximum). During the second phase, the saturated KH billow continues to evolve in a two-dimensional fashion. This phase ends at $t_d$, which characterizes the onset of the three-dimensional secondary instability. Quantitively, $t_{d}$ is defined by the time during which the viscous dissipation rate $\varepsilon (t)$ doubles from its initial value. Shortly after $t_d$, the three-dimensional secondary instability starts to grow, as shown in the curve of $\mathcal {K}_{3d}$ in figure 3(b), until $\mathcal {K}_{3d}$ reaches its maximum value at $t_{3d}$. The fourth stage represents the decay of three-dimensional turbulence until the flow becomes laminar at $t_{end}$, which we take to be defined as the time at which $\mathcal {K}_{3d}$ falls below $10\,\%$ of its peak value.

Figure 3.

Evolution of $\mathcal {K}$, $\mathcal {K}_{2d}$, $\mathcal {K}_{3d}$ and various components of $PE$, $BPE$ normalized by the initial kinetic energy $\mathcal {K}_0$ as a function of time in simulation number 2. The four vertical dashed lines represent the values for the four characteristic times $t_{2dmax}, t_d, t_{3dmax}, t_{end}$.

Visualizations of the salinity field and temperature field at these characteristic times for simulation number 2 are illustrated in figure 4. The primary KH billow can be clearly observed for both the salinity field and the temperature field at both $t_{2dmax}$ (shown in figure 4a,b) and $t_d$ (shown in figure 4c,d) when the flow is dominated by the two-dimensional dynamics. The development of secondary instabilities then drives the system into a fully turbulent state, as depicted in figure 4(e, f). It is important to note that, although temperature and salinity fields display essentially identical structures at $t_{2dmax}$ and $t_d$, they appear significantly different in the fully turbulent stage: the turbulent patches are much smaller in the salinity field than in the temperature field. The much smaller diffusivity for the salinity field allows for the existence of finer structure in the turbulence when compared with the temperature field. Finally, at $t_{end}$, the three-dimensional turbulence decays and the flow collapses into a laminar state which is characteristic of both the salinity and temperature fields in figure 4(g,h).

Figure 4.

Iso-surfaces for both the salinity fields (a,c,e,g) and the temperature fields (b,d, f,h) at four different characteristic times $t_{2dmax}$ (a,b), $t_d$ (c,d), $t_{3dmax}$ (e, f), $t_{end}$ (g,h).

The development and collapse of the KH billow eventually mixes the physical properties of the flow by transforming a significant fraction of the initial kinetic energy of the initial shear flow into background potential energies. In order to evaluate the variation of background energies, we have sorted both the temperature field and the salinity field utilizing the parallel sorting algorithm proposed in Salehipour et al. (Reference Salehipour, Peltier and Mashayek2015) to obtain the background potential energies for our DNS data in the evolution process. In figure 3(c), we plot the evolution of the background potential energies $BPE_\varTheta$, $BPE_S$, $BPE$ that can be compared with the conventional potential energies $PE_\varTheta$, $PE_S$, $PE$ (all have had their initial values subtracted and are normalized by $\mathcal {K}_0$) for simulation number 2. As we discussed in § 2, the kinetic energies in our current doubly diffusive system continue to extract energy from $BPE_\varTheta$ and transfer energy to $BPE_S$, leading to monotonic decrease of $BPE_\varTheta$ and monotonic increase of $BPE_S$. The total background potential energy is then determined by the summation of $BPE_\varTheta$ and $BPE_S$. Since the density stratification is dominated by the stably stratified salinity field, $BPE$ experiences an overall increase with time for this specific run (an example involving a decrease in $BPE$ will be discussed in the strongly stratified case in what follows).

Despite the fact that the stratification is mainly determined by salinity, the temperature field mixes more effectively than salinity considering the fact that the molecular diffusivity for temperature is 10 times higher than that for salinity in our DNSs. This can be quickly verified by referring to figure 3(c): from $t=0$ to $t=t_{end}$, $BPE_S$ increases by total amount of 0.0095$\mathcal {K}_0$ whereas $BPE_\varTheta$ decreases by the total amount of 0.0038$\mathcal {K}_0$. The ratio of their relative variations $\gamma ^{tot}$ can then be straightforwardly evaluated to have the value of 2.5, which is much smaller than the density ratio $R_\rho =5$, demonstrating that mixing in the temperature field leads to a more significant change in its background potential energy compared with salinity. We can also directly compare the variations of irreversible diapycnal diffusivities for salinity and temperature. In figure 5(a), we plot the evolution of irreversible flux for temperature $\mathcal{M}_\varTheta$, salinity $\mathcal{M}_S$ and density $\mathcal{M}$, respectively, (all non-dimensionalized by $U_0^3/h$) for simulation number 2. The associated evolution of bulk-averaged diapycnal diffusivities is plotted in figure 5(b). It is clear that $K^{irr}_\varTheta$ is significantly higher than $K^{irr}_S$ in all different stages of evolution, especially at approximately $t=t_d$ just before the onset of the secondary instabilities. The diffusivity ratio $d$ for the evolution is shown in figure 5(c). Consistently, $d$ is smaller than 1 except for the time near $t_{3dmax}$ at which the three-dimensional turbulence reaches the maximum amplitude. The combined diapycnal diffusivities for density $K^{irr}_\rho$ can then be determined by $K^{irr}_\varTheta$ and $K^{irr}_S$ based on (2.14). Generally speaking, $K^{irr}_\rho$ is close to $K^{irr}_S$ since salinity is the dominant component in determining the stratification. However, stronger $K^{irr}_\varTheta$ represents the stronger negative part of the density flux induced by temperature so that $K^{irr}_\rho$ will be influenced to be smaller.

