1. Introduction
When a lighter fluid is accelerated into a heavier fluid or conversely when a heavier fluid is decelerated by a lighter fluid, the well-known Rayleigh–Taylor instability (RTI) occurs and the small perturbations at the interface grow exponentially (Rayleigh Reference Rayleigh1882; Taylor Reference Taylor1950). The evolution of RTI induces the so-called Rayleigh–Taylor (RT) turbulence, which widely occurs in natural phenomena and industrial applications, including atmospheric physics (Caproni et al. Reference Caproni, Lanfranchi, da Silva and Falceta-Gonçalves2015), deep ocean circulation (Cui & Street Reference Cui and Street2004), supernova explosions (Gelfand, Slane & Zhang Reference Gelfand, Slane and Zhang2009) and inertial confinement fusion (Bodner Reference Bodner1974). As a canonical problem of non-stationary and non-isotropic turbulence, RT turbulence has served as a compelling fluid-mechanical prototype for studying the evolution of flow into turbulence, driven by full-scale energy injection from gravity. Particularly, an important issue in RT turbulence is understanding how the energy is generated, transferred and dissipated. By leveraging theoretical analysis and high-fidelity simulations that fully resolve all turbulence scales, the research community has extensively studied the energy transport process in standard non-rotating RT turbulence (see e.g. Zhou Reference Zhou2017a , Reference Zhoub ; Zhou et al. Reference Zhou2021).
In particular, the energy budget has been demonstrated as an effective tool for elucidating energy conservation in RT turbulence. Specifically, by evaluating the energy budget, Cabot & Cook (Reference Cabot and Cook2006) showed that kinetic energy is converted from potential energy and dissipated by viscous effects for incompressible RT turbulence. However, upon further examination of the compressibility effects on RT turbulence under density stratification, pressure–dilatation work emerges as a key contributor to the generation of kinetic energy in the early stage of RT evolution (Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a ). Moreover, along with energy generation, the scale-to-scale transfer of energy is fundamentally important to the dynamics of the flow. Indeed, RT turbulence exhibits the coexistence of fine-scale structures and large patches of mixed fluids, and therefore the scale-to-scale transfer has attracted much research interest. Based on a theoretical model of the energy transfer processes, Chertkov (Reference Chertkov2003) adopted the Bolgiano–Obukhov (BO59) scenario to resolve the small-scale spatial–temporal correlations of the velocity and temperature fields. Furthermore, in the inertial range of two-dimensional (2-D) RT turbulence, a backward transfer mechanism for kinetic energy was revealed by Celani, Mazzino & Vozella (Reference Celani, Mazzino and Vozella2006), which was further demonstrated by numerical investigations on the scale-to-scale transfer of kinetic energy and enstrophy of Zhou et al. (Reference Zhou, Huang, Lu, Liu and Ni2016a ). However, in three-dimensional (3-D) RT turbulence, the classical forward cascade of kinetic energy and enstrophy is recovered, namely they are transferred from large to small scales (Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a ). Soulard (Reference Soulard2012) adapted the Monin–Yaglom relation to RT turbulence both in 3-D cases, which confirms the Kolmogorov–Obukhov theory, and 2-D cases, where it recovers the BO59 scenario. Moreover, it is shown in Zhao, Betti & Aluie (Reference Zhao, Betti and Aluie2022) that energy transfers downscale in 3-D cases and upscale in 2-D cases. Another significant aspect of RT turbulence is the dynamics of vorticity and associated generation and transfer of enstrophy (Sengupta, Ulloa & Joshi Reference Sengupta, Ulloa and Joshi2023). It has been found that baroclinicity acts as the primary mechanism for the generation of enstrophy in the initial stage, while the vortex stretching and tilting term assumes a dominant role in the late stage (Cabot & Cook Reference Cabot and Cook2006; Zhao & Li Reference Zhao and Li2025). Furthermore, based on the directional enstrophy transport equation, Zhao et al. (Reference Zhao, N.-S. and X.-Y.2020a ) identified that horizontal enstrophy is primarily generated by the baroclinic effect, while vertical enstrophy arises from vortex stretching and tilting.
As discussed above, dedicated efforts have been made to obtain a mechanistic understanding of kinetic energy generation and transfer in non-rotating RT turbulence. However, for geophysics (Pouquet & Marino Reference Pouquet and Marino2013), and some engineering applications (Chen, Schrock & Peterson Reference Chen, Schrock and Peterson1997), the system rotation plays a significant role in flow dynamics. Rotating RT turbulence has recently received considerable attention due to its occurrence in astrophysical-scale flows (Scase & Sengupta Reference Scase and Sengupta2021). Additionally, there are numerous scenarios where it is desirable to stabilise a system against RTI through long-range effects (Baldwin, Scase & Hill Reference Baldwin, Scase and Hill2015). For example, rotation may potentially facilitate further control of RTI in inertial confinement fusion (Betti et al. Reference Betti, Goncharov, McCrory and Verdon1998; Zhang et al. Reference Zhang, Betti, Yan and Aluie2020) and magnetised target fusion (Buyko et al. Reference Buyko, Garanin, Mokhov and Yakubov1997). Indeed, the presence of rotation introduces a restoring force, i.e. the Coriolis force, on fluid elements moving perpendicular to the axis of rotation. In terms of the evolution of rotating RT flows, previous numerical and experimental investigations have demonstrated the Coriolis-force-induced stabilising effect on RTI. On the one hand, a series of experiments carried out by Baldwin et al. (Reference Baldwin, Scase and Hill2015) demonstrated a rotation-induced inhibitory effect on the growth of RTI. In particular, long-wavelength instabilities are stabilised by system rotation. These conclusions are further substantiated in experiments conducted by Scase et al. (Reference Scase, Baldwin and Hill2017, Reference Scase, Baldwin and Hill2020). On the other hand, numerical studies have facilitated a better understanding of the stabilising effects of rotation in RT flows. Specifically, Carnevale et al. (Reference Carnevale, Orlandi, Zhou and Kloosterziel2002) revealed that the formation of vortex rings can be hindered by introducing rotation at the early stage. As a consequence, RTI can be effectively prevented if a sufficiently strong rotation is applied. Moreover, Boffetta, Mazzino & Musacchio (Reference Boffetta, Mazzino and Musacchio2016) revealed the main fundamental features of rotating RT turbulence. They found that the rotation significantly weakens the intensity of turbulent velocity fluctuations, thereby impeding the growth of the mixing layer (Boffetta et al. Reference Boffetta, Mazzino and Musacchio2016). Additionally, in the presence of rotation the conversion of potential energy into turbulent kinetic energy is found to be less effective (Boffetta et al. Reference Boffetta, Mazzino and Musacchio2016). However, the process of fluid mixing is facilitated by the action of rotation (Wei et al. Reference Wei, Li, Wang, Yang, Zhu, Qian and Luo2022). Analytical and experimental studies have also considered rotating RT flow in which the rotation axis is perpendicular to the direction of external acceleration inducing RTI. It is observed that rotation exerts a stabilising influence on RTI across arbitrary Atwood numbers (Tao et al. Reference Tao, He, Ye and Busse2013; Scase & Hill Reference Scase and Hill2018; Scase & Sengupta Reference Scase and Sengupta2021). Here, the Atwood number is defined as
$A_t=(\rho _h-\rho _l)/(\rho _h+\rho _l)$
with
$\rho _h$
and
$\rho _l$
being the heavy and light fluid densities, respectively. In addition, experimental results show that RTI can be suppressed by the rotation of a sufficient angular velocity at the inner surface of an imploding liquid shell (Huneault, Plant & Higgins Reference Huneault, Plant and Higgins2019). More recently, Zhao, Aluie & Li (Reference Zhao, Aluie and Li2025) investigated the role of small-scale structures in turbulent RT flows, and revealed that eliminating intense small-scale motion leads to the formation of more organised and coherent flow structures, with reduced mixing and enhanced anisotropy. This suggests that RT flows under rotation, where the suppression of small-scale motions plays a critical role, may exhibit unique flow structures.
Despite the fact that the stabilising effect of rotation has been extensively studied, little is known about the dynamics of energy and enstrophy in rotating RT turbulence. Indeed, across a broad range of flow systems, complex behaviours of energy transfer have been observed in rotating turbulence. An overview by Alexakis & Biferale (Reference Alexakis and Biferale2018) indicated that rotation causes a transition in energy transfer for rotating turbulence, shifting from a forward energy cascade to a split energy cascade. The split energy transfer refers to the phenomenon where part of the injected energy goes to small and part to large scales. For turbulence in a confined thin fluid layer, Boffetta (Reference Boffetta2023) demonstrated that confinement and rotation produce an inverse cascade of energy by a similar mechanism which suppresses enstrophy production at large scales. Interestingly, based on laboratory experiments of wave-driven rotating turbulence, Brunet, Gallet & Cortet (Reference Brunet, Gallet and Cortet2020) showed that resonant quartets of inertial waves can trigger an instability known as the ‘quartetic instability’. The quartetic instability induces a transfer of energy to the slow modes, thus restricting the domain of validity of weak turbulence theory (Brunet et al. Reference Brunet, Gallet and Cortet2020). Shaltiel et al. (Reference Shaltiel, Salhov, Gat and Sharon2024) employed a short, abrupt increase in energy injection rate into steady strongly driven rotating turbulent flow as a probe for energy transfer in the system. They found that the inverse cascade of energy is mediated by inertial waves that propagate within the flow with small, but non-vanishing frequency (Shaltiel et al. Reference Shaltiel, Salhov, Gat and Sharon2024).
Moreover, rotation has been shown to introduce novel mechanisms of energy transfer. Specifically, rotation is invariably accompanied by helicity, which is defined as
$H=\boldsymbol{u\boldsymbol{\cdot }\omega }$
(Yang & Wu Reference Yang and Wu2012). Here
$\boldsymbol{u}$
and
$\boldsymbol{\omega }$
are velocity and vorticity, respectively. Typically, helicity is used to quantify the degree of knottedness or linkage of vortex lines (Moffatt Reference Moffatt1969) and the degree of broken symmetry (Betchov Reference Betchov1961). In inviscid flows with localised vorticity distributions, the total helicity is invariant (Polifke & Shtilman Reference Polifke and Shtilman1989). In unstably stratified turbulence, Agoua et al. (Reference Agoua, Favier, Delache, Briard and Bos2021) demonstrated that helicity generation can be piloted by the interaction between skew diffusion and helicity. For this mechanism to be effective, the flow requires an unstable density gradient and a body force that enables the system to attain a close-to-2-D flow state (Agoua et al. Reference Agoua, Favier, Delache, Briard and Bos2021). Indeed, the energy transfer process in turbulence is investigated by scrutinising the behaviour of helicity. Specifically, Kraichnan (Reference Kraichnan1973) presented the absolute equilibrium spectrum of the Euler equation and demonstrated that helicity acts to suppress overall energy transfer. Research on the influence of helicity on turbulence at high Reynolds numbers dates back to André & Lesieur (Reference André and Lesieur1977). They demonstrated that in isotropic turbulence at high Reynolds numbers, the helicity cascade inhibits energy transfer towards large wavenumbers, consistent with the conclusion of Kraichnan (Reference Kraichnan1973). Biferale, Musacchio & Toschi (Reference Biferale, Musacchio and Toschi2012) presented theoretical and numerical evidence that a decimated version of the Navier–Stokes equations, such that only modes with a given sign of helicity are retained, displays an inverse energy transfer mechanism. In decaying homogeneous helical turbulence, Briard & Gomez (Reference Briard and Gomez2017) demonstrated that helicity slows down the kinetic energy transfer in the early stage of the decay, while the effects of helicity on the kinetic energy decay are rather weak once the turbulence is fully developed. Subsequently, Buzzicotti et al. (Reference Buzzicotti, Aluie, Biferale and Linkmann2018a
) found that the upscale cascade at wavenumbers close to the forcing scale is generated by increasingly dominant homochiral interactions in turbulence under rotation. In addition, Buzzicotti et al. (Reference Buzzicotti, Aluie, Biferale and Linkmann2018a
) further revealed that while the mean transfer is produced inside regions of strain, the rare but extreme events of energy transfer occur primarily inside the large-scale column vortices. A series of studies on 3-D helical isotropic turbulence have also shown that helicity can impede the forward energy cascade and even result in the reversal of energy transfer (Moffatt Reference Moffatt2014; Stepanov et al. Reference Stepanov, Golbraikh, Frick and Shestakov2015; Sahoo, Alexakis & Biferale Reference Sahoo, Alexakis and Biferale2017; Słomka & Dunkel Reference Słomka and Dunkel2017). Furthermore, a two-channel theory of the helicity cascade has been developed: the first channel is mainly due to the vortex twisting process, and the second channel is mainly associated with the vortex stretching process. The latter is proposed as a novel mechanism of the hindered or even inverse energy cascades in rotating turbulence (Yan, Li & Yu Reference Yan, Li and Yu2020a
,
Reference Yan, Li, Yu, Wang and Chenb
). The aforementioned observations indicate that the mixing layer evolution and the flow features in rotating RT turbulence, governed by the combined effects of helicity, anisotropy and non-stationarity, are complex and warrant further investigation. To this end, regarding the generation and transfer of kinetic energy and enstrophy, intriguing mechanisms that are dramatically different from those of non-rotating RT turbulence can be expected.
In the present study, direct numerical simulations (DNS) of 3-D rotating RT turbulence are performed, aiming to provide a mechanistic understanding of the rotation-induced alteration in the dynamics of kinetic energy and enstropy. The remainder of this paper is organised as follows. The details of the numerical simulation are described in § 2. The generation of kinetic energy and enstrophy is reported in § 3. The scale-to-scale transfer of kinetic energy is analysed in § 4. Finally, concluding remarks are presented in § 5.
2. Numerical simulations
2.1. Governing equations and numerical methods
We consider the rotating RT turbulence generated in a single-species two-density fluid. The density
$\rho _I^*=(\rho _h^*+\rho _l^*)/2$
and pressure
$p_I^*$
at the initial interface are chosen as characteristic scales, where
$\rho _h^*$
and
$\rho _l^*$
are the densities of heavy and light fluids, respectively. Then, the characteristic velocity and temperature are given by
$U_I^*=\sqrt {p_I^*/\rho _I^*}$
and
$T_I^*=p_I^*/\mathcal{R}^*\rho _I^*$
with
$\mathcal{R}^*$
being the perfect gas constant. The vertical length (
$L_z^*$
) is used as the characteristic length. Note that the superscript ‘
$*$
’ denotes dimensional physical quantities. The initial interface coincides with the horizontal plane at
$z_I=0$
, which is perpendicular to the directions of gravity
$\boldsymbol{g^*}=(0,0,-g^*)$
and of the uniform rotation speed
$\boldsymbol{\varOmega ^* }=(0,0,\varOmega ^* )$
. Here,
$g^*$
is the gravitational acceleration and
$\varOmega ^*$
is the angular speed of the reference frame. According to previous studies on flow under gravity, the centrifugal force can be considered negligible compared with gravity (Khan, Tak & Sharma Reference Khan, Tak and Sharma2011; Mizerski & Tobias Reference Mizerski and Tobias2011; Horn & Schmid Reference Horn and Schmid2017; Liu et al. Reference Liu, Wan, Yan, Sun and Sun2019; Wang et al. Reference Wang, Liu, Wan and Sun2021). Furthermore, energy transfer in rotating turbulence is significantly influenced by the Coriolis force (Alexakis & Biferale Reference Alexakis and Biferale2018; Buzzicotti et al. Reference Buzzicotti, Aluie, Biferale and Linkmann2018a
,
Reference Buzzicotti, Clark Di Leoni and Biferaleb
). To this end, we omit the centrifugal term and concentrate solely on the effects of the Coriolis force in the present study, similar to previous research on rotating RT turbulence (Boffetta et al. Reference Boffetta, Mazzino and Musacchio2016; Scase, Baldwin & Hill Reference Scase, Baldwin and Hill2017; Wei et al. Reference Wei, Li, Wang, Yang, Zhu, Qian and Luo2022). Thus, the dimensionless Navier–Stokes equations in a reference frame rotating about the
$z$
axis are expressed as follows (Khan et al. Reference Khan, Tak and Sharma2011; Mizerski & Tobias Reference Mizerski and Tobias2011; Vidal & Cébron Reference Vidal and Cébron2020):
Here
$\rho$
is fluid density,
$p$
is pressure,
$\boldsymbol{u}=(u_1,u_2,u_3)$
is velocity,
$T$
is temperature and
$e=T/(\gamma -1)+u_i u_i/{2}$
denotes the specific total energy with the ratio of specific heat
$\gamma =1.4$
. The viscous stress
$\sigma _{\textit{ij}}$
is defined as
where the viscosity coefficient
$\mu =(T)^{3/2}({1+c})/({T+c})$
is computed by the Sutherland law with
$c=T_s^*/T_I^*$
, where the Sutherland temperature
$T_s^*=124\,\mathrm{K}$
and characteristic temperature
$T_I^*=273.15\,\mathrm{K}$
. The dilatation related to the compression/expansion motions of fluid parcels is
$\theta =\partial u_i / \partial x_i$
. The dimensionless parameters in (2.1)–(2.3), namely the Reynolds, Prandtl, Froude and Rossby numbers, are defined as
respectively. Moreover, the ideal gas law
$p=\rho T$
is adopted to close the governing equations.
In the present study, we have developed an in-house code with high-order accuracy schemes to solve the above governing equations. Specifically, the convection terms are discretised by the seven-order weighted essentially non-oscillatory scheme, and the diffusion terms are discretised using the six-order central difference scheme. The time derivatives are approximated by the standard third-order Runge–Kutta method. In addition, we have carefully examined the numerical approach used in this study and verified that the simulated results are reliable based on our previous work on RT flows (Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a , Reference Zhao, Wang, Liu and Lub ; Fu et al. Reference Fu, Zhao, Xu, Wang, Liu, Wan and Lu2022, Reference Fu, Zhao, Wang, Liu, Wan and Lu2023).
2.2. Flow-field initialisation and grid resolution tests
In this study, the governing equations are solved in a 3-D rectangular domain in Cartesian coordinates. The heavy fluid is set at uniform temperature
$T_{\textit{up}}=1/(1+A_t)$
above the interface (
$z\gt 0$
) and the lighter fluid is maintained at
$T_{\textit{low}}=1/(1-A_t)$
below the interface (
$z\lt 0$
), where the Atwood number
$A_t$
is defined as
The initial velocity field is set to be zero, and initial hydrostatic equilibrium requires
By combining the ideal gas equation of state with initial temperature to solve the above equation, the initial density can be determined as follows:
\begin{equation} \rho (z)=\left \{ \begin{aligned} &(1-A_t)\mathrm{exp}\left ( - \frac {1-A_t}{\textit{Fr}}z\right ), \quad z\lt 0 , \\ &(1+A_t)\mathrm{exp}\left ( - \frac {1+A_t}{\textit{Fr}}z\right ),\quad z\gt 0 . \end{aligned} \right . \end{equation}
Moreover, to prevent strong discontinuities in the initial field, and following previous studies (Gauthier Reference Gauthier2017; Luo & Wang Reference Luo and Wang2021), the initial density distribution is set as
where
$H^+(z)=1+\mathrm{erf}(z)/2,\ H^-(z)=1-\mathrm{erf}(z)/2$
,
$\mathrm{erf}(z)$
is the error function and
$\delta$
is the initial mixing thickness. For the present simulations,
$\delta$
is set to
$\delta =8\varDelta$
, where
$\varDelta$
is the grid width. The interface at
$z=0$
is perturbed via a multi-mode perturbation given by
$I(x,y)=\sum _{k_x , k_y} A_m\cos (2\pi k_x x/L_x + \phi _k)\cos (2\pi k_y y/L_y + \psi _k)$
. Here,
$L_x=0.5$
and
$L_y=0.5$
are the horizontal lengths. The perturbation wavenumbers
$k_x$
and
$k_y$
in the
$x$
and
$y$
directions are chosen within the range of
$30\leqslant k_x$
(or
$k_y$
)
$\leqslant 60$
, and
$\phi _k$
and
$\psi _k$
are random phases (Boffetta et al. Reference Boffetta, Mazzino, Musacchio and Vozella2009; Qiu, Liu & Zhou Reference Qiu, Liu and Zhou2014). According to a previous study (Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a
), the amplitude is chosen as
$A_m=10^{-4}$
.
In the present study, we alter the significance of rotation by varying
$Ro$
. We have chosen four values of
$Ro=\infty$
,
$1.0$
,
$0.5$
and
$0.1$
, ranging from non-rotation to strong rotation, to systematically investigate the effects of rotation on RT turbulence. The other dimensionless parameters are fixed as
$A_t=0.6,\ Fr=1,\ Re=10^5,\ Pr=0.72$
to ensure similar fluid properties in simulations.
The calculations are performed with periodic boundary conditions in the
$x$
and
$y$
directions. Aligning with commonly used boundary settings in RT turbulence simulations (Boffetta et al. Reference Boffetta, Mazzino and Musacchio2016; Gauthier Reference Gauthier2017; Hu, Zhang & Tian Reference Hu, Zhang and Tian2020; Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a
; Fu et al. Reference Fu, Zhao, Wang, Liu and Lu2025), the boundary conditions at the top and bottom of the domain are set to prevent mass and heat flux at these boundaries. Specifically, there is no flow and therefore no mass flux at the top and bottom of the domain, as the vertical velocity and density at the
$z$
-direction boundaries are fixed at their initial values. The vertical derivatives of temperature and horizontal velocity at the
$z$
-direction boundaries are set to zero and there is no heat flux at the top and bottom. Moreover, a sufficiently large computational domain of
$D=\{ (x,y,z) | 0\leqslant x\leqslant 0.5, 0\leqslant y \leqslant 0.5, -0.5 \leqslant z \leqslant 0.5\}$
and a uniform grid of
$N_x \times N_y \times N_z = 650\times 650\times 1300$
are used to obtain reliable results. These simulation parameters are summarised in table 1.
Simulation parameters for DNS at the end of simulations.