Figure 5.

Evolution of irreversible fluxes $\mathcal{M}_\varTheta$, $\mathcal{M}_S$, $\mathcal{M}$ (a) irreversible diapycnal diffusivities $K^{irr}_\varTheta$, $K^{irr}_S$, $K^{irr}_\rho$ (non-dimensionalized by molecular viscosity $\nu$) (b), diffusivity ratio $d$ (c), flux coefficients $\varGamma ^{irr}_\varTheta$, $\varGamma ^{irr}_S$, $\varGamma ^{irr}_\rho$ (d) and dissipation ratios for scalars $\varepsilon _\varTheta$, $\varepsilon _S$ (non-dimensionalized by dimensional units of $\Delta \varTheta ^2 U_0^2/h$ and $\Delta S^2 U_0^2/h$ separately) (e) as a function of time in simulation number 2.

The fact that the temperature mixes more effectively than salinity can also be verified in their flux coefficients in figure 5(d). The irreversible flux coefficient for temperature $\varGamma ^{irr}_\varTheta$ reaches its peak of approximately 0.4 before the onset of three-dimensional secondary instability, and drops to the value of approximately 0.1 in the fully turbulent stage. While the value of $\varGamma ^{irr}_\varTheta$ in the lifecycle remains comparable to the canonical value of 0.2, the irreversible flux coefficient for salinity $\varGamma ^{irr}_S$ is always considerably lower than 0.2. This again emphasizes the idea that different flux coefficients should be assumed for temperature and salinity separately due to their different values of molecular diffusivity. The combined flux coefficient for density can also be determined through the relation (2.14b). Similar to the evolution of $K^{irr}_\rho$, $\varGamma ^{irr}_\rho$ is also close to $\varGamma ^{irr}_S$. The finite differences between $\varGamma ^{irr}_S$ and $\varGamma ^{irr}_\rho$ are mostly minor in the fully turbulent regime, and keep increasing as turbulence dies at the end of the simulation lifecycle. In figure 5(e) we also show the time evolution of the dissipation ratio for temperature $\varepsilon _\varTheta \equiv |\boldsymbol {\nabla } \varTheta |^2/(RePr)$ and salinity $\varepsilon _S \equiv |\boldsymbol {\nabla } S|^2/(ReSc)$, which are non-dimensionalized by dimensional units of $\Delta \varTheta ^2 U_0^2/h$ and $\Delta S^2 U_0^2/h$ separately. These physical quantities also reflect the strength of mixing in the turbulence lifecycle and their evolutions are consistent with the evolution of diapycnal diffusivities for both scalars, as in figure 5(b).

4.2. Influences of bulk Richardson number $J$ and density ratio $R_\rho$

Having discussed the typical characteristics of the evolution of KH billows and the mixing properties of turbulence in this doubly diffusive system, we will focus next upon the influence of the governing parameters $J$ and $R_\rho$ on the detailed characteristics of turbulent mixing that were discussed above in general terms for simulation number 2.

To demonstrate the specific influences of these two governing parameters, we show in figure 6(a–d) the evolution of the total kinetic energy $\mathcal {K}$, the background potential energy $BPE$, the buoyancy Reynolds number $Re_b$ and the irreversible flux coefficients for density $\varGamma ^{irr}_\rho$ separately for two different bulk Richardson numbers $J=0.05$ and $J=0.12$ at different values of $R_\rho$. By comparing the evolution of kinetic energy and background potential energy in figures 6(a) and 6(b), it will be clear that a larger proportion of energy is transferred from the kinetic energy reservoir to the background potential energies in the weaker stratification case $J=0.05$, compared with the stronger stratification $J=0.12$. This is consistent with the role played by the bulk Richardson number discussed in Caulfield & Peltier (Reference Caulfield and Peltier2000). The weaker stratification also naturally leads to a higher $Re_b$ (shown figure 6c) at the peak of turbulence intensity compared with the stronger stratification case, although $Re_b$ in both cases remains at a relatively low value due to the small value of $Re$ implemented in these simulations. It is also worth noting that the irreversible flux coefficient for density is also significantly higher in the turbulent phase for $J=0.05$ compared with $J=0.12$, as shown in figure 6(d). This decrease of flux coefficient with $J$ is also consistent with previous DNSs, and is often referred to as the right flank of the non-monotonic functional dependence of flux coefficient on the gradient Richardson number (e.g. Caulfield Reference Caulfield2021).

Figure 6.

Evolution of total kinetic energy $\mathcal {K}$ (a), background potential energy $BPE$ (b), buoyancy Reynolds number $Re_b$ (c) and irreversible flux coefficients for density $\varGamma ^{irr}_\rho$ (d) as a function of time in simulations with different governing parameters of bulk Richardson number $J$ and density ratio $R_\rho$.

With this understanding of the effect of $J$, we turn next to an exploration of the effect of $R_\rho$ on the evolution of KH billow turbulence in the doubly diffusive system, where $R_\rho$ represents the importance of the salinity field relative to the temperature field to stratification. By comparing the evolution of $BPE$ in figure 6(b), we are able to characterize the different behaviours of $BPE$ for simulations with different $R_\rho$. At relatively small density ratio $R_\rho =2$, we note that the background potential energy is decreasing prior to $t=100$ (before the onset of the three-dimensional secondary instability). Furthermore, in the special case of $R_\rho =2$ at $J=0.12$, the total background potential energy experiences a decreasing trend again after $t=170$ and falls below its initial value at approximately $t=200$. This period of decreasing $BPE$ may also be verified in figure 6(d) where it is associated with negative flux coefficient.

For the general comparison between different simulations shown in figure 6(b), we can conclude that a smaller $R_\rho$ always leads to a lower net increase of $BPE$ relative to its initial value. This can be qualitatively understood as follows: the constant increase of $BPE_S$ is competing with the constant decrease of $BPE_\varTheta$ in the evolution of $BPE$, and the smaller $R_\rho$ suggests that $BPE_\varTheta$ is playing a more important role in influencing $BPE$, which makes it easier for $BPE$ to decrease or to remain at a relatively low value. In fact, a quantitative explanation for the arguments above can be reached through an analysis of the total irreversible buoyancy flux

(4.3a)\begin{align} \mathcal{M} &= N^2K^{irr}_\rho, \end{align}

(4.3b)\begin{align} &= N^2\left(\frac{-1}{R_\rho-1} K_\varTheta^{irr}+\frac{R_\rho}{R_\rho-1} K^{irr}_S\right). \end{align}

In the above equations, (4.3b) is derived by substituting the relationship between $K^{irr}_\rho$, $K^{irr}_S$ and $K^{irr}_\varTheta$ that we have shown previously in (2.14b) into (4.3a). As we have demonstrated in the last subsection, $K_\varTheta ^{irr}$ is always higher than $K^{irr}_S$, especially when the turbulence is weak. In (4.3b), $N^2$ is fixed since we have employed the same bulk Richardson number $J$ in the simulations, the variation of $R_\rho$ influences the relative importance of $K_\varTheta ^{irr}$ and $K_S^{irr}$ to influence the instantaneous buoyancy flux $\mathcal {M}$. In the case of large $R_\rho$, $K^{irr}_\rho$ is close to the value of $K^{irr}_S$. When $R_\rho$ is sufficiently small, on the other hand, $\mathcal {M}$ can be negative when it is dominated by the negative term in (2.14b), leading to a decreasing $BPE$, as shown in the two curves with $R_\rho =2$ in figure 6(b) which we mentioned above. Generally speaking, the differences between $K^{irr}_S$ and $K_\varTheta ^{irr}$ are most significant when the buoyancy Reynolds number is small, which explains why these time intervals of decreasing $BPE$ occur either at the early or late stage of the KH evolution. In § 5 we will provide a detailed analysis of a parametrization scheme that is suitable for $K^{irr}_S$ and $K_\varTheta ^{irr}$ in our system based on the buoyancy Reynolds number $Re_b$, so that the detailed value of buoyancy flux in (4.3) can be better quantified.

While we have compared the time evolution of the KH billow under different parameters above, it is also beneficial for us to compare the overall effect of mixing that is accumulated in the entire evolution cycle. To do this, we firstly define the accumulated irreversible fluxes $\mathcal {M}^{acc}_\varTheta$, $\mathcal {M}^{acc}_S$, $\mathcal {M}^{acc}$, accumulated viscous dissipation ratio $\varepsilon ^{acc}$ and accumulated flux ratios $\varGamma ^{acc}_\varTheta$, $\varGamma ^{acc}_S$, $\varGamma ^{acc}_\rho$ as the time-integral of the associated physical quantities, following:

(4.4a)$$\begin{gather} \mathcal{M}^{acc}_\varTheta= \int_0^{t_{end}}\mathcal{M}_\varTheta \,{\rm d} t, \end{gather}$$

(4.4b)$$\begin{gather}\mathcal{M}^{acc}_S= \int_0^{t_{end}}\mathcal{M}_S \,{\rm d} t, \end{gather}$$

(4.4c)$$\begin{gather}\mathcal{M}^{acc}=\mathcal{M}^{acc}_\varTheta+\mathcal{M}^{acc}_S, \end{gather}$$

(4.4d)$$\begin{gather}\varepsilon^{acc}= \int_0^{t_{end}}\varepsilon \,{\rm d} t, \end{gather}$$

(4.4e)$$\begin{gather}\varGamma^{acc}_\varTheta= \frac{\mathcal{M}^{acc}_\varTheta}{\varepsilon^{acc}} \frac{R_\rho-1}{-1}, \end{gather}$$