It is important to note that periodic boundaries in the horizontal directions are valid because the present study neglects centrifugal acceleration and focuses solely on the Coriolis force. Specifically, the Coriolis force is independent of the location of the rotation axis, and its effect on flow is homogeneous in the horizontal directions (Boffetta et al. Reference Boffetta, Mazzino and Musacchio2016; Buzzicotti et al. Reference Buzzicotti, Aluie, Biferale and Linkmann2018a , Reference Buzzicotti, Clark Di Leoni and Biferaleb ; Wei et al. Reference Wei, Li, Wang, Yang, Zhu, Qian and Luo2022). Consequently, in line with previous studies on rotating RT turbulence (Boffetta et al. Reference Boffetta, Mazzino and Musacchio2016; Wei et al. Reference Wei, Li, Wang, Yang, Zhu, Qian and Luo2022), periodic boundaries are set in the horizontal directions. Additionally, according to the experiments on rotating RTI conducted by Baldwin et al. (Reference Baldwin, Scase and Hill2015) and Scase et al. (Reference Scase, Baldwin and Hill2017), where the rotation axis is within the flow domain and centrifugal acceleration is accounted for, the flow exhibits radial inhomogeneity within the horizontal plane. Therefore, in a more realistic experimental scenario, where the rotation axis is within the flow domain and centrifugal acceleration is included, the results of RT turbulence presented in this work may be influenced by the radial inhomogeneity induced by centrifugal acceleration.
To ensure the reliability of statistics by means of DNS, the smallest-scale motion must be captured. Hence, the grid width should be smaller than the Kolmogorov length scale (Yeung & Pope Reference Yeung and Pope1989). We examined the Kolmogorov length scale
$\eta$
in the mixing zone, which is calculated by
\begin{equation} \eta = \frac {1}{\sqrt {\textit{Re}}}\frac {1}{\sqrt {\langle \rho \rangle }}\left (\frac {\langle \mu \rangle ^3}{\langle \varepsilon \rangle }\right )^{1/4}, \end{equation}
where
$\langle \boldsymbol{\cdot }\rangle$
denotes spatial averaging in the mixing region and
$\varepsilon$
represents the viscous dissipation. Here, the width of the mixing region is determined by
$h=h_s + h_b$
, where
$h_s$
and
$h_b$
denote the widths of spike and bubble mixing layers, respectively. Following Gauthier (Reference Gauthier2017),
$h_s$
and
$h_b$
are calculated as
\begin{equation} {h_s}=\int _{-\frac {L_z}{2}}^{0}f([T])\mathrm{d}z, \,\quad {h_b}=\int _{0}^{\frac {L_z}{2}}f([T])\mathrm{d}z, \end{equation}
where
$[ \boldsymbol{\cdot }]$
represents spatial averaging in the horizontal plane and the mixing function
$f(T)=4(T_{\textit{low}}-T)(T-T_{\textit{up}})/(T_{\textit{low}}-T_{\textit{up}})^2$
. As can be seen from this expression, the mixing function
$f(T)$
ranges from
$0$
to
$1$
, with
$0$
and
$1$
indicating complete segregation and complete mixing, respectively. As shown in table 1, it is confirmed that
$\eta /\varDelta \geqslant 1.0$
for all simulations, where
$\varDelta$
represents the grid width in the computational domain. This ensures that the grid resolution is adequate (Yeung & Pope Reference Yeung and Pope1989; Zhao et al. Reference Zhao, Betti and Aluie2022). Moreover, in contrast to the non-rotation case, the Kolmogorov length scale becomes larger with decreasing
$Ro$
.
2.3. Fundamental features in mixing layer
In this subsection, we examine the fundamental features of rotating RT turbulence, including the flow Reynolds number and key statistical measures such as mixing width, mixed mass and the helicity of the mixing layer.
The criteria for transitioning to turbulent mixing have consistently been a significant point of concern for RT flows, involving both spatial and temporal scales (Zhou et al. Reference Zhou2003; Zhou Reference Zhou2017a
). According to Dimotakis (Reference Dimotakis2000), fully developed turbulent flow requires an outer-scale Reynolds number of
${\textit{Re}}_H \gtrsim 10^4$
, or a Taylor Reynolds number of
${\textit{Re}}_\lambda \gtrsim 100$
. To this end, we selected the temporal evolution of outer-scale Reynolds number and Taylor microscale Reynolds number to evaluate whether the flow has developed a turbulent field (Cook & Zhou Reference Cook and Zhou2002b
; Yan et al. Reference Yan, Fu, Wang, Yu and Li2022). The outer-scale Reynolds number
${\textit{Re}}_H$
is defined as
as listed in table 1. Here,
$h_{\textit{vis}}$
denotes the visual width of the mixing region, defined as the width of area corresponding to
$f([T]) \geqslant 0.01$
. As the RTI evolves, the values of
${\textit{Re}}_H$
eventually exceed
$10^4$
in all cases, thereby the simulated flows are indeed turbulent (Dimotakis Reference Dimotakis2000). Compared with the non-rotating case, stronger rotation reduces the value of
${\textit{Re}}_H$
in the final RT turbulence. This clearly demonstrates the inhibitory effect of rotation on RT mixing-layer growth.
Additionally, we have provided the final physical simulation parameters
${\textit{Re}}_\lambda$
in table 1. The Taylor microscale Reynolds number
${\textit{Re}}_\lambda$
is defined as
Here no summation is implied over
$i$
, where
$i = x,y,z$
. The Taylor microscale
$\lambda _i$
is defined as
\begin{equation} \lambda _i = \sqrt {\frac {\langle u_iu_i \rangle }{\langle (\partial {u_x}/\partial {x})^2+(\partial {u_y}/\partial {y})^2+(\partial {u_z}/\partial {z})^2\rangle }}. \end{equation}
It is convenient to combine
$x$
and
$y$
components with horizontal isotropy, namely
${\textit{Re}}_{\lambda ,xy} = ({\textit{Re}}_{\lambda ,x} + {\textit{Re}}_{\lambda ,y})/2$
. As shown in table 1, increasing rotational intensity significantly decreases
${\textit{Re}}_{\lambda ,z}$
, indirectly indicating that rotation plays a crucial role in suppressing the growth of the mixing layer in the vertical direction. Furthermore, the ratio
${\textit{Re}}_{\lambda ,z}/{\textit{Re}}_{\lambda ,xy}$
decreases notably with increasing rotation. Additionally, helicity is another key statistic in rotating turbulence. As shown in table 1, the mean helicity
$\langle H \rangle$
in the mixing layer is close to zero in non-rotating RT turbulence. As
$Ro$
decreases,
$\langle H \rangle$
increases, highlighting the importance of studying helicity statistics in rotating RT turbulence (Yan et al. Reference Yan, Li, Yu, Wang and Chen2020b
).
It is noted that buoyancy and rotation are two competing effects in rotating RT turbulence. These two effects can indeed be characterised by the buoyancy frequency
$N$
and the angular frequency of rotation
$\varOmega$
. Here, the buoyancy frequency of the mixing layer is given by
$N=\sqrt {-g/\langle \rho \rangle \langle \partial \rho /\partial z\rangle }$
and is an angular frequency (Lawrie & Dalziel Reference Lawrie and Dalziel2011; Dalziel & Mouet Reference Dalziel and Mouet2021). The angular frequency of rotation in the present study is given by
$\varOmega =1/Ro$
. Moreover, the ratio
$N/\varOmega$
reflects the significance of buoyancy relative to rotation. Clearly, the regimes of RT turbulence can be classified into one where buoyancy is significant and another where rotation dominates, based on
$N/\varOmega$
. Figure 1 clearly shows that, for rotating RT turbulence, the ratio
$N/\varOmega$
decreases over time due to the growth of the mixing layer and the reduction of its density gradient. Furthermore, at the time when
${\textit{Re}}_H=12\,000$
, the ratio
$N/\varOmega$
decreases as
$Ro$
decreases, ultimately reaching
$O(10^{-1})$
at
$Ro=0.1$
. This indicates that the RT turbulence regimes for the present settings of
$Ro$
vary from one where buoyancy is significant (
$Ro=\infty$
) to one where rotation dominates (
$Ro=0.1$
).
Temporal evolution of
$N/\varOmega$
for cases
$Ro=1.0$
,
$Ro=0.5$
and
$Ro=0.1$
. Here, open symbols represent the ratio at the time when
${\textit{Re}}_H=12\,000$
.

Moreover, to highlight the associated rotation-induced structural changes, figure 2 shows snapshots of the instantaneous density field for three representative cases at instants with mixing width
$h\approx 0.3L_z$
. With the introduction of rotation, the flow structures become more coherent and elongated in the direction of gravitational acceleration. Vertical filamentary structures can be observed for the strong rotation case with
$Ro=0.1$
. These observations are consistent with the results of Baldwin et al. (Reference Baldwin, Scase and Hill2015) and Boffetta et al. (Reference Boffetta, Mazzino and Musacchio2016).
Snapshots of the instantaneous density field for cases
$(a)$
$Ro=\infty$
,
$(b)$
$Ro=1.0$
and
$(c)$
$Ro=0.1$
at instants with mixing width
$h\approx 0.3L_z$
. Here red (blue) represents heavy (light) fluid.

To determine the wavenumber of vertical filamentary structures observed in density fields, the spectra of the density gradient components
$\rho _{i}\,(i=x,y,z)$
at the
$z=0$
plane are illustrated in figure 3. Here, the spectra are normalised by the mean squared magnitude of the density gradient. The spectra of
$\rho _x$
and
$\rho _y$
with strong rotation show a significant peak at approximately
$k(L_x/2\pi )\approx 10$
, indicating the wavenumber of the dominant vertical filamentary structure in RT turbulence with
$Ro=0.1$
. Furthermore, the spectra in the vertical direction at low and intermediate wavenumbers are stronger than those in the horizontal directions for all cases. This anisotropy in the density gradient corresponds to the non-uniform mean density field within the mixing layer along the vertical direction. The difference in the density gradient between the vertical (
$\rho _z$
) and horizontal (
$\rho _x$
or
$\rho _y$
) directions at
$Ro=0.1$
is smaller than at
$Ro=\infty$
and
$1.0$
at intermediate wavenumbers, corresponding to the vertical filamentary structures. This phenomenon indicates that strong rotation reduces the anisotropy of the density gradient at intermediate scales. This reduction is associated with the vertical filamentary structures observed in RT turbulence under strong rotation, which amplify the horizontal components of the density gradient at intermediate wavenumbers.
Spectra of density gradients at the
$z=0$
plane for cases with
$(a)\ Ro=\infty$
,
$(b)\ Ro=1.0$
and
$(c)\ Ro=0.1$
. Here the horizontal wavenumber
$k=\sqrt {k_x^2 + k_y^2}$
and the spectra are calculated at the time when
${\textit{Re}}_H=12\,000$
.

The mixing width is a key measurement that characterises the spatial extent of the bubble and spike regions in RT turbulence. Figure 4(
$a$
) shows the temporal evolution of the positions of frontiers of the mixing layer
$h_{\textit{low}}$
and
$h_{\textit{up}}$
for different
$Ro$
cases. Here,
$h_{\textit{low}}$
and
$h_{\textit{up}}$
can be determined based on the widths of spike mixing layer
$h_s$
and of bubble mixing layer
$h_b$
, i.e.
$h_{\textit{low}}=z_I-h_s$
and
$h_{\textit{up}}=z_I+h_b$
. Similar to non-rotating RT turbulence, the growth of mixing layers in rotating cases eventually follows a quadratic law
$h_i=\alpha _i A_t g t^2$
with
$i=b,s$
(Youngs Reference Youngs1984). However, stronger rotation delays the onset of quadratic growth and suppresses the overall growth of the mixing layer, as indicated by the decrease in
$\alpha _s$
and
$\alpha _b$
with decreasing
$Ro$
. Figure 4(
$b$
) reveals an increasing asymmetry in the mixing layer as
$Ro$
decreases, because stronger rotation leads to more pronounced inhibition effects of bubble growth.
Temporal evolution of
$(a)$
the positions of frontiers of the mixing layer,
$(b)$
the ratio of
$h_s / h_b$
,
$(c)$
the mixed mass
$M$
and
$(d)$
the normalised mixed mass
$\varPsi$
for cases
$Ro=\infty$
(red squares),
$Ro=1.0$
(green stars),
$Ro=0.5$
(yellow circles) and
$Ro=0.1$
(blue triangles).