(4.4f)$$\begin{gather}\varGamma^{acc}_S= \frac{\mathcal{M}^{acc}_S}{\varepsilon^{acc}} \frac{R_\rho-1}{R_\rho}, \end{gather}$$

(4.4g)$$\begin{gather}\varGamma^{acc}_\rho= \frac{\mathcal{M}^{acc}}{\varepsilon^{acc}}. \end{gather}$$

These accumulated quantities have been evaluated for our simulations to be shown in table 2. Consistent with our discussions above, simulations with $J=0.05$ lead to stronger turbulence and stronger mixing compared with $J=0.12$, which is reflected in the higher values of $|\mathcal {M}^{acc}_\varTheta |$, $\mathcal {M}^{acc}_S$, $\mathcal {M}^{acc}$ and higher $\varepsilon ^{acc}$. The influence of variation of $R_\rho$ we discussed above can also be confirmed in table 2: table 2 shows that simulations with higher $R_\rho$ will have higher values of $\mathcal {M}^{acc}$, which has been well explained in our discussions above using (4.3b). Besides this, it can also be observed that a larger $R_\rho$ will lead to smaller values of both $|\mathcal {M}^{acc}_\varTheta |$ and $\mathcal {M}^{acc}_S$. This can in fact also be explained simply by noting that the two coefficients $1/(R_\rho -1)$ and $R_\rho /(R_\rho -1)$ in (4.3b) are both decreasing functions of $R_\rho$. As $R_\rho$ goes from small values to large values, the system becomes more and more dominated by the salinity stratification and $\mathcal {M}^{acc}_S$ gradually converge to their values in the single-component cases with the corresponding $J$.

Table 2.

Accumulated irreversible heat fluxes $\mathcal {M}^{acc}_\varTheta$, irreversible salt flux $\mathcal {M}^{acc}_S$, total irreversible flux $\mathcal {M}^{acc}$, accumulated viscous dissipation $\varepsilon ^{acc}$, irreversible temperature flux coefficient $\varGamma ^{acc}_\varTheta$, irreversible salt flux coefficient $\varGamma ^{acc}_S$ and total irreversible flux coefficient $\varGamma ^{acc}$ evaluated for our numerical simulations with different $R_\rho$ and $J$. Here, $\mathcal {M}^{acc}_\varTheta$, $\mathcal {M}^{acc}_S$, $\mathcal {M}^{acc}$ and $\varepsilon ^{acc}$ have been non-dimensionalized by the initial kinetic energy $\mathcal {K}_0$.

Although we have explained how the accumulated buoyancy fluxes vary significantly with $R_\rho$, the accumulated flux coefficients for individual components $\varGamma ^{acc}_\varTheta$ and $\varGamma ^{acc}_S$ are not strong functions of $R_\rho$, as shown in table 2. This suggests that $R_\rho$ only influences the overall flux coefficient $\varGamma ^{acc}_\rho$ by changing the participation between two scalars without influencing their individual flux coefficients much. This will be one of the most important conclusions drawn from our analysis, which will be discussed in detail in § 5.

4.3. Secondary instabilities in the doubly diffusive system

In our discussions above, we have assumed that three-dimensional secondary instabilities that control the transition to three-dimensional turbulence may be fully represented in a numerical domain that includes only a single wavelength of the fastest growing mode of linear instability in the streamwise direction. As shown in Mashayek & Peltier (Reference Mashayek and Peltier2013), the path to turbulence can potentially influence the mixing in the system. To this end, we will investigate the detailed secondary instability that our simulations are susceptible to. In the single-component case, the characteristics of these secondary instabilities have been summarized in the work of Mashayek & Peltier (Reference Mashayek and Peltier2012a,Reference Mashayek and Peltierb). In this subsection we will firstly provide a brief review of these secondary instabilities, followed by an analysis of the secondary instability mechanism(s) that govern the turbulence transition in our DNS-based analyses.

The first candidate from the secondary instability ‘zoo’ is the amalgamation instability or pairing instability (Winant & Browand Reference Winant and Browand1974; Pierrehumbert & Widnall Reference Pierrehumbert and Widnall1982; Klaassen & Peltier Reference Klaassen and Peltier1989), which is characterized by the vortex pairing of nearby KH billows. However, vortex merging events have rarely been observed in either oceanographic or atmospheric environments since they are always suppressed by other candidate modes of secondary instability at high Reynolds number. An example of such competing secondary instabilities is the shear-aligned convective instability (Davis & Peltier Reference Davis and Peltier1979; Klaassen & Peltier Reference Klaassen and Peltier1985) which arises due to the overturning of the statically unstable regions inside the vortex cores created by the roll-up of iso-density surfaces during billow growth. Another well-studied secondary instability is the secondary shear instability of the vorticity braid that connects adjacent billows in a horizontally periodic array of such structures (Corcos & Sherman Reference Corcos and Sherman1976; Staquet Reference Staquet1995, Reference Staquet2000). The newest member of the ‘zoo’ of secondary instabilities is the instability (usually named stagnation point instability) which only exist at sufficiently high Reynolds number. Driven by the strain-related deformation of the background flow, the instability grows at the stagnation point on the braid and produces a region of recirculation near the stagnation point which then evolves into turbulence (see Mashayek & Peltier Reference Mashayek and Peltier2013; Salehipour et al. Reference Salehipour, Peltier and Mashayek2015).