In order to quantify the mixing process, we evaluate the temporal evolution of mixed mass
$M$
and the normalised mixed mass
$\varPsi$
following Zhou et al. (Reference Zhou, Cabot and Thornber2016b
). Specifically, the mixed mass is defined as
where
$Y_h$
and
$Y_l$
are mass fractions,
$\rho$
is the mixing density and
$V$
is the volume of the mixing layer. Indeed, the mass fractions are the same as the mole fractions for a single-species fluid (Linden, Redondo & Youngs Reference Linden, Redondo and Youngs1994). Hence, in the present study,
$Y_h$
and
$Y_l$
can be calculated by
$Y_h=(T_{\textit{low}}-T)/(T_{\textit{low}}-T_{\textit{up}})$
and
$Y_l=(T-T_{\textit{up}})/(T_{\textit{low}}-T_{\textit{up}})$
. The normalised mixed mass
$\varPsi$
is defined as
Obviously, the normalised mixed mass
$\varPsi$
ranging from
$0$
to
$1$
was introduced by Zhou et al. (Reference Zhou, Cabot and Thornber2016b
) to measure the efficiency of fluid mixing within the mixing layer. It is shown by Zhou et al. (Reference Zhou, Cabot and Thornber2016b
) that
$\varPsi$
is effective in providing information on the efficiency of mixed mass generation for a given mixed width. In other words, for a mixing layer with a given mixed width, a larger
$\varPsi$
is related to greater mixed mass.
Similar to the mixing width growth, a quadratic growth is also revealed for the mixed mass (see figure 4
$c$
). This finding of quadratic growth in mixed mass aligns with previous results for non-rotating RT turbulence (Zhou et al. Reference Zhou, Cabot and Thornber2016a
). A non-monotonic trend of
$M$
is evident in figure 4
$(c)$
. The mixed mass
$M$
at
$Ro=1.0$
exceeds that of the non-rotating case (
$Ro=\infty$
). However, as
$Ro$
decreases from
$1.0$
to
$0.1$
,
$M$
also decreases. This non-monotonic trend of
$M$
can be attributed to the width of the mixing layer and the efficiency of fluid mixing. Specifically, as shown in figure 4
$(a)$
, the width of the mixing layer is monotonically suppressed by rotation as
$Ro$
decreases, resulting in a reduction in mixed mass at a given efficiency of fluid mixing. However, as shown in figure 4
$(d)$
, the normalised mixed mass
$\varPsi$
increases monotonically as
$Ro$
decreases, which is consistent with the findings of Wei et al. (Reference Wei, Li, Wang, Yang, Zhu, Qian and Luo2022). This indicates that the efficiency of fluid mixing improves as
$Ro$
decreases, resulting in more effective mixing and a greater mixed mass for a given mixed width. Consequently, as
$Ro$
decreases, the mixed mass within the mixing layer exhibits a non-monotonic trend.
3. Generation of kinetic energy and enstrophy
3.1. Generation of kinetic energy
In this subsection, the generation of kinetic energy is examined using three representative cases. Typically, the global kinetic energy
$E_k=\int _D (1/2) \rho u_i u_i \,\mathrm{d}V$
measures the intensity of velocity in a flow system. Similarly,
$E_{kz}=\int _D (1/2) \rho u_3 u_3 \,\mathrm{d}V$
represents the global kinetic energy contributed by the vertical components of velocity. Here,
$D$
denotes the computation domain defined in § 2.2. In an RTI-induced flow, gravitational potential energy is mostly converted into kinetic energy associated with the vertical velocity (Cabot & Zhou Reference Cabot and Zhou2013). As shown in figure 5
$(a)$
, the evolution of RTI leads to an increase in both
$E_k$
and
$E_{kz}$
over time for all cases. Notably, the growth rates of
$E_k$
and
$E_{kz}$
diminish with decreasing
$Ro$
, which is consistent with the results of Boffetta et al. (Reference Boffetta, Mazzino and Musacchio2016) and Wei et al. (Reference Wei, Li, Wang, Yang, Zhu, Qian and Luo2022). To quantify the distribution of kinetic energy in various directions, we evaluate the ratio
$E_{kz}/E_k$
. Figure 5
$(b)$
shows that
$E_{kz}/E_k\gtrapprox 50\,\%$
across all cases, corresponding to roughly half of the kinetic energy in the vertical component, which indicates a strong anisotropy of velocity in rotating RT flow. However, the decrease in the ratio
$E_{kz}/E_k$
with deceasing
$Ro$
indicates that the inhibition of vertical growth in the RT mixing layer, caused by increasing rotation, results in the global kinetic energy contributed by the vertical velocity components becoming closer to that contributed by the horizontal components in strongly rotating RT turbulence.
Temporal evolution of
$(a)$
kinetic energy
$E_k$
and vertical kinetic energy
$E_{kz}$
and
$(b)$
the ratio
$E_{kz}/E_k$
for cases
$Ro=\infty$
(red squares),
$Ro=1.0$
(green stars) and
$Ro=0.1$
(blue triangles).

In compressible RT flows, the conversion of potential energy and pressure–dilatation work are two mechanisms that generate kinetic energy (Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a ). Specifically, the kinetic energy budget can be examined by analysing the following kinetic energy transport equation:
Since there is no net flux across the boundaries, all the pure divergence terms on the left-hand side are eliminated by integrating (3.1) over the computational domain (Cabot & Cook Reference Cabot and Cook2006). The obtained equation can be expressed as follows:
Here,
$P={ (1/Fr)}\int _{D}\rho z \,\mathrm{d}V$
represents the potential energy and
$\varepsilon = ({1}/{\textit{Re}})\int _{D} \sigma _{\textit{ij}} ({ {\partial u_i}/{\partial x_j}}) \,\mathrm{d}V$
is the energy dissipation rate. The pressure–dilatation power term is
$W= \int _{D} p\theta \,\mathrm{d}V$
, which plays an important role in the local conversion rate between kinetic energy and internal energy (Wang et al. Reference Wang, Wan, Chen and Chen2018). Specifically, positive
$W$
denotes the conversion of internal energy into kinetic energy. Conversely, negative
$W$
indicates the conversion of kinetic energy back into internal energy.
As depicted in figure 6,
$\mathrm{d}E_k/\mathrm{d}t$
remains positive while
$\mathrm{d}P/\mathrm{d}t$
remains negative throughout the flow evolution in all three cases. That is, the potential energy is continuously converted into kinetic energy, regardless of the intensity of rotation. Furthermore, the rotation-induced suppression effect is again demonstrated by the slower growth of
$\mathrm{d}E_k/\mathrm{d}t$
and
$-\mathrm{d}P/\mathrm{d}t$
with
$Ro=0.1$
compared with those with
$Ro=\infty$
. Moreover, in the non-rotating case
$Ro=\infty$
, the pressure–dilatation power
$W$
remains positive, indicating a continuous conversion of internal energy into kinetic energy. This phenomenon is consistent with the previous results observed for compressible RT turbulence (Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a
). In contrast, for the rotating cases (
$Ro=1$
and
$Ro=0.1$
), both positive and negative values of
$W$
are observed as the RTI evolves, which indicates that kinetic and internal energies are converted into each other. The differing behaviours of
$W$
in RT turbulence with and without rotation are related to flow compressibility (Wang et al. Reference Wang, Wan, Chen and Chen2018). Specifically, according to the results for compressible RT turbulence (Luo & Wang Reference Luo and Wang2021; Fu et al. Reference Fu, Zhao, Wang, Liu and Lu2025), both positive and negative
$W$
values are observed when flow compressibility is weak, indicating that kinetic and internal energies are converted into each other. The value of
$W$
remains positive in RT turbulence with strong flow compressibility, indicating that internal energy is converted into kinetic energy (Luo & Wang Reference Luo and Wang2021; Fu et al. Reference Fu, Zhao, Wang, Liu and Lu2025). In the present simulations, the average turbulent Mach number over the mixing region at the end of the simulation decreases from
$0.12$
for
$Ro=\infty$
to
$0.05$
for
$Ro=0.1$
, indicating that rotation suppresses flow compressibility in RT turbulence. Thus, consistent with the results in compressible RT turbulence, the
$W$
value is both positive and negative in the case of rotation with weak compressibility, while it almost remains positive in the absence of rotation with stronger compressibility.
Temporal evolution of the average terms of the kinetic energy budget equation versus time for cases
$(a)\,Ro=\infty$
,
$(b)\,Ro=1.0$
and
$(c)\,Ro=0.1$
.

To better illustrate the effect of rotation on the efficiency of kinetic energy production, we track the temporal evolution of the efficiency of kinetic energy production for three simulation cases. Specifically, the overall energy balance can be quantified by taking a temporal integration of (3.2) from
$0$
to
$t$
, yielding
Here,
$\Delta E_k$
represents the total kinetic energy,
$\mathcal{E}=\int ^{t}_{0} \varepsilon (t')\text{d}t'$
denotes the total dissipated energy, the total pressure–dilatation work
$\mathcal{W} = \int ^{t}_{0} W(t')\text{d}t'$
quantifies kinetic energy converted from internal energy and
$- \Delta P$
signifies the total converted potential energy. Consequently, the efficiency of kinetic energy production is defined as
$\varXi = \Delta E_k / (\mathcal{W} - \Delta P)$
. Figure 7 depicts the temporal evolution of the efficiency
$\varXi$
in three cases. For case
$Ro=\infty$
,
$\varXi$
increases gradually from
$20\,\%$
to over
$60\,\%$
, followed by a slight decrease. For case
$Ro=1.0$
,
$\varXi$
exhibits a similar trend but with marginally lower values. Also, nearly half of the energy is converted to kinetic energy, while the remainder is eventually dissipated by viscosity. However, for case
$Ro=0.1$
,
$\varXi$
drops to
$10\,\%$
, highlighting that rotation reduces the efficiency of kinetic energy production and enhances energy dissipation.
Temporal evolution of the efficiency of kinetic energy production
$\varXi$
for for cases
$Ro=\infty$
(red squares),
$Ro=1.0$
(green stars) and
$Ro=0.1$
(blue triangles).

To understand the mechanisms of rotational inhibition on kinetic energy production efficiency, we further scrutinise the energy dissipation. Specifically, the energy dissipation rate
$\varepsilon$
can be classified into components based on viscous stress:
\begin{align} \varepsilon _{\textit{NS}}&=\frac {1}{\textit{Re}}\int _{D} \left(\sigma _{11} {\frac {\partial u_1}{\partial x_1}} + \sigma _{22} {\frac {\partial u_2}{\partial x_2}} + \sigma _{33} {\frac {\partial u_3}{\partial x_3}}\right) \mathrm{d}V,\nonumber \\ \varepsilon _{{\textit{SS}}_{xy}}&=\frac {1}{\textit{Re}}\int _{D} \left(\sigma _{12} {\frac {\partial u_1}{\partial x_2}} + \sigma _{21} {\frac {\partial u_2}{\partial x_1}}\right) \mathrm{d}V, \nonumber\\ \varepsilon _{{\textit{SS}}_{\textit{xz}}}&=\frac {1}{\textit{Re}}\int _{D} \left(\sigma _{13} {\frac {\partial u_1}{\partial x_3}} + \sigma _{31} {\frac {\partial u_3}{\partial x_1}}\right)\mathrm{d}V, \nonumber \\ \varepsilon _{{\textit{SS}}_{\textit{yz}}}&=\frac {1}{\textit{Re}}\int _{D} \left(\sigma _{23} {\frac {\partial u_2}{\partial x_3}} + \sigma _{32} {\frac {\partial u_3}{\partial x_2}}\right) \mathrm{d}V, \end{align}
where
$\varepsilon _{\textit{NS}},\ \varepsilon _{{\textit{SS}}_{xy}},\ \varepsilon _{{\textit{SS}}_{\textit{xz}}}$
and
$\varepsilon _{{\textit{SS}}_{\textit{yz}}}$
correspond to dissipation due to normal viscous stress and shear viscous stress in three different planes, respectively.
The relative importance of each energy dissipation rate component in the total dissipation rate is quantified by
$\phi _{i}=\varepsilon _{i} / \varepsilon$
. As shown in figure 8
$(a)$
, for the case
$Ro=\infty$
,
$\phi _{\textit{NS}}$
is approximately
$40\,\%$
, while
$\phi _{{\textit{SS}}_{xy}},\ \phi _{{\textit{SS}}_{\textit{xz}}}$
and
$\phi _{{\textit{SS}}_{\textit{yz}}}$
are around
$20\,\%$
. However, for the case
$Ro=0.1$
, figure 8
$(b)$
reveals a notable decrease in
$\phi _{\textit{NS}}$
and
$\phi _{{\textit{SS}}_{xy}}$
, accompanied by a significant increase in
$\phi _{{\textit{SS}}_{\textit{xz}}}$
and
$\phi _{{\textit{SS}}_{\textit{yz}}}$
. These results indicate that rotational effects enhance the vertical viscous dissipation components
$\phi _{{\textit{SS}}_{\textit{xz}}}$
and
$\phi _{{\textit{SS}}_{\textit{yz}}}$
within the mixing layer. This enhancement subsequently decelerates the development of vertical flow velocities and diminishes potential energy transfer into the system, thereby effectively restraining the growth of the mixing layer.
Temporal evolution of
$\phi _{i} = \varepsilon _{i}/\varepsilon$
for cases
$(a)$
$Ro=\infty$
and
$(b)$
$Ro=0.1$
, where
$\varepsilon _{i}$
is defined by (3.4).

3.2. Generation of enstrophy
A key mechanism in turbulence is vortex stretching, which facilitates the redistribution of velocity perturbations across different length scales. In 3-D RT turbulence, previous studies have shown that during the initial stages, baroclinicity is the dominant mechanism for generating enstrophy, while at later times, vortex stretching becomes the dominant mechanism (Cabot Reference Cabot2006). However, the evolution of average enstrophy
$\varOmega =\langle 1/2 \omega _i \omega _i \rangle$
within the mixing layer remains not fully understood in rotating RT turbulence. This measure is significant for characterising the flow and is useful for understanding the energy cascade (Suman et al. Reference Suman, Sundaram, Puttam, Sengupta and Sengupta2022). Moreover, to evaluate the anisotropy of vorticity in rotating RT turbulence, the average vertical enstrophy, given by
$\varOmega _z =\langle 1/2 \omega _3 \omega _3 \rangle$
, is also examined. Therefore, the ratio
$\varOmega _z/\varOmega$
is the anisotropy of vorticity, which can be used to describe small-scale anisotropy (Cabot & Zhou Reference Cabot and Zhou2013). Figure 9
$(a)$
illustrates the temporal evolution of both
$\varOmega$
and
$\varOmega _z$
in all cases. As expected, enstrophy and vertical enstrophy grow throughout rotating RTI evolution. In particular, the growth rate of enstrophy diminishes with decreasing
$Ro$
. Interestingly,
$\varOmega _z$
in case
$Ro=1.0$
attains the fastest growth at the early stage. However, at late stage
$\varOmega _z$
in case
$Ro=\infty$
still has the highest value. Despite the differences in enstrophy, the small-scale anisotropy is similar under different rotations. As shown in figure 9
$(b)$
, for all cases
$\varOmega _z/\varOmega$
tends to be
$1/3$
. This indicates that the flow’s small-scale motions are approaching isotropy (Cabot & Zhou Reference Cabot and Zhou2013). The aforementioned intriguing phenomena prompt an investigation into the underlying mechanisms responsible for the generations of enstrophy and of vertical enstrophy in rotating RT turbulence.
Temporal evolution of
$(a)$
average enstrophy
$\varOmega$
and its vertical component
$\varOmega _z$
within the mixing layer and
$(b)$
the ratio
$\varOmega _z/\varOmega$
for cases
$Ro=\infty$
,
$Ro=1.0$
and
$Ro=0.1$
.

To elucidate the mechanisms driving enstrophy generation in rotating RT turbulence, the enstrophy transport equation is derived from the momentum equation and is expressed as follows:
In the equation above, the various contributions to the enstrophy rate on the right-hand side are defined as follows:
\begin{align} P_\omega &= \omega _i S_{\textit{ij}} \omega _j , \nonumber\\ D_\omega &= -\omega _i \omega _i \frac {\partial u_k}{\partial x_k} , \nonumber\\ B_\omega &= \frac {1}{\rho ^2}\varepsilon _{\textit{ijk}}\omega _i\frac {\partial \rho }{\partial x_j}\frac {\partial p }{\partial x_k}, \nonumber\\ V_\omega &= \frac {1}{\textit{Re}} \varepsilon _{\textit{ijk}}\omega _i\frac {\partial }{\partial x_j}\left ( \frac {1}{\rho }\frac {\partial \sigma _{lk}}{\partial x_l}\right )\! , \nonumber\\ {\textit{Ro}}_\omega &= -\frac {2}{\textit{Ro}}\left ( \omega _3 \frac {\partial u_k}{\partial x_k} - \omega _i \frac {\partial u_i}{\partial x_3}\right )\! , \end{align}
where
$\varepsilon _{\textit{ijk}}$
is the alternating tensor. The physical meanings of each term are as follows:
$P_\omega$
is the vortex stretching and tilting term, which disperses and spreads velocity perturbations across various length scales, and serves as an important measurement for understanding the energy cascade (Sengupta et al. Reference Sengupta, Ulloa and Joshi2023);
$D_\omega$
represents the dilatation term, indicating enstrophy amplification or dissipation due to compression or expansion;
$B_\omega$
is the baroclinic term, reflecting the contribution to enstrophy from the misalignment of gradients of pressure and density;
$V_\omega$
denotes the viscous term; and
${\textit{Ro}}_\omega$
accounts for the Coriolis force term. Evaluating the relative contribution of these terms to the growth rate of enstrophy enables the quantification of key mechanisms during flow evolution.
As shown in figure 10
$(a{-}c)$
, it is identified that the sum of these five terms is consistently positive but decreases with decreasing
$Ro$
. Hence, the enstrophy increases during flow evolution, but stronger rotation can reduce its growth rate. The time series of
$\left \langle \text{d} (1/2\omega _i\omega _i)/\text{d}t \right \rangle$
and
$\left \langle \text{d}(1/2\omega _3\omega _3)/\text{d}t \right \rangle$
are also provided. It is evident that
$\left \langle \text{d} (1/2\omega _i\omega _i)/\text{d}t \right \rangle$
and
$\left \langle \text{d}(1/2\omega _3\omega _3)/\text{d}t \right \rangle$
agree well with terms
$\langle Sum \rangle$
, indicating that the enstrophy transport equations remain strictly complete and that the current grid and temporal resolutions adequately resolve these equations. A careful examination of different terms reveals a competition between the baroclinic mechanism and the vortex stretching and tilting mechanism as the primary contributors to enstrophy generation during the evolution of RTI. Specifically, for
$t\leqslant 1.5$
,
$B_\omega$
is dominant over
$P_\omega$
, and it becomes less significant for
$t \geqslant 2.0$
. Thus, baroclinic effects primarily drive the enstrophy generation in the early stage of flow evolution, while vortex stretching and tilting become more important in the later stage. Additionally,
$P_\omega$
is analytically zero in 2-D flow. Hence, the vortex stretching and tilting term is significantly reduced as
$Ro$
decreases (see figure 10
$c$
), indicating that rotation tends to make the enstrophy generation in 3-D RT turbulence behave more like that in 2-D flow (Cabot Reference Cabot2006). In addition, the Coriolis force term
${\textit{Ro}}_\omega$
does not increase significantly with enhanced rotation. This is because the enhancement of rotation is accompanied by the significant suppression of velocity, which in turn leads to a substantial decline in the velocity gradient. For the parameter range currently considered, the dilatation term is minimal. Hence, compressibility has little impact on enstrophy generation. Eventually, the enstrophy is dissipated by viscosity, as indicated by the negative viscous term.
Temporal evolution of
$(a{-}c)$
the average terms on the right-hand side of the enstrophy transport equation and of
$(d{-}f)$
the average terms on the right-hand side of the vertical enstrophy transport equation for cases
$Ro=\infty$
,
$Ro=1.0$
and
$Ro=0.1$
. Here,
$\langle P_\omega \rangle$
is the vortex stretching and tilting term,
$\langle D_\omega \rangle$
is the dilatation term,
$\langle B_\omega \rangle$
is the baroclinic term,
$\langle V_\omega \rangle$
is the viscous term and
$\langle {\textit{Ro}}_\omega \rangle$
denotes the Coriolis force term and
$\langle Sum \rangle$
represents the sum of these terms. The open circles in
$(a{-}c)$
represent the time series of
$\langle \text{d} (1/2\omega _i\omega _i)/\text{d}t \rangle$
, while those in
$(d{-}f)$
represent the time series of
$\langle \text{d}(1/2\omega _3\omega _3)/\text{d}t \rangle$
.