In the evolution of KH billows, the route to turbulence is strongly dependent on the Reynolds number of the background flow. For the particular value of $Re=600$ selected for our DNSs, transition to the fully turbulent state is usually obtained through the onset of secondary shear-aligned convective instability in the singly stratified system (see DNSs of Caulfield & Peltier (Reference Caulfield and Peltier2000), for example). However, it is not yet clear whether this is still true in our doubly diffusive system, considering that the introduction of a second stratified component might influence the buoyancy force that causes convective instability. To this end, we performed the same non-separable secondary stability analysis following the methodology initially developed by Klaassen & Peltier (Reference Klaassen and Peltier1985). By analysing the stability properties of the primary KH billow using this methodology (both the description of this methodology and the results obtained by its application are provided in Appendix B), we demonstrated that the dominant mode of secondary instability is indeed the secondary shear-aligned convective instability.

In order to visualize the growth of the secondary instability predicted by the non-separable analysis we plot in figure 7 the streamwise vorticity iso-surfaces for simulation number 5, which contains two fastest growing wavelengths of primary KH instability so that the pairing instability would be captured if it were to emerge. However, the pairing of vortices did not occur in this longer domain and the secondary shear-aligned convective instability remains the dominant mode among the zoo of secondary instabilities. The growth of the secondary shear-aligned convective instability can be clearly identified in the convective rolls that are aligned with the background shear in figure 7(a). These convection rolls have previously been seen in the DNS analysis of Caulfield & Peltier (Reference Caulfield and Peltier2000) and Mashayek & Peltier (Reference Mashayek and Peltier2013) for example and now also in our analyses of KH billow mediated transition in the doubly diffusive system. As time evolves, the interaction between neighbouring rolls drives the system into the three-dimensional turbulent state and eventually relaminarization, as shown in figure 7(b–d).

Figure 7.

Streamwise vorticity iso-surfaces of $\omega _x=0.2$ (red) and $\omega _x=-0.2$ (blue) for simulation number 5.

5.1. Dependence of diapycnal diffusivities on governing non-dimensional parameters

It has been widely accepted that the buoyancy Reynolds number $Re_b$ is the most important non-dimensional parameter that influences the diapycnal diffusivities (e.g. Caulfield Reference Caulfield2021). We will therefore evaluate the irreversible diapycnal diffusivities $K^{irr}_\varTheta$ and $K^{irr}_S$ in the fully turbulent regime ($t_{3dmax}< t< t_{end}$) of each of our DNSs and plot them as a function of $Re_b$ at each time, as shown in the scatter plot in figure 8(a,c). The corresponding irreversible flux coefficients $\varGamma ^{irr}_\varTheta$ and $\varGamma ^{irr}_S$ are shown in figure 8(b,d) and the diffusivity ratio is shown in figure 8(e). It will be apparent that, for our simulations with $R_\rho =\infty$, the temperature field is not active in the simulation and thus the $K^{irr}_\varTheta$ data (and also $d$) are not applicable in these simulations. Simulations with different bulk Richardson numbers achieved a distribution of buoyancy Reynolds number in the range from 20 to 100, which perfectly captures the environment of the central Canada Basin region of the Arctic Ocean which is characterized by low energy turbulence with $Re_b<100$ (see the most recent estimations of Dosser et al. (Reference Dosser, Chanona, Waterman, Shibley and Timmermans2021), for example).

Figure 8.

Irreversible diapycnal diffusivities $K^{irr}_S$ (a) $K^{irr}_\varTheta$ (c), irreversible mixing efficiencies $\varGamma ^{irr}_S$ (b) $\varGamma ^{irr}_\varTheta$ (d) and diffusivity ratio (e) evaluated for the fully turbulent regime of DNSs plotted as a function of $Re_b$. Each scatter point represents the average value over a non-overlapping time interval of five non-dimensional units. The solid line shows the parametrization of the above values in the work of Bouffard & Boegman (Reference Bouffard and Boegman2013). The three vertical dashed lines represent the three critical values of $Re_b$ that separate four different regimes of the Bouffard & Boegman (Reference Bouffard and Boegman2013) parametrization scheme.