Next, we focus on the vertical enstrophy generation and further scrutinise the anisotropy of the enstrophy generation. By taking the vertical component of (3.5), the vertical enstrophy transport equation is expressed as follows:
\begin{align} \frac {\mathrm{d}}{\mathrm{d} t}\left (\frac {1}{2} \omega _3 \omega _3\right ) &= \omega _3 S_{3j} \omega _j -\omega _3 \omega _3 \frac {\partial u_k}{\partial x_k} + \frac {1}{\rho ^2}\varepsilon _{3jk}\omega _3\frac {\partial \rho }{\partial x_j}\frac {\partial p }{\partial x_k} \nonumber\\ &\quad +\frac {1}{\textit{Re}} \varepsilon _{3jk}\omega _3\frac {\partial }{\partial x_j}\left ( \frac {1}{\rho }\frac {\partial \sigma _{lk}}{\partial x_l}\right ) - \frac {2}{\textit{Ro}}\omega _3\left ( \frac {\partial u_k}{\partial x_k} - \frac {\partial u_3}{\partial x_3}\right ). \end{align}
Again, the rotation-induced suppression of vertical enstrophy generation is observed. As shown in figure 10
$(d{-}f)$
, different from the generation of enstrophy, the generation of vertical enstrophy shows that
$B_\omega$
is trivially small, while
$P_\omega$
plays a dominant role throughout the entire stage of RTI evolution, particularly for
$Ro=\infty$
and
$1.0$
. Hence, the contribution of the baroclinic term to vertical enstrophy generation is negligible, similar to the behaviour observed in previous studies on non-rotation RT turbulence (Schneider & Gauthier Reference Schneider and Gauthier2016; Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a
). On the other hand, compared with the non-rotation case
$Ro=\infty$
, the Coriolis force term slightly increases, while
$P_\omega$
experiences a slight decrease in the later stages of the weaker rotation case
$Ro=1.0$
. These findings assist in elucidating the observations in figure 9
$(a)$
. In other words,
$\varOmega _z$
experiences the fastest growth during the early stages in case
$Ro=1.0$
due to the contribution of the Coriolis force term, while it reaches the highest value in the later stages of the case
$Ro=\infty$
, driven by the significant contribution of the vortex stretching and tilting term. Furthermore, under strong rotation
$Ro=0.1$
, the Coriolis force term has greatly facilitated the generation of vertical enstrophy. Specifically,
$P_\omega$
approaches nearly zero in the later stages, and
${\textit{Ro}}_\omega$
exceeds
$P_\omega$
at
$t \geqslant 2.5$
(see figure 10
$f$
), indicating that the Coriolis force term becomes the primary source of vertical enstrophy in the strong rotation case.
4. Scale-to-scale transfer of kinetic energy
4.1. Statistics of kinetic energy transfer
To investigate the scale-to-scale transfer of kinetic energy in rotating RT turbulence, we apply a filtering technique to understand the scale interactions. For any field
$f$
, its corresponding filtering field
$\overline {f}$
is computed using a top-hat filter, which is calculated in one dimension by
\begin{equation} \overline {f}_i = \frac {1}{4n}\left (f_{i-n}+2\sum _{j = i-n+1}^{i+n-1} f_j + f_{i+n}\right )\! , \end{equation}
where the filter width is
$r=2n\Delta x$
(Wang et al. Reference Wang, Wan, Chen and Chen2018). Additionally, the Favre-filtered field
$\widetilde {f}$
is defined as
$\widetilde {f}=\overline {\rho f}/\overline {\rho }$
, which is commonly employed for flows described by the compressible Navier–Stokes equations (Aluie Reference Aluie2013; Wang et al. Reference Wang, Wan, Chen and Chen2018; Zhao et al. Reference Zhao, Betti and Aluie2022).
By filtering the kinetic energy equation, we can investigate the scale pathways of kinetic energy in rotating RT flows. The large-scale kinetic energy equation can be straightforwardly derived as
Here,
$\boldsymbol{J}_r$
represents the spatial transport of the large-scale kinetic energy and
$\varPhi _r$
is the large-scale pressure–dilatation term;
$\varPi _r$
is the subgrid-scale (SGS) flux across scale
$r$
, and represents the deformation work done by large-scale strain
$\partial \widetilde {u}_i / \partial x_j$
against the SGS stress
$\tau _{\textit{ij}}=\overline {\rho }(\widetilde {u_i u_j}-\widetilde {u}_i \widetilde {u} _j )$
;
$\varLambda _r$
is the baropycnal work done by the large-scale pressure gradient
$({1}/{\overline {\rho }})({\partial \overline {p} }/{\partial x_j})$
against the SGS mass flux
$\overline {\rho u_j} - \overline {\rho }$
$\overline {u_j}$
(Aluie Reference Aluie2011, Reference Aluie2013);
$\varepsilon _r$
is the viscous dissipation term; and
$I_r$
is the energy injected by gravity. The expressions for these terms are as follows:
\begin{align} J_{r_j} &= \overline {\rho } \frac {| \widetilde {\boldsymbol u}_r |^2 }{2} \widetilde {u}_j + \overline {pu_j} + \widetilde {u}_i\overline {\rho }(\widetilde {u_i u_j}-\widetilde {u}_i \widetilde {u} _j) - \frac {1}{\textit{Re}} \widetilde {u}_i \overline {\sigma }_{\textit{ij}} , \nonumber\\ \varPhi _r &= \overline {p} \frac {\partial \overline {u}_i }{\partial x_i} , \nonumber\\ \varPi _r &= -\overline {\rho }(\widetilde {u_i u_j}-\widetilde {u}_i \widetilde {u} _j ) \frac {\partial \widetilde {u}_i }{\partial x_j}, \nonumber\\ \varLambda _r &= \frac {1}{\overline {\rho }}\frac {\partial \overline {p} }{\partial x_j}(\overline {\rho u_j} - \overline {\rho } \, \overline {u_j}) , \nonumber\\ \varepsilon _r &= \frac {1}{\textit{Re}} \overline {\sigma }_{\textit{ij}} \frac {\partial \widetilde {u}_i }{\partial x_j} , \nonumber\\ I_r &= -\frac {1}{\textit{Fr}} \overline {\rho } \widetilde {u}_3 . \end{align}
To reveal the effects of rotation on energy conversion and cascade in rotating RT turbulence, we compare the average terms of the large-scale kinetic energy equation versus the normalised filter scale
$r/\varDelta$
at the time when
${\textit{Re}}_H$
reaches the same value. Specifically,
${\textit{Re}}_H=12\,000$
is chosen here to ensure that the flows have reached a turbulent state (Cook & Zhou Reference Cook and Zhou2002b
). At the time when
${\textit{Re}}_H=12\,000$
, the RT turbulence regimes for cases with
$Ro=\infty$
,
$Ro=1.0$
and
$Ro=0.1$
vary from one in which buoyancy is significant (
$Ro=\infty$
) to one where rotation dominates (
$Ro=0.1$
). Thus, these three cases are selected to investigate energy transfer.
As depicted in figure 11
$(a{-}c)$
, at all filtering scales, the kinetic energy is injected primarily by gravitational work
$I_r$
and pressure–dilatation work
$\varPhi _r$
, with gravitational input being the predominant factor. As
$Ro$
decreases, the contribution of gravitational work diminishes significantly, and the pressure–dilatation work also exhibits a downward trend. This also supports the fact that rotation hinders the conversion of potential and internal energy into kinetic energy. The baropycnal work is greater on large scales than on small scales, indicating its significant role in scale-to-scale kinetic energy transfer within the large-scale regime. Moreover, the baropycnal work decreases as
$Ro$
decreases, demonstrating a rotation-induced suppression effect on the baropycnal work. Strikingly, the overall SGS flux
$\varPi _r$
is always positive, indicating that kinetic energy is transferred from large to small scales, following the traditional forward cascade in 3-D turbulence (Zhou et al. Reference Zhou, Huang, Lu, Liu and Ni2016a
). The peak of the SGS flux emerges at a filter scale of
$r/\varDelta \approx 20$
. As
$Ro$
decreases, the SGS flux also decreases, approaching zero at case
$Ro=0.1$
. Hence, rotation hinders the kinetic energy transfer from large to small scales. Moreover, as expected, the dissipation of kinetic energy, influenced by viscous effects, intensifies at smaller scales.
$(a{-}c)$
The average terms of the large-scale kinetic energy equation,
$(d{-}f)$
the average terms of the vertical large-scale kinetic energy equation and
$(g{-}i)$
the average terms of the horizontal large-scale kinetic energy equation versus the normalised filter scale
$r/\varDelta$
for cases
$Ro=\infty$
,
$Ro=1.0$
and
$Ro=0.1$
at the time when
${\textit{Re}}_H=12\,000$
.

The contributions to vertical large-scale kinetic energy can be derived from (4.2) and are expressed as
\begin{align} \varPhi _z &= \overline {p} \frac {\partial \overline {u}_3 }{\partial x_3} , \nonumber\\ \varPi _z &= -\overline {\rho }(\widetilde {u_3 u_j}-\widetilde {u}_3 \widetilde {u} _j ) \frac {\partial \widetilde {u}_3 }{\partial x_j}, \nonumber\\ \varLambda _z &= \frac {1}{\overline {\rho }}\frac {\partial \overline {p} }{\partial x_3}(\overline {\rho u_3} - \overline {\rho } \, \overline {u_3}) , \nonumber\\ \varepsilon _z &= \frac {1}{\textit{Re}} \overline {\sigma _{3j}} \frac {\partial \widetilde {u}_3 }{\partial x_j} , \nonumber\\ I_z &= -\frac {1}{\textit{Fr}} \overline {\rho } \widetilde {u}_3 . \end{align}
Moreover, the budget terms of horizontal kinetic energy equation can be obtained by subtracting (4.4) from (4.3). Figures 11
$(d{-}f)$
and 11
$(g{-}i)$
show the budget terms of the large-scale vertical and horizontal kinetic energy equations, respectively. Due to the gravitational acceleration in the
$z$
direction, the gravitational input work is present only in the vertical direction. The baropycnal work
$\varLambda _r$
resides mostly in the vertical component
$\varLambda _z$
, and its horizontal component
$\varLambda _{xy}$
is negligible. It is noteworthy that the components of the pressure–dilatation term are no longer constant across scales, with the vertical component being larger at large scales and the horizontal component being predominant at small scales. Therefore, the pressure–dilatation work acts as source of vertical large-scale kinetic energy and of horizontal small-scale kinetic energy. Furthermore, the vertical component
$\varPi _z$
of SGS flux is larger than its horizontal component
$\varPi _{xy}$
, with peaks of
$\varPi _z$
occurring at larger scales compared with
$\varPi _{xy}$
. These observations provide concrete evidence that, in the scale-to-scale energy transfer of RT turbulence, the scale at which the horizontal component of SGS energy flux is most intense is smaller than that at which the vertical component is most intense. Moreover, the viscous dissipation term components
$\varepsilon _z$
and
$\varepsilon _{xy}$
contribute mainly at small scales. The ratio
$\varepsilon _z/\varepsilon _{xy}$
also increases as
$Ro$
decreases, further confirming the previous findings regarding the viscous dissipation in § 3.1.
To gain a further understanding of the transfer of energy across different scales, the spectra for velocity and vorticity are presented. Specifically, the spectra of the velocity components are presented in figure 12. Here, the spectra are normalised by the mean squared magnitude of the velocity. It is evident that the RT flows cover a wide range of length scales once the flow reaches a turbulent state. The obtained velocity spectra exhibit a power law in the inertial range, and the scaling exponent can be determined by linearly fitting the spectra within the range
$10 \leqslant k(L_x/2\pi ) \leqslant 30$
. This range corresponds to a significant positive energy flux
$\langle \varPi _r\rangle$
. The fitted exponents are approximately
$-1.74$
for the
$Ro = \infty$
and
$Ro = 1.0$
cases and
$-1.78$
for the
$Ro = 0.1$
case. These values are closer to the
$-7/4$
power law predicted by Zhou (Reference Zhou2001) for RT turbulence than to the classical Kolmogorov law of
$-5/3$
. The power law of
$-7/4$
is reasonably related to the external time scale introduced by the buoyancy (Cook & Zhou Reference Cook and Zhou2002a
). Nevertheless, the present spectra are insufficient to demonstrate that the
$-7/4$
scaling is more suitable for RT turbulence than the
$-5/3$
scaling, due to the narrow inertial range in the present simulations. To accurately determine which scaling exponent is better for the velocity spectra in the inertial range of RT turbulence, future studies should focus on simulating high-Reynolds-number conditions. Moreover, the presence of an inertial range in rotating RT turbulence indicates a traditional forward cascade, corresponding to the positive overall SGS flux. As
$Ro$
decreases, the length of the inertial range shortens in rotating RT turbulence, which is associated with the decreasing SGS flux shown in figure 11.
Spectra of velocity components at the
$z=0$
plane for cases
$(a)\ Ro=\infty$
,
$(b)\ Ro=1.0$
and
$(c)\ Ro=0.1$
. Here the horizontal wavenumber
$k=\sqrt {k_x^2 + k_y^2}$
.