Scatter plots in figure 8 shows that both $K^{irr}_\varTheta$ and $K^{irr}_S$ are almost monotonically increasing functions of $Re_b$, despite the fact that different values of $J$ and $R_\rho$ are employed in these simulations. In fact, our simulations with $J=0.05$ is characterized by higher $Re_b$ compared with the $J=0.12$ cases, due to the weaker stratification employed. Figure 8 demonstrates that the bulk Richardson number $J$ is only contributing to the diapycnal diffusivities through its influence on $Re_b$, thus there is no need to consider an explicit dependence on $J$. At the same time, different values of $R_\rho$ do not significantly change the dependence on $Re_b$ either, suggesting that $K^{irr}_\varTheta$ and $K^{irr}_S$ do not strongly depend on $R_\rho$. This is a somewhat unusual result considering that past simulations of diffusive–convection interfaces have always revealed a strong functional dependence of diapycnal diffusivities on $R_\rho$ (see Caro (Reference Caro2009), Carpenter, Sommer & Wüest (Reference Carpenter, Sommer and Wüest2012), Flanagan, Lefler & Radko (Reference Flanagan, Lefler and Radko2013) and Brown & Radko (Reference Brown and Radko2021) for example). The key differences should be understood as follows: our current system is a dynamically driven (specifically shear-driven) system and it is the turbulence generated from the background shear that causes mixing for both temperature and salinity. For these previous simulations on the diffusive interface, on the other hand, the macroscopic motions are mainly induced by the release of potential energy from the unstably stratified component (temperature component) of the double-diffusive system and such systems should be recognized as buoyancy-driven systems (the system of Brown & Radko (Reference Brown and Radko2021) is simultaneously driven by buoyancy and shear). Since $R_\rho$ controls the relatively strength of the stratification of stably stratification component over unstably stratified component, it is apparent that variations of $R_\rho$ should strongly influence the vertical mixing in buoyancy-driven systems. Therefore, no conflicts exist by showing that $K^{irr}_\varTheta$ and $K^{irr}_S$ are weakly dependent on $R_\rho$ in our dynamically driven system.

It should furthermore be mentioned that the exiting parametrization scheme of diapycnal diffusivities implemented in global ocean models has always assumed a functional dependence of $R_\rho$ (see the kappa profile parametrization (KPP) parametrization of Large, McWilliams & Doney (Reference Large, McWilliams and Doney1994) and Kelley (Reference Kelley1990), for example). Such parametrization schemes have been established based on the assumption that a series of thermohaline staircases will be formed in the diffusive–convection environment and the fluxes across the diffusive interface staircases (which have been regarded in the buoyancy-driven system as mentioned above) are strongly dependent on $R_\rho$ (see Marmorino & Caldwell (Reference Marmorino and Caldwell1976) and Linden & Shirtcliffe (Reference Linden and Shirtcliffe1978) for example). As discussed in Peltier, Ma & Chandan (Reference Peltier, Ma and Chandan2020), the conventional parametrization scheme for diapycnal diffusivity under conditions of diffusive–convection water column stratification may lead to a significant over-estimation of diapycnal diffusivities when inserted into an enhancement to diapycnal diffusivity based upon the assumption that a staircase has formed even if the turbulence level is so high that the staircase would not be able to form. In this scenario, a parametrization based on the dynamically driven system (that will be discussed in the following subsection) should be employed instead.

5.2. Comparison with the existing turbulent parametrization of Bouffard & Boegman (Reference Bouffard and Boegman2013)

In fact, the weak dependence of $K^{irr}_\varTheta$ and $K^{irr}_S$ on $R_\rho$ essentially suggests that the temperature field and salinity field are weakly coupled in the development of turbulence, they react to the background shear, stratified turbulence and buoyancy forcing as if they are the only diffusing species in the system. It is therefore of great interest to compare our results for the dependence of these diffusivities upon buoyancy Reynolds number with those previously published for single-component systems. To be useful for our purposes, such a parametrization would have to include explicit dependence on the Prandtl number (Schmitt number) to provide different parametrizations for temperature and salinity. To our knowledge, the only turbulent parametrization scheme that stresses the differences in the Prandtl number (Schmitt number) is that based upon the recent work of Bouffard & Boegman (Reference Bouffard and Boegman2013) (hereafter BB). By examining extensive sets of published data from both laboratory experiments (e.g. Jackson & Rehmann Reference Jackson and Rehmann2003; Rehmann & Koseff Reference Rehmann and Koseff2004) and DNSs (e.g. Shih et al. Reference Shih, Koseff, Ivey and Ferziger2005; Smyth et al. Reference Smyth, Nash and Moum2005) on single-component fluids with either the Prandtl number for temperature or the Schmidt number for salinity, BB extend previous parametrizations of Shih et al. (Reference Shih, Koseff, Ivey and Ferziger2005) to incorporate a proper dependence upon $Pr$ into their parametrization scheme. Their scheme therefore dependent on both $Re_b$ and $Pr$ ($Sc$) as

(5.1)\begin{equation} K^{BB}_\rho(Re_b, Pr)= \begin{cases} \kappa, & \text{if}\ Re_b<10^{{2}/{3}}Pr^{-({1}/{2})},\\ \dfrac{0.1}{Pr^{{1}/{4}}}\nu Re_b^{{3}/{2}}, & \text{if}\ 10^{{2}/{3}}Pr^{-({1}/{2})}< Re_b<(3 \ln \sqrt {Pr})^2, \\ 0.2 \nu Re_b, & \text{if}\ (3 \ln \sqrt {Pr})^2< Re_b<100, \\ 2 \nu Re_b^{{1}/{2}}, & \text{if}\ Re_b>100. \end{cases} \end{equation}

In the above parametrizations, the diapycnal diffusivities have a different power law dependence on $Re_b$ in different ranges of $Re_b$. The smallest $Re_b$ regime is the molecular regime in which molecular diffusivities are assumed. The second regime, the buoyancy-controlled regime, (which is originally included in BB) describes the regime in which mixing is strongly influenced by the Prandtl number. In this regime, the diapycnal diffusivities increase rapidly with $Re_b$ at the rate of $Re_b^{3/2}$. The third regime is the transitional regime which is consistent with the classic Osborn model with the flux coefficient fixed to 0.2. For $Re_b$ higher than 100 the system enters the energetic regime in which diffusivities scale with $Re_b^{1/2}$, in accordance with previous work of Shih et al. (Reference Shih, Koseff, Ivey and Ferziger2005).