Interestingly, the shorter inertial range under strong rotation results in a shift of the maximum energy wavenumber of velocity spectra towards larger values. Specifically, as shown in figure 12, the maximum energy wavenumber is approximately
$1$
for cases with
$Ro=\infty$
and
$Ro=1.0$
, which is similar to the velocity spectra of RT turbulence simulated by Cabot & Cook (Reference Cabot and Cook2006). In fact, for RT flow with no rotation, instability occurs across all wavenumbers. Thus, as RT turbulence with no or weak rotation evolves, the maximum energy wavenumber of the velocity spectra occurs at low wavenumbers due to the increasing dominance of larger-scale structures formed by the merging of bubbles and spikes (Casner et al. Reference Casner2016; Gauthier Reference Gauthier2017; Zhao et al. Reference Zhao, N.-S. and X.-Y.2020a
). However, in the case of strong rotation, the maximum energy wavenumber is approximately
$10^1$
, closely corresponding to the wavenumber of vertical filamentary structures identified by the density gradient spectra in figure 3. This indicates that the occurrence of the maximum energy wavenumber in velocity spectra with strong rotation at intermediate wavenumbers is related to the vertical filamentary structures.
Additionally, the spectra of the vorticity components are shown in figure 13 and are normalised by the mean squared magnitude of the vorticity. Different from the behaviour of the velocity spectra, the spectra of
$\omega _{x}$
and
$\omega _{y}$
remain nearly constant in the inertial range. The constant range of the spectra of
$\omega _{x}$
and
$\omega _{y}$
shortens as
$Ro$
decreases, corresponding to the decreasing SGS flux. Furthermore, the spectra of the shear rates, namely
$S_{xy}$
,
$S_{\textit{yz}}$
and
$S_{\textit{xz}}$
, are illustrated in figure 14. Here, the spectra are normalised by the mean squared magnitude of strain-rate tensor. It is evident the spectra of shear rates under strong rotation are significantly amplified at intermediate wavenumbers. For the RT turbulence with strong rotation (
$Ro=0.1$
), both the vorticity and the shear rate spectra exhibit significant peaks at approximately the wavenumber of the vertical filamentary structure identified by the density gradient spectra in figure 3. This observation supports the notion that the maximum energy wavenumbers of the vorticity and shear rate spectra with strong rotation, occurring at intermediate wavenumbers, are related to the vertical filamentary structures.
Spectra of vorticity components at the
$z=0$
plane for cases
$(a)\ Ro=\infty$
,
$(b)\ Ro=1.0$
and
$(c)\ Ro=0.1$
. Here the horizontal wavenumber
$k=\sqrt {k_x^2 + k_y^2}$
.

Spectra of shear rates at the
$z=0$
plane for cases
$(a)\ Ro=\infty$
,
$(b)\ Ro=1.0$
and
$(c)\ Ro=0.1$
. Here the horizontal wavenumber
$k=\sqrt {k_x^2 + k_y^2}$
.

4.2. Effects of helicity on SGS flux
As introduced in § 1, previous studies have shown that helicity plays a significant role in energy transfer of turbulence, impeding the forward energy cascade and even leading to the reversal of energy transfer (André & Lesieur Reference André and Lesieur1977; Biferale et al. Reference Biferale, Musacchio and Toschi2012; Moffatt Reference Moffatt2014; Briard & Gomez Reference Briard and Gomez2017; Sahoo et al. Reference Sahoo, Alexakis and Biferale2017; Buzzicotti et al. Reference Buzzicotti, Aluie, Biferale and Linkmann2018a ). Motivated by this finding, the inhibitory effect of helicity on energy flux is examined in rotating RT turbulence. Building on the findings regarding the effect of helicity, this subsection further explores the quantitative relationship between SGS flux and the level of helicity while examining whether this relationship remains consistent across different scales. Moreover, the effect of helicity on the Smagorinsky model is examined to evaluate the model’s applicability in rotating RT turbulence.
As shown in figure 15
$(a)$
, we evaluate the probability density function (PDF) of the normalised helicity
$H_r/H_r^{\prime }$
at filter scale
$r/\varDelta =20$
for three cases at the time when
${\textit{Re}}_H=12\,000$
. Here, the root-mean-square value of helicity is defined as
$H_r^{\prime }=\sqrt {\left \langle (H_r - \langle H_r\rangle ) ^{2}\right \rangle }$
. The filter scale
$r/\varDelta =20$
is selected because it corresponds to the peak of SGS flux emergence. Interestingly, the PDF of normalised helicity is essentially symmetric with respect to the line
$ H_r /H_r^{\prime }=0$
in the case
$Ro=\infty$
. As rotation enhances, this symmetry is broken, leading to a mean helicity that is no longer zero. Moreover, to further investigate the cause of the inhibitory effect of rotation on energy transfer, we depict the PDF of the normalised SGS flux
$\varPi _{r}/\varPi _{r}^{\prime }$
at filter scale
$r/\varDelta =20$
for three cases with
${\textit{Re}}_H=12\,000$
, as shown in figure 15
$(b)$
. Similarly,
$\varPi _{r}^{\prime }=\sqrt {\left \langle (\varPi _{r} - \langle \varPi _{r}\rangle ) ^{2}\right \rangle }$
. We observe that the negative value of SGS flux
$\varPi _r$
occurs in certain flow regions, as indicated by the negative side of the PDF. The negative value of SGS flux can be understood as the SGS backscatter of kinetic energy, i.e. the reverse flux of kinetic energy from small scales to large scales. In the non-rotating case (
$Ro=\infty$
), the PDF of the SGS flux exhibits a skewness towards the positive side, demonstrating the dominance of the regions where the kinetic energy is transferred from large to small scales. As
$Ro$
decreases, the PDF of the SGS flux tends to be shifted with a skewness towards the negative side, indicating a tendency for kinetic energy transfer from small scales to large scales.
(a–f) The average of normalised SGS flux
$\varPi _{r}/\varPi _{r}^{\prime }$
conditioned on the normalised helicity
$ H_{r}/H_{r}^{\prime }$
at filter scales
$r/\varDelta =10$
,
$20$
and
$40$
for cases with
$Ro=\infty$
,
$1.0$
and
$0.1$
at the time when
${\textit{Re}}_H=12\,000$
.

It is important to note that although the local inverse energy transfer can be observed via PDFs of SGS flux in figure 15, the overall SGS flux remains positive in all simulations, indicating that the energy cascade is always forward in rotating RT turbulence. Indeed, unlike rotating turbulence with an inverse cascade of energy (Alexakis & Biferale Reference Alexakis and Biferale2018; Buzzicotti et al. Reference Buzzicotti, Aluie, Biferale and Linkmann2018a , Reference Buzzicotti, Clark Di Leoni and Biferaleb ), the growth of large scales in RT turbulence is driven by buoyancy. Therefore, as observed in figure 11, the overall SGS flux in the mixing region of RT turbulence is positive, indicating that the energy cascade is always forward. However, for rotating turbulence with an inverse cascade of energy (Alexakis & Biferale Reference Alexakis and Biferale2018; Buzzicotti et al. Reference Buzzicotti, Aluie, Biferale and Linkmann2018a , Reference Buzzicotti, Clark Di Leoni and Biferaleb ), the external force driving the flow is applied at a specific scale. Consequently, the large scales are formed by the effect of rotation, leading to an inverse cascade at scales larger than the forced scale.
To quantitatively assess the effect of helicity on the SGS flux, we present the joint PDF of the normalised SGS flux
$\varPi _{r}/\varPi _{r}^{\prime }$
and the normalised helicity
$ H_{r}/H_{r}^{\prime }$
at a filter scale
$r/\varDelta =20$
for three cases at the time when
${\textit{Re}}_H=12\,000$
. In the non-rotating case (figure 16
$a$
), the shape of the joint PDF is nearly symmetric with respect to the line
$H_{r} /H_{r}^{\prime }=0$
. Hence, the contributions from regions with positive and negative helicity to the SGS flux are approximately equal. The fractions of the first and second quadrants of the joint PDF are larger, indicating the dominance of forward SGS flux for case
$Ro=\infty$
. As
$Ro$
decreases (figure 16
$b{,}c$
), the symmetry of the joint PDF is broken, resulting in a larger fraction on the right-hand side of the line
$H_{r} /H_{r}^{\prime }=0$
. This indicates that the positive helicity regions in the mixing layer are predominant in rotating cases. Furthermore, for the strong rotation case, the fractions in the third and fourth quadrants increase significantly, indicating a higher probability of negative SGS flux events (see figure 16
$c$
). In particular, the fraction of negative SGS fluxes is larger than the fraction of positive SGS fluxes in regions where
$\left \lvert H_{r}/H_{r}^{\prime } \right \rvert \gtrapprox 4$
, suggesting that backscatter dominates in high-helicity regions in rotating RT turbulence. This observation implies that helicity can hinder the forward energy cascade and may even promote a reversal of energy transfer, consistent with previous findings (Moffatt Reference Moffatt2014; Słomka & Dunkel Reference Słomka and Dunkel2017).
(a–i) Joint PDFs of
$\varPi _r/\varPi _r'$
and
$H_r/H_r'$
at filter scales
$r/\varDelta =10$
,
$20$
and
$40$
for cases with
$Ro=\infty$
,
$1.0$
and
$0.1$
at the time when
${\textit{Re}}_H=12\,000$
.

To further investigate the characteristics of the helicity and SGS flux at different scales, analyses are also conducted using a larger filter scale
$r/\varDelta =40$
, which lies within the inertial range shown in figure 12. Additionally, a smaller filter scale of
$r/\varDelta =10$
is also considered. It is clear from figure 15 that at filter scales of
$r/\varDelta =10$
and
$40$
, the symmetry of the PDF of normalised helicity is also disrupted as rotation increases, resulting in a mean helicity that is no longer zero. This reveals that the rotational disruption of symmetry in the helicity PDF occurs at varying filter scales of
$r/\varDelta =10$
,
$20$
and
$40$
. Moreover, similar to the observations at
$r/\varDelta =20$
, the PDF of the SGS flux tends to shift with a skewness towards the negative side as rotation increases at filter scales of
$r/\varDelta =10$
and
$40$
. This skewness towards the negative side further demonstrates that the overall SGS flux is hindered by rotation at varying filter scales of
$r/\varDelta =10$
,
$20$
and
$40$
. It should be noted that although local negative SGS flux events are enhanced by rotation at different scales, the overall SGS flux remains positive, and the energy cascade in rotating RT turbulence is forward. As shown in figure 16, the effect of rotation on the joint PDF of normalised helicity and SGS flux at scales of
$r/\varDelta =10$
and
$40$
is similar to that at
$r/\varDelta =20$
. Specifically, the fractions of positive helicity and negative SGS flux within the mixing layer of RT turbulence also increase significantly for the strong rotation case at scales of
$r/\varDelta =10$
and
$40$
.
The spectra of helicity, shown in figure 17, illustrate the characteristics of helicity across the entire range of scales. Here, the spectra are normalised by the mean squared helicity. The helicity for case
$Ro=0.1$
is clearly greater than that for cases
$Ro=\infty$
and
$1.0$
at low wavenumbers, but it is smaller at high wavenumbers compared with the same cases. This indicates that rotation enhances the generation of helicity at low wavenumbers.
Spectra of helicity at the
$z=0$
plane for cases with
$Ro=\infty$
,
$1.0$
and
$0.1$
. Here the horizontal wavenumber
$k=\sqrt {k_x^2 + k_y^2}$
and the spectra are calculated at the time when
${\textit{Re}}_H=12\,000$
.