BB's parametrization described above is evaluated at $Pr(Sc)=70$ and $Pr=7$ separately for different $Re_b$ and plotted as the solid line in figure 8(a,c). A strikingly good fit can be identified in these figures: for $K^{irr}_S$, a close match between the parametrization and our DNS data can be found, except for the tail of low $Re_b$. In fact, our DNS data strongly support the existence of a large buoyancy-controlled regime for salinity ($5< Re_b<70$) in which $K_S$ scales at $Re_b^{3/2}$. When $Re_b$ drops below approximately 3, however, $K^{irr}_S$ drops to the level of the molecular value in a fashion that is much faster than $Re_b^{3/2}$, suggesting a possible overestimation of (5.1) in the low $Re_b$ range. For the temperature field, the parametrization seems to produce a slight overestimation of the diapycnal diffusivities. However, the different power laws for the buoyancy-controlled regime and transition regime can be clearly identified in our DNS data, demonstrating the reasonableness of the partition of regimes of BB. Meanwhile, BB's prediction for flux coefficients as a function form of $Re_b$ are plotted in figure 8(b,d) to be compared with our numerical data. It can also be seen from these two plots that the functional dependence of $\varGamma ^{irr}_S$ and $\varGamma ^{irr}_\varTheta$ over $Re_b$ follows well from BB, although the values of $\varGamma ^{irr}_S$ and $\varGamma ^{irr}_\varTheta$ are somewhat smaller than the predicted values of BB. Furthermore, we compare the parametrized diffusivity ratio (shown in figure 5 of BB) with our DNS data in figure 8(c) and again find a good match. Also consistent with previous work of Smyth et al. (Reference Smyth, Nash and Moum2005), the diffusivity ratio only reaches unity when $Re_b$ reaches the level of $O(100)$, otherwise strong differences in the diffusivity ratio between temperature and salinity exist. We interpret these close fits to a parameterization scheme for single-component systems comprised of a species with Prandtl number 7 and another single diffusing species with Schmidt number much higher (70 instead of the actual Schmidt number for salt of 700) to fully validate our conclusion that in the diffusive convection regime of the Arctic Ocean the turbulent diffusivities for temperature and salinity operate independently. This is a critical conclusion as it was upon this assumption that our recently published new theory for the formation of the previously unexplained thermohaline staircases in the Arctic Ocean has been based (Ma & Peltier Reference Ma and Peltier2022).

In ending this section, further comment is warranted on two subtleties connected to the preceding analyses. First, it is important to note that the BB parameterization is based upon a combination of experimental/DNS data (e.g. Jackson & Rehmann Reference Jackson and Rehmann2003; Shih et al. Reference Shih, Koseff, Ivey and Ferziger2005) that are evaluated based on the conventional definitions of $K_\varTheta$ and $K_S$. As $K_\varTheta$ and $K_S$ are determined in quasi-steady states of these systems, it is reasonable to assume that they are consistent with the irreversible definitions $K^{irr}_\varTheta$ and $K^{irr}_S$. The KH system that has been studied here, on the other hand, is a transiently evolving system that does not reach a quasi-steady state; $K_\varTheta$ and $K_S$ are highly variable quantities that frequently obtain negative values because they are strongly influenced by the reversible stirring process of the KH billow which does not contribute to turbulent diffusivity. Therefore, we have employed the instantaneous values of the turbulence data to compute the irreversible vertical diffusivities $K^{irr}_\varTheta$ and $K^{irr}_S$ instead of $K_\varTheta$ and $K_S$ as in our parametrization study. A second issue that warrants comment concerns the question of the impact on mixing in the event that iso-surfaces of salinity and temperature are not parallel and perpendicular to the local gravitational acceleration. This is the circumstance that attends the existence of so-called thermohaline intrusions that have been suggested previously as an explanation (Bebieva & Timmermans Reference Bebieva and Timmermans2017) for the thermohaline staircases observed in the polar oceans in regions where cold and fresh water overlies relatively warm and salty water. Although our hypothesis in Ma & Peltier (Reference Ma and Peltier2022) obviated the need to invoke such exotic circumstances, there continues to be interest in what the mixing properties might be in this situation (e.g. see the model of Middleton & Taylor (Reference Middleton and Taylor2020) as well as chapter 7 of Radko (Reference Radko2013) for a review). In this circumstance, the turbulent diffusivities $K_\varTheta$ and $K_S$ can differ with the irreversible diffusivities $K^{irr}_\varTheta$ and $K^{irr}_S$ even if the system is in a quasi-steady state.