Moreover, we evaluate the average of normalised SGS flux
$\varPi _{r}/\varPi _{r}^{\prime }$
conditioned on the normalised helicity
$ H_{r}/H_{r}^{\prime }$
at filter scales
$r/\varDelta =10$
,
$20$
and
$40$
for three cases at the time when
${\textit{Re}}_H=12\,000$
in figure 18. Obviously,
$\left \langle \varPi _r/\varPi _{r}^{\prime }\vert H_{r}/H_{r}^{\prime } \right \rangle$
nearly collapses across different scales of
$r/\varDelta =10$
,
$20$
and
$40$
, indicating that the relation between helicity and SGS flux is quite similar at these scales. It is worth noting that the average normalised SGS flux is positive in regions where
$\left \lvert H_{r}/H_{r}^{\prime } \right \rvert \lessapprox 6$
(
$\left \lvert H_{r}/H_{r}^{\prime } \right \rvert \lessapprox 4$
in case
$Ro=0.1$
), while negative in regions where
$\left \lvert H_{r}/H_{r}^{\prime } \right \rvert \gtrapprox 6$
(
$\left \lvert H_{r}/H_{r}^{\prime } \right \rvert \gtrapprox 4$
in case
$Ro=0.1$
). This demonstrates that the low-helicity regions are more favourable for local forward energy transfer, while the high-helicity regions are more favourable for local inverse energy transfer. This phenomenon is more pronounced in highly helical turbulence.
The average of normalised SGS flux
$\varPi _{r}/\varPi _{r}^{\prime }$
conditioned on the normalised helicity
$ H_{r}/H_{r}^{\prime }$
at filter scales
$r/\varDelta =10$
,
$20$
and
$40$
for cases with (a–c)
$Ro=\infty$
,
$1.0$
and
$0.1$
at the time when
${\textit{Re}}_H=12\,000$
.

Based on the findings that the SGS flux is significantly affected by helicity, the Smagorinsky model is examined at various levels of normalised helicity, aiming to determine whether the classical SGS model is affected by helicity and its applicability to large-eddy simulations (LES) of rotating RT turbulence. The classical Smagorinsky model is an eddy-viscosity model, with a fundamental assumption that the anisotropic component of SGS stress
$\tau _{\textit{ij}} ^d = \tau _{\textit{ij}}-1/3\tau _{kk}\delta _{\textit{ij}}$
is linearly proportional to the anisotropic component of large-scale strain
$\widetilde {S}^d _{\textit{ij}} = \widetilde {S} _{\textit{ij}} - 1/3 \delta _{\textit{ij}}\widetilde {S} _{kk}$
(Smagorinsky Reference Smagorinsky1963; Moin et al. Reference Moin, Squires, Cabot and Lee1991). Here we define variable
$\beta _r$
to investigate the relation between the large-scale strain and the SGS stress, which is defined as
\begin{equation} \beta _r = \frac {\widetilde {\tau _{\textit{ij}} ^d} \widetilde {S}^d _{\textit{ij}}}{\left \lvert \widetilde {\tau } ^d \right \rvert \left \lvert \widetilde { S } ^d \right \rvert }, \end{equation}
where
$\lvert \widetilde {\tau ^d } \rvert = \sqrt {\widetilde {\tau _{\textit{ij}}^d} \widetilde {\tau _{\textit{ij}}^d}}$
and
$\lvert \widetilde {S ^d} \rvert = \sqrt {\widetilde {S^d _{\textit{ij}}} \widetilde {S ^d_{\textit{ij}}}}$
. Figure 19 presents the average of
$\beta _{r}$
conditioned on the normalised helicity
$ H_{r}/H_{r}^{\prime }$
for cases with
$Ro=\infty$
,
$1.0$
and
$0.1$
, covering filter scales of
$r/\varDelta =10$
,
$20$
and
$40$
. Clearly, the relation between helicity and
$\beta _r$
is nearly identical at scales
$r/\varDelta =10$
,
$20$
and
$40$
, as
$\langle \beta _r\vert H_{r}/H_{r}^{\prime } \rangle$
nearly collapses at these scales. It is observed in figure 19 that the minimum
$\langle \beta _r\vert H_{r}/H_{r}^{\prime } \rangle$
, occurring in the zero-helicity region, is only approximately
$-0.25$
in RT turbulence, which is far from the
$\langle \beta _r\rangle =-1$
in the classical Smagorinsky model. According to the results of Borue & Orszag (Reference Borue and Orszag1998), the correlation coefficient (equal to negative
$\beta _r$
) between the SGS stress and the Smagorinsky model ranges from approximately
$0.23$
to
$0.42$
with varying filter widths in 3-D isotropic turbulence. Moreover, in compressible isotropic turbulence, Wang et al. (Reference Wang, Yang, Shi, Xiao, He and Chen2013, Reference Wang, Wan, Chen and Chen2018) reported that the
$\beta _r$
value is approximately
$-0.3$
in the incompressible region. Obviously, the present
$\beta _r$
value in the zero-helicity region of rotating RT turbulence is consistent with
$\beta _r$
values observed in isotropic turbulence.
The average of
$\beta _{r}$
conditioned on the normalised helicity
$ H_{r}/H_{r}^{\prime }$
at filter scales
$r/\varDelta =10$
,
$20$
and
$40$
for cases with (a–c)
$Ro=\infty$
,
$1.0$
and
$0.1$
at the time when
${\textit{Re}}_H=12\,000$
.

Additionally, figure 19 shows that
$\langle \beta _r\vert H_{r}/H_{r}^{\prime } \rangle$
is negative in low-helicity regions, corresponding to forward energy cascade. Positive values of
$\langle \beta _r\vert H_{r}/H_{r}^{\prime } \rangle$
are again attained in high-helicity regions, revealing the reverse SGS flux of kinetic energy from small scales to large scales. In brief, the SGS stress is not antiparallel to the large-scale strain (Wang et al. Reference Wang, Wan, Chen and Chen2018). It is thus evident that both mean helicity and local helicity must be taken into account when developing SGS models for LES of highly helical turbulence. Moreover, this evidence substantiates the proposition that the assumption of linear correlation between the SGS stress and the large-scale strain is no longer valid in RT turbulence. This also clarifies the unsatisfactory a posteriori tests on LES of RT turbulence when using the classical Smagorinsky model (Luo et al. Reference Luo, Wang, Yuan, Zhou, Huang and Wang2023).
5. Conclusion
This study systematically examines the effects of rotation on RT turbulence using DNS, focusing primarily on the generation of kinetic energy and enstrophy, as well as the scale-to-scale transfer of kinetic energy. The results are briefly summarised as follows.
The fundamental features of rotating RT turbulence, including flow structures, mixing width, mixed mass and anisotropy, have been analysed numerically. Compared with non-rotating RT flow, rotation induces more coherent and elongated flow structures. As RTI evolves, the mixing width
$h$
also follows a quadratic growth law similar to the non-rotating case, but its growth is significantly delayed and suppressed under stronger rotation. In particular, a lower
$Ro$
(stronger rotation) leads to greater asymmetry in the mixing layer and more pronounced suppression of bubble growth. The mixed mass
$M$
exhibits a quadratic growth over time similar to the mixing width, but is also significantly suppressed at low
$Ro$
. Interestingly, the normalised mixed mass increases with decreasing
$Ro$
, indicating enhanced efficiency of fluid mixing within the mixing layer under stronger rotation.
For the generation of kinetic energy, it is found that the growth rates of total and vertical kinetic energy are gradually reduced with enhancing rotation. To reveal the associated mechanisms, a systematic analysis of the kinetic energy budget is conducted. In rotating RT turbulence, energy conversion efficiency significantly decreases from approximately
$50\,\%$
in the non-rotating case (
$Ro = \infty$
) to just
$10\,\%$
in the strongly rotating case (
$Ro = 0.1$
). Subsequently, an analysis of energy dissipation reveals that rotation enhances dissipation primarily via intensifying the shear stress, particularly in the
$xz$
and
$yz$
components. This enhanced dissipation suppresses the growths of vertical velocity and the mixing layer, thereby leading to a reduction in anisotropy of velocity. In addition, the spectra of shear stress reveal that the enhanced dissipation is associated with the vertical filamentary structures present in cases of strong rotation. Moreover, the pressure–dilatation power term fluctuates between positive and negative values in rotating cases, while this term remains positive in the non-rotating case.
For enstrophy generation, the growth rates of both total enstrophy and vertical enstrophy are gradually reduced with decreasing
$Ro$
. Based on the enstrophy transport equation, it is found that the baroclinic effects dominate in the early stages of flow evolution, whereas vortex stretching and tilting become the main contributors in later stages. Furthermore, in the generation of vertical enstrophy, the vortex stretching and tilting term, along with the Coriolis force term, play significant roles, while the contribution of the baroclinic term is negligible. Notably, the vortex stretching and tilting term is significantly suppressed by rotation, indicating that rotation tends to make the enstrophy generation in 3-D RT turbulence behave more like that in 2-D flow.
Regarding the scale-to-scale transfer of kinetic energy, it is found that the kinetic energy is predominantly injected by gravity and pressure–dilatation work. As
$Ro$
decreases, both gravity input and pressure–dilatation work decline. Hence, rotation hinders the conversion of potential and internal energy into kinetic energy. Moreover, the overall SGS flux is always positive, corresponding to a forward cascade. However, this flux diminishes as
$Ro$
decreases, indicating that rotation induces suppression of the scale-to-scale transfer of kinetic energy. Moreover, a detailed analysis of helicity reveals a higher likelihood of local inverse energy transfer events under enhanced rotation. In particular, low-helicity regions are conducive to forward energy cascades, whereas high-helicity regions favour local inverse energy transfer. In strongly helical turbulence (e.g.
$Ro=0.1$
), the latter becomes especially pronounced. Furthermore, helicity reduces the parallelism between large-scale strain and SGS stress, highlighting that the classical Smagorinsky model is unsuitable for LES of strong rotational RT turbulence.
Overall, the intriguing phenomena observed in this work concerning the effect of rotation on the fundamental features and energy transfer of RT turbulence warrant further study in the future. Specifically, it is noteworthy that the suppressed growth of the mixing layer and the enhanced efficiency of fluid mixing due to rotation lead to a non-monotonic trend of mixed mass with decreasing
$Ro$
in rotating RT turbulence. However, it remains unclear at what value of
$Ro$
the maximum mixed mass is achieved and whether the Atwood number
$A_t$
influences this
$Ro$
value. This warrants further investigation into rotating RT turbulence in the future and could aid in regulating fluid mixing through rotation. Moreover, this work indicates that high helicity promotes local inverse energy transfer and reduces the parallelism between large-scale strain and SGS stress in rotating RT turbulence. Therefore, modelling the SGS stress in high-helicity RT turbulence warrants further investigation, as it is essential for the LES of strong rotational RT turbulence.
Acknowledgements
The authors are very grateful to Dr Y.-S. Zhang at the Institute of Applied Physics and Computational Mathematics for useful discussions on the algorithm and code.
Funding
This work was supported by the Science Challenge Project (no. TZ2025016), by the National Natural Science Foundation of China (nos. 12388101, 12202436, 12402257) and by the Strategic Priority Research Program of the Chinese Academy of Sciences (no. XDB0500301).
Declaration of interests
The authors report no conflict of interest.





























































































































