5.3. An algorithm for the determination of diapycnal diffusivities in the stratified turbulence

In the practical measurement of turbulence and mixing in the Arctic Ocean, there are generally two most critical physical quantities that are especially important to understand: the diapycnal diffusivities for density $K_\rho$ and the vertical heat flux $F_H$. In the recent work on direct or indirect measurements in the Arctic Ocean (for example, Chanona, Waterman & Gratton Reference Chanona, Waterman and Gratton2018; Scheifele et al. Reference Scheifele, Waterman, Merckelbach and Carpenter2018, Reference Scheifele, Waterman and Carpenter2021; Chanona & Waterman Reference Chanona and Waterman2020; Dosser et al. Reference Dosser, Chanona, Waterman, Shibley and Timmermans2021), a critical level of $Re^{cr}_b=10$ or $Re^{cr}_b=20$ is usually chosen to differentiate the turbulent regime from the molecular regime. In the molecular regime the difference between the molecular diffusion for temperature and salinity is identified so that $K_\rho =R_\rho /(R_\rho -1) \kappa _s -1/(R_\rho -1) \kappa _\theta$. In the turbulent regime, however, the canonical Osborn formula $K_\rho =K_\varTheta =0.2 \nu Re_b$ we discussed in § 2 has been used to estimate both $K_\rho$ and $K_\varTheta$; $K_\varTheta$ is then further used to estimate the heat flux.

Based on our DNS results, at least two major sources of systematic errors in this standard procedure may be identified in the determination of $K_\rho$ based on the current algorithm described above. First, the water column density is mostly influenced by the salinity, whose diapycnal diffusivity $K^{irr}_S$ has a $Re_b^{3/2}$ dependence in the vast range of buoyancy-controlled regime $(0.17< Re_b<96)$ as predicted by taking $Sc=700$ in (5.1). Despite a smaller value of $Sc=70$ applied in our DNS, our data confirmed that such a 3/2 power law does exist in a wide range of the parameter space $(5< Re_b<60)$. For such a wide range of $Re_b$ (in fact a significant proportion of the turbulent measurements in the Arctic lie in this range of $Re_b$, see Dosser et al. (Reference Dosser, Chanona, Waterman, Shibley and Timmermans2021) for example), the Osborn formula suggesting a linear dependence on $Re_b$ by mistake thus can lead to a strong over-estimation of $K^{irr}_S$ (considering that $Re_b^{3/2}$ dependence and $Re_b$ dependence overlap at approximately $Re_b=100$). Second, even though $K^{irr}_\rho$ is usually similar to $K^{irr}_S$, as shown in the previous section, our rigorous derivation in (2.14) shows that $K^{irr}_\rho$ depends upon both $K^{irr}_S$ and $K^{irr}_\varTheta$ through the relationship (2.14b). Therefore, the true value of $K^{irr}_\rho$ should be even smaller than the estimation from $K^{irr}_S$, especially when $R_\rho$ is low. Such differences of $K^{irr}_\rho$ and $K^{irr}_S$ are clearly apparent in our figure 5. For the above reasons, the simplified algorithm that is currently used in the oceanographic measurement literature can lead to a large overestimate of $K_\rho$ due to the existence of two error sources both of which exaggerate $K_\rho$.

Despite the systematic errors in $K_\rho$ estimation mentioned above, the traditional method gives relatively better estimates in terms of the temperature diapycnal diffusivity $K_\varTheta$. In fact, at $Pr=7$ for the temperature field, BB's parametrization agrees with the canonical Osborn formula for a wide range of values of the buoyancy Reynolds number $(9< Re_b<100)$. However, an overestimation of $K_\varTheta$ is still present at smaller $Re_b$ $(Re_b<9)$ and therefore the estimation of the heat flux derived from $K_\varTheta$ based on Osborn's formula may still lead to exaggeration in the low turbulence environment.

Given our analysis above, we propose the following simple three-step algorithm to be employed for evaluate the diapycnal diffusivities for density as well as heat fluxes in measurements in the Arctic:

1. (i) Calculate $K_S$ and $K_\varTheta$ based on the parametrization of BB in (5.1). Replace $K_S$ with the molecular diffusivity $\kappa _s$ once $Re_b$ drops below a critical value of $Re_b^{cr}=5$.

2. (ii) Use the vertical derivatives of scalars $S_z$ and $\varTheta _z$ to evaluate $R_\rho =\beta S_z/ \alpha \varTheta _z$ to calculate $K_\rho$ in individual water columns based on (2.14b), which is restated here as

   (5.2)\begin{equation} K_\rho=\frac{R_\rho}{R_\rho-1}K_S-\frac{1}{R_\rho-1}K_\varTheta. \end{equation}

3. (iii) Infer the heat flux $F_H$ based Fick's law using the local temperature gradient $\varTheta _z$ and the estimation of $K_\varTheta$ from step (i).

In the above algorithm, a critical buoyancy Reynolds number $Re^{cr}_b$ is kept in the first step by recognizing that the BB parametrization may yield overestimation of the $K_S$ in the low $Re_b$ regime. We expect this algorithm to be employed in future estimation of diapycnal diffusivities based on the measurements of the viscous dissipation ratio.